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

Merge pull request #18544 from taosdata/feature/3.0_mhli 

refactor(sync): add trace log for elect count, become leader count, c…
This commit is contained in:
Shengliang Guan 2022-11-29 16:28:29 +08:00 committed by GitHub
parent 10c8a60147 198e18c348
commit a7cd465a6a
No known key found for this signature in database
GPG Key ID: 4AEE18F83AFDEB23
10 changed files with 58 additions and 24 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

5
include/libs/sync/sync.h
View File

@ -162,7 +162,10 @@ typedef struct SSyncFSM {
// SWal implements it // SWal implements it
typedef struct SSyncLogStore { typedef struct SSyncLogStore {
SLRUCache* pCache; SLRUCache* pCache;
void* data; int32_t cacheHit;
int32_t cacheMiss;
void* data;
int32_t (*syncLogUpdateCommitIndex)(struct SSyncLogStore* pLogStore, SyncIndex index); int32_t (*syncLogUpdateCommitIndex)(struct SSyncLogStore* pLogStore, SyncIndex index);
SyncIndex (*syncLogCommitIndex)(struct SSyncLogStore* pLogStore); SyncIndex (*syncLogCommitIndex)(struct SSyncLogStore* pLogStore);

4
source/libs/sync/inc/syncInt.h
View File

@ -192,6 +192,10 @@ typedef struct SSyncNode {
int64_t leaderTime; int64_t leaderTime;
int64_t lastReplicateTime; int64_t lastReplicateTime;
int32_t electNum;
int32_t becomeLeaderNum;
int32_t configChangeNum;
bool isStart; bool isStart;
} SSyncNode; } SSyncNode;

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

@ -205,9 +205,11 @@ int32_t syncNodeOnAppendEntries(SSyncNode* ths, const SRpcMsg* pRpcMsg) {
pLocalEntry = (SSyncRaftEntry*)taosLRUCacheValue(pCache, hLocal); pLocalEntry = (SSyncRaftEntry*)taosLRUCacheValue(pCache, hLocal);
code = 0; code = 0;
ths->pLogStore->cacheHit++;
sNTrace(ths, "hit cache index:%" PRId64 ", bytes:%u, %p", appendIndex, pLocalEntry->bytes, pLocalEntry); sNTrace(ths, "hit cache index:%" PRId64 ", bytes:%u, %p", appendIndex, pLocalEntry->bytes, pLocalEntry);
} else { } else {
ths->pLogStore->cacheMiss++;
sNTrace(ths, "miss cache index:%" PRId64, appendIndex); sNTrace(ths, "miss cache index:%" PRId64, appendIndex);
code = ths->pLogStore->syncLogGetEntry(ths->pLogStore, appendIndex, &pLocalEntry); code = ths->pLogStore->syncLogGetEntry(ths->pLogStore, appendIndex, &pLocalEntry);

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

@ -117,9 +117,11 @@ void syncMaybeAdvanceCommitIndex(SSyncNode* pSyncNode) {
if (h) { if (h) {
pEntry = (SSyncRaftEntry*)taosLRUCacheValue(pCache, h); pEntry = (SSyncRaftEntry*)taosLRUCacheValue(pCache, h);
pSyncNode->pLogStore->cacheHit++;
sNTrace(pSyncNode, "hit cache index:%" PRId64 ", bytes:%u, %p", index, pEntry->bytes, pEntry); sNTrace(pSyncNode, "hit cache index:%" PRId64 ", bytes:%u, %p", index, pEntry->bytes, pEntry);
} else { } else {
pSyncNode->pLogStore->cacheMiss++;
sNTrace(pSyncNode, "miss cache index:%" PRId64, index); sNTrace(pSyncNode, "miss cache index:%" PRId64, index);
int32_t code = pSyncNode->pLogStore->syncLogGetEntry(pSyncNode->pLogStore, index, &pEntry); int32_t code = pSyncNode->pLogStore->syncLogGetEntry(pSyncNode->pLogStore, index, &pEntry);

6
source/libs/sync/src/syncElection.c
View File

@ -61,7 +61,8 @@ static int32_t syncNodeRequestVotePeers(SSyncNode* pNode) {
} }
int32_t syncNodeElect(SSyncNode* pSyncNode) { int32_t syncNodeElect(SSyncNode* pSyncNode) {
sNTrace(pSyncNode, "begin election"); sNInfo(pSyncNode, "begin election");
pSyncNode->electNum++;
int32_t ret = 0; int32_t ret = 0;
if (pSyncNode->state == TAOS_SYNC_STATE_FOLLOWER) { if (pSyncNode->state == TAOS_SYNC_STATE_FOLLOWER) {
@ -86,7 +87,7 @@ int32_t syncNodeElect(SSyncNode* pSyncNode) {
syncNodeCandidate2Leader(pSyncNode); syncNodeCandidate2Leader(pSyncNode);
pSyncNode->pVotesGranted->toLeader = true; pSyncNode->pVotesGranted->toLeader = true;
return ret; return ret;
} }
if (pSyncNode->replicaNum == 1) { if (pSyncNode->replicaNum == 1) {
// only myself, to leader // only myself, to leader
@ -98,7 +99,6 @@ int32_t syncNodeElect(SSyncNode* pSyncNode) {
syncNodeCandidate2Leader(pSyncNode); syncNodeCandidate2Leader(pSyncNode);
pSyncNode->pVotesGranted->toLeader = true; pSyncNode->pVotesGranted->toLeader = true;
return ret; return ret;
} }
ret = syncNodeRequestVotePeers(pSyncNode); ret = syncNodeRequestVotePeers(pSyncNode);

26
source/libs/sync/src/syncMain.c
View File

@ -410,9 +410,11 @@ bool syncIsReadyForRead(int64_t rid) {
pEntry = (SSyncRaftEntry*)taosLRUCacheValue(pCache, h); pEntry = (SSyncRaftEntry*)taosLRUCacheValue(pCache, h);
code = 0; code = 0;
pSyncNode->pLogStore->cacheHit++;
sNTrace(pSyncNode, "hit cache index:%" PRId64 ", bytes:%u, %p", lastIndex, pEntry->bytes, pEntry); sNTrace(pSyncNode, "hit cache index:%" PRId64 ", bytes:%u, %p", lastIndex, pEntry->bytes, pEntry);
} else { } else {
pSyncNode->pLogStore->cacheMiss++;
sNTrace(pSyncNode, "miss cache index:%" PRId64, lastIndex); sNTrace(pSyncNode, "miss cache index:%" PRId64, lastIndex);
code = pSyncNode->pLogStore->syncLogGetEntry(pSyncNode->pLogStore, lastIndex, &pEntry); code = pSyncNode->pLogStore->syncLogGetEntry(pSyncNode->pLogStore, lastIndex, &pEntry);
@ -1008,6 +1010,10 @@ SSyncNode* syncNodeOpen(SSyncInfo* pSyncInfo) {
atomic_store_64(&pSyncNode->snapshottingIndex, SYNC_INDEX_INVALID); atomic_store_64(&pSyncNode->snapshottingIndex, SYNC_INDEX_INVALID);
pSyncNode->isStart = true; pSyncNode->isStart = true;
pSyncNode->electNum = 0;
pSyncNode->becomeLeaderNum = 0;
pSyncNode->configChangeNum = 0;
sNTrace(pSyncNode, "sync open, node:%p", pSyncNode); sNTrace(pSyncNode, "sync open, node:%p", pSyncNode);
return pSyncNode; return pSyncNode;
@ -1340,6 +1346,8 @@ void syncNodeDoConfigChange(SSyncNode* pSyncNode, SSyncCfg* pNewConfig, SyncInde
pSyncNode->pRaftCfg->cfg = *pNewConfig; pSyncNode->pRaftCfg->cfg = *pNewConfig;
pSyncNode->pRaftCfg->lastConfigIndex = lastConfigChangeIndex; pSyncNode->pRaftCfg->lastConfigIndex = lastConfigChangeIndex;
pSyncNode->configChangeNum++;
bool IamInOld = syncNodeInConfig(pSyncNode, &oldConfig); bool IamInOld = syncNodeInConfig(pSyncNode, &oldConfig);
bool IamInNew = syncNodeInConfig(pSyncNode, pNewConfig); bool IamInNew = syncNodeInConfig(pSyncNode, pNewConfig);
@ -1363,7 +1371,7 @@ void syncNodeDoConfigChange(SSyncNode* pSyncNode, SSyncCfg* pNewConfig, SyncInde
char newCfgStr[1024] = {0}; char newCfgStr[1024] = {0};
syncCfg2SimpleStr(&oldConfig, oldCfgStr, sizeof(oldCfgStr)); syncCfg2SimpleStr(&oldConfig, oldCfgStr, sizeof(oldCfgStr));
syncCfg2SimpleStr(pNewConfig, oldCfgStr, sizeof(oldCfgStr)); syncCfg2SimpleStr(pNewConfig, oldCfgStr, sizeof(oldCfgStr));
sNTrace(pSyncNode, "begin do config change, from %s to %s", oldCfgStr, oldCfgStr); sNInfo(pSyncNode, "begin do config change, from %s to %s", oldCfgStr, oldCfgStr);
if (IamInNew) { if (IamInNew) {
pSyncNode->pRaftCfg->isStandBy = 0; // change isStandBy to normal pSyncNode->pRaftCfg->isStandBy = 0; // change isStandBy to normal
@ -1495,13 +1503,13 @@ void syncNodeDoConfigChange(SSyncNode* pSyncNode, SSyncCfg* pNewConfig, SyncInde
} else { } else {
// persist cfg // persist cfg
raftCfgPersist(pSyncNode->pRaftCfg); raftCfgPersist(pSyncNode->pRaftCfg);
sNTrace(pSyncNode, "do not config change from %d to %d, index:%" PRId64 ", %s --> %s", oldConfig.replicaNum, sNInfo(pSyncNode, "do not config change from %d to %d, index:%" PRId64 ", %s --> %s", oldConfig.replicaNum,
pNewConfig->replicaNum, lastConfigChangeIndex, oldCfgStr, newCfgStr); pNewConfig->replicaNum, lastConfigChangeIndex, oldCfgStr, newCfgStr);
} }
_END: _END:
// log end config change // log end config change
sNTrace(pSyncNode, "end do config change, from %s to %s", oldCfgStr, newCfgStr); sNInfo(pSyncNode, "end do config change, from %s to %s", oldCfgStr, newCfgStr);
} }
// raft state change -------------- // raft state change --------------
@ -1598,6 +1606,8 @@ void syncNodeBecomeFollower(SSyncNode* pSyncNode, const char* debugStr) {
void syncNodeBecomeLeader(SSyncNode* pSyncNode, const char* debugStr) { void syncNodeBecomeLeader(SSyncNode* pSyncNode, const char* debugStr) {
pSyncNode->leaderTime = taosGetTimestampMs(); pSyncNode->leaderTime = taosGetTimestampMs();
pSyncNode->becomeLeaderNum++;
// reset restoreFinish // reset restoreFinish
pSyncNode->restoreFinish = false; pSyncNode->restoreFinish = false;
@ -1666,7 +1676,7 @@ void syncNodeBecomeLeader(SSyncNode* pSyncNode, const char* debugStr) {
pSyncNode->minMatchIndex = SYNC_INDEX_INVALID; pSyncNode->minMatchIndex = SYNC_INDEX_INVALID;
// trace log // trace log
sNTrace(pSyncNode, "become leader %s", debugStr); sNInfo(pSyncNode, "become leader %s", debugStr);
} }
void syncNodeCandidate2Leader(SSyncNode* pSyncNode) { void syncNodeCandidate2Leader(SSyncNode* pSyncNode) {
@ -1842,9 +1852,11 @@ SyncTerm syncNodeGetPreTerm(SSyncNode* pSyncNode, SyncIndex index) {
pPreEntry = (SSyncRaftEntry*)taosLRUCacheValue(pCache, h); pPreEntry = (SSyncRaftEntry*)taosLRUCacheValue(pCache, h);
code = 0; code = 0;
pSyncNode->pLogStore->cacheHit++;
sNTrace(pSyncNode, "hit cache index:%" PRId64 ", bytes:%u, %p", preIndex, pPreEntry->bytes, pPreEntry); sNTrace(pSyncNode, "hit cache index:%" PRId64 ", bytes:%u, %p", preIndex, pPreEntry->bytes, pPreEntry);
} else { } else {
pSyncNode->pLogStore->cacheMiss++;
sNTrace(pSyncNode, "miss cache index:%" PRId64, preIndex); sNTrace(pSyncNode, "miss cache index:%" PRId64, preIndex);
code = pSyncNode->pLogStore->syncLogGetEntry(pSyncNode->pLogStore, preIndex, &pPreEntry); code = pSyncNode->pLogStore->syncLogGetEntry(pSyncNode->pLogStore, preIndex, &pPreEntry);
@ -1971,7 +1983,7 @@ static void syncNodeEqHeartbeatTimer(void* param, void* tmrId) {
return; return;
} }
sTrace("enqueue heartbeat timer"); sTrace("vgId:%d, enqueue heartbeat timer", pNode->vgId);
code = pNode->syncEqMsg(pNode->msgcb, &rpcMsg); code = pNode->syncEqMsg(pNode->msgcb, &rpcMsg);
if (code != 0) { if (code != 0) {
sError("failed to enqueue heartbeat msg since %s", terrstr()); sError("failed to enqueue heartbeat msg since %s", terrstr());
@ -2526,9 +2538,11 @@ int32_t syncNodeDoCommit(SSyncNode* ths, SyncIndex beginIndex, SyncIndex endInde
if (h) { if (h) {
pEntry = (SSyncRaftEntry*)taosLRUCacheValue(pCache, h); pEntry = (SSyncRaftEntry*)taosLRUCacheValue(pCache, h);
ths->pLogStore->cacheHit++;
sNTrace(ths, "hit cache index:%" PRId64 ", bytes:%u, %p", i, pEntry->bytes, pEntry); sNTrace(ths, "hit cache index:%" PRId64 ", bytes:%u, %p", i, pEntry->bytes, pEntry);
} else { } else {
ths->pLogStore->cacheMiss++;
sNTrace(ths, "miss cache index:%" PRId64, i); sNTrace(ths, "miss cache index:%" PRId64, i);
code = ths->pLogStore->syncLogGetEntry(ths->pLogStore, i, &pEntry); code = ths->pLogStore->syncLogGetEntry(ths->pLogStore, i, &pEntry);

3
source/libs/sync/src/syncRaftLog.c
View File

@ -45,6 +45,9 @@ SSyncLogStore* logStoreCreate(SSyncNode* pSyncNode) {
return NULL; return NULL;
} }
pLogStore->cacheHit = 0;
pLogStore->cacheMiss = 0;
taosLRUCacheSetStrictCapacity(pLogStore->pCache, false); taosLRUCacheSetStrictCapacity(pLogStore->pCache, false);
pLogStore->data = taosMemoryMalloc(sizeof(SSyncLogStoreData)); pLogStore->data = taosMemoryMalloc(sizeof(SSyncLogStoreData));

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

@ -80,9 +80,11 @@ int32_t syncNodeReplicateOne(SSyncNode* pSyncNode, SRaftId* pDestId, bool snapsh
pEntry = (SSyncRaftEntry*)taosLRUCacheValue(pCache, h); pEntry = (SSyncRaftEntry*)taosLRUCacheValue(pCache, h);
code = 0; code = 0;
pSyncNode->pLogStore->cacheHit++;
sNTrace(pSyncNode, "hit cache index:%" PRId64 ", bytes:%u, %p", nextIndex, pEntry->bytes, pEntry); sNTrace(pSyncNode, "hit cache index:%" PRId64 ", bytes:%u, %p", nextIndex, pEntry->bytes, pEntry);
} else { } else {
pSyncNode->pLogStore->cacheMiss++;
sNTrace(pSyncNode, "miss cache index:%" PRId64, nextIndex); sNTrace(pSyncNode, "miss cache index:%" PRId64, nextIndex);
code = pSyncNode->pLogStore->syncLogGetEntry(pSyncNode->pLogStore, nextIndex, &pEntry); code = pSyncNode->pLogStore->syncLogGetEntry(pSyncNode->pLogStore, nextIndex, &pEntry);

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

@ -52,7 +52,7 @@ static void syncNodeCleanConfigIndex(SSyncNode* ths) {
} }
static int32_t syncNodeTimerRoutine(SSyncNode* ths) { static int32_t syncNodeTimerRoutine(SSyncNode* ths) {
sNTrace(ths, "timer routines"); sNInfo(ths, "timer routines");
// timer replicate // timer replicate
syncNodeReplicate(ths); syncNodeReplicate(ths);

30
source/libs/sync/src/syncUtil.c
View File

@ -242,6 +242,9 @@ void syncPrintNodeLog(const char* flags, ELogLevel level, int32_t dflag, SSyncNo
logBeginIndex = pNode->pLogStore->syncLogBeginIndex(pNode->pLogStore); logBeginIndex = pNode->pLogStore->syncLogBeginIndex(pNode->pLogStore);
} }
int32_t cacheHit = pNode->pLogStore->cacheHit;
int32_t cacheMiss = pNode->pLogStore->cacheMiss;
char cfgStr[1024]; char cfgStr[1024];
if (pNode->pRaftCfg != NULL) { if (pNode->pRaftCfg != NULL) {
syncCfg2SimpleStr(&(pNode->pRaftCfg->cfg), cfgStr, sizeof(cfgStr)); syncCfg2SimpleStr(&(pNode->pRaftCfg->cfg), cfgStr, sizeof(cfgStr));
@ -275,18 +278,18 @@ void syncPrintNodeLog(const char* flags, ELogLevel level, int32_t dflag, SSyncNo
terrno = errCode; terrno = errCode;
if (pNode != NULL && pNode->pRaftCfg != NULL) { if (pNode != NULL && pNode->pRaftCfg != NULL) {
taosPrintLog(flags, level, dflag, taosPrintLog(
"vgId:%d, sync %s " flags, level, dflag,
"%s" "vgId:%d, sync %s "
", tm:%" PRIu64 ", cmt:%" PRId64 ", fst:%" PRId64 ", lst:%" PRId64 ", min:%" PRId64 ", snap:%" PRId64 "%s"
", snap-tm:%" PRIu64 ", sby:%d, aq:%d, snaping:%" PRId64 ", r-num:%d, lcfg:%" PRId64 ", tm:%" PRIu64 ", cmt:%" PRId64 ", fst:%" PRId64 ", lst:%" PRId64 ", min:%" PRId64 ", snap:%" PRId64
", chging:%d, rsto:%d, dquorum:%d, elt:%" PRId64 ", hb:%" PRId64 ", %s, %s, %s, %s", ", snap-tm:%" PRIu64 ", elt-num:%d, bl-num:%d, cc-num:%d, hit:%d, mis:%d, aq:%d, snaping:%" PRId64
pNode->vgId, syncStr(pNode->state), eventLog, currentTerm, pNode->commitIndex, logBeginIndex, ", r-num:%d, lcfg:%" PRId64 ", chging:%d, rsto:%d, dquorum:%d, elt:%" PRId64 ", hb:%" PRId64 ", %s, %s, %s, %s",
logLastIndex, pNode->minMatchIndex, snapshot.lastApplyIndex, snapshot.lastApplyTerm, pNode->vgId, syncStr(pNode->state), eventLog, currentTerm, pNode->commitIndex, logBeginIndex, logLastIndex,
pNode->pRaftCfg->isStandBy, aqItems, pNode->snapshottingIndex, pNode->replicaNum, pNode->minMatchIndex, snapshot.lastApplyIndex, snapshot.lastApplyTerm, pNode->electNum, pNode->becomeLeaderNum,
pNode->pRaftCfg->lastConfigIndex, pNode->changing, pNode->restoreFinish, quorum, pNode->configChangeNum, cacheHit, cacheMiss, aqItems, pNode->snapshottingIndex, pNode->replicaNum,
pNode->electTimerLogicClock, pNode->heartbeatTimerLogicClockUser, peerStr, cfgStr, hbTimeStr, pNode->pRaftCfg->lastConfigIndex, pNode->changing, pNode->restoreFinish, quorum, pNode->electTimerLogicClock,
hbrTimeStr); pNode->heartbeatTimerLogicClockUser, peerStr, cfgStr, hbTimeStr, hbrTimeStr);
} }
} }
@ -438,7 +441,8 @@ void syncLogRecvAppendEntriesReply(SSyncNode* pSyncNode, const SyncAppendEntries
host, port, pMsg->term, pMsg->privateTerm, pMsg->success, pMsg->lastSendIndex, pMsg->matchIndex, s); host, port, pMsg->term, pMsg->privateTerm, pMsg->success, pMsg->lastSendIndex, pMsg->matchIndex, s);
} }
void syncLogSendHeartbeat(SSyncNode* pSyncNode, const SyncHeartbeat* pMsg, bool printX, int64_t timerElapsed, int64_t execTime) { void syncLogSendHeartbeat(SSyncNode* pSyncNode, const SyncHeartbeat* pMsg, bool printX, int64_t timerElapsed,
int64_t execTime) {
if (!(sDebugFlag & DEBUG_TRACE)) return; if (!(sDebugFlag & DEBUG_TRACE)) return;
char host[64]; char host[64];