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

refactor(sync): add pre snapshot

This commit is contained in:
Minghao Li 2022-11-02 15:38:30 +08:00
parent 3072afbacb 7d291174aa
commit 70aff8451f
136 changed files with 2800 additions and 1847 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

2
cmake/cmake.version
View File

@ -2,7 +2,7 @@
IF (DEFINED VERNUMBER)
SET(TD_VER_NUMBER ${VERNUMBER})
ELSE ()
SET(TD_VER_NUMBER "3.0.1.5")
SET(TD_VER_NUMBER "3.0.1.6")
ENDIF ()
IF (DEFINED VERCOMPATIBLE)

2
cmake/taostools_CMakeLists.txt.in
View File

@ -2,7 +2,7 @@
# taos-tools
ExternalProject_Add(taos-tools
GIT_REPOSITORY https://github.com/taosdata/taos-tools.git
GIT_TAG 7321fbb
GIT_TAG 16eb34f
SOURCE_DIR "${TD_SOURCE_DIR}/tools/taos-tools"
BINARY_DIR ""
#BUILD_IN_SOURCE TRUE

6
docs/en/12-taos-sql/03-table.md
View File

@ -7,12 +7,12 @@ title: Table
You create standard tables and subtables with the `CREATE TABLE` statement.
```sql
CREATE TABLE [IF NOT EXISTS] [db_name.]tb_name (create_definition [, create_definitionn] ...) [table_options]
CREATE TABLE [IF NOT EXISTS] [db_name.]tb_name (create_definition [, create_definition] ...) [table_options]
CREATE TABLE create_subtable_clause
CREATE TABLE [IF NOT EXISTS] [db_name.]tb_name (create_definition [, create_definitionn] ...)
[TAGS (create_definition [, create_definitionn] ...)]
CREATE TABLE [IF NOT EXISTS] [db_name.]tb_name (create_definition [, create_definition] ...)
[TAGS (create_definition [, create_definition] ...)]
[table_options]
create_subtable_clause: {

2
docs/en/12-taos-sql/04-stable.md
View File

@ -6,7 +6,7 @@ title: Supertable
## Create a Supertable
```sql
CREATE STABLE [IF NOT EXISTS] stb_name (create_definition [, create_definitionn] ...) TAGS (create_definition [, create_definition] ...) [table_options]
CREATE STABLE [IF NOT EXISTS] stb_name (create_definition [, create_definition] ...) TAGS (create_definition [, create_definition] ...) [table_options]
create_definition:
col_name column_definition

6
docs/zh/12-taos-sql/03-table.md
View File

@ -9,12 +9,12 @@ description: 对表的各种管理操作
`CREATE TABLE` 语句用于创建普通表和以超级表为模板创建子表。
```sql
CREATE TABLE [IF NOT EXISTS] [db_name.]tb_name (create_definition [, create_definitionn] ...) [table_options]
CREATE TABLE [IF NOT EXISTS] [db_name.]tb_name (create_definition [, create_definition] ...) [table_options]
CREATE TABLE create_subtable_clause
CREATE TABLE [IF NOT EXISTS] [db_name.]tb_name (create_definition [, create_definitionn] ...)
[TAGS (create_definition [, create_definitionn] ...)]
CREATE TABLE [IF NOT EXISTS] [db_name.]tb_name (create_definition [, create_definition] ...)
[TAGS (create_definition [, create_definition] ...)]
[table_options]
create_subtable_clause: {

2
docs/zh/12-taos-sql/04-stable.md
View File

@ -7,7 +7,7 @@ description: 对超级表的各种管理操作
## 创建超级表
```sql
CREATE STABLE [IF NOT EXISTS] stb_name (create_definition [, create_definitionn] ...) TAGS (create_definition [, create_definition] ...) [table_options]
CREATE STABLE [IF NOT EXISTS] stb_name (create_definition [, create_definition] ...) TAGS (create_definition [, create_definition] ...) [table_options]
create_definition:
col_name column_definition

4
docs/zh/12-taos-sql/24-show.md
View File

@ -211,10 +211,10 @@ SHOW USERS;
显示当前系统中所有用户的信息。包括用户自定义的用户和系统默认用户。
## SHOW VARIABLES
## SHOW CLUSTER VARIABLES(3.0.1.6 之前为 SHOW VARIABLES)
```sql
SHOW VARIABLES;
SHOW CLUSTER VARIABLES;
SHOW DNODE dnode_id VARIABLES;
```

3
include/common/tglobal.h
View File

@ -126,6 +126,9 @@ extern char tsSmlChildTableName[];
extern char tsSmlTagName[];
extern bool tsSmlDataFormat;
// wal
extern int64_t tsWalFsyncDataSizeLimit;
// internal
extern int32_t tsTransPullupInterval;
extern int32_t tsMqRebalanceInterval;

4
include/common/tmsg.h
View File

@ -57,9 +57,11 @@ extern int32_t tMsgDict[];
#define TMSG_SEG_SEQ(TYPE) ((TYPE)&0xff)
#define TMSG_INFO(TYPE) \
((TYPE) < TDMT_DND_MAX_MSG || (TYPE) < TDMT_MND_MAX_MSG || (TYPE) < TDMT_VND_MAX_MSG || (TYPE) < TDMT_SCH_MAX_MSG || \
(TYPE) < TDMT_STREAM_MAX_MSG || (TYPE) < TDMT_MON_MAX_MSG || (TYPE) < TDMT_SYNC_MAX_MSG) \
(TYPE) < TDMT_STREAM_MAX_MSG || (TYPE) < TDMT_MON_MAX_MSG || (TYPE) < TDMT_SYNC_MAX_MSG) || \
(TYPE) < TDMT_VND_STREAM_MSG || (TYPE) < TDMT_VND_TMQ_MSG \
? tMsgInfo[tMsgDict[TMSG_SEG_CODE(TYPE)] + TMSG_SEG_SEQ(TYPE)] \
: 0
#define TMSG_INDEX(TYPE) (tMsgDict[TMSG_SEG_CODE(TYPE)] + TMSG_SEG_SEQ(TYPE))
typedef uint16_t tmsg_t;

74
include/common/tmsgdef.h
View File

@ -138,18 +138,18 @@ enum {
TD_DEF_MSG_TYPE(TDMT_MND_GET_TABLE_INDEX, "get-table-index", NULL, NULL)
TD_DEF_MSG_TYPE(TDMT_MND_BATCH_META, "batch-meta", NULL, NULL)
TD_DEF_MSG_TYPE(TDMT_MND_TABLE_CFG, "table-cfg", NULL, NULL)
TD_DEF_MSG_TYPE(TDMT_MND_CREATE_TOPIC, "create-topic", SMCreateTopicReq, SMCreateTopicRsp)
TD_DEF_MSG_TYPE(TDMT_MND_ALTER_TOPIC, "alter-topic", NULL, NULL)
TD_DEF_MSG_TYPE(TDMT_MND_DROP_TOPIC, "drop-topic", NULL, NULL)
TD_DEF_MSG_TYPE(TDMT_MND_SUBSCRIBE, "subscribe", SCMSubscribeReq, SCMSubscribeRsp)
TD_DEF_MSG_TYPE(TDMT_MND_MQ_ASK_EP, "ask-ep", SMqAskEpReq, SMqAskEpRsp)
TD_DEF_MSG_TYPE(TDMT_MND_MQ_CONSUMER_LOST, "consumer-lost", SMqConsumerLostMsg, NULL)
TD_DEF_MSG_TYPE(TDMT_MND_MQ_CONSUMER_RECOVER, "consumer-recover", SMqConsumerRecoverMsg, NULL)
TD_DEF_MSG_TYPE(TDMT_MND_MQ_HB, "consumer-hb", NULL, NULL)
TD_DEF_MSG_TYPE(TDMT_MND_MQ_DO_REBALANCE, "do-rebalance", SMqDoRebalanceMsg, NULL)
TD_DEF_MSG_TYPE(TDMT_MND_MQ_DROP_CGROUP, "drop-cgroup", SMqDropCGroupReq, SMqDropCGroupRsp)
TD_DEF_MSG_TYPE(TDMT_MND_MQ_COMMIT_OFFSET, "mnode-commit-offset", SMqCMCommitOffsetReq, SMqCMCommitOffsetRsp)
TD_DEF_MSG_TYPE(TDMT_MND_MQ_TIMER, "mq-tmr", SMTimerReq, NULL)
TD_DEF_MSG_TYPE(TDMT_MND_TMQ_CREATE_TOPIC, "create-topic", SMCreateTopicReq, SMCreateTopicRsp)
TD_DEF_MSG_TYPE(TDMT_MND_UNUSED1, "unused", NULL, NULL)
TD_DEF_MSG_TYPE(TDMT_MND_TMQ_DROP_TOPIC, "drop-topic", NULL, NULL)
TD_DEF_MSG_TYPE(TDMT_MND_TMQ_SUBSCRIBE, "subscribe", SCMSubscribeReq, SCMSubscribeRsp)
TD_DEF_MSG_TYPE(TDMT_MND_TMQ_ASK_EP, "ask-ep", SMqAskEpReq, SMqAskEpRsp)
TD_DEF_MSG_TYPE(TDMT_MND_TMQ_CONSUMER_LOST, "consumer-lost", SMqConsumerLostMsg, NULL)
TD_DEF_MSG_TYPE(TDMT_MND_TMQ_CONSUMER_RECOVER, "consumer-recover", SMqConsumerRecoverMsg, NULL)
TD_DEF_MSG_TYPE(TDMT_MND_TMQ_HB, "consumer-hb", NULL, NULL)
TD_DEF_MSG_TYPE(TDMT_MND_TMQ_DO_REBALANCE, "do-rebalance", SMqDoRebalanceMsg, NULL)
TD_DEF_MSG_TYPE(TDMT_MND_TMQ_DROP_CGROUP, "drop-cgroup", SMqDropCGroupReq, SMqDropCGroupRsp)
TD_DEF_MSG_TYPE(TDMT_MND_UNUSED2, "unused2", NULL, NULL)
TD_DEF_MSG_TYPE(TDMT_MND_TMQ_TIMER, "mq-tmr", SMTimerReq, NULL)
TD_DEF_MSG_TYPE(TDMT_MND_TELEM_TIMER, "telem-tmr", SMTimerReq, SMTimerReq)
TD_DEF_MSG_TYPE(TDMT_MND_TRANS_TIMER, "trans-tmr", NULL, NULL)
TD_DEF_MSG_TYPE(TDMT_MND_TTL_TIMER, "ttl-tmr", NULL, NULL)
@ -186,21 +186,21 @@ enum {
TD_DEF_MSG_TYPE(TDMT_VND_CREATE_STB, "vnode-create-stb", SVCreateStbReq, NULL)
TD_DEF_MSG_TYPE(TDMT_VND_ALTER_STB, "vnode-alter-stb", NULL, NULL)
TD_DEF_MSG_TYPE(TDMT_VND_DROP_STB, "vnode-drop-stb", SVDropStbReq, NULL)
TD_DEF_MSG_TYPE(TDMT_VND_MQ_VG_CHANGE, "vnode-mq-vg-change", SMqRebVgReq, SMqRebVgRsp)
TD_DEF_MSG_TYPE(TDMT_VND_MQ_VG_DELETE, "vnode-mq-vg-delete", SMqVDeleteReq, SMqVDeleteRsp)
TD_DEF_MSG_TYPE(TDMT_VND_MQ_COMMIT_OFFSET, "vnode-commit-offset", STqOffset, STqOffset)
TD_DEF_MSG_TYPE(TDMT_VND_ADD_CHECK_INFO, "vnode-add-check-info", NULL, NULL)
TD_DEF_MSG_TYPE(TDMT_VND_DELETE_CHECK_INFO, "vnode-delete-check-info", NULL, NULL)
TD_DEF_MSG_TYPE(TDMT_VND_CREATE_TOPIC, "vnode-create-topic", NULL, NULL)
TD_DEF_MSG_TYPE(TDMT_VND_ALTER_TOPIC, "vnode-alter-topic", NULL, NULL)
TD_DEF_MSG_TYPE(TDMT_VND_DROP_TOPIC, "vnode-drop-topic", NULL, NULL)
TD_DEF_MSG_TYPE(TDMT_VND_SUBSCRIBE, "vnode-subscribe", SMVSubscribeReq, SMVSubscribeRsp)
TD_DEF_MSG_TYPE(TDMT_VND_CONSUME, "vnode-consume", SMqPollReq, SMqDataBlkRsp)
TD_DEF_MSG_TYPE(TDMT_VND_STREAM_TRIGGER, "vnode-stream-trigger", NULL, NULL)
TD_DEF_MSG_TYPE(TDMT_VND_STREAM_DISPATCH_WRITE, "vnode-stream-task-dispatch-write", NULL, NULL)
TD_DEF_MSG_TYPE(TDMT_VND_STREAM_RECOVER_STEP1, "vnode-stream-recover1", NULL, NULL)
TD_DEF_MSG_TYPE(TDMT_VND_STREAM_RECOVER_STEP2, "vnode-stream-recover2", NULL, NULL)
TD_DEF_MSG_TYPE(TDMT_VND_STREAM_RECOVER_FINISH, "vnode-stream-finish", NULL, NULL)
TD_DEF_MSG_TYPE(TDMT_VND_UNUSED1, "vnode-unused1", NULL, NULL)
TD_DEF_MSG_TYPE(TDMT_VND_UNUSED2, "vnode-unused2", NULL, NULL)
TD_DEF_MSG_TYPE(TDMT_VND_UNUSED3, "vnode-unused3", NULL, NULL)
TD_DEF_MSG_TYPE(TDMT_VND_UNUSED4, "vnode-unused4", NULL, NULL)
TD_DEF_MSG_TYPE(TDMT_VND_UNUSED5, "vnode-unused5", NULL, NULL)
TD_DEF_MSG_TYPE(TDMT_VND_UNUSED6, "vnode-unused6", NULL, NULL)
TD_DEF_MSG_TYPE(TDMT_VND_UNUSED7, "vnode-unused7", NULL, NULL)
TD_DEF_MSG_TYPE(TDMT_VND_UNUSED8, "vnode-unused8", NULL, NULL)
TD_DEF_MSG_TYPE(TDMT_VND_UNUSED9, "vnode-unused9", NULL, NULL)
TD_DEF_MSG_TYPE(TDMT_VND_UNUSED10, "vnode-unused10", NULL, NULL)
TD_DEF_MSG_TYPE(TDMT_VND_UNUSED11, "vnode-unused11", NULL, NULL)
TD_DEF_MSG_TYPE(TDMT_VND_UNUSED12, "vnode-unused12", NULL, NULL)
TD_DEF_MSG_TYPE(TDMT_VND_UNUSED13, "vnode-unused13", NULL, NULL)
TD_DEF_MSG_TYPE(TDMT_VND_UNUSED14, "vnode-unused14", NULL, NULL)
TD_DEF_MSG_TYPE(TDMT_VND_UNUSED15, "vnode-unused15", NULL, NULL)
TD_DEF_MSG_TYPE(TDMT_VND_CREATE_SMA, "vnode-create-sma", NULL, NULL)
TD_DEF_MSG_TYPE(TDMT_VND_CANCEL_SMA, "vnode-cancel-sma", NULL, NULL)
TD_DEF_MSG_TYPE(TDMT_VND_DROP_SMA, "vnode-drop-sma", NULL, NULL)
@ -232,13 +232,15 @@ enum {
TD_DEF_MSG_TYPE(TDMT_SCH_LINK_BROKEN, "link-broken", NULL, NULL)
TD_DEF_MSG_TYPE(TDMT_SCH_MAX_MSG, "sch-max", NULL, NULL)
TD_NEW_MSG_SEG(TDMT_STREAM_MSG)
TD_DEF_MSG_TYPE(TDMT_STREAM_TASK_DEPLOY, "stream-task-deploy", SStreamTaskDeployReq, SStreamTaskDeployRsp)
TD_DEF_MSG_TYPE(TDMT_STREAM_TASK_DROP, "stream-task-drop", NULL, NULL)
TD_DEF_MSG_TYPE(TDMT_STREAM_TASK_RUN, "stream-task-run", NULL, NULL)
TD_DEF_MSG_TYPE(TDMT_STREAM_TASK_DISPATCH, "stream-task-dispatch", NULL, NULL)
TD_DEF_MSG_TYPE(TDMT_STREAM_TASK_RECOVER, "stream-task-recover", NULL, NULL)
TD_DEF_MSG_TYPE(TDMT_STREAM_UNUSED1, "stream-unused1", NULL, NULL)
TD_DEF_MSG_TYPE(TDMT_STREAM_RETRIEVE, "stream-retrieve", NULL, NULL)
TD_DEF_MSG_TYPE(TDMT_STREAM_RECOVER_FINISH, "vnode-stream-finish", NULL, NULL)
TD_DEF_MSG_TYPE(TDMT_STREAM_MAX_MSG, "stream-max", NULL, NULL)
TD_NEW_MSG_SEG(TDMT_MON_MSG)
@ -274,6 +276,22 @@ enum {
TD_DEF_MSG_TYPE(TDMT_SYNC_PRE_SNAPSHOT_REPLY, "sync-pre-snapshot-reply", NULL, NULL)
TD_DEF_MSG_TYPE(TDMT_SYNC_MAX_MSG, "sync-max", NULL, NULL)
TD_NEW_MSG_SEG(TDMT_VND_STREAM_MSG)
TD_DEF_MSG_TYPE(TDMT_VND_STREAM_TRIGGER, "vnode-stream-trigger", NULL, NULL)
TD_DEF_MSG_TYPE(TDMT_VND_STREAM_RECOVER_STEP1, "vnode-stream-recover1", NULL, NULL)
TD_DEF_MSG_TYPE(TDMT_VND_STREAM_RECOVER_STEP2, "vnode-stream-recover2", NULL, NULL)
TD_DEF_MSG_TYPE(TDMT_VND_STREAM_MAX_MSG, "vnd-stream-max", NULL, NULL)
TD_NEW_MSG_SEG(TDMT_VND_TMQ_MSG)
TD_DEF_MSG_TYPE(TDMT_VND_TMQ_SUBSCRIBE, "vnode-tmq-subscribe", SMqRebVgReq, SMqRebVgRsp)
TD_DEF_MSG_TYPE(TDMT_VND_TMQ_DELETE_SUB, "vnode-tmq-delete-sub", SMqVDeleteReq, SMqVDeleteRsp)
TD_DEF_MSG_TYPE(TDMT_VND_TMQ_COMMIT_OFFSET, "vnode-tmq-commit-offset", STqOffset, STqOffset)
TD_DEF_MSG_TYPE(TDMT_VND_TMQ_ADD_CHECKINFO, "vnode-tmq-add-checkinfo", NULL, NULL)
TD_DEF_MSG_TYPE(TDMT_VND_TMQ_DEL_CHECKINFO, "vnode-del-checkinfo", NULL, NULL)
TD_DEF_MSG_TYPE(TDMT_VND_TMQ_CONSUME, "vnode-tmq-consume", SMqPollReq, SMqDataBlkRsp)
TD_DEF_MSG_TYPE(TDMT_VND_TMQ_MAX_MSG, "vnd-tmq-max", NULL, NULL)
#if defined(TD_MSG_NUMBER_)
TDMT_MAX
#endif

6
include/common/tname.h
View File

@ -73,11 +73,11 @@ bool tNameTbNameEqual(SName* left, SName* right);
typedef struct {
// input
SArray* tags; // element is SSmlKv
const char* sTableName; // super table name
uint8_t sTableNameLen; // the length of super table name
const char* stbFullName; // super table name
uint8_t stbFullNameLen; // the length of super table name
// output
char* childTableName; // must have size of TSDB_TABLE_NAME_LEN;
char* ctbShortName; // must have size of TSDB_TABLE_NAME_LEN;
uint64_t uid; // child table uid, may be useful
} RandTableName;

1
include/dnode/mnode/mnode.h
View File

@ -99,7 +99,6 @@ int32_t mndGetLoad(SMnode *pMnode, SMnodeLoad *pLoad);
*/
int32_t mndProcessRpcMsg(SRpcMsg *pMsg);
int32_t mndProcessSyncMsg(SRpcMsg *pMsg);
int32_t mndProcessSyncCtrlMsg(SRpcMsg *pMsg);
int32_t mndPreProcessQueryMsg(SRpcMsg *pMsg);
void mndPostProcessQueryMsg(SRpcMsg *pMsg);

3
include/libs/catalog/catalog.h
View File

@ -309,6 +309,9 @@ int32_t catalogGetUdfInfo(SCatalog* pCtg, SRequestConnInfo* pConn, const char* f
int32_t catalogChkAuth(SCatalog* pCtg, SRequestConnInfo* pConn, const char* user, const char* dbFName, AUTH_TYPE type,
bool* pass);
int32_t catalogChkAuthFromCache(SCatalog* pCtg, SRequestConnInfo* pConn, const char* user, const char* dbFName, AUTH_TYPE type,
bool* pass, bool* exists);
int32_t catalogUpdateUserAuthInfo(SCatalog* pCtg, SGetUserAuthRsp* pAuth);
int32_t catalogUpdateVgEpSet(SCatalog* pCtg, const char* dbFName, int32_t vgId, SEpSet* epSet);

1
include/libs/executor/executor.h
View File

@ -79,6 +79,7 @@ qTaskInfo_t qCreateStreamExecTaskInfo(void* msg, SReadHandle* readers);
*/
qTaskInfo_t qCreateQueueExecTaskInfo(void* msg, SReadHandle* readers, int32_t* numOfCols, SSchemaWrapper** pSchema);
int32_t qSetStreamOpOpen(qTaskInfo_t tinfo);
/**
* Set multiple input data blocks for the stream scan.
* @param tinfo

1
include/libs/function/functionMgt.h
View File

@ -216,6 +216,7 @@ bool fmIsMultiRowsFunc(int32_t funcId);
bool fmIsKeepOrderFunc(int32_t funcId);
bool fmIsCumulativeFunc(int32_t funcId);
bool fmIsInterpPseudoColumnFunc(int32_t funcId);
bool fmIsGroupKeyFunc(int32_t funcId);
void getLastCacheDataType(SDataType* pType);

1
include/libs/stream/tstream.h
View File

@ -441,6 +441,7 @@ typedef struct {
} SStreamRetrieveRsp;
typedef struct {
SMsgHead msgHead;
int64_t streamId;
int32_t taskId;
} SStreamRecoverStep1Req, SStreamRecoverStep2Req;

82
include/libs/sync/sync.h
View File

@ -25,8 +25,6 @@ extern "C" {
#include "tlrucache.h"
#include "tmsgcb.h"
extern bool gRaftDetailLog;
#define SYNC_RESP_TTL_MS 10000000
#define SYNC_SPEED_UP_HB_TIMER 400
#define SYNC_SPEED_UP_AFTER_MS (1000 * 20)
@ -133,27 +131,27 @@ typedef struct SSnapshotMeta {
typedef struct SSyncFSM {
void* data;
void (*FpCommitCb)(struct SSyncFSM* pFsm, const SRpcMsg* pMsg, SFsmCbMeta cbMeta);
void (*FpPreCommitCb)(struct SSyncFSM* pFsm, const SRpcMsg* pMsg, SFsmCbMeta cbMeta);
void (*FpRollBackCb)(struct SSyncFSM* pFsm, const SRpcMsg* pMsg, SFsmCbMeta cbMeta);
void (*FpCommitCb)(const struct SSyncFSM* pFsm, const SRpcMsg* pMsg, const SFsmCbMeta* pMeta);
void (*FpPreCommitCb)(const struct SSyncFSM* pFsm, const SRpcMsg* pMsg, const SFsmCbMeta* pMeta);
void (*FpRollBackCb)(const struct SSyncFSM* pFsm, const SRpcMsg* pMsg, const SFsmCbMeta* pMeta);
void (*FpRestoreFinishCb)(struct SSyncFSM* pFsm);
void (*FpReConfigCb)(struct SSyncFSM* pFsm, const SRpcMsg* pMsg, SReConfigCbMeta* cbMeta);
void (*FpLeaderTransferCb)(struct SSyncFSM* pFsm, const SRpcMsg* pMsg, SFsmCbMeta cbMeta);
void (*FpRestoreFinishCb)(const struct SSyncFSM* pFsm);
void (*FpReConfigCb)(const struct SSyncFSM* pFsm, const SRpcMsg* pMsg, const SReConfigCbMeta* pMeta);
void (*FpLeaderTransferCb)(const struct SSyncFSM* pFsm, const SRpcMsg* pMsg, const SFsmCbMeta* pMeta);
void (*FpBecomeLeaderCb)(struct SSyncFSM* pFsm);
void (*FpBecomeFollowerCb)(struct SSyncFSM* pFsm);
void (*FpBecomeLeaderCb)(const struct SSyncFSM* pFsm);
void (*FpBecomeFollowerCb)(const struct SSyncFSM* pFsm);
int32_t (*FpGetSnapshot)(struct SSyncFSM* pFsm, SSnapshot* pSnapshot, void* pReaderParam, void** ppReader);
int32_t (*FpGetSnapshotInfo)(struct SSyncFSM* pFsm, SSnapshot* pSnapshot);
int32_t (*FpGetSnapshot)(const struct SSyncFSM* pFsm, SSnapshot* pSnapshot, void* pReaderParam, void** ppReader);
int32_t (*FpGetSnapshotInfo)(const struct SSyncFSM* pFsm, SSnapshot* pSnapshot);
int32_t (*FpSnapshotStartRead)(struct SSyncFSM* pFsm, void* pReaderParam, void** ppReader);
int32_t (*FpSnapshotStopRead)(struct SSyncFSM* pFsm, void* pReader);
int32_t (*FpSnapshotDoRead)(struct SSyncFSM* pFsm, void* pReader, void** ppBuf, int32_t* len);
int32_t (*FpSnapshotStartRead)(const struct SSyncFSM* pFsm, void* pReaderParam, void** ppReader);
int32_t (*FpSnapshotStopRead)(const struct SSyncFSM* pFsm, void* pReader);
int32_t (*FpSnapshotDoRead)(const struct SSyncFSM* pFsm, void* pReader, void** ppBuf, int32_t* len);
int32_t (*FpSnapshotStartWrite)(struct SSyncFSM* pFsm, void* pWriterParam, void** ppWriter);
int32_t (*FpSnapshotStopWrite)(struct SSyncFSM* pFsm, void* pWriter, bool isApply, SSnapshot* pSnapshot);
int32_t (*FpSnapshotDoWrite)(struct SSyncFSM* pFsm, void* pWriter, void* pBuf, int32_t len);
int32_t (*FpSnapshotStartWrite)(const struct SSyncFSM* pFsm, void* pWriterParam, void** ppWriter);
int32_t (*FpSnapshotStopWrite)(const struct SSyncFSM* pFsm, void* pWriter, bool isApply, SSnapshot* pSnapshot);
int32_t (*FpSnapshotDoWrite)(const struct SSyncFSM* pFsm, void* pWriter, void* pBuf, int32_t len);
} SSyncFSM;
@ -194,48 +192,36 @@ typedef struct SSyncInfo {
SWal* pWal;
SSyncFSM* pFsm;
SMsgCb* msgcb;
int32_t (*FpSendMsg)(const SEpSet* pEpSet, SRpcMsg* pMsg);
int32_t (*FpEqMsg)(const SMsgCb* msgcb, SRpcMsg* pMsg);
int32_t (*FpEqCtrlMsg)(const SMsgCb* msgcb, SRpcMsg* pMsg);
int32_t pingMs;
int32_t electMs;
int32_t heartbeatMs;
int32_t (*syncSendMSg)(const SEpSet* pEpSet, SRpcMsg* pMsg);
int32_t (*syncEqMsg)(const SMsgCb* msgcb, SRpcMsg* pMsg);
int32_t (*syncEqCtrlMsg)(const SMsgCb* msgcb, SRpcMsg* pMsg);
} SSyncInfo;
typedef struct SSyncState {
ESyncState state;
bool restored;
} SSyncState;
int32_t syncInit();
void syncCleanUp();
bool syncIsInit();
int64_t syncOpen(SSyncInfo* pSyncInfo);
void syncStart(int64_t rid);
void syncStop(int64_t rid);
ESyncState syncGetMyRole(int64_t rid);
bool syncIsReady(int64_t rid);
const char* syncGetMyRoleStr(int64_t rid);
bool syncRestoreFinish(int64_t rid);
SyncTerm syncGetMyTerm(int64_t rid);
SyncIndex syncGetLastIndex(int64_t rid);
SyncIndex syncGetCommitIndex(int64_t rid);
SyncGroupId syncGetVgId(int64_t rid);
void syncGetEpSet(int64_t rid, SEpSet* pEpSet);
void syncGetRetryEpSet(int64_t rid, SEpSet* pEpSet);
int32_t syncPropose(int64_t rid, SRpcMsg* pMsg, bool isWeak);
// int32_t syncProposeBatch(int64_t rid, SRpcMsg** pMsgPArr, bool* pIsWeakArr, int32_t arrSize);
const char* syncStr(ESyncState state);
bool syncIsRestoreFinish(int64_t rid);
int32_t syncGetSnapshotByIndex(int64_t rid, SyncIndex index, SSnapshot* pSnapshot);
int32_t syncReconfig(int64_t rid, SSyncCfg* pNewCfg);
// build SRpcMsg, need to call syncPropose with SRpcMsg
int32_t syncReconfigBuild(int64_t rid, const SSyncCfg* pNewCfg, SRpcMsg* pRpcMsg);
int32_t syncLeaderTransfer(int64_t rid);
int32_t syncLeaderTransferTo(int64_t rid, SNodeInfo newLeader);
int32_t syncProcessMsg(int64_t rid, SRpcMsg* pMsg);
int32_t syncReconfig(int64_t rid, SSyncCfg* pCfg);
int32_t syncBeginSnapshot(int64_t rid, int64_t lastApplyIndex);
int32_t syncEndSnapshot(int64_t rid);
int32_t syncLeaderTransfer(int64_t rid);
int32_t syncStepDown(int64_t rid, SyncTerm newTerm);
SSyncNode* syncNodeAcquire(int64_t rid);
void syncNodeRelease(SSyncNode* pNode);
SSyncState syncGetState(int64_t rid);
void syncGetRetryEpSet(int64_t rid, SEpSet* pEpSet);
const char* syncStr(ESyncState state);
#ifdef __cplusplus
}

2
include/libs/wal/wal.h
View File

@ -43,7 +43,6 @@ extern "C" {
#define WAL_FILE_LEN (WAL_PATH_LEN + 32)
#define WAL_MAGIC 0xFAFBFCFDF4F3F2F1ULL
#define WAL_SCAN_BUF_SIZE (1024 * 1024 * 3)
#define WAL_RECOV_SIZE_LIMIT (100 * WAL_SCAN_BUF_SIZE)
typedef enum {
TAOS_WAL_WRITE = 1,
@ -159,6 +158,7 @@ void walCleanUp();
// handle open and ctl
SWal *walOpen(const char *path, SWalCfg *pCfg);
int32_t walAlter(SWal *, SWalCfg *pCfg);
int32_t walPersist(SWal *);
void walClose(SWal *);
// write interfaces

10
include/os/osDef.h
View File

@ -244,6 +244,16 @@ void syslog(int unused, const char *format, ...);
#define TD_CHARSET_LEN 64
#define TD_TIMEZONE_LEN 96
#ifdef WINDOWS
#define TD_PATH_MAX 260
#elif defined(PATH_MAX)
#define TD_PATH_MAX PATH_MAX
#elif defined(_XOPEN_PATH_MAX)
#define TD_PATH_MAX _XOPEN_PATH_MAX
#else
#define TD_PATH_MAX _POSIX_PATH_MAX
#endif
#ifdef __cplusplus
}
#endif

1
include/util/taoserror.h
View File

@ -451,6 +451,7 @@ int32_t* taosGetErrno();
#define TSDB_CODE_WAL_OUT_OF_MEMORY TAOS_DEF_ERROR_CODE(0, 0x1004)
#define TSDB_CODE_WAL_LOG_NOT_EXIST TAOS_DEF_ERROR_CODE(0, 0x1005)
#define TSDB_CODE_WAL_CHKSUM_MISMATCH TAOS_DEF_ERROR_CODE(0, 0x1006)
#define TSDB_CODE_WAL_LOG_INCOMPLETE TAOS_DEF_ERROR_CODE(0, 0x1007)
// tfs
#define TSDB_CODE_FS_INVLD_CFG TAOS_DEF_ERROR_CODE(0, 0x2201)

10
source/client/src/clientTmq.c
View File

@ -508,7 +508,7 @@ static int32_t tmqSendCommitReq(tmq_t* tmq, SMqClientVg* pVg, SMqClientTopic* pT
pMsgSendInfo->param = pParam;
pMsgSendInfo->paramFreeFp = taosMemoryFree;
pMsgSendInfo->fp = tmqCommitCb;
pMsgSendInfo->msgType = TDMT_VND_MQ_COMMIT_OFFSET;
pMsgSendInfo->msgType = TDMT_VND_TMQ_COMMIT_OFFSET;
// send msg
atomic_add_fetch_32(&pParamSet->waitingRspNum, 1);
@ -750,7 +750,7 @@ void tmqSendHbReq(void* param, void* tmrId) {
sendInfo->requestObjRefId = 0;
sendInfo->param = NULL;
sendInfo->fp = tmqHbCb;
sendInfo->msgType = TDMT_MND_MQ_HB;
sendInfo->msgType = TDMT_MND_TMQ_HB;
SEpSet epSet = getEpSet_s(&tmq->pTscObj->pAppInfo->mgmtEp);
@ -1038,7 +1038,7 @@ int32_t tmq_subscribe(tmq_t* tmq, const tmq_list_t* topic_list) {
sendInfo->requestObjRefId = 0;
sendInfo->param = &param;
sendInfo->fp = tmqSubscribeCb;
sendInfo->msgType = TDMT_MND_SUBSCRIBE;
sendInfo->msgType = TDMT_MND_TMQ_SUBSCRIBE;
SEpSet epSet = getEpSet_s(&tmq->pTscObj->pAppInfo->mgmtEp);
@ -1420,7 +1420,7 @@ int32_t tmqAskEp(tmq_t* tmq, bool async) {
sendInfo->requestObjRefId = 0;
sendInfo->param = pParam;
sendInfo->fp = tmqAskEpCb;
sendInfo->msgType = TDMT_MND_MQ_ASK_EP;
sendInfo->msgType = TDMT_MND_TMQ_ASK_EP;
SEpSet epSet = getEpSet_s(&tmq->pTscObj->pAppInfo->mgmtEp);
@ -1573,7 +1573,7 @@ int32_t tmqPollImpl(tmq_t* tmq, int64_t timeout) {
sendInfo->requestObjRefId = 0;
sendInfo->param = pParam;
sendInfo->fp = tmqPollCb;
sendInfo->msgType = TDMT_VND_CONSUME;
sendInfo->msgType = TDMT_VND_TMQ_CONSUME;
int64_t transporterId = 0;
/*printf("send poll\n");*/

41
source/common/src/tdatablock.c
View File

@ -1132,6 +1132,10 @@ void blockDataCleanup(SSDataBlock* pDataBlock) {
pDataBlock->info.window.ekey = 0;
pDataBlock->info.window.skey = 0;
if (pDataBlock->info.capacity == 0) {
return;
}
size_t numOfCols = taosArrayGetSize(pDataBlock->pDataBlock);
for (int32_t i = 0; i < numOfCols; ++i) {
SColumnInfoData* p = taosArrayGet(pDataBlock->pDataBlock, i);
@ -1186,6 +1190,7 @@ static int32_t doEnsureCapacity(SColumnInfoData* pColumn, const SDataBlockInfo*
}
void colInfoDataCleanup(SColumnInfoData* pColumn, uint32_t numOfRows) {
pColumn->hasNull = false;
if (IS_VAR_DATA_TYPE(pColumn->info.type)) {
pColumn->varmeta.length = 0;
if (pColumn->varmeta.offset != NULL) {
@ -2063,8 +2068,8 @@ int32_t buildSubmitReqFromDataBlock(SSubmitReq** pReq, const SSDataBlock* pDataB
default:
if (pColInfoData->info.type < TSDB_DATA_TYPE_MAX && pColInfoData->info.type > TSDB_DATA_TYPE_NULL) {
if (colDataIsNull_s(pColInfoData, j)) {
tdAppendColValToRow(&rb, PRIMARYKEY_TIMESTAMP_COL_ID + k, pCol->type, TD_VTYPE_NULL, NULL, false, offset,
k);
tdAppendColValToRow(&rb, PRIMARYKEY_TIMESTAMP_COL_ID + k, pCol->type, TD_VTYPE_NULL, NULL, false,
offset, k);
} else if (pCol->type == pColInfoData->info.type) {
tdAppendColValToRow(&rb, PRIMARYKEY_TIMESTAMP_COL_ID + k, pCol->type, TD_VTYPE_NORM, var, true, offset,
k);
@ -2137,10 +2142,26 @@ int32_t buildSubmitReqFromDataBlock(SSubmitReq** pReq, const SSDataBlock* pDataB
return TSDB_CODE_SUCCESS;
}
char* buildCtbNameByGroupId(const char* stbName, uint64_t groupId) {
ASSERT(stbName[0] != 0);
char* buildCtbNameByGroupId(const char* stbFullName, uint64_t groupId) {
ASSERT(stbFullName[0] != 0);
SArray* tags = taosArrayInit(0, sizeof(void*));
if (tags == NULL) {
return NULL;
}
SSmlKv* pTag = taosMemoryCalloc(1, sizeof(SSmlKv));
if (pTag == NULL) {
taosArrayDestroy(tags);
return NULL;
}
void* cname = taosMemoryCalloc(1, TSDB_TABLE_NAME_LEN + 1);
if (cname == NULL) {
taosArrayDestroy(tags);
taosMemoryFree(pTag);
return NULL;
}
pTag->key = "group_id";
pTag->keyLen = strlen(pTag->key);
pTag->type = TSDB_DATA_TYPE_UBIGINT;
@ -2148,13 +2169,11 @@ char* buildCtbNameByGroupId(const char* stbName, uint64_t groupId) {
pTag->length = sizeof(uint64_t);
taosArrayPush(tags, &pTag);
void* cname = taosMemoryCalloc(1, TSDB_TABLE_NAME_LEN + 1);
RandTableName rname = {
.tags = tags,
.sTableName = stbName,
.sTableNameLen = strlen(stbName),
.childTableName = cname,
.stbFullName = stbFullName,
.stbFullNameLen = strlen(stbFullName),
.ctbShortName = cname,
};
buildChildTableName(&rname);
@ -2162,8 +2181,8 @@ char* buildCtbNameByGroupId(const char* stbName, uint64_t groupId) {
taosMemoryFree(pTag);
taosArrayDestroy(tags);
ASSERT(rname.childTableName && rname.childTableName[0]);
return rname.childTableName;
ASSERT(rname.ctbShortName && rname.ctbShortName[0]);
return rname.ctbShortName;
}
void blockEncode(const SSDataBlock* pBlock, char* data, int32_t* dataLen, int32_t numOfCols, int8_t needCompress) {

8
source/common/src/tglobal.c
View File

@ -156,6 +156,9 @@ char tsCompressor[32] = "ZSTD_COMPRESSOR"; // ZSTD_COMPRESSOR or GZIP_COMPR
// udf
bool tsStartUdfd = true;
// wal
int64_t tsWalFsyncDataSizeLimit = (100 * 1024 * 1024L);
// internal
int32_t tsTransPullupInterval = 2;
int32_t tsMqRebalanceInterval = 2;
@ -423,6 +426,9 @@ static int32_t taosAddServerCfg(SConfig *pCfg) {
if (cfgAddInt32(pCfg, "uptimeInterval", tsUptimeInterval, 1, 100000, 1) != 0) return -1;
if (cfgAddInt32(pCfg, "queryRsmaTolerance", tsQueryRsmaTolerance, 0, 900000, 0) != 0) return -1;
if (cfgAddInt64(pCfg, "walFsyncDataSizeLimit", tsWalFsyncDataSizeLimit, 100 * 1024 * 1024, INT64_MAX, 0) != 0)
return -1;
if (cfgAddBool(pCfg, "udf", tsStartUdfd, 0) != 0) return -1;
if (cfgAddString(pCfg, "udfdResFuncs", tsUdfdResFuncs, 0) != 0) return -1;
if (cfgAddString(pCfg, "udfdLdLibPath", tsUdfdLdLibPath, 0) != 0) return -1;
@ -722,6 +728,8 @@ static int32_t taosSetServerCfg(SConfig *pCfg) {
tsUptimeInterval = cfgGetItem(pCfg, "uptimeInterval")->i32;
tsQueryRsmaTolerance = cfgGetItem(pCfg, "queryRsmaTolerance")->i32;
tsWalFsyncDataSizeLimit = cfgGetItem(pCfg, "walFsyncDataSizeLimit")->i64;
tsStartUdfd = cfgGetItem(pCfg, "udf")->bval;
tstrncpy(tsUdfdResFuncs, cfgGetItem(pCfg, "udfdResFuncs")->str, sizeof(tsUdfdResFuncs));
tstrncpy(tsUdfdLdLibPath, cfgGetItem(pCfg, "udfdLdLibPath")->str, sizeof(tsUdfdLdLibPath));

8
source/common/src/tname.c
View File

@ -315,7 +315,7 @@ static int compareKv(const void* p1, const void* p2) {
*/
void buildChildTableName(RandTableName* rName) {
SStringBuilder sb = {0};
taosStringBuilderAppendStringLen(&sb, rName->sTableName, rName->sTableNameLen);
taosStringBuilderAppendStringLen(&sb, rName->stbFullName, rName->stbFullNameLen);
taosArraySort(rName->tags, compareKv);
for (int j = 0; j < taosArrayGetSize(rName->tags); ++j) {
taosStringBuilderAppendChar(&sb, ',');
@ -336,11 +336,11 @@ void buildChildTableName(RandTableName* rName) {
tMD5Final(&context);
char temp[8] = {0};
rName->childTableName[0] = 't';
rName->childTableName[1] = '_';
rName->ctbShortName[0] = 't';
rName->ctbShortName[1] = '_';
for (int i = 0; i < 16; i++) {
sprintf(temp, "%02x", context.digest[i]);
strcat(rName->childTableName, temp);
strcat(rName->ctbShortName, temp);
}
taosStringBuilderDestroy(&sb);
rName->uid = *(uint64_t*)(context.digest);

14
source/dnode/mgmt/CMakeLists.txt
View File

@ -13,4 +13,18 @@ target_include_directories(
taosd
PRIVATE "${CMAKE_CURRENT_SOURCE_DIR}/node_mgmt/inc"
)
IF (TD_LINUX_64 AND JEMALLOC_ENABLED)
ADD_DEFINITIONS(-DTD_JEMALLOC_ENABLED -I${CMAKE_BINARY_DIR}/build/include -L${CMAKE_BINARY_DIR}/build/lib -Wl,-rpath,${CMAKE_BINARY_DIR}/build/lib -ljemalloc)
SET(LINK_JEMALLOC "-L${CMAKE_BINARY_DIR}/build/lib -ljemalloc")
ELSE ()
SET(LINK_JEMALLOC "")
ENDIF ()
IF (TD_LINUX_64 AND JEMALLOC_ENABLED)
ADD_DEPENDENCIES(taosd jemalloc)
target_link_libraries(taosd dnode ${LINK_JEMALLOC})
ELSE ()
target_link_libraries(taosd dnode)
ENDIF ()

24
source/dnode/mgmt/mgmt_mnode/src/mmHandle.c
View File

@ -141,15 +141,13 @@ SArray *mmGetMsgHandles() {
if (dmSetMgmtHandle(pArray, TDMT_MND_DROP_STREAM, mmPutMsgToWriteQueue, 0) == NULL) goto _OVER;
if (dmSetMgmtHandle(pArray, TDMT_MND_GET_INDEX, mmPutMsgToReadQueue, 0) == NULL) goto _OVER;
if (dmSetMgmtHandle(pArray, TDMT_MND_GET_TABLE_INDEX, mmPutMsgToReadQueue, 0) == NULL) goto _OVER;
if (dmSetMgmtHandle(pArray, TDMT_MND_CREATE_TOPIC, mmPutMsgToWriteQueue, 0) == NULL) goto _OVER;
if (dmSetMgmtHandle(pArray, TDMT_MND_ALTER_TOPIC, mmPutMsgToWriteQueue, 0) == NULL) goto _OVER;
if (dmSetMgmtHandle(pArray, TDMT_MND_DROP_TOPIC, mmPutMsgToWriteQueue, 0) == NULL) goto _OVER;
if (dmSetMgmtHandle(pArray, TDMT_MND_SUBSCRIBE, mmPutMsgToWriteQueue, 0) == NULL) goto _OVER;
if (dmSetMgmtHandle(pArray, TDMT_MND_MQ_ASK_EP, mmPutMsgToReadQueue, 0) == NULL) goto _OVER;
if (dmSetMgmtHandle(pArray, TDMT_MND_MQ_HB, mmPutMsgToReadQueue, 0) == NULL) goto _OVER;
if (dmSetMgmtHandle(pArray, TDMT_MND_MQ_DROP_CGROUP, mmPutMsgToWriteQueue, 0) == NULL) goto _OVER;
if (dmSetMgmtHandle(pArray, TDMT_MND_MQ_DROP_CGROUP_RSP, mmPutMsgToWriteQueue, 0) == NULL) goto _OVER;
if (dmSetMgmtHandle(pArray, TDMT_MND_MQ_COMMIT_OFFSET, mmPutMsgToWriteQueue, 0) == NULL) goto _OVER;
if (dmSetMgmtHandle(pArray, TDMT_MND_TMQ_CREATE_TOPIC, mmPutMsgToWriteQueue, 0) == NULL) goto _OVER;
if (dmSetMgmtHandle(pArray, TDMT_MND_TMQ_DROP_TOPIC, mmPutMsgToWriteQueue, 0) == NULL) goto _OVER;
if (dmSetMgmtHandle(pArray, TDMT_MND_TMQ_SUBSCRIBE, mmPutMsgToWriteQueue, 0) == NULL) goto _OVER;
if (dmSetMgmtHandle(pArray, TDMT_MND_TMQ_ASK_EP, mmPutMsgToReadQueue, 0) == NULL) goto _OVER;
if (dmSetMgmtHandle(pArray, TDMT_MND_TMQ_HB, mmPutMsgToReadQueue, 0) == NULL) goto _OVER;
if (dmSetMgmtHandle(pArray, TDMT_MND_TMQ_DROP_CGROUP, mmPutMsgToWriteQueue, 0) == NULL) goto _OVER;
if (dmSetMgmtHandle(pArray, TDMT_MND_TMQ_DROP_CGROUP_RSP, mmPutMsgToWriteQueue, 0) == NULL) goto _OVER;
if (dmSetMgmtHandle(pArray, TDMT_MND_KILL_TRANS, mmPutMsgToWriteQueue, 0) == NULL) goto _OVER;
if (dmSetMgmtHandle(pArray, TDMT_MND_KILL_QUERY, mmPutMsgToWriteQueue, 0) == NULL) goto _OVER;
if (dmSetMgmtHandle(pArray, TDMT_MND_KILL_CONN, mmPutMsgToWriteQueue, 0) == NULL) goto _OVER;
@ -171,10 +169,10 @@ SArray *mmGetMsgHandles() {
if (dmSetMgmtHandle(pArray, TDMT_VND_DROP_STB_RSP, mmPutMsgToWriteQueue, 0) == NULL) goto _OVER;
if (dmSetMgmtHandle(pArray, TDMT_VND_CREATE_SMA_RSP, mmPutMsgToWriteQueue, 0) == NULL) goto _OVER;
if (dmSetMgmtHandle(pArray, TDMT_VND_DROP_SMA_RSP, mmPutMsgToWriteQueue, 0) == NULL) goto _OVER;
if (dmSetMgmtHandle(pArray, TDMT_VND_MQ_VG_CHANGE_RSP, mmPutMsgToWriteQueue, 0) == NULL) goto _OVER;
if (dmSetMgmtHandle(pArray, TDMT_VND_MQ_VG_DELETE_RSP, mmPutMsgToWriteQueue, 0) == NULL) goto _OVER;
if (dmSetMgmtHandle(pArray, TDMT_VND_ADD_CHECK_INFO_RSP, mmPutMsgToWriteQueue, 0) == NULL) goto _OVER;
if (dmSetMgmtHandle(pArray, TDMT_VND_DELETE_CHECK_INFO_RSP, mmPutMsgToWriteQueue, 0) == NULL) goto _OVER;
if (dmSetMgmtHandle(pArray, TDMT_VND_TMQ_SUBSCRIBE_RSP, mmPutMsgToWriteQueue, 0) == NULL) goto _OVER;
if (dmSetMgmtHandle(pArray, TDMT_VND_TMQ_DELETE_SUB_RSP, mmPutMsgToWriteQueue, 0) == NULL) goto _OVER;
if (dmSetMgmtHandle(pArray, TDMT_VND_TMQ_ADD_CHECKINFO_RSP, mmPutMsgToWriteQueue, 0) == NULL) goto _OVER;
if (dmSetMgmtHandle(pArray, TDMT_VND_TMQ_DEL_CHECKINFO_RSP, mmPutMsgToWriteQueue, 0) == NULL) goto _OVER;
if (dmSetMgmtHandle(pArray, TDMT_SCH_DROP_TASK, mmPutMsgToFetchQueue, 1) == NULL) goto _OVER;
if (dmSetMgmtHandle(pArray, TDMT_STREAM_TASK_DEPLOY_RSP, mmPutMsgToWriteQueue, 0) == NULL) goto _OVER;
if (dmSetMgmtHandle(pArray, TDMT_STREAM_TASK_DROP_RSP, mmPutMsgToWriteQueue, 0) == NULL) goto _OVER;

20
source/dnode/mgmt/mgmt_mnode/src/mmWorker.c
View File

@ -67,24 +67,6 @@ static void mmProcessRpcMsg(SQueueInfo *pInfo, SRpcMsg *pMsg) {
taosFreeQitem(pMsg);
}
static void mmProcessSyncCtrlMsg(SQueueInfo *pInfo, SRpcMsg *pMsg) {
SMnodeMgmt *pMgmt = pInfo->ahandle;
pMsg->info.node = pMgmt->pMnode;
const STraceId *trace = &pMsg->info.traceId;
dGTrace("msg:%p, get from mnode-sync-ctrl queue", pMsg);
SMsgHead *pHead = pMsg->pCont;
pHead->contLen = ntohl(pHead->contLen);
pHead->vgId = ntohl(pHead->vgId);
int32_t code = mndProcessSyncCtrlMsg(pMsg);
dGTrace("msg:%p, is freed, code:0x%x", pMsg, code);
rpcFreeCont(pMsg->pCont);
taosFreeQitem(pMsg);
}
static void mmProcessSyncMsg(SQueueInfo *pInfo, SRpcMsg *pMsg) {
SMnodeMgmt *pMgmt = pInfo->ahandle;
pMsg->info.node = pMgmt->pMnode;
@ -252,7 +234,7 @@ int32_t mmStartWorker(SMnodeMgmt *pMgmt) {
.min = 1,
.max = 1,
.name = "mnode-sync-ctrl",
.fp = (FItem)mmProcessSyncCtrlMsg,
.fp = (FItem)mmProcessSyncMsg,
.param = pMgmt,
};
if (tSingleWorkerInit(&pMgmt->syncCtrlWorker, &scCfg) != 0) {

2
source/dnode/mgmt/mgmt_snode/src/smHandle.c
View File

@ -72,8 +72,6 @@ SArray *smGetMsgHandles() {
if (dmSetMgmtHandle(pArray, TDMT_STREAM_TASK_RUN, smPutNodeMsgToStreamQueue, 1) == NULL) goto _OVER;
if (dmSetMgmtHandle(pArray, TDMT_STREAM_TASK_DISPATCH, smPutNodeMsgToStreamQueue, 1) == NULL) goto _OVER;
if (dmSetMgmtHandle(pArray, TDMT_STREAM_TASK_DISPATCH_RSP, smPutNodeMsgToStreamQueue, 1) == NULL) goto _OVER;
if (dmSetMgmtHandle(pArray, TDMT_STREAM_TASK_RECOVER, smPutNodeMsgToStreamQueue, 1) == NULL) goto _OVER;
if (dmSetMgmtHandle(pArray, TDMT_STREAM_TASK_RECOVER_RSP, smPutNodeMsgToStreamQueue, 1) == NULL) goto _OVER;
if (dmSetMgmtHandle(pArray, TDMT_STREAM_RETRIEVE, smPutNodeMsgToStreamQueue, 1) == NULL) goto _OVER;
if (dmSetMgmtHandle(pArray, TDMT_STREAM_RETRIEVE_RSP, smPutNodeMsgToStreamQueue, 1) == NULL) goto _OVER;

15
source/dnode/mgmt/mgmt_vnode/src/vmHandle.c
View File

@ -392,12 +392,12 @@ SArray *vmGetMsgHandles() {
if (dmSetMgmtHandle(pArray, TDMT_VND_CANCEL_SMA, vmPutMsgToWriteQueue, 0) == NULL) goto _OVER;
if (dmSetMgmtHandle(pArray, TDMT_VND_DROP_SMA, vmPutMsgToWriteQueue, 0) == NULL) goto _OVER;
if (dmSetMgmtHandle(pArray, TDMT_VND_SUBMIT_RSMA, vmPutMsgToWriteQueue, 0) == NULL) goto _OVER;
if (dmSetMgmtHandle(pArray, TDMT_VND_MQ_VG_CHANGE, vmPutMsgToWriteQueue, 0) == NULL) goto _OVER;
if (dmSetMgmtHandle(pArray, TDMT_VND_MQ_VG_DELETE, vmPutMsgToWriteQueue, 0) == NULL) goto _OVER;
if (dmSetMgmtHandle(pArray, TDMT_VND_MQ_COMMIT_OFFSET, vmPutMsgToWriteQueue, 0) == NULL) goto _OVER;
if (dmSetMgmtHandle(pArray, TDMT_VND_ADD_CHECK_INFO, vmPutMsgToWriteQueue, 0) == NULL) goto _OVER;
if (dmSetMgmtHandle(pArray, TDMT_VND_DELETE_CHECK_INFO, vmPutMsgToWriteQueue, 0) == NULL) goto _OVER;
if (dmSetMgmtHandle(pArray, TDMT_VND_CONSUME, vmPutMsgToFetchQueue, 0) == NULL) goto _OVER;
if (dmSetMgmtHandle(pArray, TDMT_VND_TMQ_SUBSCRIBE, vmPutMsgToWriteQueue, 0) == NULL) goto _OVER;
if (dmSetMgmtHandle(pArray, TDMT_VND_TMQ_DELETE_SUB, vmPutMsgToWriteQueue, 0) == NULL) goto _OVER;
if (dmSetMgmtHandle(pArray, TDMT_VND_TMQ_COMMIT_OFFSET, vmPutMsgToWriteQueue, 0) == NULL) goto _OVER;
if (dmSetMgmtHandle(pArray, TDMT_VND_TMQ_ADD_CHECKINFO, vmPutMsgToWriteQueue, 0) == NULL) goto _OVER;
if (dmSetMgmtHandle(pArray, TDMT_VND_TMQ_DEL_CHECKINFO, vmPutMsgToWriteQueue, 0) == NULL) goto _OVER;
if (dmSetMgmtHandle(pArray, TDMT_VND_TMQ_CONSUME, vmPutMsgToFetchQueue, 0) == NULL) goto _OVER;
if (dmSetMgmtHandle(pArray, TDMT_VND_DELETE, vmPutMsgToWriteQueue, 0) == NULL) goto _OVER;
if (dmSetMgmtHandle(pArray, TDMT_VND_BATCH_DEL, vmPutMsgToWriteQueue, 0) == NULL) goto _OVER;
if (dmSetMgmtHandle(pArray, TDMT_VND_COMMIT, vmPutMsgToWriteQueue, 0) == NULL) goto _OVER;
@ -409,10 +409,9 @@ SArray *vmGetMsgHandles() {
if (dmSetMgmtHandle(pArray, TDMT_STREAM_TASK_RUN, vmPutMsgToStreamQueue, 0) == NULL) goto _OVER;
if (dmSetMgmtHandle(pArray, TDMT_STREAM_TASK_DISPATCH, vmPutMsgToStreamQueue, 0) == NULL) goto _OVER;
if (dmSetMgmtHandle(pArray, TDMT_STREAM_TASK_DISPATCH_RSP, vmPutMsgToStreamQueue, 0) == NULL) goto _OVER;
if (dmSetMgmtHandle(pArray, TDMT_STREAM_TASK_RECOVER, vmPutMsgToStreamQueue, 0) == NULL) goto _OVER;
if (dmSetMgmtHandle(pArray, TDMT_STREAM_TASK_RECOVER_RSP, vmPutMsgToStreamQueue, 0) == NULL) goto _OVER;
if (dmSetMgmtHandle(pArray, TDMT_STREAM_RETRIEVE, vmPutMsgToStreamQueue, 0) == NULL) goto _OVER;
if (dmSetMgmtHandle(pArray, TDMT_STREAM_RETRIEVE_RSP, vmPutMsgToStreamQueue, 0) == NULL) goto _OVER;
if (dmSetMgmtHandle(pArray, TDMT_STREAM_RECOVER_FINISH, vmPutMsgToStreamQueue, 0) == NULL) goto _OVER;
if (dmSetMgmtHandle(pArray, TDMT_VND_ALTER_REPLICA, vmPutMsgToMgmtQueue, 0) == NULL) goto _OVER;
if (dmSetMgmtHandle(pArray, TDMT_VND_ALTER_CONFIG, vmPutMsgToWriteQueue, 0) == NULL) goto _OVER;

18
source/dnode/mgmt/mgmt_vnode/src/vmWorker.c
View File

@ -133,22 +133,6 @@ static void vmProcessSyncQueue(SQueueInfo *pInfo, STaosQall *qall, int32_t numOf
}
}
static void vmProcessSyncCtrlQueue(SQueueInfo *pInfo, STaosQall *qall, int32_t numOfMsgs) {
SVnodeObj *pVnode = pInfo->ahandle;
SRpcMsg *pMsg = NULL;
for (int32_t i = 0; i < numOfMsgs; ++i) {
if (taosGetQitem(qall, (void **)&pMsg) == 0) continue;
const STraceId *trace = &pMsg->info.traceId;
dGTrace("vgId:%d, msg:%p get from vnode-sync queue", pVnode->vgId, pMsg);
int32_t code = vnodeProcessSyncCtrlMsg(pVnode->pImpl, pMsg, NULL); // no response here
dGTrace("vgId:%d, msg:%p is freed, code:0x%x", pVnode->vgId, pMsg, code);
rpcFreeCont(pMsg->pCont);
taosFreeQitem(pMsg);
}
}
static int32_t vmPutMsgToQueue(SVnodeMgmt *pMgmt, SRpcMsg *pMsg, EQueueType qtype) {
const STraceId *trace = &pMsg->info.traceId;
SMsgHead *pHead = pMsg->pCont;
@ -317,7 +301,7 @@ int32_t vmGetQueueSize(SVnodeMgmt *pMgmt, int32_t vgId, EQueueType qtype) {
int32_t vmAllocQueue(SVnodeMgmt *pMgmt, SVnodeObj *pVnode) {
pVnode->pWriteQ = tWWorkerAllocQueue(&pMgmt->writePool, pVnode->pImpl, (FItems)vnodeProposeWriteMsg);
pVnode->pSyncQ = tWWorkerAllocQueue(&pMgmt->syncPool, pVnode, (FItems)vmProcessSyncQueue);
pVnode->pSyncCtrlQ = tWWorkerAllocQueue(&pMgmt->syncCtrlPool, pVnode, (FItems)vmProcessSyncCtrlQueue);
pVnode->pSyncCtrlQ = tWWorkerAllocQueue(&pMgmt->syncCtrlPool, pVnode, (FItems)vmProcessSyncQueue);
pVnode->pApplyQ = tWWorkerAllocQueue(&pMgmt->applyPool, pVnode->pImpl, (FItems)vnodeApplyWriteMsg);
pVnode->pQueryQ = tQWorkerAllocQueue(&pMgmt->queryPool, pVnode, (FItem)vmProcessQueryQueue);
pVnode->pStreamQ = tQWorkerAllocQueue(&pMgmt->streamPool, pVnode, (FItem)vmProcessStreamQueue);

2
source/dnode/mgmt/node_util/src/dmUtil.c
View File

@ -55,7 +55,7 @@ void *dmSetMgmtHandle(SArray *pArray, tmsg_t msgType, void *nodeMsgFp, bool need
}
void dmGetMonitorSystemInfo(SMonSysInfo *pInfo) {
taosGetCpuUsage(&pInfo->cpu_engine, &pInfo->cpu_system);
taosGetCpuUsage(&pInfo->cpu_system, &pInfo->cpu_engine);
taosGetCpuCores(&pInfo->cpu_cores);
taosGetProcMemory(&pInfo->mem_engine);
taosGetSysMemory(&pInfo->mem_system);

2
source/dnode/mnode/impl/inc/mndInt.h
View File

@ -20,7 +20,6 @@
#include "sdb.h"
#include "sync.h"
#include "syncTools.h"
#include "tcache.h"
#include "tdatablock.h"
#include "tglobal.h"
@ -90,7 +89,6 @@ typedef struct {
int32_t errCode;
int32_t transId;
SRWLatch lock;
int8_t leaderTransferFinish;
int8_t selfIndex;
int8_t numOfReplicas;
SReplica replicas[TSDB_MAX_REPLICA];

20
source/dnode/mnode/impl/src/mndConsumer.c
View File

@ -63,12 +63,12 @@ int32_t mndInitConsumer(SMnode *pMnode) {
.deleteFp = (SdbDeleteFp)mndConsumerActionDelete,
};
mndSetMsgHandle(pMnode, TDMT_MND_SUBSCRIBE, mndProcessSubscribeReq);
mndSetMsgHandle(pMnode, TDMT_MND_MQ_HB, mndProcessMqHbReq);
mndSetMsgHandle(pMnode, TDMT_MND_MQ_ASK_EP, mndProcessAskEpReq);
mndSetMsgHandle(pMnode, TDMT_MND_MQ_TIMER, mndProcessMqTimerMsg);
mndSetMsgHandle(pMnode, TDMT_MND_MQ_CONSUMER_LOST, mndProcessConsumerLostMsg);
mndSetMsgHandle(pMnode, TDMT_MND_MQ_CONSUMER_RECOVER, mndProcessConsumerRecoverMsg);
mndSetMsgHandle(pMnode, TDMT_MND_TMQ_SUBSCRIBE, mndProcessSubscribeReq);
mndSetMsgHandle(pMnode, TDMT_MND_TMQ_HB, mndProcessMqHbReq);
mndSetMsgHandle(pMnode, TDMT_MND_TMQ_ASK_EP, mndProcessAskEpReq);
mndSetMsgHandle(pMnode, TDMT_MND_TMQ_TIMER, mndProcessMqTimerMsg);
mndSetMsgHandle(pMnode, TDMT_MND_TMQ_CONSUMER_LOST, mndProcessConsumerLostMsg);
mndSetMsgHandle(pMnode, TDMT_MND_TMQ_CONSUMER_RECOVER, mndProcessConsumerRecoverMsg);
mndAddShowRetrieveHandle(pMnode, TSDB_MGMT_TABLE_CONSUMERS, mndRetrieveConsumer);
mndAddShowFreeIterHandle(pMnode, TSDB_MGMT_TABLE_CONSUMERS, mndCancelGetNextConsumer);
@ -207,7 +207,7 @@ static int32_t mndProcessMqTimerMsg(SRpcMsg *pMsg) {
pLostMsg->consumerId = pConsumer->consumerId;
SRpcMsg pRpcMsg = {
.msgType = TDMT_MND_MQ_CONSUMER_LOST,
.msgType = TDMT_MND_TMQ_CONSUMER_LOST,
.pCont = pLostMsg,
.contLen = sizeof(SMqConsumerLostMsg),
};
@ -256,7 +256,7 @@ static int32_t mndProcessMqTimerMsg(SRpcMsg *pMsg) {
if (taosHashGetSize(pRebMsg->rebSubHash) != 0) {
mInfo("mq rebalance will be triggered");
SRpcMsg rpcMsg = {
.msgType = TDMT_MND_MQ_DO_REBALANCE,
.msgType = TDMT_MND_TMQ_DO_REBALANCE,
.pCont = pRebMsg,
.contLen = sizeof(SMqDoRebalanceMsg),
};
@ -292,7 +292,7 @@ static int32_t mndProcessMqHbReq(SRpcMsg *pMsg) {
pRecoverMsg->consumerId = consumerId;
SRpcMsg pRpcMsg = {
.msgType = TDMT_MND_MQ_CONSUMER_RECOVER,
.msgType = TDMT_MND_TMQ_CONSUMER_RECOVER,
.pCont = pRecoverMsg,
.contLen = sizeof(SMqConsumerRecoverMsg),
};
@ -331,7 +331,7 @@ static int32_t mndProcessAskEpReq(SRpcMsg *pMsg) {
pRecoverMsg->consumerId = consumerId;
SRpcMsg pRpcMsg = {
.msgType = TDMT_MND_MQ_CONSUMER_RECOVER,
.msgType = TDMT_MND_TMQ_CONSUMER_RECOVER,
.pCont = pRecoverMsg,
.contLen = sizeof(SMqConsumerRecoverMsg),
};

7
source/dnode/mnode/impl/src/mndDnode.c
View File

@ -403,6 +403,10 @@ static int32_t mndProcessStatusReq(SRpcMsg *pReq) {
bool reboot = (pDnode->rebootTime != statusReq.rebootTime);
bool needCheck = !online || dnodeChanged || reboot;
pDnode->accessTimes++;
mTrace("dnode:%d, status received, access times:%d check:%d online:%d reboot:%d changed:%d", pDnode->id,
pDnode->accessTimes, needCheck, online, reboot, dnodeChanged);
if (needCheck) {
if (statusReq.sver != tsVersion) {
if (pDnode != NULL) {
@ -424,9 +428,6 @@ static int32_t mndProcessStatusReq(SRpcMsg *pReq) {
pMnode->clusterId);
terrno = TSDB_CODE_MND_INVALID_CLUSTER_ID;
goto _OVER;
} else {
pDnode->accessTimes++;
mDebug("dnode:%d, status received, access times %d", pDnode->id, pDnode->accessTimes);
}
}

150
source/dnode/mnode/impl/src/mndMain.c
View File

@ -105,7 +105,7 @@ static void mndCalMqRebalance(SMnode *pMnode) {
int32_t contLen = 0;
void *pReq = mndBuildTimerMsg(&contLen);
if (pReq != NULL) {
SRpcMsg rpcMsg = {.msgType = TDMT_MND_MQ_TIMER, .pCont = pReq, .contLen = contLen};
SRpcMsg rpcMsg = {.msgType = TDMT_MND_TMQ_TIMER, .pCont = pReq, .contLen = contLen};
tmsgPutToQueue(&pMnode->msgCb, READ_QUEUE, &rpcMsg);
}
}
@ -428,18 +428,7 @@ SMnode *mndOpen(const char *path, const SMnodeOpt *pOption) {
void mndPreClose(SMnode *pMnode) {
if (pMnode != NULL) {
atomic_store_8(&(pMnode->syncMgmt.leaderTransferFinish), 0);
syncLeaderTransfer(pMnode->syncMgmt.sync);
#if 0
mInfo("vgId:1, mnode start leader transfer");
// wait for leader transfer finish
while (!atomic_load_8(&(pMnode->syncMgmt.leaderTransferFinish))) {
taosMsleep(10);
mInfo("vgId:1, mnode waiting for leader transfer");
}
mInfo("vgId:1, mnode finish leader transfer");
#endif
}
}
@ -474,128 +463,18 @@ void mndStop(SMnode *pMnode) {
mndCleanupTimer(pMnode);
}
int32_t mndProcessSyncCtrlMsg(SRpcMsg *pMsg) {
SMnode *pMnode = pMsg->info.node;
SSyncMgmt *pMgmt = &pMnode->syncMgmt;
int32_t code = 0;
mInfo("vgId:%d, process sync ctrl msg", 1);
if (!syncIsInit()) {
mError("failed to process sync msg:%p type:%s since syncEnv stop", pMsg, TMSG_INFO(pMsg->msgType));
terrno = TSDB_CODE_SYN_INTERNAL_ERROR;
return -1;
}
SSyncNode *pSyncNode = syncNodeAcquire(pMgmt->sync);
if (pSyncNode == NULL) {
mError("failed to process sync msg:%p type:%s since syncNode is null", pMsg, TMSG_INFO(pMsg->msgType));
terrno = TSDB_CODE_SYN_INTERNAL_ERROR;
return -1;
}
if (pMsg->msgType == TDMT_SYNC_HEARTBEAT) {
SyncHeartbeat *pSyncMsg = syncHeartbeatFromRpcMsg2(pMsg);
code = syncNodeOnHeartbeat(pSyncNode, pSyncMsg);
syncHeartbeatDestroy(pSyncMsg);
} else if (pMsg->msgType == TDMT_SYNC_HEARTBEAT_REPLY) {
SyncHeartbeatReply *pSyncMsg = syncHeartbeatReplyFromRpcMsg2(pMsg);
code = syncNodeOnHeartbeatReply(pSyncNode, pSyncMsg);
syncHeartbeatReplyDestroy(pSyncMsg);
}
syncNodeRelease(pSyncNode);
if (code != 0) {
terrno = TSDB_CODE_SYN_INTERNAL_ERROR;
}
return code;
}
int32_t mndProcessSyncMsg(SRpcMsg *pMsg) {
SMnode *pMnode = pMsg->info.node;
SSyncMgmt *pMgmt = &pMnode->syncMgmt;
int32_t code = 0;
if (!syncIsInit()) {
mError("failed to process sync msg:%p type:%s since syncEnv stop", pMsg, TMSG_INFO(pMsg->msgType));
terrno = TSDB_CODE_SYN_INTERNAL_ERROR;
return -1;
}
SSyncNode *pSyncNode = syncNodeAcquire(pMgmt->sync);
if (pSyncNode == NULL) {
mError("failed to process sync msg:%p type:%s since syncNode is null", pMsg, TMSG_INFO(pMsg->msgType));
terrno = TSDB_CODE_SYN_INTERNAL_ERROR;
return -1;
}
if (pMsg->msgType == TDMT_SYNC_TIMEOUT) {
SyncTimeout *pSyncMsg = syncTimeoutFromRpcMsg2(pMsg);
code = syncNodeOnTimer(pSyncNode, pSyncMsg);
syncTimeoutDestroy(pSyncMsg);
} else if (pMsg->msgType == TDMT_SYNC_PING) {
SyncPing *pSyncMsg = syncPingFromRpcMsg2(pMsg);
code = syncNodeOnPing(pSyncNode, pSyncMsg);
syncPingDestroy(pSyncMsg);
} else if (pMsg->msgType == TDMT_SYNC_PING_REPLY) {
SyncPingReply *pSyncMsg = syncPingReplyFromRpcMsg2(pMsg);
code = syncNodeOnPingReply(pSyncNode, pSyncMsg);
syncPingReplyDestroy(pSyncMsg);
} else if (pMsg->msgType == TDMT_SYNC_CLIENT_REQUEST) {
SyncClientRequest *pSyncMsg = syncClientRequestFromRpcMsg2(pMsg);
code = syncNodeOnClientRequest(pSyncNode, pSyncMsg, NULL);
syncClientRequestDestroy(pSyncMsg);
} else if (pMsg->msgType == TDMT_SYNC_REQUEST_VOTE) {
SyncRequestVote *pSyncMsg = syncRequestVoteFromRpcMsg2(pMsg);
code = syncNodeOnRequestVote(pSyncNode, pSyncMsg);
syncRequestVoteDestroy(pSyncMsg);
} else if (pMsg->msgType == TDMT_SYNC_REQUEST_VOTE_REPLY) {
SyncRequestVoteReply *pSyncMsg = syncRequestVoteReplyFromRpcMsg2(pMsg);
code = syncNodeOnRequestVoteReply(pSyncNode, pSyncMsg);
syncRequestVoteReplyDestroy(pSyncMsg);
} else if (pMsg->msgType == TDMT_SYNC_APPEND_ENTRIES) {
SyncAppendEntries *pSyncMsg = syncAppendEntriesFromRpcMsg2(pMsg);
code = syncNodeOnAppendEntries(pSyncNode, pSyncMsg);
syncAppendEntriesDestroy(pSyncMsg);
} else if (pMsg->msgType == TDMT_SYNC_APPEND_ENTRIES_REPLY) {
SyncAppendEntriesReply *pSyncMsg = syncAppendEntriesReplyFromRpcMsg2(pMsg);
code = syncNodeOnAppendEntriesReply(pSyncNode, pSyncMsg);
syncAppendEntriesReplyDestroy(pSyncMsg);
} else if (pMsg->msgType == TDMT_SYNC_SNAPSHOT_SEND) {
SyncSnapshotSend *pSyncMsg = syncSnapshotSendFromRpcMsg2(pMsg);
code = syncNodeOnSnapshot(pSyncNode, pSyncMsg);
syncSnapshotSendDestroy(pSyncMsg);
} else if (pMsg->msgType == TDMT_SYNC_SNAPSHOT_RSP) {
SyncSnapshotRsp *pSyncMsg = syncSnapshotRspFromRpcMsg2(pMsg);
code = syncNodeOnSnapshotReply(pSyncNode, pSyncMsg);
syncSnapshotRspDestroy(pSyncMsg);
} else if (pMsg->msgType == TDMT_SYNC_LOCAL_CMD) {
SyncLocalCmd *pSyncMsg = syncLocalCmdFromRpcMsg2(pMsg);
code = syncNodeOnLocalCmd(pSyncNode, pSyncMsg);
syncLocalCmdDestroy(pSyncMsg);
} else {
mError("failed to process msg:%p since invalid type:%s", pMsg, TMSG_INFO(pMsg->msgType));
code = -1;
}
syncNodeRelease(pSyncNode);
const STraceId *trace = &pMsg->info.traceId;
mGTrace("vgId:1, sync msg:%p will be processed, type:%s", pMsg, TMSG_INFO(pMsg->msgType));
int32_t code = syncProcessMsg(pMgmt->sync, pMsg);
if (code != 0) {
terrno = TSDB_CODE_SYN_INTERNAL_ERROR;
mGError("vgId:1, failed to process sync msg:%p type:%s since %s", pMsg, TMSG_INFO(pMsg->msgType), terrstr());
}
return code;
}
@ -609,13 +488,13 @@ static int32_t mndCheckMnodeState(SRpcMsg *pMsg) {
if (mndAcquireRpc(pMsg->info.node) == 0) return 0;
SMnode *pMnode = pMsg->info.node;
const char *role = syncGetMyRoleStr(pMnode->syncMgmt.sync);
bool restored = syncIsRestoreFinish(pMnode->syncMgmt.sync);
if (pMsg->msgType == TDMT_MND_MQ_TIMER || pMsg->msgType == TDMT_MND_TELEM_TIMER ||
SSyncState state = syncGetState(pMnode->syncMgmt.sync);
if (pMsg->msgType == TDMT_MND_TMQ_TIMER || pMsg->msgType == TDMT_MND_TELEM_TIMER ||
pMsg->msgType == TDMT_MND_TRANS_TIMER || pMsg->msgType == TDMT_MND_TTL_TIMER ||
pMsg->msgType == TDMT_MND_UPTIME_TIMER) {
mTrace("timer not process since mnode restored:%d stopped:%d, sync restored:%d role:%s ", pMnode->restored,
pMnode->stopped, restored, role);
pMnode->stopped, state.restored, syncStr(state.restored));
return -1;
}
@ -626,8 +505,8 @@ static int32_t mndCheckMnodeState(SRpcMsg *pMsg) {
mDebug(
"msg:%p, type:%s failed to process since %s, mnode restored:%d stopped:%d, sync restored:%d "
"role:%s, redirect numOfEps:%d inUse:%d",
pMsg, TMSG_INFO(pMsg->msgType), terrstr(), pMnode->restored, pMnode->stopped, restored, role, epSet.numOfEps,
epSet.inUse);
pMsg, TMSG_INFO(pMsg->msgType), terrstr(), pMnode->restored, pMnode->stopped, state.restored,
syncStr(state.restored), epSet.numOfEps, epSet.inUse);
if (epSet.numOfEps > 0) {
for (int32_t i = 0; i < epSet.numOfEps; ++i) {
@ -850,8 +729,9 @@ int32_t mndGetMonitorInfo(SMnode *pMnode, SMonClusterInfo *pClusterInfo, SMonVgr
}
int32_t mndGetLoad(SMnode *pMnode, SMnodeLoad *pLoad) {
pLoad->syncState = syncGetMyRole(pMnode->syncMgmt.sync);
pLoad->syncRestore = pMnode->restored;
SSyncState state = syncGetState(pMnode->syncMgmt.sync);
pLoad->syncState = state.state;
pLoad->syncRestore = state.restored;
mTrace("mnode current syncState is %s, syncRestore:%d", syncStr(pLoad->syncState), pLoad->syncRestore);
return 0;
}

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

@ -487,6 +487,7 @@ int32_t mndScheduleStream(SMnode* pMnode, SStreamObj* pStream) {
qDestroyQueryPlan(pPlan);
return -1;
}
pTask->fillHistory = pStream->fillHistory;
mndAddTaskToTaskSet(taskOneLevel, pTask);
// source

72
source/dnode/mnode/impl/src/mndStream.c
View File

@ -557,78 +557,6 @@ static int32_t mndPersistTaskDropReq(STrans *pTrans, SStreamTask *pTask) {
return 0;
}
#if 0
static int32_t mndPersistTaskRecoverReq(STrans *pTrans, SStreamTask *pTask) {
SMStreamTaskRecoverReq *pReq = taosMemoryCalloc(1, sizeof(SMStreamTaskRecoverReq));
if (pReq == NULL) {
terrno = TSDB_CODE_OUT_OF_MEMORY;
return -1;
}
pReq->streamId = pTask->streamId;
pReq->taskId = pTask->taskId;
int32_t len;
int32_t code;
tEncodeSize(tEncodeSMStreamTaskRecoverReq, pReq, len, code);
if (code != 0) {
return -1;
}
void *buf = taosMemoryCalloc(1, sizeof(SMsgHead) + len);
if (buf == NULL) {
return -1;
}
void *abuf = POINTER_SHIFT(buf, sizeof(SMsgHead));
SEncoder encoder;
tEncoderInit(&encoder, abuf, len);
tEncodeSMStreamTaskRecoverReq(&encoder, pReq);
((SMsgHead *)buf)->vgId = pTask->nodeId;
STransAction action = {0};
memcpy(&action.epSet, &pTask->epSet, sizeof(SEpSet));
action.pCont = buf;
action.contLen = sizeof(SMsgHead) + len;
action.msgType = TDMT_STREAM_TASK_RECOVER;
if (mndTransAppendRedoAction(pTrans, &action) != 0) {
taosMemoryFree(buf);
return -1;
}
return 0;
}
int32_t mndRecoverStreamTasks(SMnode *pMnode, STrans *pTrans, SStreamObj *pStream) {
if (pStream->isDistributed) {
int32_t lv = taosArrayGetSize(pStream->tasks);
for (int32_t i = 0; i < lv; i++) {
SArray *pTasks = taosArrayGetP(pStream->tasks, i);
int32_t sz = taosArrayGetSize(pTasks);
SStreamTask *pTask = taosArrayGetP(pTasks, 0);
if (pTask->taskLevel == TASK_LEVEL__AGG) {
ASSERT(sz == 1);
if (mndPersistTaskRecoverReq(pTrans, pTask) < 0) {
return -1;
}
} else {
continue;
}
}
} else {
int32_t lv = taosArrayGetSize(pStream->tasks);
for (int32_t i = 0; i < lv; i++) {
SArray *pTasks = taosArrayGetP(pStream->tasks, i);
int32_t sz = taosArrayGetSize(pTasks);
for (int32_t j = 0; j < sz; j++) {
SStreamTask *pTask = taosArrayGetP(pTasks, j);
if (pTask->taskLevel != TASK_LEVEL__SOURCE) break;
ASSERT(pTask->taskLevel != TASK_LEVEL__SINK);
if (mndPersistTaskRecoverReq(pTrans, pTask) < 0) {
return -1;
}
}
}
}
return 0;
}
#endif
int32_t mndDropStreamTasks(SMnode *pMnode, STrans *pTrans, SStreamObj *pStream) {
int32_t lv = taosArrayGetSize(pStream->tasks);
for (int32_t i = 0; i < lv; i++) {

14
source/dnode/mnode/impl/src/mndSubscribe.c
View File

@ -73,11 +73,11 @@ int32_t mndInitSubscribe(SMnode *pMnode) {
.deleteFp = (SdbDeleteFp)mndSubActionDelete,
};
mndSetMsgHandle(pMnode, TDMT_VND_MQ_VG_CHANGE_RSP, mndTransProcessRsp);
mndSetMsgHandle(pMnode, TDMT_VND_MQ_VG_DELETE_RSP, mndTransProcessRsp);
mndSetMsgHandle(pMnode, TDMT_MND_MQ_DO_REBALANCE, mndProcessRebalanceReq);
mndSetMsgHandle(pMnode, TDMT_MND_MQ_DROP_CGROUP, mndProcessDropCgroupReq);
mndSetMsgHandle(pMnode, TDMT_MND_MQ_DROP_CGROUP_RSP, mndTransProcessRsp);
mndSetMsgHandle(pMnode, TDMT_VND_TMQ_SUBSCRIBE_RSP, mndTransProcessRsp);
mndSetMsgHandle(pMnode, TDMT_VND_TMQ_DELETE_SUB_RSP, mndTransProcessRsp);
mndSetMsgHandle(pMnode, TDMT_MND_TMQ_DO_REBALANCE, mndProcessRebalanceReq);
mndSetMsgHandle(pMnode, TDMT_MND_TMQ_DROP_CGROUP, mndProcessDropCgroupReq);
mndSetMsgHandle(pMnode, TDMT_MND_TMQ_DROP_CGROUP_RSP, mndTransProcessRsp);
mndAddShowRetrieveHandle(pMnode, TSDB_MGMT_TABLE_SUBSCRIPTIONS, mndRetrieveSubscribe);
mndAddShowFreeIterHandle(pMnode, TSDB_MGMT_TABLE_TOPICS, mndCancelGetNextSubscribe);
@ -164,7 +164,7 @@ static int32_t mndPersistSubChangeVgReq(SMnode *pMnode, STrans *pTrans, const SM
action.epSet = mndGetVgroupEpset(pMnode, pVgObj);
action.pCont = buf;
action.contLen = tlen;
action.msgType = TDMT_VND_MQ_VG_CHANGE;
action.msgType = TDMT_VND_TMQ_SUBSCRIBE;
mndReleaseVgroup(pMnode, pVgObj);
if (mndTransAppendRedoAction(pTrans, &action) != 0) {
@ -920,7 +920,7 @@ int32_t mndDropSubByTopic(SMnode *pMnode, STrans *pTrans, const char *topicName)
action.epSet = pVgEp->epSet;
action.pCont = pReq;
action.contLen = sizeof(SMqVDeleteReq);
action.msgType = TDMT_VND_MQ_VG_DELETE;
action.msgType = TDMT_VND_TMQ_DELETE_SUB;
if (mndTransAppendRedoAction(pTrans, &action) != 0) {
taosMemoryFree(pReq);
return -1;

97
source/dnode/mnode/impl/src/mndSync.c
View File

@ -72,25 +72,25 @@ static int32_t mndSyncSendMsg(const SEpSet *pEpSet, SRpcMsg *pMsg) {
return code;
}
void mndSyncCommitMsg(struct SSyncFSM *pFsm, const SRpcMsg *pMsg, SFsmCbMeta cbMeta) {
void mndSyncCommitMsg(const SSyncFSM *pFsm, const SRpcMsg *pMsg, const SFsmCbMeta *pMeta) {
SMnode *pMnode = pFsm->data;
SSyncMgmt *pMgmt = &pMnode->syncMgmt;
SSdbRaw *pRaw = pMsg->pCont;
// delete msg handle
SRpcMsg rpcMsg = {0};
syncGetAndDelRespRpc(pMnode->syncMgmt.sync, cbMeta.seqNum, &rpcMsg.info);
rpcMsg.info = pMsg->info;
int32_t transId = sdbGetIdFromRaw(pMnode->pSdb, pRaw);
pMgmt->errCode = cbMeta.code;
pMgmt->errCode = pMeta->code;
mInfo("trans:%d, is proposed, saved:%d code:0x%x, apply index:%" PRId64 " term:%" PRIu64 " config:%" PRId64
" role:%s raw:%p",
transId, pMgmt->transId, cbMeta.code, cbMeta.index, cbMeta.term, cbMeta.lastConfigIndex, syncStr(cbMeta.state),
transId, pMgmt->transId, pMeta->code, pMeta->index, pMeta->term, pMeta->lastConfigIndex, syncStr(pMeta->state),
pRaw);
if (pMgmt->errCode == 0) {
sdbWriteWithoutFree(pMnode->pSdb, pRaw);
sdbSetApplyInfo(pMnode->pSdb, cbMeta.index, cbMeta.term, cbMeta.lastConfigIndex);
sdbSetApplyInfo(pMnode->pSdb, pMeta->index, pMeta->term, pMeta->lastConfigIndex);
}
taosWLockLatch(&pMgmt->lock);
@ -120,7 +120,7 @@ void mndSyncCommitMsg(struct SSyncFSM *pFsm, const SRpcMsg *pMsg, SFsmCbMeta cbM
}
}
int32_t mndSyncGetSnapshot(struct SSyncFSM *pFsm, SSnapshot *pSnapshot, void *pReaderParam, void **ppReader) {
int32_t mndSyncGetSnapshot(const SSyncFSM *pFsm, SSnapshot *pSnapshot, void *pReaderParam, void **ppReader) {
mInfo("start to read snapshot from sdb in atomic way");
SMnode *pMnode = pFsm->data;
return sdbStartRead(pMnode->pSdb, (SSdbIter **)ppReader, &pSnapshot->lastApplyIndex, &pSnapshot->lastApplyTerm,
@ -128,13 +128,13 @@ int32_t mndSyncGetSnapshot(struct SSyncFSM *pFsm, SSnapshot *pSnapshot, void *pR
return 0;
}
int32_t mndSyncGetSnapshotInfo(struct SSyncFSM *pFsm, SSnapshot *pSnapshot) {
int32_t mndSyncGetSnapshotInfo(const SSyncFSM *pFsm, SSnapshot *pSnapshot) {
SMnode *pMnode = pFsm->data;
sdbGetCommitInfo(pMnode->pSdb, &pSnapshot->lastApplyIndex, &pSnapshot->lastApplyTerm, &pSnapshot->lastConfigIndex);
return 0;
}
void mndRestoreFinish(struct SSyncFSM *pFsm) {
void mndRestoreFinish(const SSyncFSM *pFsm) {
SMnode *pMnode = pFsm->data;
if (!pMnode->deploy) {
@ -146,32 +146,30 @@ void mndRestoreFinish(struct SSyncFSM *pFsm) {
}
}
void mndReConfig(struct SSyncFSM *pFsm, const SRpcMsg *pMsg, SReConfigCbMeta *cbMeta) {}
int32_t mndSnapshotStartRead(struct SSyncFSM *pFsm, void *pParam, void **ppReader) {
int32_t mndSnapshotStartRead(const SSyncFSM *pFsm, void *pParam, void **ppReader) {
mInfo("start to read snapshot from sdb");
SMnode *pMnode = pFsm->data;
return sdbStartRead(pMnode->pSdb, (SSdbIter **)ppReader, NULL, NULL, NULL);
}
int32_t mndSnapshotStopRead(struct SSyncFSM *pFsm, void *pReader) {
int32_t mndSnapshotStopRead(const SSyncFSM *pFsm, void *pReader) {
mInfo("stop to read snapshot from sdb");
SMnode *pMnode = pFsm->data;
return sdbStopRead(pMnode->pSdb, pReader);
}
int32_t mndSnapshotDoRead(struct SSyncFSM *pFsm, void *pReader, void **ppBuf, int32_t *len) {
int32_t mndSnapshotDoRead(const SSyncFSM *pFsm, void *pReader, void **ppBuf, int32_t *len) {
SMnode *pMnode = pFsm->data;
return sdbDoRead(pMnode->pSdb, pReader, ppBuf, len);
}
int32_t mndSnapshotStartWrite(struct SSyncFSM *pFsm, void *pParam, void **ppWriter) {
int32_t mndSnapshotStartWrite(const SSyncFSM *pFsm, void *pParam, void **ppWriter) {
mInfo("start to apply snapshot to sdb");
SMnode *pMnode = pFsm->data;
return sdbStartWrite(pMnode->pSdb, (SSdbIter **)ppWriter);
}
int32_t mndSnapshotStopWrite(struct SSyncFSM *pFsm, void *pWriter, bool isApply, SSnapshot *pSnapshot) {
int32_t mndSnapshotStopWrite(const SSyncFSM *pFsm, void *pWriter, bool isApply, SSnapshot *pSnapshot) {
mInfo("stop to apply snapshot to sdb, apply:%d, index:%" PRId64 " term:%" PRIu64 " config:%" PRId64, isApply,
pSnapshot->lastApplyIndex, pSnapshot->lastApplyTerm, pSnapshot->lastConfigIndex);
SMnode *pMnode = pFsm->data;
@ -179,18 +177,12 @@ int32_t mndSnapshotStopWrite(struct SSyncFSM *pFsm, void *pWriter, bool isApply,
pSnapshot->lastConfigIndex);
}
int32_t mndSnapshotDoWrite(struct SSyncFSM *pFsm, void *pWriter, void *pBuf, int32_t len) {
int32_t mndSnapshotDoWrite(const SSyncFSM *pFsm, void *pWriter, void *pBuf, int32_t len) {
SMnode *pMnode = pFsm->data;
return sdbDoWrite(pMnode->pSdb, pWriter, pBuf, len);
}
void mndLeaderTransfer(struct SSyncFSM *pFsm, const SRpcMsg *pMsg, SFsmCbMeta cbMeta) {
SMnode *pMnode = pFsm->data;
atomic_store_8(&(pMnode->syncMgmt.leaderTransferFinish), 1);
mInfo("vgId:1, mnode leader transfer finish");
}
static void mndBecomeFollower(struct SSyncFSM *pFsm) {
static void mndBecomeFollower(const SSyncFSM *pFsm) {
SMnode *pMnode = pFsm->data;
mInfo("vgId:1, become follower");
@ -205,7 +197,7 @@ static void mndBecomeFollower(struct SSyncFSM *pFsm) {
taosWUnLockLatch(&pMnode->syncMgmt.lock);
}
static void mndBecomeLeader(struct SSyncFSM *pFsm) {
static void mndBecomeLeader(const SSyncFSM *pFsm) {
mInfo("vgId:1, become leader");
SMnode *pMnode = pFsm->data;
}
@ -217,8 +209,8 @@ SSyncFSM *mndSyncMakeFsm(SMnode *pMnode) {
pFsm->FpPreCommitCb = NULL;
pFsm->FpRollBackCb = NULL;
pFsm->FpRestoreFinishCb = mndRestoreFinish;
pFsm->FpLeaderTransferCb = mndLeaderTransfer;
pFsm->FpReConfigCb = mndReConfig;
pFsm->FpLeaderTransferCb = NULL;
pFsm->FpReConfigCb = NULL;
pFsm->FpBecomeLeaderCb = mndBecomeLeader;
pFsm->FpBecomeFollowerCb = mndBecomeFollower;
pFsm->FpGetSnapshot = mndSyncGetSnapshot;
@ -242,10 +234,13 @@ int32_t mndInitSync(SMnode *pMnode) {
.batchSize = 1,
.vgId = 1,
.pWal = pMnode->pWal,
.msgcb = NULL,
.FpSendMsg = mndSyncSendMsg,
.FpEqMsg = mndSyncEqMsg,
.FpEqCtrlMsg = mndSyncEqCtrlMsg,
.msgcb = &pMnode->msgCb,
.syncSendMSg = mndSyncSendMsg,
.syncEqMsg = mndSyncEqMsg,
.syncEqCtrlMsg = mndSyncEqCtrlMsg,
.pingMs = 5000,
.electMs = 3000,
.heartbeatMs = 500,
};
snprintf(syncInfo.path, sizeof(syncInfo.path), "%s%ssync", pMnode->path, TD_DIRSEP);
@ -269,11 +264,6 @@ int32_t mndInitSync(SMnode *pMnode) {
return -1;
}
// decrease election timer
setPingTimerMS(pMgmt->sync, 5000);
setElectTimerMS(pMgmt->sync, 3000);
setHeartbeatTimerMS(pMgmt->sync, 500);
mInfo("mnode-sync is opened, id:%" PRId64, pMgmt->sync);
return 0;
}
@ -289,32 +279,26 @@ void mndCleanupSync(SMnode *pMnode) {
int32_t mndSyncPropose(SMnode *pMnode, SSdbRaw *pRaw, int32_t transId) {
SSyncMgmt *pMgmt = &pMnode->syncMgmt;
SRpcMsg req = {.msgType = TDMT_MND_APPLY_MSG, .contLen = sdbGetRawTotalSize(pRaw)};
if (req.contLen <= 0) {
terrno = TSDB_CODE_APP_ERROR;
return -1;
}
pMgmt->errCode = 0;
SRpcMsg req = {.msgType = TDMT_MND_APPLY_MSG, .contLen = sdbGetRawTotalSize(pRaw)};
req.pCont = rpcMallocCont(req.contLen);
if (req.pCont == NULL) return -1;
memcpy(req.pCont, pRaw, req.contLen);
pMgmt->errCode = 0;
taosWLockLatch(&pMgmt->lock);
if (pMgmt->transId != 0) {
mError("trans:%d, can't be proposed since trans:%d alrady waiting for confirm", transId, pMgmt->transId);
mError("trans:%d, can't be proposed since trans:%d already waiting for confirm", transId, pMgmt->transId);
taosWUnLockLatch(&pMgmt->lock);
terrno = TSDB_CODE_APP_NOT_READY;
return -1;
} else {
pMgmt->transId = transId;
mInfo("trans:%d, will be proposed", pMgmt->transId);
taosWUnLockLatch(&pMgmt->lock);
}
const bool isWeak = false;
int32_t code = syncPropose(pMgmt->sync, &req, isWeak);
mInfo("trans:%d, will be proposed", transId);
pMgmt->transId = transId;
taosWUnLockLatch(&pMgmt->lock);
int32_t code = syncPropose(pMgmt->sync, &req, false);
if (code == 0) {
mInfo("trans:%d, is proposing and wait sem", pMgmt->transId);
tsem_wait(&pMgmt->syncSem);
@ -327,8 +311,8 @@ int32_t mndSyncPropose(SMnode *pMnode, SSdbRaw *pRaw, int32_t transId) {
sdbSetApplyInfo(pMnode->pSdb, req.info.conn.applyIndex, req.info.conn.applyTerm, SYNC_INDEX_INVALID);
code = 0;
} else {
taosWLockLatch(&pMgmt->lock);
mInfo("trans:%d, failed to proposed since %s", transId, terrstr());
taosWLockLatch(&pMgmt->lock);
pMgmt->transId = 0;
taosWUnLockLatch(&pMgmt->lock);
if (terrno == TSDB_CODE_SYN_NOT_LEADER) {
@ -344,13 +328,12 @@ int32_t mndSyncPropose(SMnode *pMnode, SSdbRaw *pRaw, int32_t transId) {
return code;
}
if (pMgmt->errCode != 0) terrno = pMgmt->errCode;
terrno = pMgmt->errCode;
return pMgmt->errCode;
}
void mndSyncStart(SMnode *pMnode) {
SSyncMgmt *pMgmt = &pMnode->syncMgmt;
syncSetMsgCb(pMgmt->sync, &pMnode->msgCb);
syncStart(pMgmt->sync);
mInfo("vgId:1, sync started, id:%" PRId64, pMgmt->sync);
}
@ -366,11 +349,15 @@ void mndSyncStop(SMnode *pMnode) {
}
bool mndIsLeader(SMnode *pMnode) {
SSyncMgmt *pMgmt = &pMnode->syncMgmt;
SSyncState state = syncGetState(pMnode->syncMgmt.sync);
if (!syncIsReady(pMgmt->sync)) {
// get terrno from syncIsReady
// terrno = TSDB_CODE_SYN_NOT_LEADER;
if (state.state != TAOS_SYNC_STATE_LEADER || !state.restored) {
if (state.state != TAOS_SYNC_STATE_LEADER) {
terrno = TSDB_CODE_SYN_NOT_LEADER;
} else {
terrno = TSDB_CODE_APP_NOT_READY;
}
mDebug("vgId:1, mnode not ready, state:%s, restore:%d", syncStr(state.state), state.restored);
return false;
}

13
source/dnode/mnode/impl/src/mndTopic.c
View File

@ -53,11 +53,10 @@ int32_t mndInitTopic(SMnode *pMnode) {
.deleteFp = (SdbDeleteFp)mndTopicActionDelete,
};
mndSetMsgHandle(pMnode, TDMT_MND_CREATE_TOPIC, mndProcessCreateTopicReq);
mndSetMsgHandle(pMnode, TDMT_MND_DROP_TOPIC, mndProcessDropTopicReq);
mndSetMsgHandle(pMnode, TDMT_VND_DROP_TOPIC_RSP, mndTransProcessRsp);
mndSetMsgHandle(pMnode, TDMT_VND_ADD_CHECK_INFO_RSP, mndTransProcessRsp);
mndSetMsgHandle(pMnode, TDMT_VND_DELETE_CHECK_INFO_RSP, mndTransProcessRsp);
mndSetMsgHandle(pMnode, TDMT_MND_TMQ_CREATE_TOPIC, mndProcessCreateTopicReq);
mndSetMsgHandle(pMnode, TDMT_MND_TMQ_DROP_TOPIC, mndProcessDropTopicReq);
mndSetMsgHandle(pMnode, TDMT_VND_TMQ_ADD_CHECKINFO_RSP, mndTransProcessRsp);
mndSetMsgHandle(pMnode, TDMT_VND_TMQ_DEL_CHECKINFO_RSP, mndTransProcessRsp);
mndAddShowRetrieveHandle(pMnode, TSDB_MGMT_TABLE_TOPICS, mndRetrieveTopic);
mndAddShowFreeIterHandle(pMnode, TSDB_MGMT_TABLE_TOPICS, mndCancelGetNextTopic);
@ -506,7 +505,7 @@ static int32_t mndCreateTopic(SMnode *pMnode, SRpcMsg *pReq, SCMCreateTopicReq *
action.epSet = mndGetVgroupEpset(pMnode, pVgroup);
action.pCont = buf;
action.contLen = sizeof(SMsgHead) + len;
action.msgType = TDMT_VND_ADD_CHECK_INFO;
action.msgType = TDMT_VND_TMQ_ADD_CHECKINFO;
if (mndTransAppendRedoAction(pTrans, &action) != 0) {
taosMemoryFree(buf);
sdbRelease(pSdb, pVgroup);
@ -715,7 +714,7 @@ static int32_t mndProcessDropTopicReq(SRpcMsg *pReq) {
action.epSet = mndGetVgroupEpset(pMnode, pVgroup);
action.pCont = buf;
action.contLen = sizeof(SMsgHead) + TSDB_TOPIC_FNAME_LEN;
action.msgType = TDMT_VND_DELETE_CHECK_INFO;
action.msgType = TDMT_VND_TMQ_DEL_CHECKINFO;
if (mndTransAppendRedoAction(pTrans, &action) != 0) {
taosMemoryFree(buf);
sdbRelease(pSdb, pVgroup);

2
source/dnode/qnode/src/qnode.c
View File

@ -95,7 +95,7 @@ int32_t qndProcessQueryMsg(SQnode *pQnode, int64_t ts, SRpcMsg *pMsg) {
case TDMT_SCH_DROP_TASK:
code = qWorkerProcessDropMsg(pQnode, pQnode->pQuery, pMsg, ts);
break;
case TDMT_VND_CONSUME:
case TDMT_VND_TMQ_CONSUME:
// code = tqProcessConsumeReq(pQnode->pTq, pMsg);
// break;
case TDMT_SCH_QUERY_HEARTBEAT:

38
source/dnode/snode/src/snode.c
View File

@ -261,18 +261,52 @@ int32_t sndProcessWriteMsg(SSnode *pSnode, SRpcMsg *pMsg, SRpcMsg *pRsp) {
return 0;
}
int32_t sndProcessTaskRecoverFinishReq(SSnode *pSnode, SRpcMsg *pMsg) {
char *msg = POINTER_SHIFT(pMsg->pCont, sizeof(SMsgHead));
int32_t msgLen = pMsg->contLen - sizeof(SMsgHead);
// deserialize
SStreamRecoverFinishReq req;
SDecoder decoder;
tDecoderInit(&decoder, msg, msgLen);
tDecodeSStreamRecoverFinishReq(&decoder, &req);
tDecoderClear(&decoder);
// find task
SStreamTask *pTask = streamMetaGetTask(pSnode->pMeta, req.taskId);
if (pTask == NULL) {
return -1;
}
// do process request
if (streamProcessRecoverFinishReq(pTask, req.childId) < 0) {
return -1;
}
return 0;
}
int32_t sndProcessTaskRecoverFinishRsp(SSnode *pSnode, SRpcMsg *pMsg) {
//
return 0;
}
int32_t sndProcessStreamMsg(SSnode *pSnode, SRpcMsg *pMsg) {
switch (pMsg->msgType) {
case TDMT_STREAM_TASK_RUN:
return sndProcessTaskRunReq(pSnode, pMsg);
case TDMT_STREAM_TASK_DISPATCH:
return sndProcessTaskDispatchReq(pSnode, pMsg, true);
case TDMT_STREAM_RETRIEVE:
return sndProcessTaskRetrieveReq(pSnode, pMsg);
case TDMT_STREAM_TASK_DISPATCH_RSP:
return sndProcessTaskDispatchRsp(pSnode, pMsg);
case TDMT_STREAM_RETRIEVE:
return sndProcessTaskRetrieveReq(pSnode, pMsg);
case TDMT_STREAM_RETRIEVE_RSP:
return sndProcessTaskRetrieveRsp(pSnode, pMsg);
case TDMT_STREAM_RECOVER_FINISH:
return sndProcessTaskRecoverFinishReq(pSnode, pMsg);
case TDMT_STREAM_RECOVER_FINISH_RSP:
return sndProcessTaskRecoverFinishRsp(pSnode, pMsg);
default:
ASSERT(0);
}

3
source/dnode/vnode/inc/vnode.h
View File

@ -83,7 +83,6 @@ int32_t vnodePreprocessQueryMsg(SVnode *pVnode, SRpcMsg *pMsg);
int32_t vnodeProcessWriteMsg(SVnode *pVnode, SRpcMsg *pMsg, int64_t version, SRpcMsg *pRsp);
int32_t vnodeProcessSyncMsg(SVnode *pVnode, SRpcMsg *pMsg, SRpcMsg **pRsp);
int32_t vnodeProcessSyncCtrlMsg(SVnode *pVnode, SRpcMsg *pMsg, SRpcMsg **pRsp);
int32_t vnodeProcessQueryMsg(SVnode *pVnode, SRpcMsg *pMsg);
int32_t vnodeProcessFetchMsg(SVnode *pVnode, SRpcMsg *pMsg, SQueueInfo *pInfo);
void vnodeProposeWriteMsg(SQueueInfo *pInfo, STaosQall *qall, int32_t numOfMsgs);
@ -159,7 +158,7 @@ int32_t tsdbReaderOpen(SVnode *pVnode, SQueryTableDataCond *pCond, void *pTableL
void tsdbReaderClose(STsdbReader *pReader);
bool tsdbNextDataBlock(STsdbReader *pReader);
bool tsdbTableNextDataBlock(STsdbReader *pReader, uint64_t uid);
void tsdbRetrieveDataBlockInfo(STsdbReader *pReader, SDataBlockInfo *pDataBlockInfo);
void tsdbRetrieveDataBlockInfo(const STsdbReader* pReader, int32_t* rows, uint64_t* uid, STimeWindow* pWindow);
int32_t tsdbRetrieveDatablockSMA(STsdbReader *pReader, SColumnDataAgg ***pBlockStatis, bool *allHave);
SArray *tsdbRetrieveDataBlock(STsdbReader *pTsdbReadHandle, SArray *pColumnIdList);
int32_t tsdbReaderReset(STsdbReader *pReader, SQueryTableDataCond *pCond);

1
source/dnode/vnode/src/inc/vnd.h
View File

@ -17,7 +17,6 @@
#define _TD_VND_H_
#include "sync.h"
#include "syncTools.h"
#include "ttrace.h"
#include "vnodeInt.h"

12
source/dnode/vnode/src/inc/vnodeInt.h
View File

@ -176,8 +176,8 @@ int32_t tqCheckColModifiable(STQ* pTq, int64_t tbUid, int32_t colId);
// tq-mq
int32_t tqProcessAddCheckInfoReq(STQ* pTq, int64_t version, char* msg, int32_t msgLen);
int32_t tqProcessDelCheckInfoReq(STQ* pTq, int64_t version, char* msg, int32_t msgLen);
int32_t tqProcessVgChangeReq(STQ* pTq, int64_t version, char* msg, int32_t msgLen);
int32_t tqProcessVgDeleteReq(STQ* pTq, int64_t version, char* msg, int32_t msgLen);
int32_t tqProcessSubscribeReq(STQ* pTq, int64_t version, char* msg, int32_t msgLen);
int32_t tqProcessDeleteSubReq(STQ* pTq, int64_t version, char* msg, int32_t msgLen);
int32_t tqProcessOffsetCommitReq(STQ* pTq, int64_t version, char* msg, int32_t msgLen);
int32_t tqProcessPollReq(STQ* pTq, SRpcMsg* pMsg);
// tq-stream
@ -187,11 +187,15 @@ int32_t tqProcessSubmitReq(STQ* pTq, SSubmitReq* data, int64_t ver);
int32_t tqProcessDelReq(STQ* pTq, void* pReq, int32_t len, int64_t ver);
int32_t tqProcessTaskRunReq(STQ* pTq, SRpcMsg* pMsg);
int32_t tqProcessTaskDispatchReq(STQ* pTq, SRpcMsg* pMsg, bool exec);
int32_t tqProcessTaskRecoverReq(STQ* pTq, SRpcMsg* pMsg);
int32_t tqProcessTaskDispatchRsp(STQ* pTq, SRpcMsg* pMsg);
int32_t tqProcessTaskRecoverRsp(STQ* pTq, SRpcMsg* pMsg);
// int32_t tqProcessTaskRecoverReq(STQ* pTq, SRpcMsg* pMsg);
// int32_t tqProcessTaskRecoverRsp(STQ* pTq, SRpcMsg* pMsg);
int32_t tqProcessTaskRetrieveReq(STQ* pTq, SRpcMsg* pMsg);
int32_t tqProcessTaskRetrieveRsp(STQ* pTq, SRpcMsg* pMsg);
int32_t tqProcessTaskRecover1Req(STQ* pTq, SRpcMsg* pMsg);
int32_t tqProcessTaskRecover2Req(STQ* pTq, int64_t version, char* msg, int32_t msgLen);
int32_t tqProcessTaskRecoverFinishReq(STQ* pTq, SRpcMsg* pMsg);
int32_t tqProcessTaskRecoverFinishRsp(STQ* pTq, SRpcMsg* pMsg);
SSubmitReq* tqBlockToSubmit(SVnode* pVnode, const SArray* pBlocks, const STSchema* pSchema,
SSchemaWrapper* pTagSchemaWrapper, bool createTb, int64_t suid, const char* stbFullName,

94
source/dnode/vnode/src/tq/tq.c
View File

@ -709,7 +709,7 @@ int32_t tqProcessPollReq(STQ* pTq, SRpcMsg* pMsg) {
return 0;
}
int32_t tqProcessVgDeleteReq(STQ* pTq, int64_t version, char* msg, int32_t msgLen) {
int32_t tqProcessDeleteSubReq(STQ* pTq, int64_t version, char* msg, int32_t msgLen) {
SMqVDeleteReq* pReq = (SMqVDeleteReq*)msg;
taosWLockLatch(&pTq->pushLock);
@ -767,7 +767,7 @@ int32_t tqProcessDelCheckInfoReq(STQ* pTq, int64_t version, char* msg, int32_t m
return 0;
}
int32_t tqProcessVgChangeReq(STQ* pTq, int64_t version, char* msg, int32_t msgLen) {
int32_t tqProcessSubscribeReq(STQ* pTq, int64_t version, char* msg, int32_t msgLen) {
SMqRebVgReq req = {0};
tDecodeSMqRebVgReq(msg, &req);
// todo lock
@ -982,25 +982,33 @@ int32_t tqProcessTaskDeployReq(STQ* pTq, int64_t version, char* msg, int32_t msg
// 3.go through recover steps to fill history
if (pTask->fillHistory) {
streamSetParamForRecover(pTask);
if (pTask->taskLevel == TASK_LEVEL__SOURCE) {
streamSetParamForRecover(pTask);
streamSourceRecoverPrepareStep1(pTask, version);
SStreamRecoverStep1Req req;
streamBuildSourceRecover1Req(pTask, &req);
void* serialziedReq = (void*)&req;
int32_t len = sizeof(SStreamRecoverStep1Req);
void* serializedReq = rpcMallocCont(len);
if (serializedReq == NULL) {
return -1;
}
memcpy(serializedReq, &req, len);
SRpcMsg rpcMsg = {
.contLen = len,
.pCont = serialziedReq,
.pCont = serializedReq,
.msgType = TDMT_VND_STREAM_RECOVER_STEP1,
};
tmsgPutToQueue(&pTq->pVnode->msgCb, STREAM_QUEUE, &rpcMsg);
if (tmsgPutToQueue(&pTq->pVnode->msgCb, STREAM_QUEUE, &rpcMsg) < 0) {
/*ASSERT(0);*/
}
} else if (pTask->taskLevel == TASK_LEVEL__AGG) {
streamSetParamForRecover(pTask);
streamAggRecoverPrepare(pTask);
} else if (pTask->taskLevel == TASK_LEVEL__SINK) {
// do nothing
@ -1010,8 +1018,11 @@ int32_t tqProcessTaskDeployReq(STQ* pTq, int64_t version, char* msg, int32_t msg
return 0;
}
int32_t tqProcessTaskRecover1Req(STQ* pTq, char* msg, int32_t msgLen) {
int32_t tqProcessTaskRecover1Req(STQ* pTq, SRpcMsg* pMsg) {
int32_t code;
char* msg = pMsg->pCont;
int32_t msgLen = pMsg->contLen;
SStreamRecoverStep1Req* pReq = (SStreamRecoverStep1Req*)msg;
SStreamTask* pTask = streamMetaGetTask(pTq->pStreamMeta, pReq->taskId);
if (pTask == NULL) {
@ -1035,16 +1046,24 @@ int32_t tqProcessTaskRecover1Req(STQ* pTq, char* msg, int32_t msgLen) {
return -1;
}
ASSERT(pReq->taskId == pTask->taskId);
// serialize msg
int32_t len = sizeof(SStreamRecoverStep2Req);
void* serializedReq = (void*)&req;
int32_t len = sizeof(SStreamRecoverStep1Req);
void* serializedReq = rpcMallocCont(len);
if (serializedReq == NULL) {
return -1;
}
memcpy(serializedReq, &req, len);
// dispatch msg
SRpcMsg rpcMsg = {
.code = 0,
.contLen = len,
.msgType = TDMT_VND_STREAM_RECOVER_STEP2,
.pCont = (void*)serializedReq,
.pCont = serializedReq,
};
tmsgPutToQueue(&pTq->pVnode->msgCb, WRITE_QUEUE, &rpcMsg);
@ -1087,15 +1106,15 @@ int32_t tqProcessTaskRecover2Req(STQ* pTq, int64_t version, char* msg, int32_t m
return 0;
}
int32_t tqProcessTaskRecoverFinishReq(STQ* pTq, char* msg, int32_t msgLen) {
int32_t code;
int32_t tqProcessTaskRecoverFinishReq(STQ* pTq, SRpcMsg* pMsg) {
char* msg = POINTER_SHIFT(pMsg->pCont, sizeof(SMsgHead));
int32_t msgLen = pMsg->contLen - sizeof(SMsgHead);
// deserialize
int32_t len;
SStreamRecoverFinishReq req;
SDecoder decoder;
tDecoderInit(&decoder, msg, sizeof(SStreamRecoverFinishReq));
tDecoderInit(&decoder, msg, msgLen);
tDecodeSStreamRecoverFinishReq(&decoder, &req);
tDecoderClear(&decoder);
@ -1112,6 +1131,11 @@ int32_t tqProcessTaskRecoverFinishReq(STQ* pTq, char* msg, int32_t msgLen) {
return 0;
}
int32_t tqProcessTaskRecoverFinishRsp(STQ* pTq, SRpcMsg* pMsg) {
//
return 0;
}
int32_t tqProcessDelReq(STQ* pTq, void* pReq, int32_t len, int64_t ver) {
bool failed = false;
SDecoder* pCoder = &(SDecoder){0};
@ -1235,7 +1259,7 @@ int32_t tqProcessSubmitReq(STQ* pTq, SSubmitReq* pReq, int64_t ver) {
pSubmit = streamDataSubmitNew(pReq);
if (pSubmit == NULL) {
terrno = TSDB_CODE_OUT_OF_MEMORY;
qError("failed to create data submit for stream since out of memory");
tqError("failed to create data submit for stream since out of memory");
failed = true;
}
@ -1244,18 +1268,21 @@ int32_t tqProcessSubmitReq(STQ* pTq, SSubmitReq* pReq, int64_t ver) {
if (pIter == NULL) break;
SStreamTask* pTask = *(SStreamTask**)pIter;
if (pTask->taskLevel != TASK_LEVEL__SOURCE) continue;
if (pTask->taskStatus == TASK_STATUS__RECOVER_PREPARE || pTask->taskStatus == TASK_STATUS__RECOVER1) continue;
if (pTask->taskStatus == TASK_STATUS__RECOVER_PREPARE) {
tqDebug("skip push task %d, task status %d", pTask->taskId, pTask->taskStatus);
continue;
}
qDebug("data submit enqueue stream task: %d, ver: %" PRId64, pTask->taskId, ver);
tqDebug("data submit enqueue stream task: %d, ver: %" PRId64, pTask->taskId, ver);
if (!failed) {
if (streamTaskInput(pTask, (SStreamQueueItem*)pSubmit) < 0) {
qError("stream task input failed, task id %d", pTask->taskId);
tqError("stream task input failed, task id %d", pTask->taskId);
continue;
}
if (streamSchedExec(pTask) < 0) {
qError("stream task launch failed, task id %d", pTask->taskId);
tqError("stream task launch failed, task id %d", pTask->taskId);
continue;
}
} else {
@ -1306,33 +1333,6 @@ int32_t tqProcessTaskDispatchReq(STQ* pTq, SRpcMsg* pMsg, bool exec) {
}
}
#if 0
int32_t tqProcessTaskRecoverReq(STQ* pTq, SRpcMsg* pMsg) {
SStreamTaskRecoverReq* pReq = pMsg->pCont;
int32_t taskId = pReq->taskId;
SStreamTask* pTask = streamMetaGetTask(pTq->pStreamMeta, taskId);
if (pTask) {
streamProcessRecoverReq(pTask, pReq, pMsg);
return 0;
} else {
return -1;
}
}
int32_t tqProcessTaskRecoverRsp(STQ* pTq, SRpcMsg* pMsg) {
SStreamTaskRecoverRsp* pRsp = pMsg->pCont;
int32_t taskId = pRsp->rspTaskId;
SStreamTask* pTask = streamMetaGetTask(pTq->pStreamMeta, taskId);
if (pTask) {
streamProcessRecoverRsp(pTask, pRsp);
return 0;
} else {
return -1;
}
}
#endif
int32_t tqProcessTaskDispatchRsp(STQ* pTq, SRpcMsg* pMsg) {
SStreamDispatchRsp* pRsp = POINTER_SHIFT(pMsg->pCont, sizeof(SMsgHead));
int32_t taskId = pRsp->taskId;

41
source/dnode/vnode/src/tq/tqRead.c
View File

@ -199,6 +199,9 @@ int64_t tqFetchLog(STQ* pTq, STqHandle* pHandle, int64_t* fetchOffset, SWalCkHea
goto END;
}
tqDebug("vgId:%d, taosx get msg ver %" PRId64 ", type: %s", pTq->pVnode->config.vgId, offset,
TMSG_INFO((*ppCkHead)->head.msgType));
if ((*ppCkHead)->head.msgType == TDMT_VND_SUBMIT) {
code = walFetchBody(pHandle->pWalReader, ppCkHead);
@ -216,17 +219,20 @@ int64_t tqFetchLog(STQ* pTq, STqHandle* pHandle, int64_t* fetchOffset, SWalCkHea
SWalCont* pHead = &((*ppCkHead)->head);
if (IS_META_MSG(pHead->msgType)) {
code = walFetchBody(pHandle->pWalReader, ppCkHead);
if (code < 0) {
ASSERT(0);
*fetchOffset = offset;
code = -1;
goto END;
}
if (isValValidForTable(pHandle, pHead)) {
*fetchOffset = offset;
code = 0;
goto END;
} else {
offset++;
continue;
}
}
}
@ -586,8 +592,39 @@ int32_t tqUpdateTbUidList(STQ* pTq, const SArray* tbUidList, bool isAdd) {
taosHashPut(pExec->execHandle.execDb.pFilterOutTbUid, &tbUid, sizeof(int64_t), NULL, 0);
}
}
} else if (pExec->execHandle.subType == TOPIC_SUB_TYPE__TABLE) {
if (isAdd) {
SArray* qa = taosArrayInit(4, sizeof(tb_uid_t));
SMetaReader mr = {0};
metaReaderInit(&mr, pTq->pVnode->pMeta, 0);
for (int32_t i = 0; i < taosArrayGetSize(tbUidList); ++i) {
uint64_t* id = (uint64_t*)taosArrayGet(tbUidList, i);
int32_t code = metaGetTableEntryByUid(&mr, *id);
if (code != TSDB_CODE_SUCCESS) {
qError("failed to get table meta, uid:%" PRIu64 " code:%s", *id, tstrerror(terrno));
continue;
}
tDecoderClear(&mr.coder);
if (mr.me.type != TSDB_CHILD_TABLE || mr.me.ctbEntry.suid != pExec->execHandle.execTb.suid) {
tqDebug("table uid %" PRId64 " does not add to tq handle", *id);
continue;
}
tqDebug("table uid %" PRId64 " add to tq handle", *id);
taosArrayPush(qa, id);
}
metaReaderClear(&mr);
if (taosArrayGetSize(qa) > 0) {
tqReaderAddTbUidList(pExec->execHandle.pExecReader, qa);
}
taosArrayDestroy(qa);
} else {
// tq update id
// TODO handle delete table from stb
}
} else {
ASSERT(0);
}
}
while (1) {

4
source/dnode/vnode/src/tsdb/tsdbCache.c
View File

@ -684,7 +684,7 @@ static int32_t getNextRowFromFS(void *iter, TSDBROW **ppRow) {
if (*state->pDataFReader != NULL) {
tsdbDataFReaderClose(state->pDataFReader);
resetLastBlockLoadInfo(state->pLoadInfo);
// resetLastBlockLoadInfo(state->pLoadInfo);
}
code = tsdbDataFReaderOpen(state->pDataFReader, state->pTsdb, pFileSet);
@ -764,7 +764,7 @@ static int32_t getNextRowFromFS(void *iter, TSDBROW **ppRow) {
if (--state->iBlock < 0) {
tsdbDataFReaderClose(state->pDataFReader);
*state->pDataFReader = NULL;
resetLastBlockLoadInfo(state->pLoadInfo);
// resetLastBlockLoadInfo(state->pLoadInfo);
if (state->aBlockIdx) {
taosArrayDestroy(state->aBlockIdx);

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

@ -48,7 +48,7 @@ int tsdbOpen(SVnode *pVnode, STsdb **ppTsdb, const char *dir, STsdbKeepCfg *pKee
}
pTsdb->path = (char *)&pTsdb[1];
sprintf(pTsdb->path, "%s%s%s", pVnode->path, TD_DIRSEP, dir);
snprintf(pTsdb->path, TD_PATH_MAX, "%s%s%s", pVnode->path, TD_DIRSEP, dir);
taosRealPath(pTsdb->path, NULL, slen);
pTsdb->pVnode = pVnode;
taosThreadRwlockInit(&pTsdb->rwLock, NULL);

310
source/dnode/vnode/src/tsdb/tsdbRead.c
View File

@ -35,11 +35,17 @@ typedef struct {
int32_t numOfLastFiles;
} SBlockNumber;
typedef struct SBlockIndex {
int32_t ordinalIndex;
int64_t inFileOffset;
STimeWindow window;
} SBlockIndex;
typedef struct STableBlockScanInfo {
uint64_t uid;
TSKEY lastKey;
SMapData mapData; // block info (compressed)
SArray* pBlockList; // block data index list
SArray* pBlockList; // block data index list, SArray<SBlockIndex>
SIterInfo iter; // mem buffer skip list iterator
SIterInfo iiter; // imem buffer skip list iterator
SArray* delSkyline; // delete info for this table
@ -72,6 +78,7 @@ typedef struct SIOCostSummary {
double lastBlockLoadTime;
int64_t composedBlocks;
double buildComposedBlockTime;
double createScanInfoList;
} SIOCostSummary;
typedef struct SBlockLoadSuppInfo {
@ -131,7 +138,7 @@ typedef struct SReaderStatus {
bool loadFromFile; // check file stage
bool composedDataBlock; // the returned data block is a composed block or not
SHashObj* pTableMap; // SHash<STableBlockScanInfo>
STableBlockScanInfo* pTableIter; // table iterator used in building in-memory buffer data blocks.
STableBlockScanInfo** pTableIter; // table iterator used in building in-memory buffer data blocks.
SUidOrderCheckInfo uidCheckInfo; // check all table in uid order
SFileBlockDumpInfo fBlockDumpInfo;
SDFileSet* pCurrentFileset; // current opened file set
@ -140,6 +147,12 @@ typedef struct SReaderStatus {
SDataBlockIter blockIter;
} SReaderStatus;
typedef struct SBlockInfoBuf {
int32_t currentIndex;
SArray* pData;
int32_t numPerBucket;
} SBlockInfoBuf;
struct STsdbReader {
STsdb* pTsdb;
uint64_t suid;
@ -157,7 +170,7 @@ struct STsdbReader {
STSchema* pMemSchema; // the previous schema for in-memory data, to avoid load schema too many times
SDataFReader* pFileReader;
SVersionRange verRange;
SBlockInfoBuf blockInfoBuf;
int32_t step;
STsdbReader* innerReader[2];
};
@ -225,6 +238,50 @@ static int32_t setColumnIdSlotList(STsdbReader* pReader, SSDataBlock* pBlock) {
return TSDB_CODE_SUCCESS;
}
static int32_t initBlockScanInfoBuf(SBlockInfoBuf* pBuf, int32_t numOfTables) {
int32_t num = numOfTables / pBuf->numPerBucket;
int32_t remainder = numOfTables % pBuf->numPerBucket;
if (pBuf->pData == NULL) {
pBuf->pData = taosArrayInit(num + 1, POINTER_BYTES);
}
for(int32_t i = 0; i < num; ++i) {
char* p = taosMemoryCalloc(pBuf->numPerBucket, sizeof(STableBlockScanInfo));
if (p == NULL) {
return TSDB_CODE_OUT_OF_MEMORY;
}
taosArrayPush(pBuf->pData, &p);
}
if (remainder > 0) {
char* p = taosMemoryCalloc(remainder, sizeof(STableBlockScanInfo));
if (p == NULL) {
return TSDB_CODE_OUT_OF_MEMORY;
}
taosArrayPush(pBuf->pData, &p);
}
return TSDB_CODE_SUCCESS;
}
static void clearBlockScanInfoBuf(SBlockInfoBuf* pBuf) {
size_t num = taosArrayGetSize(pBuf->pData);
for(int32_t i = 0; i < num; ++i) {
char** p = taosArrayGet(pBuf->pData, i);
taosMemoryFree(*p);
}
taosArrayDestroy(pBuf->pData);
}
static void* getPosInBlockInfoBuf(SBlockInfoBuf* pBuf, int32_t index) {
int32_t bucketIndex = index / pBuf->numPerBucket;
char** pBucket = taosArrayGet(pBuf->pData, bucketIndex);
return (*pBucket) + (index % pBuf->numPerBucket) * sizeof(STableBlockScanInfo);
}
// NOTE: speedup the whole processing by preparing the buffer for STableBlockScanInfo in batch model
static SHashObj* createDataBlockScanInfo(STsdbReader* pTsdbReader, const STableKeyInfo* idList, int32_t numOfTables) {
// allocate buffer in order to load data blocks from file
// todo use simple hash instead, optimize the memory consumption
@ -234,8 +291,24 @@ static SHashObj* createDataBlockScanInfo(STsdbReader* pTsdbReader, const STableK
return NULL;
}
int64_t st = taosGetTimestampUs();
initBlockScanInfoBuf(&pTsdbReader->blockInfoBuf, numOfTables);
for (int32_t j = 0; j < numOfTables; ++j) {
STableBlockScanInfo info = {.lastKey = 0, .uid = idList[j].uid};
STableBlockScanInfo* pScanInfo = getPosInBlockInfoBuf(&pTsdbReader->blockInfoBuf, j);
pScanInfo->uid = idList[j].uid;
if (ASCENDING_TRAVERSE(pTsdbReader->order)) {
int64_t skey = pTsdbReader->window.skey;
pScanInfo->lastKey = (skey > INT64_MIN) ? (skey - 1) : skey;
} else {
int64_t ekey = pTsdbReader->window.ekey;
pScanInfo->lastKey = (ekey < INT64_MAX) ? (ekey + 1) : ekey;
}
taosHashPut(pTableMap, &pScanInfo->uid, sizeof(uint64_t), &pScanInfo, POINTER_BYTES);
#if 0
// STableBlockScanInfo info = {.lastKey = 0, .uid = idList[j].uid};
if (ASCENDING_TRAVERSE(pTsdbReader->order)) {
int64_t skey = pTsdbReader->window.skey;
info.lastKey = (skey > INT64_MIN) ? (skey - 1) : skey;
@ -245,28 +318,33 @@ static SHashObj* createDataBlockScanInfo(STsdbReader* pTsdbReader, const STableK
}
taosHashPut(pTableMap, &info.uid, sizeof(uint64_t), &info, sizeof(info));
tsdbDebug("%p check table uid:%" PRId64 " from lastKey:%" PRId64 " %s", pTsdbReader, info.uid, info.lastKey,
#endif
tsdbTrace("%p check table uid:%" PRId64 " from lastKey:%" PRId64 " %s", pTsdbReader, pScanInfo->uid, pScanInfo->lastKey,
pTsdbReader->idStr);
}
tsdbDebug("%p create %d tables scan-info, size:%.2f Kb, %s", pTsdbReader, numOfTables,
(sizeof(STableBlockScanInfo) * numOfTables) / 1024.0, pTsdbReader->idStr);
pTsdbReader->cost.createScanInfoList = (taosGetTimestampUs() - st) / 1000.0;
tsdbDebug("%p create %d tables scan-info, size:%.2f Kb, elapsed time:%.2f ms, %s", pTsdbReader, numOfTables,
(sizeof(STableBlockScanInfo) * numOfTables) / 1024.0, pTsdbReader->cost.createScanInfoList,
pTsdbReader->idStr);
return pTableMap;
}
static void resetDataBlockScanInfo(SHashObj* pTableMap, int64_t ts) {
STableBlockScanInfo* p = NULL;
static void resetAllDataBlockScanInfo(SHashObj* pTableMap, int64_t ts) {
STableBlockScanInfo** p = NULL;
while ((p = taosHashIterate(pTableMap, p)) != NULL) {
p->iterInit = false;
p->iiter.hasVal = false;
if (p->iter.iter != NULL) {
p->iter.iter = tsdbTbDataIterDestroy(p->iter.iter);
STableBlockScanInfo* pInfo = *(STableBlockScanInfo**) p;
pInfo->iterInit = false;
pInfo->iiter.hasVal = false;
if (pInfo->iter.iter != NULL) {
pInfo->iter.iter = tsdbTbDataIterDestroy(pInfo->iter.iter);
}
p->delSkyline = taosArrayDestroy(p->delSkyline);
p->lastKey = ts;
pInfo->delSkyline = taosArrayDestroy(pInfo->delSkyline);
pInfo->lastKey = ts;
}
}
@ -287,10 +365,10 @@ static void clearBlockScanInfo(STableBlockScanInfo* p) {
tMapDataClear(&p->mapData);
}
static void destroyBlockScanInfo(SHashObj* pTableMap) {
STableBlockScanInfo* p = NULL;
static void destroyAllBlockScanInfo(SHashObj* pTableMap) {
void* p = NULL;
while ((p = taosHashIterate(pTableMap, p)) != NULL) {
clearBlockScanInfo(p);
clearBlockScanInfo(*(STableBlockScanInfo**)p);
}
taosHashCleanup(pTableMap);
@ -495,7 +573,7 @@ static int32_t tsdbReaderCreate(SVnode* pVnode, SQueryTableDataCond* pCond, STsd
pReader->verRange = getQueryVerRange(pVnode, pCond, level);
pReader->type = pCond->type;
pReader->window = updateQueryTimeWindow(pReader->pTsdb, &pCond->twindows);
pReader->blockInfoBuf.numPerBucket = 1000; // 1000 tables per bucket
ASSERT(pCond->numOfCols > 0);
limitOutputBufferSize(pCond, &pReader->capacity);
@ -566,9 +644,9 @@ static int32_t doLoadBlockIndex(STsdbReader* pReader, SDataFReader* pFileReader,
continue;
}
STableBlockScanInfo* pScanInfo = p;
STableBlockScanInfo* pScanInfo = *(STableBlockScanInfo**)p;
if (pScanInfo->pBlockList == NULL) {
pScanInfo->pBlockList = taosArrayInit(4, sizeof(int32_t));
pScanInfo->pBlockList = taosArrayInit(4, sizeof(SBlockIndex));
}
taosArrayPush(pIndexList, pBlockIdx);
@ -586,7 +664,7 @@ _end:
}
static void cleanupTableScanInfo(SHashObj* pTableMap) {
STableBlockScanInfo* px = NULL;
STableBlockScanInfo** px = NULL;
while (1) {
px = taosHashIterate(pTableMap, px);
if (px == NULL) {
@ -594,8 +672,8 @@ static void cleanupTableScanInfo(SHashObj* pTableMap) {
}
// reset the index in last block when handing a new file
tMapDataClear(&px->mapData);
taosArrayClear(px->pBlockList);
tMapDataClear(&(*px)->mapData);
taosArrayClear((*px)->pBlockList);
}
}
@ -610,7 +688,8 @@ static int32_t doLoadFileBlock(STsdbReader* pReader, SArray* pIndexList, SBlockN
for (int32_t i = 0; i < numOfTables; ++i) {
SBlockIdx* pBlockIdx = taosArrayGet(pIndexList, i);
STableBlockScanInfo* pScanInfo = taosHashGet(pReader->status.pTableMap, &pBlockIdx->uid, sizeof(int64_t));
STableBlockScanInfo* pScanInfo =
*(STableBlockScanInfo**)taosHashGet(pReader->status.pTableMap, &pBlockIdx->uid, sizeof(int64_t));
tMapDataReset(&pScanInfo->mapData);
tsdbReadDataBlk(pReader->pFileReader, pBlockIdx, &pScanInfo->mapData);
@ -630,7 +709,10 @@ static int32_t doLoadFileBlock(STsdbReader* pReader, SArray* pIndexList, SBlockN
continue;
}
void* p = taosArrayPush(pScanInfo->pBlockList, &j);
SBlockIndex bIndex = {.ordinalIndex = j, .inFileOffset = block.aSubBlock->offset};
bIndex.window = (STimeWindow) {.skey = block.minKey.ts, .ekey = block.maxKey.ts};
void* p = taosArrayPush(pScanInfo->pBlockList, &bIndex);
if (p == NULL) {
tMapDataClear(&pScanInfo->mapData);
return TSDB_CODE_OUT_OF_MEMORY;
@ -821,7 +903,8 @@ int32_t getEndPosInDataBlock(STsdbReader* pReader, SBlockData* pBlockData, SData
} else if (!asc && pReader->window.skey <= pBlock->minKey.ts) {
endPos = 0;
} else {
endPos = doBinarySearchKey(pBlockData->aTSKEY, pBlock->nRow, pos, pReader->window.ekey, pReader->order);
int64_t key = asc ? pReader->window.ekey : pReader->window.skey;
endPos = doBinarySearchKey(pBlockData->aTSKEY, pBlock->nRow, pos, key, pReader->order);
}
return endPos;
@ -852,8 +935,9 @@ static int32_t copyBlockDataToSDataBlock(STsdbReader* pReader, STableBlockScanIn
// pDumpInfo->rowIndex = pBlock->nRow - 1;
} else {
int32_t pos = asc ? pBlock->nRow - 1 : 0;
int32_t order = (pReader->order == TSDB_ORDER_ASC) ? TSDB_ORDER_DESC : TSDB_ORDER_ASC;
pDumpInfo->rowIndex = doBinarySearchKey(pBlockData->aTSKEY, pBlock->nRow, pos, pReader->window.skey, order);
int32_t order = asc ? TSDB_ORDER_DESC : TSDB_ORDER_ASC;
int64_t key = asc ? pReader->window.skey : pReader->window.ekey;
pDumpInfo->rowIndex = doBinarySearchKey(pBlockData->aTSKEY, pBlock->nRow, pos, key, order);
}
}
@ -1053,14 +1137,14 @@ static int32_t fileDataBlockOrderCompar(const void* pLeft, const void* pRight, v
static int32_t doSetCurrentBlock(SDataBlockIter* pBlockIter, const char* idStr) {
SFileDataBlockInfo* pBlockInfo = getCurrentBlockInfo(pBlockIter);
if (pBlockInfo != NULL) {
STableBlockScanInfo* pScanInfo = taosHashGet(pBlockIter->pTableMap, &pBlockInfo->uid, sizeof(pBlockInfo->uid));
STableBlockScanInfo** pScanInfo = taosHashGet(pBlockIter->pTableMap, &pBlockInfo->uid, sizeof(pBlockInfo->uid));
if (pScanInfo == NULL) {
tsdbError("failed to locate the uid:%" PRIu64 " in query table uid list, %s", pBlockInfo->uid, idStr);
return TSDB_CODE_INVALID_PARA;
}
int32_t* mapDataIndex = taosArrayGet(pScanInfo->pBlockList, pBlockInfo->tbBlockIdx);
tMapDataGetItemByIdx(&pScanInfo->mapData, *mapDataIndex, &pBlockIter->block, tGetDataBlk);
SBlockIndex* pIndex = taosArrayGet((*pScanInfo)->pBlockList, pBlockInfo->tbBlockIdx);
tMapDataGetItemByIdx(&(*pScanInfo)->mapData, pIndex->ordinalIndex, &pBlockIter->block, tGetDataBlk);
}
#if 0
@ -1073,6 +1157,7 @@ static int32_t doSetCurrentBlock(SDataBlockIter* pBlockIter, const char* idStr)
static int32_t initBlockIterator(STsdbReader* pReader, SDataBlockIter* pBlockIter, int32_t numOfBlocks) {
bool asc = ASCENDING_TRAVERSE(pReader->order);
SBlockOrderSupporter sup = {0};
pBlockIter->numOfBlocks = numOfBlocks;
taosArrayClear(pBlockIter->blockList);
pBlockIter->pTableMap = pReader->status.pTableMap;
@ -1081,8 +1166,6 @@ static int32_t initBlockIterator(STsdbReader* pReader, SDataBlockIter* pBlockIte
int32_t numOfTables = (int32_t)taosHashGetSize(pReader->status.pTableMap);
int64_t st = taosGetTimestampUs();
SBlockOrderSupporter sup = {0};
int32_t code = initBlockOrderSupporter(&sup, numOfTables);
if (code != TSDB_CODE_SUCCESS) {
return code;
@ -1096,7 +1179,7 @@ static int32_t initBlockIterator(STsdbReader* pReader, SDataBlockIter* pBlockIte
break;
}
STableBlockScanInfo* pTableScanInfo = (STableBlockScanInfo*)ptr;
STableBlockScanInfo* pTableScanInfo = *(STableBlockScanInfo**)ptr;
if (pTableScanInfo->pBlockList == NULL || taosArrayGetSize(pTableScanInfo->pBlockList) == 0) {
continue;
}
@ -1111,17 +1194,11 @@ static int32_t initBlockIterator(STsdbReader* pReader, SDataBlockIter* pBlockIte
}
sup.pDataBlockInfo[sup.numOfTables] = (SBlockOrderWrapper*)buf;
SDataBlk block = {0};
for (int32_t k = 0; k < num; ++k) {
SBlockOrderWrapper wrapper = {0};
int32_t* mapDataIndex = taosArrayGet(pTableScanInfo->pBlockList, k);
tMapDataGetItemByIdx(&pTableScanInfo->mapData, *mapDataIndex, &block, tGetDataBlk);
wrapper.uid = pTableScanInfo->uid;
wrapper.offset = block.aSubBlock[0].offset;
sup.pDataBlockInfo[sup.numOfTables][k] = wrapper;
SBlockIndex* pIndex = taosArrayGet(pTableScanInfo->pBlockList, k);
sup.pDataBlockInfo[sup.numOfTables][k] =
(SBlockOrderWrapper){.uid = pTableScanInfo->uid, .offset = pIndex->inFileOffset};
cnt++;
}
@ -1212,25 +1289,22 @@ static int32_t dataBlockPartiallyRequired(STimeWindow* pWindow, SVersionRange* p
(pVerRange->maxVer < pBlock->maxVer && pVerRange->maxVer >= pBlock->minVer);
}
static SDataBlk* getNeighborBlockOfSameTable(SFileDataBlockInfo* pBlockInfo, STableBlockScanInfo* pTableBlockScanInfo,
int32_t* nextIndex, int32_t order) {
static bool getNeighborBlockOfSameTable(SFileDataBlockInfo* pBlockInfo, STableBlockScanInfo* pTableBlockScanInfo,
int32_t* nextIndex, int32_t order, SBlockIndex* pBlockIndex) {
bool asc = ASCENDING_TRAVERSE(order);
if (asc && pBlockInfo->tbBlockIdx >= taosArrayGetSize(pTableBlockScanInfo->pBlockList) - 1) {
return NULL;
return false;
}
if (!asc && pBlockInfo->tbBlockIdx == 0) {
return NULL;
return false;
}
int32_t step = asc ? 1 : -1;
*nextIndex = pBlockInfo->tbBlockIdx + step;
SDataBlk* pBlock = taosMemoryCalloc(1, sizeof(SDataBlk));
int32_t* indexInMapdata = taosArrayGet(pTableBlockScanInfo->pBlockList, *nextIndex);
tMapDataGetItemByIdx(&pTableBlockScanInfo->mapData, *indexInMapdata, pBlock, tGetDataBlk);
return pBlock;
*pBlockIndex = *(SBlockIndex*) taosArrayGet(pTableBlockScanInfo->pBlockList, *nextIndex);
// tMapDataGetItemByIdx(&pTableBlockScanInfo->mapData, pIndex->ordinalIndex, pBlock, tGetDataBlk);
return true;
}
static int32_t findFileBlockInfoIndex(SDataBlockIter* pBlockIter, SFileDataBlockInfo* pFBlockInfo) {
@ -1272,12 +1346,12 @@ static int32_t setFileBlockActiveInBlockIter(SDataBlockIter* pBlockIter, int32_t
return TSDB_CODE_SUCCESS;
}
static bool overlapWithNeighborBlock(SDataBlk* pBlock, SDataBlk* pNeighbor, int32_t order) {
static bool overlapWithNeighborBlock(SDataBlk* pBlock, SBlockIndex* pNeighborBlockIndex, int32_t order) {
// it is the last block in current file, no chance to overlap with neighbor blocks.
if (ASCENDING_TRAVERSE(order)) {
return pBlock->maxKey.ts == pNeighbor->minKey.ts;
return pBlock->maxKey.ts == pNeighborBlockIndex->window.skey;
} else {
return pBlock->minKey.ts == pNeighbor->maxKey.ts;
return pBlock->minKey.ts == pNeighborBlockIndex->window.ekey;
}
}
@ -1365,12 +1439,13 @@ static void getBlockToLoadInfo(SDataBlockToLoadInfo* pInfo, SFileDataBlockInfo*
STableBlockScanInfo* pScanInfo, TSDBKEY keyInBuf, SLastBlockReader* pLastBlockReader,
STsdbReader* pReader) {
int32_t neighborIndex = 0;
SDataBlk* pNeighbor = getNeighborBlockOfSameTable(pBlockInfo, pScanInfo, &neighborIndex, pReader->order);
SBlockIndex bIndex = {0};
bool hasNeighbor = getNeighborBlockOfSameTable(pBlockInfo, pScanInfo, &neighborIndex, pReader->order, &bIndex);
// overlap with neighbor
if (pNeighbor) {
pInfo->overlapWithNeighborBlock = overlapWithNeighborBlock(pBlock, pNeighbor, pReader->order);
taosMemoryFree(pNeighbor);
if (hasNeighbor) {
pInfo->overlapWithNeighborBlock = overlapWithNeighborBlock(pBlock, &bIndex, pReader->order);
}
// has duplicated ts of different version in this block
@ -2228,7 +2303,7 @@ static int32_t buildComposedDataBlock(STsdbReader* pReader) {
STableBlockScanInfo* pBlockScanInfo = NULL;
if (pBlockInfo != NULL) {
pBlockScanInfo = taosHashGet(pReader->status.pTableMap, &pBlockInfo->uid, sizeof(pBlockInfo->uid));
pBlockScanInfo = *(STableBlockScanInfo**)taosHashGet(pReader->status.pTableMap, &pBlockInfo->uid, sizeof(pBlockInfo->uid));
if (pBlockScanInfo == NULL) {
code = TSDB_CODE_INVALID_PARA;
tsdbError("failed to locate the uid:%" PRIu64 " in query table uid list, total tables:%d, %s", pBlockInfo->uid,
@ -2248,7 +2323,7 @@ static int32_t buildComposedDataBlock(STsdbReader* pReader) {
}
}
} else { // file blocks not exist
pBlockScanInfo = pReader->status.pTableIter;
pBlockScanInfo = *pReader->status.pTableIter;
}
SFileBlockDumpInfo* pDumpInfo = &pReader->status.fBlockDumpInfo;
@ -2473,7 +2548,7 @@ static void extractOrderedTableUidList(SUidOrderCheckInfo* pOrderCheckInfo, SRea
void* p = taosHashIterate(pStatus->pTableMap, NULL);
while (p != NULL) {
STableBlockScanInfo* pScanInfo = p;
STableBlockScanInfo* pScanInfo = *(STableBlockScanInfo**) p;
pOrderCheckInfo->tableUidList[index++] = pScanInfo->uid;
p = taosHashIterate(pStatus->pTableMap, p);
}
@ -2547,7 +2622,7 @@ static int32_t doLoadLastBlockSequentially(STsdbReader* pReader) {
while (1) {
// load the last data block of current table
STableBlockScanInfo* pScanInfo = pStatus->pTableIter;
STableBlockScanInfo* pScanInfo = *(STableBlockScanInfo**) pStatus->pTableIter;
bool hasVal = initLastBlockReader(pLastBlockReader, pScanInfo, pReader);
if (!hasVal) {
bool hasNexTable = moveToNextTable(pOrderedCheckInfo, pStatus);
@ -2585,9 +2660,9 @@ static int32_t doBuildDataBlock(STsdbReader* pReader) {
SLastBlockReader* pLastBlockReader = pReader->status.fileIter.pLastBlockReader;
if (pBlockInfo != NULL) {
pScanInfo = taosHashGet(pReader->status.pTableMap, &pBlockInfo->uid, sizeof(pBlockInfo->uid));
pScanInfo = *(STableBlockScanInfo**)taosHashGet(pReader->status.pTableMap, &pBlockInfo->uid, sizeof(pBlockInfo->uid));
} else {
pScanInfo = pReader->status.pTableIter;
pScanInfo = *pReader->status.pTableIter;
}
if (pScanInfo == NULL) {
@ -2652,11 +2727,11 @@ static int32_t buildBlockFromBufferSequentially(STsdbReader* pReader) {
}
}
STableBlockScanInfo* pBlockScanInfo = pStatus->pTableIter;
initMemDataIterator(pBlockScanInfo, pReader);
STableBlockScanInfo** pBlockScanInfo = pStatus->pTableIter;
initMemDataIterator(*pBlockScanInfo, pReader);
int64_t endKey = (ASCENDING_TRAVERSE(pReader->order)) ? INT64_MAX : INT64_MIN;
int32_t code = buildDataBlockFromBuf(pReader, pBlockScanInfo, endKey);
int32_t code = buildDataBlockFromBuf(pReader, *pBlockScanInfo, endKey);
if (code != TSDB_CODE_SUCCESS) {
return code;
}
@ -3068,14 +3143,14 @@ static int32_t checkForNeighborFileBlock(STsdbReader* pReader, STableBlockScanIn
int32_t step = ASCENDING_TRAVERSE(pReader->order) ? 1 : -1;
int32_t nextIndex = -1;
SDataBlk* pNeighborBlock = getNeighborBlockOfSameTable(pFBlock, pScanInfo, &nextIndex, pReader->order);
if (pNeighborBlock == NULL) { // do nothing
SBlockIndex bIndex = {0};
bool hasNeighbor = getNeighborBlockOfSameTable(pFBlock, pScanInfo, &nextIndex, pReader->order, &bIndex);
if (!hasNeighbor) { // do nothing
return 0;
}
bool overlap = overlapWithNeighborBlock(pBlock, pNeighborBlock, pReader->order);
taosMemoryFree(pNeighborBlock);
bool overlap = overlapWithNeighborBlock(pBlock, &bIndex, pReader->order);
if (overlap) { // load next block
SReaderStatus* pStatus = &pReader->status;
SDataBlockIter* pBlockIter = &pStatus->blockIter;
@ -3458,18 +3533,23 @@ int32_t buildDataBlockFromBufImpl(STableBlockScanInfo* pBlockScanInfo, int64_t e
// TODO refactor: with createDataBlockScanInfo
int32_t tsdbSetTableList(STsdbReader* pReader, const void* pTableList, int32_t num) {
ASSERT(pReader != NULL);
int32_t size = taosHashGetSize(pReader->status.pTableMap);
STableBlockScanInfo* p = NULL;
STableBlockScanInfo** p = NULL;
while ((p = taosHashIterate(pReader->status.pTableMap, p)) != NULL) {
clearBlockScanInfo(p);
clearBlockScanInfo(*p);
}
// todo handle the case where size is less than the value of num
ASSERT(size >= num);
taosHashClear(pReader->status.pTableMap);
STableKeyInfo* pList = (STableKeyInfo*) pTableList;
for(int32_t i = 0; i < num; ++i) {
STableBlockScanInfo info = {.lastKey = 0, .uid = pList[i].uid};
taosHashPut(pReader->status.pTableMap, &info.uid, sizeof(uint64_t), &info, sizeof(info));
STableBlockScanInfo* pInfo = getPosInBlockInfoBuf(&pReader->blockInfoBuf, i);
pInfo->uid = pList[i].uid;
taosHashPut(pReader->status.pTableMap, &pInfo->uid, sizeof(uint64_t), &pInfo, POINTER_BYTES);
}
return TDB_CODE_SUCCESS;
@ -3496,7 +3576,6 @@ static int32_t doOpenReaderImpl(STsdbReader* pReader) {
initFilesetIterator(&pReader->status.fileIter, pReader->pReadSnap->fs.aDFileSet, pReader);
resetDataBlockIterator(&pReader->status.blockIter, pReader->order);
// resetDataBlockScanInfo(pReader->status.pTableMap, pReader->window.skey);
// no data in files, let's try buffer in memory
if (pReader->status.fileIter.numOfFiles == 0) {
@ -3674,8 +3753,9 @@ void tsdbReaderClose(STsdbReader* pReader) {
cleanupDataBlockIterator(&pReader->status.blockIter);
size_t numOfTables = taosHashGetSize(pReader->status.pTableMap);
destroyBlockScanInfo(pReader->status.pTableMap);
destroyAllBlockScanInfo(pReader->status.pTableMap);
blockDataDestroy(pReader->pResBlock);
clearBlockScanInfoBuf(&pReader->blockInfoBuf);
if (pReader->pFileReader != NULL) {
tsdbDataFReaderClose(&pReader->pFileReader);
@ -3697,15 +3777,16 @@ void tsdbReaderClose(STsdbReader* pReader) {
taosMemoryFree(pLReader);
}
tsdbDebug(
"%p :io-cost summary: head-file:%" PRIu64 ", head-file time:%.2f ms, SMA:%" PRId64
tsdbDebug("%p :io-cost summary: head-file:%" PRIu64 ", head-file time:%.2f ms, SMA:%" PRId64
" SMA-time:%.2f ms, fileBlocks:%" PRId64
", fileBlocks-load-time:%.2f ms, "
"build in-memory-block-time:%.2f ms, lastBlocks:%" PRId64 ", lastBlocks-time:%.2f ms, composed-blocks:%" PRId64
", composed-blocks-time:%.2fms, STableBlockScanInfo size:%.2f Kb %s",
pReader, pCost->headFileLoad, pCost->headFileLoadTime, pCost->smaDataLoad, pCost->smaLoadTime, pCost->numOfBlocks,
pCost->blockLoadTime, pCost->buildmemBlock, pCost->lastBlockLoad, pCost->lastBlockLoadTime, pCost->composedBlocks,
pCost->buildComposedBlockTime, numOfTables * sizeof(STableBlockScanInfo) / 1000.0, pReader->idStr);
"build in-memory-block-time:%.2f ms, lastBlocks:%" PRId64
", lastBlocks-time:%.2f ms, composed-blocks:%" PRId64
", composed-blocks-time:%.2fms, STableBlockScanInfo size:%.2f Kb, creatTime:%.2f ms, %s",
pReader, pCost->headFileLoad, pCost->headFileLoadTime, pCost->smaDataLoad, pCost->smaLoadTime,
pCost->numOfBlocks, pCost->blockLoadTime, pCost->buildmemBlock, pCost->lastBlockLoad,
pCost->lastBlockLoadTime, pCost->composedBlocks, pCost->buildComposedBlockTime,
numOfTables * sizeof(STableBlockScanInfo) / 1000.0, pCost->createScanInfoList, pReader->idStr);
taosMemoryFree(pReader->idStr);
taosMemoryFree(pReader->pSchema);
@ -3758,7 +3839,7 @@ bool tsdbNextDataBlock(STsdbReader* pReader) {
if (pReader->step == EXTERNAL_ROWS_PREV) {
// prepare for the main scan
int32_t code = doOpenReaderImpl(pReader);
resetDataBlockScanInfo(pReader->status.pTableMap, pReader->innerReader[0]->window.ekey);
resetAllDataBlockScanInfo(pReader->status.pTableMap, pReader->innerReader[0]->window.ekey);
if (code != TSDB_CODE_SUCCESS) {
return code;
@ -3775,7 +3856,7 @@ bool tsdbNextDataBlock(STsdbReader* pReader) {
if (pReader->innerReader[1] != NULL && pReader->step == EXTERNAL_ROWS_MAIN) {
// prepare for the next row scan
int32_t code = doOpenReaderImpl(pReader->innerReader[1]);
resetDataBlockScanInfo(pReader->innerReader[1]->status.pTableMap, pReader->window.ekey);
resetAllDataBlockScanInfo(pReader->innerReader[1]->status.pTableMap, pReader->window.ekey);
if (code != TSDB_CODE_SUCCESS) {
return code;
}
@ -3791,7 +3872,7 @@ bool tsdbNextDataBlock(STsdbReader* pReader) {
}
bool tsdbTableNextDataBlock(STsdbReader* pReader, uint64_t uid) {
STableBlockScanInfo* pBlockScanInfo = taosHashGet(pReader->status.pTableMap, &uid, sizeof(uid));
STableBlockScanInfo* pBlockScanInfo = *(STableBlockScanInfo**)taosHashGet(pReader->status.pTableMap, &uid, sizeof(uid));
if (pBlockScanInfo == NULL) { // no data block for the table of given uid
return false;
}
@ -3799,24 +3880,24 @@ bool tsdbTableNextDataBlock(STsdbReader* pReader, uint64_t uid) {
return true;
}
static void setBlockInfo(STsdbReader* pReader, SDataBlockInfo* pDataBlockInfo) {
ASSERT(pDataBlockInfo != NULL && pReader != NULL);
pDataBlockInfo->rows = pReader->pResBlock->info.rows;
pDataBlockInfo->uid = pReader->pResBlock->info.uid;
pDataBlockInfo->window = pReader->pResBlock->info.window;
static void setBlockInfo(const STsdbReader* pReader, int32_t* rows, uint64_t* uid, STimeWindow* pWindow) {
ASSERT(pReader != NULL);
*rows = pReader->pResBlock->info.rows;
*uid = pReader->pResBlock->info.uid;
*pWindow = pReader->pResBlock->info.window;
}
void tsdbRetrieveDataBlockInfo(STsdbReader* pReader, SDataBlockInfo* pDataBlockInfo) {
void tsdbRetrieveDataBlockInfo(const STsdbReader* pReader, int32_t* rows, uint64_t* uid, STimeWindow* pWindow) {
if (pReader->type == TIMEWINDOW_RANGE_EXTERNAL) {
if (pReader->step == EXTERNAL_ROWS_MAIN) {
setBlockInfo(pReader, pDataBlockInfo);
setBlockInfo(pReader, rows, uid, pWindow);
} else if (pReader->step == EXTERNAL_ROWS_PREV) {
setBlockInfo(pReader->innerReader[0], pDataBlockInfo);
setBlockInfo(pReader->innerReader[0], rows, uid, pWindow);
} else {
setBlockInfo(pReader->innerReader[1], pDataBlockInfo);
setBlockInfo(pReader->innerReader[1], rows, uid, pWindow);
}
} else {
setBlockInfo(pReader, pDataBlockInfo);
setBlockInfo(pReader, rows, uid, pWindow);
}
}
@ -3838,7 +3919,7 @@ int32_t tsdbRetrieveDatablockSMA(STsdbReader* pReader, SColumnDataAgg*** pBlockS
SFileDataBlockInfo* pFBlock = getCurrentBlockInfo(&pReader->status.blockIter);
SDataBlk* pBlock = getCurrentBlock(&pReader->status.blockIter);
int64_t stime = taosGetTimestampUs();
// int64_t stime = taosGetTimestampUs();
SBlockLoadSuppInfo* pSup = &pReader->suppInfo;
@ -3869,7 +3950,9 @@ int32_t tsdbRetrieveDatablockSMA(STsdbReader* pReader, SColumnDataAgg*** pBlockS
size_t numOfCols = blockDataGetNumOfCols(pReader->pResBlock);
int32_t i = 0, j = 0;
while (j < numOfCols && i < taosArrayGetSize(pSup->pColAgg)) {
size_t size = taosArrayGetSize(pSup->pColAgg);
while (j < numOfCols && i < size) {
SColumnDataAgg* pAgg = taosArrayGet(pSup->pColAgg, i);
if (pAgg->colId == pSup->colIds[j]) {
if (IS_BSMA_ON(&(pReader->pSchema->columns[i]))) {
@ -3886,14 +3969,10 @@ int32_t tsdbRetrieveDatablockSMA(STsdbReader* pReader, SColumnDataAgg*** pBlockS
}
}
double elapsed = (taosGetTimestampUs() - stime) / 1000.0;
pReader->cost.smaLoadTime += elapsed;
pReader->cost.smaDataLoad += 1;
*pBlockStatis = pSup->plist;
tsdbDebug("vgId:%d, succeed to load block SMA for uid %" PRIu64 ", elapsed time:%.2f ms, %s", 0, pFBlock->uid,
elapsed, pReader->idStr);
tsdbDebug("vgId:%d, succeed to load block SMA for uid %" PRIu64 ", %s", 0, pFBlock->uid, pReader->idStr);
return code;
}
@ -3906,7 +3985,7 @@ static SArray* doRetrieveDataBlock(STsdbReader* pReader) {
}
SFileDataBlockInfo* pBlockInfo = getCurrentBlockInfo(&pStatus->blockIter);
STableBlockScanInfo* pBlockScanInfo = taosHashGet(pStatus->pTableMap, &pBlockInfo->uid, sizeof(pBlockInfo->uid));
STableBlockScanInfo* pBlockScanInfo = *(STableBlockScanInfo**)taosHashGet(pStatus->pTableMap, &pBlockInfo->uid, sizeof(pBlockInfo->uid));
if (pBlockScanInfo == NULL) {
terrno = TSDB_CODE_INVALID_PARA;
tsdbError("failed to locate the uid:%" PRIu64 " in query table uid list, total tables:%d, %s", pBlockInfo->uid,
@ -3942,6 +4021,8 @@ int32_t tsdbReaderReset(STsdbReader* pReader, SQueryTableDataCond* pCond) {
return TSDB_CODE_SUCCESS;
}
SDataBlockIter* pBlockIter = &pReader->status.blockIter;
pReader->order = pCond->order;
pReader->type = TIMEWINDOW_RANGE_CONTAINED;
pReader->status.loadFromFile = true;
@ -3958,13 +4039,12 @@ int32_t tsdbReaderReset(STsdbReader* pReader, SQueryTableDataCond* pCond) {
int32_t numOfTables = taosHashGetSize(pReader->status.pTableMap);
initFilesetIterator(&pReader->status.fileIter, pReader->pReadSnap->fs.aDFileSet, pReader);
resetDataBlockIterator(&pReader->status.blockIter, pReader->order);
resetDataBlockIterator(pBlockIter, pReader->order);
int64_t ts = ASCENDING_TRAVERSE(pReader->order) ? pReader->window.skey - 1 : pReader->window.ekey + 1;
resetDataBlockScanInfo(pReader->status.pTableMap, ts);
resetAllDataBlockScanInfo(pReader->status.pTableMap, ts);
int32_t code = 0;
SDataBlockIter* pBlockIter = &pReader->status.blockIter;
// no data in files, let's try buffer in memory
if (pReader->status.fileIter.numOfFiles == 0) {
@ -4066,7 +4146,7 @@ int64_t tsdbGetNumOfRowsInMemTable(STsdbReader* pReader) {
pStatus->pTableIter = taosHashIterate(pStatus->pTableMap, NULL);
while (pStatus->pTableIter != NULL) {
STableBlockScanInfo* pBlockScanInfo = pStatus->pTableIter;
STableBlockScanInfo* pBlockScanInfo = *(STableBlockScanInfo**)pStatus->pTableIter;
STbData* d = NULL;
if (pReader->pTsdb->mem != NULL) {

6
source/dnode/vnode/src/vnd/vnodeCommit.c
View File

@ -212,6 +212,12 @@ int vnodeCommit(SVnode *pVnode) {
vInfo("vgId:%d, start to commit, commit ID:%" PRId64 " version:%" PRId64, TD_VID(pVnode), pVnode->state.commitID,
pVnode->state.applied);
// persist wal before starting
if (walPersist(pVnode->pWal) < 0) {
vError("vgId:%d, failed to persist wal since %s", TD_VID(pVnode), terrstr());
return -1;
}
pVnode->state.commitTerm = pVnode->state.applyTerm;
// save info

6
source/dnode/vnode/src/vnd/vnodeQuery.c
View File

@ -414,9 +414,11 @@ _exit:
}
int32_t vnodeGetLoad(SVnode *pVnode, SVnodeLoad *pLoad) {
SSyncState state = syncGetState(pVnode->sync);
pLoad->vgId = TD_VID(pVnode);
pLoad->syncState = syncGetMyRole(pVnode->sync);
pLoad->syncRestore = pVnode->restored;
pLoad->syncState = state.state;
pLoad->syncRestore = state.restored;
pLoad->cacheUsage = tsdbCacheGetUsage(pVnode);
pLoad->numOfTables = metaGetTbNum(pVnode->pMeta);
pLoad->numOfTimeSeries = metaGetTimeSeriesNum(pVnode->pMeta);

44
source/dnode/vnode/src/vnd/vnodeSvr.c
View File

@ -228,30 +228,30 @@ int32_t vnodeProcessWriteMsg(SVnode *pVnode, SRpcMsg *pMsg, int64_t version, SRp
if (vnodeProcessBatchDeleteReq(pVnode, version, pReq, len, pRsp) < 0) goto _err;
break;
/* TQ */
case TDMT_VND_MQ_VG_CHANGE:
if (tqProcessVgChangeReq(pVnode->pTq, version, POINTER_SHIFT(pMsg->pCont, sizeof(SMsgHead)),
case TDMT_VND_TMQ_SUBSCRIBE:
if (tqProcessSubscribeReq(pVnode->pTq, version, POINTER_SHIFT(pMsg->pCont, sizeof(SMsgHead)),
pMsg->contLen - sizeof(SMsgHead)) < 0) {
goto _err;
}
break;
case TDMT_VND_MQ_VG_DELETE:
if (tqProcessVgDeleteReq(pVnode->pTq, version, pMsg->pCont, pMsg->contLen) < 0) {
case TDMT_VND_TMQ_DELETE_SUB:
if (tqProcessDeleteSubReq(pVnode->pTq, version, pMsg->pCont, pMsg->contLen) < 0) {
goto _err;
}
break;
case TDMT_VND_MQ_COMMIT_OFFSET:
case TDMT_VND_TMQ_COMMIT_OFFSET:
if (tqProcessOffsetCommitReq(pVnode->pTq, version, POINTER_SHIFT(pMsg->pCont, sizeof(SMsgHead)),
pMsg->contLen - sizeof(SMsgHead)) < 0) {
goto _err;
}
break;
case TDMT_VND_ADD_CHECK_INFO:
case TDMT_VND_TMQ_ADD_CHECKINFO:
if (tqProcessAddCheckInfoReq(pVnode->pTq, version, POINTER_SHIFT(pMsg->pCont, sizeof(SMsgHead)),
pMsg->contLen - sizeof(SMsgHead)) < 0) {
goto _err;
}
break;
case TDMT_VND_DELETE_CHECK_INFO:
case TDMT_VND_TMQ_DEL_CHECKINFO:
if (tqProcessDelCheckInfoReq(pVnode->pTq, version, POINTER_SHIFT(pMsg->pCont, sizeof(SMsgHead)),
pMsg->contLen - sizeof(SMsgHead)) < 0) {
goto _err;
@ -268,6 +268,11 @@ int32_t vnodeProcessWriteMsg(SVnode *pVnode, SRpcMsg *pMsg, int64_t version, SRp
goto _err;
}
} break;
case TDMT_VND_STREAM_RECOVER_STEP2: {
if (tqProcessTaskRecover2Req(pVnode->pTq, version, pMsg->pCont, pMsg->contLen) < 0) {
goto _err;
}
} break;
case TDMT_VND_ALTER_CONFIRM:
vnodeProcessAlterConfirmReq(pVnode, version, pReq, len, pRsp);
break;
@ -280,8 +285,8 @@ int32_t vnodeProcessWriteMsg(SVnode *pVnode, SRpcMsg *pMsg, int64_t version, SRp
case TDMT_VND_COMMIT:
goto _do_commit;
default:
ASSERT(0);
break;
vError("vgId:%d, unprocessed msg, %d", TD_VID(pVnode), pMsg->msgType);
return -1;
}
vTrace("vgId:%d, process %s request, code:0x%x index:%" PRId64, TD_VID(pVnode), TMSG_INFO(pMsg->msgType), pRsp->code,
@ -356,14 +361,11 @@ int32_t vnodeProcessFetchMsg(SVnode *pVnode, SRpcMsg *pMsg, SQueueInfo *pInfo) {
return 0;
}
if (pMsg->msgType == TDMT_VND_CONSUME && !pVnode->restored) {
if (pMsg->msgType == TDMT_VND_TMQ_CONSUME && !pVnode->restored) {
vnodeRedirectRpcMsg(pVnode, pMsg);
return 0;
}
char *msgstr = POINTER_SHIFT(pMsg->pCont, sizeof(SMsgHead));
int32_t msgLen = pMsg->contLen - sizeof(SMsgHead);
switch (pMsg->msgType) {
case TDMT_SCH_FETCH:
case TDMT_SCH_MERGE_FETCH:
@ -382,7 +384,7 @@ int32_t vnodeProcessFetchMsg(SVnode *pVnode, SRpcMsg *pMsg, SQueueInfo *pInfo) {
return vnodeGetTableCfg(pVnode, pMsg, true);
case TDMT_VND_BATCH_META:
return vnodeGetBatchMeta(pVnode, pMsg);
case TDMT_VND_CONSUME:
case TDMT_VND_TMQ_CONSUME:
return tqProcessPollReq(pVnode->pTq, pMsg);
case TDMT_STREAM_TASK_RUN:
return tqProcessTaskRunReq(pVnode->pTq, pMsg);
@ -390,16 +392,18 @@ int32_t vnodeProcessFetchMsg(SVnode *pVnode, SRpcMsg *pMsg, SQueueInfo *pInfo) {
case TDMT_STREAM_TASK_DISPATCH:
return tqProcessTaskDispatchReq(pVnode->pTq, pMsg, true);
#endif
/*case TDMT_STREAM_TASK_RECOVER:*/
/*return tqProcessTaskRecoverReq(pVnode->pTq, pMsg);*/
case TDMT_STREAM_RETRIEVE:
return tqProcessTaskRetrieveReq(pVnode->pTq, pMsg);
case TDMT_STREAM_TASK_DISPATCH_RSP:
return tqProcessTaskDispatchRsp(pVnode->pTq, pMsg);
/*case TDMT_STREAM_TASK_RECOVER_RSP:*/
/*return tqProcessTaskRecoverRsp(pVnode->pTq, pMsg);*/
case TDMT_STREAM_RETRIEVE:
return tqProcessTaskRetrieveReq(pVnode->pTq, pMsg);
case TDMT_STREAM_RETRIEVE_RSP:
return tqProcessTaskRetrieveRsp(pVnode->pTq, pMsg);
case TDMT_VND_STREAM_RECOVER_STEP1:
return tqProcessTaskRecover1Req(pVnode->pTq, pMsg);
case TDMT_STREAM_RECOVER_FINISH:
return tqProcessTaskRecoverFinishReq(pVnode->pTq, pMsg);
case TDMT_STREAM_RECOVER_FINISH_RSP:
return tqProcessTaskRecoverFinishRsp(pVnode->pTq, pMsg);
default:
vError("unknown msg type:%d in fetch queue", pMsg->msgType);
return TSDB_CODE_VND_APP_ERROR;

257
source/dnode/vnode/src/vnd/vnodeSync.c
View File

@ -230,142 +230,16 @@ void vnodeApplyWriteMsg(SQueueInfo *pInfo, STaosQall *qall, int32_t numOfMsgs) {
}
}
int32_t vnodeProcessSyncCtrlMsg(SVnode *pVnode, SRpcMsg *pMsg, SRpcMsg **pRsp) {
int32_t code = 0;
const STraceId *trace = &pMsg->info.traceId;
if (!syncIsInit()) {
vGError("vgId:%d, msg:%p failed to process since sync env not start", pVnode->config.vgId, pMsg);
terrno = TSDB_CODE_APP_ERROR;
return -1;
}
SSyncNode *pSyncNode = syncNodeAcquire(pVnode->sync);
if (pSyncNode == NULL) {
vGError("vgId:%d, msg:%p failed to process since invalid sync node", pVnode->config.vgId, pMsg);
terrno = TSDB_CODE_SYN_INTERNAL_ERROR;
return -1;
}
vGTrace("vgId:%d, sync msg:%p will be processed, type:%s", pVnode->config.vgId, pMsg, TMSG_INFO(pMsg->msgType));
if (pMsg->msgType == TDMT_SYNC_HEARTBEAT) {
SyncHeartbeat *pSyncMsg = syncHeartbeatFromRpcMsg2(pMsg);
code = syncNodeOnHeartbeat(pSyncNode, pSyncMsg);
syncHeartbeatDestroy(pSyncMsg);
} else if (pMsg->msgType == TDMT_SYNC_HEARTBEAT_REPLY) {
SyncHeartbeatReply *pSyncMsg = syncHeartbeatReplyFromRpcMsg2(pMsg);
code = syncNodeOnHeartbeatReply(pSyncNode, pSyncMsg);
syncHeartbeatReplyDestroy(pSyncMsg);
} else {
vGError("vgId:%d, msg:%p failed to process since error msg type:%d", pVnode->config.vgId, pMsg, pMsg->msgType);
code = -1;
}
vTrace("vgId:%d, sync msg:%p is processed, type:%s code:0x%x", pVnode->config.vgId, pMsg, TMSG_INFO(pMsg->msgType),
code);
syncNodeRelease(pSyncNode);
if (code != 0 && terrno == 0) {
terrno = TSDB_CODE_SYN_INTERNAL_ERROR;
}
return code;
}
int32_t vnodeProcessSyncMsg(SVnode *pVnode, SRpcMsg *pMsg, SRpcMsg **pRsp) {
int32_t code = 0;
const STraceId *trace = &pMsg->info.traceId;
if (!syncIsInit()) {
vGError("vgId:%d, msg:%p failed to process since sync env not start", pVnode->config.vgId, pMsg);
terrno = TSDB_CODE_APP_ERROR;
return -1;
}
SSyncNode *pSyncNode = syncNodeAcquire(pVnode->sync);
if (pSyncNode == NULL) {
vGError("vgId:%d, msg:%p failed to process since invalid sync node", pVnode->config.vgId, pMsg);
terrno = TSDB_CODE_SYN_INTERNAL_ERROR;
return -1;
}
vGTrace("vgId:%d, sync msg:%p will be processed, type:%s", pVnode->config.vgId, pMsg, TMSG_INFO(pMsg->msgType));
if (pMsg->msgType == TDMT_SYNC_TIMEOUT) {
SyncTimeout *pSyncMsg = syncTimeoutFromRpcMsg2(pMsg);
ASSERT(pSyncMsg != NULL);
code = syncNodeOnTimer(pSyncNode, pSyncMsg);
syncTimeoutDestroy(pSyncMsg);
} else if (pMsg->msgType == TDMT_SYNC_PING) {
SyncPing *pSyncMsg = syncPingFromRpcMsg2(pMsg);
ASSERT(pSyncMsg != NULL);
code = syncNodeOnPing(pSyncNode, pSyncMsg);
syncPingDestroy(pSyncMsg);
} else if (pMsg->msgType == TDMT_SYNC_PING_REPLY) {
SyncPingReply *pSyncMsg = syncPingReplyFromRpcMsg2(pMsg);
ASSERT(pSyncMsg != NULL);
code = syncNodeOnPingReply(pSyncNode, pSyncMsg);
syncPingReplyDestroy(pSyncMsg);
} else if (pMsg->msgType == TDMT_SYNC_CLIENT_REQUEST) {
SyncClientRequest *pSyncMsg = syncClientRequestFromRpcMsg2(pMsg);
ASSERT(pSyncMsg != NULL);
code = syncNodeOnClientRequest(pSyncNode, pSyncMsg, NULL);
syncClientRequestDestroy(pSyncMsg);
} else if (pMsg->msgType == TDMT_SYNC_REQUEST_VOTE) {
SyncRequestVote *pSyncMsg = syncRequestVoteFromRpcMsg2(pMsg);
ASSERT(pSyncMsg != NULL);
code = syncNodeOnRequestVote(pSyncNode, pSyncMsg);
syncRequestVoteDestroy(pSyncMsg);
} else if (pMsg->msgType == TDMT_SYNC_REQUEST_VOTE_REPLY) {
SyncRequestVoteReply *pSyncMsg = syncRequestVoteReplyFromRpcMsg2(pMsg);
ASSERT(pSyncMsg != NULL);
code = syncNodeOnRequestVoteReply(pSyncNode, pSyncMsg);
syncRequestVoteReplyDestroy(pSyncMsg);
} else if (pMsg->msgType == TDMT_SYNC_APPEND_ENTRIES) {
SyncAppendEntries *pSyncMsg = syncAppendEntriesFromRpcMsg2(pMsg);
ASSERT(pSyncMsg != NULL);
code = syncNodeOnAppendEntries(pSyncNode, pSyncMsg);
syncAppendEntriesDestroy(pSyncMsg);
} else if (pMsg->msgType == TDMT_SYNC_APPEND_ENTRIES_REPLY) {
SyncAppendEntriesReply *pSyncMsg = syncAppendEntriesReplyFromRpcMsg2(pMsg);
ASSERT(pSyncMsg != NULL);
code = syncNodeOnAppendEntriesReply(pSyncNode, pSyncMsg);
syncAppendEntriesReplyDestroy(pSyncMsg);
} else if (pMsg->msgType == TDMT_SYNC_SNAPSHOT_SEND) {
SyncSnapshotSend *pSyncMsg = syncSnapshotSendFromRpcMsg2(pMsg);
code = syncNodeOnSnapshot(pSyncNode, pSyncMsg);
syncSnapshotSendDestroy(pSyncMsg);
} else if (pMsg->msgType == TDMT_SYNC_SNAPSHOT_RSP) {
SyncSnapshotRsp *pSyncMsg = syncSnapshotRspFromRpcMsg2(pMsg);
code = syncNodeOnSnapshotReply(pSyncNode, pSyncMsg);
syncSnapshotRspDestroy(pSyncMsg);
} else if (pMsg->msgType == TDMT_SYNC_LOCAL_CMD) {
SyncLocalCmd *pSyncMsg = syncLocalCmdFromRpcMsg2(pMsg);
code = syncNodeOnLocalCmd(pSyncNode, pSyncMsg);
syncLocalCmdDestroy(pSyncMsg);
} else {
vGError("vgId:%d, msg:%p failed to process since error msg type:%d", pVnode->config.vgId, pMsg, pMsg->msgType);
code = -1;
int32_t code = syncProcessMsg(pVnode->sync, pMsg);
if (code != 0) {
vGError("vgId:%d, failed to process sync msg:%p type:%s since %s", pVnode->config.vgId, pMsg,
TMSG_INFO(pMsg->msgType), terrstr());
}
vTrace("vgId:%d, sync msg:%p is processed, type:%s code:0x%x", pVnode->config.vgId, pMsg, TMSG_INFO(pMsg->msgType),
code);
syncNodeRelease(pSyncNode);
if (code != 0 && terrno == 0) {
terrno = TSDB_CODE_SYN_INTERNAL_ERROR;
}
return code;
}
@ -416,78 +290,61 @@ static int32_t vnodeSyncSendMsg(const SEpSet *pEpSet, SRpcMsg *pMsg) {
return code;
}
static int32_t vnodeSyncGetSnapshot(SSyncFSM *pFsm, SSnapshot *pSnapshot) {
static int32_t vnodeSyncGetSnapshot(const SSyncFSM *pFsm, SSnapshot *pSnapshot) {
vnodeGetSnapshot(pFsm->data, pSnapshot);
return 0;
}
static void vnodeSyncReconfig(struct SSyncFSM *pFsm, const SRpcMsg *pMsg, SReConfigCbMeta *cbMeta) {}
static void vnodeSyncCommitMsg(SSyncFSM *pFsm, const SRpcMsg *pMsg, SFsmCbMeta cbMeta) {
if (cbMeta.isWeak == 0) {
static void vnodeSyncApplyMsg(const SSyncFSM *pFsm, const SRpcMsg *pMsg, const SFsmCbMeta *pMeta) {
SVnode *pVnode = pFsm->data;
if (cbMeta.code == 0) {
if (pMeta->code == 0) {
SRpcMsg rpcMsg = {.msgType = pMsg->msgType, .contLen = pMsg->contLen};
rpcMsg.pCont = rpcMallocCont(rpcMsg.contLen);
memcpy(rpcMsg.pCont, pMsg->pCont, pMsg->contLen);
syncGetAndDelRespRpc(pVnode->sync, cbMeta.seqNum, &rpcMsg.info);
rpcMsg.info.conn.applyIndex = cbMeta.index;
rpcMsg.info.conn.applyTerm = cbMeta.term;
rpcMsg.info = pMsg->info;
rpcMsg.info.conn.applyIndex = pMeta->index;
rpcMsg.info.conn.applyTerm = pMeta->term;
vInfo("vgId:%d, commit-cb is excuted, fsm:%p, index:%" PRId64 ", term:%" PRIu64 ", msg-index:%" PRId64
const STraceId *trace = &pMsg->info.traceId;
vGTrace("vgId:%d, commit-cb is excuted, fsm:%p, index:%" PRId64 ", term:%" PRIu64 ", msg-index:%" PRId64
", weak:%d, code:%d, state:%d %s, type:%s",
syncGetVgId(pVnode->sync), pFsm, cbMeta.index, cbMeta.term, rpcMsg.info.conn.applyIndex, cbMeta.isWeak,
cbMeta.code, cbMeta.state, syncUtilState2String(cbMeta.state), TMSG_INFO(pMsg->msgType));
pVnode->config.vgId, pFsm, pMeta->index, pMeta->term, rpcMsg.info.conn.applyIndex, pMeta->isWeak,
pMeta->code, pMeta->state, syncStr(pMeta->state), TMSG_INFO(pMsg->msgType));
tmsgPutToQueue(&pVnode->msgCb, APPLY_QUEUE, &rpcMsg);
} else {
SRpcMsg rsp = {.code = cbMeta.code, .info = pMsg->info};
vError("vgId:%d, commit-cb execute error, type:%s, index:%" PRId64 ", error:0x%x %s", syncGetVgId(pVnode->sync),
TMSG_INFO(pMsg->msgType), cbMeta.index, cbMeta.code, tstrerror(cbMeta.code));
SRpcMsg rsp = {.code = pMeta->code, .info = pMsg->info};
vError("vgId:%d, commit-cb execute error, type:%s, index:%" PRId64 ", error:0x%x %s", pVnode->config.vgId,
TMSG_INFO(pMsg->msgType), pMeta->index, pMeta->code, tstrerror(pMeta->code));
if (rsp.info.handle != NULL) {
tmsgSendRsp(&rsp);
}
}
}
}
static void vnodeSyncPreCommitMsg(SSyncFSM *pFsm, const SRpcMsg *pMsg, SFsmCbMeta cbMeta) {
if (cbMeta.isWeak == 1) {
SVnode *pVnode = pFsm->data;
vTrace("vgId:%d, pre-commit-cb is excuted, fsm:%p, index:%" PRId64 ", weak:%d, code:%d, state:%d %s, type:%s",
syncGetVgId(pVnode->sync), pFsm, cbMeta.index, cbMeta.isWeak, cbMeta.code, cbMeta.state,
syncUtilState2String(cbMeta.state), TMSG_INFO(pMsg->msgType));
if (cbMeta.code == 0) {
SRpcMsg rpcMsg = {.msgType = pMsg->msgType, .contLen = pMsg->contLen};
rpcMsg.pCont = rpcMallocCont(rpcMsg.contLen);
memcpy(rpcMsg.pCont, pMsg->pCont, pMsg->contLen);
syncGetAndDelRespRpc(pVnode->sync, cbMeta.seqNum, &rpcMsg.info);
rpcMsg.info.conn.applyIndex = cbMeta.index;
rpcMsg.info.conn.applyTerm = cbMeta.term;
tmsgPutToQueue(&pVnode->msgCb, APPLY_QUEUE, &rpcMsg);
} else {
SRpcMsg rsp = {.code = cbMeta.code, .info = pMsg->info};
vError("vgId:%d, pre-commit-cb execute error, type:%s, error:0x%x %s", syncGetVgId(pVnode->sync),
TMSG_INFO(pMsg->msgType), cbMeta.code, tstrerror(cbMeta.code));
if (rsp.info.handle != NULL) {
tmsgSendRsp(&rsp);
}
}
static void vnodeSyncCommitMsg(const SSyncFSM *pFsm, const SRpcMsg *pMsg, const SFsmCbMeta *pMeta) {
if (pMeta->isWeak == 0) {
vnodeSyncApplyMsg(pFsm, pMsg, pMeta);
}
}
static void vnodeSyncRollBackMsg(SSyncFSM *pFsm, const SRpcMsg *pMsg, SFsmCbMeta cbMeta) {
static void vnodeSyncPreCommitMsg(const SSyncFSM *pFsm, const SRpcMsg *pMsg, const SFsmCbMeta *pMeta) {
if (pMeta->isWeak == 1) {
vnodeSyncApplyMsg(pFsm, pMsg, pMeta);
}
}
static void vnodeSyncRollBackMsg(const SSyncFSM *pFsm, const SRpcMsg *pMsg, const SFsmCbMeta *pMeta) {
SVnode *pVnode = pFsm->data;
vTrace("vgId:%d, rollback-cb is excuted, fsm:%p, index:%" PRId64 ", weak:%d, code:%d, state:%d %s, type:%s",
syncGetVgId(pVnode->sync), pFsm, cbMeta.index, cbMeta.isWeak, cbMeta.code, cbMeta.state,
syncUtilState2String(cbMeta.state), TMSG_INFO(pMsg->msgType));
pVnode->config.vgId, pFsm, pMeta->index, pMeta->isWeak, pMeta->code, pMeta->state, syncStr(pMeta->state),
TMSG_INFO(pMsg->msgType));
}
#define USE_TSDB_SNAPSHOT
static int32_t vnodeSnapshotStartRead(struct SSyncFSM *pFsm, void *pParam, void **ppReader) {
static int32_t vnodeSnapshotStartRead(const SSyncFSM *pFsm, void *pParam, void **ppReader) {
#ifdef USE_TSDB_SNAPSHOT
SVnode *pVnode = pFsm->data;
SSnapshotParam *pSnapshotParam = pParam;
@ -499,7 +356,7 @@ static int32_t vnodeSnapshotStartRead(struct SSyncFSM *pFsm, void *pParam, void
#endif
}
static int32_t vnodeSnapshotStopRead(struct SSyncFSM *pFsm, void *pReader) {
static int32_t vnodeSnapshotStopRead(const SSyncFSM *pFsm, void *pReader) {
#ifdef USE_TSDB_SNAPSHOT
SVnode *pVnode = pFsm->data;
int32_t code = vnodeSnapReaderClose(pReader);
@ -510,7 +367,7 @@ static int32_t vnodeSnapshotStopRead(struct SSyncFSM *pFsm, void *pReader) {
#endif
}
static int32_t vnodeSnapshotDoRead(struct SSyncFSM *pFsm, void *pReader, void **ppBuf, int32_t *len) {
static int32_t vnodeSnapshotDoRead(const SSyncFSM *pFsm, void *pReader, void **ppBuf, int32_t *len) {
#ifdef USE_TSDB_SNAPSHOT
SVnode *pVnode = pFsm->data;
int32_t code = vnodeSnapRead(pReader, (uint8_t **)ppBuf, len);
@ -529,7 +386,7 @@ static int32_t vnodeSnapshotDoRead(struct SSyncFSM *pFsm, void *pReader, void **
#endif
}
static int32_t vnodeSnapshotStartWrite(struct SSyncFSM *pFsm, void *pParam, void **ppWriter) {
static int32_t vnodeSnapshotStartWrite(const SSyncFSM *pFsm, void *pParam, void **ppWriter) {
#ifdef USE_TSDB_SNAPSHOT
SVnode *pVnode = pFsm->data;
SSnapshotParam *pSnapshotParam = pParam;
@ -553,7 +410,7 @@ static int32_t vnodeSnapshotStartWrite(struct SSyncFSM *pFsm, void *pParam, void
#endif
}
static int32_t vnodeSnapshotStopWrite(struct SSyncFSM *pFsm, void *pWriter, bool isApply, SSnapshot *pSnapshot) {
static int32_t vnodeSnapshotStopWrite(const SSyncFSM *pFsm, void *pWriter, bool isApply, SSnapshot *pSnapshot) {
#ifdef USE_TSDB_SNAPSHOT
SVnode *pVnode = pFsm->data;
vInfo("vgId:%d, stop write vnode snapshot, apply:%d, index:%" PRId64 " term:%" PRIu64 " config:%" PRId64,
@ -568,7 +425,7 @@ static int32_t vnodeSnapshotStopWrite(struct SSyncFSM *pFsm, void *pWriter, bool
#endif
}
static int32_t vnodeSnapshotDoWrite(struct SSyncFSM *pFsm, void *pWriter, void *pBuf, int32_t len) {
static int32_t vnodeSnapshotDoWrite(const SSyncFSM *pFsm, void *pWriter, void *pBuf, int32_t len) {
#ifdef USE_TSDB_SNAPSHOT
SVnode *pVnode = pFsm->data;
vDebug("vgId:%d, continue write vnode snapshot, len:%d", pVnode->config.vgId, len);
@ -580,9 +437,7 @@ static int32_t vnodeSnapshotDoWrite(struct SSyncFSM *pFsm, void *pWriter, void *
#endif
}
static void vnodeLeaderTransfer(struct SSyncFSM *pFsm, const SRpcMsg *pMsg, SFsmCbMeta cbMeta) {}
static void vnodeRestoreFinish(struct SSyncFSM *pFsm) {
static void vnodeRestoreFinish(const SSyncFSM *pFsm) {
SVnode *pVnode = pFsm->data;
do {
@ -602,7 +457,7 @@ static void vnodeRestoreFinish(struct SSyncFSM *pFsm) {
vDebug("vgId:%d, sync restore finished", pVnode->config.vgId);
}
static void vnodeBecomeFollower(struct SSyncFSM *pFsm) {
static void vnodeBecomeFollower(const SSyncFSM *pFsm) {
SVnode *pVnode = pFsm->data;
vDebug("vgId:%d, become follower", pVnode->config.vgId);
@ -616,7 +471,7 @@ static void vnodeBecomeFollower(struct SSyncFSM *pFsm) {
taosThreadMutexUnlock(&pVnode->lock);
}
static void vnodeBecomeLeader(struct SSyncFSM *pFsm) {
static void vnodeBecomeLeader(const SSyncFSM *pFsm) {
SVnode *pVnode = pFsm->data;
vDebug("vgId:%d, become leader", pVnode->config.vgId);
@ -638,10 +493,10 @@ static SSyncFSM *vnodeSyncMakeFsm(SVnode *pVnode) {
pFsm->FpRollBackCb = vnodeSyncRollBackMsg;
pFsm->FpGetSnapshotInfo = vnodeSyncGetSnapshot;
pFsm->FpRestoreFinishCb = vnodeRestoreFinish;
pFsm->FpLeaderTransferCb = vnodeLeaderTransfer;
pFsm->FpLeaderTransferCb = NULL;
pFsm->FpBecomeLeaderCb = vnodeBecomeLeader;
pFsm->FpBecomeFollowerCb = vnodeBecomeFollower;
pFsm->FpReConfigCb = vnodeSyncReconfig;
pFsm->FpReConfigCb = NULL;
pFsm->FpSnapshotStartRead = vnodeSnapshotStartRead;
pFsm->FpSnapshotStopRead = vnodeSnapshotStopRead;
pFsm->FpSnapshotDoRead = vnodeSnapshotDoRead;
@ -659,10 +514,13 @@ int32_t vnodeSyncOpen(SVnode *pVnode, char *path) {
.vgId = pVnode->config.vgId,
.syncCfg = pVnode->config.syncCfg,
.pWal = pVnode->pWal,
.msgcb = NULL,
.FpSendMsg = vnodeSyncSendMsg,
.FpEqMsg = vnodeSyncEqMsg,
.FpEqCtrlMsg = vnodeSyncEqCtrlMsg,
.msgcb = &pVnode->msgCb,
.syncSendMSg = vnodeSyncSendMsg,
.syncEqMsg = vnodeSyncEqMsg,
.syncEqCtrlMsg = vnodeSyncEqCtrlMsg,
.pingMs = 5000,
.electMs = 4000,
.heartbeatMs = 700,
};
snprintf(syncInfo.path, sizeof(syncInfo.path), "%s%ssync", path, TD_DIRSEP);
@ -681,15 +539,11 @@ int32_t vnodeSyncOpen(SVnode *pVnode, char *path) {
return -1;
}
setPingTimerMS(pVnode->sync, 5000);
setElectTimerMS(pVnode->sync, 4000);
setHeartbeatTimerMS(pVnode->sync, 700);
return 0;
}
void vnodeSyncStart(SVnode *pVnode) {
vDebug("vgId:%d, start sync", pVnode->config.vgId);
syncSetMsgCb(pVnode->sync, &pVnode->msgCb);
syncStart(pVnode->sync);
}
@ -698,12 +552,21 @@ void vnodeSyncClose(SVnode *pVnode) {
syncStop(pVnode->sync);
}
bool vnodeIsRoleLeader(SVnode *pVnode) { return syncGetMyRole(pVnode->sync) == TAOS_SYNC_STATE_LEADER; }
bool vnodeIsRoleLeader(SVnode *pVnode) {
SSyncState state = syncGetState(pVnode->sync);
return state.state == TAOS_SYNC_STATE_LEADER;
}
bool vnodeIsLeader(SVnode *pVnode) {
if (!syncIsReady(pVnode->sync)) {
vDebug("vgId:%d, vnode not ready, state:%s, restore:%d", pVnode->config.vgId, syncGetMyRoleStr(pVnode->sync),
syncRestoreFinish(pVnode->sync));
SSyncState state = syncGetState(pVnode->sync);
if (state.state != TAOS_SYNC_STATE_LEADER || !state.restored) {
if (state.state != TAOS_SYNC_STATE_LEADER) {
terrno = TSDB_CODE_SYN_NOT_LEADER;
} else {
terrno = TSDB_CODE_APP_NOT_READY;
}
vDebug("vgId:%d, vnode not ready, state:%s, restore:%d", pVnode->config.vgId, syncStr(state.state), state.restored);
return false;
}

30
source/libs/catalog/src/catalog.c
View File

@ -320,7 +320,7 @@ _return:
}
int32_t ctgChkAuth(SCatalog* pCtg, SRequestConnInfo* pConn, const char* user, const char* dbFName, AUTH_TYPE type,
bool* pass) {
bool* pass, bool* exists) {
bool inCache = false;
int32_t code = 0;
@ -329,6 +329,13 @@ int32_t ctgChkAuth(SCatalog* pCtg, SRequestConnInfo* pConn, const char* user, co
CTG_ERR_RET(ctgChkAuthFromCache(pCtg, (char*)user, (char*)dbFName, type, &inCache, pass));
if (inCache) {
if (exists) {
*exists = true;
}
return TSDB_CODE_SUCCESS;
} else if (exists) {
*exists = false;
return TSDB_CODE_SUCCESS;
}
@ -1032,7 +1039,7 @@ int32_t catalogChkTbMetaVersion(SCatalog* pCtg, SRequestConnInfo* pConn, SArray*
switch (tbType) {
case TSDB_CHILD_TABLE: {
SName stb = name;
strcpy(stb.tname, stbName);
tstrncpy(stb.tname, stbName, sizeof(stb.tname));
ctgRemoveTbMeta(pCtg, &stb);
break;
}
@ -1373,13 +1380,30 @@ int32_t catalogChkAuth(SCatalog* pCtg, SRequestConnInfo* pConn, const char* user
}
int32_t code = 0;
CTG_ERR_JRET(ctgChkAuth(pCtg, pConn, user, dbFName, type, pass));
CTG_ERR_JRET(ctgChkAuth(pCtg, pConn, user, dbFName, type, pass, NULL));
_return:
CTG_API_LEAVE(code);
}
int32_t catalogChkAuthFromCache(SCatalog* pCtg, SRequestConnInfo* pConn, const char* user, const char* dbFName, AUTH_TYPE type,
bool* pass, bool* exists) {
CTG_API_ENTER();
if (NULL == pCtg || NULL == pConn || NULL == user || NULL == dbFName || NULL == pass || NULL == exists) {
CTG_API_LEAVE(TSDB_CODE_CTG_INVALID_INPUT);
}
int32_t code = 0;
CTG_ERR_JRET(ctgChkAuth(pCtg, pConn, user, dbFName, type, pass, exists));
_return:
CTG_API_LEAVE(code);
}
int32_t catalogGetServerVersion(SCatalog* pCtg, SRequestConnInfo* pConn, char** pVersion) {
CTG_API_ENTER();

5
source/libs/catalog/src/ctgUtil.c
View File

@ -924,6 +924,11 @@ int32_t ctgGetVgInfosFromHashValue(SCatalog* pCtg, SCtgTaskReq* tReq, SDBVgInfo*
if (1 == vgNum) {
void* pIter = taosHashIterate(dbInfo->vgHash, NULL);
if (NULL == pIter) {
ctgError("empty vgHash, db:%s, vgroup number:%d", dbFName, vgNum);
CTG_ERR_RET(TSDB_CODE_CTG_INTERNAL_ERROR);
}
for (int32_t i = 0; i < tbNum; ++i) {
vgInfo = taosMemoryMalloc(sizeof(SVgroupInfo));
if (NULL == vgInfo) {

20
source/libs/catalog/test/catalogTests.cpp
View File

@ -2771,10 +2771,30 @@ TEST(apiTest, catalogChkAuth_test) {
ASSERT_EQ(code, 0);
bool pass = false;
bool exists = false;
code = catalogChkAuthFromCache(pCtg, mockPointer, ctgTestUsername, ctgTestDbname, AUTH_TYPE_READ, &pass, &exists);
ASSERT_EQ(code, 0);
ASSERT_EQ(exists, false);
code = catalogChkAuth(pCtg, mockPointer, ctgTestUsername, ctgTestDbname, AUTH_TYPE_READ, &pass);
ASSERT_EQ(code, 0);
ASSERT_EQ(pass, true);
while (true) {
uint64_t n = 0;
ctgdGetStatNum("runtime.numOfOpDequeue", (void *)&n);
if (n != 1) {
taosMsleep(50);
} else {
break;
}
}
code = catalogChkAuthFromCache(pCtg, mockPointer, ctgTestUsername, ctgTestDbname, AUTH_TYPE_READ, &pass, &exists);
ASSERT_EQ(code, 0);
ASSERT_EQ(pass, true);
ASSERT_EQ(exists, true);
catalogDestroy();
}

3
source/libs/executor/inc/executil.h
View File

@ -96,11 +96,12 @@ typedef struct SColMatchInfo {
int32_t matchType; // determinate the source according to col id or slot id
} SColMatchInfo;
typedef struct SExecTaskInfo SExecTaskInfo;
typedef struct STableListInfo STableListInfo;
struct SqlFunctionCtx;
int32_t createScanTableListInfo(SScanPhysiNode* pScanNode, SNodeList* pGroupTags, bool groupSort, SReadHandle* pHandle,
STableListInfo* pTableListInfo, SNode* pTagCond, SNode* pTagIndexCond, const char* id);
STableListInfo* pTableListInfo, SNode* pTagCond, SNode* pTagIndexCond, SExecTaskInfo* pTaskInfo);
STableListInfo* tableListCreate();
void* tableListDestroy(STableListInfo* pTableListInfo);

36
source/libs/executor/inc/executorimpl.h
View File

@ -89,25 +89,10 @@ typedef struct STableScanAnalyzeInfo SFileBlockLoadRecorder;
typedef struct STaskCostInfo {
int64_t created;
int64_t start;
uint64_t loadStatisTime;
uint64_t loadFileBlockTime;
uint64_t loadDataInCacheTime;
uint64_t loadStatisSize;
uint64_t loadFileBlockSize;
uint64_t loadDataInCacheSize;
uint64_t loadDataTime;
SFileBlockLoadRecorder* pRecoder;
uint64_t elapsedTime;
uint64_t winInfoSize;
uint64_t tableInfoSize;
uint64_t hashSize;
uint64_t numOfTimeWindows;
SArray* queryProfEvents; // SArray<SQueryProfEvent>
SHashObj* operatorProfResults; // map<operator_type, SQueryProfEvent>
double extractListTime;
double groupIdMapTime;
SFileBlockLoadRecorder* pRecoder;
} STaskCostInfo;
typedef struct SOperatorCostInfo {
@ -150,18 +135,13 @@ typedef struct {
SSchemaWrapper* schema;
char tbName[TSDB_TABLE_NAME_LEN];
SSDataBlock* pullOverBlk; // for streaming
SWalFilterCond cond;
int64_t lastScanUid;
int8_t recoverStep;
SQueryTableDataCond tableCond;
int64_t recoverStartVer;
int64_t recoverEndVer;
int64_t fillHistoryVer1;
int64_t fillHistoryVer2;
int8_t triggerSaved;
int64_t deleteMarkSaved;
// int8_t triggerSaved;
// int64_t deleteMarkSaved;
SStreamState* pState;
} SStreamTaskInfo;
@ -461,8 +441,10 @@ typedef struct SPartitionDataInfo {
typedef struct STimeWindowAggSupp {
int8_t calTrigger;
int64_t waterMark;
int8_t calTriggerSaved;
int64_t deleteMark;
int64_t deleteMarkSaved;
int64_t waterMark;
TSKEY maxTs;
TSKEY minTs;
SColumnInfoData timeWindowData; // query time window info for scalar function execution.
@ -921,7 +903,7 @@ int32_t getBufferPgSize(int32_t rowSize, uint32_t* defaultPgsz, uint32_t* defaul
void doSetOperatorCompleted(SOperatorInfo* pOperator);
void doFilter(const SNode* pFilterNode, SSDataBlock* pBlock, SColMatchInfo* pColMatchInfo, SFilterInfo* pFilterInfo);
int32_t addTagPseudoColumnData(SReadHandle* pHandle, SExprInfo* pPseudoExpr, int32_t numOfPseudoExpr,
SSDataBlock* pBlock, const char* idStr);
SSDataBlock* pBlock, int32_t rows, const char* idStr);
void cleanupAggSup(SAggSupporter* pAggSup);
void appendOneRowToDataBlock(SSDataBlock* pBlock, STupleHandle* pTupleHandle);

17
source/libs/executor/src/cachescanoperator.c
View File

@ -152,25 +152,27 @@ SSDataBlock* doScanCache(SOperatorInfo* pOperator) {
pInfo->indexOfBufferedRes = 0;
}
SSDataBlock* pRes = pInfo->pRes;
if (pInfo->indexOfBufferedRes < pInfo->pBufferredRes->info.rows) {
for (int32_t i = 0; i < taosArrayGetSize(pInfo->matchInfo.pList); ++i) {
SColMatchItem* pMatchInfo = taosArrayGet(pInfo->matchInfo.pList, i);
int32_t slotId = pMatchInfo->dstSlotId;
SColumnInfoData* pSrc = taosArrayGet(pInfo->pBufferredRes->pDataBlock, slotId);
SColumnInfoData* pDst = taosArrayGet(pInfo->pRes->pDataBlock, slotId);
SColumnInfoData* pDst = taosArrayGet(pRes->pDataBlock, slotId);
char* p = colDataGetData(pSrc, pInfo->indexOfBufferedRes);
bool isNull = colDataIsNull_s(pSrc, pInfo->indexOfBufferedRes);
colDataAppend(pDst, 0, p, isNull);
}
pInfo->pRes->info.uid = *(tb_uid_t*)taosArrayGet(pInfo->pUidList, pInfo->indexOfBufferedRes);
pInfo->pRes->info.rows = 1;
pRes->info.uid = *(tb_uid_t*)taosArrayGet(pInfo->pUidList, pInfo->indexOfBufferedRes);
pRes->info.rows = 1;
if (pInfo->pseudoExprSup.numOfExprs > 0) {
SExprSupp* pSup = &pInfo->pseudoExprSup;
int32_t code = addTagPseudoColumnData(&pInfo->readHandle, pSup->pExprInfo, pSup->numOfExprs, pInfo->pRes,
int32_t code = addTagPseudoColumnData(&pInfo->readHandle, pSup->pExprInfo, pSup->numOfExprs, pRes, pRes->info.rows,
GET_TASKID(pTaskInfo));
if (code != TSDB_CODE_SUCCESS) {
pTaskInfo->code = code;
@ -178,10 +180,9 @@ SSDataBlock* doScanCache(SOperatorInfo* pOperator) {
}
}
pInfo->pRes->info.groupId = getTableGroupId(pTableList, pInfo->pRes->info.uid);
pRes->info.groupId = getTableGroupId(pTableList, pRes->info.uid);
pInfo->indexOfBufferedRes += 1;
return pInfo->pRes;
return pRes;
} else {
doSetOperatorCompleted(pOperator);
return NULL;
@ -221,7 +222,7 @@ SSDataBlock* doScanCache(SOperatorInfo* pOperator) {
ASSERT((pInfo->retrieveType & CACHESCAN_RETRIEVE_LAST_ROW) == CACHESCAN_RETRIEVE_LAST_ROW);
pInfo->pRes->info.uid = *(tb_uid_t*)taosArrayGet(pInfo->pUidList, 0);
code = addTagPseudoColumnData(&pInfo->readHandle, pSup->pExprInfo, pSup->numOfExprs, pInfo->pRes,
code = addTagPseudoColumnData(&pInfo->readHandle, pSup->pExprInfo, pSup->numOfExprs, pInfo->pRes, pInfo->pRes->info.rows,
GET_TASKID(pTaskInfo));
if (code != TSDB_CODE_SUCCESS) {
pTaskInfo->code = code;

156
source/libs/executor/src/executil.c
View File

@ -775,75 +775,6 @@ end:
return code;
}
static int tableUidCompare(const void* a, const void* b) {
int64_t u1 = *(uint64_t*)a;
int64_t u2 = *(uint64_t*)b;
if (u1 == u2) {
return 0;
}
return u1 < u2 ? -1 : 1;
}
static int32_t optimizeTbnameInCond(void* metaHandle, int64_t suid, SArray* list, SNode* cond, SHashObj* tags) {
int32_t ret = -1;
if (nodeType(cond) == QUERY_NODE_OPERATOR) {
ret = optimizeTbnameInCondImpl(metaHandle, suid, list, cond);
if (ret != -1) {
metaGetTableTagsByUids(metaHandle, suid, list, tags);
removeInvalidTable(list, tags);
}
}
if (nodeType(cond) != QUERY_NODE_LOGIC_CONDITION || ((SLogicConditionNode*)cond)->condType != LOGIC_COND_TYPE_AND) {
return ret;
}
bool hasTbnameCond = false;
SLogicConditionNode* pNode = (SLogicConditionNode*)cond;
SNodeList* pList = (SNodeList*)pNode->pParameterList;
int32_t len = LIST_LENGTH(pList);
if (len <= 0) return ret;
SListCell* cell = pList->pHead;
for (int i = 0; i < len; i++) {
if (cell == NULL) break;
if (optimizeTbnameInCondImpl(metaHandle, suid, list, cell->pNode) == 0) {
hasTbnameCond = true;
break;
}
cell = cell->pNext;
}
taosArraySort(list, tableUidCompare);
taosArrayRemoveDuplicate(list, tableUidCompare, NULL);
if (hasTbnameCond) {
ret = metaGetTableTagsByUids(metaHandle, suid, list, tags);
removeInvalidTable(list, tags);
}
return ret;
}
/*
* handle invalid uid
*/
static int32_t removeInvalidTable(SArray* uids, SHashObj* tags) {
if (taosArrayGetSize(uids) <= 0) return 0;
SArray* validUid = taosArrayInit(taosArrayGetSize(uids), sizeof(int64_t));
for (int32_t i = 0; i < taosArrayGetSize(uids); i++) {
int64_t* uid = taosArrayGet(uids, i);
if (taosHashGet(tags, uid, sizeof(int64_t)) != NULL) {
taosArrayPush(validUid, uid);
}
}
taosArraySwap(uids, validUid);
taosArrayDestroy(validUid);
return 0;
}
static int32_t nameComparFn(const void* p1, const void* p2) {
const char* pName1 = *(const char**)p1;
const char* pName2 = *(const char**)p2;
@ -897,6 +828,81 @@ static SArray* getTableNameList(const SNodeListNode* pList) {
return pNewList;
}
static int tableUidCompare(const void* a, const void* b) {
uint64_t u1 = *(uint64_t*)a;
uint64_t u2 = *(uint64_t*)b;
if (u1 == u2) {
return 0;
}
return u1 < u2 ? -1 : 1;
}
static int32_t optimizeTbnameInCond(void* metaHandle, int64_t suid, SArray* list, SNode* cond, SHashObj* tags) {
int32_t ret = -1;
if (nodeType(cond) == QUERY_NODE_OPERATOR) {
ret = optimizeTbnameInCondImpl(metaHandle, suid, list, cond);
if (ret != -1) {
metaGetTableTagsByUids(metaHandle, suid, list, tags);
removeInvalidTable(list, tags);
}
}
if (nodeType(cond) != QUERY_NODE_LOGIC_CONDITION || ((SLogicConditionNode*)cond)->condType != LOGIC_COND_TYPE_AND) {
return ret;
}
bool hasTbnameCond = false;
SLogicConditionNode* pNode = (SLogicConditionNode*)cond;
SNodeList* pList = (SNodeList*)pNode->pParameterList;
int32_t len = LIST_LENGTH(pList);
if (len <= 0) {
return ret;
}
SListCell* cell = pList->pHead;
for (int i = 0; i < len; i++) {
if (cell == NULL) break;
if (optimizeTbnameInCondImpl(metaHandle, suid, list, cell->pNode) == 0) {
hasTbnameCond = true;
break;
}
cell = cell->pNext;
}
taosArraySort(list, tableUidCompare);
taosArrayRemoveDuplicate(list, tableUidCompare, NULL);
if (hasTbnameCond) {
ret = metaGetTableTagsByUids(metaHandle, suid, list, tags);
removeInvalidTable(list, tags);
}
return ret;
}
/*
* handle invalid uid
*/
static int32_t removeInvalidTable(SArray* uids, SHashObj* tags) {
if (taosArrayGetSize(uids) <= 0) return 0;
SArray* validUid = taosArrayInit(taosArrayGetSize(uids), sizeof(int64_t));
for (int32_t i = 0; i < taosArrayGetSize(uids); i++) {
int64_t* uid = taosArrayGet(uids, i);
if (taosHashGet(tags, uid, sizeof(int64_t)) != NULL) {
taosArrayPush(validUid, uid);
}
}
taosArraySwap(uids, validUid);
taosArrayDestroy(validUid);
return 0;
}
static int32_t optimizeTbnameInCondImpl(void* metaHandle, int64_t suid, SArray* list, SNode* pTagCond) {
if (nodeType(pTagCond) != QUERY_NODE_OPERATOR) {
return -1;
@ -1757,7 +1763,7 @@ STableListInfo* tableListCreate() {
goto _error;
}
pListInfo->map = taosHashInit(32, taosGetDefaultHashFunction(TSDB_DATA_TYPE_BINARY), false, HASH_ENTRY_LOCK);
pListInfo->map = taosHashInit(32, taosGetDefaultHashFunction(TSDB_DATA_TYPE_BIGINT), false, HASH_ENTRY_LOCK);
if (pListInfo->map == NULL) {
goto _error;
}
@ -1882,8 +1888,9 @@ int32_t buildGroupIdMapForAllTables(STableListInfo* pTableListInfo, SReadHandle*
int32_t createScanTableListInfo(SScanPhysiNode* pScanNode, SNodeList* pGroupTags, bool groupSort, SReadHandle* pHandle,
STableListInfo* pTableListInfo, SNode* pTagCond, SNode* pTagIndexCond,
const char* idStr) {
struct SExecTaskInfo* pTaskInfo) {
int64_t st = taosGetTimestampUs();
const char* idStr = GET_TASKID(pTaskInfo);
if (pHandle == NULL) {
qError("invalid handle, in creating operator tree, %s", idStr);
@ -1899,7 +1906,8 @@ int32_t createScanTableListInfo(SScanPhysiNode* pScanNode, SNodeList* pGroupTags
ASSERT(pTableListInfo->numOfOuputGroups == 1);
int64_t st1 = taosGetTimestampUs();
qDebug("generate queried table list completed, elapsed time:%.2f ms %s", (st1 - st) / 1000.0, idStr);
pTaskInfo->cost.extractListTime = (st1 - st) / 1000.0;
qDebug("extract queried table list completed, elapsed time:%.2f ms %s", pTaskInfo->cost.extractListTime, idStr);
if (taosArrayGetSize(pTableListInfo->pTableList) == 0) {
qDebug("no table qualified for query, %s" PRIx64, idStr);
@ -1911,8 +1919,8 @@ int32_t createScanTableListInfo(SScanPhysiNode* pScanNode, SNodeList* pGroupTags
return code;
}
int64_t st2 = taosGetTimestampUs();
qDebug("generate group id map completed, elapsed time:%.2f ms %s", (st2 - st1) / 1000.0, idStr);
pTaskInfo->cost.groupIdMapTime = (taosGetTimestampUs() - st1)/1000.0;
qDebug("generate group id map completed, elapsed time:%.2f ms %s", pTaskInfo->cost.groupIdMapTime, idStr);
return TSDB_CODE_SUCCESS;
}

104
source/libs/executor/src/executor.c
View File

@ -70,6 +70,26 @@ static int32_t doSetSMABlock(SOperatorInfo* pOperator, void* input, size_t numOf
}
}
static int32_t doSetStreamOpOpen(SOperatorInfo* pOperator, char* id) {
{
ASSERT(pOperator != NULL);
if (pOperator->operatorType != QUERY_NODE_PHYSICAL_PLAN_STREAM_SCAN) {
if (pOperator->numOfDownstream == 0) {
qError("failed to find stream scan operator to set the input data block, %s" PRIx64, id);
return TSDB_CODE_QRY_APP_ERROR;
}
if (pOperator->numOfDownstream > 1) { // not handle this in join query
qError("join not supported for stream block scan, %s" PRIx64, id);
return TSDB_CODE_QRY_APP_ERROR;
}
pOperator->status = OP_NOT_OPENED;
return doSetStreamOpOpen(pOperator->pDownstream[0], id);
}
}
return 0;
}
static int32_t doSetStreamBlock(SOperatorInfo* pOperator, void* input, size_t numOfBlocks, int32_t type, char* id) {
ASSERT(pOperator != NULL);
if (pOperator->operatorType != QUERY_NODE_PHYSICAL_PLAN_STREAM_SCAN) {
@ -117,7 +137,22 @@ static int32_t doSetStreamBlock(SOperatorInfo* pOperator, void* input, size_t nu
}
}
static FORCE_INLINE void streamInputBlockDataDestory(void* pBlock) { blockDataDestroy((SSDataBlock*)pBlock); }
int32_t qSetStreamOpOpen(qTaskInfo_t tinfo) {
if (tinfo == NULL) {
return TSDB_CODE_QRY_APP_ERROR;
}
SExecTaskInfo* pTaskInfo = (SExecTaskInfo*)tinfo;
int32_t code = doSetStreamOpOpen(pTaskInfo->pRoot, GET_TASKID(pTaskInfo));
if (code != TSDB_CODE_SUCCESS) {
qError("%s failed to set the stream block data", GET_TASKID(pTaskInfo));
} else {
qDebug("%s set the stream block successfully", GET_TASKID(pTaskInfo));
}
return code;
}
int32_t qSetMultiStreamInput(qTaskInfo_t tinfo, const void* pBlocks, size_t numOfBlocks, int32_t type) {
if (tinfo == NULL) {
@ -706,8 +741,7 @@ int32_t qStreamInput(qTaskInfo_t tinfo, void* pItem) {
int32_t qStreamSourceRecoverStep1(qTaskInfo_t tinfo, int64_t ver) {
SExecTaskInfo* pTaskInfo = (SExecTaskInfo*)tinfo;
ASSERT(pTaskInfo->execModel == OPTR_EXEC_MODEL_STREAM);
pTaskInfo->streamInfo.recoverStartVer = 0;
pTaskInfo->streamInfo.recoverEndVer = ver;
pTaskInfo->streamInfo.fillHistoryVer1 = ver;
pTaskInfo->streamInfo.recoverStep = STREAM_RECOVER_STEP__PREPARE1;
return 0;
}
@ -715,8 +749,7 @@ int32_t qStreamSourceRecoverStep1(qTaskInfo_t tinfo, int64_t ver) {
int32_t qStreamSourceRecoverStep2(qTaskInfo_t tinfo, int64_t ver) {
SExecTaskInfo* pTaskInfo = (SExecTaskInfo*)tinfo;
ASSERT(pTaskInfo->execModel == OPTR_EXEC_MODEL_STREAM);
pTaskInfo->streamInfo.recoverStartVer = pTaskInfo->streamInfo.recoverEndVer;
pTaskInfo->streamInfo.recoverEndVer = ver;
pTaskInfo->streamInfo.fillHistoryVer2 = ver;
pTaskInfo->streamInfo.recoverStep = STREAM_RECOVER_STEP__PREPARE2;
return 0;
}
@ -737,22 +770,44 @@ int32_t qStreamSetParamForRecover(qTaskInfo_t tinfo) {
pOperator->operatorType == QUERY_NODE_PHYSICAL_PLAN_STREAM_SEMI_INTERVAL ||
pOperator->operatorType == QUERY_NODE_PHYSICAL_PLAN_STREAM_FINAL_INTERVAL) {
SStreamIntervalOperatorInfo* pInfo = pOperator->info;
pTaskInfo->streamInfo.triggerSaved = pInfo->twAggSup.calTrigger;
pTaskInfo->streamInfo.deleteMarkSaved = pInfo->twAggSup.deleteMark;
ASSERT(pInfo->twAggSup.calTrigger == STREAM_TRIGGER_AT_ONCE ||
pInfo->twAggSup.calTrigger == STREAM_TRIGGER_WINDOW_CLOSE);
ASSERT(pInfo->twAggSup.calTriggerSaved == 0);
ASSERT(pInfo->twAggSup.deleteMarkSaved == 0);
qInfo("save stream param for interval: %d, %" PRId64, pInfo->twAggSup.calTrigger, pInfo->twAggSup.deleteMark);
pInfo->twAggSup.calTriggerSaved = pInfo->twAggSup.calTrigger;
pInfo->twAggSup.deleteMarkSaved = pInfo->twAggSup.deleteMark;
pInfo->twAggSup.calTrigger = STREAM_TRIGGER_AT_ONCE;
pInfo->twAggSup.deleteMark = INT64_MAX;
} else if (pOperator->operatorType == QUERY_NODE_PHYSICAL_PLAN_STREAM_SESSION ||
pOperator->operatorType == QUERY_NODE_PHYSICAL_PLAN_STREAM_SEMI_SESSION ||
pOperator->operatorType == QUERY_NODE_PHYSICAL_PLAN_STREAM_FINAL_SESSION) {
SStreamSessionAggOperatorInfo* pInfo = pOperator->info;
pTaskInfo->streamInfo.triggerSaved = pInfo->twAggSup.calTrigger;
pTaskInfo->streamInfo.deleteMarkSaved = pInfo->twAggSup.deleteMark;
ASSERT(pInfo->twAggSup.calTrigger == STREAM_TRIGGER_AT_ONCE ||
pInfo->twAggSup.calTrigger == STREAM_TRIGGER_WINDOW_CLOSE);
ASSERT(pInfo->twAggSup.calTriggerSaved == 0);
ASSERT(pInfo->twAggSup.deleteMarkSaved == 0);
qInfo("save stream param for session: %d, %" PRId64, pInfo->twAggSup.calTrigger, pInfo->twAggSup.deleteMark);
pInfo->twAggSup.calTriggerSaved = pInfo->twAggSup.calTrigger;
pInfo->twAggSup.deleteMarkSaved = pInfo->twAggSup.deleteMark;
pInfo->twAggSup.calTrigger = STREAM_TRIGGER_AT_ONCE;
pInfo->twAggSup.deleteMark = INT64_MAX;
} else if (pOperator->operatorType == QUERY_NODE_PHYSICAL_PLAN_STREAM_STATE) {
SStreamStateAggOperatorInfo* pInfo = pOperator->info;
pTaskInfo->streamInfo.triggerSaved = pInfo->twAggSup.calTrigger;
pTaskInfo->streamInfo.deleteMarkSaved = pInfo->twAggSup.deleteMark;
ASSERT(pInfo->twAggSup.calTrigger == STREAM_TRIGGER_AT_ONCE ||
pInfo->twAggSup.calTrigger == STREAM_TRIGGER_WINDOW_CLOSE);
ASSERT(pInfo->twAggSup.calTriggerSaved == 0);
ASSERT(pInfo->twAggSup.deleteMarkSaved == 0);
qInfo("save stream param for state: %d, %" PRId64, pInfo->twAggSup.calTrigger, pInfo->twAggSup.deleteMark);
pInfo->twAggSup.calTriggerSaved = pInfo->twAggSup.calTrigger;
pInfo->twAggSup.deleteMarkSaved = pInfo->twAggSup.deleteMark;
pInfo->twAggSup.calTrigger = STREAM_TRIGGER_AT_ONCE;
pInfo->twAggSup.deleteMark = INT64_MAX;
}
@ -782,21 +837,36 @@ int32_t qStreamRestoreParam(qTaskInfo_t tinfo) {
pOperator->operatorType == QUERY_NODE_PHYSICAL_PLAN_STREAM_SEMI_INTERVAL ||
pOperator->operatorType == QUERY_NODE_PHYSICAL_PLAN_STREAM_FINAL_INTERVAL) {
SStreamIntervalOperatorInfo* pInfo = pOperator->info;
ASSERT(pInfo->twAggSup.calTrigger == STREAM_TRIGGER_AT_ONCE);
ASSERT(pInfo->twAggSup.deleteMark == INT64_MAX);
pInfo->twAggSup.calTrigger = pTaskInfo->streamInfo.triggerSaved;
pInfo->twAggSup.deleteMark = pTaskInfo->streamInfo.deleteMarkSaved;
pInfo->twAggSup.calTrigger = pInfo->twAggSup.calTriggerSaved;
pInfo->twAggSup.deleteMark = pInfo->twAggSup.deleteMarkSaved;
ASSERT(pInfo->twAggSup.calTrigger == STREAM_TRIGGER_AT_ONCE ||
pInfo->twAggSup.calTrigger == STREAM_TRIGGER_WINDOW_CLOSE);
qInfo("restore stream param for interval: %d, %" PRId64, pInfo->twAggSup.calTrigger, pInfo->twAggSup.deleteMark);
} else if (pOperator->operatorType == QUERY_NODE_PHYSICAL_PLAN_STREAM_SESSION ||
pOperator->operatorType == QUERY_NODE_PHYSICAL_PLAN_STREAM_SEMI_SESSION ||
pOperator->operatorType == QUERY_NODE_PHYSICAL_PLAN_STREAM_FINAL_SESSION) {
SStreamSessionAggOperatorInfo* pInfo = pOperator->info;
ASSERT(pInfo->twAggSup.calTrigger == STREAM_TRIGGER_AT_ONCE);
ASSERT(pInfo->twAggSup.deleteMark == INT64_MAX);
pInfo->twAggSup.calTrigger = pTaskInfo->streamInfo.triggerSaved;
pInfo->twAggSup.deleteMark = pTaskInfo->streamInfo.deleteMarkSaved;
pInfo->twAggSup.calTrigger = pInfo->twAggSup.calTriggerSaved;
pInfo->twAggSup.deleteMark = pInfo->twAggSup.deleteMarkSaved;
ASSERT(pInfo->twAggSup.calTrigger == STREAM_TRIGGER_AT_ONCE ||
pInfo->twAggSup.calTrigger == STREAM_TRIGGER_WINDOW_CLOSE);
qInfo("restore stream param for session: %d, %" PRId64, pInfo->twAggSup.calTrigger, pInfo->twAggSup.deleteMark);
} else if (pOperator->operatorType == QUERY_NODE_PHYSICAL_PLAN_STREAM_STATE) {
SStreamStateAggOperatorInfo* pInfo = pOperator->info;
ASSERT(pInfo->twAggSup.calTrigger == STREAM_TRIGGER_AT_ONCE);
ASSERT(pInfo->twAggSup.deleteMark == INT64_MAX);
pInfo->twAggSup.calTrigger = pTaskInfo->streamInfo.triggerSaved;
pInfo->twAggSup.deleteMark = pTaskInfo->streamInfo.deleteMarkSaved;
pInfo->twAggSup.calTrigger = pInfo->twAggSup.calTriggerSaved;
pInfo->twAggSup.deleteMark = pInfo->twAggSup.deleteMarkSaved;
ASSERT(pInfo->twAggSup.calTrigger == STREAM_TRIGGER_AT_ONCE ||
pInfo->twAggSup.calTrigger == STREAM_TRIGGER_WINDOW_CLOSE);
qInfo("restore stream param for state: %d, %" PRId64, pInfo->twAggSup.calTrigger, pInfo->twAggSup.deleteMark);
}
// iterate operator tree

115
source/libs/executor/src/executorimpl.c
View File

@ -574,6 +574,9 @@ int32_t projectApplyFunctions(SExprInfo* pExpr, SSDataBlock* pResult, SSDataBloc
// if the source equals to the destination, it is to create a new column as the result of scalar
// function or some operators.
bool createNewColModel = (pResult == pSrcBlock);
if (createNewColModel) {
blockDataEnsureCapacity(pResult, pResult->info.rows);
}
int32_t numOfRows = 0;
@ -623,6 +626,7 @@ int32_t projectApplyFunctions(SExprInfo* pExpr, SSDataBlock* pResult, SSDataBloc
int32_t startOffset = createNewColModel ? 0 : pResult->info.rows;
ASSERT(pResult->info.capacity > 0);
colDataMergeCol(pResColData, startOffset, (int32_t*)&pResult->info.capacity, &idata, dest.numOfRows);
colDataDestroy(&idata);
@ -844,57 +848,6 @@ STimeWindow getAlignQueryTimeWindow(SInterval* pInterval, int32_t precision, int
return win;
}
#if 0
static bool overlapWithTimeWindow(STaskAttr* pQueryAttr, SDataBlockInfo* pBlockInfo) {
STimeWindow w = {0};
TSKEY sk = TMIN(pQueryAttr->window.skey, pQueryAttr->window.ekey);
TSKEY ek = TMAX(pQueryAttr->window.skey, pQueryAttr->window.ekey);
if (true) {
// getAlignQueryTimeWindow(pQueryAttr, pBlockInfo->window.skey, sk, ek, &w);
assert(w.ekey >= pBlockInfo->window.skey);
if (w.ekey < pBlockInfo->window.ekey) {
return true;
}
while (1) {
// getNextTimeWindow(pQueryAttr, &w);
if (w.skey > pBlockInfo->window.ekey) {
break;
}
assert(w.ekey > pBlockInfo->window.ekey);
if (w.skey <= pBlockInfo->window.ekey && w.skey > pBlockInfo->window.skey) {
return true;
}
}
} else {
// getAlignQueryTimeWindow(pQueryAttr, pBlockInfo->window.ekey, sk, ek, &w);
assert(w.skey <= pBlockInfo->window.ekey);
if (w.skey > pBlockInfo->window.skey) {
return true;
}
while (1) {
// getNextTimeWindow(pQueryAttr, &w);
if (w.ekey < pBlockInfo->window.skey) {
break;
}
assert(w.skey < pBlockInfo->window.skey);
if (w.ekey < pBlockInfo->window.ekey && w.ekey >= pBlockInfo->window.skey) {
return true;
}
}
}
return false;
}
#endif
int32_t loadDataBlockOnDemand(SExecTaskInfo* pTaskInfo, STableScanInfo* pTableScanInfo, SSDataBlock* pBlock,
uint32_t* status) {
*status = BLK_DATA_NOT_LOAD;
@ -1447,40 +1400,18 @@ void doBuildResultDatablock(SOperatorInfo* pOperator, SOptrBasicInfo* pbInfo, SG
}
}
static int32_t compressQueryColData(SColumnInfoData* pColRes, int32_t numOfRows, char* data, int8_t compressed) {
int32_t colSize = pColRes->info.bytes * numOfRows;
return (*(tDataTypes[pColRes->info.type].compFunc))(pColRes->pData, colSize, numOfRows, data,
colSize + COMP_OVERFLOW_BYTES, compressed, NULL, 0);
}
void queryCostStatis(SExecTaskInfo* pTaskInfo) {
STaskCostInfo* pSummary = &pTaskInfo->cost;
// uint64_t hashSize = taosHashGetMemSize(pQInfo->runtimeEnv.pResultRowHashTable);
// hashSize += taosHashGetMemSize(pRuntimeEnv->tableqinfoGroupInfo.map);
// pSummary->hashSize = hashSize;
// SResultRowPool* p = pTaskInfo->pool;
// if (p != NULL) {
// pSummary->winInfoSize = getResultRowPoolMemSize(p);
// pSummary->numOfTimeWindows = getNumOfAllocatedResultRows(p);
// } else {
// pSummary->winInfoSize = 0;
// pSummary->numOfTimeWindows = 0;
// }
SFileBlockLoadRecorder* pRecorder = pSummary->pRecoder;
if (pSummary->pRecoder != NULL) {
qDebug(
"%s :cost summary: elapsed time:%.2f ms, total blocks:%d, load block SMA:%d, load data block:%d, total "
"rows:%" PRId64 ", check rows:%" PRId64,
GET_TASKID(pTaskInfo), pSummary->elapsedTime / 1000.0, pRecorder->totalBlocks, pRecorder->loadBlockStatis,
pRecorder->loadBlocks, pRecorder->totalRows, pRecorder->totalCheckedRows);
"%s :cost summary: elapsed time:%.2f ms, extract tableList:%.2f ms, createGroupIdMap:%.2f ms, total blocks:%d, "
"load block SMA:%d, load data block:%d, total rows:%" PRId64 ", check rows:%" PRId64,
GET_TASKID(pTaskInfo), pSummary->elapsedTime / 1000.0, pSummary->extractListTime, pSummary->groupIdMapTime,
pRecorder->totalBlocks, pRecorder->loadBlockStatis, pRecorder->loadBlocks, pRecorder->totalRows,
pRecorder->totalCheckedRows);
}
// qDebug("QInfo:0x%"PRIx64" :cost summary: winResPool size:%.2f Kb, numOfWin:%"PRId64", tableInfoSize:%.2f Kb,
// hashTable:%.2f Kb", pQInfo->qId, pSummary->winInfoSize/1024.0,
// pSummary->numOfTimeWindows, pSummary->tableInfoSize/1024.0, pSummary->hashSize/1024.0);
}
// static void updateOffsetVal(STaskRuntimeEnv *pRuntimeEnv, SDataBlockInfo *pBlockInfo) {
@ -3267,6 +3198,11 @@ SOperatorInfo* createFillOperatorInfo(SOperatorInfo* downstream, SFillPhysiNode*
static SExecTaskInfo* createExecTaskInfo(uint64_t queryId, uint64_t taskId, EOPTR_EXEC_MODEL model, char* dbFName) {
SExecTaskInfo* pTaskInfo = taosMemoryCalloc(1, sizeof(SExecTaskInfo));
if (pTaskInfo == NULL) {
terrno = TSDB_CODE_OUT_OF_MEMORY;
return NULL;
}
setTaskStatus(pTaskInfo, TASK_NOT_COMPLETED);
pTaskInfo->schemaInfo.dbname = strdup(dbFName);
@ -3397,7 +3333,7 @@ SOperatorInfo* createOperatorTree(SPhysiNode* pPhyNode, SExecTaskInfo* pTaskInfo
int32_t code =
createScanTableListInfo(&pTableScanNode->scan, pTableScanNode->pGroupTags, pTableScanNode->groupSort, pHandle,
pTableListInfo, pTagCond, pTagIndexCond, idstr);
pTableListInfo, pTagCond, pTagIndexCond, pTaskInfo);
if (code) {
pTaskInfo->code = code;
qError("failed to createScanTableListInfo, code:%s, %s", tstrerror(code), idstr);
@ -3415,9 +3351,9 @@ SOperatorInfo* createOperatorTree(SPhysiNode* pPhyNode, SExecTaskInfo* pTaskInfo
pTaskInfo->cost.pRecoder = &pScanInfo->readRecorder;
} else if (QUERY_NODE_PHYSICAL_PLAN_TABLE_MERGE_SCAN == type) {
STableMergeScanPhysiNode* pTableScanNode = (STableMergeScanPhysiNode*)pPhyNode;
int32_t code =
createScanTableListInfo(&pTableScanNode->scan, pTableScanNode->pGroupTags, /*pTableScanNode->groupSort*/true, pHandle,
pTableListInfo, pTagCond, pTagIndexCond, idstr);
int32_t code = createScanTableListInfo(&pTableScanNode->scan, pTableScanNode->pGroupTags, true, pHandle,
pTableListInfo, pTagCond, pTagIndexCond, pTaskInfo);
if (code) {
pTaskInfo->code = code;
qError("failed to createScanTableListInfo, code: %s", tstrerror(code));
@ -3442,7 +3378,7 @@ SOperatorInfo* createOperatorTree(SPhysiNode* pPhyNode, SExecTaskInfo* pTaskInfo
if (pHandle->vnode) {
int32_t code =
createScanTableListInfo(&pTableScanNode->scan, pTableScanNode->pGroupTags, pTableScanNode->groupSort,
pHandle, pTableListInfo, pTagCond, pTagIndexCond, idstr);
pHandle, pTableListInfo, pTagCond, pTagIndexCond, pTaskInfo);
if (code) {
pTaskInfo->code = code;
qError("failed to createScanTableListInfo, code: %s", tstrerror(code));
@ -3469,7 +3405,7 @@ SOperatorInfo* createOperatorTree(SPhysiNode* pPhyNode, SExecTaskInfo* pTaskInfo
STagScanPhysiNode* pScanPhyNode = (STagScanPhysiNode*)pPhyNode;
int32_t code = createScanTableListInfo(pScanPhyNode, NULL, false, pHandle, pTableListInfo, pTagCond,
pTagIndexCond, idstr);
pTagIndexCond, pTaskInfo);
if (code != TSDB_CODE_SUCCESS) {
pTaskInfo->code = code;
qError("failed to getTableList, code: %s", tstrerror(code));
@ -3502,7 +3438,7 @@ SOperatorInfo* createOperatorTree(SPhysiNode* pPhyNode, SExecTaskInfo* pTaskInfo
SLastRowScanPhysiNode* pScanNode = (SLastRowScanPhysiNode*)pPhyNode;
int32_t code = createScanTableListInfo(&pScanNode->scan, pScanNode->pGroupTags, true, pHandle, pTableListInfo,
pTagCond, pTagIndexCond, idstr);
pTagCond, pTagIndexCond, pTaskInfo);
if (code != TSDB_CODE_SUCCESS) {
pTaskInfo->code = code;
return NULL;
@ -3834,10 +3770,8 @@ int32_t createExecTaskInfoImpl(SSubplan* pPlan, SExecTaskInfo** pTaskInfo, SRead
char* sql, EOPTR_EXEC_MODEL model) {
uint64_t queryId = pPlan->id.queryId;
int32_t code = TSDB_CODE_SUCCESS;
*pTaskInfo = createExecTaskInfo(queryId, taskId, model, pPlan->dbFName);
if (*pTaskInfo == NULL) {
code = TSDB_CODE_QRY_OUT_OF_MEMORY;
goto _complete;
}
@ -3856,17 +3790,16 @@ int32_t createExecTaskInfoImpl(SSubplan* pPlan, SExecTaskInfo** pTaskInfo, SRead
createOperatorTree(pPlan->pNode, *pTaskInfo, pHandle, pPlan->pTagCond, pPlan->pTagIndexCond, pPlan->user);
if (NULL == (*pTaskInfo)->pRoot) {
code = (*pTaskInfo)->code;
terrno = (*pTaskInfo)->code;
goto _complete;
}
return code;
return TSDB_CODE_SUCCESS;
_complete:
taosMemoryFree(sql);
doDestroyTask(*pTaskInfo);
terrno = code;
return code;
return terrno;
}
void doDestroyTask(SExecTaskInfo* pTaskInfo) {

7
source/libs/executor/src/groupoperator.c
View File

@ -621,9 +621,10 @@ int32_t* setupColumnOffset(const SSDataBlock* pBlock, int32_t rowCapacity) {
}
static void clearPartitionOperator(SPartitionOperatorInfo* pInfo) {
void* ite = NULL;
while ((ite = taosHashIterate(pInfo->pGroupSet, ite)) != NULL) {
taosArrayDestroy(((SDataGroupInfo*)ite)->pPageList);
int32_t size = taosArrayGetSize(pInfo->sortedGroupArray);
for (int32_t i = 0; i < size; i++) {
SDataGroupInfo* pGp = taosArrayGet(pInfo->sortedGroupArray, i);
taosArrayDestroy(pGp->pPageList);
}
taosArrayClear(pInfo->sortedGroupArray);
clearDiskbasedBuf(pInfo->pBuf);

144
source/libs/executor/src/scanoperator.c
View File

@ -331,7 +331,8 @@ static bool doLoadBlockSMA(STableScanInfo* pTableScanInfo, SSDataBlock* pBlock,
}
}
for (int32_t i = 0; i < taosArrayGetSize(pTableScanInfo->matchInfo.pList); ++i) {
size_t num = taosArrayGetSize(pTableScanInfo->matchInfo.pList);
for (int32_t i = 0; i < num; ++i) {
SColMatchItem* pColMatchInfo = taosArrayGet(pTableScanInfo->matchInfo.pList, i);
if (!pColMatchInfo->needOutput) {
continue;
@ -343,11 +344,12 @@ static bool doLoadBlockSMA(STableScanInfo* pTableScanInfo, SSDataBlock* pBlock,
return true;
}
static void doSetTagColumnData(STableScanInfo* pTableScanInfo, SSDataBlock* pBlock, SExecTaskInfo* pTaskInfo) {
static void doSetTagColumnData(STableScanInfo* pTableScanInfo, SSDataBlock* pBlock, SExecTaskInfo* pTaskInfo,
int32_t rows) {
if (pTableScanInfo->pseudoSup.numOfExprs > 0) {
SExprSupp* pSup = &pTableScanInfo->pseudoSup;
int32_t code = addTagPseudoColumnData(&pTableScanInfo->readHandle, pSup->pExprInfo, pSup->numOfExprs, pBlock,
int32_t code = addTagPseudoColumnData(&pTableScanInfo->readHandle, pSup->pExprInfo, pSup->numOfExprs, pBlock, rows,
GET_TASKID(pTaskInfo));
if (code != TSDB_CODE_SUCCESS) {
T_LONG_JMP(pTaskInfo->env, code);
@ -382,6 +384,17 @@ void applyLimitOffset(SLimitInfo* pLimitInfo, SSDataBlock* pBlock, SExecTaskInfo
}
}
static void ensureBlockCapacity(SSDataBlock* pBlock, int32_t capacity) {
// keep the value of rows temporarily
int32_t rows = pBlock->info.rows;
pBlock->info.rows = 0;
blockDataEnsureCapacity(pBlock, capacity);
// restore the rows number
pBlock->info.rows = rows;
}
static int32_t loadDataBlock(SOperatorInfo* pOperator, STableScanInfo* pTableScanInfo, SSDataBlock* pBlock,
uint32_t* status) {
SExecTaskInfo* pTaskInfo = pOperator->pTaskInfo;
@ -413,7 +426,10 @@ static int32_t loadDataBlock(SOperatorInfo* pOperator, STableScanInfo* pTableSca
qDebug("%s data block skipped, brange:%" PRId64 "-%" PRId64 ", rows:%d", GET_TASKID(pTaskInfo),
pBlockInfo->window.skey, pBlockInfo->window.ekey, pBlockInfo->rows);
doSetTagColumnData(pTableScanInfo, pBlock, pTaskInfo);
if (pTableScanInfo->pseudoSup.numOfExprs > 0) {
ensureBlockCapacity(pBlock, pBlock->info.rows);
}
doSetTagColumnData(pTableScanInfo, pBlock, pTaskInfo, 1);
pCost->skipBlocks += 1;
return TSDB_CODE_SUCCESS;
} else if (*status == FUNC_DATA_REQUIRED_STATIS_LOAD) {
@ -423,7 +439,11 @@ static int32_t loadDataBlock(SOperatorInfo* pOperator, STableScanInfo* pTableSca
if (success) { // failed to load the block sma data, data block statistics does not exist, load data block instead
qDebug("%s data block SMA loaded, brange:%" PRId64 "-%" PRId64 ", rows:%d", GET_TASKID(pTaskInfo),
pBlockInfo->window.skey, pBlockInfo->window.ekey, pBlockInfo->rows);
doSetTagColumnData(pTableScanInfo, pBlock, pTaskInfo);
if (pTableScanInfo->pseudoSup.numOfExprs > 0) {
ensureBlockCapacity(pBlock, pBlock->info.rows);
}
doSetTagColumnData(pTableScanInfo, pBlock, pTaskInfo, 1);
return TSDB_CODE_SUCCESS;
} else {
qDebug("%s failed to load SMA, since not all columns have SMA", GET_TASKID(pTaskInfo));
@ -472,8 +492,9 @@ static int32_t loadDataBlock(SOperatorInfo* pOperator, STableScanInfo* pTableSca
return terrno;
}
ensureBlockCapacity(pBlock, pBlock->info.rows);
relocateColumnData(pBlock, pTableScanInfo->matchInfo.pList, pCols, true);
doSetTagColumnData(pTableScanInfo, pBlock, pTaskInfo);
doSetTagColumnData(pTableScanInfo, pBlock, pTaskInfo, pBlock->info.rows);
// restore the previous value
pCost->totalRows -= pBlock->info.rows;
@ -513,26 +534,29 @@ static void prepareForDescendingScan(STableScanInfo* pTableScanInfo, SqlFunction
}
int32_t addTagPseudoColumnData(SReadHandle* pHandle, SExprInfo* pPseudoExpr, int32_t numOfPseudoExpr,
SSDataBlock* pBlock, const char* idStr) {
SSDataBlock* pBlock, int32_t rows, const char* idStr) {
// currently only the tbname pseudo column
if (numOfPseudoExpr == 0) {
return TSDB_CODE_SUCCESS;
}
// backup the rows
int32_t backupRows = pBlock->info.rows;
pBlock->info.rows = rows;
SMetaReader mr = {0};
metaReaderInit(&mr, pHandle->meta, 0);
int32_t code = metaGetTableEntryByUid(&mr, pBlock->info.uid);
metaReaderReleaseLock(&mr);
if (code != TSDB_CODE_SUCCESS) {
qError("failed to get table meta, uid:0x%" PRIx64 ", code:%s, %s", pBlock->info.uid, tstrerror(terrno), idStr);
metaReaderClear(&mr);
return terrno;
}
metaReaderReleaseLock(&mr);
for (int32_t j = 0; j < numOfPseudoExpr; ++j) {
SExprInfo* pExpr = &pPseudoExpr[j];
int32_t dstSlotId = pExpr->base.resSchema.slotId;
SColumnInfoData* pColInfoData = taosArrayGet(pBlock->pDataBlock, dstSlotId);
@ -572,6 +596,9 @@ int32_t addTagPseudoColumnData(SReadHandle* pHandle, SExprInfo* pPseudoExpr, int
}
metaReaderClear(&mr);
// restore the rows
pBlock->info.rows = backupRows;
return TSDB_CODE_SUCCESS;
}
@ -616,14 +643,10 @@ static SSDataBlock* doTableScanImpl(SOperatorInfo* pOperator) {
blockDataCleanup(pBlock);
SDataBlockInfo binfo = pBlock->info;
tsdbRetrieveDataBlockInfo(pTableScanInfo->dataReader, &binfo);
SDataBlockInfo* pBInfo = &pBlock->info;
tsdbRetrieveDataBlockInfo(pTableScanInfo->dataReader, &pBInfo->rows, &pBInfo->uid, &pBInfo->window);
binfo.capacity = binfo.rows;
blockDataEnsureCapacity(pBlock, binfo.rows);
pBlock->info = binfo;
ASSERT(binfo.uid != 0);
ASSERT(pBInfo->uid != 0);
pBlock->info.groupId = getTableGroupId(pTaskInfo->pTableInfoList, pBlock->info.uid);
uint32_t status = 0;
@ -1146,20 +1169,19 @@ static SSDataBlock* readPreVersionData(SOperatorInfo* pTableScanOp, uint64_t tbU
bool hasBlock = tsdbNextDataBlock(pReader);
if (hasBlock) {
SDataBlockInfo binfo = {0};
tsdbRetrieveDataBlockInfo(pReader, &binfo);
SDataBlockInfo* pBInfo = &pBlock->info;
int32_t rows = 0;
tsdbRetrieveDataBlockInfo(pReader, &rows, &pBInfo->uid, &pBInfo->window);
SArray* pCols = tsdbRetrieveDataBlock(pReader, NULL);
blockDataEnsureCapacity(pBlock, binfo.rows);
pBlock->info.window = binfo.window;
pBlock->info.uid = binfo.uid;
pBlock->info.rows = binfo.rows;
blockDataEnsureCapacity(pBlock, rows);
pBlock->info.rows = rows;
relocateColumnData(pBlock, pTableScanInfo->matchInfo.pList, pCols, true);
doSetTagColumnData(pTableScanInfo, pBlock, pTaskInfo);
doSetTagColumnData(pTableScanInfo, pBlock, pTaskInfo, rows);
pBlock->info.groupId = getTableGroupId(pTaskInfo->pTableInfoList, binfo.uid);
pBlock->info.groupId = getTableGroupId(pTaskInfo->pTableInfoList, pBInfo->uid);
}
tsdbReaderClose(pReader);
@ -1181,12 +1203,6 @@ static uint64_t getGroupIdByCol(SStreamScanInfo* pInfo, uint64_t uid, TSKEY ts,
static uint64_t getGroupIdByUid(SStreamScanInfo* pInfo, uint64_t uid) {
return getTableGroupId(pInfo->pTableScanOp->pTaskInfo->pTableInfoList, uid);
// SHashObj* map = pInfo->pTableScanOp->pTaskInfo->pTableInfoList.map;
// uint64_t* groupId = taosHashGet(map, &uid, sizeof(int64_t));
// if (groupId) {
// return *groupId;
// }
// return 0;
}
static uint64_t getGroupIdByData(SStreamScanInfo* pInfo, uint64_t uid, TSKEY ts, int64_t maxVersion) {
@ -1631,7 +1647,7 @@ static int32_t setBlockIntoRes(SStreamScanInfo* pInfo, const SSDataBlock* pBlock
// currently only the tbname pseudo column
if (pInfo->numOfPseudoExpr > 0) {
int32_t code = addTagPseudoColumnData(&pInfo->readHandle, pInfo->pPseudoExpr, pInfo->numOfPseudoExpr, pInfo->pRes,
GET_TASKID(pTaskInfo));
pInfo->pRes->info.rows, GET_TASKID(pTaskInfo));
if (code != TSDB_CODE_SUCCESS) {
blockDataFreeRes((SSDataBlock*)pBlock);
T_LONG_JMP(pTaskInfo->env, code);
@ -1641,11 +1657,11 @@ static int32_t setBlockIntoRes(SStreamScanInfo* pInfo, const SSDataBlock* pBlock
if (filter) {
doFilter(pInfo->pCondition, pInfo->pRes, NULL, NULL);
}
blockDataUpdateTsWindow(pInfo->pRes, pInfo->primaryTsIndex);
blockDataFreeRes((SSDataBlock*)pBlock);
calBlockTbName(&pInfo->tbnameCalSup, pInfo->pRes);
return 0;
}
@ -1863,7 +1879,6 @@ static SSDataBlock* doStreamScan(SOperatorInfo* pOperator) {
}
#endif
#if 1
if (pTaskInfo->streamInfo.recoverStep == STREAM_RECOVER_STEP__PREPARE1 ||
pTaskInfo->streamInfo.recoverStep == STREAM_RECOVER_STEP__PREPARE2) {
STableScanInfo* pTSInfo = pInfo->pTableScanOp->info;
@ -1875,6 +1890,11 @@ static SSDataBlock* doStreamScan(SOperatorInfo* pOperator) {
pTSInfo->cond.startVersion = pTaskInfo->streamInfo.fillHistoryVer1 + 1;
pTSInfo->cond.endVersion = pTaskInfo->streamInfo.fillHistoryVer2;
}
/*resetTableScanInfo(pTSInfo, pWin);*/
tsdbReaderClose(pTSInfo->dataReader);
pTSInfo->dataReader = NULL;
pTSInfo->scanTimes = 0;
pTSInfo->currentGroupId = -1;
pTaskInfo->streamInfo.recoverStep = STREAM_RECOVER_STEP__SCAN;
@ -1888,9 +1908,12 @@ static SSDataBlock* doStreamScan(SOperatorInfo* pOperator) {
return pBlock;
}
pTaskInfo->streamInfo.recoverStep = STREAM_RECOVER_STEP__NONE;
STableScanInfo* pTSInfo = pInfo->pTableScanOp->info;
pTSInfo->cond.startVersion = 0;
pTSInfo->cond.endVersion = -1;
return NULL;
}
#endif
size_t total = taosArrayGetSize(pInfo->pBlockLists);
// TODO: refactor
@ -2142,7 +2165,9 @@ static SSDataBlock* doRawScan(SOperatorInfo* pOperator) {
longjmp(pTaskInfo->env, TSDB_CODE_TSC_QUERY_CANCELLED);
}
tsdbRetrieveDataBlockInfo(pInfo->dataReader, &pBlock->info);
int32_t rows = 0;
tsdbRetrieveDataBlockInfo(pInfo->dataReader, &rows, &pBlock->info.uid, &pBlock->info.window);
pBlock->info.rows = rows;
SArray* pCols = tsdbRetrieveDataBlock(pInfo->dataReader, NULL);
pBlock->pDataBlock = pCols;
@ -2270,7 +2295,7 @@ SOperatorInfo* createRawScanOperatorInfo(SReadHandle* pHandle, SExecTaskInfo* pT
pInfo->vnode = pHandle->vnode;
pInfo->sContext = pHandle->sContext;
pOperator->name = "RawStreamScanOperator";
pOperator->name = "RawScanOperator";
pOperator->info = pInfo;
pOperator->pTaskInfo = pTaskInfo;
@ -4358,8 +4383,9 @@ static int32_t loadDataBlockFromOneTable2(SOperatorInfo* pOperator, STableMergeS
// currently only the tbname pseudo column
if (pTableScanInfo->pseudoSup.numOfExprs > 0) {
int32_t code = addTagPseudoColumnData(&pTableScanInfo->readHandle, pTableScanInfo->pseudoSup.pExprInfo,
pTableScanInfo->pseudoSup.numOfExprs, pBlock, GET_TASKID(pTaskInfo));
int32_t code =
addTagPseudoColumnData(&pTableScanInfo->readHandle, pTableScanInfo->pseudoSup.pExprInfo,
pTableScanInfo->pseudoSup.numOfExprs, pBlock, pBlock->info.rows, GET_TASKID(pTaskInfo));
if (code != TSDB_CODE_SUCCESS) {
T_LONG_JMP(pTaskInfo->env, code);
}
@ -4475,8 +4501,9 @@ static int32_t loadDataBlockFromOneTable(SOperatorInfo* pOperator, STableMergeSc
// currently only the tbname pseudo column
if (pTableScanInfo->pseudoSup.numOfExprs > 0) {
int32_t code = addTagPseudoColumnData(&pTableScanInfo->readHandle, pTableScanInfo->pseudoSup.pExprInfo,
pTableScanInfo->pseudoSup.numOfExprs, pBlock, GET_TASKID(pTaskInfo));
int32_t code =
addTagPseudoColumnData(&pTableScanInfo->readHandle, pTableScanInfo->pseudoSup.pExprInfo,
pTableScanInfo->pseudoSup.numOfExprs, pBlock, pBlock->info.rows, GET_TASKID(pTaskInfo));
if (code != TSDB_CODE_SUCCESS) {
T_LONG_JMP(pTaskInfo->env, code);
}
@ -4540,14 +4567,11 @@ static SSDataBlock* getTableDataBlockTemp(void* param) {
}
blockDataCleanup(pBlock);
SDataBlockInfo binfo = pBlock->info;
tsdbRetrieveDataBlockInfo(reader, &binfo);
blockDataEnsureCapacity(pBlock, binfo.rows);
pBlock->info.type = binfo.type;
pBlock->info.uid = binfo.uid;
pBlock->info.window = binfo.window;
pBlock->info.rows = binfo.rows;
int32_t rows = 0;
tsdbRetrieveDataBlockInfo(reader, &rows, &pBlock->info.uid, &pBlock->info.window);
blockDataEnsureCapacity(pBlock, rows);
pBlock->info.rows = rows;
if (pQueryCond->order == TSDB_ORDER_ASC) {
pQueryCond->twindows.skey = pBlock->info.window.ekey + 1;
@ -4603,14 +4627,11 @@ static SSDataBlock* getTableDataBlock2(void* param) {
}
blockDataCleanup(pBlock);
SDataBlockInfo binfo = pBlock->info;
tsdbRetrieveDataBlockInfo(reader, &binfo);
blockDataEnsureCapacity(pBlock, binfo.rows);
pBlock->info.type = binfo.type;
pBlock->info.uid = binfo.uid;
pBlock->info.window = binfo.window;
pBlock->info.rows = binfo.rows;
int32_t rows = 0;
tsdbRetrieveDataBlockInfo(reader, &rows, &pBlock->info.uid, &pBlock->info.window);
blockDataEnsureCapacity(pBlock, rows);
pBlock->info.rows = rows;
uint32_t status = 0;
int32_t code = loadDataBlockFromOneTable2(pOperator, pTableScanInfo, pBlock, &status);
@ -4656,14 +4677,11 @@ static SSDataBlock* getTableDataBlock(void* param) {
}
blockDataCleanup(pBlock);
SDataBlockInfo binfo = pBlock->info;
tsdbRetrieveDataBlockInfo(reader, &binfo);
blockDataEnsureCapacity(pBlock, binfo.rows);
pBlock->info.type = binfo.type;
pBlock->info.uid = binfo.uid;
pBlock->info.window = binfo.window;
pBlock->info.rows = binfo.rows;
int32_t rows = 0;
tsdbRetrieveDataBlockInfo(reader, &rows, &pBlock->info.uid, &pBlock->info.window);
blockDataEnsureCapacity(pBlock, rows);
pBlock->info.rows = rows;
uint32_t status = 0;
int32_t code = loadDataBlockFromOneTable(pOperator, pTableScanInfo, readerIdx, pBlock, &status);

2
source/libs/executor/src/timewindowoperator.c
View File

@ -3303,6 +3303,8 @@ SOperatorInfo* createStreamFinalIntervalOperatorInfo(SOperatorInfo* downstream,
// for test 315360000000
.deleteMark = 1000LL * 60LL * 60LL * 24LL * 365LL * 10LL,
// .deleteMark = INT64_MAX,
.deleteMarkSaved = 0,
.calTriggerSaved = 0,
};
ASSERT(pInfo->twAggSup.calTrigger != STREAM_TRIGGER_MAX_DELAY);
pInfo->primaryTsIndex = ((SColumnNode*)pIntervalPhyNode->window.pTspk)->slotId;

40
source/libs/function/CMakeLists.txt
View File

@ -12,6 +12,17 @@ target_include_directories(
PRIVATE "${CMAKE_CURRENT_SOURCE_DIR}/inc"
)
IF (TD_LINUX_64 AND JEMALLOC_ENABLED)
ADD_DEFINITIONS(-DTD_JEMALLOC_ENABLED -I${CMAKE_BINARY_DIR}/build/include -L${CMAKE_BINARY_DIR}/build/lib -Wl,-rpath,${CMAKE_BINARY_DIR}/build/lib -ljemalloc)
SET(LINK_JEMALLOC "-L${CMAKE_BINARY_DIR}/build/lib -ljemalloc")
ELSE ()
SET(LINK_JEMALLOC "")
ENDIF ()
IF (TD_LINUX_64 AND JEMALLOC_ENABLED)
ADD_DEPENDENCIES(function jemalloc)
ENDIF ()
target_link_libraries(
function
PRIVATE os
@ -21,7 +32,7 @@ target_link_libraries(
PRIVATE qcom
PRIVATE scalar
PRIVATE transport
PRIVATE stream
PRIVATE stream ${LINK_JEMALLOC}
PUBLIC uv_a
)
@ -37,10 +48,15 @@ target_include_directories(
"${TD_SOURCE_DIR}/include/os"
PRIVATE "${CMAKE_CURRENT_SOURCE_DIR}/inc"
)
IF (TD_LINUX_64 AND JEMALLOC_ENABLED)
ADD_DEPENDENCIES(runUdf jemalloc)
ENDIF ()
target_link_libraries(
runUdf
PUBLIC uv_a
PRIVATE os util common nodes function
PRIVATE os util common nodes function ${LINK_JEMALLOC}
)
add_library(udf1 STATIC MODULE test/udf1.c)
@ -54,8 +70,13 @@ target_include_directories(
"${TD_SOURCE_DIR}/include/os"
PRIVATE "${CMAKE_CURRENT_SOURCE_DIR}/inc"
)
IF (TD_LINUX_64 AND JEMALLOC_ENABLED)
ADD_DEPENDENCIES(udf1 jemalloc)
ENDIF ()
target_link_libraries(
udf1 PUBLIC os)
udf1 PUBLIC os ${LINK_JEMALLOC})
add_library(udf2 STATIC MODULE test/udf2.c)
target_include_directories(
@ -68,8 +89,13 @@ target_include_directories(
"${TD_SOURCE_DIR}/include/os"
PRIVATE "${CMAKE_CURRENT_SOURCE_DIR}/inc"
)
IF (TD_LINUX_64 AND JEMALLOC_ENABLED)
ADD_DEPENDENCIES(udf2 jemalloc)
ENDIF ()
target_link_libraries(
udf2 PUBLIC os
udf2 PUBLIC os ${LINK_JEMALLOC}
)
#SET(EXECUTABLE_OUTPUT_PATH ${CMAKE_BINARY_DIR}/build/bin)
@ -86,9 +112,13 @@ target_include_directories(
PRIVATE "${CMAKE_CURRENT_SOURCE_DIR}/inc"
)
IF (TD_LINUX_64 AND JEMALLOC_ENABLED)
ADD_DEPENDENCIES(udfd jemalloc)
ENDIF ()
target_link_libraries(
udfd
PUBLIC uv_a
PRIVATE os util common nodes function
PRIVATE os util common nodes function ${LINK_JEMALLOC}
)

52
source/libs/function/src/builtinsimpl.c
View File

@ -497,7 +497,7 @@ bool getCountFuncEnv(SFunctionNode* UNUSED_PARAM(pFunc), SFuncExecEnv* pEnv) {
return true;
}
static FORCE_INLINE int32_t getNumOfElems(SqlFunctionCtx* pCtx) {
static int32_t getNumOfElems(SqlFunctionCtx* pCtx) {
int32_t numOfElem = 0;
/*
@ -2194,6 +2194,56 @@ int32_t leastSQRFunction(SqlFunctionCtx* pCtx) {
break;
}
case TSDB_DATA_TYPE_UTINYINT: {
uint8_t* plist = (uint8_t*)pCol->pData;
for (int32_t i = start; i < numOfRows + pInput->startRowIndex; ++i) {
if (pCol->hasNull && colDataIsNull_f(pCol->nullbitmap, i)) {
continue;
}
numOfElem++;
LEASTSQR_CAL(param, x, plist, i, pInfo->stepVal);
}
break;
}
case TSDB_DATA_TYPE_USMALLINT: {
uint16_t* plist = (uint16_t*)pCol->pData;
for (int32_t i = start; i < numOfRows + pInput->startRowIndex; ++i) {
if (pCol->hasNull && colDataIsNull_f(pCol->nullbitmap, i)) {
continue;
}
numOfElem++;
LEASTSQR_CAL(param, x, plist, i, pInfo->stepVal);
}
break;
}
case TSDB_DATA_TYPE_UINT: {
uint32_t* plist = (uint32_t*)pCol->pData;
for (int32_t i = start; i < numOfRows + pInput->startRowIndex; ++i) {
if (pCol->hasNull && colDataIsNull_f(pCol->nullbitmap, i)) {
continue;
}
numOfElem++;
LEASTSQR_CAL(param, x, plist, i, pInfo->stepVal);
}
break;
}
case TSDB_DATA_TYPE_UBIGINT: {
uint64_t* plist = (uint64_t*)pCol->pData;
for (int32_t i = start; i < numOfRows + pInput->startRowIndex; ++i) {
if (pCol->hasNull && colDataIsNull_f(pCol->nullbitmap, i)) {
continue;
}
numOfElem++;
LEASTSQR_CAL(param, x, plist, i, pInfo->stepVal);
}
break;
}
case TSDB_DATA_TYPE_FLOAT: {
float* plist = (float*)pCol->pData;
for (int32_t i = start; i < numOfRows + pInput->startRowIndex; ++i) {

7
source/libs/function/src/functionMgt.c
View File

@ -251,6 +251,13 @@ bool fmIsSelectValueFunc(int32_t funcId) {
return FUNCTION_TYPE_SELECT_VALUE == funcMgtBuiltins[funcId].type;
}
bool fmIsGroupKeyFunc(int32_t funcId) {
if (funcId < 0 || funcId >= funcMgtBuiltinsNum) {
return false;
}
return FUNCTION_TYPE_GROUP_KEY == funcMgtBuiltins[funcId].type;
}
void fmFuncMgtDestroy() {
void* m = gFunMgtService.pFuncNameHashTable;
if (m != NULL && atomic_val_compare_exchange_ptr((void**)&gFunMgtService.pFuncNameHashTable, m, 0) == m) {

7
source/libs/function/src/tudf.c
View File

@ -124,13 +124,16 @@ static int32_t udfSpawnUdfd(SUdfdData *pData) {
char pathTaosdLdLib[512] = {0};
size_t taosdLdLibPathLen = sizeof(pathTaosdLdLib);
uv_os_getenv("LD_LIBRARY_PATH", pathTaosdLdLib, &taosdLdLibPathLen);
int ret = uv_os_getenv("LD_LIBRARY_PATH", pathTaosdLdLib, &taosdLdLibPathLen);
if (ret != UV_ENOBUFS) {
taosdLdLibPathLen = strlen(pathTaosdLdLib);
}
char udfdPathLdLib[1024] = {0};
size_t udfdLdLibPathLen = strlen(tsUdfdLdLibPath);
strncpy(udfdPathLdLib, tsUdfdLdLibPath, udfdLdLibPathLen);
udfdPathLdLib[udfdLdLibPathLen] = ':';
strncpy(udfdPathLdLib + udfdLdLibPathLen + 1, pathTaosdLdLib, sizeof(udfdPathLdLib) - udfdLdLibPathLen);
strncpy(udfdPathLdLib + udfdLdLibPathLen + 1, pathTaosdLdLib, sizeof(udfdPathLdLib) - udfdLdLibPathLen - 1);
if (udfdLdLibPathLen + taosdLdLibPathLen < 1024) {
fnInfo("udfd LD_LIBRARY_PATH: %s", udfdPathLdLib);
} else {

8
source/libs/parser/src/parTranslater.c
View File

@ -5160,7 +5160,7 @@ static int32_t translateCreateIndex(STranslateContext* pCxt, SCreateIndexStmt* p
static int32_t translateDropIndex(STranslateContext* pCxt, SDropIndexStmt* pStmt) {
SMDropSmaReq dropSmaReq = {0};
SName name;
tNameExtractFullName(toName(pCxt->pParseCxt->acctId, pCxt->pParseCxt->db, pStmt->indexName, &name), dropSmaReq.name);
tNameExtractFullName(toName(pCxt->pParseCxt->acctId, pStmt->indexDbName, pStmt->indexName, &name), dropSmaReq.name);
dropSmaReq.igNotExists = pStmt->ignoreNotExists;
return buildCmdMsg(pCxt, TDMT_MND_DROP_SMA, (FSerializeFunc)tSerializeSMDropSmaReq, &dropSmaReq);
}
@ -5281,7 +5281,7 @@ static int32_t translateCreateTopic(STranslateContext* pCxt, SCreateTopicStmt* p
code = buildCreateTopicReq(pCxt, pStmt, &createReq);
}
if (TSDB_CODE_SUCCESS == code) {
code = buildCmdMsg(pCxt, TDMT_MND_CREATE_TOPIC, (FSerializeFunc)tSerializeSCMCreateTopicReq, &createReq);
code = buildCmdMsg(pCxt, TDMT_MND_TMQ_CREATE_TOPIC, (FSerializeFunc)tSerializeSCMCreateTopicReq, &createReq);
}
tFreeSCMCreateTopicReq(&createReq);
return code;
@ -5295,7 +5295,7 @@ static int32_t translateDropTopic(STranslateContext* pCxt, SDropTopicStmt* pStmt
tNameGetFullDbName(&name, dropReq.name);
dropReq.igNotExists = pStmt->ignoreNotExists;
return buildCmdMsg(pCxt, TDMT_MND_DROP_TOPIC, (FSerializeFunc)tSerializeSMDropTopicReq, &dropReq);
return buildCmdMsg(pCxt, TDMT_MND_TMQ_DROP_TOPIC, (FSerializeFunc)tSerializeSMDropTopicReq, &dropReq);
}
static int32_t translateDropCGroup(STranslateContext* pCxt, SDropCGroupStmt* pStmt) {
@ -5307,7 +5307,7 @@ static int32_t translateDropCGroup(STranslateContext* pCxt, SDropCGroupStmt* pSt
dropReq.igNotExists = pStmt->ignoreNotExists;
strcpy(dropReq.cgroup, pStmt->cgroup);
return buildCmdMsg(pCxt, TDMT_MND_MQ_DROP_CGROUP, (FSerializeFunc)tSerializeSMDropCgroupReq, &dropReq);
return buildCmdMsg(pCxt, TDMT_MND_TMQ_DROP_CGROUP, (FSerializeFunc)tSerializeSMDropCgroupReq, &dropReq);
}
static int32_t translateAlterLocal(STranslateContext* pCxt, SAlterLocalStmt* pStmt) {

6
source/libs/planner/src/planLogicCreater.c
View File

@ -587,6 +587,10 @@ static int32_t createIndefRowsFuncLogicNode(SLogicPlanContext* pCxt, SSelectStmt
return code;
}
static bool isInterpFunc(int32_t funcId) {
return fmIsInterpFunc(funcId) || fmIsInterpPseudoColumnFunc(funcId) || fmIsGroupKeyFunc(funcId);
}
static int32_t createInterpFuncLogicNode(SLogicPlanContext* pCxt, SSelectStmt* pSelect, SLogicNode** pLogicNode) {
if (!pSelect->hasInterpFunc) {
return TSDB_CODE_SUCCESS;
@ -602,7 +606,7 @@ static int32_t createInterpFuncLogicNode(SLogicPlanContext* pCxt, SSelectStmt* p
pInterpFunc->node.resultDataOrder = pInterpFunc->node.requireDataOrder;
// interp functions and _group_key functions
int32_t code = nodesCollectFuncs(pSelect, SQL_CLAUSE_SELECT, fmIsVectorFunc, &pInterpFunc->pFuncs);
int32_t code = nodesCollectFuncs(pSelect, SQL_CLAUSE_SELECT, isInterpFunc, &pInterpFunc->pFuncs);
if (TSDB_CODE_SUCCESS == code) {
code = rewriteExprsForSelect(pInterpFunc->pFuncs, pSelect, SQL_CLAUSE_SELECT);
}

1
source/libs/qcom/src/queryUtil.c
View File

@ -456,6 +456,7 @@ int32_t cloneDbVgInfo(SDBVgInfo* pSrc, SDBVgInfo** pDst) {
(*pDst)->vgHash = taosHashInit(taosHashGetSize(pSrc->vgHash), taosGetDefaultHashFunction(TSDB_DATA_TYPE_INT), true,
HASH_ENTRY_LOCK);
if (NULL == (*pDst)->vgHash) {
taosMemoryFreeClear(*pDst);
return TSDB_CODE_TSC_OUT_OF_MEMORY;
}

20
source/libs/scalar/src/filter.c
View File

@ -1087,7 +1087,7 @@ int32_t filterAddUnitImpl(SFilterInfo *info, uint8_t optr, SFilterFieldId *left,
if (tmp == NULL) {
return TSDB_CODE_OUT_OF_MEMORY;
}
info->units = tmp;
info->units = (SFilterUnit*)tmp;
memset(info->units + psize, 0, sizeof(*info->units) * FILTER_DEFAULT_UNIT_SIZE);
}
@ -1633,12 +1633,12 @@ void filterDumpInfoToString(SFilterInfo *info, const char *msg, int32_t options)
SValueNode *var = (SValueNode *)field->desc;
SDataType *dType = &var->node.resType;
if (dType->type == TSDB_DATA_TYPE_VALUE_ARRAY) {
qDebug("VAL%d => [type:TS][val:[%" PRIi64 "] - [%" PRId64 "]]", i, *(int64_t *)field->data,
*(((int64_t *)field->data) + 1));
} else {
//if (dType->type == TSDB_DATA_TYPE_VALUE_ARRAY) {
// qDebug("VAL%d => [type:TS][val:[%" PRIi64 "] - [%" PRId64 "]]", i, *(int64_t *)field->data,
// *(((int64_t *)field->data) + 1));
//} else {
qDebug("VAL%d => [type:%d][val:%" PRIx64 "]", i, dType->type, var->datum.i); // TODO
}
//}
} else if (field->data) {
qDebug("VAL%d => [type:NIL][val:NIL]", i); // TODO
}
@ -4059,11 +4059,13 @@ bool filterExecute(SFilterInfo *info, SSDataBlock *pSrc, SColumnInfoData **p, SC
SArray *pList = taosArrayInit(1, POINTER_BYTES);
taosArrayPush(pList, &pSrc);
FLT_ERR_RET(scalarCalculate(info->sclCtx.node, pList, &output));
*p = output.columnData;
int32_t code = scalarCalculate(info->sclCtx.node, pList, &output);
taosArrayDestroy(pList);
FLT_ERR_RET(code);
*p = output.columnData;
if (output.numOfQualified == output.numOfRows) {
*pResultStatus = FILTER_RESULT_ALL_QUALIFIED;
} else if (output.numOfQualified == 0) {

4
source/libs/scalar/src/scalar.c
View File

@ -896,6 +896,10 @@ int32_t sclExecCaseWhen(SCaseWhenNode *node, SScalarCtx *ctx, SScalarParam *outp
SCL_ERR_JRET(sclGetNodeRes(pWhenThen->pWhen, ctx, &pWhen));
SCL_ERR_JRET(sclGetNodeRes(pWhenThen->pThen, ctx, &pThen));
if (NULL == pWhen || NULL == pThen) {
sclError("invalid when/then in whenThen list");
SCL_ERR_JRET(TSDB_CODE_INVALID_PARA);
}
if (pCase) {
vectorCompare(pCase, pWhen, &comp, TSDB_ORDER_ASC, OP_TYPE_EQUAL);

3
source/libs/stream/inc/streamInc.h
View File

@ -43,6 +43,9 @@ int32_t streamBroadcastToChildren(SStreamTask* pTask, const SSDataBlock* pBlock)
int32_t tEncodeStreamRetrieveReq(SEncoder* pEncoder, const SStreamRetrieveReq* pReq);
int32_t streamDispatchOneRecoverFinishReq(SStreamTask* pTask, const SStreamRecoverFinishReq* pReq, int32_t vgId,
SEpSet* pEpSet);
SStreamQueueItem* streamMergeQueueItem(SStreamQueueItem* dst, SStreamQueueItem* elem);
void streamFreeQitem(SStreamQueueItem* data);

20
source/libs/stream/src/streamDispatch.c
View File

@ -239,7 +239,7 @@ int32_t streamDispatchOneRecoverFinishReq(SStreamTask* pTask, const SStreamRecov
msg.contLen = tlen + sizeof(SMsgHead);
msg.pCont = buf;
msg.msgType = TDMT_VND_STREAM_RECOVER_FINISH;
msg.msgType = TDMT_STREAM_RECOVER_FINISH;
tmsgSendReq(pEpSet, &msg);
@ -250,7 +250,7 @@ FAIL:
return code;
}
int32_t streamDispatchOneReq(SStreamTask* pTask, const SStreamDispatchReq* pReq, int32_t vgId, SEpSet* pEpSet) {
int32_t streamDispatchOneDataReq(SStreamTask* pTask, const SStreamDispatchReq* pReq, int32_t vgId, SEpSet* pEpSet) {
void* buf = NULL;
int32_t code = -1;
SRpcMsg msg = {0};
@ -292,13 +292,19 @@ FAIL:
int32_t streamSearchAndAddBlock(SStreamTask* pTask, SStreamDispatchReq* pReqs, SSDataBlock* pDataBlock, int32_t vgSz,
int64_t groupId) {
char* ctbName;
char* ctbName = taosMemoryCalloc(1, TSDB_TABLE_FNAME_LEN);
if (ctbName == NULL) {
return -1;
}
if (pDataBlock->info.parTbName[0]) {
ctbName = taosMemoryCalloc(1, TSDB_TABLE_NAME_LEN);
snprintf(ctbName, TSDB_TABLE_NAME_LEN, "%s.%s", pTask->shuffleDispatcher.dbInfo.db, pDataBlock->info.parTbName);
} else {
ctbName = buildCtbNameByGroupId(pTask->shuffleDispatcher.stbFullName, groupId);
char* ctbShortName = buildCtbNameByGroupId(pTask->shuffleDispatcher.stbFullName, groupId);
snprintf(ctbName, TSDB_TABLE_NAME_LEN, "%s.%s", pTask->shuffleDispatcher.dbInfo.db, ctbShortName);
taosMemoryFree(ctbShortName);
}
SArray* vgInfo = pTask->shuffleDispatcher.dbInfo.pVgroupInfos;
/*uint32_t hashValue = MurmurHash3_32(ctbName, strlen(ctbName));*/
@ -365,7 +371,7 @@ int32_t streamDispatchAllBlocks(SStreamTask* pTask, const SStreamDataBlock* pDat
qDebug("dispatch from task %d (child id %d) to down stream task %d in vnode %d", pTask->taskId, pTask->selfChildId,
downstreamTaskId, vgId);
if (streamDispatchOneReq(pTask, &req, vgId, pEpSet) < 0) {
if (streamDispatchOneDataReq(pTask, &req, vgId, pEpSet) < 0) {
goto FAIL_FIXED_DISPATCH;
}
code = 0;
@ -427,7 +433,7 @@ int32_t streamDispatchAllBlocks(SStreamTask* pTask, const SStreamDataBlock* pDat
if (pReqs[i].blockNum > 0) {
// send
SVgroupInfo* pVgInfo = taosArrayGet(vgInfo, i);
if (streamDispatchOneReq(pTask, &pReqs[i], pVgInfo->vgId, &pVgInfo->epSet) < 0) {
if (streamDispatchOneDataReq(pTask, &pReqs[i], pVgInfo->vgId, &pVgInfo->epSet) < 0) {
goto FAIL_SHUFFLE_DISPATCH;
}
}

7
source/libs/stream/src/streamExec.c
View File

@ -90,6 +90,8 @@ int32_t streamScanExec(SStreamTask* pTask, int32_t batchSz) {
void* exec = pTask->exec.executor;
qSetStreamOpOpen(exec);
while (1) {
SArray* pRes = taosArrayInit(0, sizeof(SSDataBlock));
if (pRes == NULL) {
@ -127,7 +129,10 @@ int32_t streamScanExec(SStreamTask* pTask, int32_t batchSz) {
qRes->type = STREAM_INPUT__DATA_BLOCK;
qRes->blocks = pRes;
streamTaskOutput(pTask, qRes);
// TODO stream sched dispatch
if (pTask->outputType == TASK_OUTPUT__FIXED_DISPATCH || pTask->outputType == TASK_OUTPUT__SHUFFLE_DISPATCH) {
streamDispatch(pTask);
}
}
return 0;
}

22
source/libs/stream/src/streamRecover.c
View File

@ -36,6 +36,7 @@ int32_t streamSourceRecoverPrepareStep1(SStreamTask* pTask, int64_t ver) {
}
int32_t streamBuildSourceRecover1Req(SStreamTask* pTask, SStreamRecoverStep1Req* pReq) {
pReq->msgHead.vgId = pTask->nodeId;
pReq->streamId = pTask->streamId;
pReq->taskId = pTask->taskId;
return 0;
@ -44,10 +45,10 @@ int32_t streamBuildSourceRecover1Req(SStreamTask* pTask, SStreamRecoverStep1Req*
int32_t streamSourceRecoverScanStep1(SStreamTask* pTask) {
//
return streamScanExec(pTask, 100);
// TODO next: dispatch msg to launch scan step2
}
int32_t streamBuildSourceRecover2Req(SStreamTask* pTask, SStreamRecoverStep2Req* pReq) {
pReq->msgHead.vgId = pTask->nodeId;
pReq->streamId = pTask->streamId;
pReq->taskId = pTask->taskId;
return 0;
@ -64,11 +65,20 @@ int32_t streamSourceRecoverScanStep2(SStreamTask* pTask, int64_t ver) {
int32_t streamDispatchRecoverFinishReq(SStreamTask* pTask) {
SStreamRecoverFinishReq req = {
.streamId = pTask->streamId,
.taskId = pTask->taskId,
.childId = pTask->selfChildId,
};
// serialize
if (pTask->outputType == TASK_OUTPUT__FIXED_DISPATCH) {
req.taskId = pTask->fixedEpDispatcher.taskId;
streamDispatchOneRecoverFinishReq(pTask, &req, pTask->fixedEpDispatcher.nodeId, &pTask->fixedEpDispatcher.epSet);
} else if (pTask->outputType == TASK_OUTPUT__SHUFFLE_DISPATCH) {
SArray* vgInfo = pTask->shuffleDispatcher.dbInfo.pVgroupInfos;
int32_t vgSz = taosArrayGetSize(vgInfo);
for (int32_t i = 0; i < vgSz; i++) {
SVgroupInfo* pVgInfo = taosArrayGet(vgInfo, i);
req.taskId = pVgInfo->taskId;
streamDispatchOneRecoverFinishReq(pTask, &req, pVgInfo->vgId, &pVgInfo->epSet);
}
}
return 0;
}
@ -76,9 +86,9 @@ int32_t streamDispatchRecoverFinishReq(SStreamTask* pTask) {
// agg
int32_t streamAggRecoverPrepare(SStreamTask* pTask) {
void* exec = pTask->exec.executor;
if (qStreamSetParamForRecover(exec) < 0) {
return -1;
}
/*if (qStreamSetParamForRecover(exec) < 0) {*/
/*return -1;*/
/*}*/
pTask->recoverWaitingChild = taosArrayGetSize(pTask->childEpInfo);
return 0;
}
@ -96,11 +106,13 @@ int32_t streamAggChildrenRecoverFinish(SStreamTask* pTask) {
}
int32_t streamProcessRecoverFinishReq(SStreamTask* pTask, int32_t childId) {
if (pTask->taskLevel == TASK_LEVEL__AGG) {
int32_t left = atomic_sub_fetch_32(&pTask->recoverWaitingChild, 1);
ASSERT(left >= 0);
if (left == 0) {
streamAggChildrenRecoverFinish(pTask);
}
}
return 0;
}

1
source/libs/sync/inc/syncEnv.h
View File

@ -52,6 +52,7 @@ typedef struct SSyncEnv {
} SSyncEnv;
SSyncEnv* syncEnv();
bool syncIsInit();
int64_t syncNodeAdd(SSyncNode* pNode);
void syncNodeRemove(int64_t rid);

8
source/libs/sync/inc/syncInt.h
View File

@ -24,6 +24,7 @@ extern "C" {
#include "syncTools.h"
#include "taosdef.h"
#include "tlog.h"
#include "trpc.h"
#include "ttimer.h"
// clang-format off
@ -106,9 +107,9 @@ typedef struct SSyncNode {
// sync io
SWal* pWal;
const SMsgCb* msgcb;
int32_t (*FpSendMsg)(const SEpSet* pEpSet, SRpcMsg* pMsg);
int32_t (*FpEqMsg)(const SMsgCb* msgcb, SRpcMsg* pMsg);
int32_t (*FpEqCtrlMsg)(const SMsgCb* msgcb, SRpcMsg* pMsg);
int32_t (*syncSendMSg)(const SEpSet* pEpSet, SRpcMsg* pMsg);
int32_t (*syncEqMsg)(const SMsgCb* msgcb, SRpcMsg* pMsg);
int32_t (*syncEqCtrlMsg)(const SMsgCb* msgcb, SRpcMsg* pMsg);
// init internal
SNodeInfo myNodeInfo;
@ -301,7 +302,6 @@ int32_t syncGetSnapshotMeta(int64_t rid, struct SSnapshotMeta* sMeta);
int32_t syncGetSnapshotMetaByIndex(int64_t rid, SyncIndex snapshotIndex, struct SSnapshotMeta* sMeta);
bool syncNodeCanChange(SSyncNode* pSyncNode);
bool syncNodeCheckNewConfig(SSyncNode* pSyncNode, const SSyncCfg* pNewCfg);
int32_t syncNodeLeaderTransfer(SSyncNode* pSyncNode);
int32_t syncNodeLeaderTransferTo(SSyncNode* pSyncNode, SNodeInfo newLeader);

13
include/libs/sync/syncTools.h → source/libs/sync/inc/syncTools.h
View File

@ -20,30 +20,17 @@
extern "C" {
#endif
#include "trpc.h"
// ------------------ ds -------------------
typedef struct SRaftId {
SyncNodeId addr;
SyncGroupId vgId;
} SRaftId;
int32_t syncGetRespRpc(int64_t rid, uint64_t index, SRpcMsg* msg);
int32_t syncGetAndDelRespRpc(int64_t rid, uint64_t index, SRpcHandleInfo* pInfo);
void syncSetMsgCb(int64_t rid, const SMsgCb* msgcb);
char* sync2SimpleStr(int64_t rid);
// set timer ms
void setPingTimerMS(int64_t rid, int32_t pingTimerMS);
void setElectTimerMS(int64_t rid, int32_t electTimerMS);
void setHeartbeatTimerMS(int64_t rid, int32_t hbTimerMS);
// for compatibility, the same as syncPropose
int32_t syncForwardToPeer(int64_t rid, SRpcMsg* pMsg, bool isWeak);
// utils
const char* syncUtilState2String(ESyncState state);
// ------------------ for debug -------------------
void syncRpcMsgPrint(SRpcMsg* pMsg);
void syncRpcMsgPrint2(char* s, SRpcMsg* pMsg);

2
source/libs/sync/inc/syncUtil.h
View File

@ -49,7 +49,7 @@ int32_t syncUtilQuorum(int32_t replicaNum);
cJSON* syncUtilNodeInfo2Json(const SNodeInfo* p);
cJSON* syncUtilRaftId2Json(const SRaftId* p);
char* syncUtilRaftId2Str(const SRaftId* p);
const char* syncUtilState2String(ESyncState state);
const char* syncStr(ESyncState state);
bool syncUtilCanPrint(char c);
char* syncUtilprintBin(char* ptr, uint32_t len);
char* syncUtilprintBin2(char* ptr, uint32_t len);

1
source/libs/sync/src/syncEnv.c
View File

@ -80,6 +80,7 @@ SSyncNode *syncNodeAcquire(int64_t rid) {
SSyncNode *pNode = taosAcquireRef(gNodeRefId, rid);
if (pNode == NULL) {
sTrace("failed to acquire node from refId:%" PRId64, rid);
terrno = TSDB_CODE_SYN_INTERNAL_ERROR;
}
return pNode;

476
source/libs/sync/src/syncMain.c
View File

@ -13,6 +13,7 @@
* along with this program. If not, see <http://www.gnu.org/licenses/>.
*/
#define _DEFAULT_SOURCE
#include "sync.h"
#include "syncAppendEntries.h"
#include "syncAppendEntriesReply.h"
@ -34,8 +35,6 @@
#include "syncUtil.h"
#include "syncVoteMgr.h"
// ------ local funciton ---------
// enqueue message ----
static void syncNodeEqPingTimer(void* param, void* tmrId);
static void syncNodeEqElectTimer(void* param, void* tmrId);
static void syncNodeEqHeartbeatTimer(void* param, void* tmrId);
@ -44,162 +43,155 @@ static int32_t syncNodeAppendNoop(SSyncNode* ths);
static void syncNodeEqPeerHeartbeatTimer(void* param, void* tmrId);
static bool syncIsConfigChanged(const SSyncCfg* pOldCfg, const SSyncCfg* pNewCfg);
// process message ----
int32_t syncNodeOnPing(SSyncNode* ths, SyncPing* pMsg);
int32_t syncNodeOnPingReply(SSyncNode* ths, SyncPingReply* pMsg);
int64_t syncOpen(SSyncInfo* pSyncInfo) {
SSyncNode* pNode = syncNodeOpen(pSyncInfo);
if (pNode == NULL) {
SSyncNode* pSyncNode = syncNodeOpen(pSyncInfo);
if (pSyncNode == NULL) {
sError("vgId:%d, failed to open sync node", pSyncInfo->vgId);
return -1;
}
pNode->rid = syncNodeAdd(pNode);
if (pNode->rid < 0) {
syncNodeClose(pNode);
pSyncNode->rid = syncNodeAdd(pSyncNode);
if (pSyncNode->rid < 0) {
syncNodeClose(pSyncNode);
return -1;
}
return pNode->rid;
pSyncNode->pingBaseLine = pSyncInfo->pingMs;
pSyncNode->pingTimerMS = pSyncInfo->pingMs;
pSyncNode->electBaseLine = pSyncInfo->electMs;
pSyncNode->hbBaseLine = pSyncInfo->heartbeatMs;
pSyncNode->heartbeatTimerMS = pSyncInfo->heartbeatMs;
pSyncNode->msgcb = pSyncInfo->msgcb;
return pSyncNode->rid;
}
void syncStart(int64_t rid) {
SSyncNode* pNode = syncNodeAcquire(rid);
if (pNode != NULL) {
syncNodeStart(pNode);
syncNodeRelease(pNode);
SSyncNode* pSyncNode = syncNodeAcquire(rid);
if (pSyncNode != NULL) {
syncNodeStart(pSyncNode);
syncNodeRelease(pSyncNode);
}
}
void syncStop(int64_t rid) {
SSyncNode* pNode = syncNodeAcquire(rid);
if (pNode != NULL) {
syncNodeRelease(pNode);
SSyncNode* pSyncNode = syncNodeAcquire(rid);
if (pSyncNode != NULL) {
syncNodeRelease(pSyncNode);
syncNodeRemove(rid);
}
}
bool syncNodeCheckNewConfig(SSyncNode* pSyncNode, const SSyncCfg* pNewCfg) {
bool IamInNew = syncNodeInConfig(pSyncNode, pNewCfg);
if (!IamInNew) {
return false;
}
if (pNewCfg->replicaNum > pSyncNode->replicaNum + 1) {
return false;
}
if (pNewCfg->replicaNum < pSyncNode->replicaNum - 1) {
return false;
}
return true;
}
int32_t syncReconfigBuild(int64_t rid, const SSyncCfg* pNewCfg, SRpcMsg* pRpcMsg) {
SSyncNode* pSyncNode = syncNodeAcquire(rid);
if (pSyncNode == NULL) {
terrno = TSDB_CODE_SYN_INTERNAL_ERROR;
return -1;
}
ASSERT(rid == pSyncNode->rid);
int32_t ret = 0;
if (!syncNodeCheckNewConfig(pSyncNode, pNewCfg)) {
syncNodeRelease(pSyncNode);
terrno = TSDB_CODE_SYN_NEW_CONFIG_ERROR;
sError("invalid new config. vgId:%d", pSyncNode->vgId);
return -1;
}
char* newconfig = syncCfg2Str((SSyncCfg*)pNewCfg);
pRpcMsg->msgType = TDMT_SYNC_CONFIG_CHANGE;
pRpcMsg->info.noResp = 1;
pRpcMsg->contLen = strlen(newconfig) + 1;
pRpcMsg->pCont = rpcMallocCont(pRpcMsg->contLen);
snprintf(pRpcMsg->pCont, pRpcMsg->contLen, "%s", newconfig);
taosMemoryFree(newconfig);
syncNodeRelease(pSyncNode);
return ret;
static bool syncNodeCheckNewConfig(SSyncNode* pSyncNode, const SSyncCfg* pCfg) {
if (!syncNodeInConfig(pSyncNode, pCfg)) return false;
return abs(pCfg->replicaNum - pSyncNode->replicaNum) <= 1;
}
int32_t syncReconfig(int64_t rid, SSyncCfg* pNewCfg) {
SSyncNode* pSyncNode = syncNodeAcquire(rid);
if (pSyncNode == NULL) {
terrno = TSDB_CODE_SYN_INTERNAL_ERROR;
return -1;
}
ASSERT(rid == pSyncNode->rid);
if (pSyncNode == NULL) return -1;
if (!syncNodeCheckNewConfig(pSyncNode, pNewCfg)) {
syncNodeRelease(pSyncNode);
terrno = TSDB_CODE_SYN_NEW_CONFIG_ERROR;
sError("invalid new config. vgId:%d", pSyncNode->vgId);
sError("vgId:%d, failed to reconfig since invalid new config", pSyncNode->vgId);
return -1;
}
#if 0
char* newconfig = syncCfg2Str((SSyncCfg*)pNewCfg);
int32_t ret = 0;
SRpcMsg rpcMsg = {0};
rpcMsg.msgType = TDMT_SYNC_CONFIG_CHANGE;
rpcMsg.info.noResp = 1;
rpcMsg.contLen = strlen(newconfig) + 1;
rpcMsg.pCont = rpcMallocCont(rpcMsg.contLen);
snprintf(rpcMsg.pCont, rpcMsg.contLen, "%s", newconfig);
taosMemoryFree(newconfig);
ret = syncNodePropose(pSyncNode, &rpcMsg, false);
syncNodeRelease(pSyncNode);
return ret;
#else
syncNodeUpdateNewConfigIndex(pSyncNode, pNewCfg);
syncNodeDoConfigChange(pSyncNode, pNewCfg, SYNC_INDEX_INVALID);
if (pSyncNode->state == TAOS_SYNC_STATE_LEADER) {
syncNodeStopHeartbeatTimer(pSyncNode);
for (int32_t i = 0; i < TSDB_MAX_REPLICA; ++i) {
syncHbTimerInit(pSyncNode, &(pSyncNode->peerHeartbeatTimerArr[i]), (pSyncNode->replicasId)[i]);
syncHbTimerInit(pSyncNode, &pSyncNode->peerHeartbeatTimerArr[i], pSyncNode->replicasId[i]);
}
syncNodeStartHeartbeatTimer(pSyncNode);
syncNodeReplicate(pSyncNode);
}
syncNodeRelease(pSyncNode);
return 0;
#endif
}
int32_t syncProcessMsg(int64_t rid, SRpcMsg* pMsg) {
int32_t code = -1;
if (!syncIsInit()) return code;
SSyncNode* pSyncNode = syncNodeAcquire(rid);
if (pSyncNode == NULL) return code;
if (pMsg->msgType == TDMT_SYNC_HEARTBEAT) {
SyncHeartbeat* pSyncMsg = syncHeartbeatFromRpcMsg2(pMsg);
code = syncNodeOnHeartbeat(pSyncNode, pSyncMsg);
syncHeartbeatDestroy(pSyncMsg);
} else if (pMsg->msgType == TDMT_SYNC_HEARTBEAT_REPLY) {
SyncHeartbeatReply* pSyncMsg = syncHeartbeatReplyFromRpcMsg2(pMsg);
code = syncNodeOnHeartbeatReply(pSyncNode, pSyncMsg);
syncHeartbeatReplyDestroy(pSyncMsg);
} else if (pMsg->msgType == TDMT_SYNC_TIMEOUT) {
SyncTimeout* pSyncMsg = syncTimeoutFromRpcMsg2(pMsg);
code = syncNodeOnTimer(pSyncNode, pSyncMsg);
syncTimeoutDestroy(pSyncMsg);
} else if (pMsg->msgType == TDMT_SYNC_PING) {
SyncPing* pSyncMsg = syncPingFromRpcMsg2(pMsg);
code = syncNodeOnPing(pSyncNode, pSyncMsg);
syncPingDestroy(pSyncMsg);
} else if (pMsg->msgType == TDMT_SYNC_PING_REPLY) {
SyncPingReply* pSyncMsg = syncPingReplyFromRpcMsg2(pMsg);
code = syncNodeOnPingReply(pSyncNode, pSyncMsg);
syncPingReplyDestroy(pSyncMsg);
} else if (pMsg->msgType == TDMT_SYNC_CLIENT_REQUEST) {
SyncClientRequest* pSyncMsg = syncClientRequestFromRpcMsg2(pMsg);
code = syncNodeOnClientRequest(pSyncNode, pSyncMsg, NULL);
syncClientRequestDestroy(pSyncMsg);
} else if (pMsg->msgType == TDMT_SYNC_REQUEST_VOTE) {
SyncRequestVote* pSyncMsg = syncRequestVoteFromRpcMsg2(pMsg);
code = syncNodeOnRequestVote(pSyncNode, pSyncMsg);
syncRequestVoteDestroy(pSyncMsg);
} else if (pMsg->msgType == TDMT_SYNC_REQUEST_VOTE_REPLY) {
SyncRequestVoteReply* pSyncMsg = syncRequestVoteReplyFromRpcMsg2(pMsg);
code = syncNodeOnRequestVoteReply(pSyncNode, pSyncMsg);
syncRequestVoteReplyDestroy(pSyncMsg);
} else if (pMsg->msgType == TDMT_SYNC_APPEND_ENTRIES) {
SyncAppendEntries* pSyncMsg = syncAppendEntriesFromRpcMsg2(pMsg);
code = syncNodeOnAppendEntries(pSyncNode, pSyncMsg);
syncAppendEntriesDestroy(pSyncMsg);
} else if (pMsg->msgType == TDMT_SYNC_APPEND_ENTRIES_REPLY) {
SyncAppendEntriesReply* pSyncMsg = syncAppendEntriesReplyFromRpcMsg2(pMsg);
code = syncNodeOnAppendEntriesReply(pSyncNode, pSyncMsg);
syncAppendEntriesReplyDestroy(pSyncMsg);
} else if (pMsg->msgType == TDMT_SYNC_SNAPSHOT_SEND) {
SyncSnapshotSend* pSyncMsg = syncSnapshotSendFromRpcMsg2(pMsg);
code = syncNodeOnSnapshot(pSyncNode, pSyncMsg);
syncSnapshotSendDestroy(pSyncMsg);
} else if (pMsg->msgType == TDMT_SYNC_SNAPSHOT_RSP) {
SyncSnapshotRsp* pSyncMsg = syncSnapshotRspFromRpcMsg2(pMsg);
code = syncNodeOnSnapshotReply(pSyncNode, pSyncMsg);
syncSnapshotRspDestroy(pSyncMsg);
} else if (pMsg->msgType == TDMT_SYNC_LOCAL_CMD) {
SyncLocalCmd* pSyncMsg = syncLocalCmdFromRpcMsg2(pMsg);
code = syncNodeOnLocalCmd(pSyncNode, pSyncMsg);
syncLocalCmdDestroy(pSyncMsg);
} else {
sError("vgId:%d, failed to process msg:%p since invalid type:%s", pSyncNode->vgId, pMsg, TMSG_INFO(pMsg->msgType));
code = -1;
}
syncNodeRelease(pSyncNode);
return code;
}
int32_t syncLeaderTransfer(int64_t rid) {
SSyncNode* pSyncNode = syncNodeAcquire(rid);
if (pSyncNode == NULL) {
terrno = TSDB_CODE_SYN_INTERNAL_ERROR;
return -1;
}
ASSERT(rid == pSyncNode->rid);
if (pSyncNode == NULL) return -1;
int32_t ret = syncNodeLeaderTransfer(pSyncNode);
syncNodeRelease(pSyncNode);
return ret;
}
int32_t syncLeaderTransferTo(int64_t rid, SNodeInfo newLeader) {
SSyncNode* pSyncNode = syncNodeAcquire(rid);
if (pSyncNode == NULL) {
terrno = TSDB_CODE_SYN_INTERNAL_ERROR;
return -1;
}
ASSERT(rid == pSyncNode->rid);
int32_t ret = syncNodeLeaderTransferTo(pSyncNode, newLeader);
syncNodeRelease(pSyncNode);
return ret;
}
SyncIndex syncMinMatchIndex(SSyncNode* pSyncNode) {
SyncIndex minMatchIndex = SYNC_INDEX_INVALID;
@ -513,50 +505,20 @@ int32_t syncForwardToPeer(int64_t rid, SRpcMsg* pMsg, bool isWeak) {
return ret;
}
ESyncState syncGetMyRole(int64_t rid) {
SSyncNode* pSyncNode = syncNodeAcquire(rid);
if (pSyncNode == NULL) {
return TAOS_SYNC_STATE_ERROR;
}
ASSERT(rid == pSyncNode->rid);
ESyncState state = pSyncNode->state;
SSyncState syncGetState(int64_t rid) {
SSyncState state = {.state = TAOS_SYNC_STATE_ERROR};
SSyncNode* pSyncNode = syncNodeAcquire(rid);
if (pSyncNode != NULL) {
state.state = pSyncNode->state;
state.restored = pSyncNode->restoreFinish;
syncNodeRelease(pSyncNode);
}
return state;
}
bool syncIsReady(int64_t rid) {
SSyncNode* pSyncNode = syncNodeAcquire(rid);
if (pSyncNode == NULL) {
return false;
}
ASSERT(rid == pSyncNode->rid);
bool b = (pSyncNode->state == TAOS_SYNC_STATE_LEADER) && pSyncNode->restoreFinish;
syncNodeRelease(pSyncNode);
// if false, set error code
if (false == b) {
if (pSyncNode->state != TAOS_SYNC_STATE_LEADER) {
terrno = TSDB_CODE_SYN_NOT_LEADER;
} else {
terrno = TSDB_CODE_APP_NOT_READY;
}
}
return b;
}
bool syncIsRestoreFinish(int64_t rid) {
SSyncNode* pSyncNode = syncNodeAcquire(rid);
if (pSyncNode == NULL) {
return false;
}
ASSERT(rid == pSyncNode->rid);
bool b = pSyncNode->restoreFinish;
syncNodeRelease(pSyncNode);
return b;
}
#if 0
int32_t syncGetSnapshotByIndex(int64_t rid, SyncIndex index, SSnapshot* pSnapshot) {
if (index < SYNC_INDEX_BEGIN) {
return -1;
@ -626,6 +588,7 @@ int32_t syncGetSnapshotMetaByIndex(int64_t rid, SyncIndex snapshotIndex, struct
syncNodeRelease(pSyncNode);
return 0;
}
#endif
SyncIndex syncNodeGetSnapshotConfigIndex(SSyncNode* pSyncNode, SyncIndex snapshotLastApplyIndex) {
ASSERT(pSyncNode->pRaftCfg->configIndexCount >= 1);
@ -643,23 +606,7 @@ SyncIndex syncNodeGetSnapshotConfigIndex(SSyncNode* pSyncNode, SyncIndex snapsho
return lastIndex;
}
const char* syncGetMyRoleStr(int64_t rid) {
const char* s = syncUtilState2String(syncGetMyRole(rid));
return s;
}
bool syncRestoreFinish(int64_t rid) {
SSyncNode* pSyncNode = syncNodeAcquire(rid);
if (pSyncNode == NULL) {
return false;
}
ASSERT(rid == pSyncNode->rid);
bool restoreFinish = pSyncNode->restoreFinish;
syncNodeRelease(pSyncNode);
return restoreFinish;
}
#if 0
SyncTerm syncGetMyTerm(int64_t rid) {
SSyncNode* pSyncNode = syncNodeAcquire(rid);
if (pSyncNode == NULL) {
@ -727,6 +674,7 @@ void syncGetEpSet(int64_t rid, SEpSet* pEpSet) {
syncNodeRelease(pSyncNode);
}
#endif
void syncGetRetryEpSet(int64_t rid, SEpSet* pEpSet) {
SSyncNode* pSyncNode = syncNodeAcquire(rid);
@ -750,31 +698,7 @@ void syncGetRetryEpSet(int64_t rid, SEpSet* pEpSet) {
syncNodeRelease(pSyncNode);
}
int32_t syncGetRespRpc(int64_t rid, uint64_t index, SRpcMsg* msg) {
SSyncNode* pSyncNode = syncNodeAcquire(rid);
if (pSyncNode == NULL) {
return TAOS_SYNC_STATE_ERROR;
}
ASSERT(rid == pSyncNode->rid);
SRespStub stub;
int32_t ret = syncRespMgrGet(pSyncNode->pSyncRespMgr, index, &stub);
if (ret == 1) {
memcpy(msg, &(stub.rpcMsg), sizeof(SRpcMsg));
}
syncNodeRelease(pSyncNode);
return ret;
}
int32_t syncGetAndDelRespRpc(int64_t rid, uint64_t index, SRpcHandleInfo* pInfo) {
SSyncNode* pSyncNode = syncNodeAcquire(rid);
if (pSyncNode == NULL) {
return TAOS_SYNC_STATE_ERROR;
}
ASSERT(rid == pSyncNode->rid);
static void syncGetAndDelRespRpc(SSyncNode* pSyncNode, uint64_t index, SRpcHandleInfo* pInfo) {
SRespStub stub;
int32_t ret = syncRespMgrGetAndDel(pSyncNode->pSyncRespMgr, index, &stub);
if (ret == 1) {
@ -782,20 +706,6 @@ int32_t syncGetAndDelRespRpc(int64_t rid, uint64_t index, SRpcHandleInfo* pInfo)
}
sTrace("vgId:%d, get seq:%" PRIu64 " rpc handle:%p", pSyncNode->vgId, index, pInfo->handle);
syncNodeRelease(pSyncNode);
return ret;
}
void syncSetMsgCb(int64_t rid, const SMsgCb* msgcb) {
SSyncNode* pSyncNode = syncNodeAcquire(rid);
if (pSyncNode == NULL) {
sTrace("syncSetQ get pSyncNode is NULL, rid:%" PRId64, rid);
return;
}
ASSERT(rid == pSyncNode->rid);
pSyncNode->msgcb = msgcb;
syncNodeRelease(pSyncNode);
}
char* sync2SimpleStr(int64_t rid) {
@ -811,41 +721,6 @@ char* sync2SimpleStr(int64_t rid) {
return s;
}
void setPingTimerMS(int64_t rid, int32_t pingTimerMS) {
SSyncNode* pSyncNode = syncNodeAcquire(rid);
if (pSyncNode == NULL) {
return;
}
ASSERT(rid == pSyncNode->rid);
pSyncNode->pingBaseLine = pingTimerMS;
pSyncNode->pingTimerMS = pingTimerMS;
syncNodeRelease(pSyncNode);
}
void setElectTimerMS(int64_t rid, int32_t electTimerMS) {
SSyncNode* pSyncNode = syncNodeAcquire(rid);
if (pSyncNode == NULL) {
return;
}
ASSERT(rid == pSyncNode->rid);
pSyncNode->electBaseLine = electTimerMS;
syncNodeRelease(pSyncNode);
}
void setHeartbeatTimerMS(int64_t rid, int32_t hbTimerMS) {
SSyncNode* pSyncNode = syncNodeAcquire(rid);
if (pSyncNode == NULL) {
return;
}
ASSERT(rid == pSyncNode->rid);
pSyncNode->hbBaseLine = hbTimerMS;
pSyncNode->heartbeatTimerMS = hbTimerMS;
syncNodeRelease(pSyncNode);
}
int32_t syncPropose(int64_t rid, SRpcMsg* pMsg, bool isWeak) {
SSyncNode* pSyncNode = syncNodeAcquire(rid);
if (pSyncNode == NULL) {
@ -942,7 +817,7 @@ int32_t syncNodePropose(SSyncNode* pSyncNode, SRpcMsg* pMsg, bool isWeak) {
}
} else {
if (pSyncNode->FpEqMsg != NULL && (*pSyncNode->FpEqMsg)(pSyncNode->msgcb, &rpcMsg) == 0) {
if (pSyncNode->syncEqMsg != NULL && (*pSyncNode->syncEqMsg)(pSyncNode->msgcb, &rpcMsg) == 0) {
ret = 0;
} else {
ret = -1;
@ -957,7 +832,7 @@ int32_t syncNodePropose(SSyncNode* pSyncNode, SRpcMsg* pMsg, bool isWeak) {
} else {
ret = -1;
terrno = TSDB_CODE_SYN_NOT_LEADER;
sError("vgId:%d, sync propose not leader, %s, type:%s", pSyncNode->vgId, syncUtilState2String(pSyncNode->state),
sError("vgId:%d, sync propose not leader, %s, type:%s", pSyncNode->vgId, syncStr(pSyncNode->state),
TMSG_INFO(pMsg->msgType));
goto _END;
}
@ -1073,9 +948,9 @@ SSyncNode* syncNodeOpen(SSyncInfo* pSyncInfo) {
pSyncNode->pWal = pSyncInfo->pWal;
pSyncNode->msgcb = pSyncInfo->msgcb;
pSyncNode->FpSendMsg = pSyncInfo->FpSendMsg;
pSyncNode->FpEqMsg = pSyncInfo->FpEqMsg;
pSyncNode->FpEqCtrlMsg = pSyncInfo->FpEqCtrlMsg;
pSyncNode->syncSendMSg = pSyncInfo->syncSendMSg;
pSyncNode->syncEqMsg = pSyncInfo->syncEqMsg;
pSyncNode->syncEqCtrlMsg = pSyncInfo->syncEqCtrlMsg;
// init raft config
pSyncNode->pRaftCfg = raftCfgOpen(pSyncNode->configPath);
@ -1591,12 +1466,12 @@ int32_t syncNodeRestartHeartbeatTimer(SSyncNode* pSyncNode) {
int32_t syncNodeSendMsgById(const SRaftId* destRaftId, SSyncNode* pSyncNode, SRpcMsg* pMsg) {
SEpSet epSet;
syncUtilraftId2EpSet(destRaftId, &epSet);
if (pSyncNode->FpSendMsg != NULL) {
if (pSyncNode->syncSendMSg != NULL) {
// htonl
syncUtilMsgHtoN(pMsg->pCont);
pMsg->info.noResp = 1;
pSyncNode->FpSendMsg(&epSet, pMsg);
pSyncNode->syncSendMSg(&epSet, pMsg);
} else {
sError("vgId:%d, sync send msg by id error, fp-send-msg is null", pSyncNode->vgId);
return -1;
@ -1608,12 +1483,12 @@ int32_t syncNodeSendMsgById(const SRaftId* destRaftId, SSyncNode* pSyncNode, SRp
int32_t syncNodeSendMsgByInfo(const SNodeInfo* nodeInfo, SSyncNode* pSyncNode, SRpcMsg* pMsg) {
SEpSet epSet;
syncUtilnodeInfo2EpSet(nodeInfo, &epSet);
if (pSyncNode->FpSendMsg != NULL) {
if (pSyncNode->syncSendMSg != NULL) {
// htonl
syncUtilMsgHtoN(pMsg->pCont);
pMsg->info.noResp = 1;
pSyncNode->FpSendMsg(&epSet, pMsg);
pSyncNode->syncSendMSg(&epSet, pMsg);
} else {
sError("vgId:%d, sync send msg by info error, fp-send-msg is null", pSyncNode->vgId);
}
@ -1637,13 +1512,13 @@ cJSON* syncNode2Json(const SSyncNode* pSyncNode) {
snprintf(u64buf, sizeof(u64buf), "%p", pSyncNode->msgcb);
cJSON_AddStringToObject(pRoot, "rpcClient", u64buf);
snprintf(u64buf, sizeof(u64buf), "%p", pSyncNode->FpSendMsg);
cJSON_AddStringToObject(pRoot, "FpSendMsg", u64buf);
snprintf(u64buf, sizeof(u64buf), "%p", pSyncNode->syncSendMSg);
cJSON_AddStringToObject(pRoot, "syncSendMSg", u64buf);
snprintf(u64buf, sizeof(u64buf), "%p", pSyncNode->msgcb);
cJSON_AddStringToObject(pRoot, "queue", u64buf);
snprintf(u64buf, sizeof(u64buf), "%p", pSyncNode->FpEqMsg);
cJSON_AddStringToObject(pRoot, "FpEqMsg", u64buf);
snprintf(u64buf, sizeof(u64buf), "%p", pSyncNode->syncEqMsg);
cJSON_AddStringToObject(pRoot, "syncEqMsg", u64buf);
// init internal
cJSON* pMe = syncUtilNodeInfo2Json(&pSyncNode->myNodeInfo);
@ -1683,7 +1558,7 @@ cJSON* syncNode2Json(const SSyncNode* pSyncNode) {
// tla+ server vars
cJSON_AddNumberToObject(pRoot, "state", pSyncNode->state);
cJSON_AddStringToObject(pRoot, "state_str", syncUtilState2String(pSyncNode->state));
cJSON_AddStringToObject(pRoot, "state_str", syncStr(pSyncNode->state));
cJSON_AddItemToObject(pRoot, "pRaftStore", raftStore2Json(pSyncNode->pRaftStore));
// tla+ candidate vars
@ -1823,7 +1698,7 @@ inline void syncNodeEventLog(const SSyncNode* pSyncNode, char* str) {
"stgy:%d, bch:%d, "
"r-num:%d, "
"lcfg:%" PRId64 ", chging:%d, rsto:%d, dquorum:%d, elt:%" PRId64 ", hb:%" PRId64 ", %s, %s",
pSyncNode->vgId, syncUtilState2String(pSyncNode->state), str, pSyncNode->pRaftStore->currentTerm,
pSyncNode->vgId, syncStr(pSyncNode->state), str, pSyncNode->pRaftStore->currentTerm,
pSyncNode->commitIndex, logBeginIndex, logLastIndex, pSyncNode->minMatchIndex, snapshot.lastApplyIndex,
snapshot.lastApplyTerm, pSyncNode->pRaftCfg->isStandBy, pSyncNode->pRaftCfg->snapshotStrategy,
pSyncNode->pRaftCfg->batchSize, pSyncNode->replicaNum, pSyncNode->pRaftCfg->lastConfigIndex,
@ -1847,7 +1722,7 @@ inline void syncNodeEventLog(const SSyncNode* pSyncNode, char* str) {
"stgy:%d, bch:%d, "
"r-num:%d, "
"lcfg:%" PRId64 ", chging:%d, rsto:%d, dquorum:%d, elt:%" PRId64 ", hb:%" PRId64 ", %s, %s",
pSyncNode->vgId, syncUtilState2String(pSyncNode->state), str, pSyncNode->pRaftStore->currentTerm,
pSyncNode->vgId, syncStr(pSyncNode->state), str, pSyncNode->pRaftStore->currentTerm,
pSyncNode->commitIndex, logBeginIndex, logLastIndex, pSyncNode->minMatchIndex, snapshot.lastApplyIndex,
snapshot.lastApplyTerm, pSyncNode->pRaftCfg->isStandBy, pSyncNode->pRaftCfg->snapshotStrategy,
pSyncNode->pRaftCfg->batchSize, pSyncNode->replicaNum, pSyncNode->pRaftCfg->lastConfigIndex,
@ -1901,7 +1776,7 @@ inline void syncNodeErrorLog(const SSyncNode* pSyncNode, char* str) {
"stgy:%d, bch:%d, "
"r-num:%d, "
"lcfg:%" PRId64 ", chging:%d, rsto:%d, dquorum:%d, elt:%" PRId64 ", hb:%" PRId64 ", %s",
pSyncNode->vgId, syncUtilState2String(pSyncNode->state), str, pSyncNode->pRaftStore->currentTerm,
pSyncNode->vgId, syncStr(pSyncNode->state), str, pSyncNode->pRaftStore->currentTerm,
pSyncNode->commitIndex, logBeginIndex, logLastIndex, pSyncNode->minMatchIndex, snapshot.lastApplyIndex,
snapshot.lastApplyTerm, pSyncNode->pRaftCfg->isStandBy, pSyncNode->pRaftCfg->snapshotStrategy,
pSyncNode->pRaftCfg->batchSize, pSyncNode->replicaNum, pSyncNode->pRaftCfg->lastConfigIndex,
@ -1923,7 +1798,7 @@ inline void syncNodeErrorLog(const SSyncNode* pSyncNode, char* str) {
"stgy:%d, bch:%d, "
"r-num:%d, "
"lcfg:%" PRId64 ", chging:%d, rsto:%d, dquorum:%d, elt:%" PRId64 ", hb:%" PRId64 ", %s",
pSyncNode->vgId, syncUtilState2String(pSyncNode->state), str, pSyncNode->pRaftStore->currentTerm,
pSyncNode->vgId, syncStr(pSyncNode->state), str, pSyncNode->pRaftStore->currentTerm,
pSyncNode->commitIndex, logBeginIndex, logLastIndex, pSyncNode->minMatchIndex, snapshot.lastApplyIndex,
snapshot.lastApplyTerm, pSyncNode->pRaftCfg->isStandBy, pSyncNode->pRaftCfg->snapshotStrategy,
pSyncNode->pRaftCfg->batchSize, pSyncNode->replicaNum, pSyncNode->pRaftCfg->lastConfigIndex,
@ -1955,9 +1830,9 @@ inline char* syncNode2SimpleStr(const SSyncNode* pSyncNode) {
", sby:%d, "
"r-num:%d, "
"lcfg:%" PRId64 ", chging:%d, rsto:%d",
pSyncNode->vgId, syncUtilState2String(pSyncNode->state), pSyncNode->pRaftStore->currentTerm,
pSyncNode->commitIndex, logBeginIndex, logLastIndex, snapshot.lastApplyIndex, pSyncNode->pRaftCfg->isStandBy,
pSyncNode->replicaNum, pSyncNode->pRaftCfg->lastConfigIndex, pSyncNode->changing, pSyncNode->restoreFinish);
pSyncNode->vgId, syncStr(pSyncNode->state), pSyncNode->pRaftStore->currentTerm, pSyncNode->commitIndex,
logBeginIndex, logLastIndex, snapshot.lastApplyIndex, pSyncNode->pRaftCfg->isStandBy, pSyncNode->replicaNum,
pSyncNode->pRaftCfg->lastConfigIndex, pSyncNode->changing, pSyncNode->restoreFinish);
return s;
}
@ -2237,7 +2112,13 @@ void syncNodeUpdateTermWithoutStepDown(SSyncNode* pSyncNode, SyncTerm term) {
}
void syncNodeStepDown(SSyncNode* pSyncNode, SyncTerm newTerm) {
ASSERT(pSyncNode->pRaftStore->currentTerm <= newTerm);
if (pSyncNode->pRaftStore->currentTerm > newTerm) {
char logBuf[128];
snprintf(logBuf, sizeof(logBuf), "step down, ignore, new-term:%" PRIu64 ", current-term:%" PRIu64, newTerm,
pSyncNode->pRaftStore->currentTerm);
syncNodeEventLog(pSyncNode, logBuf);
return;
}
do {
char logBuf[128];
@ -2658,8 +2539,8 @@ static void syncNodeEqPingTimer(void* param, void* tmrId) {
SRpcMsg rpcMsg;
syncTimeout2RpcMsg(pSyncMsg, &rpcMsg);
syncRpcMsgLog2((char*)"==syncNodeEqPingTimer==", &rpcMsg);
if (pSyncNode->FpEqMsg != NULL) {
int32_t code = pSyncNode->FpEqMsg(pSyncNode->msgcb, &rpcMsg);
if (pSyncNode->syncEqMsg != NULL) {
int32_t code = pSyncNode->syncEqMsg(pSyncNode->msgcb, &rpcMsg);
if (code != 0) {
sError("vgId:%d, sync enqueue ping msg error, code:%d", pSyncNode->vgId, code);
rpcFreeCont(rpcMsg.pCont);
@ -2667,7 +2548,7 @@ static void syncNodeEqPingTimer(void* param, void* tmrId) {
return;
}
} else {
sTrace("syncNodeEqPingTimer pSyncNode->FpEqMsg is NULL");
sTrace("syncNodeEqPingTimer pSyncNode->syncEqMsg is NULL");
}
syncTimeoutDestroy(pSyncMsg);
@ -2692,8 +2573,8 @@ static void syncNodeEqElectTimer(void* param, void* tmrId) {
pSyncNode->vgId, pSyncNode);
SRpcMsg rpcMsg;
syncTimeout2RpcMsg(pSyncMsg, &rpcMsg);
if (pSyncNode->FpEqMsg != NULL && pSyncNode->msgcb != NULL && pSyncNode->msgcb->putToQueueFp != NULL) {
int32_t code = pSyncNode->FpEqMsg(pSyncNode->msgcb, &rpcMsg);
if (pSyncNode->syncEqMsg != NULL && pSyncNode->msgcb != NULL && pSyncNode->msgcb->putToQueueFp != NULL) {
int32_t code = pSyncNode->syncEqMsg(pSyncNode->msgcb, &rpcMsg);
if (code != 0) {
sError("vgId:%d, sync enqueue elect msg error, code:%d", pSyncNode->vgId, code);
rpcFreeCont(rpcMsg.pCont);
@ -2709,7 +2590,7 @@ static void syncNodeEqElectTimer(void* param, void* tmrId) {
} while (0);
} else {
sTrace("syncNodeEqElectTimer FpEqMsg is NULL");
sTrace("syncNodeEqElectTimer syncEqMsg is NULL");
}
syncTimeoutDestroy(pSyncMsg);
@ -2741,8 +2622,8 @@ static void syncNodeEqHeartbeatTimer(void* param, void* tmrId) {
SRpcMsg rpcMsg;
syncTimeout2RpcMsg(pSyncMsg, &rpcMsg);
syncRpcMsgLog2((char*)"==syncNodeEqHeartbeatTimer==", &rpcMsg);
if (pSyncNode->FpEqMsg != NULL) {
int32_t code = pSyncNode->FpEqMsg(pSyncNode->msgcb, &rpcMsg);
if (pSyncNode->syncEqMsg != NULL) {
int32_t code = pSyncNode->syncEqMsg(pSyncNode->msgcb, &rpcMsg);
if (code != 0) {
sError("vgId:%d, sync enqueue timer msg error, code:%d", pSyncNode->vgId, code);
rpcFreeCont(rpcMsg.pCont);
@ -2750,7 +2631,7 @@ static void syncNodeEqHeartbeatTimer(void* param, void* tmrId) {
return;
}
} else {
sError("vgId:%d, enqueue msg cb ptr (i.e. FpEqMsg) not set.", pSyncNode->vgId);
sError("vgId:%d, enqueue msg cb ptr (i.e. syncEqMsg) not set.", pSyncNode->vgId);
}
syncTimeoutDestroy(pSyncMsg);
@ -2797,8 +2678,8 @@ static void syncNodeEqPeerHeartbeatTimer(void* param, void* tmrId) {
// eq msg
#if 0
if (pSyncNode->FpEqCtrlMsg != NULL) {
int32_t code = pSyncNode->FpEqCtrlMsg(pSyncNode->msgcb, &rpcMsg);
if (pSyncNode->syncEqCtrlMsg != NULL) {
int32_t code = pSyncNode->syncEqCtrlMsg(pSyncNode->msgcb, &rpcMsg);
if (code != 0) {
sError("vgId:%d, sync ctrl enqueue timer msg error, code:%d", pSyncNode->vgId, code);
rpcFreeCont(rpcMsg.pCont);
@ -2806,7 +2687,7 @@ static void syncNodeEqPeerHeartbeatTimer(void* param, void* tmrId) {
return;
}
} else {
sError("vgId:%d, enqueue ctrl msg cb ptr (i.e. FpEqMsg) not set.", pSyncNode->vgId);
sError("vgId:%d, enqueue ctrl msg cb ptr (i.e. syncEqMsg) not set.", pSyncNode->vgId);
}
#endif
@ -2846,10 +2727,10 @@ static int32_t syncNodeEqNoop(SSyncNode* ths) {
SRpcMsg rpcMsg = {0};
syncClientRequest2RpcMsg(pSyncMsg, &rpcMsg);
if (ths->FpEqMsg != NULL) {
ths->FpEqMsg(ths->msgcb, &rpcMsg);
if (ths->syncEqMsg != NULL) {
ths->syncEqMsg(ths->msgcb, &rpcMsg);
} else {
sTrace("syncNodeEqNoop pSyncNode->FpEqMsg is NULL");
sTrace("syncNodeEqNoop pSyncNode->syncEqMsg is NULL");
}
syncEntryDestory(pEntry);
@ -2960,8 +2841,8 @@ int32_t syncNodeOnHeartbeat(SSyncNode* ths, SyncHeartbeat* pMsg) {
SRpcMsg rpcMsgLocalCmd;
syncLocalCmd2RpcMsg(pSyncMsg, &rpcMsgLocalCmd);
if (ths->FpEqMsg != NULL && ths->msgcb != NULL) {
int32_t code = ths->FpEqMsg(ths->msgcb, &rpcMsgLocalCmd);
if (ths->syncEqMsg != NULL && ths->msgcb != NULL) {
int32_t code = ths->syncEqMsg(ths->msgcb, &rpcMsgLocalCmd);
if (code != 0) {
sError("vgId:%d, sync enqueue step-down msg error, code:%d", ths->vgId, code);
rpcFreeCont(rpcMsgLocalCmd.pCont);
@ -3143,17 +3024,18 @@ int32_t syncDoLeaderTransfer(SSyncNode* ths, SRpcMsg* pRpcMsg, SSyncRaftEntry* p
}
if (ths->pFsm->FpLeaderTransferCb != NULL) {
SFsmCbMeta cbMeta = {0};
cbMeta.code = 0;
cbMeta.currentTerm = ths->pRaftStore->currentTerm;
cbMeta.flag = 0;
cbMeta.index = pEntry->index;
cbMeta.lastConfigIndex = syncNodeGetSnapshotConfigIndex(ths, cbMeta.index);
cbMeta.isWeak = pEntry->isWeak;
cbMeta.seqNum = pEntry->seqNum;
cbMeta.state = ths->state;
cbMeta.term = pEntry->term;
ths->pFsm->FpLeaderTransferCb(ths->pFsm, pRpcMsg, cbMeta);
SFsmCbMeta cbMeta = {
cbMeta.code = 0,
cbMeta.currentTerm = ths->pRaftStore->currentTerm,
cbMeta.flag = 0,
cbMeta.index = pEntry->index,
cbMeta.lastConfigIndex = syncNodeGetSnapshotConfigIndex(ths, pEntry->index),
cbMeta.isWeak = pEntry->isWeak,
cbMeta.seqNum = pEntry->seqNum,
cbMeta.state = ths->state,
cbMeta.term = pEntry->term,
};
ths->pFsm->FpLeaderTransferCb(ths->pFsm, pRpcMsg, &cbMeta);
}
syncLeaderTransferDestroy(pSyncLeaderTransfer);
@ -3331,18 +3213,20 @@ int32_t syncNodeDoCommit(SSyncNode* ths, SyncIndex beginIndex, SyncIndex endInde
// execute fsm in apply thread, or execute outside syncPropose
if (internalExecute) {
SFsmCbMeta cbMeta = {0};
cbMeta.index = pEntry->index;
cbMeta.lastConfigIndex = syncNodeGetSnapshotConfigIndex(ths, cbMeta.index);
cbMeta.isWeak = pEntry->isWeak;
cbMeta.code = 0;
cbMeta.state = ths->state;
cbMeta.seqNum = pEntry->seqNum;
cbMeta.term = pEntry->term;
cbMeta.currentTerm = ths->pRaftStore->currentTerm;
cbMeta.flag = flag;
SFsmCbMeta cbMeta = {
.index = pEntry->index,
.lastConfigIndex = syncNodeGetSnapshotConfigIndex(ths, pEntry->index),
.isWeak = pEntry->isWeak,
.code = 0,
.state = ths->state,
.seqNum = pEntry->seqNum,
.term = pEntry->term,
.currentTerm = ths->pRaftStore->currentTerm,
.flag = flag,
};
ths->pFsm->FpCommitCb(ths->pFsm, &rpcMsg, cbMeta);
syncGetAndDelRespRpc(ths, cbMeta.seqNum, &rpcMsg.info);
ths->pFsm->FpCommitCb(ths->pFsm, &rpcMsg, &cbMeta);
}
}

2
source/libs/sync/src/syncMessage.c
View File

@ -2710,7 +2710,7 @@ cJSON* syncApplyMsg2Json(const SyncApplyMsg* pMsg) {
cJSON_AddNumberToObject(pRoot, "fsmMeta.isWeak", pMsg->fsmMeta.isWeak);
cJSON_AddNumberToObject(pRoot, "fsmMeta.code", pMsg->fsmMeta.code);
cJSON_AddNumberToObject(pRoot, "fsmMeta.state", pMsg->fsmMeta.state);
cJSON_AddStringToObject(pRoot, "fsmMeta.state.str", syncUtilState2String(pMsg->fsmMeta.state));
cJSON_AddStringToObject(pRoot, "fsmMeta.state.str", syncStr(pMsg->fsmMeta.state));
snprintf(u64buf, sizeof(u64buf), "%" PRIu64, pMsg->fsmMeta.seqNum);
cJSON_AddStringToObject(pRoot, "fsmMeta.seqNum", u64buf);

21
source/libs/sync/src/syncRespMgr.c
View File

@ -145,16 +145,17 @@ void syncRespCleanByTTL(SSyncRespMgr *pObj, int64_t ttl, bool rsp) {
taosArrayPush(delIndexArray, pSeqNum);
cnt++;
SFsmCbMeta cbMeta = {0};
cbMeta.index = SYNC_INDEX_INVALID;
cbMeta.lastConfigIndex = SYNC_INDEX_INVALID;
cbMeta.isWeak = false;
cbMeta.code = TSDB_CODE_SYN_TIMEOUT;
cbMeta.state = pSyncNode->state;
cbMeta.seqNum = *pSeqNum;
cbMeta.term = SYNC_TERM_INVALID;
cbMeta.currentTerm = pSyncNode->pRaftStore->currentTerm;
cbMeta.flag = 0;
SFsmCbMeta cbMeta = {
cbMeta.index = SYNC_INDEX_INVALID,
cbMeta.lastConfigIndex = SYNC_INDEX_INVALID,
cbMeta.isWeak = false,
cbMeta.code = TSDB_CODE_SYN_TIMEOUT,
cbMeta.state = pSyncNode->state,
cbMeta.seqNum = *pSeqNum,
cbMeta.term = SYNC_TERM_INVALID,
cbMeta.currentTerm = pSyncNode->pRaftStore->currentTerm,
cbMeta.flag = 0,
};
pStub->rpcMsg.pCont = NULL;
pStub->rpcMsg.contLen = 0;

24
source/libs/sync/src/syncUtil.c
View File

@ -176,30 +176,6 @@ char* syncUtilRaftId2Str(const SRaftId* p) {
return serialized;
}
const char* syncUtilState2String(ESyncState state) {
/*
if (state == TAOS_SYNC_STATE_FOLLOWER) {
return "TAOS_SYNC_STATE_FOLLOWER";
} else if (state == TAOS_SYNC_STATE_CANDIDATE) {
return "TAOS_SYNC_STATE_CANDIDATE";
} else if (state == TAOS_SYNC_STATE_LEADER) {
return "TAOS_SYNC_STATE_LEADER";
} else {
return "TAOS_SYNC_STATE_UNKNOWN";
}
*/
if (state == TAOS_SYNC_STATE_FOLLOWER) {
return "follower";
} else if (state == TAOS_SYNC_STATE_CANDIDATE) {
return "candidate";
} else if (state == TAOS_SYNC_STATE_LEADER) {
return "leader";
} else {
return "state_error";
}
}
bool syncUtilCanPrint(char c) {
if (c >= 32 && c <= 126) {
return true;

21
source/libs/sync/test/syncConfigChangeSnapshotTest.cpp
View File

@ -47,8 +47,8 @@ void CommitCb(struct SSyncFSM* pFsm, const SRpcMsg* pMsg, SFsmCbMeta cbMeta) {
snprintf(logBuf, sizeof(logBuf),
"==callback== ==CommitCb== pFsm:%p, index:%" PRId64 ", isWeak:%d, code:%d, state:%d %s, flag:%" PRIu64
", term:%" PRIu64 " \n",
pFsm, cbMeta.index, cbMeta.isWeak, cbMeta.code, cbMeta.state, syncUtilState2String(cbMeta.state),
cbMeta.flag, cbMeta.term);
pFsm, cbMeta.index, cbMeta.isWeak, cbMeta.code, cbMeta.state, syncStr(cbMeta.state), cbMeta.flag,
cbMeta.term);
syncRpcMsgLog2(logBuf, (SRpcMsg*)pMsg);
} else {
sTrace("==callback== ==CommitCb== do not apply again %" PRId64, cbMeta.index);
@ -57,10 +57,10 @@ void CommitCb(struct SSyncFSM* pFsm, const SRpcMsg* pMsg, SFsmCbMeta cbMeta) {
void PreCommitCb(struct SSyncFSM* pFsm, const SRpcMsg* pMsg, SFsmCbMeta cbMeta) {
char logBuf[256] = {0};
snprintf(
logBuf, sizeof(logBuf),
"==callback== ==PreCommitCb== pFsm:%p, index:%" PRId64 ", isWeak:%d, code:%d, state:%d %s flag:%" PRIu64 "\n",
pFsm, cbMeta.index, cbMeta.isWeak, cbMeta.code, cbMeta.state, syncUtilState2String(cbMeta.state), cbMeta.flag);
snprintf(logBuf, sizeof(logBuf),
"==callback== ==PreCommitCb== pFsm:%p, index:%" PRId64 ", isWeak:%d, code:%d, state:%d %s flag:%" PRIu64
"\n",
pFsm, cbMeta.index, cbMeta.isWeak, cbMeta.code, cbMeta.state, syncStr(cbMeta.state), cbMeta.flag);
syncRpcMsgLog2(logBuf, (SRpcMsg*)pMsg);
}
@ -68,8 +68,7 @@ void RollBackCb(struct SSyncFSM* pFsm, const SRpcMsg* pMsg, SFsmCbMeta cbMeta) {
char logBuf[256];
snprintf(logBuf, sizeof(logBuf),
"==callback== ==RollBackCb== pFsm:%p, index:%" PRId64 ", isWeak:%d, code:%d, state:%d %s flag:%" PRIu64 "\n",
pFsm, cbMeta.index, cbMeta.isWeak, cbMeta.code, cbMeta.state, syncUtilState2String(cbMeta.state),
cbMeta.flag);
pFsm, cbMeta.index, cbMeta.isWeak, cbMeta.code, cbMeta.state, syncStr(cbMeta.state), cbMeta.flag);
syncRpcMsgLog2(logBuf, (SRpcMsg*)pMsg);
}
@ -158,6 +157,7 @@ SSyncFSM* createFsm() {
SSyncFSM* pFsm = (SSyncFSM*)taosMemoryMalloc(sizeof(SSyncFSM));
memset(pFsm, 0, sizeof(*pFsm));
#if 0
pFsm->FpCommitCb = CommitCb;
pFsm->FpPreCommitCb = PreCommitCb;
pFsm->FpRollBackCb = RollBackCb;
@ -172,6 +172,7 @@ SSyncFSM* createFsm() {
pFsm->FpSnapshotDoWrite = SnapshotDoWrite;
pFsm->FpReConfigCb = ReConfigCb;
#endif
return pFsm;
}
@ -195,8 +196,8 @@ int64_t createSyncNode(int32_t replicaNum, int32_t myIndex, int32_t vgId, SWal*
SSyncInfo syncInfo;
syncInfo.vgId = vgId;
syncInfo.msgcb = &gSyncIO->msgcb;
syncInfo.FpSendMsg = syncIOSendMsg;
syncInfo.FpEqMsg = syncIOEqMsg;
syncInfo.syncSendMSg = syncIOSendMsg;
syncInfo.syncEqMsg = syncIOEqMsg;
syncInfo.pFsm = createFsm();
snprintf(syncInfo.path, sizeof(syncInfo.path), "%s_sync_replica%d_index%d", path, replicaNum, myIndex);
syncInfo.pWal = pWal;

28
source/libs/sync/test/syncConfigChangeTest.cpp
View File

@ -33,7 +33,7 @@ void init() {
void cleanup() { walCleanUp(); }
void CommitCb(struct SSyncFSM* pFsm, const SRpcMsg* pMsg, SFsmCbMeta cbMeta) {
void CommitCb(const struct SSyncFSM* pFsm, const SRpcMsg* pMsg, const SFsmCbMeta cbMeta) {
SyncIndex beginIndex = SYNC_INDEX_INVALID;
if (pFsm->FpGetSnapshotInfo != NULL) {
SSnapshot snapshot;
@ -45,33 +45,31 @@ void CommitCb(struct SSyncFSM* pFsm, const SRpcMsg* pMsg, SFsmCbMeta cbMeta) {
char logBuf[256] = {0};
snprintf(logBuf, sizeof(logBuf),
"==callback== ==CommitCb== pFsm:%p, index:%" PRId64 ", isWeak:%d, code:%d, state:%d %s flag:%" PRIu64 "\n",
pFsm, cbMeta.index, cbMeta.isWeak, cbMeta.code, cbMeta.state, syncUtilState2String(cbMeta.state),
cbMeta.flag);
pFsm, cbMeta.index, cbMeta.isWeak, cbMeta.code, cbMeta.state, syncStr(cbMeta.state), cbMeta.flag);
syncRpcMsgLog2(logBuf, (SRpcMsg*)pMsg);
} else {
sTrace("==callback== ==CommitCb== do not apply again %" PRId64, cbMeta.index);
}
}
void PreCommitCb(struct SSyncFSM* pFsm, const SRpcMsg* pMsg, SFsmCbMeta cbMeta) {
void PreCommitCb(const struct SSyncFSM* pFsm, const SRpcMsg* pMsg, SFsmCbMeta cbMeta) {
char logBuf[256] = {0};
snprintf(
logBuf, sizeof(logBuf),
"==callback== ==PreCommitCb== pFsm:%p, index:%" PRId64 ", isWeak:%d, code:%d, state:%d %s flag:%" PRIu64 "\n",
pFsm, cbMeta.index, cbMeta.isWeak, cbMeta.code, cbMeta.state, syncUtilState2String(cbMeta.state), cbMeta.flag);
snprintf(logBuf, sizeof(logBuf),
"==callback== ==PreCommitCb== pFsm:%p, index:%" PRId64 ", isWeak:%d, code:%d, state:%d %s flag:%" PRIu64
"\n",
pFsm, cbMeta.index, cbMeta.isWeak, cbMeta.code, cbMeta.state, syncStr(cbMeta.state), cbMeta.flag);
syncRpcMsgLog2(logBuf, (SRpcMsg*)pMsg);
}
void RollBackCb(struct SSyncFSM* pFsm, const SRpcMsg* pMsg, SFsmCbMeta cbMeta) {
void RollBackCb(const struct SSyncFSM* pFsm, const SRpcMsg* pMsg, SFsmCbMeta cbMeta) {
char logBuf[256];
snprintf(logBuf, sizeof(logBuf),
"==callback== ==RollBackCb== pFsm:%p, index:%" PRId64 ", isWeak:%d, code:%d, state:%d %s flag:%" PRIu64 "\n",
pFsm, cbMeta.index, cbMeta.isWeak, cbMeta.code, cbMeta.state, syncUtilState2String(cbMeta.state),
cbMeta.flag);
pFsm, cbMeta.index, cbMeta.isWeak, cbMeta.code, cbMeta.state, syncStr(cbMeta.state), cbMeta.flag);
syncRpcMsgLog2(logBuf, (SRpcMsg*)pMsg);
}
int32_t GetSnapshotCb(struct SSyncFSM* pFsm, SSnapshot* pSnapshot) {
int32_t GetSnapshotCb(const struct SSyncFSM* pFsm, SSnapshot* pSnapshot) {
pSnapshot->data = NULL;
pSnapshot->lastApplyIndex = gSnapshotLastApplyIndex;
pSnapshot->lastApplyTerm = 100;
@ -89,6 +87,7 @@ SSyncFSM* createFsm() {
SSyncFSM* pFsm = (SSyncFSM*)taosMemoryMalloc(sizeof(SSyncFSM));
memset(pFsm, 0, sizeof(*pFsm));
#if 0
pFsm->FpCommitCb = CommitCb;
pFsm->FpPreCommitCb = PreCommitCb;
pFsm->FpRollBackCb = RollBackCb;
@ -97,6 +96,7 @@ SSyncFSM* createFsm() {
pFsm->FpRestoreFinishCb = RestoreFinishCb;
pFsm->FpReConfigCb = ReConfigCb;
#endif
return pFsm;
}
@ -120,8 +120,8 @@ int64_t createSyncNode(int32_t replicaNum, int32_t myIndex, int32_t vgId, SWal*
SSyncInfo syncInfo;
syncInfo.vgId = vgId;
syncInfo.msgcb = &gSyncIO->msgcb;
syncInfo.FpSendMsg = syncIOSendMsg;
syncInfo.FpEqMsg = syncIOEqMsg;
syncInfo.syncSendMSg = syncIOSendMsg;
syncInfo.syncEqMsg = syncIOEqMsg;
syncInfo.pFsm = createFsm();
snprintf(syncInfo.path, sizeof(syncInfo.path), "%s_sync_replica%d_index%d", path, replicaNum, myIndex);
syncInfo.pWal = pWal;

4
source/libs/sync/test/syncElectTest.cpp
View File

@ -45,8 +45,8 @@ SSyncNode* createSyncNode(int32_t replicaNum, int32_t myIndex, int32_t vgId, SWa
SSyncInfo syncInfo;
syncInfo.vgId = vgId;
syncInfo.msgcb = &gSyncIO->msgcb;
syncInfo.FpSendMsg = syncIOSendMsg;
syncInfo.FpEqMsg = syncIOEqMsg;
syncInfo.syncSendMSg = syncIOSendMsg;
syncInfo.syncEqMsg = syncIOEqMsg;
syncInfo.pFsm = NULL;
snprintf(syncInfo.path, sizeof(syncInfo.path), "%s_sync_replica%d_index%d", path, replicaNum, myIndex);
syncInfo.pWal = pWal;

4
source/libs/sync/test/syncEncodeTest.cpp
View File

@ -32,8 +32,8 @@ SSyncNode *pSyncNode;
SSyncNode *syncNodeInit() {
syncInfo.vgId = 1234;
syncInfo.msgcb = &gSyncIO->msgcb;
syncInfo.FpSendMsg = syncIOSendMsg;
syncInfo.FpEqMsg = syncIOEqMsg;
syncInfo.syncSendMSg = syncIOSendMsg;
syncInfo.syncEqMsg = syncIOEqMsg;
syncInfo.pFsm = pFsm;
snprintf(syncInfo.path, sizeof(syncInfo.path), "%s", "./");

6
source/libs/sync/test/syncEnqTest.cpp
View File

@ -25,8 +25,8 @@ SSyncFSM* pFsm;
SSyncNode* syncNodeInit() {
syncInfo.vgId = 1234;
syncInfo.FpSendMsg = syncIOSendMsg;
syncInfo.FpEqMsg = syncIOEqMsg;
syncInfo.syncSendMSg = syncIOSendMsg;
syncInfo.syncEqMsg = syncIOEqMsg;
syncInfo.pFsm = pFsm;
snprintf(syncInfo.path, sizeof(syncInfo.path), "%s", "./");
@ -97,7 +97,7 @@ int main(int argc, char** argv) {
SyncPingReply* pSyncMsg = syncPingReplyBuild2(&pSyncNode->myRaftId, &pSyncNode->myRaftId, 1000, "syncEnqTest");
SRpcMsg rpcMsg;
syncPingReply2RpcMsg(pSyncMsg, &rpcMsg);
pSyncNode->FpEqMsg(pSyncNode->msgcb, &rpcMsg);
pSyncNode->syncEqMsg(pSyncNode->msgcb, &rpcMsg);
taosMsleep(1000);
}

6
source/libs/sync/test/syncIOSendMsgTest.cpp
View File

@ -26,8 +26,8 @@ SSyncFSM* pFsm;
SSyncNode* syncNodeInit() {
syncInfo.vgId = 1234;
syncInfo.msgcb = &gSyncIO->msgcb;
syncInfo.FpSendMsg = syncIOSendMsg;
syncInfo.FpEqMsg = syncIOEqMsg;
syncInfo.syncSendMSg = syncIOSendMsg;
syncInfo.syncEqMsg = syncIOEqMsg;
syncInfo.pFsm = pFsm;
snprintf(syncInfo.path, sizeof(syncInfo.path), "%s", "./");
@ -103,7 +103,7 @@ int main(int argc, char** argv) {
SEpSet epSet;
syncUtilnodeInfo2EpSet(&pSyncNode->myNodeInfo, &epSet);
rpcMsg.info.noResp = 1;
pSyncNode->FpSendMsg(&epSet, &rpcMsg);
pSyncNode->syncSendMSg(&epSet, &rpcMsg);
taosMsleep(1000);
}

4
source/libs/sync/test/syncInitTest.cpp
View File

@ -26,8 +26,8 @@ SSyncFSM* pFsm;
SSyncNode* syncNodeInit() {
syncInfo.vgId = 1234;
syncInfo.msgcb = &gSyncIO->msgcb;
syncInfo.FpSendMsg = syncIOSendMsg;
syncInfo.FpEqMsg = syncIOEqMsg;
syncInfo.syncSendMSg = syncIOSendMsg;
syncInfo.syncEqMsg = syncIOEqMsg;
syncInfo.pFsm = pFsm;
snprintf(syncInfo.path, sizeof(syncInfo.path), "%s", "./sync_init_test");

4
source/libs/sync/test/syncPingSelfTest.cpp
View File

@ -26,8 +26,8 @@ SSyncFSM* pFsm;
SSyncNode* syncNodeInit() {
syncInfo.vgId = 1234;
syncInfo.msgcb = &gSyncIO->msgcb;
syncInfo.FpSendMsg = syncIOSendMsg;
syncInfo.FpEqMsg = syncIOEqMsg;
syncInfo.syncSendMSg = syncIOSendMsg;
syncInfo.syncEqMsg = syncIOEqMsg;
syncInfo.pFsm = pFsm;
snprintf(syncInfo.path, sizeof(syncInfo.path), "%s", "./");

4
source/libs/sync/test/syncPingTimerTest.cpp
View File

@ -26,8 +26,8 @@ SSyncFSM* pFsm;
SSyncNode* syncNodeInit() {
syncInfo.vgId = 1234;
syncInfo.msgcb = &gSyncIO->msgcb;
syncInfo.FpSendMsg = syncIOSendMsg;
syncInfo.FpEqMsg = syncIOEqMsg;
syncInfo.syncSendMSg = syncIOSendMsg;
syncInfo.syncEqMsg = syncIOEqMsg;
syncInfo.pFsm = pFsm;
snprintf(syncInfo.path, sizeof(syncInfo.path), "%s", "./");

4
source/libs/sync/test/syncPingTimerTest2.cpp
View File

@ -26,8 +26,8 @@ SSyncFSM* pFsm;
SSyncNode* syncNodeInit() {
syncInfo.vgId = 1234;
syncInfo.msgcb = &gSyncIO->msgcb;
syncInfo.FpSendMsg = syncIOSendMsg;
syncInfo.FpEqMsg = syncIOEqMsg;
syncInfo.syncSendMSg = syncIOSendMsg;
syncInfo.syncEqMsg = syncIOEqMsg;
syncInfo.pFsm = pFsm;
snprintf(syncInfo.path, sizeof(syncInfo.path), "%s", "./");

2
source/libs/sync/test/syncRaftLogTest2.cpp
View File

@ -25,7 +25,7 @@ const char* pWalPath = "./syncLogStoreTest_wal";
SyncIndex gSnapshotLastApplyIndex;
SyncIndex gSnapshotLastApplyTerm;
int32_t GetSnapshotCb(struct SSyncFSM* pFsm, SSnapshot* pSnapshot) {
int32_t GetSnapshotCb(const struct SSyncFSM* pFsm, SSnapshot* pSnapshot) {
pSnapshot->data = NULL;
pSnapshot->lastApplyIndex = gSnapshotLastApplyIndex;
pSnapshot->lastApplyTerm = gSnapshotLastApplyTerm;

2
source/libs/sync/test/syncRaftLogTest3.cpp
View File

@ -27,7 +27,7 @@ const char* pWalPath = "./syncLogStoreTest_wal";
SyncIndex gSnapshotLastApplyIndex;
SyncIndex gSnapshotLastApplyTerm;
int32_t GetSnapshotCb(struct SSyncFSM* pFsm, SSnapshot* pSnapshot) {
int32_t GetSnapshotCb(const struct SSyncFSM* pFsm, SSnapshot* pSnapshot) {
pSnapshot->data = NULL;
pSnapshot->lastApplyIndex = gSnapshotLastApplyIndex;
pSnapshot->lastApplyTerm = gSnapshotLastApplyTerm;

Some files were not shown because too many files have changed in this diff Show More