tq.c 41.2 KB
Newer Older
H
refact  
Hongze Cheng 已提交
1 2 3 4 5 6 7 8 9 10 11 12 13
/*
 * 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/>.
S
Shengliang Guan 已提交
14 15
 */

H
Hongze Cheng 已提交
16
#include "tq.h"
S
Shengliang Guan 已提交
17

dengyihao's avatar
dengyihao 已提交
18 19 20
// 0: not init
// 1: already inited
// 2: wait to be inited or cleaup
21
#define WAL_READ_TASKS_ID       (-1)
22

23 24
static int32_t tqInitialize(STQ* pTq);

L
Liu Jicong 已提交
25
int32_t tqInit() {
L
Liu Jicong 已提交
26 27 28 29 30 31
  int8_t old;
  while (1) {
    old = atomic_val_compare_exchange_8(&tqMgmt.inited, 0, 2);
    if (old != 2) break;
  }

32 33 34 35 36 37
  if (old == 0) {
    tqMgmt.timer = taosTmrInit(10000, 100, 10000, "TQ");
    if (tqMgmt.timer == NULL) {
      atomic_store_8(&tqMgmt.inited, 0);
      return -1;
    }
38 39 40
    if (streamInit() < 0) {
      return -1;
    }
L
Liu Jicong 已提交
41
    atomic_store_8(&tqMgmt.inited, 1);
42
  }
43

L
Liu Jicong 已提交
44 45
  return 0;
}
L
Liu Jicong 已提交
46

47
void tqCleanUp() {
L
Liu Jicong 已提交
48 49 50 51 52 53 54 55
  int8_t old;
  while (1) {
    old = atomic_val_compare_exchange_8(&tqMgmt.inited, 1, 2);
    if (old != 2) break;
  }

  if (old == 1) {
    taosTmrCleanUp(tqMgmt.timer);
L
Liu Jicong 已提交
56
    streamCleanUp();
L
Liu Jicong 已提交
57 58
    atomic_store_8(&tqMgmt.inited, 0);
  }
59
}
L
Liu Jicong 已提交
60

61
static void destroyTqHandle(void* data) {
62 63 64
  STqHandle* pData = (STqHandle*)data;
  qDestroyTask(pData->execHandle.task);
  if (pData->execHandle.subType == TOPIC_SUB_TYPE__COLUMN) {
L
Liu Jicong 已提交
65
    taosMemoryFreeClear(pData->execHandle.execCol.qmsg);
66
  } else if (pData->execHandle.subType == TOPIC_SUB_TYPE__DB) {
67
    tqCloseReader(pData->execHandle.pTqReader);
68 69
    walCloseReader(pData->pWalReader);
    taosHashCleanup(pData->execHandle.execDb.pFilterOutTbUid);
L
Liu Jicong 已提交
70
  } else if (pData->execHandle.subType == TOPIC_SUB_TYPE__TABLE) {
71
    walCloseReader(pData->pWalReader);
72
    tqCloseReader(pData->execHandle.pTqReader);
73 74
    taosMemoryFreeClear(pData->execHandle.execTb.qmsg);
    nodesDestroyNode(pData->execHandle.execTb.node);
75
  }
76 77 78 79
  if(pData->msg != NULL) {
    rpcFreeCont(pData->msg->pCont);
    taosMemoryFree(pData->msg);
    pData->msg = NULL;
H
Haojun Liao 已提交
80
  }
L
Liu Jicong 已提交
81 82
}

83 84 85 86 87
static bool tqOffsetLessOrEqual(const STqOffset* pLeft, const STqOffset* pRight) {
  return pLeft->val.type == TMQ_OFFSET__LOG && pRight->val.type == TMQ_OFFSET__LOG &&
         pLeft->val.version <= pRight->val.version;
}

L
Liu Jicong 已提交
88
STQ* tqOpen(const char* path, SVnode* pVnode) {
89
  STQ* pTq = taosMemoryCalloc(1, sizeof(STQ));
L
Liu Jicong 已提交
90
  if (pTq == NULL) {
S
Shengliang Guan 已提交
91
    terrno = TSDB_CODE_OUT_OF_MEMORY;
L
Liu Jicong 已提交
92 93
    return NULL;
  }
94

95
  pTq->path = taosStrdup(path);
L
Liu Jicong 已提交
96
  pTq->pVnode = pVnode;
L
Liu Jicong 已提交
97
  pTq->walLogLastVer = pVnode->pWal->vers.lastVer;
98

99
  pTq->pHandle = taosHashInit(64, MurmurHash3_32, true, HASH_ENTRY_LOCK);
100
  taosHashSetFreeFp(pTq->pHandle, destroyTqHandle);
101

102
  taosInitRWLatch(&pTq->lock);
103
  pTq->pPushMgr = taosHashInit(64, MurmurHash3_32, false, HASH_NO_LOCK);
L
Liu Jicong 已提交
104

105
  pTq->pCheckInfo = taosHashInit(64, MurmurHash3_32, true, HASH_ENTRY_LOCK);
L
Liu Jicong 已提交
106
  taosHashSetFreeFp(pTq->pCheckInfo, (FDelete)tDeleteSTqCheckInfo);
L
Liu Jicong 已提交
107

108 109 110 111 112 113 114
  int32_t code = tqInitialize(pTq);
  if (code != TSDB_CODE_SUCCESS) {
    tqClose(pTq);
    return NULL;
  } else {
    return pTq;
  }
115 116 117
}

int32_t tqInitialize(STQ* pTq) {
L
Liu Jicong 已提交
118
  if (tqMetaOpen(pTq) < 0) {
119
    return -1;
120 121
  }

L
Liu Jicong 已提交
122 123
  pTq->pOffsetStore = tqOffsetOpen(pTq);
  if (pTq->pOffsetStore == NULL) {
124
    return -1;
125 126
  }

127
  pTq->pStreamMeta = streamMetaOpen(pTq->path, pTq, (FTaskExpand*)tqExpandTask, pTq->pVnode->config.vgId);
L
Liu Jicong 已提交
128
  if (pTq->pStreamMeta == NULL) {
129
    return -1;
L
Liu Jicong 已提交
130 131
  }

132 133
  // the version is kept in task's meta data
  // todo check if this version is required or not
134 135
  if (streamLoadTasks(pTq->pStreamMeta, walGetCommittedVer(pTq->pVnode->pWal)) < 0) {
    return -1;
L
Liu Jicong 已提交
136 137
  }

138
  return 0;
L
Liu Jicong 已提交
139
}
L
Liu Jicong 已提交
140

L
Liu Jicong 已提交
141
void tqClose(STQ* pTq) {
142 143
  if (pTq == NULL) {
    return;
H
Hongze Cheng 已提交
144
  }
145 146 147 148 149 150 151 152 153

  tqOffsetClose(pTq->pOffsetStore);
  taosHashCleanup(pTq->pHandle);
  taosHashCleanup(pTq->pPushMgr);
  taosHashCleanup(pTq->pCheckInfo);
  taosMemoryFree(pTq->path);
  tqMetaClose(pTq);
  streamMetaClose(pTq->pStreamMeta);
  taosMemoryFree(pTq);
L
Liu Jicong 已提交
154
}
L
Liu Jicong 已提交
155

H
Haojun Liao 已提交
156 157 158 159 160 161 162 163 164 165 166 167 168
void tqNotifyClose(STQ* pTq) {
  if (pTq != NULL) {
    taosWLockLatch(&pTq->pStreamMeta->lock);

    void* pIter = NULL;
    while (1) {
      pIter = taosHashIterate(pTq->pStreamMeta->pTasks, pIter);
      if (pIter == NULL) {
        break;
      }

      SStreamTask* pTask = *(SStreamTask**)pIter;
      tqDebug("vgId:%d s-task:%s set dropping flag", pTq->pStreamMeta->vgId, pTask->id.idStr);
169 170 171
      pTask->status.taskStatus = TASK_STATUS__STOP;

      int64_t st = taosGetTimestampMs();
H
Haojun Liao 已提交
172
      qKillTask(pTask->exec.pExecutor, TSDB_CODE_SUCCESS);
173 174
      int64_t el = taosGetTimestampMs() - st;
      tqDebug("vgId:%d s-task:%s is closed in %" PRId64 "ms", pTq->pStreamMeta->vgId, pTask->id.idStr, el);
H
Haojun Liao 已提交
175 176 177 178 179 180
    }

    taosWUnLockLatch(&pTq->pStreamMeta->lock);
  }
}

H
Haojun Liao 已提交
181 182
static int32_t doSendDataRsp(const SRpcHandleInfo* pRpcHandleInfo, const SMqDataRsp* pRsp, int32_t epoch,
                             int64_t consumerId, int32_t type) {
L
Liu Jicong 已提交
183 184
  int32_t len = 0;
  int32_t code = 0;
H
Haojun Liao 已提交
185 186 187 188 189 190

  if (type == TMQ_MSG_TYPE__POLL_RSP) {
    tEncodeSize(tEncodeSMqDataRsp, pRsp, len, code);
  } else if (type == TMQ_MSG_TYPE__TAOSX_RSP) {
    tEncodeSize(tEncodeSTaosxRsp, (STaosxRsp*)pRsp, len, code);
  }
L
Liu Jicong 已提交
191 192 193 194 195 196 197 198 199 200 201

  if (code < 0) {
    return -1;
  }

  int32_t tlen = sizeof(SMqRspHead) + len;
  void*   buf = rpcMallocCont(tlen);
  if (buf == NULL) {
    return -1;
  }

H
Haojun Liao 已提交
202 203 204
  ((SMqRspHead*)buf)->mqMsgType = type;
  ((SMqRspHead*)buf)->epoch = epoch;
  ((SMqRspHead*)buf)->consumerId = consumerId;
L
Liu Jicong 已提交
205 206 207 208 209

  void* abuf = POINTER_SHIFT(buf, sizeof(SMqRspHead));

  SEncoder encoder = {0};
  tEncoderInit(&encoder, abuf, len);
H
Haojun Liao 已提交
210 211 212 213

  if (type == TMQ_MSG_TYPE__POLL_RSP) {
    tEncodeSMqDataRsp(&encoder, pRsp);
  } else if (type == TMQ_MSG_TYPE__TAOSX_RSP) {
X
Xiaoyu Wang 已提交
214
    tEncodeSTaosxRsp(&encoder, (STaosxRsp*)pRsp);
H
Haojun Liao 已提交
215 216
  }

L
Liu Jicong 已提交
217 218 219
  tEncoderClear(&encoder);

  SRpcMsg rsp = {
H
Haojun Liao 已提交
220
      .info = *pRpcHandleInfo,
L
Liu Jicong 已提交
221 222 223 224 225 226 227 228 229
      .pCont = buf,
      .contLen = tlen,
      .code = 0,
  };

  tmsgSendRsp(&rsp);
  return 0;
}

230 231 232 233 234 235
int32_t tqPushDataRsp(STQ* pTq, STqHandle* pHandle) {
  SMqDataRsp dataRsp = {0};
  dataRsp.head.consumerId = pHandle->consumerId;
  dataRsp.head.epoch = pHandle->epoch;
  dataRsp.head.mqMsgType = TMQ_MSG_TYPE__POLL_RSP;
  doSendDataRsp(&pHandle->msg->info, &dataRsp, pHandle->epoch, pHandle->consumerId, TMQ_MSG_TYPE__POLL_RSP);
L
Liu Jicong 已提交
236

wmmhello's avatar
wmmhello 已提交
237 238
  char buf1[80] = {0};
  char buf2[80] = {0};
239 240
  tFormatOffset(buf1, tListLen(buf1), &dataRsp.reqOffset);
  tFormatOffset(buf2, tListLen(buf2), &dataRsp.rspOffset);
H
Haojun Liao 已提交
241
  tqDebug("vgId:%d, from consumer:0x%" PRIx64 " (epoch %d) push rsp, block num: %d, req:%s, rsp:%s",
242
          TD_VID(pTq->pVnode), dataRsp.head.consumerId, dataRsp.head.epoch, dataRsp.blockNum, buf1, buf2);
L
Liu Jicong 已提交
243 244 245
  return 0;
}

H
Haojun Liao 已提交
246 247
int32_t tqSendDataRsp(STQ* pTq, const SRpcMsg* pMsg, const SMqPollReq* pReq, const SMqDataRsp* pRsp, int32_t type) {
  doSendDataRsp(&pMsg->info, pRsp, pReq->epoch, pReq->consumerId, type);
248 249 250 251 252 253

  char buf1[80] = {0};
  char buf2[80] = {0};
  tFormatOffset(buf1, 80, &pRsp->reqOffset);
  tFormatOffset(buf2, 80, &pRsp->rspOffset);

X
Xiaoyu Wang 已提交
254
  tqDebug("vgId:%d consumer:0x%" PRIx64 " (epoch %d) send rsp, block num:%d, req:%s, rsp:%s, reqId:0x%" PRIx64,
H
Haojun Liao 已提交
255
          TD_VID(pTq->pVnode), pReq->consumerId, pReq->epoch, pRsp->blockNum, buf1, buf2, pReq->reqId);
H
Haojun Liao 已提交
256

257 258 259
  return 0;
}

260
int32_t tqProcessOffsetCommitReq(STQ* pTq, int64_t sversion, char* msg, int32_t msgLen) {
261
  STqOffset offset = {0};
X
Xiaoyu Wang 已提交
262
  int32_t   vgId = TD_VID(pTq->pVnode);
263

X
Xiaoyu Wang 已提交
264 265
  SDecoder decoder;
  tDecoderInit(&decoder, (uint8_t*)msg, msgLen);
266 267 268
  if (tDecodeSTqOffset(&decoder, &offset) < 0) {
    return -1;
  }
269

270 271
  tDecoderClear(&decoder);

wmmhello's avatar
wmmhello 已提交
272
  if (offset.val.type == TMQ_OFFSET__SNAPSHOT_DATA || offset.val.type == TMQ_OFFSET__SNAPSHOT_META) {
L
Liu Jicong 已提交
273
    tqDebug("receive offset commit msg to %s on vgId:%d, offset(type:snapshot) uid:%" PRId64 ", ts:%" PRId64,
H
Haojun Liao 已提交
274
            offset.subKey, vgId, offset.val.uid, offset.val.ts);
L
Liu Jicong 已提交
275
  } else if (offset.val.type == TMQ_OFFSET__LOG) {
X
Xiaoyu Wang 已提交
276 277
    tqDebug("receive offset commit msg to %s on vgId:%d, offset(type:log) version:%" PRId64, offset.subKey, vgId,
            offset.val.version);
278
    if (offset.val.version + 1 == sversion) {
279 280
      offset.val.version += 1;
    }
281
  } else {
282 283
    tqError("invalid commit offset type:%d", offset.val.type);
    return -1;
284
  }
285 286 287 288

  STqOffset* pSavedOffset = tqOffsetRead(pTq->pOffsetStore, offset.subKey);
  if (pSavedOffset != NULL && tqOffsetLessOrEqual(&offset, pSavedOffset)) {
    return 0;  // no need to update the offset value
289 290
  }

291
  // save the new offset value
292 293
  if (tqOffsetWrite(pTq->pOffsetStore, &offset) < 0) {
    return -1;
294
  }
295 296

  if (offset.val.type == TMQ_OFFSET__LOG) {
297
    STqHandle* pHandle = taosHashGet(pTq->pHandle, offset.subKey, strlen(offset.subKey));
298 299
    if (pHandle && (walRefVer(pHandle->pRef, offset.val.version) < 0)) {
      return -1;
300 301 302
    }
  }

303 304 305
  return 0;
}

L
Liu Jicong 已提交
306
int32_t tqCheckColModifiable(STQ* pTq, int64_t tbUid, int32_t colId) {
L
Liu Jicong 已提交
307
  void* pIter = NULL;
308

L
Liu Jicong 已提交
309
  while (1) {
310
    pIter = taosHashIterate(pTq->pCheckInfo, pIter);
311 312 313 314
    if (pIter == NULL) {
      break;
    }

315
    STqCheckInfo* pCheck = (STqCheckInfo*)pIter;
316

L
Liu Jicong 已提交
317 318
    if (pCheck->ntbUid == tbUid) {
      int32_t sz = taosArrayGetSize(pCheck->colIdList);
L
Liu Jicong 已提交
319
      for (int32_t i = 0; i < sz; i++) {
L
Liu Jicong 已提交
320 321
        int16_t forbidColId = *(int16_t*)taosArrayGet(pCheck->colIdList, i);
        if (forbidColId == colId) {
322
          taosHashCancelIterate(pTq->pCheckInfo, pIter);
L
Liu Jicong 已提交
323 324 325 326 327
          return -1;
        }
      }
    }
  }
328

L
Liu Jicong 已提交
329 330 331
  return 0;
}

332
int32_t tqProcessPollReq(STQ* pTq, SRpcMsg* pMsg) {
X
Xiaoyu Wang 已提交
333
  SMqPollReq req = {0};
334 335 336 337 338 339 340 341 342 343 344 345 346 347 348 349 350 351 352
  if (tDeserializeSMqPollReq(pMsg->pCont, pMsg->contLen, &req) < 0) {
    tqError("tDeserializeSMqPollReq %d failed", pMsg->contLen);
    terrno = TSDB_CODE_INVALID_MSG;
    return -1;
  }

  int64_t      consumerId = req.consumerId;
  int32_t      reqEpoch = req.epoch;
  STqOffsetVal reqOffset = req.reqOffset;
  int32_t      vgId = TD_VID(pTq->pVnode);

  // 1. find handle
  STqHandle* pHandle = taosHashGet(pTq->pHandle, req.subKey, strlen(req.subKey));
  if (pHandle == NULL) {
    tqError("tmq poll: consumer:0x%" PRIx64 " vgId:%d subkey %s not found", consumerId, vgId, req.subKey);
    terrno = TSDB_CODE_INVALID_MSG;
    return -1;
  }

353
  // 2. check re-balance status
354
  taosRLockLatch(&pTq->lock);
355 356 357 358
  if (pHandle->consumerId != consumerId) {
    tqDebug("ERROR tmq poll: consumer:0x%" PRIx64 " vgId:%d, subkey %s, mismatch for saved handle consumer:0x%" PRIx64,
            consumerId, TD_VID(pTq->pVnode), req.subKey, pHandle->consumerId);
    terrno = TSDB_CODE_TMQ_CONSUMER_MISMATCH;
359
    taosRUnLockLatch(&pTq->lock);
360 361
    return -1;
  }
362
  taosRUnLockLatch(&pTq->lock);
363

364
  // 3. update the epoch value
365
  taosWLockLatch(&pTq->lock);
H
Haojun Liao 已提交
366 367
  int32_t savedEpoch = pHandle->epoch;
  if (savedEpoch < reqEpoch) {
X
Xiaoyu Wang 已提交
368 369
    tqDebug("tmq poll: consumer:0x%" PRIx64 " epoch update from %d to %d by poll req", consumerId, savedEpoch,
            reqEpoch);
370
    pHandle->epoch = reqEpoch;
H
Haojun Liao 已提交
371
  }
372
  taosWUnLockLatch(&pTq->lock);
373 374 375

  char buf[80];
  tFormatOffset(buf, 80, &reqOffset);
H
Haojun Liao 已提交
376 377
  tqDebug("tmq poll: consumer:0x%" PRIx64 " (epoch %d), subkey %s, recv poll req vgId:%d, req:%s, reqId:0x%" PRIx64,
          consumerId, req.epoch, pHandle->subKey, vgId, buf, req.reqId);
378

379
  return tqExtractDataForMq(pTq, pHandle, &req, pMsg);
380 381
}

382
int32_t tqProcessDeleteSubReq(STQ* pTq, int64_t sversion, char* msg, int32_t msgLen) {
L
Liu Jicong 已提交
383
  SMqVDeleteReq* pReq = (SMqVDeleteReq*)msg;
L
Liu Jicong 已提交
384

L
Liu Jicong 已提交
385
  tqDebug("vgId:%d, tq process delete sub req %s", pTq->pVnode->config.vgId, pReq->subKey);
wmmhello's avatar
wmmhello 已提交
386
  int32_t code = 0;
wmmhello's avatar
wmmhello 已提交
387 388 389 390 391 392
//  taosWLockLatch(&pTq->lock);
//  int32_t code = taosHashRemove(pTq->pPushMgr, pReq->subKey, strlen(pReq->subKey));
//  if (code != 0) {
//    tqDebug("vgId:%d, tq remove push handle %s", pTq->pVnode->config.vgId, pReq->subKey);
//  }
//  taosWUnLockLatch(&pTq->lock);
L
Liu Jicong 已提交
393

L
Liu Jicong 已提交
394 395
  STqHandle* pHandle = taosHashGet(pTq->pHandle, pReq->subKey, strlen(pReq->subKey));
  if (pHandle) {
X
Xiaoyu Wang 已提交
396
    // walCloseRef(pHandle->pWalReader->pWal, pHandle->pRef->refId);
L
Liu Jicong 已提交
397 398 399 400 401 402 403
    if (pHandle->pRef) {
      walCloseRef(pTq->pVnode->pWal, pHandle->pRef->refId);
    }
    code = taosHashRemove(pTq->pHandle, pReq->subKey, strlen(pReq->subKey));
    if (code != 0) {
      tqError("cannot process tq delete req %s, since no such handle", pReq->subKey);
    }
L
Liu Jicong 已提交
404
  }
405

L
Liu Jicong 已提交
406 407
  code = tqOffsetDelete(pTq->pOffsetStore, pReq->subKey);
  if (code != 0) {
408
    tqError("cannot process tq delete req %s, since no such offset in cache", pReq->subKey);
L
Liu Jicong 已提交
409
  }
L
Liu Jicong 已提交
410

L
Liu Jicong 已提交
411
  if (tqMetaDeleteHandle(pTq, pReq->subKey) < 0) {
L
Liu Jicong 已提交
412
    tqError("cannot process tq delete req %s, since no such offset in tdb", pReq->subKey);
413
  }
L
Liu Jicong 已提交
414
  return 0;
L
Liu Jicong 已提交
415 416
}

417
int32_t tqProcessAddCheckInfoReq(STQ* pTq, int64_t sversion, char* msg, int32_t msgLen) {
418 419
  STqCheckInfo info = {0};
  SDecoder     decoder;
X
Xiaoyu Wang 已提交
420
  tDecoderInit(&decoder, (uint8_t*)msg, msgLen);
421
  if (tDecodeSTqCheckInfo(&decoder, &info) < 0) {
L
Liu Jicong 已提交
422 423 424 425
    terrno = TSDB_CODE_OUT_OF_MEMORY;
    return -1;
  }
  tDecoderClear(&decoder);
426 427 428 429 430
  if (taosHashPut(pTq->pCheckInfo, info.topic, strlen(info.topic), &info, sizeof(STqCheckInfo)) < 0) {
    terrno = TSDB_CODE_OUT_OF_MEMORY;
    return -1;
  }
  if (tqMetaSaveCheckInfo(pTq, info.topic, msg, msgLen) < 0) {
L
Liu Jicong 已提交
431 432 433 434 435 436
    terrno = TSDB_CODE_OUT_OF_MEMORY;
    return -1;
  }
  return 0;
}

437
int32_t tqProcessDelCheckInfoReq(STQ* pTq, int64_t sversion, char* msg, int32_t msgLen) {
438 439 440 441 442 443 444 445 446 447 448
  if (taosHashRemove(pTq->pCheckInfo, msg, strlen(msg)) < 0) {
    terrno = TSDB_CODE_OUT_OF_MEMORY;
    return -1;
  }
  if (tqMetaDeleteCheckInfo(pTq, msg) < 0) {
    terrno = TSDB_CODE_OUT_OF_MEMORY;
    return -1;
  }
  return 0;
}

449
int32_t tqProcessSubscribeReq(STQ* pTq, int64_t sversion, char* msg, int32_t msgLen) {
450
  int ret = 0;
L
Liu Jicong 已提交
451
  SMqRebVgReq req = {0};
L
Liu Jicong 已提交
452
  tDecodeSMqRebVgReq(msg, &req);
L
Liu Jicong 已提交
453

454 455 456
  SVnode* pVnode = pTq->pVnode;
  int32_t vgId = TD_VID(pVnode);

457
  tqDebug("vgId:%d, tq process sub req:%s, Id:0x%" PRIx64 " -> Id:0x%" PRIx64, pVnode->config.vgId, req.subKey,
458
          req.oldConsumerId, req.newConsumerId);
L
Liu Jicong 已提交
459

460
  STqHandle* pHandle = taosHashGet(pTq->pHandle, req.subKey, strlen(req.subKey));
L
Liu Jicong 已提交
461
  if (pHandle == NULL) {
L
Liu Jicong 已提交
462
    if (req.oldConsumerId != -1) {
463
      tqError("vgId:%d, build new consumer handle %s for consumer:0x%" PRIx64 ", but old consumerId:0x%" PRIx64,
464
              req.vgId, req.subKey, req.newConsumerId, req.oldConsumerId);
L
Liu Jicong 已提交
465
    }
466

L
Liu Jicong 已提交
467
    if (req.newConsumerId == -1) {
468
      tqError("vgId:%d, tq invalid re-balance request, new consumerId %" PRId64 "", req.vgId, req.newConsumerId);
469
      goto end;
L
Liu Jicong 已提交
470
    }
471

L
Liu Jicong 已提交
472 473
    STqHandle tqHandle = {0};
    pHandle = &tqHandle;
L
Liu Jicong 已提交
474

L
Liu Jicong 已提交
475 476 477
    memcpy(pHandle->subKey, req.subKey, TSDB_SUBSCRIBE_KEY_LEN);
    pHandle->consumerId = req.newConsumerId;
    pHandle->epoch = -1;
L
Liu Jicong 已提交
478

L
Liu Jicong 已提交
479
    pHandle->execHandle.subType = req.subType;
L
Liu Jicong 已提交
480
    pHandle->fetchMeta = req.withMeta;
wmmhello's avatar
wmmhello 已提交
481

482
    // TODO version should be assigned and refed during preprocess
483
    SWalRef* pRef = walRefCommittedVer(pVnode->pWal);
484
    if (pRef == NULL) {
485 486
      ret = -1;
      goto end;
487
    }
H
Haojun Liao 已提交
488

489 490
    int64_t ver = pRef->refVer;
    pHandle->pRef = pRef;
L
Liu Jicong 已提交
491

492
    SReadHandle handle = {
493
        .meta = pVnode->pMeta, .vnode = pVnode, .initTableReader = true, .initTqReader = true, .version = ver};
wmmhello's avatar
wmmhello 已提交
494
    pHandle->snapshotVer = ver;
495

L
Liu Jicong 已提交
496
    if (pHandle->execHandle.subType == TOPIC_SUB_TYPE__COLUMN) {
L
Liu Jicong 已提交
497
      pHandle->execHandle.execCol.qmsg = req.qmsg;
L
Liu Jicong 已提交
498
      req.qmsg = NULL;
499

X
Xiaoyu Wang 已提交
500 501
      pHandle->execHandle.task = qCreateQueueExecTaskInfo(pHandle->execHandle.execCol.qmsg, &handle, vgId,
                                                          &pHandle->execHandle.numOfCols, req.newConsumerId);
L
Liu Jicong 已提交
502
      void* scanner = NULL;
503
      qExtractStreamScanner(pHandle->execHandle.task, &scanner);
504
      pHandle->execHandle.pTqReader = qExtractReaderFromStreamScanner(scanner);
L
Liu Jicong 已提交
505
    } else if (pHandle->execHandle.subType == TOPIC_SUB_TYPE__DB) {
506
      pHandle->pWalReader = walOpenReader(pVnode->pWal, NULL);
507
      pHandle->execHandle.pTqReader = tqReaderOpen(pVnode);
508

L
Liu Jicong 已提交
509
      pHandle->execHandle.execDb.pFilterOutTbUid =
L
Liu Jicong 已提交
510
          taosHashInit(64, taosGetDefaultHashFunction(TSDB_DATA_TYPE_BIGINT), false, HASH_NO_LOCK);
511 512
      buildSnapContext(handle.meta, handle.version, 0, pHandle->execHandle.subType, pHandle->fetchMeta,
                       (SSnapContext**)(&handle.sContext));
513

514
      pHandle->execHandle.task = qCreateQueueExecTaskInfo(NULL, &handle, vgId, NULL, req.newConsumerId);
L
Liu Jicong 已提交
515
    } else if (pHandle->execHandle.subType == TOPIC_SUB_TYPE__TABLE) {
516
      pHandle->pWalReader = walOpenReader(pVnode->pWal, NULL);
wmmhello's avatar
wmmhello 已提交
517
      pHandle->execHandle.execTb.suid = req.suid;
518 519 520 521 522 523 524
      pHandle->execHandle.execTb.qmsg = req.qmsg;
      req.qmsg = NULL;

      if (nodesStringToNode(pHandle->execHandle.execTb.qmsg, &pHandle->execHandle.execTb.node) != 0) {
        tqError("nodesStringToNode error in sub stable, since %s", terrstr());
        return -1;
      }
wmmhello's avatar
wmmhello 已提交
525

526 527 528 529 530 531
      SArray* tbUidList = NULL;
      ret = qGetTableList(req.suid, pVnode->pMeta, pVnode, pHandle->execHandle.execTb.node, NULL, &tbUidList);
      if(ret != TDB_CODE_SUCCESS) {
        tqError("qGetTableList error:%d handle %s consumer:0x%" PRIx64, ret, req.subKey, pHandle->consumerId);
        taosArrayDestroy(tbUidList);
        goto end;
L
Liu Jicong 已提交
532
      }
533
      tqDebug("vgId:%d, tq try to get ctb for stb subscribe, suid:%" PRId64, pVnode->config.vgId, req.suid);
534
      pHandle->execHandle.pTqReader = tqReaderOpen(pVnode);
535
      tqReaderSetTbUidList(pHandle->execHandle.pTqReader, tbUidList);
L
Liu Jicong 已提交
536
      taosArrayDestroy(tbUidList);
wmmhello's avatar
wmmhello 已提交
537

L
Liu Jicong 已提交
538 539
      buildSnapContext(handle.meta, handle.version, req.suid, pHandle->execHandle.subType, pHandle->fetchMeta,
                       (SSnapContext**)(&handle.sContext));
540
      pHandle->execHandle.task = qCreateQueueExecTaskInfo(NULL, &handle, vgId, NULL, req.newConsumerId);
L
Liu Jicong 已提交
541
    }
H
Haojun Liao 已提交
542

543
    taosHashPut(pTq->pHandle, req.subKey, strlen(req.subKey), pHandle, sizeof(STqHandle));
544 545
    tqDebug("try to persist handle %s consumer:0x%" PRIx64, req.subKey,
            pHandle->consumerId);
546 547
    ret = tqMetaSaveHandle(pTq, req.subKey, pHandle);
    goto end;
L
Liu Jicong 已提交
548
  } else {
549 550 551
    if (pHandle->consumerId == req.newConsumerId) {  // do nothing
      tqInfo("vgId:%d consumer:0x%" PRIx64 " remains, no switch occurs", req.vgId, req.newConsumerId);
      atomic_add_fetch_32(&pHandle->epoch, 1);
552

553 554 555
    } else {
      tqInfo("vgId:%d switch consumer from Id:0x%" PRIx64 " to Id:0x%" PRIx64, req.vgId, pHandle->consumerId,
             req.newConsumerId);
556
      atomic_store_64(&pHandle->consumerId, req.newConsumerId);
wmmhello's avatar
wmmhello 已提交
557
      atomic_store_32(&pHandle->epoch, 0);
558 559 560 561 562 563
    }
    // kill executing task
    qTaskInfo_t pTaskInfo = pHandle->execHandle.task;
    if (pTaskInfo != NULL) {
      qKillTask(pTaskInfo, TSDB_CODE_SUCCESS);
    }
564

565 566 567
    taosWLockLatch(&pTq->lock);
    // remove if it has been register in the push manager, and return one empty block to consumer
    tqUnregisterPushHandle(pTq, pHandle);
568

569

570 571
    if (pHandle->execHandle.subType == TOPIC_SUB_TYPE__COLUMN) {
      qStreamCloseTsdbReader(pTaskInfo);
L
Liu Jicong 已提交
572
    }
573 574 575 576

    taosWUnLockLatch(&pTq->lock);
    ret = tqMetaSaveHandle(pTq, req.subKey, pHandle);
    goto end;
L
Liu Jicong 已提交
577
  }
L
Liu Jicong 已提交
578

579
end:
H
Haojun Liao 已提交
580
  taosMemoryFree(req.qmsg);
581
  return ret;
L
Liu Jicong 已提交
582
}
583

584
int32_t tqExpandTask(STQ* pTq, SStreamTask* pTask, int64_t ver) {
585
  int32_t vgId = TD_VID(pTq->pVnode);
586
  pTask->id.idStr = createStreamTaskIdStr(pTask->id.streamId, pTask->id.taskId);
L
Liu Jicong 已提交
587
  pTask->refCnt = 1;
588
  pTask->status.schedStatus = TASK_SCHED_STATUS__INACTIVE;
L
Liu Jicong 已提交
589 590
  pTask->inputQueue = streamQueueOpen();
  pTask->outputQueue = streamQueueOpen();
L
Liu Jicong 已提交
591 592

  if (pTask->inputQueue == NULL || pTask->outputQueue == NULL) {
L
Liu Jicong 已提交
593
    return -1;
L
Liu Jicong 已提交
594 595
  }

L
Liu Jicong 已提交
596 597
  pTask->inputStatus = TASK_INPUT_STATUS__NORMAL;
  pTask->outputStatus = TASK_OUTPUT_STATUS__NORMAL;
598
  pTask->pMsgCb = &pTq->pVnode->msgCb;
599
  pTask->pMeta = pTq->pStreamMeta;
600
  pTask->chkInfo.version = ver;
601
  pTask->chkInfo.currentVer = ver;
602

603
  // expand executor
604
  pTask->status.taskStatus = (pTask->fillHistory)? TASK_STATUS__WAIT_DOWNSTREAM:TASK_STATUS__NORMAL;
605

606
  if (pTask->taskLevel == TASK_LEVEL__SOURCE) {
607
    pTask->pState = streamStateOpen(pTq->pStreamMeta->path, pTask, false, -1, -1);
608 609 610 611
    if (pTask->pState == NULL) {
      return -1;
    }

612
    SReadHandle handle = {
613
        .meta = pTq->pVnode->pMeta, .vnode = pTq->pVnode, .initTqReader = 1, .pStateBackend = pTask->pState};
614

615 616
    pTask->exec.pExecutor = qCreateStreamExecTaskInfo(pTask->exec.qmsg, &handle, vgId);
    if (pTask->exec.pExecutor == NULL) {
L
Liu Jicong 已提交
617 618
      return -1;
    }
619

620
  } else if (pTask->taskLevel == TASK_LEVEL__AGG) {
621
    pTask->pState = streamStateOpen(pTq->pStreamMeta->path, pTask, false, -1, -1);
622 623 624
    if (pTask->pState == NULL) {
      return -1;
    }
625

626 627 628 629 630
    int32_t numOfVgroups = (int32_t)taosArrayGetSize(pTask->childEpInfo);
    SReadHandle mgHandle = { .vnode = NULL, .numOfVgroups = numOfVgroups, .pStateBackend = pTask->pState};

    pTask->exec.pExecutor = qCreateStreamExecTaskInfo(pTask->exec.qmsg, &mgHandle, vgId);
    if (pTask->exec.pExecutor == NULL) {
L
Liu Jicong 已提交
631 632
      return -1;
    }
L
Liu Jicong 已提交
633
  }
L
Liu Jicong 已提交
634 635

  // sink
L
Liu Jicong 已提交
636
  /*pTask->ahandle = pTq->pVnode;*/
637
  if (pTask->outputType == TASK_OUTPUT__SMA) {
L
Liu Jicong 已提交
638
    pTask->smaSink.vnode = pTq->pVnode;
L
Liu Jicong 已提交
639
    pTask->smaSink.smaSink = smaHandleRes;
640
  } else if (pTask->outputType == TASK_OUTPUT__TABLE) {
L
Liu Jicong 已提交
641
    pTask->tbSink.vnode = pTq->pVnode;
L
Liu Jicong 已提交
642
    pTask->tbSink.tbSinkFunc = tqSinkToTablePipeline2;
L
Liu Jicong 已提交
643

X
Xiaoyu Wang 已提交
644
    int32_t   ver1 = 1;
5
54liuyao 已提交
645
    SMetaInfo info = {0};
dengyihao's avatar
dengyihao 已提交
646
    int32_t   code = metaGetInfo(pTq->pVnode->pMeta, pTask->tbSink.stbUid, &info, NULL);
5
54liuyao 已提交
647
    if (code == TSDB_CODE_SUCCESS) {
H
Haojun Liao 已提交
648
      ver1 = info.skmVer;
5
54liuyao 已提交
649
    }
L
Liu Jicong 已提交
650

651 652
    SSchemaWrapper* pschemaWrapper = pTask->tbSink.pSchemaWrapper;
    pTask->tbSink.pTSchema = tBuildTSchema(pschemaWrapper->pSchema, pschemaWrapper->nCols, ver1);
wmmhello's avatar
wmmhello 已提交
653
    if(pTask->tbSink.pTSchema == NULL) {
wmmhello's avatar
wmmhello 已提交
654
      return -1;
wmmhello's avatar
wmmhello 已提交
655
    }
L
Liu Jicong 已提交
656
  }
657

658
  if (pTask->taskLevel == TASK_LEVEL__SOURCE) {
659
    pTask->exec.pWalReader = walOpenReader(pTq->pVnode->pWal, NULL);
660 661
  }

662
  streamSetupTrigger(pTask);
663

664
  tqInfo("vgId:%d expand stream task, s-task:%s, checkpoint ver:%" PRId64 " child id:%d, level:%d", vgId, pTask->id.idStr,
665
         pTask->chkInfo.version, pTask->selfChildId, pTask->taskLevel);
666 667 668

  // next valid version will add one
  pTask->chkInfo.version += 1;
L
Liu Jicong 已提交
669
  return 0;
L
Liu Jicong 已提交
670
}
L
Liu Jicong 已提交
671

672 673 674 675 676 677
int32_t tqProcessStreamTaskCheckReq(STQ* pTq, SRpcMsg* pMsg) {
  char*               msgStr = pMsg->pCont;
  char*               msgBody = POINTER_SHIFT(msgStr, sizeof(SMsgHead));
  int32_t             msgLen = pMsg->contLen - sizeof(SMsgHead);
  SStreamTaskCheckReq req;
  SDecoder            decoder;
X
Xiaoyu Wang 已提交
678
  tDecoderInit(&decoder, (uint8_t*)msgBody, msgLen);
679 680 681 682 683 684 685 686 687 688 689 690
  tDecodeSStreamTaskCheckReq(&decoder, &req);
  tDecoderClear(&decoder);
  int32_t             taskId = req.downstreamTaskId;
  SStreamTaskCheckRsp rsp = {
      .reqId = req.reqId,
      .streamId = req.streamId,
      .childId = req.childId,
      .downstreamNodeId = req.downstreamNodeId,
      .downstreamTaskId = req.downstreamTaskId,
      .upstreamNodeId = req.upstreamNodeId,
      .upstreamTaskId = req.upstreamTaskId,
  };
691

L
Liu Jicong 已提交
692
  SStreamTask* pTask = streamMetaAcquireTask(pTq->pStreamMeta, taskId);
693

694
  if (pTask) {
695
    rsp.status = streamTaskCheckStatus(pTask);
696 697 698 699 700 701
    streamMetaReleaseTask(pTq->pStreamMeta, pTask);

    tqDebug("tq recv task check req(reqId:0x%" PRIx64
            ") %d at node %d task status:%d, check req from task %d at node %d, rsp status %d",
            rsp.reqId, rsp.downstreamTaskId, rsp.downstreamNodeId, pTask->status.taskStatus, rsp.upstreamTaskId,
            rsp.upstreamNodeId, rsp.status);
702 703
  } else {
    rsp.status = 0;
704 705 706 707
    tqDebug("tq recv task check(taskId:%d not built yet) req(reqId:0x%" PRIx64
            ") %d at node %d, check req from task %d at node %d, rsp status %d",
            taskId, rsp.reqId, rsp.downstreamTaskId, rsp.downstreamNodeId, rsp.upstreamTaskId, rsp.upstreamNodeId,
            rsp.status);
708 709 710 711 712 713 714
  }

  SEncoder encoder;
  int32_t  code;
  int32_t  len;
  tEncodeSize(tEncodeSStreamTaskCheckRsp, &rsp, len, code);
  if (code < 0) {
L
Liu Jicong 已提交
715
    tqError("unable to encode rsp %d", __LINE__);
L
Liu Jicong 已提交
716
    return -1;
717
  }
L
Liu Jicong 已提交
718

719 720 721 722 723 724 725 726
  void* buf = rpcMallocCont(sizeof(SMsgHead) + len);
  ((SMsgHead*)buf)->vgId = htonl(req.upstreamNodeId);

  void* abuf = POINTER_SHIFT(buf, sizeof(SMsgHead));
  tEncoderInit(&encoder, (uint8_t*)abuf, len);
  tEncodeSStreamTaskCheckRsp(&encoder, &rsp);
  tEncoderClear(&encoder);

727
  SRpcMsg rspMsg = { .code = 0, .pCont = buf, .contLen = sizeof(SMsgHead) + len, .info = pMsg->info };
728 729 730 731
  tmsgSendRsp(&rspMsg);
  return 0;
}

732
int32_t tqProcessStreamTaskCheckRsp(STQ* pTq, int64_t sversion, char* msg, int32_t msgLen) {
733 734 735 736 737 738 739 740 741 742 743
  int32_t             code;
  SStreamTaskCheckRsp rsp;

  SDecoder decoder;
  tDecoderInit(&decoder, (uint8_t*)msg, msgLen);
  code = tDecodeSStreamTaskCheckRsp(&decoder, &rsp);
  if (code < 0) {
    tDecoderClear(&decoder);
    return -1;
  }

744
  tDecoderClear(&decoder);
745
  tqDebug("tq recv task check rsp(reqId:0x%" PRIx64 ") %d at node %d check req from task %d at node %d, status %d",
746 747
          rsp.reqId, rsp.downstreamTaskId, rsp.downstreamNodeId, rsp.upstreamTaskId, rsp.upstreamNodeId, rsp.status);

L
Liu Jicong 已提交
748
  SStreamTask* pTask = streamMetaAcquireTask(pTq->pStreamMeta, rsp.upstreamTaskId);
749 750 751 752
  if (pTask == NULL) {
    return -1;
  }

753
  code = streamProcessTaskCheckRsp(pTask, &rsp, sversion);
L
Liu Jicong 已提交
754 755
  streamMetaReleaseTask(pTq->pStreamMeta, pTask);
  return code;
756 757
}

758
int32_t tqProcessTaskDeployReq(STQ* pTq, int64_t sversion, char* msg, int32_t msgLen) {
759 760 761 762 763
  int32_t code;
#if 0
  code = streamMetaAddSerializedTask(pTq->pStreamMeta, version, msg, msgLen);
  if (code < 0) return code;
#endif
5
54liuyao 已提交
764 765 766
  if (tsDisableStream) {
    return 0;
  }
767 768 769 770 771 772

  // 1.deserialize msg and build task
  SStreamTask* pTask = taosMemoryCalloc(1, sizeof(SStreamTask));
  if (pTask == NULL) {
    return -1;
  }
773

774 775
  SDecoder decoder;
  tDecoderInit(&decoder, (uint8_t*)msg, msgLen);
776
  code = tDecodeStreamTask(&decoder, pTask);
777 778 779 780 781
  if (code < 0) {
    tDecoderClear(&decoder);
    taosMemoryFree(pTask);
    return -1;
  }
782

783 784
  tDecoderClear(&decoder);

785
  // 2.save task, use the newest commit version as the initial start version of stream task.
786
  taosWLockLatch(&pTq->pStreamMeta->lock);
787
  code = streamMetaAddDeployedTask(pTq->pStreamMeta, sversion, pTask);
788
  if (code < 0) {
789 790
    tqError("vgId:%d failed to add s-task:%s, total:%d", TD_VID(pTq->pVnode), pTask->id.idStr,
            streamMetaGetNumOfTasks(pTq->pStreamMeta));
791
    taosWUnLockLatch(&pTq->pStreamMeta->lock);
792 793 794
    return -1;
  }

795 796
  taosWUnLockLatch(&pTq->pStreamMeta->lock);

797 798
  // 3.go through recover steps to fill history
  if (pTask->fillHistory) {
799
    streamTaskCheckDownstream(pTask, sversion);
800 801
  }

802 803
  tqDebug("vgId:%d s-task:%s is deployed and add meta from mnd, status:%d, total:%d", TD_VID(pTq->pVnode),
          pTask->id.idStr, pTask->status.taskStatus, streamMetaGetNumOfTasks(pTq->pStreamMeta));
804 805 806
  return 0;
}

L
Liu Jicong 已提交
807 808 809 810 811
int32_t tqProcessTaskRecover1Req(STQ* pTq, SRpcMsg* pMsg) {
  int32_t code;
  char*   msg = pMsg->pCont;
  int32_t msgLen = pMsg->contLen;

812
  SStreamRecoverStep1Req* pReq = (SStreamRecoverStep1Req*)msg;
L
Liu Jicong 已提交
813
  SStreamTask*            pTask = streamMetaAcquireTask(pTq->pStreamMeta, pReq->taskId);
814 815 816 817 818
  if (pTask == NULL) {
    return -1;
  }

  // check param
819
  int64_t fillVer1 = pTask->chkInfo.version;
820
  if (fillVer1 <= 0) {
L
Liu Jicong 已提交
821
    streamMetaReleaseTask(pTq->pStreamMeta, pTask);
822 823 824 825
    return -1;
  }

  // do recovery step 1
H
Haojun Liao 已提交
826 827
  tqDebug("s-task:%s start recover step 1 scan", pTask->id.idStr);
  int64_t st = taosGetTimestampMs();
828

H
Haojun Liao 已提交
829
  streamSourceRecoverScanStep1(pTask);
830
  if (atomic_load_8(&pTask->status.taskStatus) == TASK_STATUS__DROPPING) {
L
Liu Jicong 已提交
831 832 833 834
    streamMetaReleaseTask(pTq->pStreamMeta, pTask);
    return 0;
  }

H
Haojun Liao 已提交
835 836 837
  double el = (taosGetTimestampMs() - st) / 1000.0;
  tqDebug("s-task:%s recover step 1 ended, elapsed time:%.2fs", pTask->id.idStr, el);

838 839 840 841
  // build msg to launch next step
  SStreamRecoverStep2Req req;
  code = streamBuildSourceRecover2Req(pTask, &req);
  if (code < 0) {
L
Liu Jicong 已提交
842
    streamMetaReleaseTask(pTq->pStreamMeta, pTask);
843 844 845
    return -1;
  }

L
Liu Jicong 已提交
846
  streamMetaReleaseTask(pTq->pStreamMeta, pTask);
L
Liu Jicong 已提交
847

848
  if (atomic_load_8(&pTask->status.taskStatus) == TASK_STATUS__DROPPING) {
L
Liu Jicong 已提交
849 850 851
    return 0;
  }

852
  // serialize msg
L
Liu Jicong 已提交
853 854 855 856 857 858 859 860
  int32_t len = sizeof(SStreamRecoverStep1Req);

  void* serializedReq = rpcMallocCont(len);
  if (serializedReq == NULL) {
    return -1;
  }

  memcpy(serializedReq, &req, len);
861 862

  // dispatch msg
H
Haojun Liao 已提交
863
  tqDebug("s-task:%s start recover block stage", pTask->id.idStr);
864

H
Haojun Liao 已提交
865 866
  SRpcMsg rpcMsg = {
      .code = 0, .contLen = len, .msgType = TDMT_VND_STREAM_RECOVER_BLOCKING_STAGE, .pCont = serializedReq};
867 868 869 870
  tmsgPutToQueue(&pTq->pVnode->msgCb, WRITE_QUEUE, &rpcMsg);
  return 0;
}

871
int32_t tqProcessTaskRecover2Req(STQ* pTq, int64_t sversion, char* msg, int32_t msgLen) {
H
Haojun Liao 已提交
872 873
  int32_t code = 0;

874
  SStreamRecoverStep2Req* pReq = (SStreamRecoverStep2Req*)msg;
L
Liu Jicong 已提交
875
  SStreamTask*            pTask = streamMetaAcquireTask(pTq->pStreamMeta, pReq->taskId);
876 877 878 879 880
  if (pTask == NULL) {
    return -1;
  }

  // do recovery step 2
881
  code = streamSourceRecoverScanStep2(pTask, sversion);
882
  if (code < 0) {
L
Liu Jicong 已提交
883
    streamMetaReleaseTask(pTq->pStreamMeta, pTask);
884 885 886
    return -1;
  }

887
  if (atomic_load_8(&pTask->status.taskStatus) == TASK_STATUS__DROPPING) {
L
Liu Jicong 已提交
888 889 890 891
    streamMetaReleaseTask(pTq->pStreamMeta, pTask);
    return 0;
  }

892 893 894
  // restore param
  code = streamRestoreParam(pTask);
  if (code < 0) {
L
Liu Jicong 已提交
895
    streamMetaReleaseTask(pTq->pStreamMeta, pTask);
896 897 898 899 900 901
    return -1;
  }

  // set status normal
  code = streamSetStatusNormal(pTask);
  if (code < 0) {
L
Liu Jicong 已提交
902
    streamMetaReleaseTask(pTq->pStreamMeta, pTask);
903 904 905 906 907 908
    return -1;
  }

  // dispatch recover finish req to all related downstream task
  code = streamDispatchRecoverFinishReq(pTask);
  if (code < 0) {
L
Liu Jicong 已提交
909
    streamMetaReleaseTask(pTq->pStreamMeta, pTask);
910 911 912
    return -1;
  }

L
Liu Jicong 已提交
913 914 915
  atomic_store_8(&pTask->fillHistory, 0);
  streamMetaSaveTask(pTq->pStreamMeta, pTask);

L
Liu Jicong 已提交
916 917
  streamMetaReleaseTask(pTq->pStreamMeta, pTask);

918 919 920
  return 0;
}

L
Liu Jicong 已提交
921 922 923
int32_t tqProcessTaskRecoverFinishReq(STQ* pTq, SRpcMsg* pMsg) {
  char*   msg = POINTER_SHIFT(pMsg->pCont, sizeof(SMsgHead));
  int32_t msgLen = pMsg->contLen - sizeof(SMsgHead);
924 925

  // deserialize
926 927 928
  SStreamRecoverFinishReq req;

  SDecoder decoder;
X
Xiaoyu Wang 已提交
929
  tDecoderInit(&decoder, (uint8_t*)msg, msgLen);
930 931 932
  tDecodeSStreamRecoverFinishReq(&decoder, &req);
  tDecoderClear(&decoder);

933
  // find task
L
Liu Jicong 已提交
934
  SStreamTask* pTask = streamMetaAcquireTask(pTq->pStreamMeta, req.taskId);
935 936 937
  if (pTask == NULL) {
    return -1;
  }
938
  // do process request
939
  if (streamProcessRecoverFinishReq(pTask, req.childId) < 0) {
L
Liu Jicong 已提交
940
    streamMetaReleaseTask(pTq->pStreamMeta, pTask);
941 942 943
    return -1;
  }

L
Liu Jicong 已提交
944
  streamMetaReleaseTask(pTq->pStreamMeta, pTask);
945
  return 0;
L
Liu Jicong 已提交
946
}
L
Liu Jicong 已提交
947

L
Liu Jicong 已提交
948 949 950 951 952
int32_t tqProcessTaskRecoverFinishRsp(STQ* pTq, SRpcMsg* pMsg) {
  //
  return 0;
}

L
Liu Jicong 已提交
953 954 955 956 957 958 959 960 961 962 963 964 965 966 967 968
int32_t tqProcessDelReq(STQ* pTq, void* pReq, int32_t len, int64_t ver) {
  bool        failed = false;
  SDecoder*   pCoder = &(SDecoder){0};
  SDeleteRes* pRes = &(SDeleteRes){0};

  pRes->uidList = taosArrayInit(0, sizeof(tb_uid_t));
  if (pRes->uidList == NULL) {
    terrno = TSDB_CODE_OUT_OF_MEMORY;
    failed = true;
  }

  tDecoderInit(pCoder, pReq, len);
  tDecodeDeleteRes(pCoder, pRes);
  tDecoderClear(pCoder);

  int32_t sz = taosArrayGetSize(pRes->uidList);
L
Liu Jicong 已提交
969
  if (sz == 0 || pRes->affectedRows == 0) {
L
Liu Jicong 已提交
970 971 972 973 974 975 976 977 978 979 980
    taosArrayDestroy(pRes->uidList);
    return 0;
  }
  SSDataBlock* pDelBlock = createSpecialDataBlock(STREAM_DELETE_DATA);
  blockDataEnsureCapacity(pDelBlock, sz);
  pDelBlock->info.rows = sz;
  pDelBlock->info.version = ver;

  for (int32_t i = 0; i < sz; i++) {
    // start key column
    SColumnInfoData* pStartCol = taosArrayGet(pDelBlock->pDataBlock, START_TS_COLUMN_INDEX);
981
    colDataSetVal(pStartCol, i, (const char*)&pRes->skey, false);  // end key column
L
Liu Jicong 已提交
982
    SColumnInfoData* pEndCol = taosArrayGet(pDelBlock->pDataBlock, END_TS_COLUMN_INDEX);
983
    colDataSetVal(pEndCol, i, (const char*)&pRes->ekey, false);
L
Liu Jicong 已提交
984 985 986
    // uid column
    SColumnInfoData* pUidCol = taosArrayGet(pDelBlock->pDataBlock, UID_COLUMN_INDEX);
    int64_t*         pUid = taosArrayGet(pRes->uidList, i);
987
    colDataSetVal(pUidCol, i, (const char*)pUid, false);
L
Liu Jicong 已提交
988

989 990 991
    colDataSetNULL(taosArrayGet(pDelBlock->pDataBlock, GROUPID_COLUMN_INDEX), i);
    colDataSetNULL(taosArrayGet(pDelBlock->pDataBlock, CALCULATE_START_TS_COLUMN_INDEX), i);
    colDataSetNULL(taosArrayGet(pDelBlock->pDataBlock, CALCULATE_END_TS_COLUMN_INDEX), i);
L
Liu Jicong 已提交
992 993
  }

L
Liu Jicong 已提交
994 995
  taosArrayDestroy(pRes->uidList);

L
Liu Jicong 已提交
996 997 998
  int32_t* pRef = taosMemoryMalloc(sizeof(int32_t));
  *pRef = 1;

999 1000
  taosWLockLatch(&pTq->pStreamMeta->lock);

L
Liu Jicong 已提交
1001 1002 1003
  void* pIter = NULL;
  while (1) {
    pIter = taosHashIterate(pTq->pStreamMeta->pTasks, pIter);
1004 1005 1006 1007
    if (pIter == NULL) {
      break;
    }

L
Liu Jicong 已提交
1008
    SStreamTask* pTask = *(SStreamTask**)pIter;
1009 1010 1011
    if (pTask->taskLevel != TASK_LEVEL__SOURCE) {
      continue;
    }
L
Liu Jicong 已提交
1012

1013
    qDebug("s-task:%s delete req enqueue, ver: %" PRId64, pTask->id.idStr, ver);
L
Liu Jicong 已提交
1014

L
Liu Jicong 已提交
1015
    if (!failed) {
S
Shengliang Guan 已提交
1016
      SStreamRefDataBlock* pRefBlock = taosAllocateQitem(sizeof(SStreamRefDataBlock), DEF_QITEM, 0);
L
Liu Jicong 已提交
1017 1018 1019 1020 1021
      pRefBlock->type = STREAM_INPUT__REF_DATA_BLOCK;
      pRefBlock->pBlock = pDelBlock;
      pRefBlock->dataRef = pRef;
      atomic_add_fetch_32(pRefBlock->dataRef, 1);

1022
      if (tAppendDataToInputQueue(pTask, (SStreamQueueItem*)pRefBlock) < 0) {
L
Liu Jicong 已提交
1023
        atomic_sub_fetch_32(pRef, 1);
L
Liu Jicong 已提交
1024
        taosFreeQitem(pRefBlock);
L
Liu Jicong 已提交
1025 1026
        continue;
      }
L
Liu Jicong 已提交
1027

L
Liu Jicong 已提交
1028
      if (streamSchedExec(pTask) < 0) {
1029
        qError("s-task:%s stream task launch failed", pTask->id.idStr);
L
Liu Jicong 已提交
1030 1031
        continue;
      }
L
Liu Jicong 已提交
1032

L
Liu Jicong 已提交
1033 1034 1035 1036
    } else {
      streamTaskInputFail(pTask);
    }
  }
L
Liu Jicong 已提交
1037

1038 1039
  taosWUnLockLatch(&pTq->pStreamMeta->lock);

L
Liu Jicong 已提交
1040 1041
  int32_t ref = atomic_sub_fetch_32(pRef, 1);
  if (ref == 0) {
L
Liu Jicong 已提交
1042
    blockDataDestroy(pDelBlock);
L
Liu Jicong 已提交
1043 1044 1045 1046
    taosMemoryFree(pRef);
  }

#if 0
S
Shengliang Guan 已提交
1047
    SStreamDataBlock* pStreamBlock = taosAllocateQitem(sizeof(SStreamDataBlock), DEF_QITEM, 0);
L
Liu Jicong 已提交
1048 1049 1050 1051 1052 1053 1054 1055
    pStreamBlock->type = STREAM_INPUT__DATA_BLOCK;
    pStreamBlock->blocks = taosArrayInit(0, sizeof(SSDataBlock));
    SSDataBlock block = {0};
    assignOneDataBlock(&block, pDelBlock);
    block.info.type = STREAM_DELETE_DATA;
    taosArrayPush(pStreamBlock->blocks, &block);

    if (!failed) {
1056
      if (tAppendDataToInputQueue(pTask, (SStreamQueueItem*)pStreamBlock) < 0) {
1057
        qError("stream task input del failed, task id %d", pTask->id.taskId);
L
Liu Jicong 已提交
1058 1059 1060 1061
        continue;
      }

      if (streamSchedExec(pTask) < 0) {
1062
        qError("stream task launch failed, task id %d", pTask->id.taskId);
L
Liu Jicong 已提交
1063 1064 1065 1066 1067 1068
        continue;
      }
    } else {
      streamTaskInputFail(pTask);
    }
  }
L
Liu Jicong 已提交
1069
  blockDataDestroy(pDelBlock);
L
Liu Jicong 已提交
1070
#endif
L
Liu Jicong 已提交
1071 1072 1073
  return 0;
}

1074 1075 1076 1077
int32_t tqProcessSubmitReqForSubscribe(STQ* pTq) {
  int32_t vgId = TD_VID(pTq->pVnode);

  taosWLockLatch(&pTq->lock);
1078 1079 1080 1081 1082

  if (taosHashGetSize(pTq->pPushMgr) > 0) {
    void* pIter = taosHashIterate(pTq->pPushMgr, NULL);

    while (pIter) {
wmmhello's avatar
wmmhello 已提交
1083
      STqHandle* pHandle = *(STqHandle**)pIter;
1084 1085 1086
      tqDebug("vgId:%d start set submit for pHandle:%p, consumer:0x%" PRIx64, vgId, pHandle, pHandle->consumerId);

      if (ASSERT(pHandle->msg != NULL)) {
wmmhello's avatar
wmmhello 已提交
1087 1088 1089 1090 1091 1092 1093 1094
        tqError("pHandle->msg should not be null");
        break;
      }else{
        SRpcMsg msg = {.msgType = TDMT_VND_TMQ_CONSUME, .pCont = pHandle->msg->pCont, .contLen = pHandle->msg->contLen, .info = pHandle->msg->info};
        tmsgPutToQueue(&pTq->pVnode->msgCb, QUERY_QUEUE, &msg);
        taosMemoryFree(pHandle->msg);
        pHandle->msg = NULL;
      }
1095

wmmhello's avatar
wmmhello 已提交
1096
      pIter = taosHashIterate(pTq->pPushMgr, pIter);
1097
    }
1098

wmmhello's avatar
wmmhello 已提交
1099
    taosHashClear(pTq->pPushMgr);
1100
  }
1101

1102 1103
  // unlock
  taosWUnLockLatch(&pTq->lock);
1104
  return 0;
L
Liu Jicong 已提交
1105 1106
}

L
Liu Jicong 已提交
1107 1108
int32_t tqProcessTaskRunReq(STQ* pTq, SRpcMsg* pMsg) {
  SStreamTaskRunReq* pReq = pMsg->pCont;
1109 1110 1111 1112

  int32_t taskId = pReq->taskId;
  int32_t vgId = TD_VID(pTq->pVnode);

1113 1114
  if (taskId == WAL_READ_TASKS_ID) {  // all tasks are extracted submit data from the wal
    tqStreamTasksScanWal(pTq);
L
Liu Jicong 已提交
1115
    return 0;
1116
  }
1117

1118
  SStreamTask* pTask = streamMetaAcquireTask(pTq->pStreamMeta, taskId);
1119 1120 1121 1122 1123
  if (pTask != NULL) {
    if (pTask->status.taskStatus == TASK_STATUS__NORMAL) {
      tqDebug("vgId:%d s-task:%s start to process block from wal, last chk point:%" PRId64, vgId,
              pTask->id.idStr, pTask->chkInfo.version);
      streamProcessRunReq(pTask);
1124
    } else {
1125
      tqDebug("vgId:%d s-task:%s ignore run req since not in ready state", vgId, pTask->id.idStr);
1126
    }
1127 1128 1129 1130 1131 1132 1133

    streamMetaReleaseTask(pTq->pStreamMeta, pTask);
    tqStartStreamTasks(pTq);
    return 0;
  } else {
    tqError("vgId:%d failed to found s-task, taskId:%d", vgId, taskId);
    return -1;
L
Liu Jicong 已提交
1134
  }
L
Liu Jicong 已提交
1135 1136
}

L
Liu Jicong 已提交
1137
int32_t tqProcessTaskDispatchReq(STQ* pTq, SRpcMsg* pMsg, bool exec) {
1138 1139 1140 1141 1142
  char*              msgStr = pMsg->pCont;
  char*              msgBody = POINTER_SHIFT(msgStr, sizeof(SMsgHead));
  int32_t            msgLen = pMsg->contLen - sizeof(SMsgHead);
  SStreamDispatchReq req;
  SDecoder           decoder;
L
Liu Jicong 已提交
1143
  tDecoderInit(&decoder, (uint8_t*)msgBody, msgLen);
1144
  tDecodeStreamDispatchReq(&decoder, &req);
L
Liu Jicong 已提交
1145

1146
  SStreamTask* pTask = streamMetaAcquireTask(pTq->pStreamMeta, req.taskId);
L
Liu Jicong 已提交
1147
  if (pTask) {
1148
    SRpcMsg rsp = { .info = pMsg->info, .code = 0 };
L
Liu Jicong 已提交
1149
    streamProcessDispatchReq(pTask, &req, &rsp, exec);
L
Liu Jicong 已提交
1150
    streamMetaReleaseTask(pTq->pStreamMeta, pTask);
L
Liu Jicong 已提交
1151
    return 0;
1152
  } else {
L
liuyao 已提交
1153
    tDeleteStreamDispatchReq(&req);
1154
    return -1;
L
Liu Jicong 已提交
1155
  }
L
Liu Jicong 已提交
1156 1157
}

L
Liu Jicong 已提交
1158 1159
int32_t tqProcessTaskDispatchRsp(STQ* pTq, SRpcMsg* pMsg) {
  SStreamDispatchRsp* pRsp = POINTER_SHIFT(pMsg->pCont, sizeof(SMsgHead));
1160
  int32_t             taskId = ntohl(pRsp->upstreamTaskId);
L
Liu Jicong 已提交
1161
  SStreamTask*        pTask = streamMetaAcquireTask(pTq->pStreamMeta, taskId);
1162
  tqDebug("recv dispatch rsp, code:%x", pMsg->code);
L
Liu Jicong 已提交
1163
  if (pTask) {
1164
    streamProcessDispatchRsp(pTask, pRsp, pMsg->code);
L
Liu Jicong 已提交
1165
    streamMetaReleaseTask(pTq->pStreamMeta, pTask);
L
Liu Jicong 已提交
1166
    return 0;
1167 1168
  } else {
    return -1;
L
Liu Jicong 已提交
1169
  }
L
Liu Jicong 已提交
1170
}
L
Liu Jicong 已提交
1171

1172
int32_t tqProcessTaskDropReq(STQ* pTq, int64_t sversion, char* msg, int32_t msgLen) {
L
Liu Jicong 已提交
1173
  SVDropStreamTaskReq* pReq = (SVDropStreamTaskReq*)msg;
1174
  streamMetaRemoveTask(pTq->pStreamMeta, pReq->taskId);
L
Liu Jicong 已提交
1175
  return 0;
L
Liu Jicong 已提交
1176
}
L
Liu Jicong 已提交
1177 1178 1179 1180 1181 1182 1183

int32_t tqProcessTaskRetrieveReq(STQ* pTq, SRpcMsg* pMsg) {
  char*              msgStr = pMsg->pCont;
  char*              msgBody = POINTER_SHIFT(msgStr, sizeof(SMsgHead));
  int32_t            msgLen = pMsg->contLen - sizeof(SMsgHead);
  SStreamRetrieveReq req;
  SDecoder           decoder;
1184
  tDecoderInit(&decoder, (uint8_t*)msgBody, msgLen);
L
Liu Jicong 已提交
1185
  tDecodeStreamRetrieveReq(&decoder, &req);
L
Liu Jicong 已提交
1186
  tDecoderClear(&decoder);
L
Liu Jicong 已提交
1187
  int32_t      taskId = req.dstTaskId;
L
Liu Jicong 已提交
1188
  SStreamTask* pTask = streamMetaAcquireTask(pTq->pStreamMeta, taskId);
L
Liu Jicong 已提交
1189
  if (pTask) {
1190
    SRpcMsg rsp = { .info = pMsg->info, .code = 0 };
L
Liu Jicong 已提交
1191
    streamProcessRetrieveReq(pTask, &req, &rsp);
L
Liu Jicong 已提交
1192
    streamMetaReleaseTask(pTq->pStreamMeta, pTask);
L
Liu Jicong 已提交
1193
    tDeleteStreamRetrieveReq(&req);
L
Liu Jicong 已提交
1194
    return 0;
L
Liu Jicong 已提交
1195
  } else {
L
liuyao 已提交
1196
    tDeleteStreamRetrieveReq(&req);
L
Liu Jicong 已提交
1197
    return -1;
L
Liu Jicong 已提交
1198 1199 1200 1201 1202 1203 1204
  }
}

int32_t tqProcessTaskRetrieveRsp(STQ* pTq, SRpcMsg* pMsg) {
  //
  return 0;
}
L
Liu Jicong 已提交
1205

1206 1207 1208 1209 1210 1211
int32_t vnodeEnqueueStreamMsg(SVnode* pVnode, SRpcMsg* pMsg) {
  STQ*      pTq = pVnode->pTq;
  SMsgHead* msgStr = pMsg->pCont;
  char*     msgBody = POINTER_SHIFT(msgStr, sizeof(SMsgHead));
  int32_t   msgLen = pMsg->contLen - sizeof(SMsgHead);
  int32_t   code = 0;
L
Liu Jicong 已提交
1212 1213 1214

  SStreamDispatchReq req;
  SDecoder           decoder;
1215
  tDecoderInit(&decoder, (uint8_t*)msgBody, msgLen);
L
Liu Jicong 已提交
1216 1217
  if (tDecodeStreamDispatchReq(&decoder, &req) < 0) {
    code = TSDB_CODE_MSG_DECODE_ERROR;
L
Liu Jicong 已提交
1218
    tDecoderClear(&decoder);
L
Liu Jicong 已提交
1219 1220
    goto FAIL;
  }
L
Liu Jicong 已提交
1221
  tDecoderClear(&decoder);
L
Liu Jicong 已提交
1222

L
Liu Jicong 已提交
1223
  int32_t taskId = req.taskId;
L
Liu Jicong 已提交
1224

L
Liu Jicong 已提交
1225
  SStreamTask* pTask = streamMetaAcquireTask(pTq->pStreamMeta, taskId);
L
Liu Jicong 已提交
1226
  if (pTask) {
1227
    SRpcMsg rsp = { .info = pMsg->info, .code = 0 };
L
Liu Jicong 已提交
1228
    streamProcessDispatchReq(pTask, &req, &rsp, false);
L
Liu Jicong 已提交
1229
    streamMetaReleaseTask(pTq->pStreamMeta, pTask);
L
Liu Jicong 已提交
1230 1231
    rpcFreeCont(pMsg->pCont);
    taosFreeQitem(pMsg);
1232
    return 0;
5
54liuyao 已提交
1233 1234
  } else {
    tDeleteStreamDispatchReq(&req);
L
Liu Jicong 已提交
1235
  }
L
Liu Jicong 已提交
1236

1237 1238
  code = TSDB_CODE_STREAM_TASK_NOT_EXIST;

L
Liu Jicong 已提交
1239
FAIL:
1240 1241 1242 1243
  if (pMsg->info.handle == NULL) return -1;

  SMsgHead* pRspHead = rpcMallocCont(sizeof(SMsgHead) + sizeof(SStreamDispatchRsp));
  if (pRspHead == NULL) {
1244
    SRpcMsg rsp = { .code = TSDB_CODE_OUT_OF_MEMORY, .info = pMsg->info };
1245 1246 1247 1248 1249 1250 1251 1252 1253 1254 1255 1256 1257 1258 1259 1260
    tqDebug("send dispatch error rsp, code: %x", code);
    tmsgSendRsp(&rsp);
    rpcFreeCont(pMsg->pCont);
    taosFreeQitem(pMsg);
    return -1;
  }

  pRspHead->vgId = htonl(req.upstreamNodeId);
  SStreamDispatchRsp* pRsp = POINTER_SHIFT(pRspHead, sizeof(SMsgHead));
  pRsp->streamId = htobe64(req.streamId);
  pRsp->upstreamTaskId = htonl(req.upstreamTaskId);
  pRsp->upstreamNodeId = htonl(req.upstreamNodeId);
  pRsp->downstreamNodeId = htonl(pVnode->config.vgId);
  pRsp->downstreamTaskId = htonl(req.taskId);
  pRsp->inputStatus = TASK_OUTPUT_STATUS__NORMAL;

L
Liu Jicong 已提交
1261
  SRpcMsg rsp = {
1262
      .code = code, .info = pMsg->info, .contLen = sizeof(SMsgHead) + sizeof(SStreamDispatchRsp), .pCont = pRspHead};
1263
  tqDebug("send dispatch error rsp, code: %x", code);
L
Liu Jicong 已提交
1264
  tmsgSendRsp(&rsp);
L
Liu Jicong 已提交
1265 1266
  rpcFreeCont(pMsg->pCont);
  taosFreeQitem(pMsg);
1267
  return -1;
L
Liu Jicong 已提交
1268
}
L
Liu Jicong 已提交
1269

1270
int32_t tqCheckLogInWal(STQ* pTq, int64_t sversion) { return sversion <= pTq->walLogLastVer; }
1271

1272
int32_t tqStartStreamTasks(STQ* pTq) {
1273
  int32_t      vgId = TD_VID(pTq->pVnode);
1274
  SStreamMeta* pMeta = pTq->pStreamMeta;
1275

1276
  taosWLockLatch(&pMeta->lock);
1277

1278
  int32_t numOfTasks = taosArrayGetSize(pMeta->pTaskList);
1279 1280 1281 1282 1283 1284
  if (numOfTasks == 0) {
    tqInfo("vgId:%d no stream tasks exists", vgId);
    taosWUnLockLatch(&pTq->pStreamMeta->lock);
    return 0;
  }

1285
  pMeta->walScanCounter += 1;
1286

1287 1288
  if (pMeta->walScanCounter > 1) {
    tqDebug("vgId:%d wal read task has been launched, remain scan times:%d", vgId, pMeta->walScanCounter);
1289 1290 1291 1292
    taosWUnLockLatch(&pTq->pStreamMeta->lock);
    return 0;
  }

1293 1294 1295 1296
  SStreamTaskRunReq* pRunReq = rpcMallocCont(sizeof(SStreamTaskRunReq));
  if (pRunReq == NULL) {
    terrno = TSDB_CODE_OUT_OF_MEMORY;
    tqError("vgId:%d failed restore stream tasks, code:%s", vgId, terrstr(terrno));
1297
    taosWUnLockLatch(&pTq->pStreamMeta->lock);
1298 1299 1300
    return -1;
  }

H
Haojun Liao 已提交
1301
  tqDebug("vgId:%d start wal scan stream tasks, tasks:%d", vgId, numOfTasks);
1302 1303
  pRunReq->head.vgId = vgId;
  pRunReq->streamId = 0;
1304
  pRunReq->taskId = WAL_READ_TASKS_ID;
1305 1306 1307

  SRpcMsg msg = {.msgType = TDMT_STREAM_TASK_RUN, .pCont = pRunReq, .contLen = sizeof(SStreamTaskRunReq)};
  tmsgPutToQueue(&pTq->pVnode->msgCb, STREAM_QUEUE, &msg);
1308
  taosWUnLockLatch(&pTq->pStreamMeta->lock);
1309 1310 1311

  return 0;
}