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

fix serialize

This commit is contained in:
Liu Jicong 2022-03-31 16:45:16 +08:00
parent ec97115b83
commit d02850382c
2 changed files with 149 additions and 139 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

2
source/dnode/vnode/src/vnd/vnodeWrite.c
View File

@ -165,7 +165,7 @@ int vnodeApplyWMsg(SVnode *pVnode, SRpcMsg *pMsg, SRpcMsg **pRsp) {
// } // }
break; break;
case TDMT_VND_SUBMIT: case TDMT_VND_SUBMIT:
printf("vnode %d write data %ld\n", pVnode->vgId, ver); /*printf("vnode %d write data %ld\n", pVnode->vgId, ver);*/
if (pVnode->config.streamMode == 0) { if (pVnode->config.streamMode == 0) {
if (tsdbInsertData(pVnode->pTsdb, (SSubmitReq *)ptr, NULL) < 0) { if (tsdbInsertData(pVnode->pTsdb, (SSubmitReq *)ptr, NULL) < 0) {
// TODO: handle error // TODO: handle error

166
source/libs/scheduler/src/scheduler.c
View File

@ -86,7 +86,7 @@ void schFreeRpcCtx(SRpcCtx *pCtx) {
} }
} }
void schFreeTask(SSchTask* pTask) { void schFreeTask(SSchTask *pTask) {
if (pTask->candidateAddrs) { if (pTask->candidateAddrs) {
taosArrayDestroy(pTask->candidateAddrs); taosArrayDestroy(pTask->candidateAddrs);
} }
@ -125,12 +125,14 @@ int32_t schValidateTaskReceivedMsgType(SSchJob *pJob, SSchTask *pTask, int32_t m
case TDMT_SCH_LINK_BROKEN: case TDMT_SCH_LINK_BROKEN:
return TSDB_CODE_SUCCESS; return TSDB_CODE_SUCCESS;
case TDMT_VND_QUERY_RSP: // query_rsp may be processed later than ready_rsp case TDMT_VND_QUERY_RSP: // query_rsp may be processed later than ready_rsp
if (lastMsgType != reqMsgType && -1 != lastMsgType && TDMT_VND_FETCH != lastMsgType) { if (lastMsgType != reqMsgType && -1 != lastMsgType && TDMT_VND_FETCH != lastMsgType) {
SCH_TASK_DLOG("rsp msg type mis-match, last sent msgType:%s, rspType:%s", TMSG_INFO(lastMsgType), TMSG_INFO(msgType)); SCH_TASK_DLOG("rsp msg type mis-match, last sent msgType:%s, rspType:%s", TMSG_INFO(lastMsgType),
TMSG_INFO(msgType));
} }
if (taskStatus != JOB_TASK_STATUS_EXECUTING && taskStatus != JOB_TASK_STATUS_PARTIAL_SUCCEED) { if (taskStatus != JOB_TASK_STATUS_EXECUTING && taskStatus != JOB_TASK_STATUS_PARTIAL_SUCCEED) {
SCH_TASK_DLOG("rsp msg conflicted with task status, status:%s, rspType:%s", jobTaskStatusStr(taskStatus), TMSG_INFO(msgType)); SCH_TASK_DLOG("rsp msg conflicted with task status, status:%s, rspType:%s", jobTaskStatusStr(taskStatus),
TMSG_INFO(msgType));
} }
SCH_SET_TASK_LASTMSG_TYPE(pTask, -1); SCH_SET_TASK_LASTMSG_TYPE(pTask, -1);
@ -138,25 +140,29 @@ int32_t schValidateTaskReceivedMsgType(SSchJob *pJob, SSchTask *pTask, int32_t m
case TDMT_VND_RES_READY_RSP: case TDMT_VND_RES_READY_RSP:
reqMsgType = TDMT_VND_QUERY; reqMsgType = TDMT_VND_QUERY;
if (lastMsgType != reqMsgType && -1 != lastMsgType) { if (lastMsgType != reqMsgType && -1 != lastMsgType) {
SCH_TASK_ELOG("rsp msg type mis-match, last sent msgType:%s, rspType:%s", (lastMsgType > 0 ? TMSG_INFO(lastMsgType) : "null"), TMSG_INFO(msgType)); SCH_TASK_ELOG("rsp msg type mis-match, last sent msgType:%s, rspType:%s",
(lastMsgType > 0 ? TMSG_INFO(lastMsgType) : "null"), TMSG_INFO(msgType));
SCH_ERR_RET(TSDB_CODE_SCH_STATUS_ERROR); SCH_ERR_RET(TSDB_CODE_SCH_STATUS_ERROR);
} }
if (taskStatus != JOB_TASK_STATUS_EXECUTING && taskStatus != JOB_TASK_STATUS_PARTIAL_SUCCEED) { if (taskStatus != JOB_TASK_STATUS_EXECUTING && taskStatus != JOB_TASK_STATUS_PARTIAL_SUCCEED) {
SCH_TASK_ELOG("rsp msg conflicted with task status, status:%s, rspType:%s", jobTaskStatusStr(taskStatus), TMSG_INFO(msgType)); SCH_TASK_ELOG("rsp msg conflicted with task status, status:%s, rspType:%s", jobTaskStatusStr(taskStatus),
TMSG_INFO(msgType));
SCH_ERR_RET(TSDB_CODE_SCH_STATUS_ERROR); SCH_ERR_RET(TSDB_CODE_SCH_STATUS_ERROR);
} }
SCH_SET_TASK_LASTMSG_TYPE(pTask, -1); SCH_SET_TASK_LASTMSG_TYPE(pTask, -1);
return TSDB_CODE_SUCCESS; return TSDB_CODE_SUCCESS;
case TDMT_VND_FETCH_RSP: case TDMT_VND_FETCH_RSP:
if (lastMsgType != reqMsgType && -1 != lastMsgType) { if (lastMsgType != reqMsgType && -1 != lastMsgType) {
SCH_TASK_ELOG("rsp msg type mis-match, last sent msgType:%s, rspType:%s", TMSG_INFO(lastMsgType), TMSG_INFO(msgType)); SCH_TASK_ELOG("rsp msg type mis-match, last sent msgType:%s, rspType:%s", TMSG_INFO(lastMsgType),
TMSG_INFO(msgType));
SCH_ERR_RET(TSDB_CODE_SCH_STATUS_ERROR); SCH_ERR_RET(TSDB_CODE_SCH_STATUS_ERROR);
} }
if (taskStatus != JOB_TASK_STATUS_EXECUTING && taskStatus != JOB_TASK_STATUS_PARTIAL_SUCCEED) { if (taskStatus != JOB_TASK_STATUS_EXECUTING && taskStatus != JOB_TASK_STATUS_PARTIAL_SUCCEED) {
SCH_TASK_ELOG("rsp msg conflicted with task status, status:%s, rspType:%s", jobTaskStatusStr(taskStatus), TMSG_INFO(msgType)); SCH_TASK_ELOG("rsp msg conflicted with task status, status:%s, rspType:%s", jobTaskStatusStr(taskStatus),
TMSG_INFO(msgType));
SCH_ERR_RET(TSDB_CODE_SCH_STATUS_ERROR); SCH_ERR_RET(TSDB_CODE_SCH_STATUS_ERROR);
} }
@ -171,12 +177,14 @@ int32_t schValidateTaskReceivedMsgType(SSchJob *pJob, SSchTask *pTask, int32_t m
} }
if (lastMsgType != reqMsgType) { if (lastMsgType != reqMsgType) {
SCH_TASK_ELOG("rsp msg type mis-match, last sent msgType:%s, rspType:%s", TMSG_INFO(lastMsgType), TMSG_INFO(msgType)); SCH_TASK_ELOG("rsp msg type mis-match, last sent msgType:%s, rspType:%s", TMSG_INFO(lastMsgType),
TMSG_INFO(msgType));
SCH_ERR_RET(TSDB_CODE_SCH_STATUS_ERROR); SCH_ERR_RET(TSDB_CODE_SCH_STATUS_ERROR);
} }
if (taskStatus != JOB_TASK_STATUS_EXECUTING && taskStatus != JOB_TASK_STATUS_PARTIAL_SUCCEED) { if (taskStatus != JOB_TASK_STATUS_EXECUTING && taskStatus != JOB_TASK_STATUS_PARTIAL_SUCCEED) {
SCH_TASK_ELOG("rsp msg conflicted with task status, status:%s, rspType:%s", jobTaskStatusStr(taskStatus), TMSG_INFO(msgType)); SCH_TASK_ELOG("rsp msg conflicted with task status, status:%s, rspType:%s", jobTaskStatusStr(taskStatus),
TMSG_INFO(msgType));
SCH_ERR_RET(TSDB_CODE_SCH_STATUS_ERROR); SCH_ERR_RET(TSDB_CODE_SCH_STATUS_ERROR);
} }
@ -654,7 +662,8 @@ int32_t schTaskCheckSetRetry(SSchJob *pJob, SSchTask *pTask, int32_t errCode, bo
if (SCH_IS_DATA_SRC_TASK(pTask)) { if (SCH_IS_DATA_SRC_TASK(pTask)) {
if (pTask->tryTimes >= SCH_TASK_NUM_OF_EPS(&pTask->plan->execNode)) { if (pTask->tryTimes >= SCH_TASK_NUM_OF_EPS(&pTask->plan->execNode)) {
*needRetry = false; *needRetry = false;
SCH_TASK_DLOG("task no more retry since all ep tried, tryTimes:%d, epNum:%d", pTask->tryTimes, SCH_TASK_NUM_OF_EPS(&pTask->plan->execNode)); SCH_TASK_DLOG("task no more retry since all ep tried, tryTimes:%d, epNum:%d", pTask->tryTimes,
SCH_TASK_NUM_OF_EPS(&pTask->plan->execNode));
return TSDB_CODE_SUCCESS; return TSDB_CODE_SUCCESS;
} }
} else { } else {
@ -662,7 +671,8 @@ int32_t schTaskCheckSetRetry(SSchJob *pJob, SSchTask *pTask, int32_t errCode, bo
if ((pTask->candidateIdx + 1) >= candidateNum) { if ((pTask->candidateIdx + 1) >= candidateNum) {
*needRetry = false; *needRetry = false;
SCH_TASK_DLOG("task no more retry since all candiates tried, candidateIdx:%d, candidateNum:%d", pTask->candidateIdx, candidateNum); SCH_TASK_DLOG("task no more retry since all candiates tried, candidateIdx:%d, candidateNum:%d",
pTask->candidateIdx, candidateNum);
return TSDB_CODE_SUCCESS; return TSDB_CODE_SUCCESS;
} }
} }
@ -706,9 +716,8 @@ int32_t schUpdateHbConnection(SQueryNodeEpId *epId, SSchTrans *trans) {
memcpy(&hb->trans, trans, sizeof(*trans)); memcpy(&hb->trans, trans, sizeof(*trans));
SCH_UNLOCK(SCH_WRITE, &hb->lock); SCH_UNLOCK(SCH_WRITE, &hb->lock);
qDebug("hb connection updated, sId:%" PRIx64 ", nodeId:%d, fqdn:%s, port:%d, instance:%p, handle:%p", qDebug("hb connection updated, sId:%" PRIx64 ", nodeId:%d, fqdn:%s, port:%d, instance:%p, handle:%p", schMgmt.sId,
schMgmt.sId, epId->nodeId, epId->ep.fqdn, epId->ep.port, trans->transInst, epId->nodeId, epId->ep.fqdn, epId->ep.port, trans->transInst, trans->transHandle);
trans->transHandle);
return TSDB_CODE_SUCCESS; return TSDB_CODE_SUCCESS;
} }
@ -965,7 +974,8 @@ int32_t schHandleResponseMsg(SSchJob *pJob, SSchTask *pTask, int32_t msgType, ch
int8_t status = 0; int8_t status = 0;
if (schJobNeedToStop(pJob, &status)) { if (schJobNeedToStop(pJob, &status)) {
SCH_TASK_ELOG("rsp not processed cause of job status, job status:%s, rspCode:0x%x", jobTaskStatusStr(status), rspCode); SCH_TASK_ELOG("rsp not processed cause of job status, job status:%s, rspCode:0x%x", jobTaskStatusStr(status),
rspCode);
SCH_RET(atomic_load_32(&pJob->errCode)); SCH_RET(atomic_load_32(&pJob->errCode));
} }
@ -1011,21 +1021,21 @@ int32_t schHandleResponseMsg(SSchJob *pJob, SSchTask *pTask, int32_t msgType, ch
break; break;
} }
case TDMT_VND_QUERY_RSP: { case TDMT_VND_QUERY_RSP: {
SQueryTableRsp rsp = {0}; SQueryTableRsp rsp = {0};
if (msg) { if (msg) {
tDeserializeSQueryTableRsp(msg, msgSize, &rsp); tDeserializeSQueryTableRsp(msg, msgSize, &rsp);
SCH_ERR_JRET(rsp.code); SCH_ERR_JRET(rsp.code);
} }
SCH_ERR_JRET(rspCode); SCH_ERR_JRET(rspCode);
if (NULL == msg) { if (NULL == msg) {
SCH_ERR_JRET(TSDB_CODE_QRY_INVALID_INPUT); SCH_ERR_JRET(TSDB_CODE_QRY_INVALID_INPUT);
} }
//SCH_ERR_JRET(schBuildAndSendMsg(pJob, pTask, NULL, TDMT_VND_RES_READY)); // SCH_ERR_JRET(schBuildAndSendMsg(pJob, pTask, NULL, TDMT_VND_RES_READY));
break; break;
} }
case TDMT_VND_RES_READY_RSP: { case TDMT_VND_RES_READY_RSP: {
SResReadyRsp *rsp = (SResReadyRsp *)msg; SResReadyRsp *rsp = (SResReadyRsp *)msg;
@ -1088,14 +1098,14 @@ _return:
} }
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;
SSchTask *pTask = NULL; SSchTask *pTask = NULL;
SSchJob *pJob = schAcquireJob(pParam->refId); SSchJob *pJob = schAcquireJob(pParam->refId);
if (NULL == pJob) { if (NULL == pJob) {
qWarn("QID:0x%" PRIx64 ",TID:0x%" PRIx64 "taosAcquireRef job failed, may be dropped, refId:%" PRIx64, qWarn("QID:0x%" PRIx64 ",TID:0x%" PRIx64 "taosAcquireRef job failed, may be dropped, refId:%" PRIx64,
pParam->queryId, pParam->taskId, pParam->refId); pParam->queryId, pParam->taskId, pParam->refId);
SCH_ERR_JRET(TSDB_CODE_QRY_JOB_FREED); SCH_ERR_JRET(TSDB_CODE_QRY_JOB_FREED);
} }
@ -1173,7 +1183,8 @@ int32_t schHandleHbCallback(void *param, const SDataBuf *pMsg, int32_t code) {
SCH_ERR_RET(schUpdateHbConnection(&rsp.epId, &trans)); SCH_ERR_RET(schUpdateHbConnection(&rsp.epId, &trans));
int32_t taskNum = (int32_t)taosArrayGetSize(rsp.taskStatus); int32_t taskNum = (int32_t)taosArrayGetSize(rsp.taskStatus);
qDebug("%d task status in hb rsp, nodeId:%d, fqdn:%s, port:%d", taskNum, rsp.epId.nodeId, rsp.epId.ep.fqdn, rsp.epId.ep.port); qDebug("%d task status in hb rsp, nodeId:%d, fqdn:%s, port:%d", taskNum, rsp.epId.nodeId, rsp.epId.ep.fqdn,
rsp.epId.ep.port);
for (int32_t i = 0; i < taskNum; ++i) { for (int32_t i = 0; i < taskNum; ++i) {
STaskStatus *taskStatus = taosArrayGet(rsp.taskStatus, i); STaskStatus *taskStatus = taosArrayGet(rsp.taskStatus, i);
@ -1188,7 +1199,8 @@ int32_t schHandleHbCallback(void *param, const SDataBuf *pMsg, int32_t code) {
// TODO // TODO
SCH_JOB_DLOG("TID:0x%" PRIx64 " task status in server: %s", taskStatus->taskId, jobTaskStatusStr(taskStatus->status)); SCH_JOB_DLOG("TID:0x%" PRIx64 " task status in server: %s", taskStatus->taskId,
jobTaskStatusStr(taskStatus->status));
schReleaseJob(taskStatus->refId); schReleaseJob(taskStatus->refId);
} }
@ -1208,7 +1220,7 @@ int32_t schHandleLinkBrokenCallback(void *param, const SDataBuf *pMsg, int32_t c
if (head->isHbParam) { if (head->isHbParam) {
SSchHbCallbackParam *hbParam = (SSchHbCallbackParam *)param; SSchHbCallbackParam *hbParam = (SSchHbCallbackParam *)param;
SSchTrans trans = {.transInst = hbParam->transport, .transHandle = NULL}; SSchTrans trans = {.transInst = hbParam->transport, .transHandle = NULL};
SCH_ERR_RET(schUpdateHbConnection(&hbParam->nodeEpId, &trans)); SCH_ERR_RET(schUpdateHbConnection(&hbParam->nodeEpId, &trans));
SCH_ERR_RET(schBuildAndSendHbMsg(&hbParam->nodeEpId)); SCH_ERR_RET(schBuildAndSendHbMsg(&hbParam->nodeEpId));
@ -1219,7 +1231,6 @@ int32_t schHandleLinkBrokenCallback(void *param, const SDataBuf *pMsg, int32_t c
return TSDB_CODE_SUCCESS; return TSDB_CODE_SUCCESS;
} }
int32_t schGetCallbackFp(int32_t msgType, __async_send_cb_fn_t *fp) { int32_t schGetCallbackFp(int32_t msgType, __async_send_cb_fn_t *fp) {
switch (msgType) { switch (msgType) {
case TDMT_VND_CREATE_TABLE: case TDMT_VND_CREATE_TABLE:
@ -1259,7 +1270,7 @@ void schFreeRpcCtxVal(const void *arg) {
return; return;
} }
SMsgSendInfo* pMsgSendInfo = (SMsgSendInfo *)arg; SMsgSendInfo *pMsgSendInfo = (SMsgSendInfo *)arg;
taosMemoryFreeClear(pMsgSendInfo->param); taosMemoryFreeClear(pMsgSendInfo->param);
taosMemoryFreeClear(pMsgSendInfo); taosMemoryFreeClear(pMsgSendInfo);
} }
@ -1301,10 +1312,9 @@ int32_t schMakeHbCallbackParam(SSchJob *pJob, SSchTask *pTask, void **pParam) {
return TSDB_CODE_SUCCESS; return TSDB_CODE_SUCCESS;
} }
int32_t schMakeBrokenLinkVal(SSchJob *pJob, SSchTask *pTask, SRpcBrokenlinkVal *brokenVal, bool isHb) { int32_t schMakeBrokenLinkVal(SSchJob *pJob, SSchTask *pTask, SRpcBrokenlinkVal *brokenVal, bool isHb) {
int32_t code = 0; int32_t code = 0;
SMsgSendInfo* pMsgSendInfo = NULL; SMsgSendInfo *pMsgSendInfo = NULL;
pMsgSendInfo = taosMemoryCalloc(1, sizeof(SMsgSendInfo)); pMsgSendInfo = taosMemoryCalloc(1, sizeof(SMsgSendInfo));
if (NULL == pMsgSendInfo) { if (NULL == pMsgSendInfo) {
@ -1318,7 +1328,7 @@ int32_t schMakeBrokenLinkVal(SSchJob *pJob, SSchTask *pTask, SRpcBrokenlinkVal *
SCH_ERR_JRET(schMakeTaskCallbackParam(pJob, pTask, &pMsgSendInfo->param)); SCH_ERR_JRET(schMakeTaskCallbackParam(pJob, pTask, &pMsgSendInfo->param));
} }
int32_t msgType = TDMT_SCH_LINK_BROKEN; int32_t msgType = TDMT_SCH_LINK_BROKEN;
__async_send_cb_fn_t fp = NULL; __async_send_cb_fn_t fp = NULL;
SCH_ERR_JRET(schGetCallbackFp(msgType, &fp)); SCH_ERR_JRET(schGetCallbackFp(msgType, &fp));
@ -1340,9 +1350,9 @@ _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;
SSchTaskCallbackParam *param = NULL; SSchTaskCallbackParam *param = NULL;
SMsgSendInfo* pMsgSendInfo = NULL; SMsgSendInfo *pMsgSendInfo = NULL;
pCtx->args = taosHashInit(1, taosGetDefaultHashFunction(TSDB_DATA_TYPE_INT), false, HASH_ENTRY_LOCK); pCtx->args = taosHashInit(1, taosGetDefaultHashFunction(TSDB_DATA_TYPE_INT), false, HASH_ENTRY_LOCK);
if (NULL == pCtx->args) { if (NULL == pCtx->args) {
@ -1362,7 +1372,7 @@ int32_t schMakeQueryRpcCtx(SSchJob *pJob, SSchTask *pTask, SRpcCtx *pCtx) {
SCH_ERR_JRET(TSDB_CODE_QRY_OUT_OF_MEMORY); SCH_ERR_JRET(TSDB_CODE_QRY_OUT_OF_MEMORY);
} }
int32_t msgType = TDMT_VND_RES_READY_RSP; int32_t msgType = TDMT_VND_RES_READY_RSP;
__async_send_cb_fn_t fp = NULL; __async_send_cb_fn_t fp = NULL;
SCH_ERR_JRET(schGetCallbackFp(TDMT_VND_RES_READY, &fp)); SCH_ERR_JRET(schGetCallbackFp(TDMT_VND_RES_READY, &fp));
@ -1394,11 +1404,11 @@ _return:
} }
int32_t schMakeHbRpcCtx(SSchJob *pJob, SSchTask *pTask, SRpcCtx *pCtx) { int32_t schMakeHbRpcCtx(SSchJob *pJob, SSchTask *pTask, SRpcCtx *pCtx) {
int32_t code = 0; int32_t code = 0;
SSchHbCallbackParam *param = NULL; SSchHbCallbackParam *param = NULL;
SMsgSendInfo* pMsgSendInfo = NULL; SMsgSendInfo *pMsgSendInfo = NULL;
SQueryNodeAddr *addr = taosArrayGet(pTask->candidateAddrs, pTask->candidateIdx); SQueryNodeAddr *addr = taosArrayGet(pTask->candidateAddrs, pTask->candidateIdx);
SQueryNodeEpId epId = {0}; SQueryNodeEpId epId = {0};
epId.nodeId = addr->nodeId; epId.nodeId = addr->nodeId;
memcpy(&epId.ep, SCH_GET_CUR_EP(addr), sizeof(SEp)); memcpy(&epId.ep, SCH_GET_CUR_EP(addr), sizeof(SEp));
@ -1421,7 +1431,7 @@ int32_t schMakeHbRpcCtx(SSchJob *pJob, SSchTask *pTask, SRpcCtx *pCtx) {
SCH_ERR_JRET(TSDB_CODE_QRY_OUT_OF_MEMORY); SCH_ERR_JRET(TSDB_CODE_QRY_OUT_OF_MEMORY);
} }
int32_t msgType = TDMT_VND_QUERY_HEARTBEAT_RSP; int32_t msgType = TDMT_VND_QUERY_HEARTBEAT_RSP;
__async_send_cb_fn_t fp = NULL; __async_send_cb_fn_t fp = NULL;
SCH_ERR_JRET(schGetCallbackFp(TDMT_VND_QUERY_HEARTBEAT, &fp)); SCH_ERR_JRET(schGetCallbackFp(TDMT_VND_QUERY_HEARTBEAT, &fp));
@ -1450,9 +1460,8 @@ _return:
SCH_RET(code); SCH_RET(code);
} }
int32_t schRegisterHbConnection(SSchJob *pJob, SSchTask *pTask, SQueryNodeEpId *epId, bool *exist) { int32_t schRegisterHbConnection(SSchJob *pJob, SSchTask *pTask, SQueryNodeEpId *epId, bool *exist) {
int32_t code = 0; int32_t code = 0;
SSchHbTrans hb = {0}; SSchHbTrans hb = {0};
hb.trans.transInst = pJob->transport; hb.trans.transInst = pJob->transport;
@ -1475,8 +1484,6 @@ int32_t schRegisterHbConnection(SSchJob *pJob, SSchTask *pTask, SQueryNodeEpId *
return TSDB_CODE_SUCCESS; return TSDB_CODE_SUCCESS;
} }
int32_t schCloneCallbackParam(SSchCallbackParamHeader *pSrc, SSchCallbackParamHeader **pDst) { int32_t schCloneCallbackParam(SSchCallbackParamHeader *pSrc, SSchCallbackParamHeader **pDst) {
if (pSrc->isHbParam) { if (pSrc->isHbParam) {
SSchHbCallbackParam *dst = taosMemoryMalloc(sizeof(SSchHbCallbackParam)); SSchHbCallbackParam *dst = taosMemoryMalloc(sizeof(SSchHbCallbackParam));
@ -1505,7 +1512,7 @@ int32_t schCloneCallbackParam(SSchCallbackParamHeader *pSrc, SSchCallbackParamHe
int32_t schCloneSMsgSendInfo(void *src, void **dst) { int32_t schCloneSMsgSendInfo(void *src, void **dst) {
SMsgSendInfo *pSrc = src; SMsgSendInfo *pSrc = src;
int32_t code = 0; int32_t code = 0;
SMsgSendInfo *pDst = taosMemoryMalloc(sizeof(*pSrc)); SMsgSendInfo *pDst = taosMemoryMalloc(sizeof(*pSrc));
if (NULL == pDst) { if (NULL == pDst) {
qError("malloc SMsgSendInfo for rpcCtx failed, len:%d", (int32_t)sizeof(*pSrc)); qError("malloc SMsgSendInfo for rpcCtx failed, len:%d", (int32_t)sizeof(*pSrc));
@ -1541,10 +1548,10 @@ int32_t schCloneHbRpcCtx(SRpcCtx *pSrc, SRpcCtx *pDst) {
} }
SRpcCtxVal dst = {0}; SRpcCtxVal dst = {0};
void *pIter = taosHashIterate(pSrc->args, NULL); void *pIter = taosHashIterate(pSrc->args, NULL);
while (pIter) { while (pIter) {
SRpcCtxVal *pVal = (SRpcCtxVal *)pIter; SRpcCtxVal *pVal = (SRpcCtxVal *)pIter;
int32_t *msgType = taosHashGetKey(pIter, NULL); int32_t *msgType = taosHashGetKey(pIter, NULL);
dst = *pVal; dst = *pVal;
dst.val = NULL; dst.val = NULL;
@ -1568,8 +1575,8 @@ _return:
SCH_RET(code); SCH_RET(code);
} }
int32_t schAsyncSendMsg(SSchJob *pJob, SSchTask *pTask, void *transport, SEpSet *epSet, int32_t msgType, void *msg,
int32_t schAsyncSendMsg(SSchJob *pJob, SSchTask *pTask, void *transport, SEpSet* epSet, int32_t msgType, void *msg, uint32_t msgSize, bool persistHandle, SRpcCtx *ctx) { uint32_t msgSize, bool persistHandle, SRpcCtx *ctx) {
int32_t code = 0; int32_t code = 0;
SSchTrans *trans = (SSchTrans *)transport; SSchTrans *trans = (SSchTrans *)transport;
@ -1601,11 +1608,11 @@ int32_t schAsyncSendMsg(SSchJob *pJob, SSchTask *pTask, void *transport, SEpSet*
pMsgSendInfo->msgType = msgType; pMsgSendInfo->msgType = msgType;
pMsgSendInfo->fp = fp; pMsgSendInfo->fp = fp;
qDebug("start to send %s msg to node[%d,%s,%d], refId:%" PRIx64 "instance:%p, handle:%p", qDebug("start to send %s msg to node[%d,%s,%d], refId:%" PRIx64 "instance:%p, handle:%p", TMSG_INFO(msgType),
TMSG_INFO(msgType), ntohl(((SMsgHead *)msg)->vgId), epSet->eps[epSet->inUse].fqdn, epSet->eps[epSet->inUse].port, ntohl(((SMsgHead *)msg)->vgId), epSet->eps[epSet->inUse].fqdn, epSet->eps[epSet->inUse].port, pJob->refId,
pJob->refId, trans->transInst, trans->transHandle); trans->transInst, trans->transHandle);
int64_t transporterId = 0; int64_t transporterId = 0;
code = asyncSendMsgToServerExt(trans->transInst, epSet, &transporterId, pMsgSendInfo, persistHandle, ctx); code = asyncSendMsgToServerExt(trans->transInst, epSet, &transporterId, pMsgSendInfo, persistHandle, ctx);
if (code) { if (code) {
SCH_ERR_JRET(code); SCH_ERR_JRET(code);
@ -1623,18 +1630,19 @@ _return:
int32_t schBuildAndSendHbMsg(SQueryNodeEpId *nodeEpId) { int32_t schBuildAndSendHbMsg(SQueryNodeEpId *nodeEpId) {
SSchedulerHbReq req = {0}; SSchedulerHbReq req = {0};
int32_t code = 0; int32_t code = 0;
SRpcCtx rpcCtx = {0}; SRpcCtx rpcCtx = {0};
SSchTrans trans = {0}; SSchTrans trans = {0};
int32_t msgType = TDMT_VND_QUERY_HEARTBEAT; int32_t msgType = TDMT_VND_QUERY_HEARTBEAT;
req.header.vgId = htonl(nodeEpId->nodeId); req.header.vgId = nodeEpId->nodeId;
req.sId = schMgmt.sId; req.sId = schMgmt.sId;
memcpy(&req.epId, nodeEpId, sizeof(SQueryNodeEpId)); memcpy(&req.epId, nodeEpId, sizeof(SQueryNodeEpId));
SSchHbTrans *hb = taosHashGet(schMgmt.hbConnections, nodeEpId, sizeof(SQueryNodeEpId)); SSchHbTrans *hb = taosHashGet(schMgmt.hbConnections, nodeEpId, sizeof(SQueryNodeEpId));
if (NULL == hb) { if (NULL == hb) {
qError("taosHashGet hb connection failed, nodeId:%d, fqdn:%s, port:%d", nodeEpId->nodeId, nodeEpId->ep.fqdn, nodeEpId->ep.port); qError("taosHashGet hb connection failed, nodeId:%d, fqdn:%s, port:%d", nodeEpId->nodeId, nodeEpId->ep.fqdn,
nodeEpId->ep.port);
SCH_ERR_RET(code); SCH_ERR_RET(code);
} }
@ -1685,16 +1693,17 @@ int32_t schBuildAndSendHbMsg(SQueryNodeEpId *nodeEpId) {
pMsgSendInfo->msgType = msgType; pMsgSendInfo->msgType = msgType;
pMsgSendInfo->fp = fp; pMsgSendInfo->fp = fp;
int64_t transporterId = 0; int64_t transporterId = 0;
SEpSet epSet = {.inUse = 0, .numOfEps = 1}; SEpSet epSet = {.inUse = 0, .numOfEps = 1};
memcpy(&epSet.eps[0], &nodeEpId->ep, sizeof(nodeEpId->ep)); memcpy(&epSet.eps[0], &nodeEpId->ep, sizeof(nodeEpId->ep));
qDebug("start to send hb msg, instance:%p, handle:%p, fqdn:%s, port:%d", trans.transInst, trans.transHandle, nodeEpId->ep.fqdn, nodeEpId->ep.port); qDebug("start to send hb msg, instance:%p, handle:%p, fqdn:%s, port:%d", trans.transInst, trans.transHandle,
nodeEpId->ep.fqdn, nodeEpId->ep.port);
code = asyncSendMsgToServerExt(trans.transInst, &epSet, &transporterId, pMsgSendInfo, true, &rpcCtx); code = asyncSendMsgToServerExt(trans.transInst, &epSet, &transporterId, pMsgSendInfo, true, &rpcCtx);
if (code) { if (code) {
qError("fail to send hb msg, instance:%p, handle:%p, fqdn:%s, port:%d, error:%x - %s", qError("fail to send hb msg, instance:%p, handle:%p, fqdn:%s, port:%d, error:%x - %s", trans.transInst,
trans.transInst, trans.transHandle, nodeEpId->ep.fqdn, nodeEpId->ep.port, code, tstrerror(code)); trans.transHandle, nodeEpId->ep.fqdn, nodeEpId->ep.port, code, tstrerror(code));
SCH_ERR_JRET(code); SCH_ERR_JRET(code);
} }
@ -1712,11 +1721,11 @@ _return:
int32_t schBuildAndSendMsg(SSchJob *pJob, SSchTask *pTask, SQueryNodeAddr *addr, int32_t msgType) { int32_t schBuildAndSendMsg(SSchJob *pJob, SSchTask *pTask, SQueryNodeAddr *addr, int32_t msgType) {
uint32_t msgSize = 0; uint32_t msgSize = 0;
void *msg = NULL; void *msg = NULL;
int32_t code = 0; int32_t code = 0;
bool isCandidateAddr = false; bool isCandidateAddr = false;
bool persistHandle = false; bool persistHandle = false;
SRpcCtx rpcCtx = {0}; SRpcCtx rpcCtx = {0};
if (NULL == addr) { if (NULL == addr) {
addr = taosArrayGet(pTask->candidateAddrs, pTask->candidateIdx); addr = taosArrayGet(pTask->candidateAddrs, pTask->candidateIdx);
@ -1856,7 +1865,8 @@ int32_t schBuildAndSendMsg(SSchJob *pJob, SSchTask *pTask, SQueryNodeAddr *addr,
SCH_SET_TASK_LASTMSG_TYPE(pTask, msgType); SCH_SET_TASK_LASTMSG_TYPE(pTask, msgType);
SSchTrans trans = {.transInst = pJob->transport, .transHandle = SCH_GET_TASK_HANDLE(pTask)}; SSchTrans trans = {.transInst = pJob->transport, .transHandle = SCH_GET_TASK_HANDLE(pTask)};
SCH_ERR_JRET(schAsyncSendMsg(pJob, pTask, &trans, &epSet, msgType, msg, msgSize, persistHandle, (rpcCtx.args ? &rpcCtx : NULL))); SCH_ERR_JRET(schAsyncSendMsg(pJob, pTask, &trans, &epSet, msgType, msg, msgSize, persistHandle,
(rpcCtx.args ? &rpcCtx : NULL)));
if (msgType == TDMT_VND_QUERY) { if (msgType == TDMT_VND_QUERY) {
SCH_ERR_RET(schRecordTaskExecNode(pJob, pTask, addr, trans.transHandle)); SCH_ERR_RET(schRecordTaskExecNode(pJob, pTask, addr, trans.transHandle));