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

fix a bug

This commit is contained in:
Hongze Cheng 2022-04-15 07:49:33 +00:00
parent 98b98c8b85
commit bc178c247d
1 changed files with 37 additions and 38 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

75
source/libs/scheduler/src/scheduler.c
View File

@ -14,12 +14,12 @@
*/ */
#include "catalog.h" #include "catalog.h"
#include "command.h"
#include "query.h" #include "query.h"
#include "schedulerInt.h" #include "schedulerInt.h"
#include "tmsg.h" #include "tmsg.h"
#include "tref.h" #include "tref.h"
#include "trpc.h" #include "trpc.h"
#include "command.h"
SSchedulerMgmt schMgmt = {0}; SSchedulerMgmt schMgmt = {0};
@ -68,8 +68,8 @@ int32_t schInitTask(SSchJob *pJob, SSchTask *pTask, SSubplan *pPlan, SSchLevel *
} }
int32_t schInitJob(SSchJob **pSchJob, SQueryPlan *pDag, void *transport, SArray *pNodeList, const char *sql, int32_t schInitJob(SSchJob **pSchJob, SQueryPlan *pDag, void *transport, SArray *pNodeList, const char *sql,
int64_t startTs, bool syncSchedule) { int64_t startTs, bool syncSchedule) {
int32_t code = 0; int32_t code = 0;
SSchJob *pJob = taosMemoryCalloc(1, sizeof(SSchJob)); SSchJob *pJob = taosMemoryCalloc(1, sizeof(SSchJob));
if (NULL == pJob) { if (NULL == pJob) {
qError("QID:%" PRIx64 " calloc %d failed", pDag->queryId, (int32_t)sizeof(SSchJob)); qError("QID:%" PRIx64 " calloc %d failed", pDag->queryId, (int32_t)sizeof(SSchJob));
@ -141,7 +141,6 @@ _return:
SCH_RET(code); SCH_RET(code);
} }
void schFreeRpcCtx(SRpcCtx *pCtx) { void schFreeRpcCtx(SRpcCtx *pCtx) {
if (NULL == pCtx) { if (NULL == pCtx) {
return; return;
@ -1047,12 +1046,12 @@ _return:
int32_t schProcessOnExplainDone(SSchJob *pJob, SSchTask *pTask, SRetrieveTableRsp *pRsp) { int32_t schProcessOnExplainDone(SSchJob *pJob, SSchTask *pTask, SRetrieveTableRsp *pRsp) {
SCH_TASK_DLOG("got explain rsp, rows:%d, complete:%d", htonl(pRsp->numOfRows), pRsp->completed); SCH_TASK_DLOG("got explain rsp, rows:%d, complete:%d", htonl(pRsp->numOfRows), pRsp->completed);
atomic_store_32(&pJob->resNumOfRows, htonl(pRsp->numOfRows)); atomic_store_32(&pJob->resNumOfRows, htonl(pRsp->numOfRows));
atomic_store_ptr(&pJob->resData, pRsp); atomic_store_ptr(&pJob->resData, pRsp);
SCH_SET_TASK_STATUS(pTask, JOB_TASK_STATUS_SUCCEED); SCH_SET_TASK_STATUS(pTask, JOB_TASK_STATUS_SUCCEED);
schProcessOnDataFetched(pJob); schProcessOnDataFetched(pJob);
return TSDB_CODE_SUCCESS; return TSDB_CODE_SUCCESS;
@ -1146,7 +1145,7 @@ int32_t schHandleResponseMsg(SSchJob *pJob, SSchTask *pTask, int32_t msgType, ch
if (NULL == msg) { if (NULL == msg) {
SCH_ERR_JRET(TSDB_CODE_QRY_INVALID_INPUT); SCH_ERR_JRET(TSDB_CODE_QRY_INVALID_INPUT);
} }
if (!SCH_IS_EXPLAIN_JOB(pJob)) { if (!SCH_IS_EXPLAIN_JOB(pJob)) {
SCH_TASK_ELOG("invalid msg received for none explain query, msg type:%s", TMSG_INFO(msgType)); SCH_TASK_ELOG("invalid msg received for none explain query, msg type:%s", TMSG_INFO(msgType));
SCH_ERR_JRET(TSDB_CODE_QRY_INVALID_INPUT); SCH_ERR_JRET(TSDB_CODE_QRY_INVALID_INPUT);
@ -1180,13 +1179,13 @@ int32_t schHandleResponseMsg(SSchJob *pJob, SSchTask *pTask, int32_t msgType, ch
} }
if (SCH_IS_EXPLAIN_JOB(pJob)) { if (SCH_IS_EXPLAIN_JOB(pJob)) {
if (rsp->completed) { if (rsp->completed) {
SRetrieveTableRsp *pRsp = NULL; SRetrieveTableRsp *pRsp = NULL;
SCH_ERR_JRET(qExecExplainEnd(pJob->explainCtx, &pRsp)); SCH_ERR_JRET(qExecExplainEnd(pJob->explainCtx, &pRsp));
if (pRsp) { if (pRsp) {
SCH_ERR_JRET(schProcessOnExplainDone(pJob, pTask, pRsp)); SCH_ERR_JRET(schProcessOnExplainDone(pJob, pTask, pRsp));
} }
return TSDB_CODE_SUCCESS; return TSDB_CODE_SUCCESS;
} }
@ -1238,23 +1237,24 @@ _return:
} }
int32_t schGetTaskFromTaskList(SHashObj *pTaskList, uint64_t taskId, SSchTask **pTask) { int32_t schGetTaskFromTaskList(SHashObj *pTaskList, uint64_t taskId, SSchTask **pTask) {
int32_t s = taosHashGetSize(pTaskList); int32_t s = taosHashGetSize(pTaskList);
if (s <= 0) { if (s <= 0) {
return TSDB_CODE_SUCCESS; return TSDB_CODE_SUCCESS;
} }
SSchTask **task = taosHashGet(pTaskList, &taskId, sizeof(taskId));
if (NULL == task || NULL == (*task)) {
return TSDB_CODE_SUCCESS;
}
*pTask = *task; SSchTask **task = taosHashGet(pTaskList, &taskId, sizeof(taskId));
if (NULL == task || NULL == (*task)) {
return TSDB_CODE_SUCCESS;
}
return TSDB_CODE_SUCCESS; *pTask = *task;
return TSDB_CODE_SUCCESS;
} }
int32_t schUpdateTaskExecNodeHandle(SSchTask *pTask, void *handle, int32_t rspCode) { int32_t schUpdateTaskExecNodeHandle(SSchTask *pTask, void *handle, int32_t rspCode) {
if (rspCode || NULL == pTask->execNodes || taosArrayGetSize(pTask->execNodes) > 1 || taosArrayGetSize(pTask->execNodes) <= 0) { if (rspCode || NULL == pTask->execNodes || taosArrayGetSize(pTask->execNodes) > 1 ||
taosArrayGetSize(pTask->execNodes) <= 0) {
return TSDB_CODE_SUCCESS; return TSDB_CODE_SUCCESS;
} }
@ -1264,7 +1264,6 @@ int32_t schUpdateTaskExecNodeHandle(SSchTask *pTask, void *handle, int32_t rspCo
return TSDB_CODE_SUCCESS; return TSDB_CODE_SUCCESS;
} }
int32_t schHandleCallback(void *param, const SDataBuf *pMsg, int32_t msgType, int32_t rspCode) { int32_t schHandleCallback(void *param, const SDataBuf *pMsg, int32_t msgType, int32_t rspCode) {
int32_t code = 0; int32_t code = 0;
SSchTaskCallbackParam *pParam = (SSchTaskCallbackParam *)param; SSchTaskCallbackParam *pParam = (SSchTaskCallbackParam *)param;
@ -1282,13 +1281,15 @@ int32_t schHandleCallback(void *param, const SDataBuf *pMsg, int32_t msgType, in
if (TDMT_VND_EXPLAIN_RSP == msgType) { if (TDMT_VND_EXPLAIN_RSP == msgType) {
schGetTaskFromTaskList(pJob->succTasks, pParam->taskId, &pTask); schGetTaskFromTaskList(pJob->succTasks, pParam->taskId, &pTask);
} else { } else {
SCH_JOB_ELOG("task not found in execTask list, refId:%" PRIx64 ", taskId:%" PRIx64, pParam->refId, pParam->taskId); SCH_JOB_ELOG("task not found in execTask list, refId:%" PRIx64 ", taskId:%" PRIx64, pParam->refId,
pParam->taskId);
SCH_ERR_JRET(TSDB_CODE_SCH_INTERNAL_ERROR); SCH_ERR_JRET(TSDB_CODE_SCH_INTERNAL_ERROR);
} }
} }
if (NULL == pTask) { if (NULL == pTask) {
SCH_JOB_ELOG("task not found in execList & succList, refId:%" PRIx64 ", taskId:%" PRIx64, pParam->refId, pParam->taskId); SCH_JOB_ELOG("task not found in execList & succList, refId:%" PRIx64 ", taskId:%" PRIx64, pParam->refId,
pParam->taskId);
SCH_ERR_JRET(TSDB_CODE_SCH_INTERNAL_ERROR); SCH_ERR_JRET(TSDB_CODE_SCH_INTERNAL_ERROR);
} }
@ -1444,7 +1445,7 @@ int32_t schGetCallbackFp(int32_t msgType, __async_send_cb_fn_t *fp) {
} }
int32_t schGenerateTaskCallBackAHandle(SSchJob *pJob, SSchTask *pTask, int32_t msgType, SMsgSendInfo **pMsgSendInfo) { int32_t schGenerateTaskCallBackAHandle(SSchJob *pJob, SSchTask *pTask, int32_t msgType, SMsgSendInfo **pMsgSendInfo) {
int32_t code = 0; int32_t code = 0;
SMsgSendInfo *msgSendInfo = taosMemoryCalloc(1, sizeof(SMsgSendInfo)); SMsgSendInfo *msgSendInfo = taosMemoryCalloc(1, sizeof(SMsgSendInfo));
if (NULL == msgSendInfo) { if (NULL == msgSendInfo) {
SCH_TASK_ELOG("calloc %d failed", (int32_t)sizeof(SMsgSendInfo)); SCH_TASK_ELOG("calloc %d failed", (int32_t)sizeof(SMsgSendInfo));
@ -1565,7 +1566,7 @@ _return:
} }
int32_t schMakeQueryRpcCtx(SSchJob *pJob, SSchTask *pTask, SRpcCtx *pCtx) { int32_t schMakeQueryRpcCtx(SSchJob *pJob, SSchTask *pTask, SRpcCtx *pCtx) {
int32_t code = 0; int32_t code = 0;
SMsgSendInfo *pReadyMsgSendInfo = NULL; SMsgSendInfo *pReadyMsgSendInfo = NULL;
SMsgSendInfo *pExplainMsgSendInfo = NULL; SMsgSendInfo *pExplainMsgSendInfo = NULL;
@ -1578,7 +1579,7 @@ int32_t schMakeQueryRpcCtx(SSchJob *pJob, SSchTask *pTask, SRpcCtx *pCtx) {
SCH_ERR_JRET(schGenerateTaskCallBackAHandle(pJob, pTask, TDMT_VND_RES_READY, &pReadyMsgSendInfo)); SCH_ERR_JRET(schGenerateTaskCallBackAHandle(pJob, pTask, TDMT_VND_RES_READY, &pReadyMsgSendInfo));
SCH_ERR_JRET(schGenerateTaskCallBackAHandle(pJob, pTask, TDMT_VND_EXPLAIN, &pExplainMsgSendInfo)); SCH_ERR_JRET(schGenerateTaskCallBackAHandle(pJob, pTask, TDMT_VND_EXPLAIN, &pExplainMsgSendInfo));
int32_t msgType = TDMT_VND_RES_READY_RSP; int32_t msgType = TDMT_VND_RES_READY_RSP;
SRpcCtxVal ctxVal = {.val = pReadyMsgSendInfo, .clone = schCloneSMsgSendInfo, .freeFunc = schFreeRpcCtxVal}; SRpcCtxVal ctxVal = {.val = pReadyMsgSendInfo, .clone = schCloneSMsgSendInfo, .freeFunc = schFreeRpcCtxVal};
if (taosHashPut(pCtx->args, &msgType, sizeof(msgType), &ctxVal, sizeof(ctxVal))) { if (taosHashPut(pCtx->args, &msgType, sizeof(msgType), &ctxVal, sizeof(ctxVal))) {
SCH_TASK_ELOG("taosHashPut msg %d to rpcCtx failed", msgType); SCH_TASK_ELOG("taosHashPut msg %d to rpcCtx failed", msgType);
@ -1599,7 +1600,7 @@ int32_t schMakeQueryRpcCtx(SSchJob *pJob, SSchTask *pTask, SRpcCtx *pCtx) {
_return: _return:
taosHashCleanup(pCtx->args); taosHashCleanup(pCtx->args);
if (pReadyMsgSendInfo) { if (pReadyMsgSendInfo) {
taosMemoryFreeClear(pReadyMsgSendInfo->param); taosMemoryFreeClear(pReadyMsgSendInfo->param);
taosMemoryFreeClear(pReadyMsgSendInfo); taosMemoryFreeClear(pReadyMsgSendInfo);
@ -1818,7 +1819,7 @@ _return:
taosMemoryFreeClear(pMsgSendInfo->param); taosMemoryFreeClear(pMsgSendInfo->param);
taosMemoryFreeClear(pMsgSendInfo); taosMemoryFreeClear(pMsgSendInfo);
} }
SCH_RET(code); SCH_RET(code);
} }
@ -2319,7 +2320,7 @@ _return:
} }
int32_t schExecStaticExplain(void *transport, SArray *pNodeList, SQueryPlan *pDag, int64_t *job, const char *sql, int32_t schExecStaticExplain(void *transport, SArray *pNodeList, SQueryPlan *pDag, int64_t *job, const char *sql,
bool syncSchedule) { bool syncSchedule) {
qDebug("QID:0x%" PRIx64 " job started", pDag->queryId); qDebug("QID:0x%" PRIx64 " job started", pDag->queryId);
int32_t code = 0; int32_t code = 0;
@ -2608,7 +2609,7 @@ int32_t schedulerFetchRows(int64_t job, void **pData) {
if (!(pJob->attr.explainMode == EXPLAIN_MODE_STATIC)) { if (!(pJob->attr.explainMode == EXPLAIN_MODE_STATIC)) {
SCH_ERR_JRET(schFetchFromRemote(pJob)); SCH_ERR_JRET(schFetchFromRemote(pJob));
tsem_wait(&pJob->rspSem); tsem_wait(&pJob->rspSem);
} }
} else { } else {
SCH_JOB_ELOG("job status error for fetch, status:%s", jobTaskStatusStr(status)); SCH_JOB_ELOG("job status error for fetch, status:%s", jobTaskStatusStr(status));
SCH_ERR_JRET(TSDB_CODE_SCH_STATUS_ERROR); SCH_ERR_JRET(TSDB_CODE_SCH_STATUS_ERROR);
@ -2670,11 +2671,11 @@ int32_t schedulerGetTasksStatus(int64_t job, SArray *pSub) {
for (int32_t i = pJob->levelNum - 1; i >= 0; --i) { for (int32_t i = pJob->levelNum - 1; i >= 0; --i) {
SSchLevel *pLevel = taosArrayGet(pJob->levels, i); SSchLevel *pLevel = taosArrayGet(pJob->levels, i);
for (int32_t m = 0; m < pLevel->taskNum; ++m) { for (int32_t m = 0; m < pLevel->taskNum; ++m) {
SSchTask *pTask = taosArrayGet(pLevel->subTasks, m); SSchTask *pTask = taosArrayGet(pLevel->subTasks, m);
SQuerySubDesc subDesc = {.tid = pTask->taskId, .status = pTask->status}; SQuerySubDesc subDesc = {.tid = pTask->taskId, .status = pTask->status};
taosArrayPush(pSub, &subDesc); taosArrayPush(pSub, &subDesc);
} }
} }
@ -2682,7 +2683,6 @@ int32_t schedulerGetTasksStatus(int64_t job, SArray *pSub) {
return TSDB_CODE_SUCCESS; return TSDB_CODE_SUCCESS;
} }
int32_t scheduleCancelJob(int64_t job) { int32_t scheduleCancelJob(int64_t job) {
SSchJob *pJob = schAcquireJob(job); SSchJob *pJob = schAcquireJob(job);
if (NULL == pJob) { if (NULL == pJob) {
@ -2738,11 +2738,10 @@ void schedulerDestroy(void) {
while (pJob) { while (pJob) {
taosRemoveRef(schMgmt.jobRef, pJob->refId); taosRemoveRef(schMgmt.jobRef, pJob->refId);
pJob = taosIterateRef(schMgmt.jobRef, pJob->refId); pJob = taosIterateRef(schMgmt.jobRef, pJob);
} }
taosCloseRef(schMgmt.jobRef); taosCloseRef(schMgmt.jobRef);
schMgmt.jobRef = 0; schMgmt.jobRef = 0;
} }
} }