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

update interface

This commit is contained in:
yihaoDeng 2024-11-29 14:56:29 +08:00
parent 90b60d9d01
commit f7cb4b9b09
14 changed files with 204 additions and 155 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

4
include/common/tmsgcb.h
View File

@ -47,7 +47,7 @@ typedef int32_t (*GetQueueSizeFp)(void* pMgmt, int32_t vgId, EQueueType qtype);
typedef int32_t (*SendReqFp)(const SEpSet* pEpSet, SRpcMsg* pMsg);
typedef void (*SendRspFp)(SRpcMsg* pMsg);
typedef void (*RegisterBrokenLinkArgFp)(struct SRpcMsg* pMsg);
typedef void (*ReleaseHandleFp)(SRpcHandleInfo* pHandle, int8_t type);
typedef void (*ReleaseHandleFp)(SRpcHandleInfo* pHandle, int8_t type, int32_t status);
typedef void (*ReportStartup)(const char* name, const char* desc);
typedef struct {
@ -76,7 +76,7 @@ int32_t tmsgSendReq(const SEpSet* epSet, SRpcMsg* pMsg);
int32_t tmsgSendSyncReq(const SEpSet* epSet, SRpcMsg* pMsg);
void tmsgSendRsp(SRpcMsg* pMsg);
void tmsgRegisterBrokenLinkArg(SRpcMsg* pMsg);
void tmsgReleaseHandle(SRpcHandleInfo* pHandle, int8_t type);
void tmsgReleaseHandle(SRpcHandleInfo* pHandle, int8_t type, int32_t code);
void tmsgReportStartup(const char* name, const char* desc);
bool tmsgUpdateDnodeInfo(int32_t* dnodeId, int64_t* clusterId, char* fqdn, uint16_t* port);
void tmsgUpdateDnodeEpSet(SEpSet* epset);

2
include/libs/transport/trpc.h
View File

@ -171,7 +171,7 @@ void *rpcReallocCont(void *ptr, int64_t contLen);
int32_t rpcSendRequest(void *thandle, const SEpSet *pEpSet, SRpcMsg *pMsg, int64_t *rid);
int32_t rpcSendResponse(const SRpcMsg *pMsg);
int32_t rpcRegisterBrokenLinkArg(SRpcMsg *msg);
int32_t rpcReleaseHandle(void *handle, int8_t type); // just release conn to rpc instance, no close sock
int32_t rpcReleaseHandle(void *handle, int8_t type, int32_t code); // just release conn to rpc instance, no close sock
// These functions will not be called in the child process
int32_t rpcSendRequestWithCtx(void *thandle, const SEpSet *pEpSet, SRpcMsg *pMsg, int64_t *rid, SRpcCtx *ctx);

4
source/dnode/mgmt/node_mgmt/src/dmTransport.c
View File

@ -368,7 +368,9 @@ static inline int32_t dmSendSyncReq(const SEpSet *pEpSet, SRpcMsg *pMsg) {
static inline void dmRegisterBrokenLinkArg(SRpcMsg *pMsg) { (void)rpcRegisterBrokenLinkArg(pMsg); }
static inline void dmReleaseHandle(SRpcHandleInfo *pHandle, int8_t type) { (void)rpcReleaseHandle(pHandle, type); }
static inline void dmReleaseHandle(SRpcHandleInfo *pHandle, int8_t type, int32_t status) {
(void)rpcReleaseHandle(pHandle, type, status);
}
static bool rpcRfp(int32_t code, tmsg_t msgType) {
if (code == TSDB_CODE_RPC_NETWORK_UNAVAIL || code == TSDB_CODE_RPC_BROKEN_LINK || code == TSDB_CODE_MNODE_NOT_FOUND ||

45
source/libs/qworker/src/qwUtil.c
View File

@ -286,7 +286,7 @@ void qwFreeTaskHandle(SQWTaskCtx *ctx) {
void qwFreeSinkHandle(SQWTaskCtx *ctx) {
// Note: free/kill may in RC
void* osinkHandle = atomic_load_ptr(&ctx->sinkHandle);
void *osinkHandle = atomic_load_ptr(&ctx->sinkHandle);
if (osinkHandle && osinkHandle == atomic_val_compare_exchange_ptr(&ctx->sinkHandle, osinkHandle, NULL)) {
QW_SINK_ENABLE_MEMPOOL(ctx);
dsDestroyDataSinker(osinkHandle);
@ -312,7 +312,7 @@ int32_t qwKillTaskHandle(SQWTaskCtx *ctx, int32_t rspCode) {
void qwFreeTaskCtx(QW_FPARAMS_DEF, SQWTaskCtx *ctx) {
if (ctx->ctrlConnInfo.handle) {
tmsgReleaseHandle(&ctx->ctrlConnInfo, TAOS_CONN_SERVER);
tmsgReleaseHandle(&ctx->ctrlConnInfo, TAOS_CONN_SERVER, 0);
}
ctx->ctrlConnInfo.handle = NULL;
@ -336,9 +336,8 @@ static void freeExplainExecItem(void *param) {
taosMemoryFree(pInfo->verboseInfo);
}
int32_t qwSendExplainResponse(QW_FPARAMS_DEF, SQWTaskCtx *ctx) {
int32_t code = TSDB_CODE_SUCCESS;
int32_t code = TSDB_CODE_SUCCESS;
qTaskInfo_t taskHandle = ctx->taskHandle;
ctx->explainRsped = true;
@ -387,8 +386,6 @@ _return:
return code;
}
int32_t qwDropTaskCtx(QW_FPARAMS_DEF) {
char id[sizeof(qId) + sizeof(cId) + sizeof(tId) + sizeof(eId)] = {0};
QW_SET_QTID(id, qId, cId, tId, eId);
@ -477,7 +474,6 @@ _return:
QW_RET(code);
}
int32_t qwHandleDynamicTaskEnd(QW_FPARAMS_DEF) {
char id[sizeof(qId) + sizeof(cId) + sizeof(tId) + sizeof(eId)] = {0};
QW_SET_QTID(id, qId, cId, tId, eId);
@ -538,16 +534,17 @@ void qwSetHbParam(int64_t refId, SQWHbParam **pParam) {
}
int32_t qwSaveTbVersionInfo(qTaskInfo_t pTaskInfo, SQWTaskCtx *ctx) {
char dbFName[TSDB_DB_FNAME_LEN];
char tbName[TSDB_TABLE_NAME_LEN];
char dbFName[TSDB_DB_FNAME_LEN];
char tbName[TSDB_TABLE_NAME_LEN];
STbVerInfo tbInfo;
int32_t i = 0;
int32_t code = TSDB_CODE_SUCCESS;
bool tbGet = false;
int32_t i = 0;
int32_t code = TSDB_CODE_SUCCESS;
bool tbGet = false;
while (true) {
tbGet = false;
code = qGetQueryTableSchemaVersion(pTaskInfo, dbFName, TSDB_DB_FNAME_LEN, tbName, TSDB_TABLE_NAME_LEN, &tbInfo.sversion, &tbInfo.tversion, i, &tbGet);
code = qGetQueryTableSchemaVersion(pTaskInfo, dbFName, TSDB_DB_FNAME_LEN, tbName, TSDB_TABLE_NAME_LEN,
&tbInfo.sversion, &tbInfo.tversion, i, &tbGet);
if (TSDB_CODE_SUCCESS != code || !tbGet) {
break;
}
@ -727,21 +724,21 @@ void qwClearExpiredSch(SQWorker *mgmt, SArray *pExpiredSch) {
}
}
void qwDestroyJobInfo(void* job) {
void qwDestroyJobInfo(void *job) {
if (NULL == job) {
return;
}
SQWJobInfo* pJob = (SQWJobInfo*)job;
SQWJobInfo *pJob = (SQWJobInfo *)job;
taosMemoryFreeClear(pJob->memInfo);
taosHashCleanup(pJob->pSessions);
pJob->pSessions = NULL;
}
bool qwStopTask(QW_FPARAMS_DEF, SQWTaskCtx *ctx) {
bool qwStopTask(QW_FPARAMS_DEF, SQWTaskCtx *ctx) {
int32_t code = TSDB_CODE_SUCCESS;
bool taskFreed = false;
bool taskFreed = false;
QW_LOCK(QW_WRITE, &ctx->lock);
@ -781,7 +778,7 @@ bool qwStopTask(QW_FPARAMS_DEF, SQWTaskCtx *ctx) {
}
bool qwRetireTask(QW_FPARAMS_DEF) {
SQWTaskCtx *ctx = NULL;
SQWTaskCtx *ctx = NULL;
int32_t code = qwAcquireTaskCtx(QW_FPARAMS(), &ctx);
if (TSDB_CODE_SUCCESS != code) {
@ -795,17 +792,18 @@ bool qwRetireTask(QW_FPARAMS_DEF) {
return retired;
}
bool qwRetireJob(SQWJobInfo* pJob) {
bool qwRetireJob(SQWJobInfo *pJob) {
if (NULL == pJob) {
return false;
}
bool retired = true;
void* pIter = taosHashIterate(pJob->pSessions, NULL);
bool retired = true;
void *pIter = taosHashIterate(pJob->pSessions, NULL);
while (pIter) {
SQWSessionInfo* pSession = (SQWSessionInfo*)pIter;
SQWSessionInfo *pSession = (SQWSessionInfo *)pIter;
if (!qwRetireTask((SQWorker *)pSession->mgmt, pSession->sId, pSession->qId, pSession->cId, pSession->tId, pSession->rId, pSession->eId)) {
if (!qwRetireTask((SQWorker *)pSession->mgmt, pSession->sId, pSession->qId, pSession->cId, pSession->tId,
pSession->rId, pSession->eId)) {
retired = false;
}
@ -814,4 +812,3 @@ bool qwRetireJob(SQWJobInfo* pJob) {
return retired;
}

94
source/libs/qworker/src/qworker.c
View File

@ -18,9 +18,8 @@ SQWorkerMgmt gQwMgmt = {
.qwNum = 0,
};
TdThreadOnce gQueryPoolInit = PTHREAD_ONCE_INIT;
SQueryMgmt gQueryMgmt = {0};
TdThreadOnce gQueryPoolInit = PTHREAD_ONCE_INIT;
SQueryMgmt gQueryMgmt = {0};
void qwStopAllTasks(SQWorker *mgmt) {
uint64_t qId, cId, tId, sId;
@ -53,7 +52,7 @@ int32_t qwProcessHbLinkBroken(SQWorker *mgmt, SQWMsg *qwMsg, SSchedulerHbReq *re
sch->hbBrokenTs = taosGetTimestampMs();
if (qwMsg->connInfo.handle == sch->hbConnInfo.handle) {
tmsgReleaseHandle(&sch->hbConnInfo, TAOS_CONN_SERVER);
tmsgReleaseHandle(&sch->hbConnInfo, TAOS_CONN_SERVER, 0);
sch->hbConnInfo.handle = NULL;
sch->hbConnInfo.ahandle = NULL;
@ -71,7 +70,7 @@ int32_t qwProcessHbLinkBroken(SQWorker *mgmt, SQWMsg *qwMsg, SSchedulerHbReq *re
int32_t qwHandleTaskComplete(QW_FPARAMS_DEF, SQWTaskCtx *ctx) {
qTaskInfo_t taskHandle = ctx->taskHandle;
int32_t code = TSDB_CODE_SUCCESS;
int32_t code = TSDB_CODE_SUCCESS;
ctx->queryExecDone = true;
@ -716,7 +715,8 @@ _return:
}
if (code) {
(void)qwUpdateTaskStatus(QW_FPARAMS(), JOB_TASK_STATUS_FAIL, ctx->dynamicTask); // already in error, ignore new error
(void)qwUpdateTaskStatus(QW_FPARAMS(), JOB_TASK_STATUS_FAIL,
ctx->dynamicTask); // already in error, ignore new error
}
QW_UNLOCK(QW_WRITE, &ctx->lock);
@ -795,7 +795,8 @@ int32_t qwProcessQuery(QW_FPARAMS_DEF, SQWMsg *qwMsg, char *sql) {
}
taosEnableMemPoolUsage(ctx->memPoolSession);
code = qCreateExecTask(qwMsg->node, mgmt->nodeId, tId, plan, &pTaskInfo, &sinkHandle, qwMsg->msgInfo.compressMsg, sql, OPTR_EXEC_MODEL_BATCH);
code = qCreateExecTask(qwMsg->node, mgmt->nodeId, tId, plan, &pTaskInfo, &sinkHandle, qwMsg->msgInfo.compressMsg, sql,
OPTR_EXEC_MODEL_BATCH);
taosDisableMemPoolUsage();
if (code) {
@ -915,11 +916,11 @@ int32_t qwProcessCQuery(QW_FPARAMS_DEF, SQWMsg *qwMsg) {
qwMsg->connInfo = ctx->dataConnInfo;
code = qwBuildAndSendFetchRsp(ctx->fetchMsgType + 1, &qwMsg->connInfo, NULL, 0, code);
if (TSDB_CODE_SUCCESS != code) {
QW_TASK_ELOG("fetch rsp send fail, handle:%p, code:%x - %s, dataLen:%d", qwMsg->connInfo.handle, code, tstrerror(code),
0);
QW_TASK_ELOG("fetch rsp send fail, handle:%p, code:%x - %s, dataLen:%d", qwMsg->connInfo.handle, code,
tstrerror(code), 0);
} else {
QW_TASK_DLOG("fetch rsp send, handle:%p, code:%x - %s, dataLen:%d", qwMsg->connInfo.handle, code, tstrerror(code),
0);
QW_TASK_DLOG("fetch rsp send, handle:%p, code:%x - %s, dataLen:%d", qwMsg->connInfo.handle, code,
tstrerror(code), 0);
}
}
@ -1024,8 +1025,8 @@ _return:
if (!rsped) {
code = qwBuildAndSendFetchRsp(qwMsg->msgType + 1, &qwMsg->connInfo, rsp, dataLen, code);
if (TSDB_CODE_SUCCESS != code) {
QW_TASK_ELOG("fetch rsp send fail, msgType:%s, handle:%p, code:%x - %s, dataLen:%d", TMSG_INFO(qwMsg->msgType + 1),
qwMsg->connInfo.handle, code, tstrerror(code), dataLen);
QW_TASK_ELOG("fetch rsp send fail, msgType:%s, handle:%p, code:%x - %s, dataLen:%d",
TMSG_INFO(qwMsg->msgType + 1), qwMsg->connInfo.handle, code, tstrerror(code), dataLen);
} else {
QW_TASK_DLOG("fetch rsp send, msgType:%s, handle:%p, code:%x - %s, dataLen:%d", TMSG_INFO(qwMsg->msgType + 1),
qwMsg->connInfo.handle, code, tstrerror(code), dataLen);
@ -1076,9 +1077,10 @@ _return:
if (code) {
if (ctx) {
QW_UPDATE_RSP_CODE(ctx, code);
(void)qwUpdateTaskStatus(QW_FPARAMS(), JOB_TASK_STATUS_FAIL, ctx->dynamicTask); // task already failed, no more error handling
(void)qwUpdateTaskStatus(QW_FPARAMS(), JOB_TASK_STATUS_FAIL,
ctx->dynamicTask); // task already failed, no more error handling
} else {
tmsgReleaseHandle(&qwMsg->connInfo, TAOS_CONN_SERVER);
tmsgReleaseHandle(&qwMsg->connInfo, TAOS_CONN_SERVER, 0);
}
}
@ -1088,7 +1090,7 @@ _return:
if (ctx) {
if (qwMsg->connInfo.handle != ctx->ctrlConnInfo.handle) {
tmsgReleaseHandle(&qwMsg->connInfo, TAOS_CONN_SERVER);
tmsgReleaseHandle(&qwMsg->connInfo, TAOS_CONN_SERVER, 0);
}
qwReleaseTaskCtx(mgmt, ctx);
@ -1130,7 +1132,8 @@ _return:
if (code) {
if (ctx) {
QW_UPDATE_RSP_CODE(ctx, code);
(void)qwUpdateTaskStatus(QW_FPARAMS(), JOB_TASK_STATUS_FAIL, ctx->dynamicTask); // task already failed, no more error handling
(void)qwUpdateTaskStatus(QW_FPARAMS(), JOB_TASK_STATUS_FAIL,
ctx->dynamicTask); // task already failed, no more error handling
}
}
@ -1162,7 +1165,7 @@ int32_t qwProcessHb(SQWorker *mgmt, SQWMsg *qwMsg, SSchedulerHbReq *req) {
QW_LOCK(QW_WRITE, &sch->hbConnLock);
if (sch->hbConnInfo.handle) {
tmsgReleaseHandle(&sch->hbConnInfo, TAOS_CONN_SERVER);
tmsgReleaseHandle(&sch->hbConnInfo, TAOS_CONN_SERVER, 0);
sch->hbConnInfo.handle = NULL;
}
@ -1171,8 +1174,9 @@ int32_t qwProcessHb(SQWorker *mgmt, SQWMsg *qwMsg, SSchedulerHbReq *req) {
QW_UNLOCK(QW_WRITE, &sch->hbConnLock);
QW_DLOG("hb connection updated, clientId:%" PRIx64 ", nodeId:%d, fqdn:%s, port:%d, handle:%p, ahandle:%p", req->clientId,
req->epId.nodeId, req->epId.ep.fqdn, req->epId.ep.port, qwMsg->connInfo.handle, qwMsg->connInfo.ahandle);
QW_DLOG("hb connection updated, clientId:%" PRIx64 ", nodeId:%d, fqdn:%s, port:%d, handle:%p, ahandle:%p",
req->clientId, req->epId.nodeId, req->epId.ep.fqdn, req->epId.ep.port, qwMsg->connInfo.handle,
qwMsg->connInfo.ahandle);
qwReleaseScheduler(QW_READ, mgmt);
@ -1182,7 +1186,7 @@ _return:
code = qwBuildAndSendHbRsp(&qwMsg->connInfo, &rsp, code);
if (code) {
tmsgReleaseHandle(&qwMsg->connInfo, TAOS_CONN_SERVER);
tmsgReleaseHandle(&qwMsg->connInfo, TAOS_CONN_SERVER, 0);
qwMsg->connInfo.handle = NULL;
}
@ -1223,7 +1227,7 @@ void qwProcessHbTimerEvent(void *param, void *tmrId) {
if (taosTmrReset(qwProcessHbTimerEvent, QW_DEFAULT_HEARTBEAT_MSEC, param, mgmt->timer, &mgmt->hbTimer)) {
qError("reset qworker hb timer error, timer stoppped");
}
(void)qwRelease(refId); // ignore error
(void)qwRelease(refId); // ignore error
return;
}
@ -1237,7 +1241,7 @@ void qwProcessHbTimerEvent(void *param, void *tmrId) {
if (taosTmrReset(qwProcessHbTimerEvent, QW_DEFAULT_HEARTBEAT_MSEC, param, mgmt->timer, &mgmt->hbTimer)) {
qError("reset qworker hb timer error, timer stoppped");
}
(void)qwRelease(refId); // ignore error
(void)qwRelease(refId); // ignore error
return;
}
@ -1279,7 +1283,7 @@ _return:
QW_UNLOCK(QW_READ, &mgmt->schLock);
for (int32_t j = 0; j < i; ++j) {
(void)qwBuildAndSendHbRsp(&rspList[j].connInfo, &rspList[j].rsp, code); // ignore error
(void)qwBuildAndSendHbRsp(&rspList[j].connInfo, &rspList[j].rsp, code); // ignore error
/*QW_DLOG("hb rsp send, handle:%p, code:%x - %s, taskNum:%d", rspList[j].connInfo.handle, code, tstrerror(code),*/
/*(rspList[j].rsp.taskStatus ? (int32_t)taosArrayGetSize(rspList[j].rsp.taskStatus) : 0));*/
tFreeSSchedulerHbRsp(&rspList[j].rsp);
@ -1296,7 +1300,7 @@ _return:
qError("reset qworker hb timer error, timer stoppped");
}
(void)qwRelease(refId); // ignore error
(void)qwRelease(refId); // ignore error
}
int32_t qwProcessDelete(QW_FPARAMS_DEF, SQWMsg *qwMsg, SDeleteRes *pRes) {
@ -1432,7 +1436,7 @@ int32_t qWorkerInit(int8_t nodeType, int32_t nodeId, void **qWorkerMgmt, const S
_return:
if (mgmt->refId >= 0) {
(void)qwRelease(mgmt->refId); // ignore error
(void)qwRelease(mgmt->refId); // ignore error
} else {
taosHashCleanup(mgmt->schHash);
taosHashCleanup(mgmt->ctxHash);
@ -1622,54 +1626,56 @@ _return:
QW_RET(code);
}
void qWorkerRetireJob(uint64_t jobId, uint64_t clientId, int32_t errCode) {
char id[sizeof(jobId) + sizeof(clientId) + 1] = {0};
QW_SET_QCID(id, jobId, clientId);
SQWJobInfo* pJob = (SQWJobInfo*)taosHashGet(gQueryMgmt.pJobInfo, id, sizeof(id));
SQWJobInfo *pJob = (SQWJobInfo *)taosHashGet(gQueryMgmt.pJobInfo, id, sizeof(id));
if (NULL == pJob) {
qError("QID:0x%" PRIx64 " CID:0x%" PRIx64 " fail to get job from job hash", jobId, clientId);
return;
}
if (0 == atomic_val_compare_exchange_32(&pJob->errCode, 0, errCode) && 0 == atomic_val_compare_exchange_8(&pJob->retired, 0, 1)) {
qDebug("QID:0x%" PRIx64 " CID:0x%" PRIx64 " mark retired, errCode: 0x%x, allocSize:%" PRId64,
jobId, clientId, errCode, atomic_load_64(&pJob->memInfo->allocMemSize));
if (0 == atomic_val_compare_exchange_32(&pJob->errCode, 0, errCode) &&
0 == atomic_val_compare_exchange_8(&pJob->retired, 0, 1)) {
qDebug("QID:0x%" PRIx64 " CID:0x%" PRIx64 " mark retired, errCode: 0x%x, allocSize:%" PRId64, jobId, clientId,
errCode, atomic_load_64(&pJob->memInfo->allocMemSize));
(void)qwRetireJob(pJob);
} else {
qDebug("QID:0x%" PRIx64 " already retired, retired: %d, errCode: 0x%x, allocSize:%" PRId64, jobId, atomic_load_8(&pJob->retired), atomic_load_32(&pJob->errCode), atomic_load_64(&pJob->memInfo->allocMemSize));
qDebug("QID:0x%" PRIx64 " already retired, retired: %d, errCode: 0x%x, allocSize:%" PRId64, jobId,
atomic_load_8(&pJob->retired), atomic_load_32(&pJob->errCode), atomic_load_64(&pJob->memInfo->allocMemSize));
}
}
void qWorkerRetireJobs(int64_t retireSize, int32_t errCode) {
SQWJobInfo* pJob = (SQWJobInfo*)taosHashIterate(gQueryMgmt.pJobInfo, NULL);
int32_t jobNum = 0;
int64_t retiredSize = 0;
SQWJobInfo *pJob = (SQWJobInfo *)taosHashIterate(gQueryMgmt.pJobInfo, NULL);
int32_t jobNum = 0;
int64_t retiredSize = 0;
while (retiredSize < retireSize && NULL != pJob) {
if (atomic_load_8(&pJob->retired)) {
pJob = (SQWJobInfo*)taosHashIterate(gQueryMgmt.pJobInfo, pJob);
pJob = (SQWJobInfo *)taosHashIterate(gQueryMgmt.pJobInfo, pJob);
continue;
}
if (0 == atomic_val_compare_exchange_32(&pJob->errCode, 0, errCode) && 0 == atomic_val_compare_exchange_8(&pJob->retired, 0, 1)) {
if (0 == atomic_val_compare_exchange_32(&pJob->errCode, 0, errCode) &&
0 == atomic_val_compare_exchange_8(&pJob->retired, 0, 1)) {
int64_t aSize = atomic_load_64(&pJob->memInfo->allocMemSize);
bool retired = qwRetireJob(pJob);
bool retired = qwRetireJob(pJob);
if (retired) {
retiredSize += aSize;
}
jobNum++;
qDebug("QID:0x%" PRIx64 " CID:0x%" PRIx64 " job mark retired in batch, retired:%d, usedSize:%" PRId64 ", retireSize:%" PRId64,
pJob->memInfo->jobId, pJob->memInfo->clientId, retired, aSize, retireSize);
qDebug("QID:0x%" PRIx64 " CID:0x%" PRIx64 " job mark retired in batch, retired:%d, usedSize:%" PRId64
", retireSize:%" PRId64,
pJob->memInfo->jobId, pJob->memInfo->clientId, retired, aSize, retireSize);
}
pJob = (SQWJobInfo*)taosHashIterate(gQueryMgmt.pJobInfo, pJob);
pJob = (SQWJobInfo *)taosHashIterate(gQueryMgmt.pJobInfo, pJob);
}
qDebug("total %d jobs mark retired, direct retiredSize:%" PRId64 " targetRetireSize:%" PRId64, jobNum, retiredSize, retireSize);
qDebug("total %d jobs mark retired, direct retiredSize:%" PRId64 " targetRetireSize:%" PRId64, jobNum, retiredSize,
retireSize);
}

6
source/libs/scheduler/src/schRemote.c
View File

@ -509,7 +509,7 @@ int32_t schHandleDropCallback(void *param, SDataBuf *pMsg, int32_t code) {
qDebug("QID:0x%" PRIx64 ",CID:0x%" PRIx64 ",TID:0x%" PRIx64 " drop task rsp received, code:0x%x", pParam->queryId,
pParam->clientId, pParam->taskId, code);
// called if drop task rsp received code
(void)rpcReleaseHandle(pMsg->handle, TAOS_CONN_CLIENT); // ignore error
(void)rpcReleaseHandle(pMsg->handle, TAOS_CONN_CLIENT, 0); // ignore error
if (pMsg->handle == NULL) {
qError("sch handle is NULL, may be already released and mem lea");
@ -535,7 +535,7 @@ int32_t schHandleNotifyCallback(void *param, SDataBuf *pMsg, int32_t code) {
int32_t schHandleLinkBrokenCallback(void *param, SDataBuf *pMsg, int32_t code) {
SSchCallbackParamHeader *head = (SSchCallbackParamHeader *)param;
(void)rpcReleaseHandle(pMsg->handle, TAOS_CONN_CLIENT); // ignore error
(void)rpcReleaseHandle(pMsg->handle, TAOS_CONN_CLIENT, 0); // ignore error
qDebug("handle %p is broken", pMsg->handle);
@ -565,7 +565,7 @@ int32_t schHandleHbCallback(void *param, SDataBuf *pMsg, int32_t code) {
if (code) {
qError("hb rsp error:%s", tstrerror(code));
(void)rpcReleaseHandle(pMsg->handle, TAOS_CONN_CLIENT); // ignore error
(void)rpcReleaseHandle(pMsg->handle, TAOS_CONN_CLIENT, 0); // ignore error
SCH_ERR_JRET(code);
}

37
source/libs/scheduler/src/schUtil.c
View File

@ -22,7 +22,7 @@
#include "tref.h"
#include "trpc.h"
FORCE_INLINE int32_t schAcquireJob(int64_t refId, SSchJob** ppJob) {
FORCE_INLINE int32_t schAcquireJob(int64_t refId, SSchJob **ppJob) {
qDebug("sch acquire jobId:0x%" PRIx64, refId);
*ppJob = (SSchJob *)taosAcquireRef(schMgmt.jobRef, refId);
if (NULL == *ppJob) {
@ -41,7 +41,7 @@ FORCE_INLINE int32_t schReleaseJob(int64_t refId) {
return taosReleaseRef(schMgmt.jobRef, refId);
}
FORCE_INLINE int32_t schReleaseJobEx(int64_t refId, int32_t* released) {
FORCE_INLINE int32_t schReleaseJobEx(int64_t refId, int32_t *released) {
if (0 == refId) {
return TSDB_CODE_SUCCESS;
}
@ -50,7 +50,7 @@ FORCE_INLINE int32_t schReleaseJobEx(int64_t refId, int32_t* released) {
return taosReleaseRefEx(schMgmt.jobRef, refId, released);
}
int32_t schDumpEpSet(SEpSet *pEpSet, char** ppRes) {
int32_t schDumpEpSet(SEpSet *pEpSet, char **ppRes) {
*ppRes = NULL;
if (NULL == pEpSet) {
return TSDB_CODE_SUCCESS;
@ -89,7 +89,7 @@ char *schGetOpStr(SCH_OP_TYPE type) {
}
void schFreeHbTrans(SSchHbTrans *pTrans) {
(void)rpcReleaseHandle((void *)pTrans->trans.pHandleId, TAOS_CONN_CLIENT);
(void)rpcReleaseHandle((void *)pTrans->trans.pHandleId, TAOS_CONN_CLIENT, 0);
schFreeRpcCtx(&pTrans->rpcCtx);
}
@ -202,11 +202,12 @@ void schDeregisterTaskHb(SSchJob *pJob, SSchTask *pTask) {
SQueryNodeAddr *addr = taosArrayGet(pTask->candidateAddrs, pTask->candidateIdx);
if (NULL == addr) {
SCH_TASK_ELOG("fail to get the %dth condidateAddr in task, totalNum:%d", pTask->candidateIdx, (int32_t)taosArrayGetSize(pTask->candidateAddrs));
SCH_TASK_ELOG("fail to get the %dth condidateAddr in task, totalNum:%d", pTask->candidateIdx,
(int32_t)taosArrayGetSize(pTask->candidateAddrs));
return;
}
SQueryNodeEpId epId = {0};
SQueryNodeEpId epId = {0};
epId.nodeId = addr->nodeId;
@ -240,11 +241,12 @@ int32_t schEnsureHbConnection(SSchJob *pJob, SSchTask *pTask) {
SQueryNodeAddr *addr = taosArrayGet(pTask->candidateAddrs, pTask->candidateIdx);
if (NULL == addr) {
SCH_TASK_ELOG("fail to get the %dth condidateAddr in task, totalNum:%d", pTask->candidateIdx, (int32_t)taosArrayGetSize(pTask->candidateAddrs));
SCH_TASK_ELOG("fail to get the %dth condidateAddr in task, totalNum:%d", pTask->candidateIdx,
(int32_t)taosArrayGetSize(pTask->candidateAddrs));
return TSDB_CODE_SCH_INTERNAL_ERROR;
}
SQueryNodeEpId epId = {0};
SQueryNodeEpId epId = {0};
epId.nodeId = addr->nodeId;
@ -276,8 +278,8 @@ int32_t schUpdateHbConnection(SQueryNodeEpId *epId, SSchTrans *trans) {
SCH_UNLOCK(SCH_WRITE, &hb->lock);
SCH_UNLOCK(SCH_READ, &schMgmt.hbLock);
qDebug("hb connection updated, nodeId:%d, fqdn:%s, port:%d, pTrans:%p, pHandle:%p",
epId->nodeId, epId->ep.fqdn, epId->ep.port, trans->pTrans, trans->pHandle);
qDebug("hb connection updated, nodeId:%d, fqdn:%s, port:%d, pTrans:%p, pHandle:%p", epId->nodeId, epId->ep.fqdn,
epId->ep.port, trans->pTrans, trans->pHandle);
return TSDB_CODE_SUCCESS;
}
@ -323,7 +325,8 @@ uint64_t schGenUUID(void) {
uint64_t pid = taosGetPId();
int32_t val = atomic_add_fetch_32(&requestSerialId, 1);
uint64_t id = ((uint64_t)((hashId & 0x0FFF)) << 52) | ((pid & 0x0FFF) << 40) | ((ts & 0xFFFFFF) << 16) | (val & 0xFFFF);
uint64_t id =
((uint64_t)((hashId & 0x0FFF)) << 52) | ((pid & 0x0FFF) << 40) | ((ts & 0xFFFFFF) << 16) | (val & 0xFFFF);
return id;
}
#endif
@ -373,7 +376,7 @@ void schGetTaskFromList(SHashObj *pTaskList, uint64_t taskId, SSchTask **pTask)
*pTask = *task;
}
int32_t schValidateSubplan(SSchJob *pJob, SSubplan* pSubplan, int32_t level, int32_t idx, int32_t taskNum) {
int32_t schValidateSubplan(SSchJob *pJob, SSubplan *pSubplan, int32_t level, int32_t idx, int32_t taskNum) {
if (NULL == pSubplan) {
SCH_JOB_ELOG("fail to get the %dth subplan, taskNum: %d, level: %d", idx, taskNum, level);
SCH_ERR_RET(TSDB_CODE_QRY_INVALID_INPUT);
@ -396,11 +399,13 @@ int32_t schValidateSubplan(SSchJob *pJob, SSubplan* pSubplan, int32_t level, int
if (SCH_IS_DATA_BIND_PLAN(pSubplan)) {
if (pSubplan->execNode.epSet.numOfEps <= 0) {
SCH_JOB_ELOG("no execNode specifed for data src plan %d, numOfEps:%d", pSubplan->subplanType, pSubplan->execNode.epSet.numOfEps);
SCH_JOB_ELOG("no execNode specifed for data src plan %d, numOfEps:%d", pSubplan->subplanType,
pSubplan->execNode.epSet.numOfEps);
SCH_ERR_RET(TSDB_CODE_SCH_DATA_SRC_EP_MISS);
}
if (pSubplan->execNode.epSet.inUse >= pSubplan->execNode.epSet.numOfEps) {
SCH_JOB_ELOG("invalid epset inUse %d for data src plan %d, numOfEps:%d", pSubplan->execNode.epSet.inUse, pSubplan->subplanType, pSubplan->execNode.epSet.numOfEps);
SCH_JOB_ELOG("invalid epset inUse %d for data src plan %d, numOfEps:%d", pSubplan->execNode.epSet.inUse,
pSubplan->subplanType, pSubplan->execNode.epSet.numOfEps);
SCH_ERR_RET(TSDB_CODE_QRY_INVALID_INPUT);
}
}
@ -418,7 +423,7 @@ int32_t schValidateSubplan(SSchJob *pJob, SSubplan* pSubplan, int32_t level, int
return TSDB_CODE_SUCCESS;
}
void schStopTaskDelayTimer(SSchJob *pJob, SSchTask* pTask, bool syncOp) {
void schStopTaskDelayTimer(SSchJob *pJob, SSchTask *pTask, bool syncOp) {
if (!taosTmrStopA(&pTask->delayTimer)) {
if (syncOp) {
while (!taosTmrIsStopped(&pTask->delayTimer)) {
@ -429,5 +434,3 @@ void schStopTaskDelayTimer(SSchJob *pJob, SSchTask* pTask, bool syncOp) {
}
}
}

4
source/libs/transport/inc/transComm.h
View File

@ -310,8 +310,8 @@ void transRefCliHandle(void* handle);
int32_t transUnrefCliHandle(void* handle);
int32_t transGetRefCount(void* handle);
int32_t transReleaseCliHandle(void* handle);
int32_t transReleaseSrvHandle(void* handle);
int32_t transReleaseCliHandle(void* handle, int32_t status);
int32_t transReleaseSrvHandle(void* handle, int32_t status);
int32_t transSendRequest(void* pInit, const SEpSet* pEpSet, STransMsg* pReq, STransCtx* pCtx);
int32_t transSendRecv(void* pInit, const SEpSet* pEpSet, STransMsg* pReq, STransMsg* pRsp);

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

@ -69,7 +69,9 @@ void tmsgSendRsp(SRpcMsg* pMsg) {
void tmsgRegisterBrokenLinkArg(SRpcMsg* pMsg) { (*defaultMsgCb.registerBrokenLinkArgFp)(pMsg); }
void tmsgReleaseHandle(SRpcHandleInfo* pHandle, int8_t type) { (*defaultMsgCb.releaseHandleFp)(pHandle, type); }
void tmsgReleaseHandle(SRpcHandleInfo* pHandle, int8_t type, int32_t status) {
(*defaultMsgCb.releaseHandleFp)(pHandle, type, status);
}
void tmsgReportStartup(const char* name, const char* desc) { (*defaultMsgCb.reportStartupFp)(name, desc); }

6
source/libs/transport/src/trans.c
View File

@ -23,7 +23,7 @@ void (*taosCloseHandle[])(void* arg) = {transCloseServer, transCloseClient};
void (*taosRefHandle[])(void* handle) = {transRefSrvHandle, transRefCliHandle};
void (*taosUnRefHandle[])(void* handle) = {transUnrefSrvHandle, NULL};
int (*transReleaseHandle[])(void* handle) = {transReleaseSrvHandle, transReleaseCliHandle};
int (*transReleaseHandle[])(void* handle, int32_t status) = {transReleaseSrvHandle, transReleaseCliHandle};
static int32_t transValidLocalFqdn(const char* localFqdn, uint32_t* ip) {
int32_t code = taosGetIpv4FromFqdn(localFqdn, ip);
@ -217,7 +217,9 @@ void rpcRefHandle(void* handle, int8_t type) { (*taosRefHandle[type])(handle); }
void rpcUnrefHandle(void* handle, int8_t type) { (*taosUnRefHandle[type])(handle); }
int32_t rpcRegisterBrokenLinkArg(SRpcMsg* msg) { return transRegisterMsg(msg); }
int32_t rpcReleaseHandle(void* handle, int8_t type) { return (*transReleaseHandle[type])(handle); }
int32_t rpcReleaseHandle(void* handle, int8_t type, int32_t status) {
return (*transReleaseHandle[type])(handle, status);
}
// client only
int32_t rpcSetDefaultAddr(void* thandle, const char* ip, const char* fqdn) {

53
source/libs/transport/src/transCli.c
View File

@ -555,9 +555,40 @@ int32_t cliBuildRespFromCont(SCliReq* pReq, STransMsg* pResp, STransMsgHead* pHe
pResp->info.handle = (void*)qid;
return 0;
}
int8_t cliMayNotifyUserOnRecvReleaseExcept(SCliConn* conn, STransMsgHead* pHead, SCliReq* pReq) {
int32_t code = 0;
if (pHead->code == 0 || pHead->msgType != TDMT_SCH_TASK_RELEASE) {
return 0;
}
// no ahandle, no need to notify user
if (pReq == NULL || pReq->ctx == NULL || pReq->ctx->ahandle == NULL) {
return 0;
}
SCliThrd* pThrd = conn->hostThrd;
STransMsg resp = {.code = pHead->code};
int64_t qId = taosHton64(pHead->qid);
STraceId* trace = &pHead->traceId;
int64_t seqNum = taosHton64(pHead->seqNum);
code = cliBuildExceptResp(pThrd, pReq, &resp);
if (code != 0) {
tGWarn("%s conn %p failed to build except resp for req:%" PRId64 " since %s", CONN_GET_INST_LABEL(conn), conn, qId,
tstrerror(code));
}
code = cliNotifyCb(conn, NULL, &resp);
if (code != 0) {
tGWarn("%s conn %p failed to notify user for req:%" PRId64 " since %s", CONN_GET_INST_LABEL(conn), conn, qId,
tstrerror(code));
}
destroyReq(pReq);
return 1;
}
int32_t cliHandleState_mayHandleReleaseResp(SCliConn* conn, STransMsgHead* pHead) {
int32_t code = 0;
SCliThrd* pThrd = conn->hostThrd;
int8_t notifyUser = 0;
if (pHead->msgType == TDMT_SCH_TASK_RELEASE || pHead->msgType == TDMT_SCH_TASK_RELEASE + 1) {
int64_t qId = taosHton64(pHead->qid);
STraceId* trace = &pHead->traceId;
@ -597,7 +628,11 @@ int32_t cliHandleState_mayHandleReleaseResp(SCliConn* conn, STransMsgHead* pHead
removeReqFromSendQ(pReq);
STraceId* trace = &pReq->msg.info.traceId;
tGDebug("start to free msg %p", pReq);
destroyReqWrapper(pReq, pThrd);
if (cliMayNotifyUserOnRecvReleaseExcept(conn, pHead, pReq)) {
} else {
destroyReqWrapper(pReq, pThrd);
}
}
taosMemoryFree(pHead);
return 1;
@ -1259,7 +1294,8 @@ static void cliHandleException(SCliConn* conn) {
if (conn->registered) {
int8_t ref = transGetRefCount(conn);
if (ref == 0 && !uv_is_closing((uv_handle_t*)conn->stream)) {
tTrace("%s conn %p fd %d,%d,%d,%p uv_closed", CONN_GET_INST_LABEL(conn), conn, conn->stream->u.fd, conn->stream->io_watcher.fd, conn->stream->accepted_fd, conn->stream->queued_fds);
tTrace("%s conn %p fd %d,%d,%d,%p uv_closed", CONN_GET_INST_LABEL(conn), conn, conn->stream->u.fd,
conn->stream->io_watcher.fd, conn->stream->accepted_fd, conn->stream->queued_fds);
uv_close((uv_handle_t*)conn->stream, cliDestroy);
}
}
@ -1785,9 +1821,7 @@ FORCE_INLINE int32_t cliBuildExceptResp(SCliThrd* pThrd, SCliReq* pReq, STransMs
STrans* pInst = pThrd->pInst;
SReqCtx* pCtx = pReq ? pReq->ctx : NULL;
STransMsg resp = {0};
// resp.code = (conn->connnected ? TSDB_CODE_RPC_BROKEN_LINK : TSDB_CODE_RPC_NETWORK_UNAVAIL);
SReqCtx* pCtx = pReq ? pReq->ctx : NULL;
pResp->msgType = pReq ? pReq->msg.msgType + 1 : 0;
pResp->info.cliVer = pInst->compatibilityVer;
pResp->info.ahandle = pCtx ? pCtx->ahandle : 0;
@ -3103,15 +3137,18 @@ SCliThrd* transGetWorkThrd(STrans* trans, int64_t handle) {
SCliThrd* pThrd = transGetWorkThrdFromHandle(trans, handle);
return pThrd;
}
int32_t transReleaseCliHandle(void* handle) {
int32_t transReleaseCliHandle(void* handle, int32_t status) {
int32_t code = 0;
SCliThrd* pThrd = transGetWorkThrdFromHandle(NULL, (int64_t)handle);
if (pThrd == NULL) {
return TSDB_CODE_RPC_BROKEN_LINK;
}
STransMsg tmsg = {
.msgType = TDMT_SCH_TASK_RELEASE, .info.handle = handle, .info.ahandle = (void*)0, .info.qId = (int64_t)handle};
STransMsg tmsg = {.msgType = TDMT_SCH_TASK_RELEASE,
.info.handle = handle,
.info.ahandle = (void*)0,
.info.qId = (int64_t)handle,
code = status};
TRACE_SET_MSGID(&tmsg.info.traceId, tGenIdPI64());

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

@ -1858,7 +1858,7 @@ void transUnrefSrvHandle(void* handle) {
}
}
int32_t transReleaseSrvHandle(void* handle) {
int32_t transReleaseSrvHandle(void* handle, int32_t status) {
int32_t code = 0;
SRpcHandleInfo* info = handle;
SExHandle* exh = info->handle;
@ -1871,7 +1871,7 @@ int32_t transReleaseSrvHandle(void* handle) {
ASYNC_ERR_JRET(pThrd);
STransMsg tmsg = {.msgType = TDMT_SCH_TASK_RELEASE,
.code = 0,
.code = status,
.info.handle = exh,
.info.ahandle = NULL,
.info.refId = refId,

6
source/libs/transport/test/transUT.cpp
View File

@ -96,7 +96,7 @@ class Client {
}
void SendAndRecvNoHandle(SRpcMsg *req, SRpcMsg *resp) {
if (req->info.handle != NULL) {
rpcReleaseHandle(req->info.handle, TAOS_CONN_CLIENT);
rpcReleaseHandle(req->info.handle, TAOS_CONN_CLIENT, 0);
req->info.handle = NULL;
}
SendAndRecv(req, resp);
@ -191,7 +191,7 @@ static void processReleaseHandleCb(void *parent, SRpcMsg *pMsg, SEpSet *pEpSet)
rpcMsg.code = 0;
rpcSendResponse(&rpcMsg);
rpcReleaseHandle(&pMsg->info, TAOS_CONN_SERVER);
rpcReleaseHandle(&pMsg->info, TAOS_CONN_SERVER, 0);
}
static void processRegisterFailure(void *parent, SRpcMsg *pMsg, SEpSet *pEpSet) {
{
@ -366,7 +366,7 @@ TEST_F(TransEnv, cliPersistHandle) {
//}
handle = resp.info.handle;
}
rpcReleaseHandle(handle, TAOS_CONN_CLIENT);
rpcReleaseHandle(handle, TAOS_CONN_CLIENT, 0);
for (int i = 0; i < 10; i++) {
SRpcMsg req = {0};
req.msgType = 1;

4
source/libs/transport/test/transUT2.cpp
View File

@ -106,7 +106,7 @@ class Client {
}
void SendAndRecvNoHandle(SRpcMsg *req, SRpcMsg *resp) {
if (req->info.handle != NULL) {
rpcReleaseHandle(req->info.handle, TAOS_CONN_CLIENT);
rpcReleaseHandle(req->info.handle, TAOS_CONN_CLIENT, 0);
req->info.handle = NULL;
}
SendAndRecv(req, resp);
@ -201,7 +201,7 @@ static void processReleaseHandleCb(void *parent, SRpcMsg *pMsg, SEpSet *pEpSet)
rpcMsg.code = 0;
rpcSendResponse(&rpcMsg);
rpcReleaseHandle(&pMsg->info, TAOS_CONN_SERVER);
rpcReleaseHandle(&pMsg->info, TAOS_CONN_SERVER, 0);
}
static void processRegisterFailure(void *parent, SRpcMsg *pMsg, SEpSet *pEpSet) {
// {