提交 64b540be 编写于 作者: S Shengliang Guan

refactor: adjust vnode propose msg

上级 f00709f3
...@@ -107,7 +107,7 @@ static void vmProcessSyncQueue(SQueueInfo *pInfo, STaosQall *qall, int32_t numOf ...@@ -107,7 +107,7 @@ static void vmProcessSyncQueue(SQueueInfo *pInfo, STaosQall *qall, int32_t numOf
const STraceId *trace = &pMsg->info.traceId; const STraceId *trace = &pMsg->info.traceId;
dGTrace("vgId:%d, msg:%p get from vnode-sync queue", pVnode->vgId, pMsg); dGTrace("vgId:%d, msg:%p get from vnode-sync queue", pVnode->vgId, pMsg);
int32_t code = vnodeProcessSyncReq(pVnode->pImpl, pMsg, NULL); // no response here int32_t code = vnodeProcessSyncMsg(pVnode->pImpl, pMsg, NULL); // no response here
dGTrace("vgId:%d, msg:%p is freed, code:0x%x", pVnode->vgId, pMsg, code); dGTrace("vgId:%d, msg:%p is freed, code:0x%x", pVnode->vgId, pMsg, code);
rpcFreeCont(pMsg->pCont); rpcFreeCont(pMsg->pCont);
taosFreeQitem(pMsg); taosFreeQitem(pMsg);
......
...@@ -52,10 +52,10 @@ int32_t vnodeCreate(const char *path, SVnodeCfg *pCfg, STfs *pTfs); ...@@ -52,10 +52,10 @@ int32_t vnodeCreate(const char *path, SVnodeCfg *pCfg, STfs *pTfs);
void vnodeDestroy(const char *path, STfs *pTfs); void vnodeDestroy(const char *path, STfs *pTfs);
SVnode *vnodeOpen(const char *path, STfs *pTfs, SMsgCb msgCb); SVnode *vnodeOpen(const char *path, STfs *pTfs, SMsgCb msgCb);
void vnodeClose(SVnode *pVnode); void vnodeClose(SVnode *pVnode);
int32_t vnodePreprocessReq(SVnode *pVnode, SRpcMsg *pMsg); int32_t vnodePreProcessReq(SVnode *pVnode, SRpcMsg *pMsg);
int32_t vnodeProcessWriteReq(SVnode *pVnode, SRpcMsg *pMsg, int64_t version, SRpcMsg *pRsp); int32_t vnodeProcessWriteReq(SVnode *pVnode, SRpcMsg *pMsg, int64_t version, SRpcMsg *pRsp);
int32_t vnodeProcessCMsg(SVnode *pVnode, SRpcMsg *pMsg, SRpcMsg **pRsp); int32_t vnodeProcessCMsg(SVnode *pVnode, SRpcMsg *pMsg, SRpcMsg **pRsp);
int32_t vnodeProcessSyncReq(SVnode *pVnode, SRpcMsg *pMsg, SRpcMsg **pRsp); int32_t vnodeProcessSyncMsg(SVnode *pVnode, SRpcMsg *pMsg, SRpcMsg **pRsp);
int32_t vnodePreprocessQueryMsg(SVnode *pVnode, SRpcMsg *pMsg); int32_t vnodePreprocessQueryMsg(SVnode *pVnode, SRpcMsg *pMsg);
int32_t vnodeProcessQueryMsg(SVnode *pVnode, SRpcMsg *pMsg); int32_t vnodeProcessQueryMsg(SVnode *pVnode, SRpcMsg *pMsg);
int32_t vnodeProcessFetchMsg(SVnode *pVnode, SRpcMsg *pMsg, SQueueInfo *pInfo); int32_t vnodeProcessFetchMsg(SVnode *pVnode, SRpcMsg *pMsg, SQueueInfo *pInfo);
......
...@@ -240,7 +240,7 @@ struct SVnode { ...@@ -240,7 +240,7 @@ struct SVnode {
SSink* pSink; SSink* pSink;
tsem_t canCommit; tsem_t canCommit;
int64_t sync; int64_t sync;
int32_t syncCount; int32_t blockCount;
tsem_t syncSem; tsem_t syncSem;
SQHandle* pQuery; SQHandle* pQuery;
}; };
......
...@@ -85,6 +85,7 @@ STqOffsetStore* tqOffsetOpen(STQ* pTq) { ...@@ -85,6 +85,7 @@ STqOffsetStore* tqOffsetOpen(STQ* pTq) {
void tqOffsetClose(STqOffsetStore* pStore) { void tqOffsetClose(STqOffsetStore* pStore) {
tqOffsetSnapshot(pStore); tqOffsetSnapshot(pStore);
taosHashCleanup(pStore->pHash); taosHashCleanup(pStore->pHash);
taosMemoryFree(pStore);
} }
STqOffset* tqOffsetRead(STqOffsetStore* pStore, const char* subscribeKey) { STqOffset* tqOffsetRead(STqOffsetStore* pStore, const char* subscribeKey) {
......
...@@ -81,7 +81,7 @@ SVnode *vnodeOpen(const char *path, STfs *pTfs, SMsgCb msgCb) { ...@@ -81,7 +81,7 @@ SVnode *vnodeOpen(const char *path, STfs *pTfs, SMsgCb msgCb) {
pVnode->state.applied = info.state.committed; pVnode->state.applied = info.state.committed;
pVnode->pTfs = pTfs; pVnode->pTfs = pTfs;
pVnode->msgCb = msgCb; pVnode->msgCb = msgCb;
pVnode->syncCount = 0; pVnode->blockCount = 0;
tsem_init(&pVnode->syncSem, 0, 0); tsem_init(&pVnode->syncSem, 0, 0);
tsem_init(&(pVnode->canCommit), 0, 1); tsem_init(&(pVnode->canCommit), 0, 1);
......
...@@ -28,7 +28,7 @@ static int32_t vnodeProcessAlterHasnRangeReq(SVnode *pVnode, int64_t version, vo ...@@ -28,7 +28,7 @@ static int32_t vnodeProcessAlterHasnRangeReq(SVnode *pVnode, int64_t version, vo
static int32_t vnodeProcessWriteMsg(SVnode *pVnode, int64_t version, SRpcMsg *pMsg, SRpcMsg *pRsp); static int32_t vnodeProcessWriteMsg(SVnode *pVnode, int64_t version, SRpcMsg *pMsg, SRpcMsg *pRsp);
static int32_t vnodeProcessDropTtlTbReq(SVnode *pVnode, int64_t version, void *pReq, int32_t len, SRpcMsg *pRsp); static int32_t vnodeProcessDropTtlTbReq(SVnode *pVnode, int64_t version, void *pReq, int32_t len, SRpcMsg *pRsp);
int32_t vnodePreprocessReq(SVnode *pVnode, SRpcMsg *pMsg) { int32_t vnodePreProcessReq(SVnode *pVnode, SRpcMsg *pMsg) {
int32_t code = 0; int32_t code = 0;
SDecoder dc = {0}; SDecoder dc = {0};
......
...@@ -25,12 +25,12 @@ static inline bool vnodeIsMsgWeak(tmsg_t type) { return false; } ...@@ -25,12 +25,12 @@ static inline bool vnodeIsMsgWeak(tmsg_t type) { return false; }
static inline void vnodeAccumBlockMsg(SVnode *pVnode, tmsg_t type) { static inline void vnodeAccumBlockMsg(SVnode *pVnode, tmsg_t type) {
if (!vnodeIsMsgBlock(type)) return; if (!vnodeIsMsgBlock(type)) return;
int32_t count = atomic_add_fetch_32(&pVnode->syncCount, 1); int32_t count = atomic_add_fetch_32(&pVnode->blockCount, 1);
vTrace("vgId:%d, accum block, count:%d type:%s", pVnode->config.vgId, count, TMSG_INFO(type)); vTrace("vgId:%d, accum block, count:%d type:%s", pVnode->config.vgId, count, TMSG_INFO(type));
} }
static inline void vnodeWaitBlockMsg(SVnode *pVnode) { static inline void vnodeWaitBlockMsg(SVnode *pVnode) {
int32_t count = atomic_load_32(&pVnode->syncCount); int32_t count = atomic_load_32(&pVnode->blockCount);
if (count <= 0) return; if (count <= 0) return;
vTrace("vgId:%d, wait block finish, count:%d", pVnode->config.vgId, count); vTrace("vgId:%d, wait block finish, count:%d", pVnode->config.vgId, count);
...@@ -40,10 +40,10 @@ static inline void vnodeWaitBlockMsg(SVnode *pVnode) { ...@@ -40,10 +40,10 @@ static inline void vnodeWaitBlockMsg(SVnode *pVnode) {
static inline void vnodePostBlockMsg(SVnode *pVnode, tmsg_t type) { static inline void vnodePostBlockMsg(SVnode *pVnode, tmsg_t type) {
if (!vnodeIsMsgBlock(type)) return; if (!vnodeIsMsgBlock(type)) return;
int32_t count = atomic_load_32(&pVnode->syncCount); int32_t count = atomic_load_32(&pVnode->blockCount);
if (count <= 0) return; if (count <= 0) return;
count = atomic_sub_fetch_32(&pVnode->syncCount, 1); count = atomic_sub_fetch_32(&pVnode->blockCount, 1);
vTrace("vgId:%d, post block, count:%d type:%s", pVnode->config.vgId, count, TMSG_INFO(type)); vTrace("vgId:%d, post block, count:%d type:%s", pVnode->config.vgId, count, TMSG_INFO(type));
if (count <= 0) { if (count <= 0) {
tsem_post(&pVnode->syncSem); tsem_post(&pVnode->syncSem);
...@@ -84,8 +84,10 @@ static int32_t vnodeProcessAlterReplicaReq(SVnode *pVnode, SRpcMsg *pMsg) { ...@@ -84,8 +84,10 @@ static int32_t vnodeProcessAlterReplicaReq(SVnode *pVnode, SRpcMsg *pMsg) {
terrno = TSDB_CODE_INVALID_MSG; terrno = TSDB_CODE_INVALID_MSG;
return TSDB_CODE_INVALID_MSG; return TSDB_CODE_INVALID_MSG;
} }
STraceId *trace = &pMsg->info.traceId;
const STraceId *trace = &pMsg->info.traceId;
vGTrace("vgId:%d, start to alter vnode replica to %d, handle:%p", TD_VID(pVnode), req.replica, pMsg->info.handle); vGTrace("vgId:%d, start to alter vnode replica to %d, handle:%p", TD_VID(pVnode), req.replica, pMsg->info.handle);
SSyncCfg cfg = {.replicaNum = req.replica, .myIndex = req.selfIndex}; SSyncCfg cfg = {.replicaNum = req.replica, .myIndex = req.selfIndex};
for (int32_t r = 0; r < req.replica; ++r) { for (int32_t r = 0; r < req.replica; ++r) {
SNodeInfo *pNode = &cfg.nodeInfo[r]; SNodeInfo *pNode = &cfg.nodeInfo[r];
...@@ -126,68 +128,49 @@ void vnodeProposeMsg(SQueueInfo *pInfo, STaosQall *qall, int32_t numOfMsgs) { ...@@ -126,68 +128,49 @@ void vnodeProposeMsg(SQueueInfo *pInfo, STaosQall *qall, int32_t numOfMsgs) {
for (int32_t m = 0; m < numOfMsgs; m++) { for (int32_t m = 0; m < numOfMsgs; m++) {
if (taosGetQitem(qall, (void **)&pMsg) == 0) continue; if (taosGetQitem(qall, (void **)&pMsg) == 0) continue;
STraceId *trace = &pMsg->info.traceId; const STraceId *trace = &pMsg->info.traceId;
vGTrace("vgId:%d, msg:%p get from vnode-write queue handle:%p", vgId, pMsg, pMsg->info.handle); vGTrace("vgId:%d, msg:%p get from vnode-write queue handle:%p", vgId, pMsg, pMsg->info.handle);
if (pMsg->msgType == TDMT_VND_ALTER_REPLICA) { code = vnodePreProcessReq(pVnode, pMsg);
code = vnodeProcessAlterReplicaReq(pVnode, pMsg); if (code != 0) {
vError("vgId:%d, msg:%p failed to pre-process since %s", vgId, pMsg, terrstr());
} else { } else {
code = vnodePreprocessReq(pVnode, pMsg); if (pMsg->msgType == TDMT_VND_ALTER_REPLICA) {
if (code != 0) { code = vnodeProcessAlterReplicaReq(pVnode, pMsg);
vError("vgId:%d, failed to pre-process msg:%p since %s", vgId, pMsg, terrstr());
} else { } else {
code = syncPropose(pVnode->sync, pMsg, vnodeIsMsgWeak(pMsg->msgType)); code = syncPropose(pVnode->sync, pMsg, vnodeIsMsgWeak(pMsg->msgType));
if (code == 1) { if (code > 0) {
do {
static int32_t cnt = 0;
if (cnt++ % 1000 == 1) {
vInfo("vgId:%d, msg:%p apply right now, apply index:%ld, msgtype:%s,%d", vgId, pMsg,
pMsg->info.conn.applyIndex, TMSG_INFO(pMsg->msgType), pMsg->msgType);
}
} while (0);
SRpcMsg rsp = {.code = pMsg->code, .info = pMsg->info}; SRpcMsg rsp = {.code = pMsg->code, .info = pMsg->info};
if (vnodeProcessWriteReq(pVnode, pMsg, pMsg->info.conn.applyIndex, &rsp) < 0) { if (vnodeProcessWriteReq(pVnode, pMsg, pMsg->info.conn.applyIndex, &rsp) < 0) {
rsp.code = terrno; rsp.code = terrno;
vInfo("vgId:%d, msg:%p failed to apply right now since %s", vgId, pMsg, terrstr()); vError("vgId:%d, msg:%p failed to apply right now since %s", vgId, pMsg, terrstr());
}
if (rsp.info.handle != NULL) {
tmsgSendRsp(&rsp);
} }
tmsgSendRsp(&rsp);
} }
} }
} }
if (code == 0) { if (code == 0) {
vnodeAccumBlockMsg(pVnode, pMsg->msgType); vnodeAccumBlockMsg(pVnode, pMsg->msgType);
} else if (code == -1 && terrno == TSDB_CODE_SYN_NOT_LEADER) { } else if (code < 0) {
SEpSet newEpSet = {0}; if (terrno == TSDB_CODE_SYN_NOT_LEADER) {
syncGetRetryEpSet(pVnode->sync, &newEpSet); SEpSet newEpSet = {0};
syncGetRetryEpSet(pVnode->sync, &newEpSet);
/* vGTrace("vgId:%d, msg:%p is redirect since not leader, numOfEps:%d inUse:%d", vgId, pMsg, newEpSet.numOfEps,
syncGetEpSet(pVnode->sync, &newEpSet); newEpSet.inUse);
SEp *pEp = &newEpSet.eps[newEpSet.inUse]; for (int32_t i = 0; i < newEpSet.numOfEps; ++i) {
if (pEp->port == tsServerPort && strcmp(pEp->fqdn, tsLocalFqdn) == 0) { vGTrace("vgId:%d, msg:%p redirect:%d ep:%s:%u", vgId, pMsg, i, newEpSet.eps[i].fqdn, newEpSet.eps[i].port);
newEpSet.inUse = (newEpSet.inUse + 1) % newEpSet.numOfEps; }
} pMsg->info.hasEpSet = 1;
*/ SRpcMsg rsp = {.code = TSDB_CODE_RPC_REDIRECT, .info = pMsg->info};
tmsgSendRedirectRsp(&rsp, &newEpSet);
vGTrace("vgId:%d, msg:%p is redirect since not leader, numOfEps:%d inUse:%d", vgId, pMsg, newEpSet.numOfEps, } else {
newEpSet.inUse);
for (int32_t i = 0; i < newEpSet.numOfEps; ++i) {
vGTrace("vgId:%d, msg:%p redirect:%d ep:%s:%u", vgId, pMsg, i, newEpSet.eps[i].fqdn, newEpSet.eps[i].port);
}
pMsg->info.hasEpSet = 1;
SRpcMsg rsp = {.code = TSDB_CODE_RPC_REDIRECT, .info = pMsg->info};
tmsgSendRedirectRsp(&rsp, &newEpSet);
} else {
if (code != 1) {
if (terrno != 0) code = terrno; if (terrno != 0) code = terrno;
vError("vgId:%d, msg:%p failed to propose since %s, code:0x%x", vgId, pMsg, tstrerror(code), code); vError("vgId:%d, msg:%p failed to propose since %s, code:0x%x", vgId, pMsg, tstrerror(code), code);
SRpcMsg rsp = {.code = code, .info = pMsg->info}; SRpcMsg rsp = {.code = code, .info = pMsg->info};
tmsgSendRsp(&rsp); tmsgSendRsp(&rsp);
} }
} else {
} }
vGTrace("vgId:%d, msg:%p is freed, code:0x%x", vgId, pMsg, code); vGTrace("vgId:%d, msg:%p is freed, code:0x%x", vgId, pMsg, code);
...@@ -206,7 +189,7 @@ void vnodeApplyMsg(SQueueInfo *pInfo, STaosQall *qall, int32_t numOfMsgs) { ...@@ -206,7 +189,7 @@ void vnodeApplyMsg(SQueueInfo *pInfo, STaosQall *qall, int32_t numOfMsgs) {
for (int32_t i = 0; i < numOfMsgs; ++i) { for (int32_t i = 0; i < numOfMsgs; ++i) {
if (taosGetQitem(qall, (void **)&pMsg) == 0) continue; if (taosGetQitem(qall, (void **)&pMsg) == 0) continue;
STraceId *trace = &pMsg->info.traceId; const STraceId *trace = &pMsg->info.traceId;
vGTrace("vgId:%d, msg:%p get from vnode-apply queue, type:%s handle:%p", vgId, pMsg, TMSG_INFO(pMsg->msgType), vGTrace("vgId:%d, msg:%p get from vnode-apply queue, type:%s handle:%p", vgId, pMsg, TMSG_INFO(pMsg->msgType),
pMsg->info.handle); pMsg->info.handle);
...@@ -229,7 +212,7 @@ void vnodeApplyMsg(SQueueInfo *pInfo, STaosQall *qall, int32_t numOfMsgs) { ...@@ -229,7 +212,7 @@ void vnodeApplyMsg(SQueueInfo *pInfo, STaosQall *qall, int32_t numOfMsgs) {
} }
} }
int32_t vnodeProcessSyncReq(SVnode *pVnode, SRpcMsg *pMsg, SRpcMsg **pRsp) { int32_t vnodeProcessSyncMsg(SVnode *pVnode, SRpcMsg *pMsg, SRpcMsg **pRsp) {
int32_t ret = 0; int32_t ret = 0;
if (syncEnvIsStart()) { if (syncEnvIsStart()) {
...@@ -247,7 +230,7 @@ int32_t vnodeProcessSyncReq(SVnode *pVnode, SRpcMsg *pMsg, SRpcMsg **pRsp) { ...@@ -247,7 +230,7 @@ int32_t vnodeProcessSyncReq(SVnode *pVnode, SRpcMsg *pMsg, SRpcMsg **pRsp) {
} }
if (gRaftDetailLog) { if (gRaftDetailLog) {
char logBuf[512] = {0}; char logBuf[512] = {0};
snprintf(logBuf, sizeof(logBuf), "==vnodeProcessSyncReq== msgType:%d, syncNode: %s", pMsg->msgType, snprintf(logBuf, sizeof(logBuf), "==vnodeProcessSyncMsg== msgType:%d, syncNode: %s", pMsg->msgType,
syncNodeStr); syncNodeStr);
syncRpcMsgLog2(logBuf, pMsg); syncRpcMsgLog2(logBuf, pMsg);
} }
...@@ -313,7 +296,7 @@ int32_t vnodeProcessSyncReq(SVnode *pVnode, SRpcMsg *pMsg, SRpcMsg **pRsp) { ...@@ -313,7 +296,7 @@ int32_t vnodeProcessSyncReq(SVnode *pVnode, SRpcMsg *pMsg, SRpcMsg **pRsp) {
SRpcMsg rsp = {.code = ret, .info = pMsg->info}; SRpcMsg rsp = {.code = ret, .info = pMsg->info};
tmsgSendRsp(&rsp); tmsgSendRsp(&rsp);
} else { } else {
vError("==vnodeProcessSyncReq== error msg type:%d", pRpcMsg->msgType); vError("==vnodeProcessSyncMsg== error msg type:%d", pRpcMsg->msgType);
ret = -1; ret = -1;
} }
...@@ -380,14 +363,14 @@ int32_t vnodeProcessSyncReq(SVnode *pVnode, SRpcMsg *pMsg, SRpcMsg **pRsp) { ...@@ -380,14 +363,14 @@ int32_t vnodeProcessSyncReq(SVnode *pVnode, SRpcMsg *pMsg, SRpcMsg **pRsp) {
SRpcMsg rsp = {.code = ret, .info = pMsg->info}; SRpcMsg rsp = {.code = ret, .info = pMsg->info};
tmsgSendRsp(&rsp); tmsgSendRsp(&rsp);
} else { } else {
vError("==vnodeProcessSyncReq== error msg type:%d", pRpcMsg->msgType); vError("==vnodeProcessSyncMsg== error msg type:%d", pRpcMsg->msgType);
ret = -1; ret = -1;
} }
} }
syncNodeRelease(pSyncNode); syncNodeRelease(pSyncNode);
} else { } else {
vError("==vnodeProcessSyncReq== error syncEnv stop"); vError("==vnodeProcessSyncMsg== error syncEnv stop");
ret = -1; ret = -1;
} }
......
Markdown is supported
0% .
You are about to add 0 people to the discussion. Proceed with caution.
先完成此消息的编辑!
想要评论请 注册