Merge pull request #21960 from taosdata/fix/3_liaohj

fix(stream): fix error for extract msg from inputQ

source/libs/stream/src/streamExec.c

@@ -399,81 +399,78 @@ static int32_t streamTransferStateToStreamTask(SStreamTask* pTask) {
   return TSDB_CODE_SUCCESS;
 }
 
-/**
- * todo: the batch of blocks should be tuned dynamic, according to the total elapsed time of each batch of blocks, the
- * appropriate batch of blocks should be handled in 5 to 10 sec.
- */
-int32_t streamExecForAll(SStreamTask* pTask) {
-  const char* id = pTask->id.idStr;
-
-  while (1) {
-    int32_t batchSize = 1;
-    int16_t times = 0;
-
-    SStreamQueueItem* pInput = NULL;
-
-    // merge multiple input data if possible in the input queue.
-    qDebug("s-task:%s start to extract data block from inputQ, status:%s", id, streamGetTaskStatusStr(pTask->status.taskStatus));
+static int32_t extractMsgFromInputQ(SStreamTask* pTask, SStreamQueueItem** pInput, int32_t* numOfBlocks,
+                                    const char* id) {
+  int32_t retryTimes = 0;
+  int32_t MAX_RETRY_TIMES = 5;
 
   while (1) {
-      // downstream task's input queue is blocked, stop immediately
-      if (streamTaskShouldPause(&pTask->status) || (pTask->outputStatus == TASK_OUTPUT_STATUS__BLOCKED) ||
-          streamTaskShouldStop(&pTask->status)) {
-        if (batchSize > 1) {
-          break;
-        } else {
-          qDebug("123 %s", pTask->id.idStr);
-          return 0;
-        }
+    if (streamTaskShouldPause(&pTask->status)) {
+      qDebug("s-task:%s task should pause, input blocks:%d", pTask->id.idStr, *numOfBlocks);
+      return TSDB_CODE_SUCCESS;
     }
 
     SStreamQueueItem* qItem = streamQueueNextItem(pTask->inputQueue);
     if (qItem == NULL) {
-        if (pTask->info.taskLevel == TASK_LEVEL__SOURCE && batchSize < MIN_STREAM_EXEC_BATCH_NUM && times < 5) {
-          times++;
+      if (pTask->info.taskLevel == TASK_LEVEL__SOURCE && (++retryTimes) < MAX_RETRY_TIMES) {
         taosMsleep(10);
-          qDebug("===stream===try again batchSize:%d", batchSize);
+        qDebug("===stream===try again batchSize:%d, retry:%d", *numOfBlocks, retryTimes);
         continue;
       }
 
-        qDebug("===stream===break batchSize:%d", batchSize);
-        break;
+      qDebug("===stream===break batchSize:%d", *numOfBlocks);
+      return TSDB_CODE_SUCCESS;
     }
 
-      if (pInput == NULL) {
-        pInput = qItem;
-        streamQueueProcessSuccess(pTask->inputQueue);
+    // do not merge blocks for sink node
     if (pTask->info.taskLevel == TASK_LEVEL__SINK) {
-          break;
+      *numOfBlocks = 1;
+      *pInput = qItem;
+      return TSDB_CODE_SUCCESS;
     }
+
+    if (*pInput == NULL) {
+      ASSERT((*numOfBlocks) == 0);
+      *pInput = qItem;
     } else {
       // todo we need to sort the data block, instead of just appending into the array list.
-        void* newRet = NULL;
-        if ((newRet = streamMergeQueueItem(pInput, qItem)) == NULL) {
+      void* newRet = streamMergeQueueItem(*pInput, qItem);
+      if (newRet == NULL) {
+        qError("s-task:%s failed to merge blocks from inputQ, numOfBlocks:%d", id, *numOfBlocks);
         streamQueueProcessFail(pTask->inputQueue);
-          break;
-        } else {
-          batchSize++;
-          pInput = newRet;
-          streamQueueProcessSuccess(pTask->inputQueue);
-
-          if (batchSize > MAX_STREAM_EXEC_BATCH_NUM) {
-            qDebug("s-task:%s batch size limit:%d reached, start to process blocks", id,
-                   MAX_STREAM_EXEC_BATCH_NUM);
-            break;
-          }
-        }
+        return TSDB_CODE_SUCCESS;
       }
+
+      *pInput = newRet;
     }
 
-    if (streamTaskShouldStop(&pTask->status)) {
-      if (pInput) {
-        streamFreeQitem(pInput);
+    *numOfBlocks += 1;
+    streamQueueProcessSuccess(pTask->inputQueue);
+
+    if (*numOfBlocks >= MAX_STREAM_EXEC_BATCH_NUM) {
+      qDebug("s-task:%s batch size limit:%d reached, start to process blocks", id, MAX_STREAM_EXEC_BATCH_NUM);
+      return TSDB_CODE_SUCCESS;
     }
-      return 0;
   }
+}
+
+/**
+ * todo: the batch of blocks should be tuned dynamic, according to the total elapsed time of each batch of blocks, the
+ * appropriate batch of blocks should be handled in 5 to 10 sec.
+ */
+int32_t streamExecForAll(SStreamTask* pTask) {
+  const char* id = pTask->id.idStr;
+
+  while (1) {
+    int32_t batchSize = 0;
+    SStreamQueueItem* pInput = NULL;
+
+    // merge multiple input data if possible in the input queue.
+    qDebug("s-task:%s start to extract data block from inputQ", id);
 
+    /*int32_t code = */extractMsgFromInputQ(pTask, &pInput, &batchSize, id);
     if (pInput == NULL) {
+      ASSERT(batchSize == 0);
       if (pTask->info.fillHistory && pTask->status.transferState) {
         int32_t code = streamTransferStateToStreamTask(pTask);
       }
@@ -532,8 +529,8 @@ int32_t streamExecForAll(SStreamTask* pTask) {
     streamTaskExecImpl(pTask, pInput, &resSize, &totalBlocks);
 
     double  el = (taosGetTimestampMs() - st) / 1000.0;
-    qDebug("s-task:%s batch of (%d)input blocks exec end, elapsed time:%.2fs, result size:%.2fMiB, numOfBlocks:%d",
-           id, batchSize, el, resSize / 1048576.0, totalBlocks);
+    qDebug("s-task:%s batch of input blocks exec end, elapsed time:%.2fs, result size:%.2fMiB, numOfBlocks:%d",
+           id, el, resSize / 1048576.0, totalBlocks);
 
     streamFreeQitem(pInput);
   }