xxq250
/
homework-jianmu
1
0
Fork 0
Code Issues Pull Requests 1 Actions Packages Projects Releases Wiki Activity

Merge remote-tracking branch 'origin/3.0' into feature/mnode

This commit is contained in:
Shengliang Guan 2022-02-09 14:38:26 +08:00
parent 9a164f48be b17e99371b
commit 22f45b4d41
76 changed files with 1940 additions and 5316 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

1
include/common/tmsg.h
View File

@ -726,6 +726,7 @@ typedef struct {
char tbName[TSDB_TABLE_NAME_LEN]; char tbName[TSDB_TABLE_NAME_LEN];
char stbName[TSDB_TABLE_NAME_LEN]; char stbName[TSDB_TABLE_NAME_LEN];
char dbFName[TSDB_DB_FNAME_LEN]; char dbFName[TSDB_DB_FNAME_LEN];
uint64_t dbId;
int32_t numOfTags; int32_t numOfTags;
int32_t numOfColumns; int32_t numOfColumns;
int8_t precision; int8_t precision;

13
include/libs/catalog/catalog.h
View File

@ -32,6 +32,15 @@ extern "C" {
struct SCatalog; struct SCatalog;
enum {
CTG_DBG_DB_NUM = 1,
CTG_DBG_META_NUM,
CTG_DBG_STB_NUM,
CTG_DBG_DB_RENT_NUM,
CTG_DBG_STB_RENT_NUM,
};
typedef struct SCatalogReq { typedef struct SCatalogReq {
SArray *pTableName; // element is SNAME SArray *pTableName; // element is SNAME
SArray *pUdf; // udf name SArray *pUdf; // udf name
@ -99,7 +108,7 @@ int32_t catalogGetDBVgroupVersion(struct SCatalog* pCatalog, const char* dbName,
*/ */
int32_t catalogGetDBVgroup(struct SCatalog* pCatalog, void *pTransporter, const SEpSet* pMgmtEps, const char* pDBName, bool forceUpdate, SArray** pVgroupList); int32_t catalogGetDBVgroup(struct SCatalog* pCatalog, void *pTransporter, const SEpSet* pMgmtEps, const char* pDBName, bool forceUpdate, SArray** pVgroupList);
int32_t catalogUpdateDBVgroup(struct SCatalog* pCatalog, const char* dbName, SDBVgroupInfo* dbInfo); int32_t catalogUpdateDBVgroup(struct SCatalog* pCatalog, const char* dbName, uint64_t dbId, SDBVgroupInfo* dbInfo);
int32_t catalogRemoveDB(struct SCatalog* pCatalog, const char* dbName, uint64_t dbId); int32_t catalogRemoveDB(struct SCatalog* pCatalog, const char* dbName, uint64_t dbId);
@ -127,6 +136,8 @@ int32_t catalogGetTableMeta(struct SCatalog* pCatalog, void * pTransporter, cons
*/ */
int32_t catalogGetSTableMeta(struct SCatalog* pCatalog, void * pTransporter, const SEpSet* pMgmtEps, const SName* pTableName, STableMeta** pTableMeta); int32_t catalogGetSTableMeta(struct SCatalog* pCatalog, void * pTransporter, const SEpSet* pMgmtEps, const SName* pTableName, STableMeta** pTableMeta);
int32_t catalogUpdateSTableMeta(struct SCatalog* pCatalog, STableMetaRsp *rspMsg);
/** /**
* Force renew a table's local cached meta data. * Force renew a table's local cached meta data.

5
include/libs/qcom/query.h
View File

@ -81,7 +81,6 @@ typedef struct STableMeta {
} STableMeta; } STableMeta;
typedef struct SDBVgroupInfo { typedef struct SDBVgroupInfo {
uint64_t dbId;
int32_t vgVersion; int32_t vgVersion;
int8_t hashMethod; int8_t hashMethod;
SHashObj *vgHash; //key:vgId, value:SVgroupInfo SHashObj *vgHash; //key:vgId, value:SVgroupInfo
@ -89,6 +88,7 @@ typedef struct SDBVgroupInfo {
typedef struct SUseDbOutput { typedef struct SUseDbOutput {
char db[TSDB_DB_FNAME_LEN]; char db[TSDB_DB_FNAME_LEN];
uint64_t dbId;
SDBVgroupInfo *dbVgroup; SDBVgroupInfo *dbVgroup;
} SUseDbOutput; } SUseDbOutput;
@ -102,6 +102,7 @@ enum {
typedef struct STableMetaOutput { typedef struct STableMetaOutput {
int32_t metaType; int32_t metaType;
uint64_t dbId;
char dbFName[TSDB_DB_FNAME_LEN]; char dbFName[TSDB_DB_FNAME_LEN];
char ctbName[TSDB_TABLE_NAME_LEN]; char ctbName[TSDB_TABLE_NAME_LEN];
char tbName[TSDB_TABLE_NAME_LEN]; char tbName[TSDB_TABLE_NAME_LEN];
@ -159,6 +160,8 @@ void initQueryModuleMsgHandle();
const SSchema* tGetTbnameColumnSchema(); const SSchema* tGetTbnameColumnSchema();
bool tIsValidSchema(struct SSchema* pSchema, int32_t numOfCols, int32_t numOfTags); bool tIsValidSchema(struct SSchema* pSchema, int32_t numOfCols, int32_t numOfTags);
int32_t queryCreateTableMetaFromMsg(STableMetaRsp* msg, bool isSuperTable, STableMeta **pMeta);
extern int32_t (*queryBuildMsg[TDMT_MAX])(void* input, char **msg, int32_t msgSize, int32_t *msgLen); extern int32_t (*queryBuildMsg[TDMT_MAX])(void* input, char **msg, int32_t msgSize, int32_t *msgLen);
extern int32_t (*queryProcessMsgRsp[TDMT_MAX])(void* output, char *msg, int32_t msgSize); extern int32_t (*queryProcessMsgRsp[TDMT_MAX])(void* output, char *msg, int32_t msgSize);

159
include/libs/sync/sync.h
View File

@ -1,159 +0,0 @@
/*
* Copyright (c) 2019 TAOS Data, Inc. <cli@taosdata.com>
*
* This program is free software: you can use, redistribute, and/or modify
* it under the terms of the GNU Affero General Public License, version 3
* or later ("AGPL"), as published by the Free Software Foundation.
*
* This program is distributed in the hope that it will be useful, but WITHOUT
* ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or
* FITNESS FOR A PARTICULAR PURPOSE.
*
* You should have received a copy of the GNU Affero General Public License
* along with this program. If not, see <http://www.gnu.org/licenses/>.
*/
#ifndef _TD_LIBS_SYNC_H
#define _TD_LIBS_SYNC_H
#ifdef __cplusplus
extern "C" {
#endif
#include <stdint.h>
#include "taosdef.h"
typedef int32_t SyncNodeId;
typedef int32_t SyncGroupId;
typedef int64_t SyncIndex;
typedef uint64_t SyncTerm;
typedef enum {
TAOS_SYNC_STATE_FOLLOWER = 0,
TAOS_SYNC_STATE_CANDIDATE = 1,
TAOS_SYNC_STATE_LEADER = 2,
} ESyncState;
typedef struct {
void* data;
size_t len;
} SSyncBuffer;
typedef struct {
SyncNodeId nodeId;
uint16_t nodePort; // node sync Port
char nodeFqdn[TSDB_FQDN_LEN]; // node FQDN
} SNodeInfo;
typedef struct {
int32_t selfIndex;
int32_t replica;
SNodeInfo nodeInfo[TSDB_MAX_REPLICA];
} SSyncCluster;
typedef struct {
int32_t selfIndex;
int32_t replica;
SNodeInfo node[TSDB_MAX_REPLICA];
ESyncState role[TSDB_MAX_REPLICA];
} SNodesRole;
typedef struct SSyncFSM {
void* pData;
// apply committed log, bufs will be free by sync module
int32_t (*applyLog)(struct SSyncFSM* fsm, SyncIndex index, const SSyncBuffer* buf, void* pData);
// cluster commit callback
int32_t (*onClusterChanged)(struct SSyncFSM* fsm, const SSyncCluster* cluster, void* pData);
// fsm return snapshot in ppBuf, bufs will be free by sync module
// TODO: getSnapshot SHOULD be async?
int32_t (*getSnapshot)(struct SSyncFSM* fsm, SSyncBuffer** ppBuf, int32_t* objId, bool* isLast);
// fsm apply snapshot with pBuf data
int32_t (*applySnapshot)(struct SSyncFSM* fsm, SSyncBuffer* pBuf, int32_t objId, bool isLast);
// call when restore snapshot and log done
int32_t (*onRestoreDone)(struct SSyncFSM* fsm);
void (*onRollback)(struct SSyncFSM* fsm, SyncIndex index, const SSyncBuffer* buf);
void (*onRoleChanged)(struct SSyncFSM* fsm, const SNodesRole* pRole);
} SSyncFSM;
typedef struct SSyncLogStore {
void* pData;
// write log with given index
int32_t (*logWrite)(struct SSyncLogStore* logStore, SyncIndex index, SSyncBuffer* pBuf);
/**
* read log from given index(included) with limit, return the actual num in nBuf,
* pBuf will be free in sync module
**/
int32_t (*logRead)(struct SSyncLogStore* logStore, SyncIndex index, int limit,
SSyncBuffer* pBuf, int* nBuf);
// mark log with given index has been commtted
int32_t (*logCommit)(struct SSyncLogStore* logStore, SyncIndex index);
// prune log before given index(not included)
int32_t (*logPrune)(struct SSyncLogStore* logStore, SyncIndex index);
// rollback log after given index(included)
int32_t (*logRollback)(struct SSyncLogStore* logStore, SyncIndex index);
// return last index of log
SyncIndex (*logLastIndex)(struct SSyncLogStore* logStore);
} SSyncLogStore;
typedef struct SStateManager {
void* pData;
// save serialized server state data, buffer will be free by Sync
int32_t (*saveServerState)(struct SStateManager* stateMng, const char* buffer, int n);
// read serialized server state data, buffer will be free by Sync
int32_t (*readServerState)(struct SStateManager* stateMng, char** ppBuffer, int* n);
// save serialized cluster state data, buffer will be free by Sync
void (*saveClusterState)(struct SStateManager* stateMng, const char* buffer, int n);
// read serialized cluster state data, buffer will be free by Sync
int32_t (*readClusterState)(struct SStateManager* stateMng, char** ppBuffer, int* n);
} SStateManager;
typedef struct {
SyncGroupId vgId;
SyncIndex appliedIndex;
SSyncCluster syncCfg;
SSyncFSM fsm;
SSyncLogStore logStore;
SStateManager stateManager;
} SSyncInfo;
struct SSyncNode;
typedef struct SSyncNode SSyncNode;
int32_t syncInit();
void syncCleanUp();
SSyncNode* syncStart(const SSyncInfo*);
void syncReconfig(const SSyncNode*, const SSyncCluster*);
void syncStop(const SSyncNode*);
int32_t syncPropose(SSyncNode* syncNode, const SSyncBuffer* pBuf, void* pData, bool isWeak);
int32_t syncAddNode(SSyncNode syncNode, const SNodeInfo *pNode);
int32_t syncRemoveNode(SSyncNode syncNode, const SNodeInfo *pNode);
extern int32_t sDebugFlag;
#ifdef __cplusplus
}
#endif
#endif /*_TD_LIBS_SYNC_H*/

21
include/nodes/nodes.h
View File

@ -54,6 +54,9 @@ typedef enum ENodeType {
QUERY_NODE_NODE_LIST, QUERY_NODE_NODE_LIST,
QUERY_NODE_FILL, QUERY_NODE_FILL,
// only for parser
QUERY_NODE_RAW_EXPR,
QUERY_NODE_SET_OPERATOR, QUERY_NODE_SET_OPERATOR,
QUERY_NODE_SELECT_STMT, QUERY_NODE_SELECT_STMT,
QUERY_NODE_SHOW_STMT QUERY_NODE_SHOW_STMT
@ -78,10 +81,12 @@ typedef struct SNodeList {
SListCell* pTail; SListCell* pTail;
} SNodeList; } SNodeList;
typedef struct SNameStr { typedef struct SRawExprNode {
int32_t len; ENodeType nodeType;
char* pName; char* p;
} SNameStr; uint32_t n;
SNode* pNode;
} SRawExprNode;
typedef struct SDataType { typedef struct SDataType {
uint8_t type; uint8_t type;
@ -114,7 +119,7 @@ typedef struct SColumnNode {
} SColumnNode; } SColumnNode;
typedef struct SValueNode { typedef struct SValueNode {
SExprNode type; // QUERY_NODE_VALUE SExprNode node; // QUERY_NODE_VALUE
char* literal; char* literal;
} SValueNode; } SValueNode;
@ -146,7 +151,7 @@ typedef enum EOperatorType {
} EOperatorType; } EOperatorType;
typedef struct SOperatorNode { typedef struct SOperatorNode {
SExprNode type; // QUERY_NODE_OPERATOR SExprNode node; // QUERY_NODE_OPERATOR
EOperatorType opType; EOperatorType opType;
SNode* pLeft; SNode* pLeft;
SNode* pRight; SNode* pRight;
@ -332,6 +337,10 @@ void nodesCloneNode(const SNode* pNode);
int32_t nodesNodeToString(const SNode* pNode, char** pStr, int32_t* pLen); int32_t nodesNodeToString(const SNode* pNode, char** pStr, int32_t* pLen);
int32_t nodesStringToNode(const char* pStr, SNode** pNode); int32_t nodesStringToNode(const char* pStr, SNode** pNode);
bool nodesIsArithmeticOp(const SOperatorNode* pOp);
bool nodesIsComparisonOp(const SOperatorNode* pOp);
bool nodesIsJsonOp(const SOperatorNode* pOp);
bool nodesIsTimeorderQuery(const SNode* pQuery); bool nodesIsTimeorderQuery(const SNode* pQuery);
bool nodesIsTimelineQuery(const SNode* pQuery); bool nodesIsTimelineQuery(const SNode* pQuery);

3
include/util/taoserror.h
View File

@ -436,6 +436,8 @@ int32_t* taosGetErrno();
#define TSDB_CODE_CTG_NOT_READY TAOS_DEF_ERROR_CODE(0, 0x2402) //catalog is not ready #define TSDB_CODE_CTG_NOT_READY TAOS_DEF_ERROR_CODE(0, 0x2402) //catalog is not ready
#define TSDB_CODE_CTG_MEM_ERROR TAOS_DEF_ERROR_CODE(0, 0x2403) //catalog memory error #define TSDB_CODE_CTG_MEM_ERROR TAOS_DEF_ERROR_CODE(0, 0x2403) //catalog memory error
#define TSDB_CODE_CTG_SYS_ERROR TAOS_DEF_ERROR_CODE(0, 0x2404) //catalog system error #define TSDB_CODE_CTG_SYS_ERROR TAOS_DEF_ERROR_CODE(0, 0x2404) //catalog system error
#define TSDB_CODE_CTG_DB_DROPPED TAOS_DEF_ERROR_CODE(0, 0x2405) //Database is dropped
#define TSDB_CODE_CTG_OUT_OF_SERVICE TAOS_DEF_ERROR_CODE(0, 0x2406) //catalog is out of service
//scheduler //scheduler
#define TSDB_CODE_SCH_STATUS_ERROR TAOS_DEF_ERROR_CODE(0, 0x2501) //scheduler status error #define TSDB_CODE_SCH_STATUS_ERROR TAOS_DEF_ERROR_CODE(0, 0x2501) //scheduler status error
@ -445,6 +447,7 @@ int32_t* taosGetErrno();
#define TSDB_CODE_PARSER_INVALID_COLUMN TAOS_DEF_ERROR_CODE(0, 0x2601) //invalid column name #define TSDB_CODE_PARSER_INVALID_COLUMN TAOS_DEF_ERROR_CODE(0, 0x2601) //invalid column name
#define TSDB_CODE_PARSER_TABLE_NOT_EXIST TAOS_DEF_ERROR_CODE(0, 0x2602) //table not exist #define TSDB_CODE_PARSER_TABLE_NOT_EXIST TAOS_DEF_ERROR_CODE(0, 0x2602) //table not exist
#define TSDB_CODE_PARSER_AMBIGUOUS_COLUMN TAOS_DEF_ERROR_CODE(0, 0x2603) //ambiguous column #define TSDB_CODE_PARSER_AMBIGUOUS_COLUMN TAOS_DEF_ERROR_CODE(0, 0x2603) //ambiguous column
#define TSDB_CODE_PARSER_WRONG_VALUE_TYPE TAOS_DEF_ERROR_CODE(0, 0x2604) //wrong value type
#ifdef __cplusplus #ifdef __cplusplus
} }

58
source/client/src/clientHb.c
View File

@ -44,7 +44,6 @@ static int32_t hbProcessDBInfoRsp(void *value, int32_t valueLen, struct SCatalog
code = catalogRemoveDB(pCatalog, rsp->db, rsp->uid); code = catalogRemoveDB(pCatalog, rsp->db, rsp->uid);
} else { } else {
SDBVgroupInfo vgInfo = {0}; SDBVgroupInfo vgInfo = {0};
vgInfo.dbId = rsp->uid;
vgInfo.vgVersion = rsp->vgVersion; vgInfo.vgVersion = rsp->vgVersion;
vgInfo.hashMethod = rsp->hashMethod; vgInfo.hashMethod = rsp->hashMethod;
vgInfo.vgHash = taosHashInit(rsp->vgNum, taosGetDefaultHashFunction(TSDB_DATA_TYPE_INT), true, HASH_ENTRY_LOCK); vgInfo.vgHash = taosHashInit(rsp->vgNum, taosGetDefaultHashFunction(TSDB_DATA_TYPE_INT), true, HASH_ENTRY_LOCK);
@ -69,7 +68,7 @@ static int32_t hbProcessDBInfoRsp(void *value, int32_t valueLen, struct SCatalog
} }
} }
code = catalogUpdateDBVgroup(pCatalog, rsp->db, &vgInfo); code = catalogUpdateDBVgroup(pCatalog, rsp->db, rsp->uid, &vgInfo);
if (code) { if (code) {
taosHashCleanup(vgInfo.vgHash); taosHashCleanup(vgInfo.vgHash);
} }
@ -101,50 +100,33 @@ static int32_t hbProcessStbInfoRsp(void *value, int32_t valueLen, struct SCatalo
tscDebug("hb remove stb, db:%s, stb:%s", rsp->dbFName, rsp->stbName); tscDebug("hb remove stb, db:%s, stb:%s", rsp->dbFName, rsp->stbName);
code = catalogRemoveSTableMeta(pCatalog, rsp->dbFName, rsp->stbName, rsp->suid); catalogRemoveSTableMeta(pCatalog, rsp->dbFName, rsp->stbName, rsp->suid);
} else { } else {
tscDebug("hb update stb, db:%s, stb:%s", rsp->dbFName, rsp->stbName);
rsp->numOfTags = ntohl(rsp->numOfTags); rsp->numOfTags = ntohl(rsp->numOfTags);
rsp->sversion = ntohl(rsp->sversion);
rsp->tversion = ntohl(rsp->tversion);
rsp->tuid = be64toh(rsp->tuid);
rsp->vgId = ntohl(rsp->vgId);
SSchema* pSchema = rsp->pSchema;
schemaNum = rsp->numOfColumns + rsp->numOfTags; schemaNum = rsp->numOfColumns + rsp->numOfTags;
/*
rsp->vgNum = ntohl(rsp->vgNum);
rsp->uid = be64toh(rsp->uid);
SDBVgroupInfo vgInfo = {0}; for (int i = 0; i < schemaNum; ++i) {
vgInfo.dbId = rsp->uid; pSchema->bytes = ntohl(pSchema->bytes);
vgInfo.vgVersion = rsp->vgVersion; pSchema->colId = ntohl(pSchema->colId);
vgInfo.hashMethod = rsp->hashMethod;
vgInfo.vgHash = taosHashInit(rsp->vgNum, taosGetDefaultHashFunction(TSDB_DATA_TYPE_INT), true, HASH_ENTRY_LOCK); pSchema++;
if (NULL == vgInfo.vgHash) {
tscError("hash init[%d] failed", rsp->vgNum);
return TSDB_CODE_TSC_OUT_OF_MEMORY;
} }
for (int32_t i = 0; i < rsp->vgNum; ++i) { if (rsp->pSchema[0].colId != PRIMARYKEY_TIMESTAMP_COL_ID) {
rsp->vgroupInfo[i].vgId = ntohl(rsp->vgroupInfo[i].vgId); tscError("invalid colId[%d] for the first column in table meta rsp msg", rsp->pSchema[0].colId);
rsp->vgroupInfo[i].hashBegin = ntohl(rsp->vgroupInfo[i].hashBegin); return TSDB_CODE_TSC_INVALID_VALUE;
rsp->vgroupInfo[i].hashEnd = ntohl(rsp->vgroupInfo[i].hashEnd); }
for (int32_t n = 0; n < rsp->vgroupInfo[i].epset.numOfEps; ++n) { catalogUpdateSTableMeta(pCatalog, rsp);
rsp->vgroupInfo[i].epset.eps[n].port = ntohs(rsp->vgroupInfo[i].epset.eps[n].port);
}
if (0 != taosHashPut(vgInfo.vgHash, &rsp->vgroupInfo[i].vgId, sizeof(rsp->vgroupInfo[i].vgId), &rsp->vgroupInfo[i], sizeof(rsp->vgroupInfo[i]))) {
tscError("hash push failed, errno:%d", errno);
taosHashCleanup(vgInfo.vgHash);
return TSDB_CODE_TSC_OUT_OF_MEMORY;
}
}
code = catalogUpdateDBVgroup(pCatalog, rsp->db, &vgInfo);
if (code) {
taosHashCleanup(vgInfo.vgHash);
}
*/
}
if (code) {
return code;
} }
msgLen += sizeof(STableMetaRsp) + schemaNum * sizeof(SSchema); msgLen += sizeof(STableMetaRsp) + schemaNum * sizeof(SSchema);

1
source/dnode/mnode/impl/src/mndStb.c
View File

@ -1178,6 +1178,7 @@ static int32_t mndProcessStbMetaReq(SMnodeMsg *pReq) {
strcpy(pMeta->dbFName, pStb->db); strcpy(pMeta->dbFName, pStb->db);
strcpy(pMeta->tbName, pInfo->tbName); strcpy(pMeta->tbName, pInfo->tbName);
strcpy(pMeta->stbName, pInfo->tbName); strcpy(pMeta->stbName, pInfo->tbName);
pMeta->dbId = htobe64(pDb->uid);
pMeta->numOfTags = htonl(pStb->numOfTags); pMeta->numOfTags = htonl(pStb->numOfTags);
pMeta->numOfColumns = htonl(pStb->numOfColumns); pMeta->numOfColumns = htonl(pStb->numOfColumns);
pMeta->precision = pDb->cfg.precision; pMeta->precision = pDb->cfg.precision;

2
source/dnode/vnode/src/tsdb/tsdbCommit.c
View File

@ -1327,7 +1327,7 @@ static int tsdbMergeMemData(SCommitH *pCommith, SCommitIter *pIter, int bidx) {
int nBlocks = pCommith->readh.pBlkIdx->numOfBlocks; int nBlocks = pCommith->readh.pBlkIdx->numOfBlocks;
SBlock * pBlock = pCommith->readh.pBlkInfo->blocks + bidx; SBlock * pBlock = pCommith->readh.pBlkInfo->blocks + bidx;
TSKEY keyLimit; TSKEY keyLimit;
int16_t colId = 0; int16_t colId = PRIMARYKEY_TIMESTAMP_COL_ID;
SMergeInfo mInfo; SMergeInfo mInfo;
SBlock subBlocks[TSDB_MAX_SUBBLOCKS]; SBlock subBlocks[TSDB_MAX_SUBBLOCKS];
SBlock block, supBlock; SBlock block, supBlock;

2
source/dnode/vnode/src/tsdb/tsdbReadImpl.c
View File

@ -472,7 +472,7 @@ static int tsdbLoadBlockDataImpl(SReadH *pReadh, SBlock *pBlock, SDataCols *pDat
continue; continue;
} }
int16_t tcolId = 0; int16_t tcolId = PRIMARYKEY_TIMESTAMP_COL_ID;
uint32_t toffset = TSDB_KEY_COL_OFFSET; uint32_t toffset = TSDB_KEY_COL_OFFSET;
int32_t tlen = pBlock->keyLen; int32_t tlen = pBlock->keyLen;

1
source/dnode/vnode/src/vnd/vnodeBufferPool.c
View File

@ -185,6 +185,7 @@ static void vBufPoolDestroyMA(SMemAllocatorFactory *pMAF, SMemAllocator *pMA) {
free(pMA); free(pMA);
if (--pVMA->_ref.val == 0) { if (--pVMA->_ref.val == 0) {
TD_DLIST_POP(&(pVnode->pBufPool->incycle), pVMA); TD_DLIST_POP(&(pVnode->pBufPool->incycle), pVMA);
vmaReset(pVMA);
TD_DLIST_APPEND(&(pVnode->pBufPool->free), pVMA); TD_DLIST_APPEND(&(pVnode->pBufPool->free), pVMA);
} }
} }

25
source/libs/catalog/inc/catalogInt.h
View File

@ -48,18 +48,22 @@ enum {
}; };
typedef struct SCtgDebug { typedef struct SCtgDebug {
int32_t lockDebug; bool lockDebug;
bool cacheDebug;
uint32_t showCachePeriodSec;
} SCtgDebug; } SCtgDebug;
typedef struct SCtgTbMetaCache { typedef struct SCtgTbMetaCache {
SRWLatch stbLock; SRWLatch stbLock;
SHashObj *cache; //key:tbname, value:STableMeta SRWLatch metaLock; // RC between cache destroy and all other operations
SHashObj *metaCache; //key:tbname, value:STableMeta
SHashObj *stbCache; //key:suid, value:STableMeta* SHashObj *stbCache; //key:suid, value:STableMeta*
} SCtgTbMetaCache; } SCtgTbMetaCache;
typedef struct SCtgDBCache { typedef struct SCtgDBCache {
SRWLatch vgLock; SRWLatch vgLock;
uint64_t dbId;
int8_t deleted; int8_t deleted;
SDBVgroupInfo *vgInfo; SDBVgroupInfo *vgInfo;
SCtgTbMetaCache tbCache; SCtgTbMetaCache tbCache;
@ -81,6 +85,7 @@ typedef struct SCtgRentMgmt {
typedef struct SCatalog { typedef struct SCatalog {
uint64_t clusterId; uint64_t clusterId;
SRWLatch dbLock;
SHashObj *dbCache; //key:dbname, value:SCtgDBCache SHashObj *dbCache; //key:dbname, value:SCtgDBCache
SCtgRentMgmt dbRent; SCtgRentMgmt dbRent;
SCtgRentMgmt stbRent; SCtgRentMgmt stbRent;
@ -105,6 +110,8 @@ typedef struct SCatalogStat {
} SCatalogStat; } SCatalogStat;
typedef struct SCatalogMgmt { typedef struct SCatalogMgmt {
bool exit;
SRWLatch lock;
SHashObj *pCluster; //key: clusterId, value: SCatalog* SHashObj *pCluster; //key: clusterId, value: SCatalog*
SCatalogStat stat; SCatalogStat stat;
SCatalogCfg cfg; SCatalogCfg cfg;
@ -132,11 +139,8 @@ typedef uint32_t (*tableNameHashFp)(const char *, uint32_t);
#define ctgDebug(param, ...) qDebug("CTG:%p " param, pCatalog, __VA_ARGS__) #define ctgDebug(param, ...) qDebug("CTG:%p " param, pCatalog, __VA_ARGS__)
#define ctgTrace(param, ...) qTrace("CTG:%p " param, pCatalog, __VA_ARGS__) #define ctgTrace(param, ...) qTrace("CTG:%p " param, pCatalog, __VA_ARGS__)
#define CTG_ERR_RET(c) do { int32_t _code = c; if (_code != TSDB_CODE_SUCCESS) { terrno = _code; return _code; } } while (0)
#define CTG_RET(c) do { int32_t _code = c; if (_code != TSDB_CODE_SUCCESS) { terrno = _code; } return _code; } while (0)
#define CTG_ERR_JRET(c) do { code = c; if (code != TSDB_CODE_SUCCESS) { terrno = code; goto _return; } } while (0)
#define CTG_LOCK_DEBUG(...) do { if (gCTGDebug.lockDebug) { qDebug(__VA_ARGS__); } } while (0) #define CTG_LOCK_DEBUG(...) do { if (gCTGDebug.lockDebug) { qDebug(__VA_ARGS__); } } while (0)
#define CTG_CACHE_DEBUG(...) do { if (gCTGDebug.cacheDebug) { qDebug(__VA_ARGS__); } } while (0)
#define TD_RWLATCH_WRITE_FLAG_COPY 0x40000000 #define TD_RWLATCH_WRITE_FLAG_COPY 0x40000000
@ -172,6 +176,15 @@ typedef uint32_t (*tableNameHashFp)(const char *, uint32_t);
} \ } \
} while (0) } while (0)
#define CTG_ERR_RET(c) do { int32_t _code = c; if (_code != TSDB_CODE_SUCCESS) { terrno = _code; return _code; } } while (0)
#define CTG_RET(c) do { int32_t _code = c; if (_code != TSDB_CODE_SUCCESS) { terrno = _code; } return _code; } while (0)
#define CTG_ERR_JRET(c) do { code = c; if (code != TSDB_CODE_SUCCESS) { terrno = code; goto _return; } } while (0)
#define CTG_API_ENTER() do { CTG_LOCK(CTG_READ, &ctgMgmt.lock); if (atomic_load_8(&ctgMgmt.exit)) { CTG_RET(TSDB_CODE_CTG_OUT_OF_SERVICE); } } while (0)
#define CTG_API_LEAVE(c) do { int32_t __code = c; CTG_UNLOCK(CTG_READ, &ctgMgmt.lock); CTG_RET(__code); } while (0)
#ifdef __cplusplus #ifdef __cplusplus
} }

891
source/libs/catalog/src/catalog.c
View File

File diff suppressed because it is too large Load Diff

256
source/libs/catalog/test/catalogTests.cpp
View File

@ -39,6 +39,7 @@ namespace {
extern "C" int32_t ctgGetTableMetaFromCache(struct SCatalog *pCatalog, const SName *pTableName, STableMeta **pTableMeta, extern "C" int32_t ctgGetTableMetaFromCache(struct SCatalog *pCatalog, const SName *pTableName, STableMeta **pTableMeta,
int32_t *exist); int32_t *exist);
extern "C" int32_t ctgUpdateTableMetaCache(struct SCatalog *pCatalog, STableMetaOutput *output); extern "C" int32_t ctgUpdateTableMetaCache(struct SCatalog *pCatalog, STableMetaOutput *output);
extern "C" int32_t ctgDbgGetClusterCacheNum(struct SCatalog* pCatalog, int32_t type);
void ctgTestSetPrepareTableMeta(); void ctgTestSetPrepareTableMeta();
void ctgTestSetPrepareCTableMeta(); void ctgTestSetPrepareCTableMeta();
@ -49,7 +50,7 @@ bool ctgTestStop = false;
bool ctgTestEnableSleep = false; bool ctgTestEnableSleep = false;
bool ctgTestDeadLoop = false; bool ctgTestDeadLoop = false;
int32_t ctgTestPrintNum = 200000; int32_t ctgTestPrintNum = 200000;
int32_t ctgTestMTRunSec = 30; int32_t ctgTestMTRunSec = 5;
int32_t ctgTestCurrentVgVersion = 0; int32_t ctgTestCurrentVgVersion = 0;
int32_t ctgTestVgVersion = 1; int32_t ctgTestVgVersion = 1;
@ -107,6 +108,7 @@ void ctgTestInitLogFile() {
const int32_t maxLogFileNum = 10; const int32_t maxLogFileNum = 10;
tsAsyncLog = 0; tsAsyncLog = 0;
qDebugFlag = 159;
char temp[128] = {0}; char temp[128] = {0};
sprintf(temp, "%s/%s", tsLogDir, defaultLogFileNamePrefix); sprintf(temp, "%s/%s", tsLogDir, defaultLogFileNamePrefix);
@ -185,7 +187,6 @@ void ctgTestBuildDBVgroup(SDBVgroupInfo **pdbVgroup) {
ctgTestCurrentVgVersion = dbVgroup->vgVersion; ctgTestCurrentVgVersion = dbVgroup->vgVersion;
dbVgroup->hashMethod = 0; dbVgroup->hashMethod = 0;
dbVgroup->dbId = ctgTestDbId;
dbVgroup->vgHash = taosHashInit(ctgTestVgNum, taosGetDefaultHashFunction(TSDB_DATA_TYPE_INT), true, HASH_ENTRY_LOCK); dbVgroup->vgHash = taosHashInit(ctgTestVgNum, taosGetDefaultHashFunction(TSDB_DATA_TYPE_INT), true, HASH_ENTRY_LOCK);
vgNum = ctgTestGetVgNumFromVgVersion(dbVgroup->vgVersion); vgNum = ctgTestGetVgNumFromVgVersion(dbVgroup->vgVersion);
@ -209,6 +210,45 @@ void ctgTestBuildDBVgroup(SDBVgroupInfo **pdbVgroup) {
*pdbVgroup = dbVgroup; *pdbVgroup = dbVgroup;
} }
void ctgTestBuildSTableMetaRsp(STableMetaRsp *rspMsg) {
strcpy(rspMsg->dbFName, ctgTestDbname);
sprintf(rspMsg->tbName, "%s", ctgTestSTablename);
sprintf(rspMsg->stbName, "%s", ctgTestSTablename);
rspMsg->numOfTags = ctgTestTagNum;
rspMsg->numOfColumns = ctgTestColNum;
rspMsg->precision = 1 + 1;
rspMsg->tableType = TSDB_SUPER_TABLE;
rspMsg->update = 1 + 1;
rspMsg->sversion = ctgTestSVersion + 1;
rspMsg->tversion = ctgTestTVersion + 1;
rspMsg->suid = ctgTestSuid + 1;
rspMsg->tuid = ctgTestSuid + 1;
rspMsg->vgId = 1;
SSchema *s = NULL;
s = &rspMsg->pSchema[0];
s->type = TSDB_DATA_TYPE_TIMESTAMP;
s->colId = 1;
s->bytes = 8;
strcpy(s->name, "ts");
s = &rspMsg->pSchema[1];
s->type = TSDB_DATA_TYPE_INT;
s->colId = 2;
s->bytes = 4;
strcpy(s->name, "col1s");
s = &rspMsg->pSchema[2];
s->type = TSDB_DATA_TYPE_BINARY;
s->colId = 3;
s->bytes = 12 + 1;
strcpy(s->name, "tag1s");
return;
}
void ctgTestPrepareDbVgroups(void *shandle, SEpSet *pEpSet, SRpcMsg *pMsg, SRpcMsg *pRsp) { void ctgTestPrepareDbVgroups(void *shandle, SEpSet *pEpSet, SRpcMsg *pMsg, SRpcMsg *pRsp) {
SUseDbRsp *rspMsg = NULL; // todo SUseDbRsp *rspMsg = NULL; // todo
@ -592,7 +632,7 @@ void *ctgTestGetDbVgroupThread(void *param) {
return NULL; return NULL;
} }
void *ctgTestSetDbVgroupThread(void *param) { void *ctgTestSetSameDbVgroupThread(void *param) {
struct SCatalog *pCtg = (struct SCatalog *)param; struct SCatalog *pCtg = (struct SCatalog *)param;
int32_t code = 0; int32_t code = 0;
SDBVgroupInfo *dbVgroup = NULL; SDBVgroupInfo *dbVgroup = NULL;
@ -600,7 +640,7 @@ void *ctgTestSetDbVgroupThread(void *param) {
while (!ctgTestStop) { while (!ctgTestStop) {
ctgTestBuildDBVgroup(&dbVgroup); ctgTestBuildDBVgroup(&dbVgroup);
code = catalogUpdateDBVgroup(pCtg, ctgTestDbname, dbVgroup); code = catalogUpdateDBVgroup(pCtg, ctgTestDbname, ctgTestDbId, dbVgroup);
if (code) { if (code) {
assert(0); assert(0);
} }
@ -616,6 +656,32 @@ void *ctgTestSetDbVgroupThread(void *param) {
return NULL; return NULL;
} }
void *ctgTestSetDiffDbVgroupThread(void *param) {
struct SCatalog *pCtg = (struct SCatalog *)param;
int32_t code = 0;
SDBVgroupInfo *dbVgroup = NULL;
int32_t n = 0;
while (!ctgTestStop) {
ctgTestBuildDBVgroup(&dbVgroup);
code = catalogUpdateDBVgroup(pCtg, ctgTestDbname, ctgTestDbId++, dbVgroup);
if (code) {
assert(0);
}
if (ctgTestEnableSleep) {
usleep(rand() % 5);
}
if (++n % ctgTestPrintNum == 0) {
printf("Set:%d\n", n);
}
}
return NULL;
}
void *ctgTestGetCtableMetaThread(void *param) { void *ctgTestGetCtableMetaThread(void *param) {
struct SCatalog *pCtg = (struct SCatalog *)param; struct SCatalog *pCtg = (struct SCatalog *)param;
int32_t code = 0; int32_t code = 0;
@ -681,6 +747,8 @@ TEST(tableMeta, normalTable) {
void *mockPointer = (void *)0x1; void *mockPointer = (void *)0x1;
SVgroupInfo vgInfo = {0}; SVgroupInfo vgInfo = {0};
ctgTestInitLogFile();
ctgTestSetPrepareDbVgroups(); ctgTestSetPrepareDbVgroups();
initQueryModuleMsgHandle(); initQueryModuleMsgHandle();
@ -771,6 +839,8 @@ TEST(tableMeta, childTableCase) {
void *mockPointer = (void *)0x1; void *mockPointer = (void *)0x1;
SVgroupInfo vgInfo = {0}; SVgroupInfo vgInfo = {0};
ctgTestInitLogFile();
ctgTestSetPrepareDbVgroupsAndChildMeta(); ctgTestSetPrepareDbVgroupsAndChildMeta();
initQueryModuleMsgHandle(); initQueryModuleMsgHandle();
@ -964,6 +1034,124 @@ TEST(tableMeta, superTableCase) {
catalogDestroy(); catalogDestroy();
} }
TEST(tableMeta, rmStbMeta) {
struct SCatalog *pCtg = NULL;
void *mockPointer = (void *)0x1;
SVgroupInfo vgInfo = {0};
ctgTestInitLogFile();
ctgTestSetPrepareDbVgroupsAndSuperMeta();
initQueryModuleMsgHandle();
int32_t code = catalogInit(NULL);
ASSERT_EQ(code, 0);
// sendCreateDbMsg(pConn->pTransporter, &pConn->pAppInfo->mgmtEp.epSet);
code = catalogGetHandle(ctgTestClusterId, &pCtg);
ASSERT_EQ(code, 0);
SName n = {.type = TSDB_TABLE_NAME_T, .acctId = 1};
strcpy(n.dbname, "db1");
strcpy(n.tname, ctgTestSTablename);
STableMeta *tableMeta = NULL;
code = catalogGetTableMeta(pCtg, mockPointer, (const SEpSet *)mockPointer, &n, &tableMeta);
ASSERT_EQ(code, 0);
ASSERT_EQ(tableMeta->vgId, 0);
ASSERT_EQ(tableMeta->tableType, TSDB_SUPER_TABLE);
ASSERT_EQ(tableMeta->sversion, ctgTestSVersion);
ASSERT_EQ(tableMeta->tversion, ctgTestTVersion);
ASSERT_EQ(tableMeta->uid, ctgTestSuid);
ASSERT_EQ(tableMeta->suid, ctgTestSuid);
ASSERT_EQ(tableMeta->tableInfo.numOfColumns, ctgTestColNum);
ASSERT_EQ(tableMeta->tableInfo.numOfTags, ctgTestTagNum);
ASSERT_EQ(tableMeta->tableInfo.precision, 1);
ASSERT_EQ(tableMeta->tableInfo.rowSize, 12);
code = catalogRemoveSTableMeta(pCtg, "1.db1", ctgTestSTablename, ctgTestSuid);
ASSERT_EQ(code, 0);
ASSERT_EQ(ctgDbgGetClusterCacheNum(pCtg, CTG_DBG_DB_NUM), 1);
ASSERT_EQ(ctgDbgGetClusterCacheNum(pCtg, CTG_DBG_META_NUM), 0);
ASSERT_EQ(ctgDbgGetClusterCacheNum(pCtg, CTG_DBG_STB_NUM), 0);
ASSERT_EQ(ctgDbgGetClusterCacheNum(pCtg, CTG_DBG_DB_RENT_NUM), 1);
ASSERT_EQ(ctgDbgGetClusterCacheNum(pCtg, CTG_DBG_STB_RENT_NUM), 0);
catalogDestroy();
}
TEST(tableMeta, updateStbMeta) {
struct SCatalog *pCtg = NULL;
void *mockPointer = (void *)0x1;
SVgroupInfo vgInfo = {0};
ctgTestInitLogFile();
ctgTestSetPrepareDbVgroupsAndSuperMeta();
initQueryModuleMsgHandle();
int32_t code = catalogInit(NULL);
ASSERT_EQ(code, 0);
// sendCreateDbMsg(pConn->pTransporter, &pConn->pAppInfo->mgmtEp.epSet);
code = catalogGetHandle(ctgTestClusterId, &pCtg);
ASSERT_EQ(code, 0);
SName n = {.type = TSDB_TABLE_NAME_T, .acctId = 1};
strcpy(n.dbname, "db1");
strcpy(n.tname, ctgTestSTablename);
STableMeta *tableMeta = NULL;
code = catalogGetTableMeta(pCtg, mockPointer, (const SEpSet *)mockPointer, &n, &tableMeta);
ASSERT_EQ(code, 0);
ASSERT_EQ(tableMeta->vgId, 0);
ASSERT_EQ(tableMeta->tableType, TSDB_SUPER_TABLE);
ASSERT_EQ(tableMeta->sversion, ctgTestSVersion);
ASSERT_EQ(tableMeta->tversion, ctgTestTVersion);
ASSERT_EQ(tableMeta->uid, ctgTestSuid);
ASSERT_EQ(tableMeta->suid, ctgTestSuid);
ASSERT_EQ(tableMeta->tableInfo.numOfColumns, ctgTestColNum);
ASSERT_EQ(tableMeta->tableInfo.numOfTags, ctgTestTagNum);
ASSERT_EQ(tableMeta->tableInfo.precision, 1);
ASSERT_EQ(tableMeta->tableInfo.rowSize, 12);
tfree(tableMeta);
STableMetaRsp rsp = {0};
ctgTestBuildSTableMetaRsp(&rsp);
code = catalogUpdateSTableMeta(pCtg, &rsp);
ASSERT_EQ(code, 0);
ASSERT_EQ(ctgDbgGetClusterCacheNum(pCtg, CTG_DBG_DB_NUM), 1);
ASSERT_EQ(ctgDbgGetClusterCacheNum(pCtg, CTG_DBG_META_NUM), 1);
ASSERT_EQ(ctgDbgGetClusterCacheNum(pCtg, CTG_DBG_STB_NUM), 1);
ASSERT_EQ(ctgDbgGetClusterCacheNum(pCtg, CTG_DBG_DB_RENT_NUM), 1);
ASSERT_EQ(ctgDbgGetClusterCacheNum(pCtg, CTG_DBG_STB_RENT_NUM), 1);
code = catalogGetTableMeta(pCtg, mockPointer, (const SEpSet *)mockPointer, &n, &tableMeta);
ASSERT_EQ(code, 0);
ASSERT_EQ(tableMeta->vgId, 0);
ASSERT_EQ(tableMeta->tableType, TSDB_SUPER_TABLE);
ASSERT_EQ(tableMeta->sversion, ctgTestSVersion + 1);
ASSERT_EQ(tableMeta->tversion, ctgTestTVersion + 1);
ASSERT_EQ(tableMeta->uid, ctgTestSuid + 1);
ASSERT_EQ(tableMeta->suid, ctgTestSuid + 1);
ASSERT_EQ(tableMeta->tableInfo.numOfColumns, ctgTestColNum);
ASSERT_EQ(tableMeta->tableInfo.numOfTags, ctgTestTagNum);
ASSERT_EQ(tableMeta->tableInfo.precision, 1 + 1);
ASSERT_EQ(tableMeta->tableInfo.rowSize, 12);
tfree(tableMeta);
catalogDestroy();
}
TEST(tableDistVgroup, normalTable) { TEST(tableDistVgroup, normalTable) {
struct SCatalog *pCtg = NULL; struct SCatalog *pCtg = NULL;
void *mockPointer = (void *)0x1; void *mockPointer = (void *)0x1;
@ -1109,7 +1297,7 @@ TEST(dbVgroup, getSetDbVgroupCase) {
taosArrayDestroy(vgList); taosArrayDestroy(vgList);
ctgTestBuildDBVgroup(&dbVgroup); ctgTestBuildDBVgroup(&dbVgroup);
code = catalogUpdateDBVgroup(pCtg, ctgTestDbname, dbVgroup); code = catalogUpdateDBVgroup(pCtg, ctgTestDbname, ctgTestDbId, dbVgroup);
ASSERT_EQ(code, 0); ASSERT_EQ(code, 0);
code = catalogGetTableHashVgroup(pCtg, mockPointer, (const SEpSet *)mockPointer, &n, &vgInfo); code = catalogGetTableHashVgroup(pCtg, mockPointer, (const SEpSet *)mockPointer, &n, &vgInfo);
@ -1128,7 +1316,7 @@ TEST(dbVgroup, getSetDbVgroupCase) {
catalogDestroy(); catalogDestroy();
} }
TEST(multiThread, getSetDbVgroupCase) { TEST(multiThread, getSetRmSameDbVgroup) {
struct SCatalog *pCtg = NULL; struct SCatalog *pCtg = NULL;
void *mockPointer = (void *)0x1; void *mockPointer = (void *)0x1;
SVgroupInfo vgInfo = {0}; SVgroupInfo vgInfo = {0};
@ -1159,10 +1347,10 @@ TEST(multiThread, getSetDbVgroupCase) {
pthread_attr_init(&thattr); pthread_attr_init(&thattr);
pthread_t thread1, thread2; pthread_t thread1, thread2;
pthread_create(&(thread1), &thattr, ctgTestSetDbVgroupThread, pCtg); pthread_create(&(thread1), &thattr, ctgTestSetSameDbVgroupThread, pCtg);
sleep(1); sleep(1);
pthread_create(&(thread1), &thattr, ctgTestGetDbVgroupThread, pCtg); pthread_create(&(thread2), &thattr, ctgTestGetDbVgroupThread, pCtg);
while (true) { while (true) {
if (ctgTestDeadLoop) { if (ctgTestDeadLoop) {
@ -1179,6 +1367,58 @@ TEST(multiThread, getSetDbVgroupCase) {
catalogDestroy(); catalogDestroy();
} }
TEST(multiThread, getSetRmDiffDbVgroup) {
struct SCatalog *pCtg = NULL;
void *mockPointer = (void *)0x1;
SVgroupInfo vgInfo = {0};
SVgroupInfo *pvgInfo = NULL;
SDBVgroupInfo dbVgroup = {0};
SArray *vgList = NULL;
ctgTestStop = false;
ctgTestInitLogFile();
ctgTestSetPrepareDbVgroups();
initQueryModuleMsgHandle();
// sendCreateDbMsg(pConn->pTransporter, &pConn->pAppInfo->mgmtEp.epSet);
int32_t code = catalogInit(NULL);
ASSERT_EQ(code, 0);
code = catalogGetHandle(ctgTestClusterId, &pCtg);
ASSERT_EQ(code, 0);
SName n = {.type = TSDB_TABLE_NAME_T, .acctId = 1};
strcpy(n.dbname, "db1");
strcpy(n.tname, ctgTestTablename);
pthread_attr_t thattr;
pthread_attr_init(&thattr);
pthread_t thread1, thread2;
pthread_create(&(thread1), &thattr, ctgTestSetDiffDbVgroupThread, pCtg);
sleep(1);
pthread_create(&(thread2), &thattr, ctgTestGetDbVgroupThread, pCtg);
while (true) {
if (ctgTestDeadLoop) {
sleep(1);
} else {
sleep(ctgTestMTRunSec);
break;
}
}
ctgTestStop = true;
sleep(1);
catalogDestroy();
}
TEST(multiThread, ctableMeta) { TEST(multiThread, ctableMeta) {
struct SCatalog *pCtg = NULL; struct SCatalog *pCtg = NULL;

1
source/libs/parser/inc/astCreateContext.h
View File

@ -28,7 +28,6 @@ typedef struct SAstCreateContext {
bool notSupport; bool notSupport;
bool valid; bool valid;
SNode* pRootNode; SNode* pRootNode;
SHashObj* pResourceHash;
} SAstCreateContext; } SAstCreateContext;
int32_t createAstCreateContext(SParseContext* pQueryCxt, SAstCreateContext* pCxt); int32_t createAstCreateContext(SParseContext* pQueryCxt, SAstCreateContext* pCxt);

16
source/libs/parser/inc/astCreateFuncs.h
View File

@ -13,11 +13,6 @@
* along with this program. If not, see <http://www.gnu.org/licenses/>. * along with this program. If not, see <http://www.gnu.org/licenses/>.
*/ */
#include "nodes.h"
#include "nodesShowStmts.h"
#include "astCreateContext.h"
#include "ttoken.h"
#ifndef _TD_AST_CREATE_FUNCS_H_ #ifndef _TD_AST_CREATE_FUNCS_H_
#define _TD_AST_CREATE_FUNCS_H_ #define _TD_AST_CREATE_FUNCS_H_
@ -25,15 +20,24 @@
extern "C" { extern "C" {
#endif #endif
#include "nodes.h"
#include "nodesShowStmts.h"
#include "astCreateContext.h"
#include "ttoken.h"
extern SToken nil_token; extern SToken nil_token;
SNode* createRawExprNode(SAstCreateContext* pCxt, const SToken* pToken, SNode* pNode);
SNode* createRawExprNodeExt(SAstCreateContext* pCxt, const SToken* pStart, const SToken* pEnd, SNode* pNode);
SNode* releaseRawExprNode(SAstCreateContext* pCxt, SNode* pNode);
SToken getTokenFromRawExprNode(SAstCreateContext* pCxt, SNode* pNode);
SNodeList* createNodeList(SAstCreateContext* pCxt, SNode* pNode); SNodeList* createNodeList(SAstCreateContext* pCxt, SNode* pNode);
SNodeList* addNodeToList(SAstCreateContext* pCxt, SNodeList* pList, SNode* pNode); SNodeList* addNodeToList(SAstCreateContext* pCxt, SNodeList* pList, SNode* pNode);
SNode* createColumnNode(SAstCreateContext* pCxt, const SToken* pTableAlias, const SToken* pColumnName); SNode* createColumnNode(SAstCreateContext* pCxt, const SToken* pTableAlias, const SToken* pColumnName);
SNode* createValueNode(SAstCreateContext* pCxt, int32_t dataType, const SToken* pLiteral); SNode* createValueNode(SAstCreateContext* pCxt, int32_t dataType, const SToken* pLiteral);
SNode* createDurationValueNode(SAstCreateContext* pCxt, const SToken* pLiteral); SNode* createDurationValueNode(SAstCreateContext* pCxt, const SToken* pLiteral);
SNode* addMinusSign(SAstCreateContext* pCxt, SNode* pNode);
SNode* setProjectionAlias(SAstCreateContext* pCxt, SNode* pNode, const SToken* pAlias); SNode* setProjectionAlias(SAstCreateContext* pCxt, SNode* pNode, const SToken* pAlias);
SNode* createLogicConditionNode(SAstCreateContext* pCxt, ELogicConditionType type, SNode* pParam1, SNode* pParam2); SNode* createLogicConditionNode(SAstCreateContext* pCxt, ELogicConditionType type, SNode* pParam1, SNode* pParam2);
SNode* createOperatorNode(SAstCreateContext* pCxt, EOperatorType type, SNode* pLeft, SNode* pRight); SNode* createOperatorNode(SAstCreateContext* pCxt, EOperatorType type, SNode* pLeft, SNode* pRight);

107
source/libs/parser/inc/new_sql.y
View File

@ -67,19 +67,19 @@ cmd ::= SHOW DATABASES.
cmd ::= query_expression(A). { PARSER_TRACE; pCxt->pRootNode = A; } cmd ::= query_expression(A). { PARSER_TRACE; pCxt->pRootNode = A; }
/************************************************ literal *************************************************************/ /************************************************ literal *************************************************************/
literal(A) ::= NK_INTEGER(B). { PARSER_TRACE; A = createValueNode(pCxt, TSDB_DATA_TYPE_BIGINT, &B); } literal(A) ::= NK_INTEGER(B). { PARSER_TRACE; A = createRawExprNode(pCxt, &B, createValueNode(pCxt, TSDB_DATA_TYPE_BIGINT, &B)); }
literal(A) ::= NK_FLOAT(B). { PARSER_TRACE; A = createValueNode(pCxt, TSDB_DATA_TYPE_DOUBLE, &B); } literal(A) ::= NK_FLOAT(B). { PARSER_TRACE; A = createRawExprNode(pCxt, &B, createValueNode(pCxt, TSDB_DATA_TYPE_DOUBLE, &B)); }
literal(A) ::= NK_STRING(B). { PARSER_TRACE; A = createValueNode(pCxt, TSDB_DATA_TYPE_BINARY, &B); } literal(A) ::= NK_STRING(B). { PARSER_TRACE; A = createRawExprNode(pCxt, &B, createValueNode(pCxt, TSDB_DATA_TYPE_BINARY, &B)); }
literal(A) ::= NK_BOOL(B). { PARSER_TRACE; A = createValueNode(pCxt, TSDB_DATA_TYPE_BOOL, &B); } literal(A) ::= NK_BOOL(B). { PARSER_TRACE; A = createRawExprNode(pCxt, &B, createValueNode(pCxt, TSDB_DATA_TYPE_BOOL, &B)); }
literal(A) ::= TIMESTAMP NK_STRING(B). { PARSER_TRACE; A = createValueNode(pCxt, TSDB_DATA_TYPE_TIMESTAMP, &B); } literal(A) ::= TIMESTAMP(B) NK_STRING(C). { PARSER_TRACE; A = createRawExprNodeExt(pCxt, &B, &C, createValueNode(pCxt, TSDB_DATA_TYPE_TIMESTAMP, &C)); }
literal(A) ::= duration_literal(B). { PARSER_TRACE; A = B; } literal(A) ::= duration_literal(B). { PARSER_TRACE; A = B; }
duration_literal(A) ::= NK_VARIABLE(B). { PARSER_TRACE; A = createDurationValueNode(pCxt, &B); } duration_literal(A) ::= NK_VARIABLE(B). { PARSER_TRACE; A = createRawExprNode(pCxt, &B, createDurationValueNode(pCxt, &B)); }
%type literal_list { SNodeList* } %type literal_list { SNodeList* }
%destructor literal_list { PARSER_DESTRUCTOR_TRACE; nodesDestroyList($$); } %destructor literal_list { PARSER_DESTRUCTOR_TRACE; nodesDestroyList($$); }
literal_list(A) ::= literal(B). { PARSER_TRACE; A = createNodeList(pCxt, B); } literal_list(A) ::= literal(B). { PARSER_TRACE; A = createNodeList(pCxt, releaseRawExprNode(pCxt, B)); }
literal_list(A) ::= literal_list(B) NK_COMMA literal(C). { PARSER_TRACE; A = addNodeToList(pCxt, B, C); } literal_list(A) ::= literal_list(B) NK_COMMA literal(C). { PARSER_TRACE; A = addNodeToList(pCxt, B, releaseRawExprNode(pCxt, C)); }
/************************************************ names and identifiers ***********************************************/ /************************************************ names and identifiers ***********************************************/
%type db_name { SToken } %type db_name { SToken }
@ -111,37 +111,70 @@ expression(A) ::= literal(B).
//expression(A) ::= NK_QUESTION(B). { PARSER_TRACE; A = B; } //expression(A) ::= NK_QUESTION(B). { PARSER_TRACE; A = B; }
//expression(A) ::= pseudo_column(B). { PARSER_TRACE; A = B; } //expression(A) ::= pseudo_column(B). { PARSER_TRACE; A = B; }
expression(A) ::= column_reference(B). { PARSER_TRACE; A = B; } expression(A) ::= column_reference(B). { PARSER_TRACE; A = B; }
expression(A) ::= function_name(B) NK_LP expression_list(C) NK_RP. { PARSER_TRACE; A = createFunctionNode(pCxt, &B, C); } expression(A) ::= function_name(B) NK_LP expression_list(C) NK_RP(D). { PARSER_TRACE; A = createRawExprNodeExt(pCxt, &B, &D, createFunctionNode(pCxt, &B, C)); }
//expression(A) ::= cast_expression(B). { PARSER_TRACE; A = B; } //expression(A) ::= cast_expression(B). { PARSER_TRACE; A = B; }
//expression(A) ::= case_expression(B). { PARSER_TRACE; A = B; } //expression(A) ::= case_expression(B). { PARSER_TRACE; A = B; }
expression(A) ::= subquery(B). { PARSER_TRACE; A = B; } expression(A) ::= subquery(B). { PARSER_TRACE; A = B; }
expression(A) ::= NK_LP expression(B) NK_RP. { PARSER_TRACE; A = B; } expression(A) ::= NK_LP(B) expression(C) NK_RP(D). { PARSER_TRACE; A = createRawExprNodeExt(pCxt, &B, &D, releaseRawExprNode(pCxt, C)); }
expression(A) ::= NK_PLUS expression(B). { PARSER_TRACE; A = B; } expression(A) ::= NK_PLUS(B) expression(C). {
expression(A) ::= NK_MINUS expression(B). { PARSER_TRACE; A = createOperatorNode(pCxt, OP_TYPE_SUB, B, NULL); } PARSER_TRACE;
expression(A) ::= expression(B) NK_PLUS expression(C). { PARSER_TRACE; A = createOperatorNode(pCxt, OP_TYPE_ADD, B, C); } SToken t = getTokenFromRawExprNode(pCxt, C);
expression(A) ::= expression(B) NK_MINUS expression(C). { PARSER_TRACE; A = createOperatorNode(pCxt, OP_TYPE_SUB, B, C); } A = createRawExprNodeExt(pCxt, &B, &t, releaseRawExprNode(pCxt, C));
expression(A) ::= expression(B) NK_STAR expression(C). { PARSER_TRACE; A = createOperatorNode(pCxt, OP_TYPE_MULTI, B, C); } }
expression(A) ::= expression(B) NK_SLASH expression(C). { PARSER_TRACE; A = createOperatorNode(pCxt, OP_TYPE_DIV, B, C); } expression(A) ::= NK_MINUS(B) expression(C). {
expression(A) ::= expression(B) NK_REM expression(C). { PARSER_TRACE; A = createOperatorNode(pCxt, OP_TYPE_MOD, B, C); } PARSER_TRACE;
SToken t = getTokenFromRawExprNode(pCxt, C);
A = createRawExprNodeExt(pCxt, &B, &t, createOperatorNode(pCxt, OP_TYPE_SUB, releaseRawExprNode(pCxt, C), NULL));
}
expression(A) ::= expression(B) NK_PLUS expression(C). {
PARSER_TRACE;
SToken s = getTokenFromRawExprNode(pCxt, B);
SToken e = getTokenFromRawExprNode(pCxt, C);
A = createRawExprNodeExt(pCxt, &s, &e, createOperatorNode(pCxt, OP_TYPE_ADD, releaseRawExprNode(pCxt, B), releaseRawExprNode(pCxt, C)));
}
expression(A) ::= expression(B) NK_MINUS expression(C). {
PARSER_TRACE;
SToken s = getTokenFromRawExprNode(pCxt, B);
SToken e = getTokenFromRawExprNode(pCxt, C);
A = createRawExprNodeExt(pCxt, &s, &e, createOperatorNode(pCxt, OP_TYPE_SUB, releaseRawExprNode(pCxt, B), releaseRawExprNode(pCxt, C)));
}
expression(A) ::= expression(B) NK_STAR expression(C). {
PARSER_TRACE;
SToken s = getTokenFromRawExprNode(pCxt, B);
SToken e = getTokenFromRawExprNode(pCxt, C);
A = createRawExprNodeExt(pCxt, &s, &e, createOperatorNode(pCxt, OP_TYPE_MULTI, releaseRawExprNode(pCxt, B), releaseRawExprNode(pCxt, C)));
}
expression(A) ::= expression(B) NK_SLASH expression(C). {
PARSER_TRACE;
SToken s = getTokenFromRawExprNode(pCxt, B);
SToken e = getTokenFromRawExprNode(pCxt, C);
A = createRawExprNodeExt(pCxt, &s, &e, createOperatorNode(pCxt, OP_TYPE_DIV, releaseRawExprNode(pCxt, B), releaseRawExprNode(pCxt, C)));
}
expression(A) ::= expression(B) NK_REM expression(C). {
PARSER_TRACE;
SToken s = getTokenFromRawExprNode(pCxt, B);
SToken e = getTokenFromRawExprNode(pCxt, C);
A = createRawExprNodeExt(pCxt, &s, &e, createOperatorNode(pCxt, OP_TYPE_MOD, releaseRawExprNode(pCxt, B), releaseRawExprNode(pCxt, C)));
}
%type expression_list { SNodeList* } %type expression_list { SNodeList* }
%destructor expression_list { PARSER_DESTRUCTOR_TRACE; nodesDestroyList($$); } %destructor expression_list { PARSER_DESTRUCTOR_TRACE; nodesDestroyList($$); }
expression_list(A) ::= expression(B). { PARSER_TRACE; A = createNodeList(pCxt, B); } expression_list(A) ::= expression(B). { PARSER_TRACE; A = createNodeList(pCxt, releaseRawExprNode(pCxt, B)); }
expression_list(A) ::= expression_list(B) NK_COMMA expression(C). { PARSER_TRACE; A = addNodeToList(pCxt, B, C); } expression_list(A) ::= expression_list(B) NK_COMMA expression(C). { PARSER_TRACE; A = addNodeToList(pCxt, B, releaseRawExprNode(pCxt, C)); }
column_reference(A) ::= column_name(B). { PARSER_TRACE; A = createColumnNode(pCxt, NULL, &B); } column_reference(A) ::= column_name(B). { PARSER_TRACE; A = createRawExprNode(pCxt, &B, createColumnNode(pCxt, NULL, &B)); }
column_reference(A) ::= table_name(B) NK_DOT column_name(C). { PARSER_TRACE; A = createColumnNode(pCxt, &B, &C); } column_reference(A) ::= table_name(B) NK_DOT column_name(C). { PARSER_TRACE; A = createRawExprNodeExt(pCxt, &B, &C, createColumnNode(pCxt, &B, &C)); }
//pseudo_column(A) ::= NK_NOW. { PARSER_TRACE; A = createFunctionNode(pCxt, NULL, NULL); } //pseudo_column(A) ::= NK_NOW. { PARSER_TRACE; A = createFunctionNode(pCxt, NULL, NULL); }
/************************************************ predicate ***********************************************************/ /************************************************ predicate ***********************************************************/
predicate(A) ::= expression(B) compare_op(C) expression(D). { PARSER_TRACE; A = createOperatorNode(pCxt, C, B, D); } predicate(A) ::= expression(B) compare_op(C) expression(D). { PARSER_TRACE; A = createOperatorNode(pCxt, C, releaseRawExprNode(pCxt, B), releaseRawExprNode(pCxt, D)); }
//predicate(A) ::= expression(B) compare_op sub_type expression(B). //predicate(A) ::= expression(B) compare_op sub_type expression(B).
predicate(A) ::= expression(B) BETWEEN expression(C) AND expression(D). { PARSER_TRACE; A = createBetweenAnd(pCxt, B, C, D); } predicate(A) ::= expression(B) BETWEEN expression(C) AND expression(D). { PARSER_TRACE; A = createBetweenAnd(pCxt, releaseRawExprNode(pCxt, B), releaseRawExprNode(pCxt, C), releaseRawExprNode(pCxt, D)); }
predicate(A) ::= expression(B) NOT BETWEEN expression(C) AND expression(D). { PARSER_TRACE; A = createNotBetweenAnd(pCxt, C, B, D); } predicate(A) ::= expression(B) NOT BETWEEN expression(C) AND expression(D). { PARSER_TRACE; A = createNotBetweenAnd(pCxt, releaseRawExprNode(pCxt, C), releaseRawExprNode(pCxt, B), releaseRawExprNode(pCxt, D)); }
predicate(A) ::= expression(B) IS NULL. { PARSER_TRACE; A = createIsNullCondNode(pCxt, B, true); } predicate(A) ::= expression(B) IS NULL. { PARSER_TRACE; A = createIsNullCondNode(pCxt, releaseRawExprNode(pCxt, B), true); }
predicate(A) ::= expression(B) IS NOT NULL. { PARSER_TRACE; A = createIsNullCondNode(pCxt, B, false); } predicate(A) ::= expression(B) IS NOT NULL. { PARSER_TRACE; A = createIsNullCondNode(pCxt, releaseRawExprNode(pCxt, B), false); }
predicate(A) ::= expression(B) in_op(C) in_predicate_value(D). { PARSER_TRACE; A = createOperatorNode(pCxt, C, B, D); } predicate(A) ::= expression(B) in_op(C) in_predicate_value(D). { PARSER_TRACE; A = createOperatorNode(pCxt, C, releaseRawExprNode(pCxt, B), D); }
%type compare_op { EOperatorType } %type compare_op { EOperatorType }
%destructor compare_op { PARSER_DESTRUCTOR_TRACE; } %destructor compare_op { PARSER_DESTRUCTOR_TRACE; }
@ -186,7 +219,7 @@ table_reference(A) ::= joined_table(B).
table_primary(A) ::= table_name(B) alias_opt(C). { PARSER_TRACE; A = createRealTableNode(pCxt, NULL, &B, &C); } table_primary(A) ::= table_name(B) alias_opt(C). { PARSER_TRACE; A = createRealTableNode(pCxt, NULL, &B, &C); }
table_primary(A) ::= db_name(B) NK_DOT table_name(C) alias_opt(D). { PARSER_TRACE; A = createRealTableNode(pCxt, &B, &C, &D); } table_primary(A) ::= db_name(B) NK_DOT table_name(C) alias_opt(D). { PARSER_TRACE; A = createRealTableNode(pCxt, &B, &C, &D); }
table_primary(A) ::= subquery(B) alias_opt(C). { PARSER_TRACE; A = createTempTableNode(pCxt, B, &C); } table_primary(A) ::= subquery(B) alias_opt(C). { PARSER_TRACE; A = createTempTableNode(pCxt, releaseRawExprNode(pCxt, B), &C); }
table_primary(A) ::= parenthesized_joined_table(B). { PARSER_TRACE; A = B; } table_primary(A) ::= parenthesized_joined_table(B). { PARSER_TRACE; A = B; }
%type alias_opt { SToken } %type alias_opt { SToken }
@ -236,9 +269,13 @@ select_list(A) ::= select_sublist(B).
select_sublist(A) ::= select_item(B). { PARSER_TRACE; A = createNodeList(pCxt, B); } select_sublist(A) ::= select_item(B). { PARSER_TRACE; A = createNodeList(pCxt, B); }
select_sublist(A) ::= select_sublist(B) NK_COMMA select_item(C). { PARSER_TRACE; A = addNodeToList(pCxt, B, C); } select_sublist(A) ::= select_sublist(B) NK_COMMA select_item(C). { PARSER_TRACE; A = addNodeToList(pCxt, B, C); }
select_item(A) ::= expression(B). { PARSER_TRACE; A = B; } select_item(A) ::= expression(B). {
select_item(A) ::= expression(B) column_alias(C). { PARSER_TRACE; A = setProjectionAlias(pCxt, B, &C); } PARSER_TRACE;
select_item(A) ::= expression(B) AS column_alias(C). { PARSER_TRACE; A = setProjectionAlias(pCxt, B, &C); } SToken t = getTokenFromRawExprNode(pCxt, B);
A = setProjectionAlias(pCxt, releaseRawExprNode(pCxt, B), &t);
}
select_item(A) ::= expression(B) column_alias(C). { PARSER_TRACE; A = setProjectionAlias(pCxt, releaseRawExprNode(pCxt, B), &C); }
select_item(A) ::= expression(B) AS column_alias(C). { PARSER_TRACE; A = setProjectionAlias(pCxt, releaseRawExprNode(pCxt, B), &C); }
select_item(A) ::= table_name(B) NK_DOT NK_STAR(C). { PARSER_TRACE; A = createColumnNode(pCxt, &B, &C); } select_item(A) ::= table_name(B) NK_DOT NK_STAR(C). { PARSER_TRACE; A = createColumnNode(pCxt, &B, &C); }
where_clause_opt(A) ::= . { PARSER_TRACE; A = NULL; } where_clause_opt(A) ::= . { PARSER_TRACE; A = NULL; }
@ -251,8 +288,8 @@ partition_by_clause_opt(A) ::= PARTITION BY expression_list(B).
twindow_clause_opt(A) ::= . { PARSER_TRACE; A = NULL; } twindow_clause_opt(A) ::= . { PARSER_TRACE; A = NULL; }
twindow_clause_opt(A) ::= twindow_clause_opt(A) ::=
SESSION NK_LP column_reference(B) NK_COMMA NK_INTEGER(C) NK_RP. { PARSER_TRACE; A = createSessionWindowNode(pCxt, B, &C); } SESSION NK_LP column_reference(B) NK_COMMA NK_INTEGER(C) NK_RP. { PARSER_TRACE; A = createSessionWindowNode(pCxt, releaseRawExprNode(pCxt, B), &C); }
twindow_clause_opt(A) ::= STATE_WINDOW NK_LP column_reference(B) NK_RP. { PARSER_TRACE; A = createStateWindowNode(pCxt, B); } twindow_clause_opt(A) ::= STATE_WINDOW NK_LP column_reference(B) NK_RP. { PARSER_TRACE; A = createStateWindowNode(pCxt, releaseRawExprNode(pCxt, B)); }
twindow_clause_opt(A) ::= twindow_clause_opt(A) ::=
INTERVAL NK_LP duration_literal(B) NK_RP sliding_opt(C) fill_opt(D). { PARSER_TRACE; A = createIntervalWindowNode(pCxt, B, NULL, C, D); } INTERVAL NK_LP duration_literal(B) NK_RP sliding_opt(C) fill_opt(D). { PARSER_TRACE; A = createIntervalWindowNode(pCxt, B, NULL, C, D); }
twindow_clause_opt(A) ::= twindow_clause_opt(A) ::=
@ -317,7 +354,7 @@ limit_clause_opt(A) ::= LIMIT NK_INTEGER(B) OFFSET NK_INTEGER(C).
limit_clause_opt(A) ::= LIMIT NK_INTEGER(C) NK_COMMA NK_INTEGER(B). { PARSER_TRACE; A = createLimitNode(pCxt, &B, &C); } limit_clause_opt(A) ::= LIMIT NK_INTEGER(C) NK_COMMA NK_INTEGER(B). { PARSER_TRACE; A = createLimitNode(pCxt, &B, &C); }
/************************************************ subquery ************************************************************/ /************************************************ subquery ************************************************************/
subquery(A) ::= NK_LP query_expression(B) NK_RP. { PARSER_TRACE; A = B; } subquery(A) ::= NK_LP(B) query_expression(C) NK_RP(D). { PARSER_TRACE; A = createRawExprNodeExt(pCxt, &B, &D, C); }
/************************************************ search_condition ****************************************************/ /************************************************ search_condition ****************************************************/
search_condition(A) ::= boolean_value_expression(B). { PARSER_TRACE; A = B; } search_condition(A) ::= boolean_value_expression(B). { PARSER_TRACE; A = B; }
@ -330,7 +367,7 @@ sort_specification_list(A) ::=
sort_specification_list(B) NK_COMMA sort_specification(C). { PARSER_TRACE; A = addNodeToList(pCxt, B, C); } sort_specification_list(B) NK_COMMA sort_specification(C). { PARSER_TRACE; A = addNodeToList(pCxt, B, C); }
sort_specification(A) ::= sort_specification(A) ::=
expression(B) ordering_specification_opt(C) null_ordering_opt(D). { PARSER_TRACE; A = createOrderByExprNode(pCxt, B, C, D); } expression(B) ordering_specification_opt(C) null_ordering_opt(D). { PARSER_TRACE; A = createOrderByExprNode(pCxt, releaseRawExprNode(pCxt, B), C, D); }
%type ordering_specification_opt EOrder %type ordering_specification_opt EOrder
%destructor ordering_specification_opt { PARSER_DESTRUCTOR_TRACE; } %destructor ordering_specification_opt { PARSER_DESTRUCTOR_TRACE; }

50
source/libs/parser/src/astCreateFuncs.c
View File

@ -24,6 +24,14 @@
} \ } \
} while (0) } while (0)
#define CHECK_RAW_EXPR_NODE(node) \
do { \
if (NULL == (node) || QUERY_NODE_RAW_EXPR != nodeType(node)) { \
pCxt->valid = false; \
return NULL; \
} \
} while (0)
SToken nil_token = { .type = TK_NIL, .n = 0, .z = NULL }; SToken nil_token = { .type = TK_NIL, .n = 0, .z = NULL };
static bool checkDbName(SAstCreateContext* pCxt, const SToken* pDbName) { static bool checkDbName(SAstCreateContext* pCxt, const SToken* pDbName) {
@ -50,6 +58,37 @@ static bool checkColumnName(SAstCreateContext* pCxt, const SToken* pColumnName)
return pCxt->valid; return pCxt->valid;
} }
SNode* createRawExprNode(SAstCreateContext* pCxt, const SToken* pToken, SNode* pNode) {
SRawExprNode* target = (SRawExprNode*)nodesMakeNode(QUERY_NODE_RAW_EXPR);
CHECK_OUT_OF_MEM(target);
target->p = pToken->z;
target->n = pToken->n;
target->pNode = pNode;
return (SNode*)target;
}
SNode* createRawExprNodeExt(SAstCreateContext* pCxt, const SToken* pStart, const SToken* pEnd, SNode* pNode) {
SRawExprNode* target = (SRawExprNode*)nodesMakeNode(QUERY_NODE_RAW_EXPR);
CHECK_OUT_OF_MEM(target);
target->p = pStart->z;
target->n = (pEnd->z + pEnd->n) - pStart->z;
target->pNode = pNode;
return (SNode*)target;
}
SNode* releaseRawExprNode(SAstCreateContext* pCxt, SNode* pNode) {
CHECK_RAW_EXPR_NODE(pNode);
SNode* tmp = ((SRawExprNode*)pNode)->pNode;
tfree(pNode);
return tmp;
}
SToken getTokenFromRawExprNode(SAstCreateContext* pCxt, SNode* pNode) {
SRawExprNode* target = (SRawExprNode*)pNode;
SToken t = { .type = 0, .z = target->p, .n = target->n};
return t;
}
SNodeList* createNodeList(SAstCreateContext* pCxt, SNode* pNode) { SNodeList* createNodeList(SAstCreateContext* pCxt, SNode* pNode) {
SNodeList* list = nodesMakeList(); SNodeList* list = nodesMakeList();
CHECK_OUT_OF_MEM(list); CHECK_OUT_OF_MEM(list);
@ -76,21 +115,20 @@ SNode* createColumnNode(SAstCreateContext* pCxt, const SToken* pTableAlias, cons
SNode* createValueNode(SAstCreateContext* pCxt, int32_t dataType, const SToken* pLiteral) { SNode* createValueNode(SAstCreateContext* pCxt, int32_t dataType, const SToken* pLiteral) {
SValueNode* val = (SValueNode*)nodesMakeNode(QUERY_NODE_VALUE); SValueNode* val = (SValueNode*)nodesMakeNode(QUERY_NODE_VALUE);
CHECK_OUT_OF_MEM(val); CHECK_OUT_OF_MEM(val);
// todo val->literal = strndup(pLiteral->z, pLiteral->n);
CHECK_OUT_OF_MEM(val->literal);
val->node.resType.type = dataType;
val->node.resType.bytes = tDataTypes[TSDB_DATA_TYPE_BOOL].bytes;
return (SNode*)val; return (SNode*)val;
} }
SNode* createDurationValueNode(SAstCreateContext* pCxt, const SToken* pLiteral) { SNode* createDurationValueNode(SAstCreateContext* pCxt, const SToken* pLiteral) {
SValueNode* val = (SValueNode*)nodesMakeNode(QUERY_NODE_VALUE); SValueNode* val = (SValueNode*)nodesMakeNode(QUERY_NODE_VALUE);
CHECK_OUT_OF_MEM(val); CHECK_OUT_OF_MEM(val);
// todo // todo : calc, for example, 10s
return (SNode*)val; return (SNode*)val;
} }
SNode* addMinusSign(SAstCreateContext* pCxt, SNode* pNode) {
// todo
}
SNode* createLogicConditionNode(SAstCreateContext* pCxt, ELogicConditionType type, SNode* pParam1, SNode* pParam2) { SNode* createLogicConditionNode(SAstCreateContext* pCxt, ELogicConditionType type, SNode* pParam1, SNode* pParam2) {
SLogicConditionNode* cond = (SLogicConditionNode*)nodesMakeNode(QUERY_NODE_LOGIC_CONDITION); SLogicConditionNode* cond = (SLogicConditionNode*)nodesMakeNode(QUERY_NODE_LOGIC_CONDITION);
CHECK_OUT_OF_MEM(cond); CHECK_OUT_OF_MEM(cond);

146
source/libs/parser/src/new_sql.c
View File

@ -1517,23 +1517,24 @@ static YYACTIONTYPE yy_reduce(
{ PARSER_TRACE; pCxt->pRootNode = yymsp[0].minor.yy168; } { PARSER_TRACE; pCxt->pRootNode = yymsp[0].minor.yy168; }
break; break;
case 2: /* literal ::= NK_INTEGER */ case 2: /* literal ::= NK_INTEGER */
{ PARSER_TRACE; yylhsminor.yy168 = createValueNode(pCxt, TSDB_DATA_TYPE_BIGINT, &yymsp[0].minor.yy0); } { PARSER_TRACE; yylhsminor.yy168 = createRawExprNode(pCxt, &yymsp[0].minor.yy0, createValueNode(pCxt, TSDB_DATA_TYPE_BIGINT, &yymsp[0].minor.yy0)); }
yymsp[0].minor.yy168 = yylhsminor.yy168; yymsp[0].minor.yy168 = yylhsminor.yy168;
break; break;
case 3: /* literal ::= NK_FLOAT */ case 3: /* literal ::= NK_FLOAT */
{ PARSER_TRACE; yylhsminor.yy168 = createValueNode(pCxt, TSDB_DATA_TYPE_DOUBLE, &yymsp[0].minor.yy0); } { PARSER_TRACE; yylhsminor.yy168 = createRawExprNode(pCxt, &yymsp[0].minor.yy0, createValueNode(pCxt, TSDB_DATA_TYPE_DOUBLE, &yymsp[0].minor.yy0)); }
yymsp[0].minor.yy168 = yylhsminor.yy168; yymsp[0].minor.yy168 = yylhsminor.yy168;
break; break;
case 4: /* literal ::= NK_STRING */ case 4: /* literal ::= NK_STRING */
{ PARSER_TRACE; yylhsminor.yy168 = createValueNode(pCxt, TSDB_DATA_TYPE_BINARY, &yymsp[0].minor.yy0); } { PARSER_TRACE; yylhsminor.yy168 = createRawExprNode(pCxt, &yymsp[0].minor.yy0, createValueNode(pCxt, TSDB_DATA_TYPE_BINARY, &yymsp[0].minor.yy0)); }
yymsp[0].minor.yy168 = yylhsminor.yy168; yymsp[0].minor.yy168 = yylhsminor.yy168;
break; break;
case 5: /* literal ::= NK_BOOL */ case 5: /* literal ::= NK_BOOL */
{ PARSER_TRACE; yylhsminor.yy168 = createValueNode(pCxt, TSDB_DATA_TYPE_BOOL, &yymsp[0].minor.yy0); } { PARSER_TRACE; yylhsminor.yy168 = createRawExprNode(pCxt, &yymsp[0].minor.yy0, createValueNode(pCxt, TSDB_DATA_TYPE_BOOL, &yymsp[0].minor.yy0)); }
yymsp[0].minor.yy168 = yylhsminor.yy168; yymsp[0].minor.yy168 = yylhsminor.yy168;
break; break;
case 6: /* literal ::= TIMESTAMP NK_STRING */ case 6: /* literal ::= TIMESTAMP NK_STRING */
{ PARSER_TRACE; yymsp[-1].minor.yy168 = createValueNode(pCxt, TSDB_DATA_TYPE_TIMESTAMP, &yymsp[0].minor.yy0); } { PARSER_TRACE; yylhsminor.yy168 = createRawExprNodeExt(pCxt, &yymsp[-1].minor.yy0, &yymsp[0].minor.yy0, createValueNode(pCxt, TSDB_DATA_TYPE_TIMESTAMP, &yymsp[0].minor.yy0)); }
yymsp[-1].minor.yy168 = yylhsminor.yy168;
break; break;
case 7: /* literal ::= duration_literal */ case 7: /* literal ::= duration_literal */
case 17: /* expression ::= literal */ yytestcase(yyruleno==17); case 17: /* expression ::= literal */ yytestcase(yyruleno==17);
@ -1545,7 +1546,6 @@ static YYACTIONTYPE yy_reduce(
case 61: /* table_reference ::= table_primary */ yytestcase(yyruleno==61); case 61: /* table_reference ::= table_primary */ yytestcase(yyruleno==61);
case 62: /* table_reference ::= joined_table */ yytestcase(yyruleno==62); case 62: /* table_reference ::= joined_table */ yytestcase(yyruleno==62);
case 66: /* table_primary ::= parenthesized_joined_table */ yytestcase(yyruleno==66); case 66: /* table_primary ::= parenthesized_joined_table */ yytestcase(yyruleno==66);
case 82: /* select_item ::= expression */ yytestcase(yyruleno==82);
case 110: /* query_expression_body ::= query_primary */ yytestcase(yyruleno==110); case 110: /* query_expression_body ::= query_primary */ yytestcase(yyruleno==110);
case 112: /* query_primary ::= query_specification */ yytestcase(yyruleno==112); case 112: /* query_primary ::= query_specification */ yytestcase(yyruleno==112);
case 124: /* search_condition ::= boolean_value_expression */ yytestcase(yyruleno==124); case 124: /* search_condition ::= boolean_value_expression */ yytestcase(yyruleno==124);
@ -1553,21 +1553,17 @@ static YYACTIONTYPE yy_reduce(
yymsp[0].minor.yy168 = yylhsminor.yy168; yymsp[0].minor.yy168 = yylhsminor.yy168;
break; break;
case 8: /* duration_literal ::= NK_VARIABLE */ case 8: /* duration_literal ::= NK_VARIABLE */
{ PARSER_TRACE; yylhsminor.yy168 = createDurationValueNode(pCxt, &yymsp[0].minor.yy0); } { PARSER_TRACE; yylhsminor.yy168 = createRawExprNode(pCxt, &yymsp[0].minor.yy0, createDurationValueNode(pCxt, &yymsp[0].minor.yy0)); }
yymsp[0].minor.yy168 = yylhsminor.yy168; yymsp[0].minor.yy168 = yylhsminor.yy168;
break; break;
case 9: /* literal_list ::= literal */ case 9: /* literal_list ::= literal */
case 29: /* expression_list ::= expression */ yytestcase(yyruleno==29); case 29: /* expression_list ::= expression */ yytestcase(yyruleno==29);
case 80: /* select_sublist ::= select_item */ yytestcase(yyruleno==80); { PARSER_TRACE; yylhsminor.yy192 = createNodeList(pCxt, releaseRawExprNode(pCxt, yymsp[0].minor.yy168)); }
case 125: /* sort_specification_list ::= sort_specification */ yytestcase(yyruleno==125);
{ PARSER_TRACE; yylhsminor.yy192 = createNodeList(pCxt, yymsp[0].minor.yy168); }
yymsp[0].minor.yy192 = yylhsminor.yy192; yymsp[0].minor.yy192 = yylhsminor.yy192;
break; break;
case 10: /* literal_list ::= literal_list NK_COMMA literal */ case 10: /* literal_list ::= literal_list NK_COMMA literal */
case 30: /* expression_list ::= expression_list NK_COMMA expression */ yytestcase(yyruleno==30); case 30: /* expression_list ::= expression_list NK_COMMA expression */ yytestcase(yyruleno==30);
case 81: /* select_sublist ::= select_sublist NK_COMMA select_item */ yytestcase(yyruleno==81); { PARSER_TRACE; yylhsminor.yy192 = addNodeToList(pCxt, yymsp[-2].minor.yy192, releaseRawExprNode(pCxt, yymsp[0].minor.yy168)); }
case 126: /* sort_specification_list ::= sort_specification_list NK_COMMA sort_specification */ yytestcase(yyruleno==126);
{ PARSER_TRACE; yylhsminor.yy192 = addNodeToList(pCxt, yymsp[-2].minor.yy192, yymsp[0].minor.yy168); }
yymsp[-2].minor.yy192 = yylhsminor.yy192; yymsp[-2].minor.yy192 = yylhsminor.yy192;
break; break;
case 11: /* db_name ::= NK_ID */ case 11: /* db_name ::= NK_ID */
@ -1580,74 +1576,106 @@ static YYACTIONTYPE yy_reduce(
yymsp[0].minor.yy241 = yylhsminor.yy241; yymsp[0].minor.yy241 = yylhsminor.yy241;
break; break;
case 19: /* expression ::= function_name NK_LP expression_list NK_RP */ case 19: /* expression ::= function_name NK_LP expression_list NK_RP */
{ PARSER_TRACE; yylhsminor.yy168 = createFunctionNode(pCxt, &yymsp[-3].minor.yy241, yymsp[-1].minor.yy192); } { PARSER_TRACE; yylhsminor.yy168 = createRawExprNodeExt(pCxt, &yymsp[-3].minor.yy241, &yymsp[0].minor.yy0, createFunctionNode(pCxt, &yymsp[-3].minor.yy241, yymsp[-1].minor.yy192)); }
yymsp[-3].minor.yy168 = yylhsminor.yy168; yymsp[-3].minor.yy168 = yylhsminor.yy168;
break; break;
case 21: /* expression ::= NK_LP expression NK_RP */ case 21: /* expression ::= NK_LP expression NK_RP */
case 57: /* boolean_primary ::= NK_LP boolean_value_expression NK_RP */ yytestcase(yyruleno==57); { PARSER_TRACE; yylhsminor.yy168 = createRawExprNodeExt(pCxt, &yymsp[-2].minor.yy0, &yymsp[0].minor.yy0, releaseRawExprNode(pCxt, yymsp[-1].minor.yy168)); }
case 70: /* parenthesized_joined_table ::= NK_LP joined_table NK_RP */ yytestcase(yyruleno==70); yymsp[-2].minor.yy168 = yylhsminor.yy168;
case 71: /* parenthesized_joined_table ::= NK_LP parenthesized_joined_table NK_RP */ yytestcase(yyruleno==71);
case 123: /* subquery ::= NK_LP query_expression NK_RP */ yytestcase(yyruleno==123);
{ PARSER_TRACE; yymsp[-2].minor.yy168 = yymsp[-1].minor.yy168; }
break; break;
case 22: /* expression ::= NK_PLUS expression */ case 22: /* expression ::= NK_PLUS expression */
case 58: /* from_clause ::= FROM table_reference_list */ yytestcase(yyruleno==58); {
case 87: /* where_clause_opt ::= WHERE search_condition */ yytestcase(yyruleno==87); PARSER_TRACE;
case 108: /* having_clause_opt ::= HAVING search_condition */ yytestcase(yyruleno==108); SToken t = getTokenFromRawExprNode(pCxt, yymsp[0].minor.yy168);
{ PARSER_TRACE; yymsp[-1].minor.yy168 = yymsp[0].minor.yy168; } yylhsminor.yy168 = createRawExprNodeExt(pCxt, &yymsp[-1].minor.yy0, &t, releaseRawExprNode(pCxt, yymsp[0].minor.yy168));
}
yymsp[-1].minor.yy168 = yylhsminor.yy168;
break; break;
case 23: /* expression ::= NK_MINUS expression */ case 23: /* expression ::= NK_MINUS expression */
{ PARSER_TRACE; yymsp[-1].minor.yy168 = createOperatorNode(pCxt, OP_TYPE_SUB, yymsp[0].minor.yy168, NULL); } {
PARSER_TRACE;
SToken t = getTokenFromRawExprNode(pCxt, yymsp[0].minor.yy168);
yylhsminor.yy168 = createRawExprNodeExt(pCxt, &yymsp[-1].minor.yy0, &t, createOperatorNode(pCxt, OP_TYPE_SUB, releaseRawExprNode(pCxt, yymsp[0].minor.yy168), NULL));
}
yymsp[-1].minor.yy168 = yylhsminor.yy168;
break; break;
case 24: /* expression ::= expression NK_PLUS expression */ case 24: /* expression ::= expression NK_PLUS expression */
{ PARSER_TRACE; yylhsminor.yy168 = createOperatorNode(pCxt, OP_TYPE_ADD, yymsp[-2].minor.yy168, yymsp[0].minor.yy168); } {
PARSER_TRACE;
SToken s = getTokenFromRawExprNode(pCxt, yymsp[-2].minor.yy168);
SToken e = getTokenFromRawExprNode(pCxt, yymsp[0].minor.yy168);
yylhsminor.yy168 = createRawExprNodeExt(pCxt, &s, &e, createOperatorNode(pCxt, OP_TYPE_ADD, releaseRawExprNode(pCxt, yymsp[-2].minor.yy168), releaseRawExprNode(pCxt, yymsp[0].minor.yy168)));
}
yymsp[-2].minor.yy168 = yylhsminor.yy168; yymsp[-2].minor.yy168 = yylhsminor.yy168;
break; break;
case 25: /* expression ::= expression NK_MINUS expression */ case 25: /* expression ::= expression NK_MINUS expression */
{ PARSER_TRACE; yylhsminor.yy168 = createOperatorNode(pCxt, OP_TYPE_SUB, yymsp[-2].minor.yy168, yymsp[0].minor.yy168); } {
PARSER_TRACE;
SToken s = getTokenFromRawExprNode(pCxt, yymsp[-2].minor.yy168);
SToken e = getTokenFromRawExprNode(pCxt, yymsp[0].minor.yy168);
yylhsminor.yy168 = createRawExprNodeExt(pCxt, &s, &e, createOperatorNode(pCxt, OP_TYPE_SUB, releaseRawExprNode(pCxt, yymsp[-2].minor.yy168), releaseRawExprNode(pCxt, yymsp[0].minor.yy168)));
}
yymsp[-2].minor.yy168 = yylhsminor.yy168; yymsp[-2].minor.yy168 = yylhsminor.yy168;
break; break;
case 26: /* expression ::= expression NK_STAR expression */ case 26: /* expression ::= expression NK_STAR expression */
{ PARSER_TRACE; yylhsminor.yy168 = createOperatorNode(pCxt, OP_TYPE_MULTI, yymsp[-2].minor.yy168, yymsp[0].minor.yy168); } {
PARSER_TRACE;
SToken s = getTokenFromRawExprNode(pCxt, yymsp[-2].minor.yy168);
SToken e = getTokenFromRawExprNode(pCxt, yymsp[0].minor.yy168);
yylhsminor.yy168 = createRawExprNodeExt(pCxt, &s, &e, createOperatorNode(pCxt, OP_TYPE_MULTI, releaseRawExprNode(pCxt, yymsp[-2].minor.yy168), releaseRawExprNode(pCxt, yymsp[0].minor.yy168)));
}
yymsp[-2].minor.yy168 = yylhsminor.yy168; yymsp[-2].minor.yy168 = yylhsminor.yy168;
break; break;
case 27: /* expression ::= expression NK_SLASH expression */ case 27: /* expression ::= expression NK_SLASH expression */
{ PARSER_TRACE; yylhsminor.yy168 = createOperatorNode(pCxt, OP_TYPE_DIV, yymsp[-2].minor.yy168, yymsp[0].minor.yy168); } {
PARSER_TRACE;
SToken s = getTokenFromRawExprNode(pCxt, yymsp[-2].minor.yy168);
SToken e = getTokenFromRawExprNode(pCxt, yymsp[0].minor.yy168);
yylhsminor.yy168 = createRawExprNodeExt(pCxt, &s, &e, createOperatorNode(pCxt, OP_TYPE_DIV, releaseRawExprNode(pCxt, yymsp[-2].minor.yy168), releaseRawExprNode(pCxt, yymsp[0].minor.yy168)));
}
yymsp[-2].minor.yy168 = yylhsminor.yy168; yymsp[-2].minor.yy168 = yylhsminor.yy168;
break; break;
case 28: /* expression ::= expression NK_REM expression */ case 28: /* expression ::= expression NK_REM expression */
{ PARSER_TRACE; yylhsminor.yy168 = createOperatorNode(pCxt, OP_TYPE_MOD, yymsp[-2].minor.yy168, yymsp[0].minor.yy168); } {
PARSER_TRACE;
SToken s = getTokenFromRawExprNode(pCxt, yymsp[-2].minor.yy168);
SToken e = getTokenFromRawExprNode(pCxt, yymsp[0].minor.yy168);
yylhsminor.yy168 = createRawExprNodeExt(pCxt, &s, &e, createOperatorNode(pCxt, OP_TYPE_MOD, releaseRawExprNode(pCxt, yymsp[-2].minor.yy168), releaseRawExprNode(pCxt, yymsp[0].minor.yy168)));
}
yymsp[-2].minor.yy168 = yylhsminor.yy168; yymsp[-2].minor.yy168 = yylhsminor.yy168;
break; break;
case 31: /* column_reference ::= column_name */ case 31: /* column_reference ::= column_name */
{ PARSER_TRACE; yylhsminor.yy168 = createColumnNode(pCxt, NULL, &yymsp[0].minor.yy241); } { PARSER_TRACE; yylhsminor.yy168 = createRawExprNode(pCxt, &yymsp[0].minor.yy241, createColumnNode(pCxt, NULL, &yymsp[0].minor.yy241)); }
yymsp[0].minor.yy168 = yylhsminor.yy168; yymsp[0].minor.yy168 = yylhsminor.yy168;
break; break;
case 32: /* column_reference ::= table_name NK_DOT column_name */ case 32: /* column_reference ::= table_name NK_DOT column_name */
{ PARSER_TRACE; yylhsminor.yy168 = createColumnNode(pCxt, &yymsp[-2].minor.yy241, &yymsp[0].minor.yy241); } { PARSER_TRACE; yylhsminor.yy168 = createRawExprNodeExt(pCxt, &yymsp[-2].minor.yy241, &yymsp[0].minor.yy241, createColumnNode(pCxt, &yymsp[-2].minor.yy241, &yymsp[0].minor.yy241)); }
yymsp[-2].minor.yy168 = yylhsminor.yy168; yymsp[-2].minor.yy168 = yylhsminor.yy168;
break; break;
case 33: /* predicate ::= expression compare_op expression */ case 33: /* predicate ::= expression compare_op expression */
case 38: /* predicate ::= expression in_op in_predicate_value */ yytestcase(yyruleno==38); { PARSER_TRACE; yylhsminor.yy168 = createOperatorNode(pCxt, yymsp[-1].minor.yy228, releaseRawExprNode(pCxt, yymsp[-2].minor.yy168), releaseRawExprNode(pCxt, yymsp[0].minor.yy168)); }
{ PARSER_TRACE; yylhsminor.yy168 = createOperatorNode(pCxt, yymsp[-1].minor.yy228, yymsp[-2].minor.yy168, yymsp[0].minor.yy168); }
yymsp[-2].minor.yy168 = yylhsminor.yy168; yymsp[-2].minor.yy168 = yylhsminor.yy168;
break; break;
case 34: /* predicate ::= expression BETWEEN expression AND expression */ case 34: /* predicate ::= expression BETWEEN expression AND expression */
{ PARSER_TRACE; yylhsminor.yy168 = createBetweenAnd(pCxt, yymsp[-4].minor.yy168, yymsp[-2].minor.yy168, yymsp[0].minor.yy168); } { PARSER_TRACE; yylhsminor.yy168 = createBetweenAnd(pCxt, releaseRawExprNode(pCxt, yymsp[-4].minor.yy168), releaseRawExprNode(pCxt, yymsp[-2].minor.yy168), releaseRawExprNode(pCxt, yymsp[0].minor.yy168)); }
yymsp[-4].minor.yy168 = yylhsminor.yy168; yymsp[-4].minor.yy168 = yylhsminor.yy168;
break; break;
case 35: /* predicate ::= expression NOT BETWEEN expression AND expression */ case 35: /* predicate ::= expression NOT BETWEEN expression AND expression */
{ PARSER_TRACE; yylhsminor.yy168 = createNotBetweenAnd(pCxt, yymsp[-2].minor.yy168, yymsp[-5].minor.yy168, yymsp[0].minor.yy168); } { PARSER_TRACE; yylhsminor.yy168 = createNotBetweenAnd(pCxt, releaseRawExprNode(pCxt, yymsp[-2].minor.yy168), releaseRawExprNode(pCxt, yymsp[-5].minor.yy168), releaseRawExprNode(pCxt, yymsp[0].minor.yy168)); }
yymsp[-5].minor.yy168 = yylhsminor.yy168; yymsp[-5].minor.yy168 = yylhsminor.yy168;
break; break;
case 36: /* predicate ::= expression IS NULL */ case 36: /* predicate ::= expression IS NULL */
{ PARSER_TRACE; yylhsminor.yy168 = createIsNullCondNode(pCxt, yymsp[-2].minor.yy168, true); } { PARSER_TRACE; yylhsminor.yy168 = createIsNullCondNode(pCxt, releaseRawExprNode(pCxt, yymsp[-2].minor.yy168), true); }
yymsp[-2].minor.yy168 = yylhsminor.yy168; yymsp[-2].minor.yy168 = yylhsminor.yy168;
break; break;
case 37: /* predicate ::= expression IS NOT NULL */ case 37: /* predicate ::= expression IS NOT NULL */
{ PARSER_TRACE; yylhsminor.yy168 = createIsNullCondNode(pCxt, yymsp[-3].minor.yy168, false); } { PARSER_TRACE; yylhsminor.yy168 = createIsNullCondNode(pCxt, releaseRawExprNode(pCxt, yymsp[-3].minor.yy168), false); }
yymsp[-3].minor.yy168 = yylhsminor.yy168; yymsp[-3].minor.yy168 = yylhsminor.yy168;
break; break;
case 38: /* predicate ::= expression in_op in_predicate_value */
{ PARSER_TRACE; yylhsminor.yy168 = createOperatorNode(pCxt, yymsp[-1].minor.yy228, releaseRawExprNode(pCxt, yymsp[-2].minor.yy168), yymsp[0].minor.yy168); }
yymsp[-2].minor.yy168 = yylhsminor.yy168;
break;
case 39: /* compare_op ::= NK_LT */ case 39: /* compare_op ::= NK_LT */
{ PARSER_TRACE; yymsp[0].minor.yy228 = OP_TYPE_LOWER_THAN; } { PARSER_TRACE; yymsp[0].minor.yy228 = OP_TYPE_LOWER_THAN; }
break; break;
@ -1698,6 +1726,16 @@ static YYACTIONTYPE yy_reduce(
{ PARSER_TRACE; yylhsminor.yy168 = createLogicConditionNode(pCxt, LOGIC_COND_TYPE_AND, yymsp[-2].minor.yy168, yymsp[0].minor.yy168); } { PARSER_TRACE; yylhsminor.yy168 = createLogicConditionNode(pCxt, LOGIC_COND_TYPE_AND, yymsp[-2].minor.yy168, yymsp[0].minor.yy168); }
yymsp[-2].minor.yy168 = yylhsminor.yy168; yymsp[-2].minor.yy168 = yylhsminor.yy168;
break; break;
case 57: /* boolean_primary ::= NK_LP boolean_value_expression NK_RP */
case 70: /* parenthesized_joined_table ::= NK_LP joined_table NK_RP */ yytestcase(yyruleno==70);
case 71: /* parenthesized_joined_table ::= NK_LP parenthesized_joined_table NK_RP */ yytestcase(yyruleno==71);
{ PARSER_TRACE; yymsp[-2].minor.yy168 = yymsp[-1].minor.yy168; }
break;
case 58: /* from_clause ::= FROM table_reference_list */
case 87: /* where_clause_opt ::= WHERE search_condition */ yytestcase(yyruleno==87);
case 108: /* having_clause_opt ::= HAVING search_condition */ yytestcase(yyruleno==108);
{ PARSER_TRACE; yymsp[-1].minor.yy168 = yymsp[0].minor.yy168; }
break;
case 60: /* table_reference_list ::= table_reference_list NK_COMMA table_reference */ case 60: /* table_reference_list ::= table_reference_list NK_COMMA table_reference */
{ PARSER_TRACE; yylhsminor.yy168 = createJoinTableNode(pCxt, JOIN_TYPE_INNER, yymsp[-2].minor.yy168, yymsp[0].minor.yy168, NULL); } { PARSER_TRACE; yylhsminor.yy168 = createJoinTableNode(pCxt, JOIN_TYPE_INNER, yymsp[-2].minor.yy168, yymsp[0].minor.yy168, NULL); }
yymsp[-2].minor.yy168 = yylhsminor.yy168; yymsp[-2].minor.yy168 = yylhsminor.yy168;
@ -1711,7 +1749,7 @@ static YYACTIONTYPE yy_reduce(
yymsp[-3].minor.yy168 = yylhsminor.yy168; yymsp[-3].minor.yy168 = yylhsminor.yy168;
break; break;
case 65: /* table_primary ::= subquery alias_opt */ case 65: /* table_primary ::= subquery alias_opt */
{ PARSER_TRACE; yylhsminor.yy168 = createTempTableNode(pCxt, yymsp[-1].minor.yy168, &yymsp[0].minor.yy241); } { PARSER_TRACE; yylhsminor.yy168 = createTempTableNode(pCxt, releaseRawExprNode(pCxt, yymsp[-1].minor.yy168), &yymsp[0].minor.yy241); }
yymsp[-1].minor.yy168 = yylhsminor.yy168; yymsp[-1].minor.yy168 = yylhsminor.yy168;
break; break;
case 67: /* alias_opt ::= */ case 67: /* alias_opt ::= */
@ -1758,12 +1796,30 @@ static YYACTIONTYPE yy_reduce(
{ PARSER_TRACE; yylhsminor.yy192 = yymsp[0].minor.yy192; } { PARSER_TRACE; yylhsminor.yy192 = yymsp[0].minor.yy192; }
yymsp[0].minor.yy192 = yylhsminor.yy192; yymsp[0].minor.yy192 = yylhsminor.yy192;
break; break;
case 80: /* select_sublist ::= select_item */
case 125: /* sort_specification_list ::= sort_specification */ yytestcase(yyruleno==125);
{ PARSER_TRACE; yylhsminor.yy192 = createNodeList(pCxt, yymsp[0].minor.yy168); }
yymsp[0].minor.yy192 = yylhsminor.yy192;
break;
case 81: /* select_sublist ::= select_sublist NK_COMMA select_item */
case 126: /* sort_specification_list ::= sort_specification_list NK_COMMA sort_specification */ yytestcase(yyruleno==126);
{ PARSER_TRACE; yylhsminor.yy192 = addNodeToList(pCxt, yymsp[-2].minor.yy192, yymsp[0].minor.yy168); }
yymsp[-2].minor.yy192 = yylhsminor.yy192;
break;
case 82: /* select_item ::= expression */
{
PARSER_TRACE;
SToken t = getTokenFromRawExprNode(pCxt, yymsp[0].minor.yy168);
yylhsminor.yy168 = setProjectionAlias(pCxt, releaseRawExprNode(pCxt, yymsp[0].minor.yy168), &t);
}
yymsp[0].minor.yy168 = yylhsminor.yy168;
break;
case 83: /* select_item ::= expression column_alias */ case 83: /* select_item ::= expression column_alias */
{ PARSER_TRACE; yylhsminor.yy168 = setProjectionAlias(pCxt, yymsp[-1].minor.yy168, &yymsp[0].minor.yy241); } { PARSER_TRACE; yylhsminor.yy168 = setProjectionAlias(pCxt, releaseRawExprNode(pCxt, yymsp[-1].minor.yy168), &yymsp[0].minor.yy241); }
yymsp[-1].minor.yy168 = yylhsminor.yy168; yymsp[-1].minor.yy168 = yylhsminor.yy168;
break; break;
case 84: /* select_item ::= expression AS column_alias */ case 84: /* select_item ::= expression AS column_alias */
{ PARSER_TRACE; yylhsminor.yy168 = setProjectionAlias(pCxt, yymsp[-2].minor.yy168, &yymsp[0].minor.yy241); } { PARSER_TRACE; yylhsminor.yy168 = setProjectionAlias(pCxt, releaseRawExprNode(pCxt, yymsp[-2].minor.yy168), &yymsp[0].minor.yy241); }
yymsp[-2].minor.yy168 = yylhsminor.yy168; yymsp[-2].minor.yy168 = yylhsminor.yy168;
break; break;
case 85: /* select_item ::= table_name NK_DOT NK_STAR */ case 85: /* select_item ::= table_name NK_DOT NK_STAR */
@ -1790,10 +1846,10 @@ static YYACTIONTYPE yy_reduce(
{ PARSER_TRACE; yymsp[-2].minor.yy192 = yymsp[0].minor.yy192; } { PARSER_TRACE; yymsp[-2].minor.yy192 = yymsp[0].minor.yy192; }
break; break;
case 91: /* twindow_clause_opt ::= SESSION NK_LP column_reference NK_COMMA NK_INTEGER NK_RP */ case 91: /* twindow_clause_opt ::= SESSION NK_LP column_reference NK_COMMA NK_INTEGER NK_RP */
{ PARSER_TRACE; yymsp[-5].minor.yy168 = createSessionWindowNode(pCxt, yymsp[-3].minor.yy168, &yymsp[-1].minor.yy0); } { PARSER_TRACE; yymsp[-5].minor.yy168 = createSessionWindowNode(pCxt, releaseRawExprNode(pCxt, yymsp[-3].minor.yy168), &yymsp[-1].minor.yy0); }
break; break;
case 92: /* twindow_clause_opt ::= STATE_WINDOW NK_LP column_reference NK_RP */ case 92: /* twindow_clause_opt ::= STATE_WINDOW NK_LP column_reference NK_RP */
{ PARSER_TRACE; yymsp[-3].minor.yy168 = createStateWindowNode(pCxt, yymsp[-1].minor.yy168); } { PARSER_TRACE; yymsp[-3].minor.yy168 = createStateWindowNode(pCxt, releaseRawExprNode(pCxt, yymsp[-1].minor.yy168)); }
break; break;
case 93: /* twindow_clause_opt ::= INTERVAL NK_LP duration_literal NK_RP sliding_opt fill_opt */ case 93: /* twindow_clause_opt ::= INTERVAL NK_LP duration_literal NK_RP sliding_opt fill_opt */
{ PARSER_TRACE; yymsp[-5].minor.yy168 = createIntervalWindowNode(pCxt, yymsp[-3].minor.yy168, NULL, yymsp[-1].minor.yy168, yymsp[0].minor.yy168); } { PARSER_TRACE; yymsp[-5].minor.yy168 = createIntervalWindowNode(pCxt, yymsp[-3].minor.yy168, NULL, yymsp[-1].minor.yy168, yymsp[0].minor.yy168); }
@ -1850,8 +1906,12 @@ static YYACTIONTYPE yy_reduce(
case 122: /* limit_clause_opt ::= LIMIT NK_INTEGER NK_COMMA NK_INTEGER */ yytestcase(yyruleno==122); case 122: /* limit_clause_opt ::= LIMIT NK_INTEGER NK_COMMA NK_INTEGER */ yytestcase(yyruleno==122);
{ PARSER_TRACE; yymsp[-3].minor.yy168 = createLimitNode(pCxt, &yymsp[0].minor.yy0, &yymsp[-2].minor.yy0); } { PARSER_TRACE; yymsp[-3].minor.yy168 = createLimitNode(pCxt, &yymsp[0].minor.yy0, &yymsp[-2].minor.yy0); }
break; break;
case 123: /* subquery ::= NK_LP query_expression NK_RP */
{ PARSER_TRACE; yylhsminor.yy168 = createRawExprNodeExt(pCxt, &yymsp[-2].minor.yy0, &yymsp[0].minor.yy0, yymsp[-1].minor.yy168); }
yymsp[-2].minor.yy168 = yylhsminor.yy168;
break;
case 127: /* sort_specification ::= expression ordering_specification_opt null_ordering_opt */ case 127: /* sort_specification ::= expression ordering_specification_opt null_ordering_opt */
{ PARSER_TRACE; yylhsminor.yy168 = createOrderByExprNode(pCxt, yymsp[-2].minor.yy168, yymsp[-1].minor.yy10, yymsp[0].minor.yy177); } { PARSER_TRACE; yylhsminor.yy168 = createOrderByExprNode(pCxt, releaseRawExprNode(pCxt, yymsp[-2].minor.yy168), yymsp[-1].minor.yy10, yymsp[0].minor.yy177); }
yymsp[-2].minor.yy168 = yylhsminor.yy168; yymsp[-2].minor.yy168 = yylhsminor.yy168;
break; break;
case 128: /* ordering_specification_opt ::= */ case 128: /* ordering_specification_opt ::= */

196
source/libs/parser/src/parserImpl.c
View File

@ -28,82 +28,11 @@ extern void NewParseFree(void*, FFree);
extern void NewParseTrace(FILE*, char*); extern void NewParseTrace(FILE*, char*);
static uint32_t toNewTokenId(uint32_t tokenId) { static uint32_t toNewTokenId(uint32_t tokenId) {
// #define 1
// #define NEW_TK_AND 2
// #define NEW_TK_UNION 3
// #define NEW_TK_ALL 4
// #define NEW_TK_MINUS 5
// #define NEW_TK_EXCEPT 6
// #define NEW_TK_INTERSECT 7
// #define NEW_TK_NK_PLUS 8
// #define NEW_TK_NK_MINUS 9
// #define NEW_TK_NK_STAR 10
// #define NEW_TK_NK_SLASH 11
// #define NEW_TK_NK_REM 12
// #define NEW_TK_SHOW 13
// #define NEW_TK_DATABASES 14
// #define NEW_TK_NK_INTEGER 15
// #define NEW_TK_NK_FLOAT 16
// #define NEW_TK_NK_STRING 17
// #define NEW_TK_NK_BOOL 18
// #define NEW_TK_TIMESTAMP 19
// #define NEW_TK_NK_VARIABLE 20
// #define NEW_TK_NK_COMMA 21
// #define NEW_TK_NK_ID 22
// #define NEW_TK_NK_LP 23
// #define NEW_TK_NK_RP 24
// #define NEW_TK_NK_DOT 25
// #define NEW_TK_BETWEEN 26
// #define NEW_TK_NOT 27
// #define NEW_TK_IS 28
// #define NEW_TK_NULL 29
// #define NEW_TK_NK_LT 30
// #define NEW_TK_NK_GT 31
// #define NEW_TK_NK_LE 32
// #define NEW_TK_NK_GE 33
// #define NEW_TK_NK_NE 34
// #define 35
// #define NEW_TK_LIKE 36
// #define NEW_TK_MATCH 37
// #define NEW_TK_NMATCH 38
// #define NEW_TK_IN 39
// #define NEW_TK_FROM 40
// #define NEW_TK_AS 41
// #define NEW_TK_JOIN 42
// #define NEW_TK_ON 43
// #define NEW_TK_INNER 44
// #define NEW_TK_SELECT 45
// #define NEW_TK_DISTINCT 46
// #define 47
// #define NEW_TK_PARTITION 48
// #define NEW_TK_BY 49
// #define NEW_TK_SESSION 50
// #define NEW_TK_STATE_WINDOW 51
// #define NEW_TK_INTERVAL 52
// #define NEW_TK_SLIDING 53
// #define NEW_TK_FILL 54
// #define NEW_TK_VALUE 55
// #define NEW_TK_NONE 56
// #define NEW_TK_PREV 57
// #define NEW_TK_LINEAR 58
// #define NEW_TK_NEXT 59
// #define NEW_TK_GROUP 60
// #define NEW_TK_HAVING 61
// #define NEW_TK_ORDER 62
// #define NEW_TK_SLIMIT 63
// #define NEW_TK_SOFFSET 64
// #define NEW_TK_LIMIT 65
// #define NEW_TK_OFFSET 66
// #define NEW_TK_NK_LR 67
// #define NEW_TK_ASC 68
// #define NEW_TK_DESC 69
// #define NEW_TK_NULLS 70
// #define NEW_TK_FIRST 71
// #define NEW_TK_LAST 72
switch (tokenId) { switch (tokenId) {
case TK_OR: case TK_OR:
return NEW_TK_OR; return NEW_TK_OR;
case TK_AND:
return NEW_TK_AND;
case TK_UNION: case TK_UNION:
return NEW_TK_UNION; return NEW_TK_UNION;
case TK_ALL: case TK_ALL:
@ -116,22 +45,62 @@ static uint32_t toNewTokenId(uint32_t tokenId) {
return NEW_TK_NK_STAR; return NEW_TK_NK_STAR;
case TK_SLASH: case TK_SLASH:
return NEW_TK_NK_SLASH; return NEW_TK_NK_SLASH;
case TK_REM:
return NEW_TK_NK_REM;
case TK_SHOW: case TK_SHOW:
return NEW_TK_SHOW; return NEW_TK_SHOW;
case TK_DATABASES: case TK_DATABASES:
return NEW_TK_DATABASES; return NEW_TK_DATABASES;
case TK_INTEGER:
return NEW_TK_NK_INTEGER;
case TK_FLOAT:
return NEW_TK_NK_FLOAT;
case TK_STRING:
return NEW_TK_NK_STRING;
case TK_BOOL:
return NEW_TK_NK_BOOL;
case TK_TIMESTAMP:
return NEW_TK_TIMESTAMP;
case TK_VARIABLE:
return NEW_TK_NK_VARIABLE;
case TK_COMMA:
return NEW_TK_NK_COMMA;
case TK_ID: case TK_ID:
return NEW_TK_NK_ID; return NEW_TK_NK_ID;
case TK_LP: case TK_LP:
return NEW_TK_NK_LP; return NEW_TK_NK_LP;
case TK_RP: case TK_RP:
return NEW_TK_NK_RP; return NEW_TK_NK_RP;
case TK_COMMA:
return NEW_TK_NK_COMMA;
case TK_DOT: case TK_DOT:
return NEW_TK_NK_DOT; return NEW_TK_NK_DOT;
case TK_BETWEEN:
return NEW_TK_BETWEEN;
case TK_NOT:
return NEW_TK_NOT;
case TK_IS:
return NEW_TK_IS;
case TK_NULL:
return NEW_TK_NULL;
case TK_LT:
return NEW_TK_NK_LT;
case TK_GT:
return NEW_TK_NK_GT;
case TK_LE:
return NEW_TK_NK_LE;
case TK_GE:
return NEW_TK_NK_GE;
case TK_NE:
return NEW_TK_NK_NE;
case TK_EQ: case TK_EQ:
return NEW_TK_NK_EQ; return NEW_TK_NK_EQ;
case TK_LIKE:
return NEW_TK_LIKE;
case TK_MATCH:
return NEW_TK_MATCH;
case TK_NMATCH:
return NEW_TK_NMATCH;
case TK_IN:
return NEW_TK_IN;
case TK_SELECT: case TK_SELECT:
return NEW_TK_SELECT; return NEW_TK_SELECT;
case TK_DISTINCT: case TK_DISTINCT:
@ -142,6 +111,38 @@ static uint32_t toNewTokenId(uint32_t tokenId) {
return NEW_TK_AS; return NEW_TK_AS;
case TK_FROM: case TK_FROM:
return NEW_TK_FROM; return NEW_TK_FROM;
case TK_JOIN:
return NEW_TK_JOIN;
// case TK_ON:
// return NEW_TK_ON;
// case TK_INNER:
// return NEW_TK_INNER;
// case TK_PARTITION:
// return NEW_TK_PARTITION;
case TK_SESSION:
return NEW_TK_SESSION;
case TK_STATE_WINDOW:
return NEW_TK_STATE_WINDOW;
case TK_INTERVAL:
return NEW_TK_INTERVAL;
case TK_SLIDING:
return NEW_TK_SLIDING;
case TK_FILL:
return NEW_TK_FILL;
// case TK_VALUE:
// return NEW_TK_VALUE;
case TK_NONE:
return NEW_TK_NONE;
case TK_PREV:
return NEW_TK_PREV;
case TK_LINEAR:
return NEW_TK_LINEAR;
// case TK_NEXT:
// return NEW_TK_NEXT;
case TK_GROUP:
return NEW_TK_GROUP;
case TK_HAVING:
return NEW_TK_HAVING;
case TK_ORDER: case TK_ORDER:
return NEW_TK_ORDER; return NEW_TK_ORDER;
case TK_BY: case TK_BY:
@ -150,6 +151,14 @@ static uint32_t toNewTokenId(uint32_t tokenId) {
return NEW_TK_ASC; return NEW_TK_ASC;
case TK_DESC: case TK_DESC:
return NEW_TK_DESC; return NEW_TK_DESC;
case TK_SLIMIT:
return NEW_TK_SLIMIT;
case TK_SOFFSET:
return NEW_TK_SOFFSET;
case TK_LIMIT:
return NEW_TK_LIMIT;
case TK_OFFSET:
return NEW_TK_OFFSET;
case TK_SPACE: case TK_SPACE:
break; break;
default: default:
@ -224,14 +233,6 @@ abort_parse:
return cxt.valid ? TSDB_CODE_SUCCESS : TSDB_CODE_FAILED; return cxt.valid ? TSDB_CODE_SUCCESS : TSDB_CODE_FAILED;
} }
// typedef struct SNamespace {
// int16_t level; // todo for correlated subquery
// char dbName[TSDB_DB_NAME_LEN];
// char tableAlias[TSDB_TABLE_NAME_LEN];
// SHashObj* pColHash; // key is colname, value is index of STableMeta.schema
// STableMeta* pMeta;
// } SNamespace;
typedef enum ESqlClause { typedef enum ESqlClause {
SQL_CLAUSE_FROM = 1, SQL_CLAUSE_FROM = 1,
SQL_CLAUSE_WHERE SQL_CLAUSE_WHERE
@ -256,6 +257,8 @@ static char* getSyntaxErrFormat(int32_t errCode) {
return "Table does not exist : %s"; return "Table does not exist : %s";
case TSDB_CODE_PARSER_AMBIGUOUS_COLUMN: case TSDB_CODE_PARSER_AMBIGUOUS_COLUMN:
return "Column ambiguously defined : %s"; return "Column ambiguously defined : %s";
case TSDB_CODE_PARSER_WRONG_VALUE_TYPE:
return "Invalid value type : %s";
default: default:
return "Unknown error"; return "Unknown error";
} }
@ -322,7 +325,8 @@ static void setColumnInfoBySchema(const STableNode* pTable, const SSchema* pColS
strcpy(pCol->node.aliasName, pColSchema->name); strcpy(pCol->node.aliasName, pColSchema->name);
} }
pCol->colId = pColSchema->colId; pCol->colId = pColSchema->colId;
pCol->colType = pColSchema->type; // pCol->colType = pColSchema->type;
pCol->node.resType.type = pColSchema->type;
pCol->node.resType.bytes = pColSchema->bytes; pCol->node.resType.bytes = pColSchema->bytes;
} }
@ -431,6 +435,30 @@ static bool translateValue(STranslateContext* pCxt, SValueNode* pVal) {
} }
static bool translateOperator(STranslateContext* pCxt, SOperatorNode* pOp) { static bool translateOperator(STranslateContext* pCxt, SOperatorNode* pOp) {
SDataType ldt = ((SExprNode*)(pOp->pLeft))->resType;
SDataType rdt = ((SExprNode*)(pOp->pRight))->resType;
if (nodesIsArithmeticOp(pOp)) {
if (TSDB_DATA_TYPE_JSON == ldt.type || TSDB_DATA_TYPE_BLOB == ldt.type ||
TSDB_DATA_TYPE_JSON == rdt.type || TSDB_DATA_TYPE_BLOB == rdt.type) {
generateSyntaxErrMsg(pCxt, TSDB_CODE_PARSER_WRONG_VALUE_TYPE, ((SExprNode*)(pOp->pRight))->aliasName);
return false;
}
pOp->node.resType.type = TSDB_DATA_TYPE_DOUBLE;
pOp->node.resType.bytes = tDataTypes[TSDB_DATA_TYPE_DOUBLE].bytes;
return true;
} else if (nodesIsComparisonOp(pOp)) {
if (TSDB_DATA_TYPE_JSON == ldt.type || TSDB_DATA_TYPE_BLOB == ldt.type ||
TSDB_DATA_TYPE_JSON == rdt.type || TSDB_DATA_TYPE_BLOB == rdt.type) {
generateSyntaxErrMsg(pCxt, TSDB_CODE_PARSER_WRONG_VALUE_TYPE, ((SExprNode*)(pOp->pRight))->aliasName);
return false;
}
pOp->node.resType.type = TSDB_DATA_TYPE_BOOL;
pOp->node.resType.bytes = tDataTypes[TSDB_DATA_TYPE_BOOL].bytes;
return true;
} else {
// todo json operator
return true;
}
return true; return true;
} }

183
source/libs/parser/test/newParserTest.cpp
View File

@ -55,10 +55,13 @@ protected:
return (TSDB_CODE_SUCCESS != translateCode); return (TSDB_CODE_SUCCESS != translateCode);
} }
if (NULL != query_.pRoot && QUERY_NODE_SELECT_STMT == nodeType(query_.pRoot)) { if (NULL != query_.pRoot && QUERY_NODE_SELECT_STMT == nodeType(query_.pRoot)) {
string sql;
selectToSql(query_.pRoot, sql);
cout << "input sql : [" << cxt_.pSql << "]" << endl; cout << "input sql : [" << cxt_.pSql << "]" << endl;
cout << "output sql : [" << sql << "]" << endl; // string sql;
// selectToSql(query_.pRoot, sql);
// cout << "output sql : [" << sql << "]" << endl;
string str;
selectToStr(query_.pRoot, str);
cout << "translate str : \n" << str << endl;
} }
return (TSDB_CODE_SUCCESS == translateCode); return (TSDB_CODE_SUCCESS == translateCode);
} }
@ -67,6 +70,162 @@ private:
static const int max_err_len = 1024; static const int max_err_len = 1024;
static const int max_sql_len = 1024 * 1024; static const int max_sql_len = 1024 * 1024;
string dataTypeToStr(const SDataType& dt) {
switch (dt.type) {
case TSDB_DATA_TYPE_NULL:
return "NULL";
case TSDB_DATA_TYPE_BOOL:
return "BOOL";
case TSDB_DATA_TYPE_TINYINT:
return "TINYINT";
case TSDB_DATA_TYPE_SMALLINT:
return "SMALLINT";
case TSDB_DATA_TYPE_INT:
return "INT";
case TSDB_DATA_TYPE_BIGINT:
return "BIGINT";
case TSDB_DATA_TYPE_FLOAT:
return "FLOAT";
case TSDB_DATA_TYPE_DOUBLE:
return "DOUBLE";
case TSDB_DATA_TYPE_BINARY:
return "BINART(" + to_string(dt.bytes) + ")";
case TSDB_DATA_TYPE_TIMESTAMP:
return "TIMESTAMP";
case TSDB_DATA_TYPE_NCHAR:
return "NCHAR(" + to_string(dt.bytes) + ")";
case TSDB_DATA_TYPE_UTINYINT:
return "UTINYINT";
case TSDB_DATA_TYPE_USMALLINT:
return "USMALLINT";
case TSDB_DATA_TYPE_UINT:
return "UINT";
case TSDB_DATA_TYPE_UBIGINT:
return "UBIGINT";
case TSDB_DATA_TYPE_VARCHAR:
return "VARCHAR(" + to_string(dt.bytes) + ")";
case TSDB_DATA_TYPE_VARBINARY:
return "VARBINARY(" + to_string(dt.bytes) + ")";
case TSDB_DATA_TYPE_JSON:
return "JSON";
case TSDB_DATA_TYPE_DECIMAL:
return "DECIMAL(" + to_string(dt.precision) + ", " + to_string(dt.scale) + ")";
case TSDB_DATA_TYPE_BLOB:
return "BLOB";
default:
break;
}
return "Unknown Data Type " + to_string(dt.type);
}
void nodeToStr(const SNode* node, string& str, bool isProject) {
if (nullptr == node) {
return;
}
switch (nodeType(node)) {
case QUERY_NODE_COLUMN: {
SColumnNode* pCol = (SColumnNode*)node;
if ('\0' != pCol->dbName[0]) {
str.append(pCol->dbName);
str.append(".");
}
if ('\0' != pCol->tableAlias[0]) {
str.append(pCol->tableAlias);
str.append(".");
}
str.append(pCol->colName);
str.append(" [" + dataTypeToStr(pCol->node.resType) + "]");
if (isProject) {
str.append(" AS " + string(pCol->node.aliasName));
}
break;
}
case QUERY_NODE_VALUE: {
SValueNode* pVal = (SValueNode*)node;
str.append(pVal->literal);
str.append(" [" + dataTypeToStr(pVal->node.resType) + "]");
if (isProject) {
str.append(" AS " + string(pVal->node.aliasName));
}
break;
}
case QUERY_NODE_OPERATOR: {
SOperatorNode* pOp = (SOperatorNode*)node;
nodeToStr(pOp->pLeft, str, false);
str.append(opTypeToStr(pOp->opType));
nodeToStr(pOp->pRight, str, false);
str.append(" [" + dataTypeToStr(pOp->node.resType) + "]");
if (isProject) {
str.append(" AS " + string(pOp->node.aliasName));
}
break;
}
default:
break;
}
}
void nodeListToStr(const SNodeList* nodelist, const string& prefix, string& str, bool isProject = false) {
SNode* node = nullptr;
FOREACH(node, nodelist) {
str.append(prefix);
nodeToStr(node, str, isProject);
str.append("\n");
}
}
void tableToStr(const SNode* node, const string& prefix, string& str) {
const STableNode* table = (const STableNode*)node;
switch (nodeType(node)) {
case QUERY_NODE_REAL_TABLE: {
SRealTableNode* realTable = (SRealTableNode*)table;
str.append(prefix);
if ('\0' != realTable->table.dbName[0]) {
str.append(realTable->table.dbName);
str.append(".");
}
str.append(realTable->table.tableName);
str.append(string(" ") + realTable->table.tableAlias);
break;
}
case QUERY_NODE_TEMP_TABLE: {
STempTableNode* tempTable = (STempTableNode*)table;
str.append(prefix + "(\n");
selectToStr(tempTable->pSubquery, str, prefix + "\t");
str.append("\n");
str.append(prefix + ") ");
str.append(tempTable->table.tableAlias);
break;
}
case QUERY_NODE_JOIN_TABLE: {
SJoinTableNode* joinTable = (SJoinTableNode*)table;
tableToStr(joinTable->pLeft, prefix, str);
str.append("\n" + prefix + "JOIN\n");
tableToStr(joinTable->pRight, prefix, str);
if (nullptr != joinTable->pOnCond) {
str.append("\n" + prefix + "\tON ");
nodeToStr(joinTable->pOnCond, str, false);
}
break;
}
default:
break;
}
}
void selectToStr(const SNode* node, string& str, const string& prefix = "") {
SSelectStmt* select = (SSelectStmt*)node;
str.append(prefix + "SELECT ");
if (select->isDistinct) {
str.append("DISTINCT");
}
str.append("\n");
nodeListToStr(select->pProjectionList, prefix + "\t", str, true);
str.append("\n" + prefix + "FROM\n");
tableToStr(select->pFromTable, prefix + "\t", str);
}
void selectToSql(const SNode* node, string& sql) { void selectToSql(const SNode* node, string& sql) {
SSelectStmt* select = (SSelectStmt*)node; SSelectStmt* select = (SSelectStmt*)node;
sql.append("SELECT "); sql.append("SELECT ");
@ -123,7 +282,7 @@ private:
} }
} }
string opTypeToSql(EOperatorType type) { string opTypeToStr(EOperatorType type) {
switch (type) { switch (type) {
case OP_TYPE_ADD: case OP_TYPE_ADD:
return " + "; return " + ";
@ -177,7 +336,7 @@ private:
case QUERY_NODE_OPERATOR: { case QUERY_NODE_OPERATOR: {
SOperatorNode* pOp = (SOperatorNode*)node; SOperatorNode* pOp = (SOperatorNode*)node;
nodeToSql(pOp->pLeft, sql); nodeToSql(pOp->pLeft, sql);
sql.append(opTypeToSql(pOp->opType)); sql.append(opTypeToStr(pOp->opType));
nodeToSql(pOp->pRight, sql); nodeToSql(pOp->pRight, sql);
break; break;
} }
@ -213,8 +372,7 @@ private:
SQuery query_; SQuery query_;
}; };
// SELECT * FROM t1 TEST_F(NewParserTest, selectSimple) {
TEST_F(NewParserTest, selectStar) {
setDatabase("root", "test"); setDatabase("root", "test");
bind("SELECT * FROM t1"); bind("SELECT * FROM t1");
@ -233,7 +391,14 @@ TEST_F(NewParserTest, selectStar) {
ASSERT_TRUE(run()); ASSERT_TRUE(run());
} }
TEST_F(NewParserTest, syntaxError) { TEST_F(NewParserTest, selectExpression) {
setDatabase("root", "test");
bind("SELECT c1 + 10, c2 FROM t1");
ASSERT_TRUE(run());
}
TEST_F(NewParserTest, selectSyntaxError) {
setDatabase("root", "test"); setDatabase("root", "test");
bind("SELECTT * FROM t1"); bind("SELECTT * FROM t1");
@ -249,7 +414,7 @@ TEST_F(NewParserTest, syntaxError) {
ASSERT_TRUE(run(TSDB_CODE_FAILED)); ASSERT_TRUE(run(TSDB_CODE_FAILED));
} }
TEST_F(NewParserTest, semanticError) { TEST_F(NewParserTest, selectSemanticError) {
setDatabase("root", "test"); setDatabase("root", "test");
bind("SELECT * FROM t10"); bind("SELECT * FROM t10");

5
source/libs/qcom/src/querymsg.c
View File

@ -119,9 +119,9 @@ int32_t queryProcessUseDBRsp(void* output, char *msg, int32_t msgSize) {
return TSDB_CODE_TSC_OUT_OF_MEMORY; return TSDB_CODE_TSC_OUT_OF_MEMORY;
} }
pOut->dbId = pRsp->uid;
pOut->dbVgroup->vgVersion = pRsp->vgVersion; pOut->dbVgroup->vgVersion = pRsp->vgVersion;
pOut->dbVgroup->hashMethod = pRsp->hashMethod; pOut->dbVgroup->hashMethod = pRsp->hashMethod;
pOut->dbVgroup->dbId = pRsp->uid;
pOut->dbVgroup->vgHash = taosHashInit(pRsp->vgNum, taosGetDefaultHashFunction(TSDB_DATA_TYPE_INT), true, HASH_ENTRY_LOCK); pOut->dbVgroup->vgHash = taosHashInit(pRsp->vgNum, taosGetDefaultHashFunction(TSDB_DATA_TYPE_INT), true, HASH_ENTRY_LOCK);
if (NULL == pOut->dbVgroup->vgHash) { if (NULL == pOut->dbVgroup->vgHash) {
qError("taosHashInit %d failed", pRsp->vgNum); qError("taosHashInit %d failed", pRsp->vgNum);
@ -159,6 +159,7 @@ _return:
} }
static int32_t queryConvertTableMetaMsg(STableMetaRsp* pMetaMsg) { static int32_t queryConvertTableMetaMsg(STableMetaRsp* pMetaMsg) {
pMetaMsg->dbId = be64toh(pMetaMsg->dbId);
pMetaMsg->numOfTags = ntohl(pMetaMsg->numOfTags); pMetaMsg->numOfTags = ntohl(pMetaMsg->numOfTags);
pMetaMsg->numOfColumns = ntohl(pMetaMsg->numOfColumns); pMetaMsg->numOfColumns = ntohl(pMetaMsg->numOfColumns);
pMetaMsg->sversion = ntohl(pMetaMsg->sversion); pMetaMsg->sversion = ntohl(pMetaMsg->sversion);
@ -258,6 +259,8 @@ int32_t queryProcessTableMetaRsp(void* output, char *msg, int32_t msgSize) {
} }
strcpy(pOut->dbFName, pMetaMsg->dbFName); strcpy(pOut->dbFName, pMetaMsg->dbFName);
pOut->dbId = pMetaMsg->dbId;
if (pMetaMsg->tableType == TSDB_CHILD_TABLE) { if (pMetaMsg->tableType == TSDB_CHILD_TABLE) {
SET_META_TYPE_BOTH_TABLE(pOut->metaType); SET_META_TYPE_BOTH_TABLE(pOut->metaType);

16
source/libs/sync/CMakeLists.txt
View File

@ -1,16 +0,0 @@
aux_source_directory(src SYNC_SRC)
add_library(sync ${SYNC_SRC})
target_link_libraries(
sync
PUBLIC common
PUBLIC transport
PUBLIC util
PUBLIC wal
)
target_include_directories(
sync
PUBLIC "${CMAKE_SOURCE_DIR}/include/libs/sync"
PRIVATE "${CMAKE_CURRENT_SOURCE_DIR}/inc"
)

146
source/libs/sync/inc/raft.h
View File

@ -1,146 +0,0 @@
/*
* Copyright (c) 2019 TAOS Data, Inc. <cli@taosdata.com>
*
* This program is free software: you can use, redistribute, and/or modify
* it under the terms of the GNU Affero General Public License, version 3
* or later ("AGPL"), as published by the Free Software Foundation.
*
* This program is distributed in the hope that it will be useful, but WITHOUT
* ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or
* FITNESS FOR A PARTICULAR PURPOSE.
*
* You should have received a copy of the GNU Affero General Public License
* along with this program. If not, see <http://www.gnu.org/licenses/>.
*/
#ifndef _TD_LIBS_SYNC_RAFT_H
#define _TD_LIBS_SYNC_RAFT_H
#include "sync.h"
#include "sync_type.h"
#include "thash.h"
#include "raft_message.h"
#include "sync_raft_impl.h"
#include "sync_raft_quorum.h"
typedef struct RaftLeaderState {
} RaftLeaderState;
typedef struct RaftCandidateState {
/* true if in pre-vote phase */
bool inPreVote;
} RaftCandidateState;
typedef struct SSyncRaftIOMethods {
// send SSyncMessage to node
int (*send)(const SSyncMessage* pMsg, const SNodeInfo* pNode);
} SSyncRaftIOMethods;
typedef int (*SyncRaftStepFp)(SSyncRaft* pRaft, const SSyncMessage* pMsg);
typedef void (*SyncRaftTickFp)(SSyncRaft* pRaft);
struct SSyncRaft {
// owner sync node
SSyncNode* pNode;
// hash map nodeId -> SNodeInfo*
SHashObj* nodeInfoMap;
SyncNodeId selfId;
SyncGroupId selfGroupId;
SSyncRaftIOMethods io;
SSyncFSM fsm;
SSyncLogStore logStore;
SStateManager stateManager;
union {
RaftLeaderState leaderState;
RaftCandidateState candidateState;
};
SyncTerm term;
SyncNodeId voteFor;
SSyncRaftLog *log;
uint64_t maxMsgSize;
uint64_t maxUncommittedSize;
SSyncRaftProgressTracker *tracker;
ESyncState state;
// isLearner is true if the local raft node is a learner.
bool isLearner;
/**
* the leader id
**/
SyncNodeId leaderId;
/**
* leadTransferee is id of the leader transfer target when its value is not zero.
* Follow the procedure defined in raft thesis 3.10.
**/
SyncNodeId leadTransferee;
/**
* Only one conf change may be pending (in the log, but not yet
* applied) at a time. This is enforced via pendingConfIndex, which
* is set to a value >= the log index of the latest pending
* configuration change (if any). Config changes are only allowed to
* be proposed if the leader's applied index is greater than this
* value.
**/
SyncIndex pendingConfigIndex;
/**
* an estimate of the size of the uncommitted tail of the Raft log. Used to
* prevent unbounded log growth. Only maintained by the leader. Reset on
* term changes.
**/
uint32_t uncommittedSize;
/**
* number of ticks since it reached last electionTimeout when it is leader
* or candidate.
* number of ticks since it reached last electionTimeout or received a
* valid message from current leader when it is a follower.
**/
uint16_t electionElapsed;
/**
* number of ticks since it reached last heartbeatTimeout.
* only leader keeps heartbeatElapsed.
**/
uint16_t heartbeatElapsed;
bool preVote;
bool checkQuorum;
int heartbeatTimeout;
int electionTimeout;
/**
* randomizedElectionTimeout is a random number between
* [electiontimeout, 2 * electiontimeout - 1]. It gets reset
* when raft changes its state to follower or candidate.
**/
int randomizedElectionTimeout;
bool disableProposalForwarding;
// current tick count since start up
uint32_t currentTick;
SyncRaftStepFp stepFp;
SyncRaftTickFp tickFp;
};
int32_t syncRaftStart(SSyncRaft* pRaft, const SSyncInfo* pInfo);
int32_t syncRaftStep(SSyncRaft* pRaft, const SSyncMessage* pMsg);
int32_t syncRaftTick(SSyncRaft* pRaft);
#endif /* _TD_LIBS_SYNC_RAFT_H */

76
source/libs/sync/inc/raft_log.h
View File

@ -1,76 +0,0 @@
/*
* Copyright (c) 2019 TAOS Data, Inc. <cli@taosdata.com>
*
* This program is free software: you can use, redistribute, and/or modify
* it under the terms of the GNU Affero General Public License, version 3
* or later ("AGPL"), as published by the Free Software Foundation.
*
* This program is distributed in the hope that it will be useful, but WITHOUT
* ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or
* FITNESS FOR A PARTICULAR PURPOSE.
*
* You should have received a copy of the GNU Affero General Public License
* along with this program. If not, see <http://www.gnu.org/licenses/>.
*/
#ifndef _TD_LIBS_SYNC_RAFT_LOG_H
#define _TD_LIBS_SYNC_RAFT_LOG_H
#include "sync.h"
#include "sync_type.h"
typedef enum ESyncRaftEntryType {
SYNC_ENTRY_TYPE_LOG = 1,
} ESyncRaftEntryType;
struct SSyncRaftEntry {
SyncTerm term;
SyncIndex index;
ESyncRaftEntryType type;
SSyncBuffer buffer;
};
struct SSyncRaftLog {
SyncIndex uncommittedConfigIndex;
SyncIndex commitIndex;
SyncIndex appliedIndex;
};
SSyncRaftLog* syncRaftLogOpen();
SyncIndex syncRaftLogLastIndex(SSyncRaftLog* pLog);
SyncIndex syncRaftLogSnapshotIndex(SSyncRaftLog* pLog);
SyncTerm syncRaftLogLastTerm(SSyncRaftLog* pLog);
void syncRaftLogAppliedTo(SSyncRaftLog* pLog, SyncIndex appliedIndex);
bool syncRaftLogIsUptodate(SSyncRaftLog* pLog, SyncIndex index, SyncTerm term);
int syncRaftLogNumOfPendingConf(SSyncRaftLog* pLog);
bool syncRaftHasUnappliedLog(SSyncRaftLog* pLog);
SyncTerm syncRaftLogTermOf(SSyncRaftLog* pLog, SyncIndex index);
int syncRaftLogAppend(SSyncRaftLog* pLog, SSyncRaftEntry *pEntries, int n);
int syncRaftLogAcquire(SSyncRaftLog* pLog, SyncIndex index, int maxMsgSize,
SSyncRaftEntry **ppEntries, int *n);
void syncRaftLogRelease(SSyncRaftLog* pLog, SyncIndex index,
SSyncRaftEntry *pEntries, int n);
bool syncRaftLogMatchTerm();
static FORCE_INLINE bool syncRaftLogIsCommitted(SSyncRaftLog* pLog, SyncIndex index) {
return pLog->commitIndex > index;
}
#endif /* _TD_LIBS_SYNC_RAFT_LOG_H */

237
source/libs/sync/inc/raft_message.h
View File

@ -1,237 +0,0 @@
/*
* Copyright (c) 2019 TAOS Data, Inc. <cli@taosdata.com>
*
* This program is free software: you can use, redistribute, and/or modify
* it under the terms of the GNU Affero General Public License, version 3
* or later ("AGPL"), as published by the Free Software Foundation.
*
* This program is distributed in the hope that it will be useful, but WITHOUT
* ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or
* FITNESS FOR A PARTICULAR PURPOSE.
*
* You should have received a copy of the GNU Affero General Public License
* along with this program. If not, see <http://www.gnu.org/licenses/>.
*/
#ifndef _TD_LIBS_SYNC_RAFT_MESSAGE_H
#define _TD_LIBS_SYNC_RAFT_MESSAGE_H
#include "sync.h"
#include "sync_type.h"
/**
* below define message type which handled by Raft.
*
* internal message, which communicate between threads, start with RAFT_MSG_INTERNAL_*.
* internal message use pointer only and stack memory, need not to be decode/encode and free.
*
* outter message start with RAFT_MSG_*, which communicate between cluster peers,
* need to implement its decode/encode functions.
**/
typedef enum ESyncRaftMessageType {
// client propose a cmd
RAFT_MSG_INTERNAL_PROP = 1,
// node election timeout
RAFT_MSG_INTERNAL_ELECTION = 2,
RAFT_MSG_VOTE = 3,
RAFT_MSG_VOTE_RESP = 4,
RAFT_MSG_APPEND = 5,
RAFT_MSG_APPEND_RESP = 6,
} ESyncRaftMessageType;
typedef struct RaftMsgInternal_Prop {
const SSyncBuffer *pBuf;
bool isWeak;
void* pData;
} RaftMsgInternal_Prop;
typedef struct RaftMsgInternal_Election {
} RaftMsgInternal_Election;
typedef struct RaftMsg_Vote {
ESyncRaftElectionType cType;
SyncIndex lastIndex;
SyncTerm lastTerm;
} RaftMsg_Vote;
typedef struct RaftMsg_VoteResp {
bool rejected;
ESyncRaftElectionType cType;
} RaftMsg_VoteResp;
typedef struct RaftMsg_Append_Entries {
// index of log entry preceeding new ones
SyncIndex index;
// term of entry at prevIndex
SyncTerm term;
// leader's commit index.
SyncIndex commitIndex;
// size of the log entries array
int nEntries;
// log entries array
SSyncRaftEntry* entries;
} RaftMsg_Append_Entries;
typedef struct RaftMsg_Append_Resp {
SyncIndex index;
} RaftMsg_Append_Resp;
typedef struct SSyncMessage {
ESyncRaftMessageType msgType;
SyncTerm term;
SyncGroupId groupId;
SyncNodeId from;
union {
RaftMsgInternal_Prop propose;
RaftMsgInternal_Election election;
RaftMsg_Vote vote;
RaftMsg_VoteResp voteResp;
RaftMsg_Append_Entries appendEntries;
RaftMsg_Append_Resp appendResp;
};
} SSyncMessage;
static FORCE_INLINE SSyncMessage* syncInitPropMsg(SSyncMessage* pMsg, const SSyncBuffer* pBuf, void* pData, bool isWeak) {
*pMsg = (SSyncMessage) {
.msgType = RAFT_MSG_INTERNAL_PROP,
.term = 0,
.propose = (RaftMsgInternal_Prop) {
.isWeak = isWeak,
.pBuf = pBuf,
.pData = pData,
},
};
return pMsg;
}
static FORCE_INLINE SSyncMessage* syncInitElectionMsg(SSyncMessage* pMsg, SyncNodeId from) {
*pMsg = (SSyncMessage) {
.msgType = RAFT_MSG_INTERNAL_ELECTION,
.term = 0,
.from = from,
.election = (RaftMsgInternal_Election) {
},
};
return pMsg;
}
static FORCE_INLINE SSyncMessage* syncNewVoteMsg(SyncGroupId groupId, SyncNodeId from,
SyncTerm term, ESyncRaftElectionType cType,
SyncIndex lastIndex, SyncTerm lastTerm) {
SSyncMessage* pMsg = (SSyncMessage*)malloc(sizeof(SSyncMessage));
if (pMsg == NULL) {
return NULL;
}
*pMsg = (SSyncMessage) {
.groupId = groupId,
.from = from,
.term = term,
.msgType = RAFT_MSG_VOTE,
.vote = (RaftMsg_Vote) {
.cType = cType,
.lastIndex = lastIndex,
.lastTerm = lastTerm,
},
};
return pMsg;
}
static FORCE_INLINE SSyncMessage* syncNewVoteRespMsg(SyncGroupId groupId, SyncNodeId from,
ESyncRaftElectionType cType, bool rejected) {
SSyncMessage* pMsg = (SSyncMessage*)malloc(sizeof(SSyncMessage));
if (pMsg == NULL) {
return NULL;
}
*pMsg = (SSyncMessage) {
.groupId = groupId,
.from = from,
.msgType = RAFT_MSG_VOTE_RESP,
.voteResp = (RaftMsg_VoteResp) {
.cType = cType,
.rejected = rejected,
},
};
return pMsg;
}
static FORCE_INLINE SSyncMessage* syncNewAppendMsg(SyncGroupId groupId, SyncNodeId from,
SyncTerm term, SyncIndex logIndex, SyncTerm logTerm,
SyncIndex commitIndex, int nEntries, SSyncRaftEntry* entries) {
SSyncMessage* pMsg = (SSyncMessage*)malloc(sizeof(SSyncMessage));
if (pMsg == NULL) {
return NULL;
}
*pMsg = (SSyncMessage) {
.groupId = groupId,
.from = from,
.term = term,
.msgType = RAFT_MSG_APPEND,
.appendEntries = (RaftMsg_Append_Entries) {
.index = logIndex,
.term = logTerm,
.commitIndex = commitIndex,
.nEntries = nEntries,
.entries = entries,
},
};
return pMsg;
}
static FORCE_INLINE SSyncMessage* syncNewEmptyAppendRespMsg(SyncGroupId groupId, SyncNodeId from, SyncTerm term) {
SSyncMessage* pMsg = (SSyncMessage*)malloc(sizeof(SSyncMessage));
if (pMsg == NULL) {
return NULL;
}
*pMsg = (SSyncMessage) {
.groupId = groupId,
.from = from,
.term = term,
.msgType = RAFT_MSG_APPEND_RESP,
.appendResp = (RaftMsg_Append_Resp) {
},
};
return pMsg;
}
static FORCE_INLINE bool syncIsInternalMsg(ESyncRaftMessageType msgType) {
return msgType == RAFT_MSG_INTERNAL_PROP ||
msgType == RAFT_MSG_INTERNAL_ELECTION;
}
static FORCE_INLINE bool syncIsPreVoteRespMsg(const SSyncMessage* pMsg) {
return pMsg->msgType == RAFT_MSG_VOTE_RESP && pMsg->voteResp.cType == SYNC_RAFT_CAMPAIGN_PRE_ELECTION;
}
static FORCE_INLINE bool syncIsPreVoteMsg(const SSyncMessage* pMsg) {
return pMsg->msgType == RAFT_MSG_VOTE && pMsg->voteResp.cType == SYNC_RAFT_CAMPAIGN_PRE_ELECTION;
}
void syncFreeMessage(const SSyncMessage* pMsg);
// message handlers
int syncRaftHandleElectionMessage(SSyncRaft* pRaft, const SSyncMessage* pMsg);
int syncRaftHandleVoteMessage(SSyncRaft* pRaft, const SSyncMessage* pMsg);
int syncRaftHandleVoteRespMessage(SSyncRaft* pRaft, const SSyncMessage* pMsg);
int syncRaftHandleAppendEntriesMessage(SSyncRaft* pRaft, const SSyncMessage* pMsg);
#endif /* _TD_LIBS_SYNC_RAFT_MESSAGE_H */

30
source/libs/sync/inc/raft_replication.h
View File

@ -1,30 +0,0 @@
/*
* Copyright (c) 2019 TAOS Data, Inc. <cli@taosdata.com>
*
* This program is free software: you can use, redistribute, and/or modify
* it under the terms of the GNU Affero General Public License, version 3
* or later ("AGPL"), as published by the Free Software Foundation.
*
* This program is distributed in the hope that it will be useful, but WITHOUT
* ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or
* FITNESS FOR A PARTICULAR PURPOSE.
*
* You should have received a copy of the GNU Affero General Public License
* along with this program. If not, see <http://www.gnu.org/licenses/>.
*/
#ifndef TD_SYNC_RAFT_REPLICATION_H
#define TD_SYNC_RAFT_REPLICATION_H
#include "sync.h"
#include "syncInt.h"
#include "sync_type.h"
// syncRaftMaybeSendAppend sends an append RPC with new entries to the given peer,
// if necessary. Returns true if a message was sent. The sendIfEmpty
// argument controls whether messages with no entries will be sent
// ("empty" messages are useful to convey updated Commit indexes, but
// are undesirable when we're sending multiple messages in a batch).
bool syncRaftMaybeSendAppend(SSyncRaft* pRaft, SSyncRaftProgress* progress, bool sendIfEmpty);
#endif /* TD_SYNC_RAFT_REPLICATION_H */

115
source/libs/sync/inc/raft_unstable_log.h
View File

@ -1,115 +0,0 @@
/*
* Copyright (c) 2019 TAOS Data, Inc. <cli@taosdata.com>
*
* This program is free software: you can use, redistribute, and/or modify
* it under the terms of the GNU Affero General Public License, version 3
* or later ("AGPL"), as published by the Free Software Foundation.
*
* This program is distributed in the hope that it will be useful, but WITHOUT
* ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or
* FITNESS FOR A PARTICULAR PURPOSE.
*
* You should have received a copy of the GNU Affero General Public License
* along with this program. If not, see <http://www.gnu.org/licenses/>.
*/
#ifndef TD_SYNC_RAFT_UNSTABLE_LOG_H
#define TD_SYNC_RAFT_UNSTABLE_LOG_H
#include "sync_type.h"
/* in-memory unstable raft log storage */
struct SSyncRaftUnstableLog {
#if 0
/* Circular buffer of log entries */
RaftEntry *entries;
/* size of Circular buffer */
int size;
/* Indexes of used slots [front, back) */
int front, back;
/* Index of first entry is offset + 1 */
SyncIndex offset;
/* meta data of snapshot */
SSyncRaftUnstableLog snapshot;
#endif
};
/**
* return index of last in memory log, return 0 if log is empty
**/
//SyncIndex syncRaftLogLastIndex(SSyncRaftUnstableLog* pLog);
#if 0
void raftLogInit(RaftLog* pLog);
void raftLogClose(RaftLog* pLog);
/**
* When startup populating log entrues loaded from disk,
* init raft memory log with snapshot index,term and log start idnex.
**/
/*
void raftLogStart(RaftLog* pLog,
RaftSnapshotMeta snapshot,
SyncIndex startIndex);
*/
/**
* Get the number of entries the log.
**/
int raftLogNumEntries(const RaftLog* pLog);
/**
* return last term of in memory log, return 0 if log is empty
**/
SyncTerm raftLogLastTerm(RaftLog* pLog);
/**
* return term of log with the given index, return 0 if the term of index cannot be found
* , errCode will save the error code.
**/
SyncTerm raftLogTermOf(RaftLog* pLog, SyncIndex index, RaftCode* errCode);
/**
* Get the last index of the most recent snapshot. Return 0 if there are no *
* snapshots.
**/
SyncIndex raftLogSnapshotIndex(RaftLog* pLog);
/* Append a new entry to the log. */
int raftLogAppend(RaftLog* pLog,
SyncTerm term,
const SSyncBuffer *buf);
/**
* acquire log from given index onwards.
**/
/*
int raftLogAcquire(RaftLog* pLog,
SyncIndex index,
RaftEntry **ppEntries,
int *n);
void raftLogRelease(RaftLog* pLog,
SyncIndex index,
RaftEntry *pEntries,
int n);
*/
/* Delete all entries from the given index (included) onwards. */
void raftLogTruncate(RaftLog* pLog, SyncIndex index);
/**
* when taking a new snapshot, the function will update the last snapshot information and delete
* all entries up last_index - trailing (included). If the log contains no entry
* a last_index - trailing, then no entry will be deleted.
**/
void raftLogSnapshot(RaftLog* pLog, SyncIndex index, SyncIndex trailing);
#endif
#endif /* TD_SYNC_RAFT_UNSTABLE_LOG_H */

71
source/libs/sync/inc/syncInt.h
View File

@ -1,71 +0,0 @@
/*
* Copyright (c) 2019 TAOS Data, Inc. <cli@taosdata.com>
*
* This program is free software: you can use, redistribute, and/or modify
* it under the terms of the GNU Affero General Public License, version 3
* or later ("AGPL"), as published by the Free Software Foundation.
*
* This program is distributed in the hope that it will be useful, but WITHOUT
* ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or
* FITNESS FOR A PARTICULAR PURPOSE.
*
* You should have received a copy of the GNU Affero General Public License
* along with this program. If not, see <http://www.gnu.org/licenses/>.
*/
#ifndef _TD_LIBS_SYNC_INT_H
#define _TD_LIBS_SYNC_INT_H
#include "thash.h"
#include "os.h"
#include "sync.h"
#include "sync_type.h"
#include "raft.h"
#include "tlog.h"
#define TAOS_SYNC_MAX_WORKER 3
typedef struct SSyncWorker {
pthread_t thread;
} SSyncWorker;
struct SSyncNode {
pthread_mutex_t mutex;
int32_t refCount;
SyncGroupId vgId;
SSyncRaft raft;
void* syncTimer;
};
typedef struct SSyncManager {
pthread_mutex_t mutex;
// sync server rpc
void* serverRpc;
// rpc server hash table base on FQDN:port key
SHashObj* rpcServerTable;
// sync client rpc
void* clientRpc;
// worker threads
SSyncWorker worker[TAOS_SYNC_MAX_WORKER];
// vgroup hash table
SHashObj* vgroupTable;
// timer manager
void* syncTimerManager;
} SSyncManager;
extern SSyncManager* gSyncManager;
#define syncFatal(...) do { if (sDebugFlag & DEBUG_FATAL) { taosPrintLog("SYNC FATAL ", 255, __VA_ARGS__); }} while(0)
#define syncError(...) do { if (sDebugFlag & DEBUG_ERROR) { taosPrintLog("SYNC ERROR ", 255, __VA_ARGS__); }} while(0)
#define syncWarn(...) do { if (sDebugFlag & DEBUG_WARN) { taosPrintLog("SYNC WARN ", 255, __VA_ARGS__); }} while(0)
#define syncInfo(...) do { if (sDebugFlag & DEBUG_INFO) { taosPrintLog("SYNC ", 255, __VA_ARGS__); }} while(0)
#define syncDebug(...) do { if (sDebugFlag & DEBUG_DEBUG) { taosPrintLog("SYNC ", sDebugFlag, __VA_ARGS__); }} while(0)
#define syncTrace(...) do { if (sDebugFlag & DEBUG_TRACE) { taosPrintLog("SYNC ", sDebugFlag, __VA_ARGS__); }} while(0)
#endif /* _TD_LIBS_SYNC_INT_H */

25
source/libs/sync/inc/sync_const.h
View File

@ -1,25 +0,0 @@
/*
* Copyright (c) 2019 TAOS Data, Inc. <cli@taosdata.com>
*
* This program is free software: you can use, redistribute, and/or modify
* it under the terms of the GNU Affero General Public License, version 3
* or later ("AGPL"), as published by the Free Software Foundation.
*
* This program is distributed in the hope that it will be useful, but WITHOUT
* ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or
* FITNESS FOR A PARTICULAR PURPOSE.
*
* You should have received a copy of the GNU Affero General Public License
* along with this program. If not, see <http://www.gnu.org/licenses/>.
*/
#ifndef _TD_LIBS_SYNC_CONST_H
#define _TD_LIBS_SYNC_CONST_H
#include "sync.h"
static int kSyncRaftMaxInflghtMsgs = 20;
static SyncIndex kMaxCommitIndex = UINT64_MAX;
#endif /* _TD_LIBS_SYNC_CONST_H */

47
source/libs/sync/inc/sync_raft_config_change.h
View File

@ -1,47 +0,0 @@
/*
* Copyright (c) 2019 TAOS Data, Inc. <cli@taosdata.com>
*
* This program is free software: you can use, redistribute, and/or modify
* it under the terms of the GNU Affero General Public License, version 3
* or later ("AGPL"), as published by the Free Software Foundation.
*
* This program is distributed in the hope that it will be useful, but WITHOUT
* ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or
* FITNESS FOR A PARTICULAR PURPOSE.
*
* You should have received a copy of the GNU Affero General Public License
* along with this program. If not, see <http://www.gnu.org/licenses/>.
*/
#ifndef TD_SYNC_RAFT_CONFIG_CHANGE_H
#define TD_SYNC_RAFT_CONFIG_CHANGE_H
#include "sync_type.h"
#include "sync_raft_proto.h"
/**
* Changer facilitates configuration changes. It exposes methods to handle
* simple and joint consensus while performing the proper validation that allows
* refusing invalid configuration changes before they affect the active
* configuration.
**/
struct SSyncRaftChanger {
SSyncRaftProgressTracker* tracker;
SyncIndex lastIndex;
};
typedef int (*configChangeFp)(SSyncRaftChanger* changer, const SSyncConfChangeSingleArray* css,
SSyncRaftProgressTrackerConfig* config, SSyncRaftProgressMap* progressMap);
// Simple carries out a series of configuration changes that (in aggregate)
// mutates the incoming majority config Voters[0] by at most one. This method
// will return an error if that is not the case, if the resulting quorum is
// zero, or if the configuration is in a joint state (i.e. if there is an
// outgoing configuration).
int syncRaftChangerSimpleConfig(SSyncRaftChanger* changer, const SSyncConfChangeSingleArray* css,
SSyncRaftProgressTrackerConfig* config, SSyncRaftProgressMap* progressMap);
int syncRaftChangerEnterJoint(SSyncRaftChanger* changer, bool autoLeave, const SSyncConfChangeSingleArray* css,
SSyncRaftProgressTrackerConfig* config, SSyncRaftProgressMap* progressMap);
#endif /* TD_SYNC_RAFT_CONFIG_CHANGE_H */

58
source/libs/sync/inc/sync_raft_impl.h
View File

@ -1,58 +0,0 @@
/*
* Copyright (c) 2019 TAOS Data, Inc. <cli@taosdata.com>
*
* This program is free software: you can use, redistribute, and/or modify
* it under the terms of the GNU Affero General Public License, version 3
* or later ("AGPL"), as published by the Free Software Foundation.
*
* This program is distributed in the hope that it will be useful, but WITHOUT
* ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or
* FITNESS FOR A PARTICULAR PURPOSE.
*
* You should have received a copy of the GNU Affero General Public License
* along with this program. If not, see <http://www.gnu.org/licenses/>.
*/
#ifndef _TD_LIBS_SYNC_RAFT_IMPL_H
#define _TD_LIBS_SYNC_RAFT_IMPL_H
#include "sync.h"
#include "sync_type.h"
#include "raft_message.h"
#include "sync_raft_quorum.h"
void syncRaftBecomeFollower(SSyncRaft* pRaft, SyncTerm term, SyncNodeId leaderId);
void syncRaftBecomePreCandidate(SSyncRaft* pRaft);
void syncRaftBecomeCandidate(SSyncRaft* pRaft);
void syncRaftBecomeLeader(SSyncRaft* pRaft);
void syncRaftStartElection(SSyncRaft* pRaft, ESyncRaftElectionType cType);
void syncRaftCampaign(SSyncRaft* pRaft, ESyncRaftElectionType cType);
void syncRaftTriggerHeartbeat(SSyncRaft* pRaft);
void syncRaftRandomizedElectionTimeout(SSyncRaft* pRaft);
bool syncRaftIsPromotable(SSyncRaft* pRaft);
bool syncRaftIsPastElectionTimeout(SSyncRaft* pRaft);
int syncRaftQuorum(SSyncRaft* pRaft);
bool syncRaftMaybeCommit(SSyncRaft* pRaft);
ESyncRaftVoteResult syncRaftPollVote(SSyncRaft* pRaft, SyncNodeId id,
bool preVote, bool accept,
int* rejectNum, int *granted);
static FORCE_INLINE bool syncRaftIsEmptyServerState(const SSyncServerState* serverState) {
return serverState->commitIndex == 0 &&
serverState->term == SYNC_NON_TERM &&
serverState->voteFor == SYNC_NON_NODE_ID;
}
void syncRaftLoadState(SSyncRaft* pRaft, const SSyncServerState* serverState);
void syncRaftBroadcastAppend(SSyncRaft* pRaft);
SNodeInfo* syncRaftGetNodeById(SSyncRaft *pRaft, SyncNodeId id);
#endif /* _TD_LIBS_SYNC_RAFT_IMPL_H */

70
source/libs/sync/inc/sync_raft_inflights.h
View File

@ -1,70 +0,0 @@
/*
* Copyright (c) 2019 TAOS Data, Inc. <cli@taosdata.com>
*
* This program is free software: you can use, redistribute, and/or modify
* it under the terms of the GNU Affero General Public License, version 3
* or later ("AGPL"), as published by the Free Software Foundation.
*
* This program is distributed in the hope that it will be useful, but WITHOUT
* ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or
* FITNESS FOR A PARTICULAR PURPOSE.
*
* You should have received a copy of the GNU Affero General Public License
* along with this program. If not, see <http: *www.gnu.org/licenses/>.
*/
#ifndef TD_SYNC_RAFT_INFLIGHTS_H
#define TD_SYNC_RAFT_INFLIGHTS_H
#include "sync.h"
// Inflights limits the number of MsgApp (represented by the largest index
// contained within) sent to followers but not yet acknowledged by them. Callers
// use Full() to check whether more messages can be sent, call Add() whenever
// they are sending a new append, and release "quota" via FreeLE() whenever an
// ack is received.
typedef struct SSyncRaftInflights {
// the starting index in the buffer
int start;
// number of inflights in the buffer
int count;
// the size of the buffer
int size;
// buffer contains the index of the last entry
// inside one message.
SyncIndex* buffer;
} SSyncRaftInflights;
SSyncRaftInflights* syncRaftOpenInflights(int size);
void syncRaftCloseInflights(SSyncRaftInflights*);
// reset frees all inflights.
static FORCE_INLINE void syncRaftInflightReset(SSyncRaftInflights* inflights) {
inflights->count = 0;
inflights->start = 0;
}
// Full returns true if no more messages can be sent at the moment.
static FORCE_INLINE bool syncRaftInflightFull(SSyncRaftInflights* inflights) {
return inflights->count == inflights->size;
}
// Add notifies the Inflights that a new message with the given index is being
// dispatched. Full() must be called prior to Add() to verify that there is room
// for one more message, and consecutive calls to add Add() must provide a
// monotonic sequence of indexes.
void syncRaftInflightAdd(SSyncRaftInflights* inflights, SyncIndex inflightIndex);
// FreeLE frees the inflights smaller or equal to the given `to` flight.
void syncRaftInflightFreeLE(SSyncRaftInflights* inflights, SyncIndex toIndex);
/**
* syncRaftInflightFreeFirstOne releases the first inflight.
* This is a no-op if nothing is inflight.
**/
void syncRaftInflightFreeFirstOne(SSyncRaftInflights* inflights);
#endif /* TD_SYNC_RAFT_INFLIGHTS_H */

49
source/libs/sync/inc/sync_raft_node_map.h
View File

@ -1,49 +0,0 @@
/*
* Copyright (c) 2019 TAOS Data, Inc. <cli@taosdata.com>
*
* This program is free software: you can use, redistribute, and/or modify
* it under the terms of the GNU Affero General Public License, version 3
* or later ("AGPL"), as published by the Free Software Foundation.
*
* This program is distributed in the hope that it will be useful, but WITHOUT
* ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or
* FITNESS FOR A PARTICULAR PURPOSE.
*
* You should have received a copy of the GNU Affero General Public License
* along with this program. If not, see <http://www.gnu.org/licenses/>.
*/
#ifndef _TD_LIBS_SYNC_RAFT_NODE_MAP_H
#define _TD_LIBS_SYNC_RAFT_NODE_MAP_H
#include "thash.h"
#include "sync.h"
#include "sync_type.h"
struct SSyncRaftNodeMap {
SHashObj* nodeIdMap;
};
void syncRaftInitNodeMap(SSyncRaftNodeMap* nodeMap);
void syncRaftFreeNodeMap(SSyncRaftNodeMap* nodeMap);
void syncRaftClearNodeMap(SSyncRaftNodeMap* nodeMap);
bool syncRaftIsInNodeMap(const SSyncRaftNodeMap* nodeMap, SyncNodeId nodeId);
void syncRaftCopyNodeMap(SSyncRaftNodeMap* from, SSyncRaftNodeMap* to);
void syncRaftUnionNodeMap(SSyncRaftNodeMap* nodeMap, SSyncRaftNodeMap* to);
void syncRaftAddToNodeMap(SSyncRaftNodeMap* nodeMap, SyncNodeId nodeId);
void syncRaftRemoveFromNodeMap(SSyncRaftNodeMap* nodeMap, SyncNodeId nodeId);
int32_t syncRaftNodeMapSize(const SSyncRaftNodeMap* nodeMap);
// return true if reach the end
bool syncRaftIterateNodeMap(const SSyncRaftNodeMap* nodeMap, SyncNodeId *pId);
bool syncRaftIsAllNodeInProgressMap(SSyncRaftNodeMap* nodeMap, SSyncRaftProgressMap* progressMap);
#endif /* _TD_LIBS_SYNC_RAFT_NODE_MAP_H */

259
source/libs/sync/inc/sync_raft_progress.h
View File

@ -1,259 +0,0 @@
/*
* Copyright (c) 2019 TAOS Data, Inc. <cli@taosdata.com>
*
* This program is free software: you can use, redistribute, and/or modify
* it under the terms of the GNU Affero General Public License, version 3
* or later ("AGPL"), as published by the Free Software Foundation.
*
* This program is distributed in the hope that it will be useful, but WITHOUT
* ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or
* FITNESS FOR A PARTICULAR PURPOSE.
*
* You should have received a copy of the GNU Affero General Public License
* along with this program. If not, see <http: *www.gnu.org/licenses/>.
*/
#ifndef TD_SYNC_RAFT_PROGRESS_H
#define TD_SYNC_RAFT_PROGRESS_H
#include "sync_type.h"
#include "sync_raft_inflights.h"
#include "thash.h"
/**
* State defines how the leader should interact with the follower.
*
* When in PROGRESS_STATE_PROBE, leader sends at most one replication message
* per heartbeat interval. It also probes actual progress of the follower.
*
* When in PROGRESS_STATE_REPLICATE, leader optimistically increases next
* to the latest entry sent after sending replication message. This is
* an optimized state for fast replicating log entries to the follower.
*
* When in PROGRESS_STATE_SNAPSHOT, leader should have sent out snapshot
* before and stops sending any replication message.
*
* PROGRESS_STATE_PROBE is the initial state.
**/
typedef enum ESyncRaftProgressState {
/**
* StateProbe indicates a follower whose last index isn't known. Such a
* follower is "probed" (i.e. an append sent periodically) to narrow down
* its last index. In the ideal (and common) case, only one round of probing
* is necessary as the follower will react with a hint. Followers that are
* probed over extended periods of time are often offline.
**/
PROGRESS_STATE_PROBE = 0,
/**
* StateReplicate is the state steady in which a follower eagerly receives
* log entries to append to its log.
**/
PROGRESS_STATE_REPLICATE,
/**
* StateSnapshot indicates a follower that needs log entries not available
* from the leader's Raft log. Such a follower needs a full snapshot to
* return to StateReplicate.
**/
PROGRESS_STATE_SNAPSHOT,
} ESyncRaftProgressState;
static const char* kProgressStateString[] = {
"Probe",
"Replicate",
"Snapshot",
};
// Progress represents a follower’s progress in the view of the leader. Leader
// maintains progresses of all followers, and sends entries to the follower
// based on its progress.
//
// NB(tbg): Progress is basically a state machine whose transitions are mostly
// strewn around `*raft.raft`. Additionally, some fields are only used when in a
// certain State. All of this isn't ideal.
struct SSyncRaftProgress {
SyncGroupId groupId;
SyncNodeId id;
int16_t refCount;
SyncIndex nextIndex;
SyncIndex matchIndex;
// State defines how the leader should interact with the follower.
//
// When in StateProbe, leader sends at most one replication message
// per heartbeat interval. It also probes actual progress of the follower.
//
// When in StateReplicate, leader optimistically increases next
// to the latest entry sent after sending replication message. This is
// an optimized state for fast replicating log entries to the follower.
//
// When in StateSnapshot, leader should have sent out snapshot
// before and stops sending any replication message.
ESyncRaftProgressState state;
// PendingSnapshot is used in StateSnapshot.
// If there is a pending snapshot, the pendingSnapshot will be set to the
// index of the snapshot. If pendingSnapshot is set, the replication process of
// this Progress will be paused. raft will not resend snapshot until the pending one
// is reported to be failed.
SyncIndex pendingSnapshotIndex;
// RecentActive is true if the progress is recently active. Receiving any messages
// from the corresponding follower indicates the progress is active.
// RecentActive can be reset to false after an election timeout.
//
// TODO(tbg): the leader should always have this set to true.
bool recentActive;
// ProbeSent is used while this follower is in StateProbe. When ProbeSent is
// true, raft should pause sending replication message to this peer until
// ProbeSent is reset. See ProbeAcked() and IsPaused().
bool probeSent;
// Inflights is a sliding window for the inflight messages.
// Each inflight message contains one or more log entries.
// The max number of entries per message is defined in raft config as MaxSizePerMsg.
// Thus inflight effectively limits both the number of inflight messages
// and the bandwidth each Progress can use.
// When inflights is Full, no more message should be sent.
// When a leader sends out a message, the index of the last
// entry should be added to inflights. The index MUST be added
// into inflights in order.
// When a leader receives a reply, the previous inflights should
// be freed by calling inflights.FreeLE with the index of the last
// received entry.
SSyncRaftInflights* inflights;
// IsLearner is true if this progress is tracked for a learner.
bool isLearner;
};
struct SSyncRaftProgressMap {
// map nodeId -> SSyncRaftProgress*
SHashObj* progressMap;
};
static FORCE_INLINE const char* syncRaftProgressStateString(const SSyncRaftProgress* progress) {
return kProgressStateString[progress->state];
}
void syncRaftResetProgress(SSyncRaft* pRaft, SSyncRaftProgress* progress);
// BecomeProbe transitions into StateProbe. Next is reset to Match+1 or,
// optionally and if larger, the index of the pending snapshot.
void syncRaftProgressBecomeProbe(SSyncRaftProgress* progress);
// BecomeReplicate transitions into StateReplicate, resetting Next to Match+1.
void syncRaftProgressBecomeReplicate(SSyncRaftProgress* progress);
// MaybeUpdate is called when an MsgAppResp arrives from the follower, with the
// index acked by it. The method returns false if the given n index comes from
// an outdated message. Otherwise it updates the progress and returns true.
bool syncRaftProgressMaybeUpdate(SSyncRaftProgress* progress, SyncIndex lastIndex);
// OptimisticUpdate signals that appends all the way up to and including index n
// are in-flight. As a result, Next is increased to n+1.
static FORCE_INLINE void syncRaftProgressOptimisticNextIndex(SSyncRaftProgress* progress, SyncIndex nextIndex) {
progress->nextIndex = nextIndex + 1;
}
// MaybeDecrTo adjusts the Progress to the receipt of a MsgApp rejection. The
// arguments are the index of the append message rejected by the follower, and
// the hint that we want to decrease to.
//
// Rejections can happen spuriously as messages are sent out of order or
// duplicated. In such cases, the rejection pertains to an index that the
// Progress already knows were previously acknowledged, and false is returned
// without changing the Progress.
//
// If the rejection is genuine, Next is lowered sensibly, and the Progress is
// cleared for sending log entries.
bool syncRaftProgressMaybeDecrTo(SSyncRaftProgress* progress,
SyncIndex rejected, SyncIndex matchHint);
// IsPaused returns whether sending log entries to this node has been throttled.
// This is done when a node has rejected recent MsgApps, is currently waiting
// for a snapshot, or has reached the MaxInflightMsgs limit. In normal
// operation, this is false. A throttled node will be contacted less frequently
// until it has reached a state in which it's able to accept a steady stream of
// log entries again.
bool syncRaftProgressIsPaused(SSyncRaftProgress* progress);
static FORCE_INLINE SyncIndex syncRaftProgressNextIndex(SSyncRaftProgress* progress) {
return progress->nextIndex;
}
static FORCE_INLINE ESyncRaftProgressState syncRaftProgressInReplicate(SSyncRaftProgress* progress) {
return progress->state == PROGRESS_STATE_REPLICATE;
}
static FORCE_INLINE ESyncRaftProgressState syncRaftProgressInSnapshot(SSyncRaftProgress* progress) {
return progress->state == PROGRESS_STATE_SNAPSHOT;
}
static FORCE_INLINE ESyncRaftProgressState syncRaftProgressInProbe(SSyncRaftProgress* progress) {
return progress->state == PROGRESS_STATE_PROBE;
}
static FORCE_INLINE bool syncRaftProgressRecentActive(SSyncRaftProgress* progress) {
return progress->recentActive;
}
void syncRaftInitProgressMap(SSyncRaftProgressMap* progressMap);
void syncRaftFreeProgressMap(SSyncRaftProgressMap* progressMap);
void syncRaftClearProgressMap(SSyncRaftProgressMap* progressMap);
void syncRaftCopyProgressMap(SSyncRaftProgressMap* from, SSyncRaftProgressMap* to);
SSyncRaftProgress* syncRaftFindProgressByNodeId(const SSyncRaftProgressMap* progressMap, SyncNodeId id);
int syncRaftAddToProgressMap(SSyncRaftProgressMap* progressMap, SSyncRaftProgress* progress);
void syncRaftRemoveFromProgressMap(SSyncRaftProgressMap* progressMap, SyncNodeId id);
bool syncRaftIsInProgressMap(SSyncRaftProgressMap* progressMap, SyncNodeId id);
/**
* return true if progress's log is up-todate
**/
bool syncRaftProgressIsUptodate(SSyncRaft* pRaft, SSyncRaftProgress* progress);
// BecomeSnapshot moves the Progress to StateSnapshot with the specified pending
// snapshot index.
void syncRaftProgressBecomeSnapshot(SSyncRaftProgress* progress, SyncIndex snapshotIndex);
void syncRaftCopyProgress(const SSyncRaftProgress* from, SSyncRaftProgress* to);
// return true if reach the end
bool syncRaftIterateProgressMap(const SSyncRaftProgressMap* progressMap, SSyncRaftProgress *pProgress);
bool syncRaftVisitProgressMap(SSyncRaftProgressMap* progressMap, visitProgressFp fp, void* arg);
#if 0
void syncRaftProgressAbortSnapshot(SSyncRaft* pRaft, int i);
SyncIndex syncRaftProgressMatchIndex(SSyncRaft* pRaft, int i);
void syncRaftProgressUpdateLastSend(SSyncRaft* pRaft, int i);
void syncRaftProgressUpdateSnapshotLastSend(SSyncRaft* pRaft, int i);
bool syncRaftProgressResetRecentRecv(SSyncRaft* pRaft, int i);
void syncRaftProgressMarkRecentRecv(SSyncRaft* pRaft, int i);
void syncRaftProgressAbortSnapshot(SSyncRaft* pRaft, int i);
#endif
#endif /* TD_SYNC_RAFT_PROGRESS_H */

132
source/libs/sync/inc/sync_raft_progress_tracker.h
View File

@ -1,132 +0,0 @@
/*
* Copyright (c) 2019 TAOS Data, Inc. <cli@taosdata.com>
*
* This program is free software: you can use, redistribute, and/or modify
* it under the terms of the GNU Affero General Public License, version 3
* or later ("AGPL"), as published by the Free Software Foundation.
*
* This program is distributed in the hope that it will be useful, but WITHOUT
* ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or
* FITNESS FOR A PARTICULAR PURPOSE.
*
* You should have received a copy of the GNU Affero General Public License
* along with this program. If not, see <http://www.gnu.org/licenses/>.
*/
#ifndef _TD_LIBS_SYNC_RAFT_PROGRESS_TRACKER_H
#define _TD_LIBS_SYNC_RAFT_PROGRESS_TRACKER_H
#include "sync_type.h"
#include "sync_raft_quorum.h"
#include "sync_raft_quorum_joint.h"
#include "sync_raft_progress.h"
#include "sync_raft_proto.h"
#include "thash.h"
// Config reflects the configuration tracked in a ProgressTracker.
struct SSyncRaftProgressTrackerConfig {
SSyncRaftQuorumJointConfig voters;
// autoLeave is true if the configuration is joint and a transition to the
// incoming configuration should be carried out automatically by Raft when
// this is possible. If false, the configuration will be joint until the
// application initiates the transition manually.
bool autoLeave;
// Learners is a set of IDs corresponding to the learners active in the
// current configuration.
//
// Invariant: Learners and Voters does not intersect, i.e. if a peer is in
// either half of the joint config, it can't be a learner; if it is a
// learner it can't be in either half of the joint config. This invariant
// simplifies the implementation since it allows peers to have clarity about
// its current role without taking into account joint consensus.
SSyncRaftNodeMap learners;
// When we turn a voter into a learner during a joint consensus transition,
// we cannot add the learner directly when entering the joint state. This is
// because this would violate the invariant that the intersection of
// voters and learners is empty. For example, assume a Voter is removed and
// immediately re-added as a learner (or in other words, it is demoted):
//
// Initially, the configuration will be
//
// voters: {1 2 3}
// learners: {}
//
// and we want to demote 3. Entering the joint configuration, we naively get
//
// voters: {1 2} & {1 2 3}
// learners: {3}
//
// but this violates the invariant (3 is both voter and learner). Instead,
// we get
//
// voters: {1 2} & {1 2 3}
// learners: {}
// next_learners: {3}
//
// Where 3 is now still purely a voter, but we are remembering the intention
// to make it a learner upon transitioning into the final configuration:
//
// voters: {1 2}
// learners: {3}
// next_learners: {}
//
// Note that next_learners is not used while adding a learner that is not
// also a voter in the joint config. In this case, the learner is added
// right away when entering the joint configuration, so that it is caught up
// as soon as possible.
SSyncRaftNodeMap learnersNext;
};
struct SSyncRaftProgressTracker {
SSyncRaftProgressTrackerConfig config;
SSyncRaftProgressMap progressMap;
// nodeid -> ESyncRaftVoteType map
SHashObj* votesMap;
int maxInflightMsgs;
SSyncRaft* pRaft;
};
SSyncRaftProgressTracker* syncRaftOpenProgressTracker(SSyncRaft* pRaft);
void syncRaftInitTrackConfig(SSyncRaftProgressTrackerConfig* config);
void syncRaftFreeTrackConfig(SSyncRaftProgressTrackerConfig* config);
void syncRaftFreeTrackConfig(SSyncRaftProgressTrackerConfig* config);
// ResetVotes prepares for a new round of vote counting via recordVote.
void syncRaftResetVotes(SSyncRaftProgressTracker*);
void syncRaftProgressVisit(SSyncRaftProgressTracker*, visitProgressFp visit, void* arg);
// RecordVote records that the node with the given id voted for this Raft
// instance if v == true (and declined it otherwise).
void syncRaftRecordVote(SSyncRaftProgressTracker* tracker, SyncNodeId id, bool grant);
void syncRaftCopyTrackerConfig(const SSyncRaftProgressTrackerConfig* from, SSyncRaftProgressTrackerConfig* to);
int syncRaftCheckTrackerConfigInProgress(SSyncRaftProgressTrackerConfig* config, SSyncRaftProgressMap* progressMap);
// TallyVotes returns the number of granted and rejected Votes, and whether the
// election outcome is known.
ESyncRaftVoteResult syncRaftTallyVotes(SSyncRaftProgressTracker* tracker, int* rejected, int *granted);
void syncRaftConfigState(SSyncRaftProgressTracker* tracker, SSyncConfigState* cs);
// Committed returns the largest log index known to be committed based on what
// the voting members of the group have acknowledged.
SyncIndex syncRaftCommittedIndex(SSyncRaftProgressTracker* tracker);
// QuorumActive returns true if the quorum is active from the view of the local
// raft state machine. Otherwise, it returns false.
bool syncRaftQuorumActive(SSyncRaftProgressTracker* tracker);
bool syncRaftIsInNodeMap(const SSyncRaftNodeMap* nodeMap, SyncNodeId nodeId);
#endif /* _TD_LIBS_SYNC_RAFT_PROGRESS_TRACKER_H */

77
source/libs/sync/inc/sync_raft_proto.h
View File

@ -1,77 +0,0 @@
/*
* Copyright (c) 2019 TAOS Data, Inc. <cli@taosdata.com>
*
* This program is free software: you can use, redistribute, and/or modify
* it under the terms of the GNU Affero General Public License, version 3
* or later ("AGPL"), as published by the Free Software Foundation.
*
* This program is distributed in the hope that it will be useful, but WITHOUT
* ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or
* FITNESS FOR A PARTICULAR PURPOSE.
*
* You should have received a copy of the GNU Affero General Public License
* along with this program. If not, see <http://www.gnu.org/licenses/>.
*/
#ifndef TD_SYNC_RAFT_PROTO_H
#define TD_SYNC_RAFT_PROTO_H
#include "sync_type.h"
#include "sync_raft_node_map.h"
typedef enum ESyncRaftConfChangeType {
SYNC_RAFT_Conf_AddNode = 0,
SYNC_RAFT_Conf_RemoveNode = 1,
SYNC_RAFT_Conf_UpdateNode = 2,
SYNC_RAFT_Conf_AddLearnerNode = 3,
} ESyncRaftConfChangeType;
// ConfChangeSingle is an individual configuration change operation. Multiple
// such operations can be carried out atomically via a ConfChangeV2.
typedef struct SSyncConfChangeSingle {
ESyncRaftConfChangeType type;
SyncNodeId nodeId;
} SSyncConfChangeSingle;
typedef struct SSyncConfChangeSingleArray {
int n;
SSyncConfChangeSingle* changes;
} SSyncConfChangeSingleArray;
typedef struct SSyncConfigState {
// The voters in the incoming config. (If the configuration is not joint,
// then the outgoing config is empty).
SSyncRaftNodeMap voters;
// The learners in the incoming config.
SSyncRaftNodeMap learners;
// The voters in the outgoing config.
SSyncRaftNodeMap votersOutgoing;
// The nodes that will become learners when the outgoing config is removed.
// These nodes are necessarily currently in nodes_joint (or they would have
// been added to the incoming config right away).
SSyncRaftNodeMap learnersNext;
// If set, the config is joint and Raft will automatically transition into
// the final config (i.e. remove the outgoing config) when this is safe.
bool autoLeave;
} SSyncConfigState;
static FORCE_INLINE bool syncRaftConfArrayIsEmpty(const SSyncConfChangeSingleArray* ary) {
return ary->n == 0;
}
static FORCE_INLINE void syncRaftInitConfArray(SSyncConfChangeSingleArray* ary) {
*ary = (SSyncConfChangeSingleArray) {
.changes = NULL,
.n = 0,
};
}
static FORCE_INLINE void syncRaftFreeConfArray(SSyncConfChangeSingleArray* ary) {
if (ary->changes != NULL) free(ary->changes);
}
#endif /* TD_SYNC_RAFT_PROTO_H */

40
source/libs/sync/inc/sync_raft_quorum.h
View File

@ -1,40 +0,0 @@
/*
* Copyright (c) 2019 TAOS Data, Inc. <cli@taosdata.com>
*
* This program is free software: you can use, redistribute, and/or modify
* it under the terms of the GNU Affero General Public License, version 3
* or later ("AGPL"), as published by the Free Software Foundation.
*
* This program is distributed in the hope that it will be useful, but WITHOUT
* ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or
* FITNESS FOR A PARTICULAR PURPOSE.
*
* You should have received a copy of the GNU Affero General Public License
* along with this program. If not, see <http://www.gnu.org/licenses/>.
*/
#ifndef TD_SYNC_RAFT_QUORUM_H
#define TD_SYNC_RAFT_QUORUM_H
/**
* ESyncRaftVoteResult indicates the outcome of a vote.
**/
typedef enum {
/**
* SYNC_RAFT_VOTE_PENDING indicates that the decision of the vote depends on future
* votes, i.e. neither "yes" or "no" has reached quorum yet.
**/
SYNC_RAFT_VOTE_PENDING = 1,
/**
* SYNC_RAFT_VOTE_LOST indicates that the quorum has voted "no".
**/
SYNC_RAFT_VOTE_LOST = 2,
/**
* SYNC_RAFT_VOTE_WON indicates that the quorum has voted "yes".
**/
SYNC_RAFT_VOTE_WON = 3,
} ESyncRaftVoteResult;
#endif /* TD_SYNC_RAFT_QUORUM_H */

84
source/libs/sync/inc/sync_raft_quorum_joint.h
View File

@ -1,84 +0,0 @@
/*
* Copyright (c) 2019 TAOS Data, Inc. <cli@taosdata.com>
*
* This program is free software: you can use, redistribute, and/or modify
* it under the terms of the GNU Affero General Public License, version 3
* or later ("AGPL"), as published by the Free Software Foundation.
*
* This program is distributed in the hope that it will be useful, but WITHOUT
* ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or
* FITNESS FOR A PARTICULAR PURPOSE.
*
* You should have received a copy of the GNU Affero General Public License
* along with this program. If not, see <http://www.gnu.org/licenses/>.
*/
#ifndef _TD_LIBS_SYNC_RAFT_QUORUM_JOINT_H
#define _TD_LIBS_SYNC_RAFT_QUORUM_JOINT_H
#include "taosdef.h"
#include "sync.h"
#include "sync_type.h"
#include "sync_raft_node_map.h"
#include "thash.h"
// JointConfig is a configuration of two groups of (possibly overlapping)
// majority configurations. Decisions require the support of both majorities.
typedef struct SSyncRaftQuorumJointConfig {
SSyncRaftNodeMap outgoing;
SSyncRaftNodeMap incoming;
} SSyncRaftQuorumJointConfig;
// IDs returns a newly initialized map representing the set of voters present
// in the joint configuration.
void syncRaftJointConfigIDs(SSyncRaftQuorumJointConfig* config, SSyncRaftNodeMap* nodeMap);
// CommittedIndex returns the largest committed index for the given joint
// quorum. An index is jointly committed if it is committed in both constituent
// majorities.
SyncIndex syncRaftJointConfigCommittedIndex(const SSyncRaftQuorumJointConfig* config, matchAckIndexerFp indexer, void* arg);
// VoteResult takes a mapping of voters to yes/no (true/false) votes and returns
// a result indicating whether the vote is pending, lost, or won. A joint quorum
// requires both majority quorums to vote in favor.
ESyncRaftVoteType syncRaftVoteResult(SSyncRaftQuorumJointConfig* config, SHashObj* votesMap);
void syncRaftInitQuorumJointConfig(SSyncRaftQuorumJointConfig* config);
static FORCE_INLINE bool syncRaftJointConfigInOutgoing(const SSyncRaftQuorumJointConfig* config, SyncNodeId id) {
return syncRaftIsInNodeMap(&config->outgoing, id);
}
static FORCE_INLINE bool syncRaftJointConfigInIncoming(const SSyncRaftQuorumJointConfig* config, SyncNodeId id) {
return syncRaftIsInNodeMap(&config->incoming, id);
}
void syncRaftJointConfigAddToIncoming(SSyncRaftQuorumJointConfig* config, SyncNodeId id);
void syncRaftJointConfigRemoveFromIncoming(SSyncRaftQuorumJointConfig* config, SyncNodeId id);
static FORCE_INLINE const SSyncRaftNodeMap* syncRaftJointConfigIncoming(const SSyncRaftQuorumJointConfig* config) {
return &config->incoming;
}
static FORCE_INLINE const SSyncRaftNodeMap* syncRaftJointConfigOutgoing(const SSyncRaftQuorumJointConfig* config) {
return &config->outgoing;
}
static FORCE_INLINE void syncRaftJointConfigClearOutgoing(SSyncRaftQuorumJointConfig* config) {
syncRaftClearNodeMap(&config->outgoing);
}
static FORCE_INLINE bool syncRaftJointConfigIsIncomingEmpty(const SSyncRaftQuorumJointConfig* config) {
return syncRaftNodeMapSize(&config->incoming) == 0;
}
static FORCE_INLINE bool syncRaftJointConfigIsOutgoingEmpty(const SSyncRaftQuorumJointConfig* config) {
return syncRaftNodeMapSize(&config->outgoing) == 0;
}
static FORCE_INLINE bool syncRaftJointConfigIsInOutgoing(const SSyncRaftQuorumJointConfig* config, SyncNodeId id) {
return syncRaftIsInNodeMap(&config->outgoing, id);
}
#endif /* _TD_LIBS_SYNC_RAFT_QUORUM_JOINT_H */

36
source/libs/sync/inc/sync_raft_quorum_majority.h
View File

@ -1,36 +0,0 @@
/*
* Copyright (c) 2019 TAOS Data, Inc. <cli@taosdata.com>
*
* This program is free software: you can use, redistribute, and/or modify
* it under the terms of the GNU Affero General Public License, version 3
* or later ("AGPL"), as published by the Free Software Foundation.
*
* This program is distributed in the hope that it will be useful, but WITHOUT
* ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or
* FITNESS FOR A PARTICULAR PURPOSE.
*
* You should have received a copy of the GNU Affero General Public License
* along with this program. If not, see <http://www.gnu.org/licenses/>.
*/
#ifndef _TD_LIBS_SYNC_RAFT_QUORUM_MAJORITY_H
#define _TD_LIBS_SYNC_RAFT_QUORUM_MAJORITY_H
#include "sync.h"
#include "sync_type.h"
#include "sync_raft_quorum.h"
#include "thash.h"
/**
* syncRaftMajorityVoteResult takes a mapping of voters to yes/no (true/false) votes and returns
* a result indicating whether the vote is pending (i.e. neither a quorum of
* yes/no has been reached), won (a quorum of yes has been reached), or lost (a
* quorum of no has been reached).
**/
ESyncRaftVoteResult syncRaftMajorityVoteResult(SSyncRaftNodeMap* config, SHashObj* votesMap);
// CommittedIndex computes the committed index from those supplied via the
// provided AckedIndexer (for the active config).
SyncIndex syncRaftMajorityConfigCommittedIndex(const SSyncRaftNodeMap* config, matchAckIndexerFp indexer, void* arg);
#endif /* _TD_LIBS_SYNC_RAFT_QUORUM_MAJORITY_H */

33
source/libs/sync/inc/sync_raft_restore.h
View File

@ -1,33 +0,0 @@
/*
* Copyright (c) 2019 TAOS Data, Inc. <cli@taosdata.com>
*
* This program is free software: you can use, redistribute, and/or modify
* it under the terms of the GNU Affero General Public License, version 3
* or later ("AGPL"), as published by the Free Software Foundation.
*
* This program is distributed in the hope that it will be useful, but WITHOUT
* ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or
* FITNESS FOR A PARTICULAR PURPOSE.
*
* You should have received a copy of the GNU Affero General Public License
* along with this program. If not, see <http://www.gnu.org/licenses/>.
*/
#ifndef TD_SYNC_RAFT_RESTORE_H
#define TD_SYNC_RAFT_RESTORE_H
#include "sync_type.h"
#include "sync_raft_proto.h"
// syncRaftRestoreConfig takes a Changer (which must represent an empty configuration), and
// runs a sequence of changes enacting the configuration described in the
// ConfState.
//
// TODO(tbg) it's silly that this takes a Changer. Unravel this by making sure
// the Changer only needs a ProgressMap (not a whole Tracker) at which point
// this can just take LastIndex and MaxInflight directly instead and cook up
// the results from that alone.
int syncRaftRestoreConfig(SSyncRaftChanger* changer, const SSyncConfigState* cs,
SSyncRaftProgressTrackerConfig* config, SSyncRaftProgressMap* progressMap);
#endif /* TD_SYNC_RAFT_RESTORE_H */

91
source/libs/sync/inc/sync_type.h
View File

@ -1,91 +0,0 @@
/*
* Copyright (c) 2019 TAOS Data, Inc. <cli@taosdata.com>
*
* This program is free software: you can use, redistribute, and/or modify
* it under the terms of the GNU Affero General Public License, version 3
* or later ("AGPL"), as published by the Free Software Foundation.
*
* This program is distributed in the hope that it will be useful, but WITHOUT
* ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or
* FITNESS FOR A PARTICULAR PURPOSE.
*
* You should have received a copy of the GNU Affero General Public License
* along with this program. If not, see <http://www.gnu.org/licenses/>.
*/
#ifndef _TD_LIBS_SYNC_TYPE_H
#define _TD_LIBS_SYNC_TYPE_H
#include <stdint.h>
#include "sync.h"
#include "osMath.h"
#define SYNC_NON_NODE_ID -1
#define SYNC_NON_TERM 0
typedef int32_t SyncTime;
typedef uint32_t SyncTick;
typedef struct SSyncRaft SSyncRaft;
typedef struct SSyncRaftProgress SSyncRaftProgress;
typedef struct SSyncRaftProgressMap SSyncRaftProgressMap;
typedef struct SSyncRaftProgressTrackerConfig SSyncRaftProgressTrackerConfig;
typedef struct SSyncRaftNodeMap SSyncRaftNodeMap;
typedef struct SSyncRaftProgressTracker SSyncRaftProgressTracker;
typedef struct SSyncRaftChanger SSyncRaftChanger;
typedef struct SSyncRaftLog SSyncRaftLog;
typedef struct SSyncRaftEntry SSyncRaftEntry;
#if 0
#ifndef TMIN
#define TMIN(x, y) (((x) < (y)) ? (x) : (y))
#endif
#ifndef TMAX
#define TMAX(x, y) (((x) > (y)) ? (x) : (y))
#endif
#endif
typedef struct SSyncServerState {
SyncNodeId voteFor;
SyncTerm term;
SyncIndex commitIndex;
} SSyncServerState;
typedef struct SSyncClusterConfig {
// Log index number of current cluster config.
SyncIndex index;
// Log index number of previous cluster config.
SyncIndex prevIndex;
// current cluster
const SSyncCluster* cluster;
} SSyncClusterConfig;
typedef enum {
SYNC_RAFT_CAMPAIGN_PRE_ELECTION = 0,
SYNC_RAFT_CAMPAIGN_ELECTION = 1,
SYNC_RAFT_CAMPAIGN_TRANSFER = 2,
} ESyncRaftElectionType;
typedef enum {
// grant the vote request
SYNC_RAFT_VOTE_RESP_GRANT = 1,
// reject the vote request
SYNC_RAFT_VOTE_RESP_REJECT = 2,
} ESyncRaftVoteType;
typedef void (*visitProgressFp)(SSyncRaftProgress* progress, void* arg);
typedef void (*matchAckIndexerFp)(SyncNodeId id, void* arg, SyncIndex* index);
#endif /* _TD_LIBS_SYNC_TYPE_H */

325
source/libs/sync/src/raft.c
View File

@ -1,325 +0,0 @@
/*
* Copyright (c) 2019 TAOS Data, Inc. <cli@taosdata.com>
*
* This program is free software: you can use, redistribute, and/or modify
* it under the terms of the GNU Affero General Public License, version 3
* or later ("AGPL"), as published by the Free Software Foundation.
*
* This program is distributed in the hope that it will be useful, but WITHOUT
* ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or
* FITNESS FOR A PARTICULAR PURPOSE.
*
* You should have received a copy of the GNU Affero General Public License
* along with this program. If not, see <http://www.gnu.org/licenses/>.
*/
#include "raft.h"
#include "sync_raft_impl.h"
#include "raft_log.h"
#include "sync_raft_restore.h"
#include "raft_replication.h"
#include "sync_raft_config_change.h"
#include "sync_raft_progress_tracker.h"
#include "syncInt.h"
#define RAFT_READ_LOG_MAX_NUM 100
static int deserializeServerStateFromBuffer(SSyncServerState* server, const char* buffer, int n);
static int deserializeClusterStateFromBuffer(SSyncConfigState* cluster, const char* buffer, int n);
static void switchToConfig(SSyncRaft* pRaft, const SSyncRaftProgressTrackerConfig* config,
const SSyncRaftProgressMap* progressMap, SSyncConfigState* cs);
static void abortLeaderTransfer(SSyncRaft* pRaft);
static bool preHandleMessage(SSyncRaft* pRaft, const SSyncMessage* pMsg);
static bool preHandleNewTermMessage(SSyncRaft* pRaft, const SSyncMessage* pMsg);
static bool preHandleOldTermMessage(SSyncRaft* pRaft, const SSyncMessage* pMsg);
int32_t syncRaftStart(SSyncRaft* pRaft, const SSyncInfo* pInfo) {
SSyncNode* pNode = pRaft->pNode;
SSyncServerState serverState;
SSyncConfigState confState;
SStateManager* stateManager;
SSyncLogStore* logStore;
SSyncFSM* fsm;
SSyncBuffer buffer[RAFT_READ_LOG_MAX_NUM];
int nBuf, limit, i;
char* buf;
int n;
SSyncRaftChanger changer;
memset(pRaft, 0, sizeof(SSyncRaft));
memcpy(&pRaft->fsm, &pInfo->fsm, sizeof(SSyncFSM));
memcpy(&pRaft->logStore, &pInfo->logStore, sizeof(SSyncLogStore));
memcpy(&pRaft->stateManager, &pInfo->stateManager, sizeof(SStateManager));
stateManager = &(pRaft->stateManager);
logStore = &(pRaft->logStore);
fsm = &(pRaft->fsm);
pRaft->nodeInfoMap = taosHashInit(TSDB_MAX_REPLICA, taosGetDefaultHashFunction(TSDB_DATA_TYPE_INT), true, HASH_ENTRY_LOCK);
if (pRaft->nodeInfoMap == NULL) {
return -1;
}
// init progress tracker
pRaft->tracker = syncRaftOpenProgressTracker(pRaft);
if (pRaft->tracker == NULL) {
return -1;
}
// open raft log
if ((pRaft->log = syncRaftLogOpen()) == NULL) {
return -1;
}
// read server state
if (stateManager->readServerState(stateManager, &buf, &n) != 0) {
syncError("readServerState for vgid %d fail", pInfo->vgId);
return -1;
}
if (deserializeServerStateFromBuffer(&serverState, buf, n) != 0) {
syncError("deserializeServerStateFromBuffer for vgid %d fail", pInfo->vgId);
return -1;
}
free(buf);
//assert(initIndex <= serverState.commitIndex);
// read config state
if (stateManager->readClusterState(stateManager, &buf, &n) != 0) {
syncError("readClusterState for vgid %d fail", pInfo->vgId);
return -1;
}
if (deserializeClusterStateFromBuffer(&confState, buf, n) != 0) {
syncError("deserializeClusterStateFromBuffer for vgid %d fail", pInfo->vgId);
return -1;
}
free(buf);
changer = (SSyncRaftChanger) {
.tracker = pRaft->tracker,
.lastIndex = syncRaftLogLastIndex(pRaft->log),
};
SSyncRaftProgressTrackerConfig config;
SSyncRaftProgressMap progressMap;
if (syncRaftRestoreConfig(&changer, &confState, &config, &progressMap) < 0) {
syncError("syncRaftRestoreConfig for vgid %d fail", pInfo->vgId);
return -1;
}
// save restored config and progress map to tracker
syncRaftCopyProgressMap(&progressMap, &pRaft->tracker->progressMap);
syncRaftCopyTrackerConfig(&config, &pRaft->tracker->config);
// free progress map and config
syncRaftFreeProgressMap(&progressMap);
syncRaftFreeTrackConfig(&config);
if (!syncRaftIsEmptyServerState(&serverState)) {
syncRaftLoadState(pRaft, &serverState);
}
if (pInfo->appliedIndex > 0) {
syncRaftLogAppliedTo(pRaft->log, pInfo->appliedIndex);
}
syncRaftBecomeFollower(pRaft, pRaft->term, SYNC_NON_NODE_ID);
syncInfo("[%d:%d] restore vgid %d state: snapshot index success",
pRaft->selfGroupId, pRaft->selfId, pInfo->vgId);
return 0;
}
int32_t syncRaftStep(SSyncRaft* pRaft, const SSyncMessage* pMsg) {
syncDebug("from %d, type:%d, term:%" PRId64 ", state:%d",
pMsg->from, pMsg->msgType, pMsg->term, pRaft->state);
if (preHandleMessage(pRaft, pMsg)) {
syncFreeMessage(pMsg);
return 0;
}
ESyncRaftMessageType msgType = pMsg->msgType;
if (msgType == RAFT_MSG_INTERNAL_ELECTION) {
syncRaftHandleElectionMessage(pRaft, pMsg);
} else if (msgType == RAFT_MSG_VOTE) {
syncRaftHandleVoteMessage(pRaft, pMsg);
} else {
pRaft->stepFp(pRaft, pMsg);
}
syncFreeMessage(pMsg);
return 0;
}
int32_t syncRaftTick(SSyncRaft* pRaft) {
pRaft->currentTick += 1;
pRaft->tickFp(pRaft);
return 0;
}
static int deserializeServerStateFromBuffer(SSyncServerState* server, const char* buffer, int n) {
return 0;
}
static int deserializeClusterStateFromBuffer(SSyncConfigState* cluster, const char* buffer, int n) {
return 0;
}
static void visitProgressMaybeSendAppend(SSyncRaftProgress* progress, void* arg) {
syncRaftMaybeSendAppend(arg, progress, false);
}
// switchToConfig reconfigures this node to use the provided configuration. It
// updates the in-memory state and, when necessary, carries out additional
// actions such as reacting to the removal of nodes or changed quorum
// requirements.
//
// The inputs usually result from restoring a ConfState or applying a ConfChange.
static void switchToConfig(SSyncRaft* pRaft, const SSyncRaftProgressTrackerConfig* config,
const SSyncRaftProgressMap* progressMap, SSyncConfigState* cs) {
SyncNodeId selfId = pRaft->selfId;
int i;
bool exist;
SSyncRaftProgress* progress = NULL;
syncRaftConfigState(pRaft->tracker, cs);
progress = syncRaftFindProgressByNodeId(&pRaft->tracker->progressMap, selfId);
exist = (progress != NULL);
// Update whether the node itself is a learner, resetting to false when the
// node is removed.
if (exist) {
pRaft->isLearner = progress->isLearner;
} else {
pRaft->isLearner = false;
}
if ((!exist || pRaft->isLearner) && pRaft->state == TAOS_SYNC_STATE_LEADER) {
// This node is leader and was removed or demoted. We prevent demotions
// at the time writing but hypothetically we handle them the same way as
// removing the leader: stepping down into the next Term.
//
// TODO(tbg): step down (for sanity) and ask follower with largest Match
// to TimeoutNow (to avoid interruption). This might still drop some
// proposals but it's better than nothing.
//
// TODO(tbg): test this branch. It is untested at the time of writing.
return;
}
// The remaining steps only make sense if this node is the leader and there
// are other nodes.
if (pRaft->state != TAOS_SYNC_STATE_LEADER || syncRaftNodeMapSize(&cs->voters) == 0) {
return;
}
if (syncRaftMaybeCommit(pRaft)) {
// If the configuration change means that more entries are committed now,
// broadcast/append to everyone in the updated config.
syncRaftBroadcastAppend(pRaft);
} else {
// Otherwise, still probe the newly added replicas; there's no reason to
// let them wait out a heartbeat interval (or the next incoming
// proposal).
syncRaftProgressVisit(pRaft->tracker, visitProgressMaybeSendAppend, pRaft);
// If the the leadTransferee was removed or demoted, abort the leadership transfer.
SyncNodeId leadTransferee = pRaft->leadTransferee;
if (leadTransferee != SYNC_NON_NODE_ID) {
if (!syncRaftIsInNodeMap(&pRaft->tracker->config.voters.incoming, leadTransferee) &&
!syncRaftIsInNodeMap(&pRaft->tracker->config.voters.outgoing, leadTransferee)) {
abortLeaderTransfer(pRaft);
}
}
}
}
static void abortLeaderTransfer(SSyncRaft* pRaft) {
pRaft->leadTransferee = SYNC_NON_NODE_ID;
}
/**
* pre-handle message, return true means no need to continue
* Handle the message term, which may result in our stepping down to a follower.
**/
static bool preHandleMessage(SSyncRaft* pRaft, const SSyncMessage* pMsg) {
// local message?
if (pMsg->term == 0) {
return false;
}
if (pMsg->term > pRaft->term) {
return preHandleNewTermMessage(pRaft, pMsg);
} else if (pMsg->term < pRaft->term) {
return preHandleOldTermMessage(pRaft, pMsg);
}
return false;
}
static bool preHandleNewTermMessage(SSyncRaft* pRaft, const SSyncMessage* pMsg) {
SyncNodeId leaderId = pMsg->from;
ESyncRaftMessageType msgType = pMsg->msgType;
if (msgType == RAFT_MSG_VOTE) {
// TODO
leaderId = SYNC_NON_NODE_ID;
}
if (syncIsPreVoteMsg(pMsg)) {
// Never change our term in response to a PreVote
} else if (syncIsPreVoteRespMsg(pMsg) && !pMsg->voteResp.rejected) {
/**
* We send pre-vote requests with a term in our future. If the
* pre-vote is granted, we will increment our term when we get a
* quorum. If it is not, the term comes from the node that
* rejected our vote so we should become a follower at the new
* term.
**/
} else {
syncInfo("[%d:%d] [term:%" PRId64 "] received a %d message with higher term from %d [term:%" PRId64 "]",
pRaft->selfGroupId, pRaft->selfId, pRaft->term, msgType, pMsg->from, pMsg->term);
syncRaftBecomeFollower(pRaft, pMsg->term, leaderId);
}
return false;
}
static bool preHandleOldTermMessage(SSyncRaft* pRaft, const SSyncMessage* pMsg) {
if (pRaft->checkQuorum && pMsg->msgType == RAFT_MSG_APPEND) {
/**
* We have received messages from a leader at a lower term. It is possible
* that these messages were simply delayed in the network, but this could
* also mean that this node has advanced its term number during a network
* partition, and it is now unable to either win an election or to rejoin
* the majority on the old term. If checkQuorum is false, this will be
* handled by incrementing term numbers in response to MsgVote with a
* higher term, but if checkQuorum is true we may not advance the term on
* MsgVote and must generate other messages to advance the term. The net
* result of these two features is to minimize the disruption caused by
* nodes that have been removed from the cluster's configuration: a
* removed node will send MsgVotes (or MsgPreVotes) which will be ignored,
* but it will not receive MsgApp or MsgHeartbeat, so it will not create
* disruptive term increases
**/
SNodeInfo* pNode = syncRaftGetNodeById(pRaft, pMsg->from);
if (pNode == NULL) {
return true;
}
SSyncMessage* msg = syncNewEmptyAppendRespMsg(pRaft->selfGroupId, pRaft->selfId, pRaft->term);
if (msg == NULL) {
return true;
}
pRaft->io.send(msg, pNode);
} else {
// ignore other cases
syncInfo("[%d:%d] [term:%" PRId64 "] ignored a %d message with lower term from %d [term:%" PRId64 "]",
pRaft->selfGroupId, pRaft->selfId, pRaft->term, pMsg->msgType, pMsg->from, pMsg->term);
}
return true;
}

48
source/libs/sync/src/raft_handle_append_entries_message.c
View File

@ -1,48 +0,0 @@
/*
* Copyright (c) 2019 TAOS Data, Inc. <cli@taosdata.com>
*
* This program is free software: you can use, redistribute, and/or modify
* it under the terms of the GNU Affero General Public License, version 3
* or later ("AGPL"), as published by the Free Software Foundation.
*
* This program is distributed in the hope that it will be useful, but WITHOUT
* ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or
* FITNESS FOR A PARTICULAR PURPOSE.
*
* You should have received a copy of the GNU Affero General Public License
* along with this program. If not, see <http://www.gnu.org/licenses/>.
*/
#include "syncInt.h"
#include "raft.h"
#include "raft_log.h"
#include "sync_raft_impl.h"
#include "raft_message.h"
int syncRaftHandleAppendEntriesMessage(SSyncRaft* pRaft, const SSyncMessage* pMsg) {
const RaftMsg_Append_Entries *appendEntries = &(pMsg->appendEntries);
SNodeInfo* pNode = syncRaftGetNodeById(pRaft, pMsg->from);
if (pNode == NULL) {
return 0;
}
SSyncMessage* pRespMsg = syncNewEmptyAppendRespMsg(pRaft->selfGroupId, pRaft->selfId, pRaft->term);
if (pRespMsg == NULL) {
return 0;
}
RaftMsg_Append_Resp *appendResp = &(pRespMsg->appendResp);
// ignore committed logs
if (syncRaftLogIsCommitted(pRaft->log, appendEntries->index)) {
appendResp->index = pRaft->log->commitIndex;
goto out;
}
syncInfo("[%d:%d] recv append from %d index %" PRId64"",
pRaft->selfGroupId, pRaft->selfId, pMsg->from, appendEntries->index);
out:
pRaft->io.send(pRespMsg, pNode);
return 0;
}

29
source/libs/sync/src/raft_handle_election_message.c
View File

@ -1,29 +0,0 @@
/*
* Copyright (c) 2019 TAOS Data, Inc. <cli@taosdata.com>
*
* This program is free software: you can use, redistribute, and/or modify
* it under the terms of the GNU Affero General Public License, version 3
* or later ("AGPL"), as published by the Free Software Foundation.
*
* This program is distributed in the hope that it will be useful, but WITHOUT
* ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or
* FITNESS FOR A PARTICULAR PURPOSE.
*
* You should have received a copy of the GNU Affero General Public License
* along with this program. If not, see <http://www.gnu.org/licenses/>.
*/
#include "syncInt.h"
#include "raft.h"
#include "raft_log.h"
#include "raft_message.h"
int syncRaftHandleElectionMessage(SSyncRaft* pRaft, const SSyncMessage* pMsg) {
if (pRaft->preVote) {
syncRaftStartElection(pRaft, SYNC_RAFT_CAMPAIGN_PRE_ELECTION);
} else {
syncRaftStartElection(pRaft, SYNC_RAFT_CAMPAIGN_ELECTION);
}
return 0;
}

61
source/libs/sync/src/raft_handle_vote_message.c
View File

@ -1,61 +0,0 @@
/*
* Copyright (c) 2019 TAOS Data, Inc. <cli@taosdata.com>
*
* This program is free software: you can use, redistribute, and/or modify
* it under the terms of the GNU Affero General Public License, version 3
* or later ("AGPL"), as published by the Free Software Foundation.
*
* This program is distributed in the hope that it will be useful, but WITHOUT
* ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or
* FITNESS FOR A PARTICULAR PURPOSE.
*
* You should have received a copy of the GNU Affero General Public License
* along with this program. If not, see <http://www.gnu.org/licenses/>.
*/
#include "syncInt.h"
#include "raft.h"
#include "sync_raft_impl.h"
#include "raft_log.h"
#include "raft_message.h"
static bool canGrantVoteMessage(SSyncRaft* pRaft, const SSyncMessage* pMsg);
int syncRaftHandleVoteMessage(SSyncRaft* pRaft, const SSyncMessage* pMsg) {
SSyncMessage* pRespMsg;
SNodeInfo* pNode = syncRaftGetNodeById(pRaft, pMsg->from);
if (pNode == NULL) {
return 0;
}
bool grant;
SyncIndex lastIndex = syncRaftLogLastIndex(pRaft->log);
SyncTerm lastTerm = syncRaftLogLastTerm(pRaft->log);
grant = canGrantVoteMessage(pRaft, pMsg);
pRespMsg = syncNewVoteRespMsg(pRaft->selfGroupId, pRaft->selfId, pMsg->vote.cType, !grant);
if (pRespMsg == NULL) {
return 0;
}
syncInfo("[%d:%d] [logterm: %" PRId64 ", index: %" PRId64 ", vote: %d] %s for %d"
"[logterm: %" PRId64 ", index: %" PRId64 "] at term %" PRId64 "",
pRaft->selfGroupId, pRaft->selfId, lastTerm, lastIndex, pRaft->voteFor,
grant ? "grant" : "reject",
pMsg->from, pMsg->vote.lastTerm, pMsg->vote.lastIndex, pRaft->term);
pRaft->io.send(pRespMsg, pNode);
return 0;
}
static bool canGrantVoteMessage(SSyncRaft* pRaft, const SSyncMessage* pMsg) {
bool canVote =
// We can vote if this is a repeat of a vote we've already cast...
pRaft->voteFor == pMsg->from ||
// ...we haven't voted and we don't think there's a leader yet in this term...
(pRaft->voteFor == SYNC_NON_NODE_ID && pRaft->leaderId == SYNC_NON_NODE_ID) ||
// ...or this is a PreVote for a future term...
(pMsg->vote.cType == SYNC_RAFT_CAMPAIGN_PRE_ELECTION && pMsg->term > pRaft->term);
// ...and we believe the candidate is up to date.
return canVote && syncRaftLogIsUptodate(pRaft->log, pMsg->vote.lastIndex, pMsg->vote.lastTerm);
}

60
source/libs/sync/src/raft_handle_vote_resp_message.c
View File

@ -1,60 +0,0 @@
/*
* Copyright (c) 2019 TAOS Data, Inc. <cli@taosdata.com>
*
* This program is free software: you can use, redistribute, and/or modify
* it under the terms of the GNU Affero General Public License, version 3
* or later ("AGPL"), as published by the Free Software Foundation.
*
* This program is distributed in the hope that it will be useful, but WITHOUT
* ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or
* FITNESS FOR A PARTICULAR PURPOSE.
*
* You should have received a copy of the GNU Affero General Public License
* along with this program. If not, see <http://www.gnu.org/licenses/>.
*/
#include "syncInt.h"
#include "raft.h"
#include "sync_raft_impl.h"
#include "raft_message.h"
int syncRaftHandleVoteRespMessage(SSyncRaft* pRaft, const SSyncMessage* pMsg) {
int granted, rejected;
int quorum;
int voterIndex;
assert(pRaft->state == TAOS_SYNC_STATE_CANDIDATE);
SNodeInfo* pNode = syncRaftGetNodeById(pRaft, pMsg->from);
if (pNode == NULL) {
syncError("[%d:%d] recv vote resp from unknown server %d", pRaft->selfGroupId, pRaft->selfId, pMsg->from);
return 0;
}
if (pRaft->state != TAOS_SYNC_STATE_CANDIDATE) {
syncError("[%d:%d] is not candidate, ignore vote resp", pRaft->selfGroupId, pRaft->selfId);
return 0;
}
ESyncRaftVoteResult result = syncRaftPollVote(pRaft, pMsg->from,
pMsg->voteResp.cType == SYNC_RAFT_CAMPAIGN_PRE_ELECTION,
!pMsg->voteResp.rejected, &rejected, &granted);
syncInfo("[%d:%d] [quorum:%d] has received %d votes and %d vote rejections",
pRaft->selfGroupId, pRaft->selfId, quorum, granted, rejected);
if (result == SYNC_RAFT_VOTE_WON) {
if (pRaft->candidateState.inPreVote) {
syncRaftCampaign(pRaft, SYNC_RAFT_CAMPAIGN_ELECTION);
} else {
syncRaftBecomeLeader(pRaft);
syncRaftBroadcastAppend(pRaft);
}
} else if (result == SYNC_RAFT_VOTE_LOST) {
// pb.MsgPreVoteResp contains future term of pre-candidate
// m.Term > r.Term; reuse r.Term
syncRaftBecomeFollower(pRaft, pRaft->term, SYNC_NON_NODE_ID);
}
return 0;
}

66
source/libs/sync/src/raft_log.c
View File

@ -1,66 +0,0 @@
/*
* Copyright (c) 2019 TAOS Data, Inc. <cli@taosdata.com>
*
* This program is free software: you can use, redistribute, and/or modify
* it under the terms of the GNU Affero General Public License, version 3
* or later ("AGPL"), as published by the Free Software Foundation.
*
* This program is distributed in the hope that it will be useful, but WITHOUT
* ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or
* FITNESS FOR A PARTICULAR PURPOSE.
*
* You should have received a copy of the GNU Affero General Public License
* along with this program. If not, see <http://www.gnu.org/licenses/>.
*/
#include "raft_log.h"
SSyncRaftLog* syncRaftLogOpen() {
return NULL;
}
SyncIndex syncRaftLogLastIndex(SSyncRaftLog* pLog) {
return 0;
}
SyncIndex syncRaftLogSnapshotIndex(SSyncRaftLog* pLog) {
return 0;
}
SyncTerm syncRaftLogLastTerm(SSyncRaftLog* pLog) {
return 0;
}
void syncRaftLogAppliedTo(SSyncRaftLog* pLog, SyncIndex appliedIndex) {
}
bool syncRaftLogIsUptodate(SSyncRaftLog* pLog, SyncIndex index, SyncTerm term) {
return true;
}
int syncRaftLogNumOfPendingConf(SSyncRaftLog* pLog) {
return 0;
}
bool syncRaftHasUnappliedLog(SSyncRaftLog* pLog) {
return pLog->commitIndex > pLog->appliedIndex;
}
SyncTerm syncRaftLogTermOf(SSyncRaftLog* pLog, SyncIndex index) {
return SYNC_NON_TERM;
}
int syncRaftLogAppend(SSyncRaftLog* pLog, SSyncRaftEntry *pEntries, int n) {
}
int syncRaftLogAcquire(SSyncRaftLog* pLog, SyncIndex index, int maxMsgSize,
SSyncRaftEntry **ppEntries, int *n) {
return 0;
}
void syncRaftLogRelease(SSyncRaftLog* pLog, SyncIndex index,
SSyncRaftEntry *pEntries, int n) {
return;
}

22
source/libs/sync/src/raft_message.c
View File

@ -1,22 +0,0 @@
/*
* Copyright (c) 2019 TAOS Data, Inc. <cli@taosdata.com>
*
* This program is free software: you can use, redistribute, and/or modify
* it under the terms of the GNU Affero General Public License, version 3
* or later ("AGPL"), as published by the Free Software Foundation.
*
* This program is distributed in the hope that it will be useful, but WITHOUT
* ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or
* FITNESS FOR A PARTICULAR PURPOSE.
*
* You should have received a copy of the GNU Affero General Public License
* along with this program. If not, see <http://www.gnu.org/licenses/>.
*/
#include "raft_message.h"
void syncFreeMessage(const SSyncMessage* pMsg) {
if (!syncIsInternalMsg(pMsg->msgType)) {
free((SSyncMessage*)pMsg);
}
}

110
source/libs/sync/src/raft_replication.c
View File

@ -1,110 +0,0 @@
/*
* Copyright (c) 2019 TAOS Data, Inc. <cli@taosdata.com>
*
* This program is free software: you can use, redistribute, and/or modify
* it under the terms of the GNU Affero General Public License, version 3
* or later ("AGPL"), as published by the Free Software Foundation.
*
* This program is distributed in the hope that it will be useful, but WITHOUT
* ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or
* FITNESS FOR A PARTICULAR PURPOSE.
*
* You should have received a copy of the GNU Affero General Public License
* along with this program. If not, see <http://www.gnu.org/licenses/>.
*/
#include "raft.h"
#include "raft_log.h"
#include "sync_raft_progress.h"
#include "syncInt.h"
#include "raft_replication.h"
static bool sendSnapshot(SSyncRaft* pRaft, SSyncRaftProgress* progress);
static bool sendAppendEntries(SSyncRaft* pRaft, SSyncRaftProgress* progress,
SyncIndex prevIndex, SyncTerm prevTerm,
SSyncRaftEntry *entries, int nEntry);
// maybeSendAppend sends an append RPC with new entries to the given peer,
// if necessary. Returns true if a message was sent. The sendIfEmpty
// argument controls whether messages with no entries will be sent
// ("empty" messages are useful to convey updated Commit indexes, but
// are undesirable when we're sending multiple messages in a batch).
bool syncRaftMaybeSendAppend(SSyncRaft* pRaft, SSyncRaftProgress* progress, bool sendIfEmpty) {
assert(pRaft->state == TAOS_SYNC_STATE_LEADER);
SyncNodeId nodeId = progress->id;
if (syncRaftProgressIsPaused(progress)) {
syncInfo("node [%d:%d] paused", pRaft->selfGroupId, nodeId);
return false;
}
SyncIndex nextIndex = syncRaftProgressNextIndex(progress);
SSyncRaftEntry *entries;
int nEntry;
SyncIndex prevIndex;
SyncTerm prevTerm;
prevIndex = nextIndex - 1;
prevTerm = syncRaftLogTermOf(pRaft->log, prevIndex);
int ret = syncRaftLogAcquire(pRaft->log, nextIndex, pRaft->maxMsgSize, &entries, &nEntry);
if (nEntry == 0 && !sendIfEmpty) {
return false;
}
if (ret != 0 || prevTerm == SYNC_NON_TERM) {
return sendSnapshot(pRaft, progress);
}
return sendAppendEntries(pRaft, progress, prevIndex, prevTerm, entries, nEntry);
}
static bool sendSnapshot(SSyncRaft* pRaft, SSyncRaftProgress* progress) {
if (!syncRaftProgressRecentActive(progress)) {
return false;
}
return true;
}
static bool sendAppendEntries(SSyncRaft* pRaft, SSyncRaftProgress* progress,
SyncIndex prevIndex, SyncTerm prevTerm,
SSyncRaftEntry *entries, int nEntry) {
SNodeInfo* pNode = syncRaftGetNodeById(pRaft, progress->id);
if (pNode == NULL) {
return false;
}
SyncIndex lastIndex;
SyncTerm logTerm = prevTerm;
SSyncMessage* msg = syncNewAppendMsg(pRaft->selfGroupId, pRaft->selfId, pRaft->term,
prevIndex, prevTerm, pRaft->log->commitIndex,
nEntry, entries);
if (msg == NULL) {
goto err_release_log;
}
if (nEntry != 0) {
switch (progress->state) {
// optimistically increase the next when in StateReplicate
case PROGRESS_STATE_REPLICATE:
lastIndex = entries[nEntry - 1].index;
syncRaftProgressOptimisticNextIndex(progress, lastIndex);
syncRaftInflightAdd(progress->inflights, lastIndex);
break;
case PROGRESS_STATE_PROBE:
progress->probeSent = true;
break;
default:
syncFatal("[%d:%d] is sending append in unhandled state %s",
pRaft->selfGroupId, pRaft->selfId, syncRaftProgressStateString(progress));
break;
}
}
pRaft->io.send(msg, pNode);
return true;
err_release_log:
syncRaftLogRelease(pRaft->log, prevIndex + 1, entries, nEntry);
return false;
}

23
source/libs/sync/src/raft_unstable_log.c
View File

@ -1,23 +0,0 @@
/*
* Copyright (c) 2019 TAOS Data, Inc. <cli@taosdata.com>
*
* This program is free software: you can use, redistribute, and/or modify
* it under the terms of the GNU Affero General Public License, version 3
* or later ("AGPL"), as published by the Free Software Foundation.
*
* This program is distributed in the hope that it will be useful, but WITHOUT
* ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or
* FITNESS FOR A PARTICULAR PURPOSE.
*
* You should have received a copy of the GNU Affero General Public License
* along with this program. If not, see <http://www.gnu.org/licenses/>.
*/
#include "sync.h"
#include "raft_unstable_log.h"
/*
SyncIndex syncRaftLogLastIndex(SSyncRaftUnstableLog* pLog) {
return 0;
}
*/

302
source/libs/sync/src/sync.c
View File

@ -1,302 +0,0 @@
/*
* Copyright (c) 2019 TAOS Data, Inc. <jhtao@taosdata.com>
*
* This program is free software: you can use, redistribute, and/or modify
* it under the terms of the GNU Affero General Public License, version 3
* or later ("AGPL"), as published by the Free Software Foundation.
*
* This program is distributed in the hope that it will be useful, but WITHOUT
* ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or
* FITNESS FOR A PARTICULAR PURPOSE.
*
* You should have received a copy of the GNU Affero General Public License
* along with this program. If not, see <http://www.gnu.org/licenses/>.
*/
#include "syncInt.h"
#include "trpc.h"
#include "ttimer.h"
SSyncManager* gSyncManager = NULL;
#define SYNC_TICK_TIMER 50
#define SYNC_ACTIVITY_TIMER 5
#define SYNC_SERVER_WORKER 2
static void syncProcessRsp(void *parent, SRpcMsg *pMsg, SEpSet *pEpSet);
static void syncProcessReqMsg(void *parent, SRpcMsg *pMsg, SEpSet *pEpSet);
static int syncInitRpcServer(SSyncManager* syncManager, const SSyncCluster* pSyncCfg);
static int syncInitRpcClient(SSyncManager* syncManager);
static int syncOpenWorkerPool(SSyncManager* syncManager);
static int syncCloseWorkerPool(SSyncManager* syncManager);
static void *syncWorkerMain(void *argv);
static void syncNodeTick(void *param, void *tmrId);
int32_t syncInit() {
if (gSyncManager != NULL) {
return 0;
}
gSyncManager = (SSyncManager*)calloc(sizeof(SSyncManager), 0);
if (gSyncManager == NULL) {
syncError("malloc SSyncManager fail");
return -1;
}
pthread_mutex_init(&gSyncManager->mutex, NULL);
// init client rpc
if (syncInitRpcClient(gSyncManager) != 0) {
syncCleanUp();
return -1;
}
// init sync timer manager
gSyncManager->syncTimerManager = taosTmrInit(1000, 50, 10000, "SYNC");
if (gSyncManager->syncTimerManager == NULL) {
syncCleanUp();
return -1;
}
// init worker pool
if (syncOpenWorkerPool(gSyncManager) != 0) {
syncCleanUp();
return -1;
}
// init vgroup hash table
gSyncManager->vgroupTable = taosHashInit(TSDB_MIN_VNODES, taosGetDefaultHashFunction(TSDB_DATA_TYPE_INT), true, HASH_ENTRY_LOCK);
if (gSyncManager->vgroupTable == NULL) {
syncCleanUp();
return -1;
}
return 0;
}
void syncCleanUp() {
if (gSyncManager == NULL) {
return;
}
pthread_mutex_lock(&gSyncManager->mutex);
if (gSyncManager->vgroupTable) {
taosHashCleanup(gSyncManager->vgroupTable);
}
if (gSyncManager->clientRpc) {
rpcClose(gSyncManager->clientRpc);
syncInfo("sync inter-sync rpc client is closed");
}
if (gSyncManager->syncTimerManager) {
taosTmrCleanUp(gSyncManager->syncTimerManager);
}
syncCloseWorkerPool(gSyncManager);
pthread_mutex_unlock(&gSyncManager->mutex);
pthread_mutex_destroy(&gSyncManager->mutex);
free(gSyncManager);
gSyncManager = NULL;
}
SSyncNode* syncStart(const SSyncInfo* pInfo) {
pthread_mutex_lock(&gSyncManager->mutex);
SSyncNode **ppNode = taosHashGet(gSyncManager->vgroupTable, &pInfo->vgId, sizeof(SyncGroupId*));
if (ppNode != NULL) {
syncInfo("vgroup %d already exist", pInfo->vgId);
pthread_mutex_unlock(&gSyncManager->mutex);
return *ppNode;
}
// init rpc server
if (syncInitRpcServer(gSyncManager, &pInfo->syncCfg) != 0) {
pthread_mutex_unlock(&gSyncManager->mutex);
return NULL;
}
SSyncNode *pNode = (SSyncNode*)malloc(sizeof(SSyncNode));
if (pNode == NULL) {
syncError("malloc vgroup %d node fail", pInfo->vgId);
pthread_mutex_unlock(&gSyncManager->mutex);
return NULL;
}
pNode->syncTimer = taosTmrStart(syncNodeTick, SYNC_TICK_TIMER, (void*)((int64_t)pInfo->vgId), gSyncManager->syncTimerManager);
// start raft
pNode->raft.pNode = pNode;
if (syncRaftStart(&pNode->raft, pInfo) != 0) {
syncError("raft start at %d node fail", pInfo->vgId);
pthread_mutex_unlock(&gSyncManager->mutex);
return NULL;
}
pthread_mutex_init(&pNode->mutex, NULL);
taosHashPut(gSyncManager->vgroupTable, &pInfo->vgId, sizeof(SyncGroupId), &pNode, sizeof(SSyncNode *));
pthread_mutex_unlock(&gSyncManager->mutex);
return NULL;
}
void syncStop(const SSyncNode* pNode) {
pthread_mutex_lock(&gSyncManager->mutex);
SSyncNode **ppNode = taosHashGet(gSyncManager->vgroupTable, &pNode->vgId, sizeof(SyncGroupId*));
if (ppNode == NULL) {
syncInfo("vgroup %d not exist", pNode->vgId);
pthread_mutex_unlock(&gSyncManager->mutex);
return;
}
assert(*ppNode == pNode);
taosTmrStop(pNode->syncTimer);
taosHashRemove(gSyncManager->vgroupTable, &pNode->vgId, sizeof(SyncGroupId));
pthread_mutex_unlock(&gSyncManager->mutex);
pthread_mutex_destroy(&((*ppNode)->mutex));
free(*ppNode);
}
int32_t syncPropose(SSyncNode* syncNode, const SSyncBuffer* pBuf, void* pData, bool isWeak) {
SSyncMessage msg;
pthread_mutex_lock(&syncNode->mutex);
int32_t ret = syncRaftStep(&syncNode->raft, syncInitPropMsg(&msg, pBuf, pData, isWeak));
pthread_mutex_unlock(&syncNode->mutex);
return ret;
}
void syncReconfig(const SSyncNode* pNode, const SSyncCluster* pCfg) {}
int32_t syncAddNode(SSyncNode syncNode, const SNodeInfo *pNode) {
return 0;
}
int32_t syncRemoveNode(SSyncNode syncNode, const SNodeInfo *pNode) {
return 0;
}
// process rpc rsp message from other sync server
static void syncProcessRsp(void *parent, SRpcMsg *pMsg, SEpSet *pEpSet) {
}
// process rpc message from other sync server
static void syncProcessReqMsg(void *parent, SRpcMsg *pMsg, SEpSet *pEpSet) {
}
static int syncInitRpcServer(SSyncManager* syncManager, const SSyncCluster* pSyncCfg) {
if (gSyncManager->rpcServerTable == NULL) {
gSyncManager->rpcServerTable = taosHashInit(TSDB_MIN_VNODES, taosGetDefaultHashFunction(TSDB_DATA_TYPE_BINARY), true, HASH_ENTRY_LOCK);
if (gSyncManager->rpcServerTable == NULL) {
syncError("init sync rpc server hash table error");
return -1;
}
}
assert(pSyncCfg->selfIndex < pSyncCfg->replica && pSyncCfg->selfIndex >= 0);
const SNodeInfo* pNode = &(pSyncCfg->nodeInfo[pSyncCfg->replica]);
char buffer[156] = {'\0'};
snprintf(buffer, sizeof(buffer), "%s:%d", &(pNode->nodeFqdn[0]), pNode->nodePort);
size_t len = strlen(buffer);
void** ppRpcServer = taosHashGet(gSyncManager->rpcServerTable, buffer, len);
if (ppRpcServer != NULL) {
// already inited
syncInfo("sync rpc server for %s already exist", buffer);
return 0;
}
SRpcInit rpcInit;
memset(&rpcInit, 0, sizeof(rpcInit));
rpcInit.localPort = pNode->nodePort;
rpcInit.label = "sync-server";
rpcInit.numOfThreads = SYNC_SERVER_WORKER;
rpcInit.cfp = syncProcessReqMsg;
rpcInit.sessions = TSDB_MAX_VNODES << 4;
rpcInit.connType = TAOS_CONN_SERVER;
rpcInit.idleTime = SYNC_ACTIVITY_TIMER * 1000;
void* rpcServer = rpcOpen(&rpcInit);
if (rpcServer == NULL) {
syncInfo("rpcOpen for sync rpc server for %s fail", buffer);
return -1;
}
taosHashPut(gSyncManager->rpcServerTable, buffer, strlen(buffer), rpcServer, len);
syncInfo("sync rpc server for %s init success", buffer);
return 0;
}
static int syncInitRpcClient(SSyncManager* syncManager) {
char secret[TSDB_PASSWORD_LEN] = "secret";
SRpcInit rpcInit;
memset(&rpcInit, 0, sizeof(rpcInit));
rpcInit.label = "sync-client";
rpcInit.numOfThreads = 1;
rpcInit.cfp = syncProcessRsp;
rpcInit.sessions = TSDB_MAX_VNODES << 4;
rpcInit.connType = TAOS_CONN_CLIENT;
rpcInit.idleTime = SYNC_ACTIVITY_TIMER * 1000;
rpcInit.user = "t";
rpcInit.ckey = "key";
rpcInit.secret = secret;
syncManager->clientRpc = rpcOpen(&rpcInit);
if (syncManager->clientRpc == NULL) {
syncError("failed to init sync rpc client");
return -1;
}
syncInfo("sync inter-sync rpc client is initialized");
return 0;
}
static int syncOpenWorkerPool(SSyncManager* syncManager) {
int i;
pthread_attr_t thattr;
pthread_attr_init(&thattr);
pthread_attr_setdetachstate(&thattr, PTHREAD_CREATE_JOINABLE);
for (i = 0; i < TAOS_SYNC_MAX_WORKER; ++i) {
SSyncWorker* pWorker = &(syncManager->worker[i]);
if (pthread_create(&(pWorker->thread), &thattr, (void *)syncWorkerMain, pWorker) != 0) {
syncError("failed to create sync worker since %s", strerror(errno));
return -1;
}
}
pthread_attr_destroy(&thattr);
return 0;
}
static int syncCloseWorkerPool(SSyncManager* syncManager) {
return 0;
}
static void *syncWorkerMain(void *argv) {
SSyncWorker* pWorker = (SSyncWorker *)argv;
taosBlockSIGPIPE();
setThreadName("syncWorker");
return NULL;
}
static void syncNodeTick(void *param, void *tmrId) {
SyncGroupId vgId = (SyncGroupId)((int64_t)param);
SSyncNode **ppNode = taosHashGet(gSyncManager->vgroupTable, &vgId, sizeof(SyncGroupId*));
if (ppNode == NULL) {
return;
}
SSyncNode *pNode = *ppNode;
pthread_mutex_lock(&pNode->mutex);
syncRaftTick(&pNode->raft);
pthread_mutex_unlock(&pNode->mutex);
pNode->syncTimer = taosTmrStart(syncNodeTick, SYNC_TICK_TIMER, (void*)(int64_t)pNode->vgId, gSyncManager->syncTimerManager);
}

409
source/libs/sync/src/sync_raft_config_change.c
View File

@ -1,409 +0,0 @@
/*
* Copyright (c) 2019 TAOS Data, Inc. <cli@taosdata.com>
*
* This program is free software: you can use, redistribute, and/or modify
* it under the terms of the GNU Affero General Public License, version 3
* or later ("AGPL"), as published by the Free Software Foundation.
*
* This program is distributed in the hope that it will be useful, but WITHOUT
* ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or
* FITNESS FOR A PARTICULAR PURPOSE.
*
* You should have received a copy of the GNU Affero General Public License
* along with this program. If not, see <http://www.gnu.org/licenses/>.
*/
#include "raft.h"
#include "syncInt.h"
#include "sync_raft_config_change.h"
#include "sync_raft_progress.h"
#include "sync_raft_progress_tracker.h"
#include "sync_raft_quorum_joint.h"
static int checkAndCopy(SSyncRaftChanger* changer, SSyncRaftProgressTrackerConfig* config, SSyncRaftProgressMap* progressMap);
static int checkAndReturn(SSyncRaftProgressTrackerConfig* config, SSyncRaftProgressMap* progressMap);
static int checkInvariants(SSyncRaftProgressTrackerConfig* config, SSyncRaftProgressMap* progressMap);
static int checkInvariants(SSyncRaftProgressTrackerConfig* config, SSyncRaftProgressMap* progressMap);
static bool hasJointConfig(const SSyncRaftProgressTrackerConfig* config);
static int applyConfig(SSyncRaftChanger* changer, SSyncRaftProgressTrackerConfig* config,
SSyncRaftProgressMap* progressMap, const SSyncConfChangeSingleArray* css);
static int symDiff(const SSyncRaftNodeMap* l, const SSyncRaftNodeMap* r);
static void initProgress(SSyncRaftChanger* changer, SSyncRaftProgressTrackerConfig* config,
SSyncRaftProgressMap* progressMap, SyncNodeId id, bool isLearner);
static void nilAwareDelete(SSyncRaftNodeMap* nodeMap, SyncNodeId id);
static void nilAwareAdd(SSyncRaftNodeMap* nodeMap, SyncNodeId id);
static void makeVoter(SSyncRaftChanger* changer, SSyncRaftProgressTrackerConfig* config,
SSyncRaftProgressMap* progressMap, SyncNodeId id);
static void makeLearner(SSyncRaftChanger* changer, SSyncRaftProgressTrackerConfig* config,
SSyncRaftProgressMap* progressMap, SyncNodeId id);
static void removeNodeId(SSyncRaftChanger* changer, SSyncRaftProgressTrackerConfig* config,
SSyncRaftProgressMap* progressMap, SyncNodeId id);
// EnterJoint verifies that the outgoing (=right) majority config of the joint
// config is empty and initializes it with a copy of the incoming (=left)
// majority config. That is, it transitions from
//
// (1 2 3)&&()
// to
// (1 2 3)&&(1 2 3).
//
// The supplied changes are then applied to the incoming majority config,
// resulting in a joint configuration that in terms of the Raft thesis[1]
// (Section 4.3) corresponds to `C_{new,old}`.
//
// [1]: https://github.com/ongardie/dissertation/blob/master/online-trim.pdf
int syncRaftChangerEnterJoint(SSyncRaftChanger* changer, bool autoLeave, const SSyncConfChangeSingleArray* css,
SSyncRaftProgressTrackerConfig* config, SSyncRaftProgressMap* progressMap) {
int ret;
ret = checkAndCopy(changer, config, progressMap);
if (ret != 0) {
return ret;
}
if (hasJointConfig(config)) {
syncError("config is already joint");
return -1;
}
if(syncRaftJointConfigIsIncomingEmpty(&config->voters) == 0) {
// We allow adding nodes to an empty config for convenience (testing and
// bootstrap), but you can't enter a joint state.
syncError("can't make a zero-voter config joint");
return -1;
}
// Clear the outgoing config.
syncRaftJointConfigClearOutgoing(&config->voters);
// Copy incoming to outgoing.
syncRaftCopyNodeMap(&config->voters.incoming, &config->voters.outgoing);
ret = applyConfig(changer, config, progressMap, css);
if (ret != 0) {
return ret;
}
config->autoLeave = autoLeave;
return checkAndReturn(config, progressMap);
}
// Simple carries out a series of configuration changes that (in aggregate)
// mutates the incoming majority config Voters[0] by at most one. This method
// will return an error if that is not the case, if the resulting quorum is
// zero, or if the configuration is in a joint state (i.e. if there is an
// outgoing configuration).
int syncRaftChangerSimpleConfig(SSyncRaftChanger* changer, const SSyncConfChangeSingleArray* css,
SSyncRaftProgressTrackerConfig* config, SSyncRaftProgressMap* progressMap) {
int ret;
ret = checkAndCopy(changer, config, progressMap);
if (ret != 0) {
return ret;
}
if (hasJointConfig(config)) {
syncError("can't apply simple config change in joint config");
return -1;
}
ret = applyConfig(changer, config, progressMap, css);
if (ret != 0) {
return ret;
}
int n = symDiff(syncRaftJointConfigIncoming(&changer->tracker->config.voters),
syncRaftJointConfigIncoming(&config->voters));
if (n > 1) {
syncError("more than one voter changed without entering joint config");
return -1;
}
return checkAndReturn(config, progressMap);
}
// apply a change to the configuration. By convention, changes to voters are
// always made to the incoming majority config Voters[0]. Voters[1] is either
// empty or preserves the outgoing majority configuration while in a joint state.
static int applyConfig(SSyncRaftChanger* changer, SSyncRaftProgressTrackerConfig* config,
SSyncRaftProgressMap* progressMap, const SSyncConfChangeSingleArray* css) {
int i;
for (i = 0; i < css->n; ++i) {
const SSyncConfChangeSingle* cs = &(css->changes[i]);
if (cs->nodeId == SYNC_NON_NODE_ID) {
continue;
}
ESyncRaftConfChangeType type = cs->type;
switch (type) {
case SYNC_RAFT_Conf_AddNode:
makeVoter(changer, config, progressMap, cs->nodeId);
break;
case SYNC_RAFT_Conf_AddLearnerNode:
makeLearner(changer, config, progressMap, cs->nodeId);
break;
case SYNC_RAFT_Conf_RemoveNode:
removeNodeId(changer, config, progressMap, cs->nodeId);
break;
case SYNC_RAFT_Conf_UpdateNode:
break;
}
}
if (syncRaftJointConfigIsIncomingEmpty(&config->voters)) {
syncError("removed all voters");
return -1;
}
return 0;
}
// makeVoter adds or promotes the given ID to be a voter in the incoming
// majority config.
static void makeVoter(SSyncRaftChanger* changer, SSyncRaftProgressTrackerConfig* config,
SSyncRaftProgressMap* progressMap, SyncNodeId id) {
SSyncRaftProgress* progress = syncRaftFindProgressByNodeId(progressMap, id);
if (progress == NULL) {
initProgress(changer, config, progressMap, id, false);
return;
}
progress->isLearner = false;
nilAwareDelete(&config->learners, id);
nilAwareDelete(&config->learnersNext, id);
syncRaftJointConfigAddToIncoming(&config->voters, id);
}
// makeLearner makes the given ID a learner or stages it to be a learner once
// an active joint configuration is exited.
//
// The former happens when the peer is not a part of the outgoing config, in
// which case we either add a new learner or demote a voter in the incoming
// config.
//
// The latter case occurs when the configuration is joint and the peer is a
// voter in the outgoing config. In that case, we do not want to add the peer
// as a learner because then we'd have to track a peer as a voter and learner
// simultaneously. Instead, we add the learner to LearnersNext, so that it will
// be added to Learners the moment the outgoing config is removed by
// LeaveJoint().
static void makeLearner(SSyncRaftChanger* changer, SSyncRaftProgressTrackerConfig* config,
SSyncRaftProgressMap* progressMap, SyncNodeId id) {
SSyncRaftProgress* progress = syncRaftFindProgressByNodeId(progressMap, id);
if (progress == NULL) {
initProgress(changer, config, progressMap, id, true);
return;
}
if (progress->isLearner) {
return;
}
// Remove any existing voter in the incoming config...
removeNodeId(changer, config, progressMap, id);
// ... but save the Progress.
syncRaftAddToProgressMap(progressMap, progress);
// Use LearnersNext if we can't add the learner to Learners directly, i.e.
// if the peer is still tracked as a voter in the outgoing config. It will
// be turned into a learner in LeaveJoint().
//
// Otherwise, add a regular learner right away.
bool inInOutgoing = syncRaftJointConfigIsInOutgoing(&config->voters, id);
if (inInOutgoing) {
nilAwareAdd(&config->learnersNext, id);
} else {
nilAwareAdd(&config->learners, id);
progress->isLearner = true;
}
}
// removeNodeId this peer as a voter or learner from the incoming config.
static void removeNodeId(SSyncRaftChanger* changer, SSyncRaftProgressTrackerConfig* config,
SSyncRaftProgressMap* progressMap, SyncNodeId id) {
SSyncRaftProgress* progress = syncRaftFindProgressByNodeId(progressMap, id);
if (progress == NULL) {
return;
}
syncRaftJointConfigRemoveFromIncoming(&config->voters, id);
nilAwareDelete(&config->learners, id);
nilAwareDelete(&config->learnersNext, id);
// If the peer is still a voter in the outgoing config, keep the Progress.
bool inInOutgoing = syncRaftJointConfigIsInOutgoing(&config->voters, id);
if (!inInOutgoing) {
syncRaftRemoveFromProgressMap(progressMap, id);
}
}
// initProgress initializes a new progress for the given node or learner.
static void initProgress(SSyncRaftChanger* changer, SSyncRaftProgressTrackerConfig* config,
SSyncRaftProgressMap* progressMap, SyncNodeId id, bool isLearner) {
if (!isLearner) {
syncRaftJointConfigAddToIncoming(&config->voters, id);
} else {
nilAwareAdd(&config->learners, id);
}
SSyncRaftProgress* pProgress = (SSyncRaftProgress*)malloc(sizeof(SSyncRaftProgress));
assert (pProgress != NULL);
*pProgress = (SSyncRaftProgress) {
// Initializing the Progress with the last index means that the follower
// can be probed (with the last index).
//
// TODO(tbg): seems awfully optimistic. Using the first index would be
// better. The general expectation here is that the follower has no log
// at all (and will thus likely need a snapshot), though the app may
// have applied a snapshot out of band before adding the replica (thus
// making the first index the better choice).
.id = id,
.groupId = changer->tracker->pRaft->selfGroupId,
.nextIndex = changer->lastIndex,
.matchIndex = 0,
.state = PROGRESS_STATE_PROBE,
.pendingSnapshotIndex = 0,
.probeSent = false,
.inflights = syncRaftOpenInflights(changer->tracker->maxInflightMsgs),
.isLearner = isLearner,
// When a node is first added, we should mark it as recently active.
// Otherwise, CheckQuorum may cause us to step down if it is invoked
// before the added node has had a chance to communicate with us.
.recentActive = true,
.refCount = 0,
};
syncRaftAddToProgressMap(progressMap, pProgress);
}
// checkInvariants makes sure that the config and progress are compatible with
// each other. This is used to check both what the Changer is initialized with,
// as well as what it returns.
static int checkInvariants(SSyncRaftProgressTrackerConfig* config, SSyncRaftProgressMap* progressMap) {
int ret = syncRaftCheckTrackerConfigInProgress(config, progressMap);
if (ret != 0) {
return ret;
}
// Any staged learner was staged because it could not be directly added due
// to a conflicting voter in the outgoing config.
SyncNodeId* pNodeId = NULL;
while (!syncRaftIterateNodeMap(&config->learnersNext, pNodeId)) {
SyncNodeId nodeId = *pNodeId;
if (!syncRaftJointConfigInOutgoing(&config->voters, nodeId)) {
syncError("[%d] is in LearnersNext, but not outgoing", nodeId);
return -1;
}
SSyncRaftProgress* progress = syncRaftFindProgressByNodeId(progressMap, nodeId);
assert(progress);
assert(progress->id == nodeId);
if (progress->isLearner) {
syncError("[%d:%d] is in LearnersNext, but is already marked as learner", progress->groupId, nodeId);
return -1;
}
}
// Conversely Learners and Voters doesn't intersect at all.
pNodeId = NULL;
while (!syncRaftIterateNodeMap(&config->learners, pNodeId)) {
SyncNodeId nodeId = *pNodeId;
if (syncRaftJointConfigInOutgoing(&config->voters, nodeId)) {
syncError("%d is in Learners and outgoing", nodeId);
return -1;
}
SSyncRaftProgress* progress = syncRaftFindProgressByNodeId(progressMap, nodeId);
assert(progress);
assert(progress->id == nodeId);
if (!progress->isLearner) {
syncError("[%d:%d] is in Learners, but is not marked as learner", progress->groupId, nodeId);
return -1;
}
}
if (!hasJointConfig(config)) {
// We enforce that empty maps are nil instead of zero.
if (syncRaftNodeMapSize(&config->learnersNext) > 0) {
syncError("cfg.LearnersNext must be nil when not joint");
return -1;
}
if (config->autoLeave) {
syncError("AutoLeave must be false when not joint");
return -1;
}
}
return 0;
}
// checkAndCopy copies the tracker's config and progress map (deeply enough for
// the purposes of the Changer) and returns those copies. It returns an error
// if checkInvariants does.
static int checkAndCopy(SSyncRaftChanger* changer, SSyncRaftProgressTrackerConfig* config, SSyncRaftProgressMap* progressMap) {
syncRaftCopyTrackerConfig(&changer->tracker->config, config);
syncRaftClearProgressMap(progressMap);
SSyncRaftProgress* pProgress = NULL;
while (!syncRaftIterateProgressMap(&changer->tracker->progressMap, pProgress)) {
syncRaftAddToProgressMap(progressMap, pProgress);
}
return checkAndReturn(config, progressMap);
}
// checkAndReturn calls checkInvariants on the input and returns either the
// resulting error or the input.
static int checkAndReturn(SSyncRaftProgressTrackerConfig* config, SSyncRaftProgressMap* progressMap) {
if (checkInvariants(config, progressMap) != 0) {
return -1;
}
return 0;
}
static bool hasJointConfig(const SSyncRaftProgressTrackerConfig* config) {
return !syncRaftJointConfigIsOutgoingEmpty(&config->voters);
}
// symdiff returns the count of the symmetric difference between the sets of
// uint64s, i.e. len( (l - r) \union (r - l)).
static int symDiff(const SSyncRaftNodeMap* l, const SSyncRaftNodeMap* r) {
int n;
int i;
int j0, j1;
const SSyncRaftNodeMap* pairs[2][2] = {
{l, r}, // count elems in l but not in r
{r, l}, // count elems in r but not in l
};
for (n = 0, i = 0; i < 2; ++i) {
const SSyncRaftNodeMap** pp = pairs[i];
const SSyncRaftNodeMap* p0 = pp[0];
const SSyncRaftNodeMap* p1 = pp[1];
SyncNodeId* pNodeId;
while (!syncRaftIterateNodeMap(p0, pNodeId)) {
if (!syncRaftIsInNodeMap(p1, *pNodeId)) {
n+=1;
}
}
}
return n;
}
// nilAwareDelete deletes from a map, nil'ing the map itself if it is empty after.
static void nilAwareDelete(SSyncRaftNodeMap* nodeMap, SyncNodeId id) {
syncRaftRemoveFromNodeMap(nodeMap, id);
}
// nilAwareAdd populates a map entry, creating the map if necessary.
static void nilAwareAdd(SSyncRaftNodeMap* nodeMap, SyncNodeId id) {
syncRaftAddToNodeMap(nodeMap, id);
}

114
source/libs/sync/src/sync_raft_election.c
View File

@ -1,114 +0,0 @@
/*
* Copyright (c) 2019 TAOS Data, Inc. <cli@taosdata.com>
*
* This program is free software: you can use, redistribute, and/or modify
* it under the terms of the GNU Affero General Public License, version 3
* or later ("AGPL"), as published by the Free Software Foundation.
*
* This program is distributed in the hope that it will be useful, but WITHOUT
* ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or
* FITNESS FOR A PARTICULAR PURPOSE.
*
* You should have received a copy of the GNU Affero General Public License
* along with this program. If not, see <http://www.gnu.org/licenses/>.
*/
#include "syncInt.h"
#include "raft.h"
#include "raft_log.h"
#include "raft_message.h"
#include "sync_raft_progress_tracker.h"
void syncRaftStartElection(SSyncRaft* pRaft, ESyncRaftElectionType cType) {
if (pRaft->state == TAOS_SYNC_STATE_LEADER) {
syncDebug("[%d:%d] ignoring RAFT_MSG_INTERNAL_ELECTION because already leader", pRaft->selfGroupId, pRaft->selfId);
return;
}
if (!syncRaftIsPromotable(pRaft)) {
syncWarn("[%d:%d] is unpromotable and can not syncRaftCampaign", pRaft->selfGroupId, pRaft->selfId);
return;
}
// if there is pending uncommitted config,cannot start election
if (syncRaftLogNumOfPendingConf(pRaft->log) > 0 && syncRaftHasUnappliedLog(pRaft->log)) {
syncWarn("[%d:%d] cannot syncRaftStartElection at term %" PRId64 " since there are still pending configuration changes to apply",
pRaft->selfGroupId, pRaft->selfId, pRaft->term);
return;
}
syncInfo("[%d:%d] is starting a new election at term %" PRId64 "", pRaft->selfGroupId, pRaft->selfId, pRaft->term);
syncRaftCampaign(pRaft, cType);
}
// syncRaftCampaign transitions the raft instance to candidate state. This must only be
// called after verifying that this is a legitimate transition.
void syncRaftCampaign(SSyncRaft* pRaft, ESyncRaftElectionType cType) {
bool preVote;
SyncTerm term;
if (syncRaftIsPromotable(pRaft)) {
syncDebug("[%d:%d] is unpromotable; syncRaftCampaign() should have been called", pRaft->selfGroupId, pRaft->selfId);
return;
}
if (cType == SYNC_RAFT_CAMPAIGN_PRE_ELECTION) {
syncRaftBecomePreCandidate(pRaft);
preVote = true;
// PreVote RPCs are sent for the next term before we've incremented r.Term.
term = pRaft->term + 1;
} else {
syncRaftBecomeCandidate(pRaft);
term = pRaft->term;
preVote = false;
}
int quorum = syncRaftQuorum(pRaft);
ESyncRaftVoteResult result = syncRaftPollVote(pRaft, pRaft->selfId, preVote, true, NULL, NULL);
if (result == SYNC_RAFT_VOTE_WON) {
// We won the election after voting for ourselves (which must mean that
// this is a single-node cluster). Advance to the next state.
if (cType == SYNC_RAFT_CAMPAIGN_PRE_ELECTION) {
syncRaftStartElection(pRaft, SYNC_RAFT_CAMPAIGN_ELECTION);
} else {
syncRaftBecomeLeader(pRaft);
}
return;
}
// broadcast vote message to other peers
int i;
SyncIndex lastIndex = syncRaftLogLastIndex(pRaft->log);
SyncTerm lastTerm = syncRaftLogLastTerm(pRaft->log);
SSyncRaftNodeMap nodeMap;
syncRaftJointConfigIDs(&pRaft->tracker->config.voters, &nodeMap);
SyncNodeId *pNodeId = NULL;
while (!syncRaftIterateNodeMap(&nodeMap, pNodeId)) {
SyncNodeId nodeId = *pNodeId;
if (nodeId == SYNC_NON_NODE_ID) {
continue;
}
if (nodeId == pRaft->selfId) {
continue;
}
SNodeInfo* pNode = syncRaftGetNodeById(pRaft, nodeId);
if (pNode == NULL) {
continue;
}
SSyncMessage* pMsg = syncNewVoteMsg(pRaft->selfGroupId, pRaft->selfId,
term, cType, lastIndex, lastTerm);
if (pMsg == NULL) {
continue;
}
syncInfo("[%d:%d] [logterm: %" PRId64 ", index: %" PRId64 "] sent vote request to %d at term %" PRId64 "",
pRaft->selfGroupId, pRaft->selfId, lastTerm,
lastIndex, nodeId, pRaft->term);
pRaft->io.send(pMsg, pNode);
}
}

369
source/libs/sync/src/sync_raft_impl.c
View File

@ -1,369 +0,0 @@
/*
* Copyright (c) 2019 TAOS Data, Inc. <cli@taosdata.com>
*
* This program is free software: you can use, redistribute, and/or modify
* it under the terms of the GNU Affero General Public License, version 3
* or later ("AGPL"), as published by the Free Software Foundation.
*
* This program is distributed in the hope that it will be useful, but WITHOUT
* ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or
* FITNESS FOR A PARTICULAR PURPOSE.
*
* You should have received a copy of the GNU Affero General Public License
* along with this program. If not, see <http://www.gnu.org/licenses/>.
*/
#include "raft.h"
#include "sync_raft_impl.h"
#include "raft_log.h"
#include "raft_replication.h"
#include "sync_raft_progress_tracker.h"
#include "syncInt.h"
static int convertClear(SSyncRaft* pRaft);
static int stepFollower(SSyncRaft* pRaft, const SSyncMessage* pMsg);
static int stepCandidate(SSyncRaft* pRaft, const SSyncMessage* pMsg);
static int stepLeader(SSyncRaft* pRaft, const SSyncMessage* pMsg);
static bool increaseUncommittedSize(SSyncRaft* pRaft, SSyncRaftEntry* entries, int n);
static int triggerAll(SSyncRaft* pRaft);
static void tickElection(SSyncRaft* pRaft);
static void tickHeartbeat(SSyncRaft* pRaft);
static void appendEntries(SSyncRaft* pRaft, SSyncRaftEntry* entries, int n);
static void abortLeaderTransfer(SSyncRaft* pRaft);
static void resetRaft(SSyncRaft* pRaft, SyncTerm term);
void syncRaftBecomeFollower(SSyncRaft* pRaft, SyncTerm term, SyncNodeId leaderId) {
convertClear(pRaft);
pRaft->stepFp = stepFollower;
resetRaft(pRaft, term);
pRaft->tickFp = tickElection;
pRaft->leaderId = leaderId;
pRaft->state = TAOS_SYNC_STATE_FOLLOWER;
syncInfo("[%d:%d] became followe at term %" PRId64 "", pRaft->selfGroupId, pRaft->selfId, pRaft->term);
}
void syncRaftBecomePreCandidate(SSyncRaft* pRaft) {
convertClear(pRaft);
/**
* Becoming a pre-candidate changes our step functions and state,
* but doesn't change anything else. In particular it does not increase
* r.Term or change r.Vote.
**/
pRaft->stepFp = stepCandidate;
pRaft->tickFp = tickElection;
pRaft->state = TAOS_SYNC_STATE_CANDIDATE;
pRaft->candidateState.inPreVote = true;
syncInfo("[%d:%d] became pre-candidate at term %" PRId64 "", pRaft->selfGroupId, pRaft->selfId, pRaft->term);
}
void syncRaftBecomeCandidate(SSyncRaft* pRaft) {
convertClear(pRaft);
pRaft->candidateState.inPreVote = false;
pRaft->stepFp = stepCandidate;
// become candidate make term+1
resetRaft(pRaft, pRaft->term + 1);
pRaft->tickFp = tickElection;
pRaft->voteFor = pRaft->selfId;
pRaft->state = TAOS_SYNC_STATE_CANDIDATE;
syncInfo("[%d:%d] became candidate at term %" PRId64 "", pRaft->selfGroupId, pRaft->selfId, pRaft->term);
}
void syncRaftBecomeLeader(SSyncRaft* pRaft) {
assert(pRaft->state != TAOS_SYNC_STATE_FOLLOWER);
pRaft->stepFp = stepLeader;
resetRaft(pRaft, pRaft->term);
pRaft->leaderId = pRaft->leaderId;
pRaft->state = TAOS_SYNC_STATE_LEADER;
SSyncRaftProgress* progress = syncRaftFindProgressByNodeId(&pRaft->tracker->progressMap, pRaft->selfId);
assert(progress != NULL);
// Followers enter replicate mode when they've been successfully probed
// (perhaps after having received a snapshot as a result). The leader is
// trivially in this state. Note that r.reset() has initialized this
// progress with the last index already.
syncRaftProgressBecomeReplicate(progress);
// Conservatively set the pendingConfIndex to the last index in the
// log. There may or may not be a pending config change, but it's
// safe to delay any future proposals until we commit all our
// pending log entries, and scanning the entire tail of the log
// could be expensive.
SyncIndex lastIndex = syncRaftLogLastIndex(pRaft->log);
pRaft->pendingConfigIndex = lastIndex;
// after become leader, send a no-op log
SSyncRaftEntry* entry = (SSyncRaftEntry*)malloc(sizeof(SSyncRaftEntry));
if (entry == NULL) {
return;
}
*entry = (SSyncRaftEntry) {
.buffer = (SSyncBuffer) {
.data = NULL,
.len = 0,
}
};
appendEntries(pRaft, entry, 1);
//syncRaftTriggerHeartbeat(pRaft);
syncInfo("[%d:%d] became leader at term %" PRId64 "", pRaft->selfGroupId, pRaft->selfId, pRaft->term);
}
void syncRaftTriggerHeartbeat(SSyncRaft* pRaft) {
triggerAll(pRaft);
}
void syncRaftRandomizedElectionTimeout(SSyncRaft* pRaft) {
// electionTimeoutTick in [3,6] tick
pRaft->randomizedElectionTimeout = taosRand() % 4 + 3;
}
bool syncRaftIsPromotable(SSyncRaft* pRaft) {
return pRaft->selfId != SYNC_NON_NODE_ID;
}
bool syncRaftIsPastElectionTimeout(SSyncRaft* pRaft) {
return pRaft->electionElapsed >= pRaft->randomizedElectionTimeout;
}
int syncRaftQuorum(SSyncRaft* pRaft) {
return 0;
//return pRaft->cluster.replica / 2 + 1;
}
ESyncRaftVoteResult syncRaftPollVote(SSyncRaft* pRaft, SyncNodeId id,
bool preVote, bool grant,
int* rejected, int *granted) {
SNodeInfo* pNode = syncRaftGetNodeById(pRaft, id);
if (pNode == NULL) {
return true;
}
if (grant) {
syncInfo("[%d:%d] received grant (pre-vote %d) from %d at term %" PRId64 "",
pRaft->selfGroupId, pRaft->selfId, preVote, id, pRaft->term);
} else {
syncInfo("[%d:%d] received rejection (pre-vote %d) from %d at term %" PRId64 "",
pRaft->selfGroupId, pRaft->selfId, preVote, id, pRaft->term);
}
syncRaftRecordVote(pRaft->tracker, pNode->nodeId, grant);
return syncRaftTallyVotes(pRaft->tracker, rejected, granted);
}
/*
if (accept) {
syncInfo("[%d:%d] received (pre-vote %d) from %d at term %" PRId64 "",
pRaft->selfGroupId, pRaft->selfId, preVote, id, pRaft->term);
} else {
syncInfo("[%d:%d] received rejection from %d at term %" PRId64 "",
pRaft->selfGroupId, pRaft->selfId, id, pRaft->term);
}
int voteIndex = syncRaftGetNodeById(pRaft, id);
assert(voteIndex < pRaft->cluster.replica && voteIndex >= 0);
assert(pRaft->candidateState.votes[voteIndex] == SYNC_RAFT_VOTE_RESP_UNKNOWN);
pRaft->candidateState.votes[voteIndex] = accept ? SYNC_RAFT_VOTE_RESP_GRANT : SYNC_RAFT_VOTE_RESP_REJECT;
int granted = 0, rejected = 0;
int i;
for (i = 0; i < pRaft->cluster.replica; ++i) {
if (pRaft->candidateState.votes[i] == SYNC_RAFT_VOTE_RESP_GRANT) granted++;
else if (pRaft->candidateState.votes[i] == SYNC_RAFT_VOTE_RESP_REJECT) rejected++;
}
if (rejectNum) *rejectNum = rejected;
return granted;
*/
void syncRaftLoadState(SSyncRaft* pRaft, const SSyncServerState* serverState) {
SyncIndex commitIndex = serverState->commitIndex;
SyncIndex lastIndex = syncRaftLogLastIndex(pRaft->log);
if (commitIndex < pRaft->log->commitIndex || commitIndex > lastIndex) {
syncFatal("[%d:%d] state.commit %"PRId64" is out of range [%" PRId64 ",%" PRId64 "",
pRaft->selfGroupId, pRaft->selfId, commitIndex, pRaft->log->commitIndex, lastIndex);
return;
}
pRaft->log->commitIndex = commitIndex;
pRaft->term = serverState->term;
pRaft->voteFor = serverState->voteFor;
}
static void visitProgressSendAppend(SSyncRaftProgress* progress, void* arg) {
SSyncRaft* pRaft = (SSyncRaft*)arg;
if (pRaft->selfId == progress->id) {
return;
}
syncRaftMaybeSendAppend(arg, progress, true);
}
// bcastAppend sends RPC, with entries to all peers that are not up-to-date
// according to the progress recorded in r.prs.
void syncRaftBroadcastAppend(SSyncRaft* pRaft) {
syncRaftProgressVisit(pRaft->tracker, visitProgressSendAppend, pRaft);
}
SNodeInfo* syncRaftGetNodeById(SSyncRaft *pRaft, SyncNodeId id) {
SNodeInfo **ppNode = taosHashGet(pRaft->nodeInfoMap, &id, sizeof(SyncNodeId*));
if (ppNode != NULL) {
return *ppNode;
}
return NULL;
}
static int convertClear(SSyncRaft* pRaft) {
}
static int stepFollower(SSyncRaft* pRaft, const SSyncMessage* pMsg) {
return 0;
}
static int stepCandidate(SSyncRaft* pRaft, const SSyncMessage* pMsg) {
/**
* Only handle vote responses corresponding to our candidacy (while in
* StateCandidate, we may get stale MsgPreVoteResp messages in this term from
* our pre-candidate state).
**/
ESyncRaftMessageType msgType = pMsg->msgType;
if (msgType == RAFT_MSG_INTERNAL_PROP) {
return 0;
}
if (msgType == RAFT_MSG_VOTE_RESP) {
syncRaftHandleVoteRespMessage(pRaft, pMsg);
return 0;
} else if (msgType == RAFT_MSG_APPEND) {
syncRaftBecomeFollower(pRaft, pMsg->term, pMsg->from);
syncRaftHandleAppendEntriesMessage(pRaft, pMsg);
}
return 0;
}
static int stepLeader(SSyncRaft* pRaft, const SSyncMessage* pMsg) {
convertClear(pRaft);
return 0;
}
// tickElection is run by followers and candidates after r.electionTimeout.
static void tickElection(SSyncRaft* pRaft) {
pRaft->electionElapsed += 1;
if (!syncRaftIsPromotable(pRaft)) {
return;
}
if (!syncRaftIsPastElectionTimeout(pRaft)) {
return;
}
// election timeout
pRaft->electionElapsed = 0;
SSyncMessage msg;
syncRaftStep(pRaft, syncInitElectionMsg(&msg, pRaft->selfId));
}
// tickHeartbeat is run by leaders to send a MsgBeat after r.heartbeatTimeout.
static void tickHeartbeat(SSyncRaft* pRaft) {
}
// TODO
static bool increaseUncommittedSize(SSyncRaft* pRaft, SSyncRaftEntry* entries, int n) {
return false;
}
static void appendEntries(SSyncRaft* pRaft, SSyncRaftEntry* entries, int n) {
SyncIndex lastIndex = syncRaftLogLastIndex(pRaft->log);
SyncTerm term = pRaft->term;
int i;
for (i = 0; i < n; ++i) {
entries[i].term = term;
entries[i].index = lastIndex + 1 + i;
}
// Track the size of this uncommitted proposal.
if (!increaseUncommittedSize(pRaft, entries, n)) {
// Drop the proposal.
return;
}
syncRaftLogAppend(pRaft->log, entries, n);
SSyncRaftProgress* progress = syncRaftFindProgressByNodeId(&pRaft->tracker->progressMap, pRaft->selfId);
assert(progress != NULL);
syncRaftProgressMaybeUpdate(progress, lastIndex);
// Regardless of syncRaftMaybeCommit's return, our caller will call bcastAppend.
syncRaftMaybeCommit(pRaft);
}
// syncRaftMaybeCommit attempts to advance the commit index. Returns true if
// the commit index changed (in which case the caller should call
// r.bcastAppend).
bool syncRaftMaybeCommit(SSyncRaft* pRaft) {
return true;
}
/**
* trigger I/O requests for newly appended log entries or heartbeats.
**/
static int triggerAll(SSyncRaft* pRaft) {
#if 0
assert(pRaft->state == TAOS_SYNC_STATE_LEADER);
int i;
for (i = 0; i < pRaft->cluster.replica; ++i) {
if (i == pRaft->cluster.selfIndex) {
continue;
}
syncRaftMaybeSendAppend(pRaft, pRaft->tracker->progressMap.progress[i], true);
}
#endif
return 0;
}
static void abortLeaderTransfer(SSyncRaft* pRaft) {
pRaft->leadTransferee = SYNC_NON_NODE_ID;
}
static void resetProgress(SSyncRaftProgress* progress, void* arg) {
syncRaftResetProgress((SSyncRaft*)arg, progress);
}
static void resetRaft(SSyncRaft* pRaft, SyncTerm term) {
if (pRaft->term != term) {
pRaft->term = term;
pRaft->voteFor = SYNC_NON_NODE_ID;
}
pRaft->leaderId = SYNC_NON_NODE_ID;
pRaft->electionElapsed = 0;
pRaft->heartbeatElapsed = 0;
syncRaftRandomizedElectionTimeout(pRaft);
abortLeaderTransfer(pRaft);
syncRaftResetVotes(pRaft->tracker);
syncRaftProgressVisit(pRaft->tracker, resetProgress, pRaft);
pRaft->pendingConfigIndex = 0;
pRaft->uncommittedSize = 0;
}

97
source/libs/sync/src/sync_raft_inflights.c
View File

@ -1,97 +0,0 @@
/*
* Copyright (c) 2019 TAOS Data, Inc. <cli@taosdata.com>
*
* This program is free software: you can use, redistribute, and/or modify
* it under the terms of the GNU Affero General Public License, version 3
* or later ("AGPL"), as published by the Free Software Foundation.
*
* This program is distributed in the hope that it will be useful, but WITHOUT
* ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or
* FITNESS FOR A PARTICULAR PURPOSE.
*
* You should have received a copy of the GNU Affero General Public License
* along with this program. If not, see <http: *www.gnu.org/licenses/>.
*/
#include "sync_raft_inflights.h"
SSyncRaftInflights* syncRaftOpenInflights(int size) {
SSyncRaftInflights* inflights = (SSyncRaftInflights*)malloc(sizeof(SSyncRaftInflights));
if (inflights == NULL) {
return NULL;
}
SyncIndex* buffer = (SyncIndex*)malloc(sizeof(SyncIndex) * size);
if (buffer == NULL) {
free(inflights);
return NULL;
}
*inflights = (SSyncRaftInflights) {
.buffer = buffer,
.count = 0,
.size = 0,
.start = 0,
};
return inflights;
}
void syncRaftCloseInflights(SSyncRaftInflights* inflights) {
free(inflights->buffer);
free(inflights);
}
// Add notifies the Inflights that a new message with the given index is being
// dispatched. Full() must be called prior to Add() to verify that there is room
// for one more message, and consecutive calls to add Add() must provide a
// monotonic sequence of indexes.
void syncRaftInflightAdd(SSyncRaftInflights* inflights, SyncIndex inflightIndex) {
assert(!syncRaftInflightFull(inflights));
int next = inflights->start + inflights->count;
int size = inflights->size;
if (next >= size) {
next -= size;
}
inflights->buffer[next] = inflightIndex;
inflights->count++;
}
// FreeLE frees the inflights smaller or equal to the given `to` flight.
void syncRaftInflightFreeLE(SSyncRaftInflights* inflights, SyncIndex toIndex) {
if (inflights->count == 0 || toIndex < inflights->buffer[inflights->start]) {
// out of the left side of the window
return;
}
int i, idx;
for (i = 0, idx = inflights->start; i < inflights->count; i++) {
if (toIndex < inflights->buffer[idx]) { // found the first large inflight
break;
}
// increase index and maybe rotate
int size = inflights->size;
idx++;
if (idx >= size) {
idx -= size;
}
}
// free i inflights and set new start index
inflights->count -= i;
inflights->start = idx;
assert(inflights->count >= 0);
if (inflights->count == 0) {
// inflights is empty, reset the start index so that we don't grow the
// buffer unnecessarily.
inflights->start = 0;
}
}
// FreeFirstOne releases the first inflight. This is a no-op if nothing is
// inflight.
void syncRaftInflightFreeFirstOne(SSyncRaftInflights* inflights) {
syncRaftInflightFreeLE(inflights, inflights->buffer[inflights->start]);
}

82
source/libs/sync/src/sync_raft_node_map.c
View File

@ -1,82 +0,0 @@
/*
* Copyright (c) 2019 TAOS Data, Inc. <cli@taosdata.com>
*
* This program is free software: you can use, redistribute, and/or modify
* it under the terms of the GNU Affero General Public License, version 3
* or later ("AGPL"), as published by the Free Software Foundation.
*
* This program is distributed in the hope that it will be useful, but WITHOUT
* ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or
* FITNESS FOR A PARTICULAR PURPOSE.
*
* You should have received a copy of the GNU Affero General Public License
* along with this program. If not, see <http://www.gnu.org/licenses/>.
*/
#include "sync_raft_node_map.h"
#include "sync_type.h"
#include "sync_raft_progress.h"
void syncRaftInitNodeMap(SSyncRaftNodeMap* nodeMap) {
nodeMap->nodeIdMap = taosHashInit(TSDB_MAX_REPLICA, taosGetDefaultHashFunction(TSDB_DATA_TYPE_INT), true, HASH_ENTRY_LOCK);
}
void syncRaftFreeNodeMap(SSyncRaftNodeMap* nodeMap) {
taosHashCleanup(nodeMap->nodeIdMap);
}
void syncRaftClearNodeMap(SSyncRaftNodeMap* nodeMap) {
taosHashClear(nodeMap->nodeIdMap);
}
bool syncRaftIsInNodeMap(const SSyncRaftNodeMap* nodeMap, SyncNodeId nodeId) {
SyncNodeId** ppId = (SyncNodeId**)taosHashGet(nodeMap->nodeIdMap, &nodeId, sizeof(SyncNodeId*));
if (ppId == NULL) {
return false;
}
return true;
}
void syncRaftCopyNodeMap(SSyncRaftNodeMap* from, SSyncRaftNodeMap* to) {
SyncNodeId *pId = NULL;
while (!syncRaftIterateNodeMap(from, pId)) {
taosHashPut(to->nodeIdMap, &pId, sizeof(SyncNodeId*), &pId, sizeof(SyncNodeId*));
}
}
bool syncRaftIterateNodeMap(const SSyncRaftNodeMap* nodeMap, SyncNodeId *pId) {
SyncNodeId **ppId = taosHashIterate(nodeMap->nodeIdMap, pId);
if (ppId == NULL) {
return true;
}
*pId = *(*ppId);
return false;
}
bool syncRaftIsAllNodeInProgressMap(SSyncRaftNodeMap* nodeMap, SSyncRaftProgressMap* progressMap) {
SyncNodeId *pId = NULL;
while (!syncRaftIterateNodeMap(nodeMap, pId)) {
if (!syncRaftIsInProgressMap(progressMap, *pId)) {
return false;
}
}
return true;
}
void syncRaftUnionNodeMap(SSyncRaftNodeMap* nodeMap, SSyncRaftNodeMap* to) {
syncRaftCopyNodeMap(nodeMap, to);
}
void syncRaftAddToNodeMap(SSyncRaftNodeMap* nodeMap, SyncNodeId nodeId) {
taosHashPut(nodeMap->nodeIdMap, &nodeId, sizeof(SyncNodeId*), &nodeId, sizeof(SyncNodeId*));
}
void syncRaftRemoveFromNodeMap(SSyncRaftNodeMap* nodeMap, SyncNodeId nodeId) {
taosHashRemove(nodeMap->nodeIdMap, &nodeId, sizeof(SyncNodeId*));
}
int32_t syncRaftNodeMapSize(const SSyncRaftNodeMap* nodeMap) {
return taosHashGetSize(nodeMap->nodeIdMap);
}

260
source/libs/sync/src/sync_raft_progress.c
View File

@ -1,260 +0,0 @@
/*
* Copyright (c) 2019 TAOS Data, Inc. <cli@taosdata.com>
*
* This program is free software: you can use, redistribute, and/or modify
* it under the terms of the GNU Affero General Public License, version 3
* or later ("AGPL"), as published by the Free Software Foundation.
*
* This program is distributed in the hope that it will be useful, but WITHOUT
* ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or
* FITNESS FOR A PARTICULAR PURPOSE.
*
* You should have received a copy of the GNU Affero General Public License
* along with this program. If not, see <http://www.gnu.org/licenses/>.
*/
#include "raft.h"
#include "raft_log.h"
#include "sync_raft_progress.h"
#include "sync_raft_progress_tracker.h"
#include "sync.h"
#include "syncInt.h"
static void copyProgress(SSyncRaftProgress* progress, void* arg);
static void refProgress(SSyncRaftProgress* progress);
static void unrefProgress(SSyncRaftProgress* progress, void*);
static void resetProgressState(SSyncRaftProgress* progress, ESyncRaftProgressState state);
static void probeAcked(SSyncRaftProgress* progress);
static void resumeProgress(SSyncRaftProgress* progress);
void syncRaftResetProgress(SSyncRaft* pRaft, SSyncRaftProgress* progress) {
if (progress->inflights) {
syncRaftCloseInflights(progress->inflights);
}
SSyncRaftInflights* inflights = syncRaftOpenInflights(pRaft->tracker->maxInflightMsgs);
if (inflights == NULL) {
return;
}
*progress = (SSyncRaftProgress) {
.matchIndex = progress->id == pRaft->selfId ? syncRaftLogLastIndex(pRaft->log) : 0,
.nextIndex = syncRaftLogLastIndex(pRaft->log) + 1,
.inflights = inflights,
.isLearner = false,
.state = PROGRESS_STATE_PROBE,
};
}
// MaybeUpdate is called when an MsgAppResp arrives from the follower, with the
// index acked by it. The method returns false if the given n index comes from
// an outdated message. Otherwise it updates the progress and returns true.
bool syncRaftProgressMaybeUpdate(SSyncRaftProgress* progress, SyncIndex lastIndex) {
bool updated = false;
if (progress->matchIndex < lastIndex) {
progress->matchIndex = lastIndex;
updated = true;
probeAcked(progress);
}
progress->nextIndex = TMAX(progress->nextIndex, lastIndex + 1);
return updated;
}
// MaybeDecrTo adjusts the Progress to the receipt of a MsgApp rejection. The
// arguments are the index of the append message rejected by the follower, and
// the hint that we want to decrease to.
//
// Rejections can happen spuriously as messages are sent out of order or
// duplicated. In such cases, the rejection pertains to an index that the
// Progress already knows were previously acknowledged, and false is returned
// without changing the Progress.
//
// If the rejection is genuine, Next is lowered sensibly, and the Progress is
// cleared for sending log entries.
bool syncRaftProgressMaybeDecrTo(SSyncRaftProgress* progress,
SyncIndex rejected, SyncIndex matchHint) {
if (progress->state == PROGRESS_STATE_REPLICATE) {
// The rejection must be stale if the progress has matched and "rejected"
// is smaller than "match".
if (rejected <= progress->matchIndex) {
syncDebug("match index is up to date,ignore");
return false;
}
// Directly decrease next to match + 1.
//
// TODO(tbg): why not use matchHint if it's larger?
progress->nextIndex = progress->matchIndex + 1;
return true;
}
// The rejection must be stale if "rejected" does not match next - 1. This
// is because non-replicating followers are probed one entry at a time.
if (rejected != progress->nextIndex - 1) {
syncDebug("rejected index %" PRId64 " different from next index %" PRId64 " -> ignore"
, rejected, progress->nextIndex);
return false;
}
progress->nextIndex = TMAX(TMIN(rejected, matchHint + 1), 1);
progress->probeSent = false;
return true;
}
// IsPaused returns whether sending log entries to this node has been throttled.
// This is done when a node has rejected recent MsgApps, is currently waiting
// for a snapshot, or has reached the MaxInflightMsgs limit. In normal
// operation, this is false. A throttled node will be contacted less frequently
// until it has reached a state in which it's able to accept a steady stream of
// log entries again.
bool syncRaftProgressIsPaused(SSyncRaftProgress* progress) {
switch (progress->state) {
case PROGRESS_STATE_PROBE:
return progress->probeSent;
case PROGRESS_STATE_REPLICATE:
return syncRaftInflightFull(progress->inflights);
case PROGRESS_STATE_SNAPSHOT:
return true;
default:
syncFatal("error sync state:%d", progress->state);
}
}
SSyncRaftProgress* syncRaftFindProgressByNodeId(const SSyncRaftProgressMap* progressMap, SyncNodeId id) {
SSyncRaftProgress** ppProgress = (SSyncRaftProgress**)taosHashGet(progressMap->progressMap, &id, sizeof(SyncNodeId*));
if (ppProgress == NULL) {
return NULL;
}
return *ppProgress;
}
int syncRaftAddToProgressMap(SSyncRaftProgressMap* progressMap, SSyncRaftProgress* progress) {
refProgress(progress);
taosHashPut(progressMap->progressMap, &progress->id, sizeof(SyncNodeId*), &progress, sizeof(SSyncRaftProgress*));
}
void syncRaftRemoveFromProgressMap(SSyncRaftProgressMap* progressMap, SyncNodeId id) {
SSyncRaftProgress** ppProgress = (SSyncRaftProgress**)taosHashGet(progressMap->progressMap, &id, sizeof(SyncNodeId*));
if (ppProgress == NULL) {
return;
}
unrefProgress(*ppProgress, NULL);
taosHashRemove(progressMap->progressMap, &id, sizeof(SyncNodeId*));
}
bool syncRaftIsInProgressMap(SSyncRaftProgressMap* progressMap, SyncNodeId id) {
return taosHashGet(progressMap->progressMap, &id, sizeof(SyncNodeId*)) != NULL;
}
bool syncRaftProgressIsUptodate(SSyncRaft* pRaft, SSyncRaftProgress* progress) {
return syncRaftLogLastIndex(pRaft->log) + 1 == progress->nextIndex;
}
// BecomeProbe transitions into StateProbe. Next is reset to Match+1 or,
// optionally and if larger, the index of the pending snapshot.
void syncRaftProgressBecomeProbe(SSyncRaftProgress* progress) {
// If the original state is StateSnapshot, progress knows that
// the pending snapshot has been sent to this peer successfully, then
// probes from pendingSnapshot + 1.
if (progress->state == PROGRESS_STATE_SNAPSHOT) {
SyncIndex pendingSnapshotIndex = progress->pendingSnapshotIndex;
resetProgressState(progress, PROGRESS_STATE_PROBE);
progress->nextIndex = TMAX(progress->matchIndex + 1, pendingSnapshotIndex + 1);
} else {
resetProgressState(progress, PROGRESS_STATE_PROBE);
progress->nextIndex = progress->matchIndex + 1;
}
}
// BecomeReplicate transitions into StateReplicate, resetting Next to Match+1.
void syncRaftProgressBecomeReplicate(SSyncRaftProgress* progress) {
resetProgressState(progress, PROGRESS_STATE_REPLICATE);
progress->nextIndex = progress->matchIndex + 1;
}
// BecomeSnapshot moves the Progress to StateSnapshot with the specified pending
// snapshot index.
void syncRaftProgressBecomeSnapshot(SSyncRaftProgress* progress, SyncIndex snapshotIndex) {
resetProgressState(progress, PROGRESS_STATE_SNAPSHOT);
progress->pendingSnapshotIndex = snapshotIndex;
}
void syncRaftCopyProgress(const SSyncRaftProgress* progress, SSyncRaftProgress* out) {
memcpy(out, progress, sizeof(SSyncRaftProgress));
}
void syncRaftInitProgressMap(SSyncRaftProgressMap* progressMap) {
progressMap->progressMap = taosHashInit(TSDB_MAX_REPLICA, taosGetDefaultHashFunction(TSDB_DATA_TYPE_INT), true, HASH_ENTRY_LOCK);
}
void syncRaftFreeProgressMap(SSyncRaftProgressMap* progressMap) {
syncRaftVisitProgressMap(progressMap, unrefProgress, NULL);
taosHashCleanup(progressMap->progressMap);
}
void syncRaftClearProgressMap(SSyncRaftProgressMap* progressMap) {
taosHashClear(progressMap->progressMap);
}
void syncRaftCopyProgressMap(SSyncRaftProgressMap* from, SSyncRaftProgressMap* to) {
syncRaftVisitProgressMap(from, copyProgress, to);
}
bool syncRaftIterateProgressMap(const SSyncRaftProgressMap* progressMap, SSyncRaftProgress *pProgress) {
SSyncRaftProgress **ppProgress = taosHashIterate(progressMap->progressMap, pProgress);
if (ppProgress == NULL) {
return true;
}
*pProgress = *(*ppProgress);
return false;
}
bool syncRaftVisitProgressMap(SSyncRaftProgressMap* progressMap, visitProgressFp fp, void* arg) {
SSyncRaftProgress *pProgress;
while (!syncRaftIterateProgressMap(progressMap, pProgress)) {
fp(pProgress, arg);
}
}
static void copyProgress(SSyncRaftProgress* progress, void* arg) {
assert(progress->refCount > 0);
SSyncRaftProgressMap* to = (SSyncRaftProgressMap*)arg;
syncRaftAddToProgressMap(to, progress);
}
static void refProgress(SSyncRaftProgress* progress) {
progress->refCount += 1;
}
static void unrefProgress(SSyncRaftProgress* progress, void* arg) {
(void)arg;
progress->refCount -= 1;
assert(progress->refCount >= 0);
if (progress->refCount == 0) {
free(progress);
}
}
// ResetState moves the Progress into the specified State, resetting ProbeSent,
// PendingSnapshot, and Inflights.
static void resetProgressState(SSyncRaftProgress* progress, ESyncRaftProgressState state) {
progress->probeSent = false;
progress->pendingSnapshotIndex = 0;
progress->state = state;
syncRaftInflightReset(progress->inflights);
}
// ProbeAcked is called when this peer has accepted an append. It resets
// ProbeSent to signal that additional append messages should be sent without
// further delay.
static void probeAcked(SSyncRaftProgress* progress) {
progress->probeSent = false;
}

156
source/libs/sync/src/sync_raft_progress_tracker.c
View File

@ -1,156 +0,0 @@
/*
* Copyright (c) 2019 TAOS Data, Inc. <cli@taosdata.com>
*
* This program is free software: you can use, redistribute, and/or modify
* it under the terms of the GNU Affero General Public License, version 3
* or later ("AGPL"), as published by the Free Software Foundation.
*
* This program is distributed in the hope that it will be useful, but WITHOUT
* ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or
* FITNESS FOR A PARTICULAR PURPOSE.
*
* You should have received a copy of the GNU Affero General Public License
* along with this program. If not, see <http://www.gnu.org/licenses/>.
*/
#include "raft.h"
#include "sync_const.h"
#include "sync_raft_progress_tracker.h"
#include "sync_raft_proto.h"
SSyncRaftProgressTracker* syncRaftOpenProgressTracker(SSyncRaft* pRaft) {
SSyncRaftProgressTracker* tracker = (SSyncRaftProgressTracker*)malloc(sizeof(SSyncRaftProgressTracker));
if (tracker == NULL) {
return NULL;
}
tracker->votesMap = taosHashInit(TSDB_MAX_REPLICA, taosGetDefaultHashFunction(TSDB_DATA_TYPE_INT), true, HASH_ENTRY_LOCK);
syncRaftInitTrackConfig(&tracker->config);
tracker->pRaft = pRaft;
tracker->maxInflightMsgs = kSyncRaftMaxInflghtMsgs;
return tracker;
}
void syncRaftInitTrackConfig(SSyncRaftProgressTrackerConfig* config) {
syncRaftInitNodeMap(&config->learners);
syncRaftInitNodeMap(&config->learnersNext);
syncRaftInitQuorumJointConfig(&config->voters);
config->autoLeave = false;
}
void syncRaftFreeTrackConfig(SSyncRaftProgressTrackerConfig* config) {
syncRaftFreeNodeMap(&config->learners);
syncRaftFreeNodeMap(&config->learnersNext);
syncRaftFreeNodeMap(&config->voters.incoming);
syncRaftFreeNodeMap(&config->voters.outgoing);
}
// ResetVotes prepares for a new round of vote counting via recordVote.
void syncRaftResetVotes(SSyncRaftProgressTracker* tracker) {
taosHashClear(tracker->votesMap);
}
void syncRaftProgressVisit(SSyncRaftProgressTracker* tracker, visitProgressFp visit, void* arg) {
syncRaftVisitProgressMap(&tracker->progressMap, visit, arg);
}
// RecordVote records that the node with the given id voted for this Raft
// instance if v == true (and declined it otherwise).
void syncRaftRecordVote(SSyncRaftProgressTracker* tracker, SyncNodeId id, bool grant) {
ESyncRaftVoteType* pType = taosHashGet(tracker->votesMap, &id, sizeof(SyncNodeId*));
if (pType != NULL) {
return;
}
taosHashPut(tracker->votesMap, &id, sizeof(SyncNodeId), &grant, sizeof(bool*));
}
void syncRaftCopyTrackerConfig(const SSyncRaftProgressTrackerConfig* from, SSyncRaftProgressTrackerConfig* to) {
memcpy(to, from, sizeof(SSyncRaftProgressTrackerConfig));
}
int syncRaftCheckTrackerConfigInProgress(SSyncRaftProgressTrackerConfig* config, SSyncRaftProgressMap* progressMap) {
// NB: intentionally allow the empty config. In production we'll never see a
// non-empty config (we prevent it from being created) but we will need to
// be able to *create* an initial config, for example during bootstrap (or
// during tests). Instead of having to hand-code this, we allow
// transitioning from an empty config into any other legal and non-empty
// config.
if (!syncRaftIsAllNodeInProgressMap(&config->voters.incoming, progressMap)) return -1;
if (!syncRaftIsAllNodeInProgressMap(&config->voters.outgoing, progressMap)) return -1;
if (!syncRaftIsAllNodeInProgressMap(&config->learners, progressMap)) return -1;
if (!syncRaftIsAllNodeInProgressMap(&config->learnersNext, progressMap)) return -1;
return 0;
}
// TallyVotes returns the number of granted and rejected Votes, and whether the
// election outcome is known.
ESyncRaftVoteResult syncRaftTallyVotes(SSyncRaftProgressTracker* tracker, int* rejected, int *granted) {
SSyncRaftProgress* progress = NULL;
int r, g;
// Make sure to populate granted/rejected correctly even if the Votes slice
// contains members no longer part of the configuration. This doesn't really
// matter in the way the numbers are used (they're informational), but might
// as well get it right.
while (!syncRaftIterateProgressMap(&tracker->progressMap, progress)) {
if (progress->id == SYNC_NON_NODE_ID) {
continue;
}
bool* v = taosHashGet(tracker->votesMap, &progress->id, sizeof(SyncNodeId*));
if (v == NULL) {
continue;
}
if (*v) {
g++;
} else {
r++;
}
}
if (rejected) *rejected = r;
if (granted) *granted = g;
return syncRaftVoteResult(&(tracker->config.voters), tracker->votesMap);
}
void syncRaftConfigState(SSyncRaftProgressTracker* tracker, SSyncConfigState* cs) {
syncRaftCopyNodeMap(&tracker->config.voters.incoming, &cs->voters);
syncRaftCopyNodeMap(&tracker->config.voters.outgoing, &cs->votersOutgoing);
syncRaftCopyNodeMap(&tracker->config.learners, &cs->learners);
syncRaftCopyNodeMap(&tracker->config.learnersNext, &cs->learnersNext);
cs->autoLeave = tracker->config.autoLeave;
}
static void matchAckIndexer(SyncNodeId id, void* arg, SyncIndex* index) {
SSyncRaftProgressTracker* tracker = (SSyncRaftProgressTracker*)arg;
SSyncRaftProgress* progress = syncRaftFindProgressByNodeId(&tracker->progressMap, id);
if (progress == NULL) {
*index = 0;
return;
}
*index = progress->matchIndex;
}
// Committed returns the largest log index known to be committed based on what
// the voting members of the group have acknowledged.
SyncIndex syncRaftCommittedIndex(SSyncRaftProgressTracker* tracker) {
return syncRaftJointConfigCommittedIndex(&tracker->config.voters, matchAckIndexer, tracker);
}
static void visitProgressActive(SSyncRaftProgress* progress, void* arg) {
SHashObj* votesMap = (SHashObj*)arg;
taosHashPut(votesMap, &progress->id, sizeof(SyncNodeId), &progress->recentActive, sizeof(bool));
}
// QuorumActive returns true if the quorum is active from the view of the local
// raft state machine. Otherwise, it returns false.
bool syncRaftQuorumActive(SSyncRaftProgressTracker* tracker) {
SHashObj* votesMap = taosHashInit(TSDB_MAX_REPLICA, taosGetDefaultHashFunction(TSDB_DATA_TYPE_INT), true, HASH_ENTRY_LOCK);
syncRaftVisitProgressMap(&tracker->progressMap, visitProgressActive, votesMap);
return syncRaftVoteResult(&tracker->config.voters, votesMap) == SYNC_RAFT_VOTE_WON;
}

75
source/libs/sync/src/sync_raft_quorum_joint.c
View File

@ -1,75 +0,0 @@
/*
* Copyright (c) 2019 TAOS Data, Inc. <cli@taosdata.com>
*
* This program is free software: you can use, redistribute, and/or modify
* it under the terms of the GNU Affero General Public License, version 3
* or later ("AGPL"), as published by the Free Software Foundation.
*
* This program is distributed in the hope that it will be useful, but WITHOUT
* ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or
* FITNESS FOR A PARTICULAR PURPOSE.
*
* You should have received a copy of the GNU Affero General Public License
* along with this program. If not, see <http://www.gnu.org/licenses/>.
*/
#include "sync_raft_node_map.h"
#include "sync_raft_quorum_majority.h"
#include "sync_raft_quorum_joint.h"
#include "sync_raft_quorum.h"
/**
* syncRaftVoteResult takes a mapping of voters to yes/no (true/false) votes and returns
* a result indicating whether the vote is pending, lost, or won. A joint quorum
* requires both majority quorums to vote in favor.
**/
ESyncRaftVoteType syncRaftVoteResult(SSyncRaftQuorumJointConfig* config, SHashObj* votesMap) {
ESyncRaftVoteResult r1 = syncRaftMajorityVoteResult(&(config->incoming), votesMap);
ESyncRaftVoteResult r2 = syncRaftMajorityVoteResult(&(config->outgoing), votesMap);
if (r1 == r2) {
// If they agree, return the agreed state.
return r1;
}
if (r1 == SYNC_RAFT_VOTE_LOST || r2 == SYNC_RAFT_VOTE_LOST) {
// If either config has lost, loss is the only possible outcome.
return SYNC_RAFT_VOTE_LOST;
}
// One side won, the other one is pending, so the whole outcome is.
return SYNC_RAFT_VOTE_PENDING;
}
void syncRaftInitQuorumJointConfig(SSyncRaftQuorumJointConfig* config) {
syncRaftInitNodeMap(&config->incoming);
syncRaftInitNodeMap(&config->outgoing);
}
void syncRaftFreeQuorumJointConfig(SSyncRaftQuorumJointConfig* config) {
syncRaftFreeNodeMap(&config->incoming);
syncRaftFreeNodeMap(&config->outgoing);
}
void syncRaftJointConfigAddToIncoming(SSyncRaftQuorumJointConfig* config, SyncNodeId id) {
syncRaftAddToNodeMap(&config->incoming, id);
}
void syncRaftJointConfigRemoveFromIncoming(SSyncRaftQuorumJointConfig* config, SyncNodeId id) {
syncRaftRemoveFromNodeMap(&config->incoming, id);
}
void syncRaftJointConfigIDs(SSyncRaftQuorumJointConfig* config, SSyncRaftNodeMap* nodeMap) {
syncRaftCopyNodeMap(&config->incoming, nodeMap);
syncRaftUnionNodeMap(&config->outgoing, nodeMap);
}
SyncIndex syncRaftJointConfigCommittedIndex(const SSyncRaftQuorumJointConfig* config, matchAckIndexerFp indexer, void* arg) {
SyncIndex index0, index1;
index0 = syncRaftMajorityConfigCommittedIndex(&config->incoming, indexer, arg);
index1 = syncRaftMajorityConfigCommittedIndex(&config->outgoing, indexer, arg);
return index0 < index1 ? index0 : index1;
}

121
source/libs/sync/src/sync_raft_quorum_majority.c
View File

@ -1,121 +0,0 @@
/*
* Copyright (c) 2019 TAOS Data, Inc. <cli@taosdata.com>
*
* This program is free software: you can use, redistribute, and/or modify
* it under the terms of the GNU Affero General Public License, version 3
* or later ("AGPL"), as published by the Free Software Foundation.
*
* This program is distributed in the hope that it will be useful, but WITHOUT
* ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or
* FITNESS FOR A PARTICULAR PURPOSE.
*
* You should have received a copy of the GNU Affero General Public License
* along with this program. If not, see <http://www.gnu.org/licenses/>.
*/
#include "sync_const.h"
#include "sync_raft_quorum.h"
#include "sync_raft_quorum_majority.h"
#include "sync_raft_node_map.h"
// VoteResult takes a mapping of voters to yes/no (true/false) votes and returns
// a result indicating whether the vote is pending (i.e. neither a quorum of
// yes/no has been reached), won (a quorum of yes has been reached), or lost (a
// quorum of no has been reached).
ESyncRaftVoteResult syncRaftMajorityVoteResult(SSyncRaftNodeMap* config, SHashObj* votesMap) {
int n = syncRaftNodeMapSize(config);
if (n == 0) {
// By convention, the elections on an empty config win. This comes in
// handy with joint quorums because it'll make a half-populated joint
// quorum behave like a majority quorum.
return SYNC_RAFT_VOTE_WON;
}
int i, g, r, missing;
i = g = r = missing = 0;
SyncNodeId* pId = NULL;
while (!syncRaftIterateNodeMap(config, pId)) {
const bool* v = (const bool*)taosHashGet(votesMap, pId, sizeof(SyncNodeId*));
if (v == NULL) {
missing += 1;
continue;
}
if (*v) {
g +=1;
} else {
r += 1;
}
}
int quorum = n / 2 + 1;
if (g >= quorum) {
return SYNC_RAFT_VOTE_WON;
}
if (g + missing >= quorum) {
return SYNC_RAFT_VOTE_PENDING;
}
return SYNC_RAFT_VOTE_LOST;
}
int compSyncIndex(const void * elem1, const void * elem2) {
SyncIndex index1 = *((SyncIndex*)elem1);
SyncIndex index2 = *((SyncIndex*)elem1);
if (index1 > index2) return 1;
if (index1 < index2) return -1;
return 0;
}
SyncIndex syncRaftMajorityConfigCommittedIndex(const SSyncRaftNodeMap* config, matchAckIndexerFp indexer, void* arg) {
int n = syncRaftNodeMapSize(config);
if (n == 0) {
// This plays well with joint quorums which, when one half is the zero
// MajorityConfig, should behave like the other half.
return kMaxCommitIndex;
}
// Use an on-stack slice to collect the committed indexes when n <= 7
// (otherwise we alloc). The alternative is to stash a slice on
// MajorityConfig, but this impairs usability (as is, MajorityConfig is just
// a map, and that's nice). The assumption is that running with a
// replication factor of >7 is rare, and in cases in which it happens
// performance is a lesser concern (additionally the performance
// implications of an allocation here are far from drastic).
SyncIndex* srt = NULL;
SyncIndex srk[TSDB_MAX_REPLICA];
if (n > TSDB_MAX_REPLICA) {
srt = (SyncIndex*)malloc(sizeof(SyncIndex) * n);
if (srt == NULL) {
return kMaxCommitIndex;
}
} else {
srt = &srk[0];
}
// Fill the slice with the indexes observed. Any unused slots will be
// left as zero; these correspond to voters that may report in, but
// haven't yet. We fill from the right (since the zeroes will end up on
// the left after sorting below anyway).
SyncNodeId *pId = NULL;
int i = 0;
SyncIndex index;
while (!syncRaftIterateNodeMap(config, pId)) {
indexer(*pId, arg, &index);
srt[i++] = index;
}
// Sort by index. Use a bespoke algorithm (copied from the stdlib's sort
// package) to keep srt on the stack.
qsort(srt, n, sizeof(SyncIndex), compSyncIndex);
// The smallest index into the array for which the value is acked by a
// quorum. In other words, from the end of the slice, move n/2+1 to the
// left (accounting for zero-indexing).
index = srt[n - (n/2 + 1)];
if (srt != &srk[0]) {
free(srt);
}
return index;
}

180
source/libs/sync/src/sync_raft_restore.c
View File

@ -1,180 +0,0 @@
/*
* Copyright (c) 2019 TAOS Data, Inc. <cli@taosdata.com>
*
* This program is free software: you can use, redistribute, and/or modify
* it under the terms of the GNU Affero General Public License, version 3
* or later ("AGPL"), as published by the Free Software Foundation.
*
* This program is distributed in the hope that it will be useful, but WITHOUT
* ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or
* FITNESS FOR A PARTICULAR PURPOSE.
*
* You should have received a copy of the GNU Affero General Public License
* along with this program. If not, see <http://www.gnu.org/licenses/>.
*/
#include "sync_raft_config_change.h"
#include "sync_raft_restore.h"
#include "sync_raft_progress_tracker.h"
static void addToConfChangeSingleArray(SSyncConfChangeSingleArray* out, int* i, const SSyncRaftNodeMap* nodeMap, ESyncRaftConfChangeType t);
static int toConfChangeSingle(const SSyncConfigState* cs, SSyncConfChangeSingleArray* out, SSyncConfChangeSingleArray* in);
// syncRaftRestoreConfig takes a Changer (which must represent an empty configuration), and
// runs a sequence of changes enacting the configuration described in the
// ConfState.
//
// TODO(tbg) it's silly that this takes a Changer. Unravel this by making sure
// the Changer only needs a ProgressMap (not a whole Tracker) at which point
// this can just take LastIndex and MaxInflight directly instead and cook up
// the results from that alone.
int syncRaftRestoreConfig(SSyncRaftChanger* changer, const SSyncConfigState* cs,
SSyncRaftProgressTrackerConfig* config, SSyncRaftProgressMap* progressMap) {
SSyncConfChangeSingleArray outgoing;
SSyncConfChangeSingleArray incoming;
SSyncConfChangeSingleArray css;
SSyncRaftProgressTracker* tracker = changer->tracker;
int i, ret;
syncRaftInitConfArray(&outgoing);
syncRaftInitConfArray(&incoming);
syncRaftInitTrackConfig(config);
syncRaftInitProgressMap(progressMap);
ret = toConfChangeSingle(cs, &outgoing, &incoming);
if (ret != 0) {
goto out;
}
if (syncRaftConfArrayIsEmpty(&outgoing)) {
// No outgoing config, so just apply the incoming changes one by one.
for (i = 0; i < incoming.n; ++i) {
css = (SSyncConfChangeSingleArray) {
.n = 1,
.changes = &incoming.changes[i],
};
ret = syncRaftChangerSimpleConfig(changer, &css, config, progressMap);
if (ret != 0) {
goto out;
}
syncRaftCopyTrackerConfig(config, &changer->tracker->config);
syncRaftCopyProgressMap(progressMap, &changer->tracker->progressMap);
}
} else {
// The ConfState describes a joint configuration.
//
// First, apply all of the changes of the outgoing config one by one, so
// that it temporarily becomes the incoming active config. For example,
// if the config is (1 2 3)&(2 3 4), this will establish (2 3 4)&().
for (i = 0; i < outgoing.n; ++i) {
css = (SSyncConfChangeSingleArray) {
.n = 1,
.changes = &outgoing.changes[i],
};
ret = syncRaftChangerSimpleConfig(changer, &css, config, progressMap);
if (ret != 0) {
goto out;
}
syncRaftCopyTrackerConfig(config, &changer->tracker->config);
syncRaftCopyProgressMap(progressMap, &changer->tracker->progressMap);
}
ret = syncRaftChangerEnterJoint(changer, cs->autoLeave, &incoming, config, progressMap);
if (ret != 0) {
goto out;
}
}
out:
syncRaftFreeConfArray(&incoming);
syncRaftFreeConfArray(&outgoing);
return ret;
}
static void addToConfChangeSingleArray(SSyncConfChangeSingleArray* out, int* i, const SSyncRaftNodeMap* nodeMap, ESyncRaftConfChangeType t) {
SyncNodeId* pId = NULL;
while (!syncRaftIterateNodeMap(nodeMap, pId)) {
out->changes[*i] = (SSyncConfChangeSingle) {
.type = t,
.nodeId = *pId,
};
*i += 1;
}
}
// toConfChangeSingle translates a conf state into 1) a slice of operations creating
// first the config that will become the outgoing one, and then the incoming one, and
// b) another slice that, when applied to the config resulted from 1), represents the
// ConfState.
static int toConfChangeSingle(const SSyncConfigState* cs, SSyncConfChangeSingleArray* out, SSyncConfChangeSingleArray* in) {
int i;
out->n = syncRaftNodeMapSize(&cs->votersOutgoing);
out->changes = (SSyncConfChangeSingle*)malloc(sizeof(SSyncConfChangeSingle) * out->n);
if (out->changes == NULL) {
out->n = 0;
return -1;
}
in->n = syncRaftNodeMapSize(&cs->votersOutgoing) +
syncRaftNodeMapSize(&cs->voters) +
syncRaftNodeMapSize(&cs->learners) +
syncRaftNodeMapSize(&cs->learnersNext);
out->changes = (SSyncConfChangeSingle*)malloc(sizeof(SSyncConfChangeSingle) * in->n);
if (in->changes == NULL) {
in->n = 0;
return -1;
}
// Example to follow along this code:
// voters=(1 2 3) learners=(5) outgoing=(1 2 4 6) learners_next=(4)
//
// This means that before entering the joint config, the configuration
// had voters (1 2 4 6) and perhaps some learners that are already gone.
// The new set of voters is (1 2 3), i.e. (1 2) were kept around, and (4 6)
// are no longer voters; however 4 is poised to become a learner upon leaving
// the joint state.
// We can't tell whether 5 was a learner before entering the joint config,
// but it doesn't matter (we'll pretend that it wasn't).
//
// The code below will construct
// outgoing = add 1; add 2; add 4; add 6
// incoming = remove 1; remove 2; remove 4; remove 6
// add 1; add 2; add 3;
// add-learner 5;
// add-learner 4;
//
// So, when starting with an empty config, after applying 'outgoing' we have
//
// quorum=(1 2 4 6)
//
// From which we enter a joint state via 'incoming'
//
// quorum=(1 2 3)&&(1 2 4 6) learners=(5) learners_next=(4)
//
// as desired.
// If there are outgoing voters, first add them one by one so that the
// (non-joint) config has them all.
i = 0;
addToConfChangeSingleArray(out, &i, &cs->votersOutgoing, SYNC_RAFT_Conf_AddNode);
assert(i == out->n);
// We're done constructing the outgoing slice, now on to the incoming one
// (which will apply on top of the config created by the outgoing slice).
i = 0;
// First, we'll remove all of the outgoing voters.
addToConfChangeSingleArray(in, &i, &cs->votersOutgoing, SYNC_RAFT_Conf_RemoveNode);
// Then we'll add the incoming voters and learners.
addToConfChangeSingleArray(in, &i, &cs->voters, SYNC_RAFT_Conf_AddNode);
addToConfChangeSingleArray(in, &i, &cs->learners, SYNC_RAFT_Conf_AddLearnerNode);
addToConfChangeSingleArray(in, &i, &cs->learnersNext, SYNC_RAFT_Conf_AddLearnerNode);
assert(i == in->n);
return 0;
}

0
source/libs/sync/test/raftTests.cpp
View File

5
source/libs/transport/CMakeLists.txt
View File

@ -14,17 +14,18 @@ target_link_libraries(
PUBLIC common PUBLIC common
) )
if (${BUILD_WITH_UV_TRANS}) if (${BUILD_WITH_UV_TRANS})
if (${BUILD_WITH_UV})
target_include_directories( target_include_directories(
transport transport
PUBLIC "${CMAKE_SOURCE_DIR}/contrib/libuv/include" PUBLIC "${CMAKE_SOURCE_DIR}/contrib/libuv/include"
) )
#LINK_DIRECTORIES("${CMAKE_SOURCE_DIR}/debug/contrib/libuv")
target_link_libraries( target_link_libraries(
transport transport
PUBLIC uv_a PUBLIC uv_a
) )
add_definitions(-DUSE_UV) add_definitions(-DUSE_UV)
endif(${BUILD_WITH_UV})
endif(${BUILD_WITH_UV_TRANS}) endif(${BUILD_WITH_UV_TRANS})
if (${BUILD_TEST}) if (${BUILD_TEST})

10
source/libs/transport/inc/transComm.h
View File

@ -134,10 +134,12 @@ typedef struct {
// int16_t numOfTry; // number of try for different servers // int16_t numOfTry; // number of try for different servers
// int8_t oldInUse; // server EP inUse passed by app // int8_t oldInUse; // server EP inUse passed by app
// int8_t redirect; // flag to indicate redirect // int8_t redirect; // flag to indicate redirect
int8_t connType; // connection type int8_t connType; // connection type
int64_t rid; // refId returned by taosAddRef int64_t rid; // refId returned by taosAddRef
SRpcMsg* pRsp; // for synchronous API
tsem_t* pSem; // for synchronous API SRpcMsg* pRsp; // for synchronous API
tsem_t* pSem; // for synchronous API
char* ip; char* ip;
uint32_t port; uint32_t port;
// SEpSet* pSet; // for synchronous API // SEpSet* pSet; // for synchronous API

4
source/libs/transport/src/rpcMain.c
View File

@ -813,8 +813,8 @@ static SRpcConn *rpcSetupConnToServer(SRpcReqContext *pContext) {
SRpcInfo *pRpc = pContext->pRpc; SRpcInfo *pRpc = pContext->pRpc;
SEpSet * pEpSet = &pContext->epSet; SEpSet * pEpSet = &pContext->epSet;
pConn = pConn = rpcGetConnFromCache(pRpc->pCache, pEpSet->eps[pEpSet->inUse].fqdn, pEpSet->eps[pEpSet->inUse].port,
rpcGetConnFromCache(pRpc->pCache, pEpSet->eps[pEpSet->inUse].fqdn, pEpSet->eps[pEpSet->inUse].port, pContext->connType); pContext->connType);
if (pConn == NULL || pConn->user[0] == 0) { if (pConn == NULL || pConn->user[0] == 0) {
pConn = rpcOpenConn(pRpc, pEpSet->eps[pEpSet->inUse].fqdn, pEpSet->eps[pEpSet->inUse].port, pContext->connType); pConn = rpcOpenConn(pRpc, pEpSet->eps[pEpSet->inUse].fqdn, pEpSet->eps[pEpSet->inUse].port, pContext->connType);
} }

34
source/libs/transport/src/trans.c
View File

@ -63,17 +63,41 @@ void rpcFreeCont(void* cont) {
} }
free((char*)cont - TRANS_MSG_OVERHEAD); free((char*)cont - TRANS_MSG_OVERHEAD);
} }
void* rpcReallocCont(void* ptr, int contLen) { return NULL; } void* rpcReallocCont(void* ptr, int contLen) {
if (ptr == NULL) {
return rpcMallocCont(contLen);
}
char* st = (char*)ptr - TRANS_MSG_OVERHEAD;
int sz = contLen + TRANS_MSG_OVERHEAD;
st = realloc(st, sz);
if (st == NULL) {
return NULL;
}
return st + TRANS_MSG_OVERHEAD;
}
void rpcSendRedirectRsp(void* thandle, const SEpSet* pEpSet) {
SRpcMsg rpcMsg;
memset(&rpcMsg, 0, sizeof(rpcMsg));
rpcMsg.contLen = sizeof(SEpSet);
rpcMsg.pCont = rpcMallocCont(rpcMsg.contLen);
if (rpcMsg.pCont == NULL) return;
memcpy(rpcMsg.pCont, pEpSet, sizeof(SEpSet));
rpcMsg.code = TSDB_CODE_RPC_REDIRECT;
rpcMsg.handle = thandle;
rpcSendResponse(&rpcMsg);
}
void rpcSendRedirectRsp(void* pConn, const SEpSet* pEpSet) {}
int rpcGetConnInfo(void* thandle, SRpcConnInfo* pInfo) { return -1; }
void rpcSendRecv(void* shandle, SEpSet* pEpSet, SRpcMsg* pReq, SRpcMsg* pRsp) { return; }
int rpcReportProgress(void* pConn, char* pCont, int contLen) { return -1; } int rpcReportProgress(void* pConn, char* pCont, int contLen) { return -1; }
void rpcCancelRequest(int64_t rid) { return; } void rpcCancelRequest(int64_t rid) { return; }
int32_t rpcInit(void) { int32_t rpcInit(void) {
// impl later // impl later
return -1; return 0;
} }
void rpcCleanup(void) { void rpcCleanup(void) {

69
source/libs/transport/src/transCli.c
View File

@ -123,9 +123,14 @@ static void clientHandleResp(SCliConn* conn) {
rpcMsg.code = pHead->code; rpcMsg.code = pHead->code;
rpcMsg.msgType = pHead->msgType; rpcMsg.msgType = pHead->msgType;
rpcMsg.ahandle = pCtx->ahandle; rpcMsg.ahandle = pCtx->ahandle;
if (pCtx->pSem == NULL) {
tDebug("conn %p handle resp", conn); tDebug("conn %p handle resp", conn);
(pRpc->cfp)(NULL, &rpcMsg, NULL); (pRpc->cfp)(NULL, &rpcMsg, NULL);
} else {
tDebug("conn %p handle resp", conn);
memcpy((char*)pCtx->pRsp, (char*)&rpcMsg, sizeof(rpcMsg));
tsem_post(pCtx->pSem);
}
conn->notifyCount += 1; conn->notifyCount += 1;
// buf's mem alread translated to rpcMsg.pCont // buf's mem alread translated to rpcMsg.pCont
@ -159,14 +164,20 @@ static void clientHandleExcept(SCliConn* pConn) {
SRpcMsg rpcMsg = {0}; SRpcMsg rpcMsg = {0};
rpcMsg.ahandle = pCtx->ahandle; rpcMsg.ahandle = pCtx->ahandle;
rpcMsg.code = TSDB_CODE_RPC_NETWORK_UNAVAIL; rpcMsg.code = TSDB_CODE_RPC_NETWORK_UNAVAIL;
// SRpcInfo* pRpc = pMsg->ctx->pRpc; if (pCtx->pSem == NULL) {
(pCtx->pTransInst->cfp)(NULL, &rpcMsg, NULL); // SRpcInfo* pRpc = pMsg->ctx->pRpc;
pConn->notifyCount += 1; (pCtx->pTransInst->cfp)(NULL, &rpcMsg, NULL);
} else {
memcpy((char*)(pCtx->pRsp), (char*)(&rpcMsg), sizeof(rpcMsg));
// SRpcMsg rpcMsg
tsem_post(pCtx->pSem);
}
destroyCmsg(pMsg); destroyCmsg(pMsg);
pConn->data = NULL; pConn->data = NULL;
// transDestroyConnCtx(pCtx); // transDestroyConnCtx(pCtx);
clientConnDestroy(pConn, true); clientConnDestroy(pConn, true);
pConn->notifyCount += 1;
} }
static void clientTimeoutCb(uv_timer_t* handle) { static void clientTimeoutCb(uv_timer_t* handle) {
@ -463,6 +474,7 @@ static void clientAsyncCb(uv_async_t* handle) {
static void* clientThread(void* arg) { static void* clientThread(void* arg) {
SCliThrdObj* pThrd = (SCliThrdObj*)arg; SCliThrdObj* pThrd = (SCliThrdObj*)arg;
setThreadName("trans-client-work");
uv_run(pThrd->loop, UV_RUN_DEFAULT); uv_run(pThrd->loop, UV_RUN_DEFAULT);
} }
@ -568,8 +580,8 @@ void taosCloseClient(void* arg) {
} }
void rpcSendRequest(void* shandle, const SEpSet* pEpSet, SRpcMsg* pMsg, int64_t* pRid) { void rpcSendRequest(void* shandle, const SEpSet* pEpSet, SRpcMsg* pMsg, int64_t* pRid) {
// impl later // impl later
char* ip = (char*)(pEpSet->fqdn[pEpSet->inUse]); char* ip = (char*)(pEpSet->eps[pEpSet->inUse].fqdn);
uint32_t port = pEpSet->port[pEpSet->inUse]; uint32_t port = pEpSet->eps[pEpSet->inUse].port;
SRpcInfo* pRpc = (SRpcInfo*)shandle; SRpcInfo* pRpc = (SRpcInfo*)shandle;
@ -609,4 +621,45 @@ void rpcSendRequest(void* shandle, const SEpSet* pEpSet, SRpcMsg* pMsg, int64_t*
// int end = taosGetTimestampUs() - start; // int end = taosGetTimestampUs() - start;
// tError("client sent to rpc, time cost: %d", (int)end); // tError("client sent to rpc, time cost: %d", (int)end);
} }
void rpcSendRecv(void* shandle, SEpSet* pEpSet, SRpcMsg* pReq, SRpcMsg* pRsp) {
char* ip = (char*)(pEpSet->eps[pEpSet->inUse].fqdn);
uint32_t port = pEpSet->eps[pEpSet->inUse].port;
SRpcInfo* pRpc = (SRpcInfo*)shandle;
STransConnCtx* pCtx = calloc(1, sizeof(STransConnCtx));
pCtx->pTransInst = (SRpcInfo*)shandle;
pCtx->ahandle = pReq->ahandle;
pCtx->msgType = pReq->msgType;
pCtx->ip = strdup(ip);
pCtx->port = port;
pCtx->pSem = calloc(1, sizeof(tsem_t));
pCtx->pRsp = pRsp;
tsem_init(pCtx->pSem, 0, 0);
int64_t index = pRpc->index;
if (pRpc->index++ >= pRpc->numOfThreads) {
pRpc->index = 0;
}
SCliMsg* cliMsg = malloc(sizeof(SCliMsg));
cliMsg->ctx = pCtx;
cliMsg->msg = *pReq;
cliMsg->st = taosGetTimestampUs();
SCliThrdObj* thrd = ((SClientObj*)pRpc->tcphandle)->pThreadObj[index % pRpc->numOfThreads];
// pthread_mutex_lock(&thrd->msgMtx);
// QUEUE_PUSH(&thrd->msg, &cliMsg->q);
// pthread_mutex_unlock(&thrd->msgMtx);
// int start = taosGetTimestampUs();
transSendAsync(thrd->asyncPool, &(cliMsg->q));
tsem_t* pSem = pCtx->pSem;
tsem_wait(pSem);
tsem_destroy(pSem);
free(pSem);
return;
}
#endif #endif

24
source/libs/transport/src/transSrv.c
View File

@ -33,6 +33,8 @@ typedef struct SSrvConn {
void* hostThrd; void* hostThrd;
void* pSrvMsg; void* pSrvMsg;
struct sockaddr peername;
// SRpcMsg sendMsg; // SRpcMsg sendMsg;
// del later // del later
char secured; char secured;
@ -487,7 +489,13 @@ void uvOnConnectionCb(uv_stream_t* q, ssize_t nread, const uv_buf_t* buf) {
uv_os_fd_t fd; uv_os_fd_t fd;
uv_fileno((const uv_handle_t*)pConn->pTcp, &fd); uv_fileno((const uv_handle_t*)pConn->pTcp, &fd);
tDebug("conn %p created, fd: %d", pConn, fd); tDebug("conn %p created, fd: %d", pConn, fd);
uv_read_start((uv_stream_t*)(pConn->pTcp), uvAllocReadBufferCb, uvOnReadCb); int namelen = sizeof(pConn->peername);
if (0 != uv_tcp_getpeername(pConn->pTcp, &pConn->peername, &namelen)) {
tError("failed to get peer name");
destroyConn(pConn, true);
} else {
uv_read_start((uv_stream_t*)(pConn->pTcp), uvAllocReadBufferCb, uvOnReadCb);
}
} else { } else {
tDebug("failed to create new connection"); tDebug("failed to create new connection");
destroyConn(pConn, true); destroyConn(pConn, true);
@ -496,6 +504,7 @@ void uvOnConnectionCb(uv_stream_t* q, ssize_t nread, const uv_buf_t* buf) {
void* acceptThread(void* arg) { void* acceptThread(void* arg) {
// opt // opt
setThreadName("trans-accept");
SServerObj* srv = (SServerObj*)arg; SServerObj* srv = (SServerObj*)arg;
uv_run(srv->loop, UV_RUN_DEFAULT); uv_run(srv->loop, UV_RUN_DEFAULT);
} }
@ -548,6 +557,7 @@ static bool addHandleToAcceptloop(void* arg) {
return true; return true;
} }
void* workerThread(void* arg) { void* workerThread(void* arg) {
setThreadName("trans-worker");
SWorkThrdObj* pThrd = (SWorkThrdObj*)arg; SWorkThrdObj* pThrd = (SWorkThrdObj*)arg;
uv_run(pThrd->loop, UV_RUN_DEFAULT); uv_run(pThrd->loop, UV_RUN_DEFAULT);
} }
@ -723,4 +733,16 @@ void rpcSendResponse(const SRpcMsg* pMsg) {
// uv_async_send(pThrd->workerAsync); // uv_async_send(pThrd->workerAsync);
} }
int rpcGetConnInfo(void* thandle, SRpcConnInfo* pInfo) {
SSrvConn* pConn = thandle;
struct sockaddr* pPeerName = &pConn->peername;
struct sockaddr_in caddr = *(struct sockaddr_in*)(pPeerName);
pInfo->clientIp = (uint32_t)(caddr.sin_addr.s_addr);
pInfo->clientPort = ntohs(caddr.sin_port);
tstrncpy(pInfo->user, pConn->user, sizeof(pInfo->user));
return 0;
}
#endif #endif

19
source/libs/transport/test/CMakeLists.txt
View File

@ -2,6 +2,7 @@ add_executable(transportTest "")
add_executable(client "") add_executable(client "")
add_executable(server "") add_executable(server "")
add_executable(transUT "") add_executable(transUT "")
add_executable(syncClient "")
target_sources(transUT target_sources(transUT
PRIVATE PRIVATE
@ -20,6 +21,10 @@ target_sources (server
PRIVATE PRIVATE
"rserver.c" "rserver.c"
) )
target_sources (syncClient
PRIVATE
"syncClient.c"
)
target_include_directories(transportTest target_include_directories(transportTest
PUBLIC PUBLIC
@ -67,7 +72,6 @@ target_include_directories(transUT
"${CMAKE_CURRENT_SOURCE_DIR}/../inc" "${CMAKE_CURRENT_SOURCE_DIR}/../inc"
) )
target_link_libraries (server target_link_libraries (server
os os
util util
@ -75,4 +79,17 @@ target_link_libraries (server
gtest_main gtest_main
transport transport
) )
target_include_directories(syncClient
PUBLIC
"${CMAKE_SOURCE_DIR}/include/libs/transport"
"${CMAKE_CURRENT_SOURCE_DIR}/../inc"
)
target_link_libraries (syncClient
os
util
common
gtest_main
transport
)

1
source/libs/transport/test/rclient.c
View File

@ -33,7 +33,6 @@ typedef struct {
pthread_t thread; pthread_t thread;
void * pRpc; void * pRpc;
} SInfo; } SInfo;
static void processResponse(void *pParent, SRpcMsg *pMsg, SEpSet *pEpSet) { static void processResponse(void *pParent, SRpcMsg *pMsg, SEpSet *pEpSet) {
SInfo *pInfo = (SInfo *)pMsg->ahandle; SInfo *pInfo = (SInfo *)pMsg->ahandle;
tDebug("thread:%d, response is received, type:%d contLen:%d code:0x%x", pInfo->index, pMsg->msgType, pMsg->contLen, tDebug("thread:%d, response is received, type:%d contLen:%d code:0x%x", pInfo->index, pMsg->msgType, pMsg->contLen,

220
source/libs/transport/test/syncClient.c Normal file
View File

@ -0,0 +1,220 @@
/*
* Copyright (c) 2019 TAOS Data, Inc. <jhtao@taosdata.com>
*
* This program is free software: you can use, redistribute, and/or modify
* it under the terms of the GNU Affero General Public License, version 3
* or later ("AGPL"), as published by the Free Software Foundation.
*
* This program is distributed in the hope that it will be useful, but WITHOUT
* ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or
* FITNESS FOR A PARTICULAR PURPOSE.
*
* You should have received a copy of the GNU Affero General Public License
* along with this program. If not, see <http://www.gnu.org/licenses/>.
*/
#include <sys/time.h>
#include <tep.h>
#include "os.h"
#include "rpcLog.h"
#include "taoserror.h"
#include "tglobal.h"
#include "trpc.h"
#include "tutil.h"
typedef struct {
int index;
SEpSet epSet;
int num;
int numOfReqs;
int msgSize;
tsem_t rspSem;
tsem_t * pOverSem;
pthread_t thread;
void * pRpc;
} SInfo;
static void processResponse(void *pParent, SRpcMsg *pMsg, SEpSet *pEpSet) {
SInfo *pInfo = (SInfo *)pMsg->ahandle;
tDebug("thread:%d, response is received, type:%d contLen:%d code:0x%x", pInfo->index, pMsg->msgType, pMsg->contLen,
pMsg->code);
if (pEpSet) pInfo->epSet = *pEpSet;
rpcFreeCont(pMsg->pCont);
// tsem_post(&pInfo->rspSem);
tsem_post(&pInfo->rspSem);
}
static int tcount = 0;
static void *sendRequest(void *param) {
SInfo * pInfo = (SInfo *)param;
SRpcMsg rpcMsg = {0};
tDebug("thread:%d, start to send request", pInfo->index);
tDebug("thread:%d, reqs: %d", pInfo->index, pInfo->numOfReqs);
int u100 = 0;
int u500 = 0;
int u1000 = 0;
int u10000 = 0;
SRpcMsg respMsg = {0};
while (pInfo->numOfReqs == 0 || pInfo->num < pInfo->numOfReqs) {
pInfo->num++;
rpcMsg.pCont = rpcMallocCont(pInfo->msgSize);
rpcMsg.contLen = pInfo->msgSize;
rpcMsg.ahandle = pInfo;
rpcMsg.msgType = 1;
// tDebug("thread:%d, send request, contLen:%d num:%d", pInfo->index, pInfo->msgSize, pInfo->num);
int64_t start = taosGetTimestampUs();
rpcSendRecv(pInfo->pRpc, &pInfo->epSet, &rpcMsg, &respMsg);
// rpcSendRequest(pInfo->pRpc, &pInfo->epSet, &rpcMsg, NULL);
if (pInfo->num % 20000 == 0) tInfo("thread:%d, %d requests have been sent", pInfo->index, pInfo->num);
// tsem_wait(&pInfo->rspSem);
// wtsem_wait(&pInfo->rspSem);
int64_t end = taosGetTimestampUs() - start;
if (end <= 100) {
u100++;
} else if (end > 100 && end <= 500) {
u500++;
} else if (end > 500 && end < 1000) {
u1000++;
} else {
u10000++;
}
tDebug("recv response succefully");
// usleep(100000000);
}
tError("send and recv sum: %d, %d, %d, %d", u100, u500, u1000, u10000);
tDebug("thread:%d, it is over", pInfo->index);
tcount++;
return NULL;
}
int main(int argc, char *argv[]) {
SRpcInit rpcInit;
SEpSet epSet = {0};
int msgSize = 128;
int numOfReqs = 0;
int appThreads = 1;
char serverIp[40] = "127.0.0.1";
char secret[20] = "mypassword";
struct timeval systemTime;
int64_t startTime, endTime;
pthread_attr_t thattr;
// server info
epSet.inUse = 0;
addEpIntoEpSet(&epSet, serverIp, 7000);
addEpIntoEpSet(&epSet, "192.168.0.1", 7000);
// client info
memset(&rpcInit, 0, sizeof(rpcInit));
rpcInit.localPort = 0;
rpcInit.label = "APP";
rpcInit.numOfThreads = 1;
rpcInit.cfp = processResponse;
rpcInit.sessions = 100;
rpcInit.idleTime = 100;
rpcInit.user = "michael";
rpcInit.secret = secret;
rpcInit.ckey = "key";
rpcInit.spi = 1;
rpcInit.connType = TAOS_CONN_CLIENT;
for (int i = 1; i < argc; ++i) {
if (strcmp(argv[i], "-p") == 0 && i < argc - 1) {
epSet.eps[0].port = atoi(argv[++i]);
} else if (strcmp(argv[i], "-i") == 0 && i < argc - 1) {
tstrncpy(epSet.eps[0].fqdn, argv[++i], sizeof(epSet.eps[0].fqdn));
} else if (strcmp(argv[i], "-t") == 0 && i < argc - 1) {
rpcInit.numOfThreads = atoi(argv[++i]);
} else if (strcmp(argv[i], "-m") == 0 && i < argc - 1) {
msgSize = atoi(argv[++i]);
} else if (strcmp(argv[i], "-s") == 0 && i < argc - 1) {
rpcInit.sessions = atoi(argv[++i]);
} else if (strcmp(argv[i], "-n") == 0 && i < argc - 1) {
numOfReqs = atoi(argv[++i]);
} else if (strcmp(argv[i], "-a") == 0 && i < argc - 1) {
appThreads = atoi(argv[++i]);
} else if (strcmp(argv[i], "-o") == 0 && i < argc - 1) {
tsCompressMsgSize = atoi(argv[++i]);
} else if (strcmp(argv[i], "-u") == 0 && i < argc - 1) {
rpcInit.user = argv[++i];
} else if (strcmp(argv[i], "-k") == 0 && i < argc - 1) {
rpcInit.secret = argv[++i];
} else if (strcmp(argv[i], "-spi") == 0 && i < argc - 1) {
rpcInit.spi = atoi(argv[++i]);
} else if (strcmp(argv[i], "-d") == 0 && i < argc - 1) {
rpcDebugFlag = atoi(argv[++i]);
} else {
printf("\nusage: %s [options] \n", argv[0]);
printf(" [-i ip]: first server IP address, default is:%s\n", serverIp);
printf(" [-p port]: server port number, default is:%d\n", epSet.eps[0].port);
printf(" [-t threads]: number of rpc threads, default is:%d\n", rpcInit.numOfThreads);
printf(" [-s sessions]: number of rpc sessions, default is:%d\n", rpcInit.sessions);
printf(" [-m msgSize]: message body size, default is:%d\n", msgSize);
printf(" [-a threads]: number of app threads, default is:%d\n", appThreads);
printf(" [-n requests]: number of requests per thread, default is:%d\n", numOfReqs);
printf(" [-o compSize]: compression message size, default is:%d\n", tsCompressMsgSize);
printf(" [-u user]: user name for the connection, default is:%s\n", rpcInit.user);
printf(" [-k secret]: password for the connection, default is:%s\n", rpcInit.secret);
printf(" [-spi SPI]: security parameter index, default is:%d\n", rpcInit.spi);
printf(" [-d debugFlag]: debug flag, default:%d\n", rpcDebugFlag);
printf(" [-h help]: print out this help\n\n");
exit(0);
}
}
taosInitLog("client.log", 100000, 10);
void *pRpc = rpcOpen(&rpcInit);
if (pRpc == NULL) {
tError("failed to initialize RPC");
return -1;
}
tInfo("client is initialized");
tInfo("threads:%d msgSize:%d requests:%d", appThreads, msgSize, numOfReqs);
gettimeofday(&systemTime, NULL);
startTime = systemTime.tv_sec * 1000000 + systemTime.tv_usec;
SInfo *pInfo = (SInfo *)calloc(1, sizeof(SInfo) * appThreads);
pthread_attr_init(&thattr);
pthread_attr_setdetachstate(&thattr, PTHREAD_CREATE_JOINABLE);
for (int i = 0; i < appThreads; ++i) {
pInfo->index = i;
pInfo->epSet = epSet;
pInfo->numOfReqs = numOfReqs;
pInfo->msgSize = msgSize;
tsem_init(&pInfo->rspSem, 0, 0);
pInfo->pRpc = pRpc;
pthread_create(&pInfo->thread, &thattr, sendRequest, pInfo);
pInfo++;
}
do {
usleep(1);
} while (tcount < appThreads);
gettimeofday(&systemTime, NULL);
endTime = systemTime.tv_sec * 1000000 + systemTime.tv_usec;
float usedTime = (endTime - startTime) / 1000.0f; // mseconds
tInfo("it takes %.3f mseconds to send %d requests to server", usedTime, numOfReqs * appThreads);
tInfo("Performance: %.3f requests per second, msgSize:%d bytes", 1000.0 * numOfReqs * appThreads / usedTime, msgSize);
int ch = getchar();
UNUSED(ch);
taosCloseLog();
return 0;
}

21
source/libs/transport/test/transUT.cc
View File

@ -15,6 +15,7 @@
#include <gtest/gtest.h> #include <gtest/gtest.h>
#include <cstdio> #include <cstdio>
#include <cstring> #include <cstring>
#include "tep.h"
#include "trpc.h" #include "trpc.h"
using namespace std; using namespace std;
@ -50,6 +51,25 @@ class TransObj {
trans = rpcOpen(&rpcInit); trans = rpcOpen(&rpcInit);
return trans != NULL ? true : false; return trans != NULL ? true : false;
} }
bool sendAndRecv() {
SEpSet epSet = {0};
epSet.inUse = 0;
addEpIntoEpSet(&epSet, "192.168.1.1", 7000);
addEpIntoEpSet(&epSet, "192.168.0.1", 7000);
if (trans == NULL) {
return false;
}
SRpcMsg rpcMsg = {0}, reqMsg = {0};
reqMsg.pCont = rpcMallocCont(10);
reqMsg.contLen = 10;
reqMsg.ahandle = NULL;
rpcSendRecv(trans, &epSet, &reqMsg, &rpcMsg);
int code = rpcMsg.code;
std::cout << tstrerror(code) << std::endl;
return true;
}
bool stop() { bool stop() {
rpcClose(trans); rpcClose(trans);
trans = NULL; trans = NULL;
@ -75,6 +95,7 @@ class TransEnv : public ::testing::Test {
}; };
TEST_F(TransEnv, test_start_stop) { TEST_F(TransEnv, test_start_stop) {
assert(tr->startCli()); assert(tr->startCli());
assert(tr->sendAndRecv());
assert(tr->stop()); assert(tr->stop());
assert(tr->startSrv()); assert(tr->startSrv());

31
source/nodes/src/nodesTraverseFuncs.c
View File

@ -75,6 +75,35 @@ static bool walkNode(SNode* pNode, ETraversalOrder order, FQueryNodeWalker walke
case QUERY_NODE_ORDER_BY_EXPR: case QUERY_NODE_ORDER_BY_EXPR:
res = walkNode(((SOrderByExprNode*)pNode)->pExpr, order, walker, pContext); res = walkNode(((SOrderByExprNode*)pNode)->pExpr, order, walker, pContext);
break; break;
case QUERY_NODE_STATE_WINDOW:
res = walkNode(((SStateWindowNode*)pNode)->pCol, order, walker, pContext);
break;
case QUERY_NODE_SESSION_WINDOW:
res = walkNode(((SSessionWindowNode*)pNode)->pCol, order, walker, pContext);
break;
case QUERY_NODE_INTERVAL_WINDOW: {
SIntervalWindowNode* pInterval = (SIntervalWindowNode*)pNode;
res = walkNode(pInterval->pInterval, order, walker, pContext);
if (res) {
res = walkNode(pInterval->pOffset, order, walker, pContext);
}
if (res) {
res = walkNode(pInterval->pSliding, order, walker, pContext);
}
if (res) {
res = walkNode(pInterval->pFill, order, walker, pContext);
}
break;
}
case QUERY_NODE_NODE_LIST:
res = walkList(((SNodeListNode*)pNode)->pNodeList, order, walker, pContext);
break;
case QUERY_NODE_FILL:
res = walkNode(((SFillNode*)pNode)->pValues, order, walker, pContext);
break;
case QUERY_NODE_RAW_EXPR:
res = walkNode(((SRawExprNode*)pNode)->pNode, order, walker, pContext);
break;
default: default:
break; break;
} }
@ -109,7 +138,7 @@ void nodesWalkNodePostOrder(SNode* pNode, FQueryNodeWalker walker, void* pContex
} }
void nodesWalkListPostOrder(SNodeList* pList, FQueryNodeWalker walker, void* pContext) { void nodesWalkListPostOrder(SNodeList* pList, FQueryNodeWalker walker, void* pContext) {
(void)walkList(pList, TRAVERSAL_PREORDER, walker, pContext); (void)walkList(pList, TRAVERSAL_POSTORDER, walker, pContext);
} }
bool nodesWalkStmt(SNode* pNode, FQueryNodeWalker walker, void* pContext) { bool nodesWalkStmt(SNode* pNode, FQueryNodeWalker walker, void* pContext) {

73
source/nodes/src/nodesUtilFuncs.c
View File

@ -58,6 +58,12 @@ SNode* nodesMakeNode(ENodeType type) {
return makeNode(type, sizeof(SSessionWindowNode)); return makeNode(type, sizeof(SSessionWindowNode));
case QUERY_NODE_INTERVAL_WINDOW: case QUERY_NODE_INTERVAL_WINDOW:
return makeNode(type, sizeof(SIntervalWindowNode)); return makeNode(type, sizeof(SIntervalWindowNode));
case QUERY_NODE_NODE_LIST:
return makeNode(type, sizeof(SNodeListNode));
case QUERY_NODE_FILL:
return makeNode(type, sizeof(SFillNode));
case QUERY_NODE_RAW_EXPR:
return makeNode(type, sizeof(SRawExprNode));
case QUERY_NODE_SET_OPERATOR: case QUERY_NODE_SET_OPERATOR:
return makeNode(type, sizeof(SSetOperator)); return makeNode(type, sizeof(SSetOperator));
case QUERY_NODE_SELECT_STMT: case QUERY_NODE_SELECT_STMT:
@ -70,8 +76,19 @@ SNode* nodesMakeNode(ENodeType type) {
return NULL; return NULL;
} }
void nodesDestroyNode(SNode* pNode) { static bool destroyNode(SNode* pNode, void* pContext) {
switch (nodeType(pNode)) {
case QUERY_NODE_VALUE:
tfree(((SValueNode*)pNode)->literal);
break;
default:
break;
}
tfree(pNode);
}
void nodesDestroyNode(SNode* pNode) {
nodesWalkNodePostOrder(pNode, destroyNode, NULL);
} }
SNodeList* nodesMakeList() { SNodeList* nodesMakeList() {
@ -103,13 +120,63 @@ SNodeList* nodesListAppend(SNodeList* pList, SNode* pNode) {
} }
void nodesDestroyList(SNodeList* pList) { void nodesDestroyList(SNodeList* pList) {
SNode* node;
FOREACH(node, pList) {
nodesDestroyNode(node);
}
tfree(pList);
}
bool nodesIsArithmeticOp(const SOperatorNode* pOp) {
switch (pOp->opType) {
case OP_TYPE_ADD:
case OP_TYPE_SUB:
case OP_TYPE_MULTI:
case OP_TYPE_DIV:
case OP_TYPE_MOD:
return true;
default:
break;
}
return false;
}
bool nodesIsComparisonOp(const SOperatorNode* pOp) {
switch (pOp->opType) {
case OP_TYPE_GREATER_THAN:
case OP_TYPE_GREATER_EQUAL:
case OP_TYPE_LOWER_THAN:
case OP_TYPE_LOWER_EQUAL:
case OP_TYPE_EQUAL:
case OP_TYPE_NOT_EQUAL:
case OP_TYPE_IN:
case OP_TYPE_NOT_IN:
case OP_TYPE_LIKE:
case OP_TYPE_NOT_LIKE:
case OP_TYPE_MATCH:
case OP_TYPE_NMATCH:
return true;
default:
break;
}
return false;
}
bool nodesIsJsonOp(const SOperatorNode* pOp) {
switch (pOp->opType) {
case OP_TYPE_JSON_GET_VALUE:
case OP_TYPE_JSON_CONTAINS:
return true;
default:
break;
}
return false;
} }
bool nodesIsTimeorderQuery(const SNode* pQuery) { bool nodesIsTimeorderQuery(const SNode* pQuery) {
return false;
} }
bool nodesIsTimelineQuery(const SNode* pQuery) { bool nodesIsTimelineQuery(const SNode* pQuery) {
return false;
} }

2
source/util/src/terror.c
View File

@ -420,6 +420,8 @@ TAOS_DEFINE_ERROR(TSDB_CODE_CTG_INVALID_INPUT, "invalid catalog input
TAOS_DEFINE_ERROR(TSDB_CODE_CTG_NOT_READY, "catalog is not ready") TAOS_DEFINE_ERROR(TSDB_CODE_CTG_NOT_READY, "catalog is not ready")
TAOS_DEFINE_ERROR(TSDB_CODE_CTG_MEM_ERROR, "catalog memory error") TAOS_DEFINE_ERROR(TSDB_CODE_CTG_MEM_ERROR, "catalog memory error")
TAOS_DEFINE_ERROR(TSDB_CODE_CTG_SYS_ERROR, "catalog system error") TAOS_DEFINE_ERROR(TSDB_CODE_CTG_SYS_ERROR, "catalog system error")
TAOS_DEFINE_ERROR(TSDB_CODE_CTG_DB_DROPPED, "Database is dropped")
TAOS_DEFINE_ERROR(TSDB_CODE_CTG_OUT_OF_SERVICE, "catalog is out of service")
//scheduler //scheduler
TAOS_DEFINE_ERROR(TSDB_CODE_SCH_STATUS_ERROR, "scheduler status error") TAOS_DEFINE_ERROR(TSDB_CODE_SCH_STATUS_ERROR, "scheduler status error")