homework-jianmu/source/libs/sync/src/syncReplication.c

261 lines
8.5 KiB
C

/*
* Copyright (c) 2019 TAOS Data, Inc. <jhtao@taosdata.com>
*
* This program is free software: you can use, redistribute, and/or modify
* it under the terms of the GNU Affero General Public License, version 3
* or later ("AGPL"), as published by the Free Software Foundation.
*
* This program is distributed in the hope that it will be useful, but WITHOUT
* ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or
* FITNESS FOR A PARTICULAR PURPOSE.
*
* You should have received a copy of the GNU Affero General Public License
* along with this program. If not, see <http://www.gnu.org/licenses/>.
*/
#define _DEFAULT_SOURCE
#include "syncReplication.h"
#include "syncIndexMgr.h"
#include "syncPipeline.h"
#include "syncRaftEntry.h"
#include "syncRaftStore.h"
#include "syncUtil.h"
// TLA+ Spec
// AppendEntries(i, j) ==
// /\ i /= j
// /\ state[i] = Leader
// /\ LET prevLogIndex == nextIndex[i][j] - 1
// prevLogTerm == IF prevLogIndex > 0 THEN
// log[i][prevLogIndex].term
// ELSE
// 0
// \* Send up to 1 entry, constrained by the end of the log.
// lastEntry == Min({Len(log[i]), nextIndex[i][j]})
// entries == SubSeq(log[i], nextIndex[i][j], lastEntry)
// IN Send([mtype |-> AppendEntriesRequest,
// mterm |-> currentTerm[i],
// mprevLogIndex |-> prevLogIndex,
// mprevLogTerm |-> prevLogTerm,
// mentries |-> entries,
// \* mlog is used as a history variable for the proof.
// \* It would not exist in a real implementation.
// mlog |-> log[i],
// mcommitIndex |-> Min({commitIndex[i], lastEntry}),
// msource |-> i,
// mdest |-> j])
// /\ UNCHANGED <<serverVars, candidateVars, leaderVars, logVars>>
int32_t syncNodeMaybeSendAppendEntries(SSyncNode* pSyncNode, const SRaftId* destRaftId, SRpcMsg* pRpcMsg);
int32_t syncNodeReplicateOne(SSyncNode* pSyncNode, SRaftId* pDestId, bool snapshot) {
ASSERT(false && "deprecated");
// next index
SyncIndex nextIndex = syncIndexMgrGetIndex(pSyncNode->pNextIndex, pDestId);
if (snapshot) {
// maybe start snapshot
SyncIndex logStartIndex = pSyncNode->pLogStore->syncLogBeginIndex(pSyncNode->pLogStore);
SyncIndex logEndIndex = pSyncNode->pLogStore->syncLogEndIndex(pSyncNode->pLogStore);
if (nextIndex < logStartIndex || nextIndex - 1 > logEndIndex) {
sNTrace(pSyncNode, "maybe start snapshot for next-index:%" PRId64 ", start:%" PRId64 ", end:%" PRId64, nextIndex,
logStartIndex, logEndIndex);
// start snapshot
int32_t code = syncNodeStartSnapshot(pSyncNode, pDestId);
}
}
// pre index, pre term
SyncIndex preLogIndex = syncNodeGetPreIndex(pSyncNode, nextIndex);
SyncTerm preLogTerm = syncNodeGetPreTerm(pSyncNode, nextIndex);
// prepare entry
SRpcMsg rpcMsg = {0};
SyncAppendEntries* pMsg = NULL;
SSyncRaftEntry* pEntry = NULL;
SLRUCache* pCache = pSyncNode->pLogStore->pCache;
LRUHandle* h = taosLRUCacheLookup(pCache, &nextIndex, sizeof(nextIndex));
int32_t code = 0;
if (h) {
pEntry = (SSyncRaftEntry*)taosLRUCacheValue(pCache, h);
code = 0;
pSyncNode->pLogStore->cacheHit++;
sNTrace(pSyncNode, "hit cache index:%" PRId64 ", bytes:%u, %p", nextIndex, pEntry->bytes, pEntry);
} else {
pSyncNode->pLogStore->cacheMiss++;
sNTrace(pSyncNode, "miss cache index:%" PRId64, nextIndex);
code = pSyncNode->pLogStore->syncLogGetEntry(pSyncNode->pLogStore, nextIndex, &pEntry);
}
if (code == 0) {
ASSERT(pEntry != NULL);
code = syncBuildAppendEntries(&rpcMsg, (int32_t)(pEntry->bytes), pSyncNode->vgId);
ASSERT(code == 0);
pMsg = rpcMsg.pCont;
memcpy(pMsg->data, pEntry, pEntry->bytes);
} else {
if (terrno == TSDB_CODE_WAL_LOG_NOT_EXIST) {
// no entry in log
code = syncBuildAppendEntries(&rpcMsg, 0, pSyncNode->vgId);
ASSERT(code == 0);
pMsg = rpcMsg.pCont;
} else {
sNError(pSyncNode, "replicate to dnode:%d error, next-index:%" PRId64, DID(pDestId), nextIndex);
return -1;
}
}
if (h) {
taosLRUCacheRelease(pCache, h, false);
} else {
syncEntryDestroy(pEntry);
}
// prepare msg
ASSERT(pMsg != NULL);
pMsg->srcId = pSyncNode->myRaftId;
pMsg->destId = *pDestId;
pMsg->term = pSyncNode->raftStore.currentTerm;
pMsg->prevLogIndex = preLogIndex;
pMsg->prevLogTerm = preLogTerm;
pMsg->commitIndex = pSyncNode->commitIndex;
pMsg->privateTerm = 0;
// pMsg->privateTerm = syncIndexMgrGetTerm(pSyncNode->pNextIndex, pDestId);
// send msg
syncNodeMaybeSendAppendEntries(pSyncNode, pDestId, &rpcMsg);
return 0;
}
int32_t syncNodeReplicate(SSyncNode* pNode) {
SSyncLogBuffer* pBuf = pNode->pLogBuf;
taosThreadMutexLock(&pBuf->mutex);
int32_t ret = syncNodeReplicateWithoutLock(pNode);
taosThreadMutexUnlock(&pBuf->mutex);
return ret;
}
int32_t syncNodeReplicateWithoutLock(SSyncNode* pNode) {
if (pNode->state != TAOS_SYNC_STATE_LEADER || pNode->replicaNum == 1) {
return -1;
}
for (int32_t i = 0; i < pNode->replicaNum; i++) {
if (syncUtilSameId(&pNode->replicasId[i], &pNode->myRaftId)) {
continue;
}
SSyncLogReplMgr* pMgr = pNode->logReplMgrs[i];
(void)syncLogReplMgrReplicateOnce(pMgr, pNode);
}
return 0;
}
int32_t syncNodeReplicateOld(SSyncNode* pSyncNode) {
if (pSyncNode->state != TAOS_SYNC_STATE_LEADER) {
return -1;
}
sNTrace(pSyncNode, "do replicate");
int32_t ret = 0;
for (int i = 0; i < pSyncNode->peersNum; ++i) {
SRaftId* pDestId = &(pSyncNode->peersId[i]);
ret = syncNodeReplicateOne(pSyncNode, pDestId, true);
if (ret != 0) {
sError("vgId:%d, do append entries error for dnode:%d", pSyncNode->vgId, DID(pDestId));
}
}
return 0;
}
int32_t syncNodeSendAppendEntries(SSyncNode* pSyncNode, const SRaftId* destRaftId, SRpcMsg* pRpcMsg) {
SyncAppendEntries* pMsg = pRpcMsg->pCont;
pMsg->destId = *destRaftId;
syncNodeSendMsgById(destRaftId, pSyncNode, pRpcMsg);
return 0;
}
int32_t syncNodeSendAppendEntriesOld(SSyncNode* pSyncNode, const SRaftId* destRaftId, SRpcMsg* pRpcMsg) {
int32_t ret = 0;
SyncAppendEntries* pMsg = pRpcMsg->pCont;
if (pMsg == NULL) {
sError("vgId:%d, sync-append-entries msg is NULL", pSyncNode->vgId);
return 0;
}
SPeerState* pState = syncNodeGetPeerState(pSyncNode, destRaftId);
if (pState == NULL) {
sError("vgId:%d, replica maybe dropped", pSyncNode->vgId);
return 0;
}
// save index, otherwise pMsg will be free by rpc
SyncIndex saveLastSendIndex = pState->lastSendIndex;
bool update = false;
if (pMsg->dataLen > 0) {
saveLastSendIndex = pMsg->prevLogIndex + 1;
update = true;
}
syncLogSendAppendEntries(pSyncNode, pMsg, "");
syncNodeSendMsgById(destRaftId, pSyncNode, pRpcMsg);
if (update) {
pState->lastSendIndex = saveLastSendIndex;
pState->lastSendTime = taosGetTimestampMs();
}
return ret;
}
int32_t syncNodeMaybeSendAppendEntries(SSyncNode* pSyncNode, const SRaftId* destRaftId, SRpcMsg* pRpcMsg) {
int32_t ret = 0;
SyncAppendEntries* pMsg = pRpcMsg->pCont;
if (syncNodeNeedSendAppendEntries(pSyncNode, destRaftId, pMsg)) {
ret = syncNodeSendAppendEntries(pSyncNode, destRaftId, pRpcMsg);
} else {
sNTrace(pSyncNode, "do not repcate to dnode:%d for index:%" PRId64, DID(destRaftId), pMsg->prevLogIndex + 1);
rpcFreeCont(pRpcMsg->pCont);
}
return ret;
}
int32_t syncNodeSendHeartbeat(SSyncNode* pSyncNode, const SRaftId* destId, SRpcMsg* pMsg) {
return syncNodeSendMsgById(destId, pSyncNode, pMsg);
}
int32_t syncNodeHeartbeatPeers(SSyncNode* pSyncNode) {
int64_t ts = taosGetTimestampMs();
for (int32_t i = 0; i < pSyncNode->peersNum; ++i) {
SRpcMsg rpcMsg = {0};
if (syncBuildHeartbeat(&rpcMsg, pSyncNode->vgId) != 0) {
sError("vgId:%d, build sync-heartbeat error", pSyncNode->vgId);
continue;
}
SyncHeartbeat* pSyncMsg = rpcMsg.pCont;
pSyncMsg->srcId = pSyncNode->myRaftId;
pSyncMsg->destId = pSyncNode->peersId[i];
pSyncMsg->term = pSyncNode->raftStore.currentTerm;
pSyncMsg->commitIndex = pSyncNode->commitIndex;
pSyncMsg->minMatchIndex = syncMinMatchIndex(pSyncNode);
pSyncMsg->privateTerm = 0;
pSyncMsg->timeStamp = ts;
// send msg
syncLogSendHeartbeat(pSyncNode, pSyncMsg, true, 0, 0);
syncNodeSendHeartbeat(pSyncNode, &pSyncMsg->destId, &rpcMsg);
}
return 0;
}