提交 d5974a8f 编写于 作者: H Haojun Liao

refactor(stream): refactor the pause/resume for fill history execution.

上级 7f3d1dc4
......@@ -45,7 +45,7 @@ enum {
TASK_STATUS__FAIL,
TASK_STATUS__STOP,
TASK_STATUS__SCAN_HISTORY, // stream task scan history data by using tsdbread in the stream scanner
TASK_STATUS__SCAN_HISTORY_WAL, // scan history data in wal
// TASK_STATUS__SCAN_HISTORY_WAL, // scan history data in wal
TASK_STATUS__HALT, // pause, but not be manipulated by user command
TASK_STATUS__PAUSE, // pause
};
......
......@@ -1067,7 +1067,6 @@ int32_t tqProcessTaskDeployReq(STQ* pTq, int64_t sversion, char* msg, int32_t ms
// 3. It's an fill history task, do nothing. wait for the main task to start it
SStreamTask* p = streamMetaAcquireTask(pStreamMeta, taskId);
if (p != NULL) { // reset the downstreamReady flag.
p->status.downstreamReady = 0;
streamTaskCheckDownstreamTasks(p);
}
......@@ -1076,12 +1075,10 @@ int32_t tqProcessTaskDeployReq(STQ* pTq, int64_t sversion, char* msg, int32_t ms
}
int32_t tqProcessTaskScanHistory(STQ* pTq, SRpcMsg* pMsg) {
int32_t code = TSDB_CODE_SUCCESS;
char* msg = pMsg->pCont;
SStreamScanHistoryReq* pReq = (SStreamScanHistoryReq*)pMsg->pCont;
SStreamMeta* pMeta = pTq->pStreamMeta;
SStreamScanHistoryReq* pReq = (SStreamScanHistoryReq*)msg;
int32_t code = TSDB_CODE_SUCCESS;
SStreamTask* pTask = streamMetaAcquireTask(pMeta, pReq->taskId);
if (pTask == NULL) {
tqError("vgId:%d failed to acquire stream task:0x%x during stream recover, task may have been destroyed",
......@@ -1089,12 +1086,14 @@ int32_t tqProcessTaskScanHistory(STQ* pTq, SRpcMsg* pMsg) {
return -1;
}
// do recovery step 1
// do recovery step1
const char* id = pTask->id.idStr;
const char* pStatus = streamGetTaskStatusStr(pTask->status.taskStatus);
tqDebug("s-task:%s start history data scan stage(step 1), status:%s", id, pStatus);
tqDebug("s-task:%s start scan-history stage(step 1), status:%s", id, pStatus);
int64_t st = taosGetTimestampMs();
if (pTask->tsInfo.step1Start == 0) {
pTask->tsInfo.step1Start = taosGetTimestampMs();
}
// we have to continue retrying to successfully execute the scan history task.
int8_t schedStatus = atomic_val_compare_exchange_8(&pTask->status.schedStatus, TASK_SCHED_STATUS__INACTIVE,
......@@ -1108,30 +1107,21 @@ int32_t tqProcessTaskScanHistory(STQ* pTq, SRpcMsg* pMsg) {
}
ASSERT(pTask->status.pauseAllowed == false);
if (pTask->info.fillHistory == 1) {
streamTaskEnablePause(pTask);
}
if (!streamTaskRecoverScanStep1Finished(pTask)) {
streamSourceScanHistoryData(pTask);
}
// disable the pause when handling the step2 scan of tsdb data.
// the whole next procedure cann't be stopped.
// todo fix it: the following procedure should be executed completed and then shutdown when trying to close vnode.
if (pTask->info.fillHistory == 1) {
streamTaskDisablePause(pTask);
}
if (streamTaskShouldStop(&pTask->status) || streamTaskShouldPause(&pTask->status)) {
tqDebug("s-task:%s is dropped or paused, abort recover in step1", id);
if (pTask->status.taskStatus == TASK_STATUS__PAUSE) {
double el = taosGetTimestampMs() - pTask->tsInfo.step1Start;
tqDebug("s-task:%s is paused in the step1, elapsed time:%.2fs, sched-status:%d", pTask->id.idStr, el,
TASK_SCHED_STATUS__INACTIVE);
atomic_store_8(&pTask->status.schedStatus, TASK_SCHED_STATUS__INACTIVE);
streamMetaReleaseTask(pMeta, pTask);
return 0;
}
double el = (taosGetTimestampMs() - st) / 1000.0;
// the following procedure should be executed, no matter status is stop/pause or not
double el = (taosGetTimestampMs() - pTask->tsInfo.step1Start) / 1000.0;
tqDebug("s-task:%s scan-history stage(step 1) ended, elapsed time:%.2fs", id, el);
if (pTask->info.fillHistory) {
......@@ -1139,10 +1129,11 @@ int32_t tqProcessTaskScanHistory(STQ* pTq, SRpcMsg* pMsg) {
SStreamTask* pStreamTask = NULL;
bool done = false;
if (!pReq->igUntreated && !streamTaskRecoverScanStep1Finished(pTask)) {
// if (!pReq->igUntreated && !streamTaskRecoverScanStep1Finished(pTask)) {
// 1. stop the related stream task, get the current scan wal version of stream task, ver.
pStreamTask = streamMetaAcquireTask(pMeta, pTask->streamTaskId.taskId);
if (pStreamTask == NULL) {
// todo delete this task, if the related stream task is dropped
qError("failed to find s-task:0x%x, it may have been destroyed, drop fill-history task:%s",
pTask->streamTaskId.taskId, pTask->id.idStr);
......@@ -1174,42 +1165,36 @@ int32_t tqProcessTaskScanHistory(STQ* pTq, SRpcMsg* pMsg) {
pRange = &pTask->dataRange.range;
int64_t latestVer = walReaderGetCurrentVer(pStreamTask->exec.pWalReader);
done = streamHistoryTaskSetVerRangeStep2(pTask, latestVer);
}
if (done) {
pTask->tsInfo.step2Start = taosGetTimestampMs();
streamTaskEndScanWAL(pTask);
streamMetaReleaseTask(pMeta, pTask);
} else {
if (!streamTaskRecoverScanStep1Finished(pTask)) {
STimeWindow* pWindow = &pTask->dataRange.window;
tqDebug("s-task:%s level:%d verRange:%" PRId64 " - %" PRId64 " window:%" PRId64 "-%" PRId64
", do secondary scan-history from WAL after halt the related stream task:%s",
id, pTask->info.taskLevel, pRange->minVer, pRange->maxVer, pWindow->skey, pWindow->ekey, id);
id, pTask->info.taskLevel, pRange->minVer, pRange->maxVer, pWindow->skey, pWindow->ekey,
pStreamTask->id.idStr);
ASSERT(pTask->status.schedStatus == TASK_SCHED_STATUS__WAITING);
pTask->tsInfo.step2Start = taosGetTimestampMs();
streamSetParamForStreamScannerStep2(pTask, pRange, pWindow);
}
if (!streamTaskRecoverScanStep2Finished(pTask)) {
pTask->status.taskStatus = TASK_STATUS__SCAN_HISTORY_WAL;
if (streamTaskShouldStop(&pTask->status) || streamTaskShouldPause(&pTask->status)) {
tqDebug("s-task:%s is dropped or paused, abort recover in step1", id);
streamMetaReleaseTask(pMeta, pTask);
return 0;
}
int64_t dstVer = pTask->dataRange.range.minVer - 1;
pTask->chkInfo.currentVer = dstVer;
walReaderSetSkipToVersion(pTask->exec.pWalReader, dstVer);
tqDebug("s-task:%s wal reader start scan from WAL ver:%" PRId64 ", set sched-status:%d", id, dstVer,
TASK_SCHED_STATUS__INACTIVE);
}
tqDebug("s-task:%s wal reader start scan WAL verRange:%" PRId64 "-%" PRId64 ", set sched-status:%d", id, dstVer,
pTask->dataRange.range.maxVer, TASK_SCHED_STATUS__INACTIVE);
atomic_store_8(&pTask->status.schedStatus, TASK_SCHED_STATUS__INACTIVE);
// set the fill-history task to be normal
if (pTask->info.fillHistory == 1) {
streamSetStatusNormal(pTask);
}
// 4. 1) transfer the ownership of executor state, 2) update the scan data range for source task.
// 5. resume the related stream task.
streamMetaReleaseTask(pMeta, pTask);
......@@ -1226,7 +1211,7 @@ int32_t tqProcessTaskScanHistory(STQ* pTq, SRpcMsg* pMsg) {
if (pTask->historyTaskId.taskId == 0) {
*pWindow = (STimeWindow){INT64_MIN, INT64_MAX};
tqDebug(
"s-task:%s scanhistory in stream time window completed, no related fill-history task, reset the time "
"s-task:%s scan-history in stream time window completed, no related fill-history task, reset the time "
"window:%" PRId64 " - %" PRId64,
id, pWindow->skey, pWindow->ekey);
qResetStreamInfoTimeWindow(pTask->exec.pExecutor);
......@@ -1422,7 +1407,7 @@ int32_t tqProcessTaskRunReq(STQ* pTq, SRpcMsg* pMsg) {
if (pTask != NULL) {
// even in halt status, the data in inputQ must be processed
int8_t st = pTask->status.taskStatus;
if (st == TASK_STATUS__NORMAL || st == TASK_STATUS__SCAN_HISTORY || st == TASK_STATUS__SCAN_HISTORY_WAL) {
if (st == TASK_STATUS__NORMAL || st == TASK_STATUS__SCAN_HISTORY/* || st == TASK_STATUS__SCAN_HISTORY_WAL*/) {
tqDebug("vgId:%d s-task:%s start to process block from inputQ, last chk point:%" PRId64, vgId, pTask->id.idStr,
pTask->chkInfo.version);
streamProcessRunReq(pTask);
......@@ -1559,7 +1544,7 @@ int32_t tqProcessTaskResumeImpl(STQ* pTq, SStreamTask* pTask, int64_t sversion,
vgId, pTask->id.idStr, pTask->chkInfo.currentVer, sversion, pTask->status.schedStatus);
}
if (level == TASK_LEVEL__SOURCE && pTask->info.fillHistory) {
if (level == TASK_LEVEL__SOURCE && pTask->info.fillHistory && pTask->status.taskStatus == TASK_STATUS__SCAN_HISTORY) {
streamStartRecoverTask(pTask, igUntreated);
} else if (level == TASK_LEVEL__SOURCE && (taosQueueItemSize(pTask->inputQueue->queue) == 0)) {
tqStartStreamTasks(pTq);
......
......@@ -211,7 +211,7 @@ int32_t doSetOffsetForWalReader(SStreamTask *pTask, int32_t vgId) {
static void checkForFillHistoryVerRange(SStreamTask* pTask, int64_t ver) {
if ((pTask->info.fillHistory == 1) && ver > pTask->dataRange.range.maxVer) {
qWarn("s-task:%s fill-history scan WAL, currentVer:%" PRId64 "reach the maximum ver:%" PRId64
qWarn("s-task:%s fill-history scan WAL, currentVer:%" PRId64 " reach the maximum ver:%" PRId64
", not scan wal anymore, set the transfer state flag",
pTask->id.idStr, ver, pTask->dataRange.range.maxVer);
pTask->status.transferState = true;
......@@ -256,14 +256,15 @@ int32_t createStreamTaskRunReq(SStreamMeta* pStreamMeta, bool* pScanIdle) {
continue;
}
if (status != TASK_STATUS__NORMAL && status != TASK_STATUS__SCAN_HISTORY_WAL) {
if (status != TASK_STATUS__NORMAL/* && status != TASK_STATUS__SCAN_HISTORY_WAL*/) {
tqDebug("s-task:%s not ready for new submit block from wal, status:%s", pTask->id.idStr, streamGetTaskStatusStr(status));
streamMetaReleaseTask(pStreamMeta, pTask);
continue;
}
if ((pTask->info.fillHistory == 1) && pTask->status.transferState) {
ASSERT(status == TASK_STATUS__SCAN_HISTORY_WAL);
// ASSERT(status == TASK_STATUS__SCAN_HISTORY_WAL);
ASSERT(status == TASK_STATUS__NORMAL);
// the maximum version of data in the WAL has reached already, the step2 is done
tqDebug("s-task:%s fill-history reach the maximum ver:%" PRId64 ", not scan wal anymore", pTask->id.idStr,
pTask->dataRange.range.maxVer);
......
......@@ -615,7 +615,7 @@ int32_t streamTryExec(SStreamTask* pTask) {
// todo the task should be commit here
if (taosQueueEmpty(pTask->inputQueue->queue)) {
// fill-history WAL scan has completed
if (pTask->status.taskStatus == TASK_STATUS__SCAN_HISTORY_WAL && pTask->status.transferState == true) {
if (pTask->info.taskLevel == TASK_LEVEL__SOURCE && pTask->status.transferState == true) {
streamTaskRecoverSetAllStepFinished(pTask);
streamTaskEndScanWAL(pTask);
} else {
......
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