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

fix(stream): ignore the dispatch failure, and set the correct rsp counter.

This commit is contained in:
Haojun Liao 2023-09-21 17:22:09 +08:00
parent 2efd155adf
commit 4925b67d4a
2 changed files with 52 additions and 27 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

2
source/libs/stream/src/streamCheckpoint.c
View File

@ -264,7 +264,6 @@ int32_t streamProcessCheckpointReadyMsg(SStreamTask* pTask) {
int32_t streamSaveAllTaskStatus(SStreamMeta* pMeta, int64_t checkpointId) {
taosWLockLatch(&pMeta->lock);
int64_t keys[2];
for (int32_t i = 0; i < taosArrayGetSize(pMeta->pTaskList); ++i) {
STaskId* pId = taosArrayGet(pMeta->pTaskList, i);
SStreamTask** ppTask = taosHashGet(pMeta->pTasksMap, pId, sizeof(*pId));
@ -305,6 +304,7 @@ int32_t streamSaveAllTaskStatus(SStreamMeta* pMeta, int64_t checkpointId) {
return TSDB_CODE_SUCCESS;
}
// todo: handle the case: during the checkpoint procedure, leader/follower changes happened.
int32_t streamTaskBuildCheckpoint(SStreamTask* pTask) {
int32_t code = 0;

77
source/libs/stream/src/streamDispatch.c
View File

@ -947,20 +947,56 @@ int32_t streamNotifyUpstreamContinue(SStreamTask* pTask) {
return 0;
}
// this message has been sent successfully, let's try next one.
static int32_t handleDispatchSuccessRsp(SStreamTask* pTask, int32_t downstreamId) {
destroyStreamDataBlock(pTask->msgInfo.pData);
pTask->msgInfo.pData = NULL;
if (pTask->msgInfo.blockingTs != 0) {
int64_t el = taosGetTimestampMs() - pTask->msgInfo.blockingTs;
stDebug("s-task:%s downstream task:0x%x resume to normal from inputQ blocking, blocking time:%" PRId64 "ms",
pTask->id.idStr, downstreamId, el);
pTask->msgInfo.blockingTs = 0;
// put data into inputQ of current task is also allowed
pTask->inputInfo.status = TASK_INPUT_STATUS__NORMAL;
}
// now ready for next data output
atomic_store_8(&pTask->outputInfo.status, TASK_OUTPUT_STATUS__NORMAL);
// otherwise, continue dispatch the first block to down stream task in pipeline
streamDispatchStreamBlock(pTask);
return 0;
}
int32_t streamProcessDispatchRsp(SStreamTask* pTask, SStreamDispatchRsp* pRsp, int32_t code) {
const char* id = pTask->id.idStr;
if (code != TSDB_CODE_SUCCESS) {
// dispatch message failed: network error, or node not available.
// in case of the input queue is full, the code will be TSDB_CODE_SUCCESS, the and pRsp>inputStatus will be set
// in case of the input queue is full, the code will be TSDB_CODE_SUCCESS, the and pRsp->inputStatus will be set
// flag. here we need to retry dispatch this message to downstream task immediately. handle the case the failure
// happened too fast.
// todo handle the shuffle dispatch failure
if (code == TSDB_CODE_STREAM_TASK_NOT_EXIST) { // destination task does not exist, not retry anymore
stWarn("s-task:%s failed to dispatch msg to task:0x%x, no retry, since it is destroyed already", id, pRsp->downstreamTaskId);
stError("s-task:%s failed to dispatch msg to task:0x%x(vgId:%d), no retry, since it is destroyed already", id,
pRsp->downstreamTaskId, pRsp->downstreamNodeId);
{// we should set the correct finish flag to make sure the shuffle dispatch will be executed completed.
if (pTask->outputInfo.type == TASK_OUTPUT__SHUFFLE_DISPATCH) {
int32_t left = atomic_sub_fetch_32(&pTask->shuffleDispatcher.waitingRspCnt, 1);
if (left > 0) { // do nothing
stError("s-task:%s add the shuffle dispatch counter to complete the dispatch process", id);
} else {
stError("s-task:%s failed to dispatch msg to task:0x%x, code:%s, retry cnt:%d", id, pRsp->downstreamTaskId,
tstrerror(code), ++pTask->msgInfo.retryCount);
stError("s-task:%s the last rsp is failed, ignore it and continue, roll-back will discard this msg", id);
handleDispatchSuccessRsp(pTask, pRsp->downstreamTaskId);
}
}
}
} else {
stError("s-task:%s failed to dispatch msg to task:0x%x(vgId:%d), code:%s, retry cnt:%d", id,
pRsp->downstreamTaskId, pRsp->downstreamNodeId, tstrerror(code), ++pTask->msgInfo.retryCount);
int32_t ret = doDispatchAllBlocks(pTask, pTask->msgInfo.pData);
if (ret != TSDB_CODE_SUCCESS) {
}
@ -969,16 +1005,20 @@ int32_t streamProcessDispatchRsp(SStreamTask* pTask, SStreamDispatchRsp* pRsp, i
return TSDB_CODE_SUCCESS;
}
stDebug("s-task:%s recv dispatch rsp from 0x%x, downstream task input status:%d code:%d", id, pRsp->downstreamTaskId,
pRsp->inputStatus, code);
// there are other dispatch message not response yet
if (pTask->outputInfo.type == TASK_OUTPUT__SHUFFLE_DISPATCH) {
int32_t leftRsp = atomic_sub_fetch_32(&pTask->shuffleDispatcher.waitingRspCnt, 1);
stDebug("s-task:%s is shuffle, left waiting rsp %d", id, leftRsp);
if (leftRsp > 0) {
stDebug("s-task:%s recv dispatch rsp from 0x%x(vgId:%d), downstream task input status:%d code:%d, waiting for %d rsp", id, pRsp->downstreamTaskId,
pRsp->downstreamNodeId, pRsp->inputStatus, code, leftRsp);
return 0;
} else {
stDebug("s-task:%s recv dispatch rsp from 0x%x(vgId:%d), downstream task input status:%d code:%d, all rsp", id,
pRsp->downstreamTaskId, pRsp->downstreamNodeId, pRsp->inputStatus, code);
}
} else {
stDebug("s-task:%s recv fix-dispatch rsp from 0x%x(vgId:%d), downstream task input status:%d code:%d", id,
pRsp->downstreamTaskId, pRsp->downstreamNodeId, pRsp->inputStatus, code);
}
// transtate msg has been sent to downstream successfully. let's transfer the fill-history task state
@ -986,6 +1026,7 @@ int32_t streamProcessDispatchRsp(SStreamTask* pTask, SStreamDispatchRsp* pRsp, i
if (p->type == STREAM_INPUT__TRANS_STATE) {
stDebug("s-task:%s dispatch transtate msg to downstream successfully, start to transfer state", id);
ASSERT(pTask->info.fillHistory == 1);
code = streamTransferStateToStreamTask(pTask);
if (code != TSDB_CODE_SUCCESS) { // todo: do nothing if error happens
}
@ -1004,6 +1045,7 @@ int32_t streamProcessDispatchRsp(SStreamTask* pTask, SStreamDispatchRsp* pRsp, i
// so the TASK_INPUT_STATUS_BLOCKED is rsp
if (pRsp->inputStatus == TASK_INPUT_STATUS__BLOCKED) {
pTask->inputInfo.status = TASK_INPUT_STATUS__BLOCKED; // block the input of current task, to push pressure to upstream
double el = 0;
if (pTask->msgInfo.blockingTs == 0) {
pTask->msgInfo.blockingTs = taosGetTimestampMs(); // record the blocking start time
@ -1018,24 +1060,7 @@ int32_t streamProcessDispatchRsp(SStreamTask* pTask, SStreamDispatchRsp* pRsp, i
streamRetryDispatchStreamBlock(pTask, DISPATCH_RETRY_INTERVAL_MS);
} else { // pipeline send data in output queue
// this message has been sent successfully, let's try next one.
destroyStreamDataBlock(pTask->msgInfo.pData);
pTask->msgInfo.pData = NULL;
if (pTask->msgInfo.blockingTs != 0) {
int64_t el = taosGetTimestampMs() - pTask->msgInfo.blockingTs;
stDebug("s-task:%s downstream task:0x%x resume to normal from inputQ blocking, blocking time:%" PRId64 "ms", id,
pRsp->downstreamTaskId, el);
pTask->msgInfo.blockingTs = 0;
// put data into inputQ of current task is also allowed
pTask->inputInfo.status = TASK_INPUT_STATUS__NORMAL;
}
// now ready for next data output
atomic_store_8(&pTask->outputInfo.status, TASK_OUTPUT_STATUS__NORMAL);
// otherwise, continue dispatch the first block to down stream task in pipeline
streamDispatchStreamBlock(pTask);
handleDispatchSuccessRsp(pTask, pRsp->downstreamTaskId);
}
return 0;