#4308

Committed 14 Jun 2025 02:06PM UTC coverage: 62.454% (-0.3%) from 62.777%

Build # #4308

Build Type

push

travis-ci

Committed by

web-flow

Commit Message

fix: taosdump windows pthread_mutex_unlock crash(3.0) (#31357)

* fix: windows pthread_mutex_unlock crash

* enh: sync from main fix taosdump crash windows

* fix: restore .github action branch to main

Run Details

153985 of 315105 branches covered (48.87%)

Branch coverage included in aggregate %.

238120 of 312727 relevant lines covered (76.14%)

6462519.65 hits per line

Source File
Press 'n' to go to next uncovered line, 'b' for previous

63.03

/source/dnode/vnode/src/tq/tqStreamTask.c

/*
 * Copyright (c) 2019 TAOS Data, Inc. <jhtao@taosdata.com>
 *
 * This program is free software: you can use, redistribute, and/or modify
 * it under the terms of the GNU Affero General Public License, version 3
 * or later ("AGPL"), as published by the Free Software Foundation.
 *
 * This program is distributed in the hope that it will be useful, but WITHOUT
 * ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or
 * FITNESS FOR A PARTICULAR PURPOSE.
 *
 * You should have received a copy of the GNU Affero General Public License
 * along with this program. If not, see <http://www.gnu.org/licenses/>.
 */

#include "tq.h"
#include "vnd.h"

#define SCAN_WAL_IDLE_DURATION    250  // idle for 500ms to do next wal scan
#define SCAN_WAL_WAIT_COUNT       2

typedef struct SBuildScanWalMsgParam {
  int64_t metaId;
  SMsgCb  msgCb;
} SBuildScanWalMsgParam;

static int32_t doScanWalForAllTasks(SStreamMeta* pStreamMeta, int32_t* pNumOfTasks);
static int32_t setWalReaderStartOffset(SStreamTask* pTask, int32_t vgId);
static bool    handleFillhistoryScanComplete(SStreamTask* pTask, int64_t ver);
static bool    taskReadyForDataFromWal(SStreamTask* pTask);
static int32_t doPutDataIntoInputQ(SStreamTask* pTask, int64_t maxVer, int32_t* numOfItems, bool* pSucc);

// extract data blocks(submit/delete) from WAL, and add them into the input queue for all the sources tasks.
int32_t tqScanWal(STQ* pTq) {
  SStreamMeta* pMeta = pTq->pStreamMeta;
  int32_t      vgId = pMeta->vgId;
  int64_t      st = taosGetTimestampMs();
  int32_t      numOfTasks = 0;
  int64_t      el = 0;
  int32_t      code = 0;

  int32_t old = atomic_val_compare_exchange_32(&pMeta->scanInfo.scanSentinel, 0, 1);
  if (old == 0) {
    tqDebug("vgId:%d try to scan wal to extract data", vgId);
  } else {
    tqDebug("vgId:%d already in wal scan, abort", vgId);
    return code;
  }

  // the scan wal interval less than 200, not scan, actually.
  if ((pMeta->scanInfo.lastScanTs > 0) && (st - pMeta->scanInfo.lastScanTs < 200)) {
    tqDebug("vgId:%d scan wal less than 200ms, do nothing", vgId);
    atomic_store_32(&pMeta->scanInfo.scanSentinel, 0);
    return code;
  }

  // check all tasks
  code = doScanWalForAllTasks(pMeta, &numOfTasks);

  pMeta->scanInfo.lastScanTs = taosGetTimestampMs();
  el = (pMeta->scanInfo.lastScanTs - st);

  if (code) {
    tqError("vgId:%d failed to scan wal for all tasks, try next time, elapsed time:%" PRId64 "ms code:%s", vgId, el,
            tstrerror(code));
  } else {
    tqDebug("vgId:%d scan wal for stream tasks completed, elapsed time:%" PRId64 "ms", vgId, el);
  }

  atomic_store_32(&pMeta->scanInfo.scanSentinel, 0);
  return code;
}

static bool waitEnoughDuration(SStreamMeta* pMeta) {
  if ((++pMeta->scanInfo.tickCounter) >= SCAN_WAL_WAIT_COUNT) {
    pMeta->scanInfo.tickCounter = 0;
    return true;
  }

  return false;
}

static void doStartScanWal(void* param, void* tmrId) {
  int32_t                vgId = 0;
  int32_t                code = 0;
  int32_t                numOfTasks = 0;
  tmr_h                  pTimer = NULL;
  int32_t                numOfItems = 0;
  STQ*                   pTq = NULL;
  SBuildScanWalMsgParam* pParam = (SBuildScanWalMsgParam*)param;

  tqTrace("start to do scan wal in tmr, metaRid:%" PRId64, pParam->metaId);

  SStreamMeta* pMeta = taosAcquireRef(streamMetaRefPool, pParam->metaId);
  if (pMeta == NULL) {
    tqError("metaRid:%" PRId64 " not valid now, stream meta has been freed", pParam->metaId);
    taosMemoryFree(pParam);
    return;
  }

  pTq = pMeta->ahandle;
  vgId = pMeta->vgId;
  code = streamTimerGetInstance(&pTimer);
  if (code) {
    tqFatal("vgId:%d failed to get tmr ctrl during sched scan wal, not scan wal, code:%s", vgId, tstrerror(code));
    taosMemoryFree(pParam);
    return;
  }

  if (pMeta->closeFlag) {
    code = taosReleaseRef(streamMetaRefPool, pParam->metaId);
    if (code == TSDB_CODE_SUCCESS) {
      tqInfo("vgId:%d jump out of scan wal timer since closed", vgId);
    } else {
      tqError("vgId:%d failed to release ref for streamMeta, rid:%" PRId64 " code:%s", vgId, pParam->metaId,
              tstrerror(code));
    }

    taosMemoryFree(pParam);
    return;
  }

  if (pMeta->role != NODE_ROLE_LEADER) {
    tqDebug("vgId:%d not leader, role:%d not scan wal anymore", vgId, pMeta->role);

    code = taosReleaseRef(streamMetaRefPool, pParam->metaId);
    if (code == TSDB_CODE_SUCCESS) {
      tqDebug("vgId:%d jump out of scan wal timer since not leader", vgId);
    } else {
      tqError("vgId:%d failed to release ref for streamMeta, rid:%" PRId64 " code:%s", vgId, pParam->metaId,
              tstrerror(code));
    }

    taosMemFree(pParam);
    return;
  }

  if (pMeta->startInfo.startAllTasks) {
    tqDebug("vgId:%d in restart procedure, not ready to scan wal", vgId);
    goto _end;
  }

  numOfItems = tmsgGetQueueSize(&pTq->pVnode->msgCb, pMeta->vgId, STREAM_QUEUE);
  bool tooMany = (numOfItems > tsThresholdItemsInStreamQueue);

  if (!waitEnoughDuration(pMeta) || tooMany) {
    if (tooMany) {
      tqDebug("vgId:%d %d items (threshold: %d) in stream_queue, not scan wal now", vgId, numOfItems,
              tsThresholdItemsInStreamQueue);
    }

    streamTmrStart(doStartScanWal, SCAN_WAL_IDLE_DURATION, pParam, pTimer, &pMeta->scanInfo.scanTimer, vgId,
                   "scan-wal");
    code = taosReleaseRef(streamMetaRefPool, pParam->metaId);
    if (code) {
      tqError("vgId:%d failed to release ref for streamMeta, rid:%" PRId64 " code:%s", vgId, pParam->metaId,
              tstrerror(code));
    }
    return;
  }

  // failed to lock, try 500ms later
  code = streamMetaTryRlock(pMeta);
  if (code == 0) {
    numOfTasks = taosArrayGetSize(pMeta->pTaskList);
    streamMetaRUnLock(pMeta);
  } else {
    numOfTasks = 0;
  }

  if (numOfTasks > 0) {
    tqDebug("vgId:%d create msg to start wal scan, numOfTasks:%d", vgId, numOfTasks);

#if 0
  //  wait for the vnode is freed, and invalid read may occur.
  taosMsleep(10000);
#endif

    code = streamTaskSchedTask(&pParam->msgCb, vgId, 0, 0, STREAM_EXEC_T_EXTRACT_WAL_DATA, false);
    if (code) {
      tqError("vgId:%d failed sched task to scan wal, code:%s", vgId, tstrerror(code));
    }
  }

_end:
  streamTmrStart(doStartScanWal, SCAN_WAL_IDLE_DURATION, pParam, pTimer, &pMeta->scanInfo.scanTimer, vgId, "scan-wal");
  tqTrace("vgId:%d try scan-wal will start in %dms", vgId, SCAN_WAL_IDLE_DURATION*SCAN_WAL_WAIT_COUNT);

  code = taosReleaseRef(streamMetaRefPool, pParam->metaId);
  if (code) {
    tqError("vgId:%d failed to release ref for streamMeta, rid:%" PRId64 " code:%s", vgId, pParam->metaId,
            tstrerror(code));
  }
}

void tqScanWalAsync(STQ* pTq) {
  SStreamMeta*           pMeta = pTq->pStreamMeta;
  int32_t                code = 0;
  int32_t                vgId = TD_VID(pTq->pVnode);
  tmr_h                  pTimer = NULL;
  SBuildScanWalMsgParam* pParam = NULL;

  // 1. the vnode should be the leader.
  // 2. the stream isn't disabled
  if ((pMeta->role != NODE_ROLE_LEADER) || tsDisableStream) {
    tqInfo("vgId:%d follower node or stream disabled, not scan wal", vgId);
    return;
  }

  pParam = taosMemoryMalloc(sizeof(SBuildScanWalMsgParam));
  if (pParam == NULL) {
    tqError("vgId:%d failed to start scan wal, stream not executes, code:%s", vgId, tstrerror(code));
    return;
  }

  pParam->metaId = pMeta->rid;
  pParam->msgCb = pTq->pVnode->msgCb;

  code = streamTimerGetInstance(&pTimer);
  if (code) {
    tqFatal("vgId:%d failed to get tmr ctrl during sched scan wal", vgId);
    taosMemoryFree(pParam);
  } else {
    streamTmrStart(doStartScanWal, SCAN_WAL_IDLE_DURATION, pParam, pTimer, &pMeta->scanInfo.scanTimer, vgId,
                   "scan-wal");
  }
}

int32_t tqStopStreamAllTasksAsync(SStreamMeta* pMeta, SMsgCb* pMsgCb) {
  return streamTaskSchedTask(pMsgCb, pMeta->vgId, 0, 0, STREAM_EXEC_T_STOP_ALL_TASKS, false);
}

int32_t setWalReaderStartOffset(SStreamTask* pTask, int32_t vgId) {
  // seek the stored version and extract data from WAL
  int64_t firstVer = walReaderGetValidFirstVer(pTask->exec.pWalReader);
  if (pTask->chkInfo.nextProcessVer < firstVer) {
    tqWarn("vgId:%d s-task:%s ver:%" PRId64 " earlier than the first ver of wal range %" PRId64 ", forward to %" PRId64,
           vgId, pTask->id.idStr, pTask->chkInfo.nextProcessVer, firstVer, firstVer);

    pTask->chkInfo.nextProcessVer = firstVer;

    // todo need retry if failed
    int32_t code = walReaderSeekVer(pTask->exec.pWalReader, pTask->chkInfo.nextProcessVer);
    if (code != TSDB_CODE_SUCCESS) {
      return code;
    }

    // append the data for the stream
    tqDebug("vgId:%d s-task:%s wal reader seek to ver:%" PRId64, vgId, pTask->id.idStr, pTask->chkInfo.nextProcessVer);
  } else {
    int64_t currentVer = walReaderGetCurrentVer(pTask->exec.pWalReader);
    if (currentVer == -1) {  // we only seek the read for the first time
      int32_t code = walReaderSeekVer(pTask->exec.pWalReader, pTask->chkInfo.nextProcessVer);
      if (code != TSDB_CODE_SUCCESS) {  // no data in wal, quit
        return code;
      }

      // append the data for the stream
      tqDebug("vgId:%d s-task:%s wal reader initial seek to ver:%" PRId64, vgId, pTask->id.idStr,
              pTask->chkInfo.nextProcessVer);
    }
  }

  int64_t skipToVer = walReaderGetSkipToVersion(pTask->exec.pWalReader);
  if (skipToVer != 0 && skipToVer > pTask->chkInfo.nextProcessVer) {
    int32_t code = walReaderSeekVer(pTask->exec.pWalReader, skipToVer);
    if (code != TSDB_CODE_SUCCESS) {  // no data in wal, quit
      return code;
    }

    tqDebug("vgId:%d s-task:%s wal reader jump to ver:%" PRId64, vgId, pTask->id.idStr, skipToVer);
  }

  return TSDB_CODE_SUCCESS;
}

// todo handle memory error
bool handleFillhistoryScanComplete(SStreamTask* pTask, int64_t ver) {
  const char* id = pTask->id.idStr;
  int64_t     maxVer = pTask->step2Range.maxVer;

  if ((pTask->info.fillHistory == 1) && ver > maxVer) {
    if (!pTask->status.appendTranstateBlock) {
      qWarn("s-task:%s fill-history scan WAL, nextProcessVer:%" PRId64 " out of the maximum ver:%" PRId64
            ", not scan wal anymore, add transfer-state block into inputQ",
            id, ver, maxVer);

      double el = (taosGetTimestampMs() - pTask->execInfo.step2Start) / 1000.0;
      qDebug("s-task:%s scan-history from WAL stage(step 2) ended, range:%" PRId64 "-%" PRId64 ", elapsed time:%.2fs",
             id, pTask->step2Range.minVer, maxVer, el);
      int32_t code = streamTaskPutTranstateIntoInputQ(pTask);
      if (code) {
        qError("s-task:%s failed to put trans-state into inputQ", id);
      }

      return true;
    } else {
      qWarn("s-task:%s fill-history scan WAL, nextProcessVer:%" PRId64 " out of the ver range:%" PRId64 "-%" PRId64
            ", not scan wal",
            id, ver, pTask->step2Range.minVer, maxVer);
    }
  }

  return false;
}

bool taskReadyForDataFromWal(SStreamTask* pTask) {
  // non-source or fill-history tasks don't need to response the WAL scan action.
  SSTaskBasicInfo* pInfo = &pTask->info;
  if ((pInfo->taskLevel != TASK_LEVEL__SOURCE) || (pTask->status.downstreamReady == 0)) {
    return false;
  }

  if (pInfo->trigger == STREAM_TRIGGER_FORCE_WINDOW_CLOSE) {
    return false;
  }

  if (pInfo->fillHistory == STREAM_RECALCUL_TASK) {
    return false;
  }

  // not in ready state, do not handle the data from wal
  SStreamTaskState pState = streamTaskGetStatus(pTask);
  if (pState.state != TASK_STATUS__READY) {
    tqTrace("s-task:%s not ready for submit block in wal, status:%s", pTask->id.idStr, pState.name);
    return false;
  }

  // fill-history task has entered into the last phase, no need to do anything
  if ((pInfo->fillHistory == STREAM_HISTORY_TASK) && pTask->status.appendTranstateBlock) {
    // 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);
    return false;
  }

  // check whether input queue is full or not
  if (streamQueueIsFull(pTask->inputq.queue)) {
    tqTrace("s-task:%s input queue is full, launch task without scanning wal", pTask->id.idStr);
    int32_t code = streamTrySchedExec(pTask, false);
    if (code) {
      tqError("s-task:%s failed to start task while inputQ is full", pTask->id.idStr);
    }
    return false;
  }

  // the input queue of downstream task is full, so the output is blocked, stopped for a while
  if (pTask->inputq.status == TASK_INPUT_STATUS__BLOCKED) {
    tqDebug("s-task:%s inputQ is blocked, do nothing", pTask->id.idStr);
    return false;
  }

  return true;
}

int32_t doPutDataIntoInputQ(SStreamTask* pTask, int64_t maxVer, int32_t* numOfItems, bool* pSucc) {
  const char* id = pTask->id.idStr;
  int32_t     numOfNewItems = 0;
  int32_t     code = 0;
  *pSucc = false;

  while (1) {
    if ((pTask->info.fillHistory == 1) && pTask->status.appendTranstateBlock) {
      *numOfItems += numOfNewItems;
      return numOfNewItems > 0;
    }

    SStreamQueueItem* pItem = NULL;
    code = extractMsgFromWal(pTask->exec.pWalReader, (void**)&pItem, maxVer, id);
    if (code != TSDB_CODE_SUCCESS || pItem == NULL) {  // failed, continue
      int64_t currentVer = walReaderGetCurrentVer(pTask->exec.pWalReader);
      bool    itemInFillhistory = handleFillhistoryScanComplete(pTask, currentVer);
      if (itemInFillhistory) {
        numOfNewItems += 1;
      }
      break;
    }

    if (pItem != NULL) {
      code = streamTaskPutDataIntoInputQ(pTask, pItem);
      if (code == TSDB_CODE_SUCCESS) {
        numOfNewItems += 1;
        int64_t ver = walReaderGetCurrentVer(pTask->exec.pWalReader);
        pTask->chkInfo.nextProcessVer = ver;
        tqDebug("s-task:%s set ver:%" PRId64 " for reader after extract data from WAL", id, ver);

        bool itemInFillhistory = handleFillhistoryScanComplete(pTask, ver);
        if (itemInFillhistory) {
          break;
        }
      } else {
        if (code == TSDB_CODE_OUT_OF_MEMORY) {
          tqError("s-task:%s failed to put data into inputQ, since out of memory", id);
        } else {
          tqTrace("s-task:%s append input queue failed, code:inputQ is full, ver:%" PRId64, id,
                  pTask->chkInfo.nextProcessVer);
          code = walReaderSeekVer(pTask->exec.pWalReader, pTask->chkInfo.nextProcessVer);
          if (code) {
            tqError("s-task:%s failed to seek ver to:%" PRId64 " in wal", id, pTask->chkInfo.nextProcessVer);
          }

          code = 0;  // reset the error code
        }

        break;
      }
    }
  }

  *numOfItems += numOfNewItems;
  *pSucc = (numOfNewItems > 0);
  return code;
}

int32_t doScanWalForAllTasks(SStreamMeta* pStreamMeta, int32_t* pNumOfTasks) {
  int32_t vgId = pStreamMeta->vgId;
  SArray* pTaskList = NULL;
  int32_t numOfTasks = 0;

  // clone the task list, to avoid the task update during scan wal files
  streamMetaWLock(pStreamMeta);
  pTaskList = taosArrayDup(pStreamMeta->pTaskList, NULL);
  streamMetaWUnLock(pStreamMeta);
  if (pTaskList == NULL) {
    tqError("vgId:%d failed to create task list dup, code:%s", vgId, tstrerror(terrno));
    return terrno;
  }

  // update the new task number
  numOfTasks = taosArrayGetSize(pTaskList);
  if (pNumOfTasks != NULL) {
    *pNumOfTasks = numOfTasks;
  }

  tqDebug("vgId:%d start to check wal to extract new submit block for %d tasks", vgId, numOfTasks);

  for (int32_t i = 0; i < numOfTasks; ++i) {
    STaskId* pTaskId = taosArrayGet(pTaskList, i);
    if (pTaskId == NULL) {
      continue;
    }

    SStreamTask* pTask = NULL;
    int32_t      code = streamMetaAcquireTask(pStreamMeta, pTaskId->streamId, pTaskId->taskId, &pTask);
    if (pTask == NULL || code != 0) {
      continue;
    }

    if (!taskReadyForDataFromWal(pTask)) {
      streamMetaReleaseTask(pStreamMeta, pTask);
      continue;
    }

    // seek the stored version and extract data from WAL
    code = setWalReaderStartOffset(pTask, vgId);
    if (code != TSDB_CODE_SUCCESS) {
      streamMetaReleaseTask(pStreamMeta, pTask);
      continue;
    }

    int32_t numOfItems = streamQueueGetNumOfItems(pTask->inputq.queue);
    int64_t maxVer = (pTask->info.fillHistory == 1) ? pTask->step2Range.maxVer : INT64_MAX;

    streamMutexLock(&pTask->lock);

    SStreamTaskState state = streamTaskGetStatus(pTask);
    if (state.state != TASK_STATUS__READY) {
      tqDebug("s-task:%s not ready for submit block from wal, status:%s", pTask->id.idStr, state.name);
      streamMutexUnlock(&pTask->lock);
      streamMetaReleaseTask(pStreamMeta, pTask);
      continue;
    }

    bool hasNewData = false;
    code = doPutDataIntoInputQ(pTask, maxVer, &numOfItems, &hasNewData);
    streamMutexUnlock(&pTask->lock);

    TAOS_UNUSED(code);

    if ((numOfItems > 0) || hasNewData) {
      code = streamTrySchedExec(pTask, false);
      if (code != TSDB_CODE_SUCCESS) {
        streamMetaReleaseTask(pStreamMeta, pTask);
        taosArrayDestroy(pTaskList);
        return code;
      }
    }

    streamMetaReleaseTask(pStreamMeta, pTask);
  }

  taosArrayDestroy(pTaskList);
  return TSDB_CODE_SUCCESS;
}

void streamMetaFreeTQDuringScanWalError(STQ* pTq) {
  SBuildScanWalMsgParam* p = taosMemoryCalloc(1, sizeof(SBuildScanWalMsgParam));
  p->metaId = pTq->pStreamMeta->rid;

  doStartScanWal(p, 0);
}

1	/*
2	* Copyright (c) 2019 TAOS Data, Inc. <jhtao@taosdata.com>
3	*
4	* This program is free software: you can use, redistribute, and/or modify
5	* it under the terms of the GNU Affero General Public License, version 3
6	* or later ("AGPL"), as published by the Free Software Foundation.
7	*
8	* This program is distributed in the hope that it will be useful, but WITHOUT
9	* ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or
10	* FITNESS FOR A PARTICULAR PURPOSE.
11	*
12	* You should have received a copy of the GNU Affero General Public License
13	* along with this program. If not, see <http://www.gnu.org/licenses/>.
14	*/
15
16	#include "tq.h"
17	#include "vnd.h"
18
19	#define SCAN_WAL_IDLE_DURATION 250 // idle for 500ms to do next wal scan
20	#define SCAN_WAL_WAIT_COUNT 2
21
22	typedef struct SBuildScanWalMsgParam {
23	int64_t metaId;
24	SMsgCb msgCb;
25	} SBuildScanWalMsgParam;
26
27	static int32_t doScanWalForAllTasks(SStreamMeta* pStreamMeta, int32_t* pNumOfTasks);
28	static int32_t setWalReaderStartOffset(SStreamTask* pTask, int32_t vgId);
29	static bool handleFillhistoryScanComplete(SStreamTask* pTask, int64_t ver);
30	static bool taskReadyForDataFromWal(SStreamTask* pTask);
31	static int32_t doPutDataIntoInputQ(SStreamTask* pTask, int64_t maxVer, int32_t* numOfItems, bool* pSucc);
32
33	// extract data blocks(submit/delete) from WAL, and add them into the input queue for all the sources tasks.
34	int32_t tqScanWal(STQ* pTq) {	59,375✔
35	SStreamMeta* pMeta = pTq->pStreamMeta;	59,375✔
36	int32_t vgId = pMeta->vgId;	59,375✔
37	int64_t st = taosGetTimestampMs();	59,395✔
38	int32_t numOfTasks = 0;	59,395✔
39	int64_t el = 0;	59,395✔
40	int32_t code = 0;	59,395✔
41
42	int32_t old = atomic_val_compare_exchange_32(&pMeta->scanInfo.scanSentinel, 0, 1);	59,395✔
43	if (old == 0) {	59,393!
44	tqDebug("vgId:%d try to scan wal to extract data", vgId);	59,393✔
45	} else {
46	tqDebug("vgId:%d already in wal scan, abort", vgId);	×
47	return code;	×
48	}
49
50	// the scan wal interval less than 200, not scan, actually.
51	if ((pMeta->scanInfo.lastScanTs > 0) && (st - pMeta->scanInfo.lastScanTs < 200)) {	59,402✔
52	tqDebug("vgId:%d scan wal less than 200ms, do nothing", vgId);	1!
53	atomic_store_32(&pMeta->scanInfo.scanSentinel, 0);	1✔
54	return code;	1✔
55	}
56
57	// check all tasks
58	code = doScanWalForAllTasks(pMeta, &numOfTasks);	59,401✔
59
60	pMeta->scanInfo.lastScanTs = taosGetTimestampMs();	59,388✔
61	el = (pMeta->scanInfo.lastScanTs - st);	59,388✔
62
63	if (code) {	59,388!
64	tqError("vgId:%d failed to scan wal for all tasks, try next time, elapsed time:%" PRId64 "ms code:%s", vgId, el,	×
65	tstrerror(code));
66	} else {
67	tqDebug("vgId:%d scan wal for stream tasks completed, elapsed time:%" PRId64 "ms", vgId, el);	59,388!
68	}
69
70	atomic_store_32(&pMeta->scanInfo.scanSentinel, 0);	59,394✔
71	return code;	59,394✔
72	}
73
74	static bool waitEnoughDuration(SStreamMeta* pMeta) {	821,234✔
75	if ((++pMeta->scanInfo.tickCounter) >= SCAN_WAL_WAIT_COUNT) {	821,234✔
76	pMeta->scanInfo.tickCounter = 0;	407,834✔
77	return true;	407,834✔
78	}
79
80	return false;	413,400✔
81	}
82
83	static void doStartScanWal(void* param, void* tmrId) {	838,269✔
84	int32_t vgId = 0;	838,269✔
85	int32_t code = 0;	838,269✔
86	int32_t numOfTasks = 0;	838,269✔
87	tmr_h pTimer = NULL;	838,269✔
88	int32_t numOfItems = 0;	838,269✔
89	STQ* pTq = NULL;	838,269✔
90	SBuildScanWalMsgParam* pParam = (SBuildScanWalMsgParam*)param;	838,269✔
91
92	tqTrace("start to do scan wal in tmr, metaRid:%" PRId64, pParam->metaId);	838,269✔
93
94	SStreamMeta* pMeta = taosAcquireRef(streamMetaRefPool, pParam->metaId);	838,269✔
95	if (pMeta == NULL) {	838,269!
96	tqError("metaRid:%" PRId64 " not valid now, stream meta has been freed", pParam->metaId);	×
97	taosMemoryFree(pParam);	×
98	return;	424,816✔
99	}
100
101	pTq = pMeta->ahandle;	838,269✔
102	vgId = pMeta->vgId;	838,269✔
103	code = streamTimerGetInstance(&pTimer);	838,269✔
104	if (code) {	838,269!
105	tqFatal("vgId:%d failed to get tmr ctrl during sched scan wal, not scan wal, code:%s", vgId, tstrerror(code));	×
106	taosMemoryFree(pParam);	×
107	return;	×
108	}
109
110	if (pMeta->closeFlag) {	838,269✔
111	code = taosReleaseRef(streamMetaRefPool, pParam->metaId);	11,362✔
112	if (code == TSDB_CODE_SUCCESS) {	11,362!
113	tqInfo("vgId:%d jump out of scan wal timer since closed", vgId);	11,362!
114	} else {
115	tqError("vgId:%d failed to release ref for streamMeta, rid:%" PRId64 " code:%s", vgId, pParam->metaId,	×
116	tstrerror(code));
117	}
118
119	taosMemoryFree(pParam);	11,362!
120	return;	11,362✔
121	}
122
123	if (pMeta->role != NODE_ROLE_LEADER) {	826,907✔
124	tqDebug("vgId:%d not leader, role:%d not scan wal anymore", vgId, pMeta->role);	54!
125
126	code = taosReleaseRef(streamMetaRefPool, pParam->metaId);	54✔
127	if (code == TSDB_CODE_SUCCESS) {	54!
128	tqDebug("vgId:%d jump out of scan wal timer since not leader", vgId);	54!
129	} else {
130	tqError("vgId:%d failed to release ref for streamMeta, rid:%" PRId64 " code:%s", vgId, pParam->metaId,	×
131	tstrerror(code));
132	}
133
134	taosMemFree(pParam);	54✔
135	return;	54✔
136	}
137
138	if (pMeta->startInfo.startAllTasks) {	826,853✔
139	tqDebug("vgId:%d in restart procedure, not ready to scan wal", vgId);	5,619!
140	goto _end;	5,619✔
141	}
142
143	numOfItems = tmsgGetQueueSize(&pTq->pVnode->msgCb, pMeta->vgId, STREAM_QUEUE);	821,234✔
144	bool tooMany = (numOfItems > tsThresholdItemsInStreamQueue);	821,234✔
145
146	if (!waitEnoughDuration(pMeta) \|\| tooMany) {	821,234!
147	if (tooMany) {	413,400!
148	tqDebug("vgId:%d %d items (threshold: %d) in stream_queue, not scan wal now", vgId, numOfItems,	×
149	tsThresholdItemsInStreamQueue);
150	}
151
152	streamTmrStart(doStartScanWal, SCAN_WAL_IDLE_DURATION, pParam, pTimer, &pMeta->scanInfo.scanTimer, vgId,	413,400✔
153	"scan-wal");
154	code = taosReleaseRef(streamMetaRefPool, pParam->metaId);	413,400✔
155	if (code) {	413,400!
156	tqError("vgId:%d failed to release ref for streamMeta, rid:%" PRId64 " code:%s", vgId, pParam->metaId,	×
157	tstrerror(code));
158	}
159	return;	413,400✔
160	}
161
162	// failed to lock, try 500ms later
163	code = streamMetaTryRlock(pMeta);	407,834✔
164	if (code == 0) {	407,834✔
165	numOfTasks = taosArrayGetSize(pMeta->pTaskList);	407,593✔
166	streamMetaRUnLock(pMeta);	407,593✔
167	} else {
168	numOfTasks = 0;	241✔
169	}
170
171	if (numOfTasks > 0) {	407,834✔
172	tqDebug("vgId:%d create msg to start wal scan, numOfTasks:%d", vgId, numOfTasks);	59,395!
173
174	#if 0
175	// wait for the vnode is freed, and invalid read may occur.
176	taosMsleep(10000);
177	#endif
178
179	code = streamTaskSchedTask(&pParam->msgCb, vgId, 0, 0, STREAM_EXEC_T_EXTRACT_WAL_DATA, false);	59,395✔
180	if (code) {	59,395!
181	tqError("vgId:%d failed sched task to scan wal, code:%s", vgId, tstrerror(code));	×
182	}
183	}
184
185	_end:	407,834✔
186	streamTmrStart(doStartScanWal, SCAN_WAL_IDLE_DURATION, pParam, pTimer, &pMeta->scanInfo.scanTimer, vgId, "scan-wal");	413,453✔
187	tqTrace("vgId:%d try scan-wal will start in %dms", vgId, SCAN_WAL_IDLE_DURATION*SCAN_WAL_WAIT_COUNT);	413,453✔
188
189	code = taosReleaseRef(streamMetaRefPool, pParam->metaId);	413,453✔
190	if (code) {	413,453!
191	tqError("vgId:%d failed to release ref for streamMeta, rid:%" PRId64 " code:%s", vgId, pParam->metaId,	×
192	tstrerror(code));
193	}
194	}
195
196	void tqScanWalAsync(STQ* pTq) {	11,413✔
197	SStreamMeta* pMeta = pTq->pStreamMeta;	11,413✔
198	int32_t code = 0;	11,413✔
199	int32_t vgId = TD_VID(pTq->pVnode);	11,413✔
200	tmr_h pTimer = NULL;	11,413✔
201	SBuildScanWalMsgParam* pParam = NULL;	11,413✔
202
203	// 1. the vnode should be the leader.
204	// 2. the stream isn't disabled
205	if ((pMeta->role != NODE_ROLE_LEADER) \|\| tsDisableStream) {	11,413!
206	tqInfo("vgId:%d follower node or stream disabled, not scan wal", vgId);	×
207	return;	×
208	}
209
210	pParam = taosMemoryMalloc(sizeof(SBuildScanWalMsgParam));	11,415!
211	if (pParam == NULL) {	11,415!
212	tqError("vgId:%d failed to start scan wal, stream not executes, code:%s", vgId, tstrerror(code));	×
213	return;	×
214	}
215
216	pParam->metaId = pMeta->rid;	11,415✔
217	pParam->msgCb = pTq->pVnode->msgCb;	11,415✔
218
219	code = streamTimerGetInstance(&pTimer);	11,415✔
220	if (code) {	11,415!
221	tqFatal("vgId:%d failed to get tmr ctrl during sched scan wal", vgId);	×
222	taosMemoryFree(pParam);	×
223	} else {
224	streamTmrStart(doStartScanWal, SCAN_WAL_IDLE_DURATION, pParam, pTimer, &pMeta->scanInfo.scanTimer, vgId,	11,415✔
225	"scan-wal");
226	}
227	}
228
229	int32_t tqStopStreamAllTasksAsync(SStreamMeta* pMeta, SMsgCb* pMsgCb) {	6,166✔
230	return streamTaskSchedTask(pMsgCb, pMeta->vgId, 0, 0, STREAM_EXEC_T_STOP_ALL_TASKS, false);	6,166✔
231	}
232
233	int32_t setWalReaderStartOffset(SStreamTask* pTask, int32_t vgId) {	71,246✔
234	// seek the stored version and extract data from WAL
235	int64_t firstVer = walReaderGetValidFirstVer(pTask->exec.pWalReader);	71,246✔
236	if (pTask->chkInfo.nextProcessVer < firstVer) {	71,246!
237	tqWarn("vgId:%d s-task:%s ver:%" PRId64 " earlier than the first ver of wal range %" PRId64 ", forward to %" PRId64,	×
238	vgId, pTask->id.idStr, pTask->chkInfo.nextProcessVer, firstVer, firstVer);
239
240	pTask->chkInfo.nextProcessVer = firstVer;	×
241
242	// todo need retry if failed
243	int32_t code = walReaderSeekVer(pTask->exec.pWalReader, pTask->chkInfo.nextProcessVer);	×
244	if (code != TSDB_CODE_SUCCESS) {	×
245	return code;	×
246	}
247
248	// append the data for the stream
249	tqDebug("vgId:%d s-task:%s wal reader seek to ver:%" PRId64, vgId, pTask->id.idStr, pTask->chkInfo.nextProcessVer);	×
250	} else {
251	int64_t currentVer = walReaderGetCurrentVer(pTask->exec.pWalReader);	71,246✔
252	if (currentVer == -1) { // we only seek the read for the first time	71,246✔
253	int32_t code = walReaderSeekVer(pTask->exec.pWalReader, pTask->chkInfo.nextProcessVer);	2,597✔
254	if (code != TSDB_CODE_SUCCESS) { // no data in wal, quit	2,597!
255	return code;	×
256	}
257
258	// append the data for the stream
259	tqDebug("vgId:%d s-task:%s wal reader initial seek to ver:%" PRId64, vgId, pTask->id.idStr,	2,597!
260	pTask->chkInfo.nextProcessVer);
261	}
262	}
263
264	int64_t skipToVer = walReaderGetSkipToVersion(pTask->exec.pWalReader);	71,246✔
265	if (skipToVer != 0 && skipToVer > pTask->chkInfo.nextProcessVer) {	71,245✔
266	int32_t code = walReaderSeekVer(pTask->exec.pWalReader, skipToVer);	1✔
267	if (code != TSDB_CODE_SUCCESS) { // no data in wal, quit	1!
268	return code;	×
269	}
270
271	tqDebug("vgId:%d s-task:%s wal reader jump to ver:%" PRId64, vgId, pTask->id.idStr, skipToVer);	1!
272	}
273
274	return TSDB_CODE_SUCCESS;	71,245✔
275	}
276
277	// todo handle memory error
278	bool handleFillhistoryScanComplete(SStreamTask* pTask, int64_t ver) {	126,671✔
279	const char* id = pTask->id.idStr;	126,671✔
280	int64_t maxVer = pTask->step2Range.maxVer;	126,671✔
281
282	if ((pTask->info.fillHistory == 1) && ver > maxVer) {	126,671✔
283	if (!pTask->status.appendTranstateBlock) {	540!
284	qWarn("s-task:%s fill-history scan WAL, nextProcessVer:%" PRId64 " out of the maximum ver:%" PRId64	540!
285	", not scan wal anymore, add transfer-state block into inputQ",
286	id, ver, maxVer);
287
288	double el = (taosGetTimestampMs() - pTask->execInfo.step2Start) / 1000.0;	540✔
289	qDebug("s-task:%s scan-history from WAL stage(step 2) ended, range:%" PRId64 "-%" PRId64 ", elapsed time:%.2fs",	540✔
290	id, pTask->step2Range.minVer, maxVer, el);
291	int32_t code = streamTaskPutTranstateIntoInputQ(pTask);	540✔
292	if (code) {	540!
293	qError("s-task:%s failed to put trans-state into inputQ", id);	×
294	}
295
296	return true;	540✔
297	} else {
298	qWarn("s-task:%s fill-history scan WAL, nextProcessVer:%" PRId64 " out of the ver range:%" PRId64 "-%" PRId64	×
299	", not scan wal",
300	id, ver, pTask->step2Range.minVer, maxVer);
301	}
302	}
303
304	return false;	126,131✔
305	}
306
307	bool taskReadyForDataFromWal(SStreamTask* pTask) {	248,349✔
308	// non-source or fill-history tasks don't need to response the WAL scan action.
309	SSTaskBasicInfo* pInfo = &pTask->info;	248,349✔
310	if ((pInfo->taskLevel != TASK_LEVEL__SOURCE) \|\| (pTask->status.downstreamReady == 0)) {	248,349✔
311	return false;	118,187✔
312	}
313
314	if (pInfo->trigger == STREAM_TRIGGER_FORCE_WINDOW_CLOSE) {	130,162✔
315	return false;	36,780✔
316	}
317
318	if (pInfo->fillHistory == STREAM_RECALCUL_TASK) {	93,382!
319	return false;	×
320	}
321
322	// not in ready state, do not handle the data from wal
323	SStreamTaskState pState = streamTaskGetStatus(pTask);	93,382✔
324	if (pState.state != TASK_STATUS__READY) {	93,382✔
325	tqTrace("s-task:%s not ready for submit block in wal, status:%s", pTask->id.idStr, pState.name);	16,063✔
326	return false;	16,063✔
327	}
328
329	// fill-history task has entered into the last phase, no need to do anything
330	if ((pInfo->fillHistory == STREAM_HISTORY_TASK) && pTask->status.appendTranstateBlock) {	77,319✔
331	// the maximum version of data in the WAL has reached already, the step2 is done
332	tqDebug("s-task:%s fill-history reach the maximum ver:%" PRId64 ", not scan wal anymore", pTask->id.idStr,	6,074!
333	pTask->dataRange.range.maxVer);
334	return false;	6,074✔
335	}
336
337	// check whether input queue is full or not
338	if (streamQueueIsFull(pTask->inputq.queue)) {	71,245!
339	tqTrace("s-task:%s input queue is full, launch task without scanning wal", pTask->id.idStr);	×
340	int32_t code = streamTrySchedExec(pTask, false);	×
341	if (code) {	×
342	tqError("s-task:%s failed to start task while inputQ is full", pTask->id.idStr);	×
343	}
344	return false;	×
345	}
346
347	// the input queue of downstream task is full, so the output is blocked, stopped for a while
348	if (pTask->inputq.status == TASK_INPUT_STATUS__BLOCKED) {	71,246!
349	tqDebug("s-task:%s inputQ is blocked, do nothing", pTask->id.idStr);	×
350	return false;	×
351	}
352
353	return true;	71,246✔
354	}
355
356	int32_t doPutDataIntoInputQ(SStreamTask* pTask, int64_t maxVer, int32_t* numOfItems, bool* pSucc) {	71,244✔
357	const char* id = pTask->id.idStr;	71,244✔
358	int32_t numOfNewItems = 0;	71,244✔
359	int32_t code = 0;	71,244✔
360	*pSucc = false;	71,244✔
361
362	while (1) {	55,427✔
363	if ((pTask->info.fillHistory == 1) && pTask->status.appendTranstateBlock) {	126,671!
364	*numOfItems += numOfNewItems;	×
365	return numOfNewItems > 0;	×
366	}
367
368	SStreamQueueItem* pItem = NULL;	126,671✔
369	code = extractMsgFromWal(pTask->exec.pWalReader, (void**)&pItem, maxVer, id);	126,671✔
370	if (code != TSDB_CODE_SUCCESS \|\| pItem == NULL) { // failed, continue	126,670✔
371	int64_t currentVer = walReaderGetCurrentVer(pTask->exec.pWalReader);	71,046✔
372	bool itemInFillhistory = handleFillhistoryScanComplete(pTask, currentVer);	71,047✔
373	if (itemInFillhistory) {	71,046✔
374	numOfNewItems += 1;	342✔
375	}
376	break;	71,046✔
377	}
378
379	if (pItem != NULL) {	55,624!
380	code = streamTaskPutDataIntoInputQ(pTask, pItem);	55,624✔
381	if (code == TSDB_CODE_SUCCESS) {	55,625!
382	numOfNewItems += 1;	55,625✔
383	int64_t ver = walReaderGetCurrentVer(pTask->exec.pWalReader);	55,625✔
384	pTask->chkInfo.nextProcessVer = ver;	55,625✔
385	tqDebug("s-task:%s set ver:%" PRId64 " for reader after extract data from WAL", id, ver);	55,625!
386
387	bool itemInFillhistory = handleFillhistoryScanComplete(pTask, ver);	55,625✔
388	if (itemInFillhistory) {	55,625✔
389	break;	198✔
390	}
391	} else {
392	if (code == TSDB_CODE_OUT_OF_MEMORY) {	×
393	tqError("s-task:%s failed to put data into inputQ, since out of memory", id);	×
394	} else {
395	tqTrace("s-task:%s append input queue failed, code:inputQ is full, ver:%" PRId64, id,	×
396	pTask->chkInfo.nextProcessVer);
397	code = walReaderSeekVer(pTask->exec.pWalReader, pTask->chkInfo.nextProcessVer);	×
398	if (code) {	×
399	tqError("s-task:%s failed to seek ver to:%" PRId64 " in wal", id, pTask->chkInfo.nextProcessVer);	×
400	}
401
402	code = 0; // reset the error code	×
403	}
404
405	break;	×
406	}
407	}
408	}
409
410	*numOfItems += numOfNewItems;	71,244✔
411	*pSucc = (numOfNewItems > 0);	71,244✔
412	return code;	71,244✔
413	}
414
415	int32_t doScanWalForAllTasks(SStreamMeta* pStreamMeta, int32_t* pNumOfTasks) {	59,394✔
416	int32_t vgId = pStreamMeta->vgId;	59,394✔
417	SArray* pTaskList = NULL;	59,394✔
418	int32_t numOfTasks = 0;	59,394✔
419
420	// clone the task list, to avoid the task update during scan wal files
421	streamMetaWLock(pStreamMeta);	59,394✔
422	pTaskList = taosArrayDup(pStreamMeta->pTaskList, NULL);	59,393✔
423	streamMetaWUnLock(pStreamMeta);	59,394✔
424	if (pTaskList == NULL) {	59,394!
425	tqError("vgId:%d failed to create task list dup, code:%s", vgId, tstrerror(terrno));	×
426	return terrno;	×
427	}
428
429	// update the new task number
430	numOfTasks = taosArrayGetSize(pTaskList);	59,394✔
431	if (pNumOfTasks != NULL) {	59,394!
432	*pNumOfTasks = numOfTasks;	59,394✔
433	}
434
435	tqDebug("vgId:%d start to check wal to extract new submit block for %d tasks", vgId, numOfTasks);	59,394!
436
437	for (int32_t i = 0; i < numOfTasks; ++i) {	312,907✔
438	STaskId* pTaskId = taosArrayGet(pTaskList, i);	253,509✔
439	if (pTaskId == NULL) {	253,509!
440	continue;	182,268✔
441	}
442
443	SStreamTask* pTask = NULL;	253,509✔
444	int32_t code = streamMetaAcquireTask(pStreamMeta, pTaskId->streamId, pTaskId->taskId, &pTask);	253,509✔
445	if (pTask == NULL \|\| code != 0) {	253,505!
446	continue;	5,156✔
447	}
448
449	if (!taskReadyForDataFromWal(pTask)) {	248,349✔
450	streamMetaReleaseTask(pStreamMeta, pTask);	177,108✔
451	continue;	177,112✔
452	}
453
454	// seek the stored version and extract data from WAL
455	code = setWalReaderStartOffset(pTask, vgId);	71,242✔
456	if (code != TSDB_CODE_SUCCESS) {	71,245!
457	streamMetaReleaseTask(pStreamMeta, pTask);	×
458	continue;	×
459	}
460
461	int32_t numOfItems = streamQueueGetNumOfItems(pTask->inputq.queue);	71,245✔
462	int64_t maxVer = (pTask->info.fillHistory == 1) ? pTask->step2Range.maxVer : INT64_MAX;	71,245✔
463
464	streamMutexLock(&pTask->lock);	71,245✔
465
466	SStreamTaskState state = streamTaskGetStatus(pTask);	71,244✔
467	if (state.state != TASK_STATUS__READY) {	71,244!
468	tqDebug("s-task:%s not ready for submit block from wal, status:%s", pTask->id.idStr, state.name);	×
469	streamMutexUnlock(&pTask->lock);	×
470	streamMetaReleaseTask(pStreamMeta, pTask);	×
471	continue;	×
472	}
473
474	bool hasNewData = false;	71,244✔
475	code = doPutDataIntoInputQ(pTask, maxVer, &numOfItems, &hasNewData);	71,244✔
476	streamMutexUnlock(&pTask->lock);	71,245✔
477
478	TAOS_UNUSED(code);
479
480	if ((numOfItems > 0) \|\| hasNewData) {	71,243!
481	code = streamTrySchedExec(pTask, false);	7,368✔
482	if (code != TSDB_CODE_SUCCESS) {	7,367!
483	streamMetaReleaseTask(pStreamMeta, pTask);	×
484	taosArrayDestroy(pTaskList);	×
485	return code;	×
486	}
487	}
488
489	streamMetaReleaseTask(pStreamMeta, pTask);	71,242✔
490	}
491
492	taosArrayDestroy(pTaskList);	59,398✔
493	return TSDB_CODE_SUCCESS;	59,389✔
494	}
495
496	void streamMetaFreeTQDuringScanWalError(STQ* pTq) {	×
497	SBuildScanWalMsgParam* p = taosMemoryCalloc(1, sizeof(SBuildScanWalMsgParam));	×
498	p->metaId = pTq->pStreamMeta->rid;	×
499
500	doStartScanWal(p, 0);	×
501	}	×

taosdata / TDengine / #4308

Source File Press 'n' to go to next uncovered line, 'b' for previous

Source File
Press 'n' to go to next uncovered line, 'b' for previous