taosdata/TDengine | Build 3599 | source/dnode/mnode/impl/src/mndVgroup.c | Coveralls

50

int32_t mndInitVgroup(SMnode *pMnode) {

×

51

  SSdbTable table = {

×

62

  mndSetMsgHandle(pMnode, TDMT_DND_CREATE_VNODE_RSP, mndTransProcessRsp);

×

63

  mndSetMsgHandle(pMnode, TDMT_VND_ALTER_REPLICA_RSP, mndTransProcessRsp);

×

64

  mndSetMsgHandle(pMnode, TDMT_VND_ALTER_CONFIG_RSP, mndTransProcessRsp);

×

65

  mndSetMsgHandle(pMnode, TDMT_VND_ALTER_CONFIRM_RSP, mndTransProcessRsp);

×

66

  mndSetMsgHandle(pMnode, TDMT_VND_ALTER_HASHRANGE_RSP, mndTransProcessRsp);

×

67

  mndSetMsgHandle(pMnode, TDMT_DND_DROP_VNODE_RSP, mndTransProcessRsp);

×

68

  mndSetMsgHandle(pMnode, TDMT_VND_COMPACT_RSP, mndTransProcessRsp);

×

69

  mndSetMsgHandle(pMnode, TDMT_VND_DISABLE_WRITE_RSP, mndTransProcessRsp);

×

70

  mndSetMsgHandle(pMnode, TDMT_SYNC_FORCE_FOLLOWER_RSP, mndTransProcessRsp);

×

71

  mndSetMsgHandle(pMnode, TDMT_DND_ALTER_VNODE_TYPE_RSP, mndTransProcessRsp);

×

72

  mndSetMsgHandle(pMnode, TDMT_DND_CHECK_VNODE_LEARNER_CATCHUP_RSP, mndTransProcessRsp);

×

73

  mndSetMsgHandle(pMnode, TDMT_SYNC_CONFIG_CHANGE_RSP, mndTransProcessRsp);

×

75

  mndSetMsgHandle(pMnode, TDMT_MND_REDISTRIBUTE_VGROUP, mndProcessRedistributeVgroupMsg);

×

76

  mndSetMsgHandle(pMnode, TDMT_MND_SPLIT_VGROUP, mndProcessSplitVgroupMsg);

×

78

  mndSetMsgHandle(pMnode, TDMT_MND_BALANCE_VGROUP, mndProcessBalanceVgroupMsg);

×

79

  mndSetMsgHandle(pMnode, TDMT_MND_BALANCE_VGROUP_LEADER, mndProcessVgroupBalanceLeaderMsg);

×

81

  mndAddShowRetrieveHandle(pMnode, TSDB_MGMT_TABLE_VGROUP, mndRetrieveVgroups);

×

82

  mndAddShowFreeIterHandle(pMnode, TSDB_MGMT_TABLE_VGROUP, mndCancelGetNextVgroup);

×

83

  mndAddShowRetrieveHandle(pMnode, TSDB_MGMT_TABLE_VNODES, mndRetrieveVnodes);

×

84

  mndAddShowFreeIterHandle(pMnode, TSDB_MGMT_TABLE_VNODES, mndCancelGetNextVnode);

×

86

  return sdbSetTable(pMnode->pSdb, table);

×

89

void mndCleanupVgroup(SMnode *pMnode) {}

×

91

SSdbRaw *mndVgroupActionEncode(SVgObj *pVgroup) {

×

92

  int32_t code = 0;

×

93

  int32_t lino = 0;

×

94

  terrno = TSDB_CODE_OUT_OF_MEMORY;

×

96

  SSdbRaw *pRaw = sdbAllocRaw(SDB_VGROUP, VGROUP_VER_NUMBER, sizeof(SVgObj) + VGROUP_RESERVE_SIZE);

×

97

  if (pRaw == NULL) goto _OVER;

×

99

  int32_t dataPos = 0;

×

100

  SDB_SET_INT32(pRaw, dataPos, pVgroup->vgId, _OVER)

×

101

  SDB_SET_INT64(pRaw, dataPos, pVgroup->createdTime, _OVER)

×

102

  SDB_SET_INT64(pRaw, dataPos, pVgroup->updateTime, _OVER)

×

103

  SDB_SET_INT32(pRaw, dataPos, pVgroup->version, _OVER)

×

104

  SDB_SET_INT32(pRaw, dataPos, pVgroup->hashBegin, _OVER)

×

105

  SDB_SET_INT32(pRaw, dataPos, pVgroup->hashEnd, _OVER)

×

106

  SDB_SET_BINARY(pRaw, dataPos, pVgroup->dbName, TSDB_DB_FNAME_LEN, _OVER)

×

107

  SDB_SET_INT64(pRaw, dataPos, pVgroup->dbUid, _OVER)

×

108

  SDB_SET_INT8(pRaw, dataPos, pVgroup->isTsma, _OVER)

×

109

  SDB_SET_INT8(pRaw, dataPos, pVgroup->replica, _OVER)

×

110

  for (int8_t i = 0; i < pVgroup->replica; ++i) {

×

111

    SVnodeGid *pVgid = &pVgroup->vnodeGid[i];

×

112

    SDB_SET_INT32(pRaw, dataPos, pVgid->dnodeId, _OVER)

×

114

  SDB_SET_INT32(pRaw, dataPos, pVgroup->syncConfChangeVer, _OVER)

×

115

  SDB_SET_RESERVE(pRaw, dataPos, VGROUP_RESERVE_SIZE, _OVER)

×

116

  SDB_SET_DATALEN(pRaw, dataPos, _OVER)

×

118

  terrno = 0;

×

120

_OVER:

×

121

  if (terrno != 0) {

×

122

    mError("vgId:%d, failed to encode to raw:%p since %s", pVgroup->vgId, pRaw, terrstr());

×

123

    sdbFreeRaw(pRaw);

×

124

    return NULL;

×

127

  mTrace("vgId:%d, encode to raw:%p, row:%p", pVgroup->vgId, pRaw, pVgroup);

×

128

  return pRaw;

×

131

SSdbRow *mndVgroupActionDecode(SSdbRaw *pRaw) {

×

132

  int32_t code = 0;

×

133

  int32_t lino = 0;

×

134

  terrno = TSDB_CODE_OUT_OF_MEMORY;

×

135

  SSdbRow *pRow = NULL;

×

136

  SVgObj  *pVgroup = NULL;

×

138

  int8_t sver = 0;

×

139

  if (sdbGetRawSoftVer(pRaw, &sver) != 0) goto _OVER;

×

141

  if (sver < 1 || sver > VGROUP_VER_NUMBER) {

×

142

    terrno = TSDB_CODE_SDB_INVALID_DATA_VER;

×

143

    goto _OVER;

×

146

  pRow = sdbAllocRow(sizeof(SVgObj));

×

147

  if (pRow == NULL) goto _OVER;

×

149

  pVgroup = sdbGetRowObj(pRow);

×

150

  if (pVgroup == NULL) goto _OVER;

×

152

  int32_t dataPos = 0;

×

153

  SDB_GET_INT32(pRaw, dataPos, &pVgroup->vgId, _OVER)

×

154

  SDB_GET_INT64(pRaw, dataPos, &pVgroup->createdTime, _OVER)

×

155

  SDB_GET_INT64(pRaw, dataPos, &pVgroup->updateTime, _OVER)

×

156

  SDB_GET_INT32(pRaw, dataPos, &pVgroup->version, _OVER)

×

157

  SDB_GET_INT32(pRaw, dataPos, &pVgroup->hashBegin, _OVER)

×

158

  SDB_GET_INT32(pRaw, dataPos, &pVgroup->hashEnd, _OVER)

×

159

  SDB_GET_BINARY(pRaw, dataPos, pVgroup->dbName, TSDB_DB_FNAME_LEN, _OVER)

×

160

  SDB_GET_INT64(pRaw, dataPos, &pVgroup->dbUid, _OVER)

×

161

  SDB_GET_INT8(pRaw, dataPos, &pVgroup->isTsma, _OVER)

×

162

  SDB_GET_INT8(pRaw, dataPos, &pVgroup->replica, _OVER)

×

163

  for (int8_t i = 0; i < pVgroup->replica; ++i) {

×

164

    SVnodeGid *pVgid = &pVgroup->vnodeGid[i];

×

165

    SDB_GET_INT32(pRaw, dataPos, &pVgid->dnodeId, _OVER)

×

166

    if (pVgroup->replica == 1) {

×

167

      pVgid->syncState = TAOS_SYNC_STATE_LEADER;

×

170

  if (dataPos + sizeof(int32_t) + VGROUP_RESERVE_SIZE <= pRaw->dataLen) {

×

171

    SDB_GET_INT32(pRaw, dataPos, &pVgroup->syncConfChangeVer, _OVER)

×

174

  SDB_GET_RESERVE(pRaw, dataPos, VGROUP_RESERVE_SIZE, _OVER)

×

176

  terrno = 0;

×

178

_OVER:

×

179

  if (terrno != 0) {

×

180

    mError("vgId:%d, failed to decode from raw:%p since %s", pVgroup == NULL ? 0 : pVgroup->vgId, pRaw, terrstr());

×

181

    taosMemoryFreeClear(pRow);

×

182

    return NULL;

×

185

  mTrace("vgId:%d, decode from raw:%p, row:%p", pVgroup->vgId, pRaw, pVgroup);

×

186

  return pRow;

×

189

static int32_t mndNewVgActionValidate(SMnode *pMnode, STrans *pTrans, SSdbRaw *pRaw) {

×

190

  SSdb    *pSdb = pMnode->pSdb;

×

191

  SSdbRow *pRow = NULL;

×

192

  SVgObj  *pVgroup = NULL;

×

193

  int      code = -1;

×

195

  pRow = mndVgroupActionDecode(pRaw);

×

196

  if (pRow == NULL) {

×

197

    code = TSDB_CODE_MND_RETURN_VALUE_NULL;

×

198

    if (terrno != 0) code = terrno;

×

199

    goto _OVER;

×

201

  pVgroup = sdbGetRowObj(pRow);

×

202

  if (pVgroup == NULL) {

×

203

    code = TSDB_CODE_MND_RETURN_VALUE_NULL;

×

204

    if (terrno != 0) code = terrno;

×

205

    goto _OVER;

×

208

  int32_t maxVgId = sdbGetMaxId(pMnode->pSdb, SDB_VGROUP);

×

209

  if (maxVgId > pVgroup->vgId) {

×

210

    mError("trans:%d, vgroup id %d already in use. maxVgId:%d", pTrans->id, pVgroup->vgId, maxVgId);

×

211

    goto _OVER;

×

214

  code = 0;

×

215

_OVER:

×

216

  if (pVgroup) mndVgroupActionDelete(pSdb, pVgroup);

×

217

  taosMemoryFreeClear(pRow);

×

218

  TAOS_RETURN(code);

×

221

static int32_t mndVgroupActionInsert(SSdb *pSdb, SVgObj *pVgroup) {

×

222

  mTrace("vgId:%d, perform insert action, row:%p", pVgroup->vgId, pVgroup);

×

223

  return 0;

×

226

static int32_t mndVgroupActionDelete(SSdb *pSdb, SVgObj *pVgroup) {

×

227

  mTrace("vgId:%d, perform delete action, row:%p", pVgroup->vgId, pVgroup);

×

228

  return 0;

×

231

static int32_t mndVgroupActionUpdate(SSdb *pSdb, SVgObj *pOld, SVgObj *pNew) {

×

232

  mTrace("vgId:%d, perform update action, old row:%p new row:%p", pOld->vgId, pOld, pNew);

×

233

  pOld->updateTime = pNew->updateTime;

×

234

  pOld->version = pNew->version;

×

235

  pOld->hashBegin = pNew->hashBegin;

×

236

  pOld->hashEnd = pNew->hashEnd;

×

237

  pOld->replica = pNew->replica;

×

238

  pOld->isTsma = pNew->isTsma;

×

239

  for (int32_t i = 0; i < pNew->replica; ++i) {

×

240

    SVnodeGid *pNewGid = &pNew->vnodeGid[i];

×

241

    for (int32_t j = 0; j < pOld->replica; ++j) {

×

242

      SVnodeGid *pOldGid = &pOld->vnodeGid[j];

×

243

      if (pNewGid->dnodeId == pOldGid->dnodeId) {

×

244

        pNewGid->syncState = pOldGid->syncState;

×

245

        pNewGid->syncRestore = pOldGid->syncRestore;

×

246

        pNewGid->syncCanRead = pOldGid->syncCanRead;

×

250

  pNew->numOfTables = pOld->numOfTables;

×

251

  pNew->numOfTimeSeries = pOld->numOfTimeSeries;

×

252

  pNew->totalStorage = pOld->totalStorage;

×

253

  pNew->compStorage = pOld->compStorage;

×

254

  pNew->pointsWritten = pOld->pointsWritten;

×

255

  pNew->compact = pOld->compact;

×

256

  memcpy(pOld->vnodeGid, pNew->vnodeGid, (TSDB_MAX_REPLICA + TSDB_MAX_LEARNER_REPLICA) * sizeof(SVnodeGid));

×

257

  pOld->syncConfChangeVer = pNew->syncConfChangeVer;

×

258

  return 0;

×

261

SVgObj *mndAcquireVgroup(SMnode *pMnode, int32_t vgId) {

×

262

  SSdb   *pSdb = pMnode->pSdb;

×

263

  SVgObj *pVgroup = sdbAcquire(pSdb, SDB_VGROUP, &vgId);

×

264

  if (pVgroup == NULL && terrno == TSDB_CODE_SDB_OBJ_NOT_THERE) {

×

265

    terrno = TSDB_CODE_MND_VGROUP_NOT_EXIST;

×

267

  return pVgroup;

×

270

void mndReleaseVgroup(SMnode *pMnode, SVgObj *pVgroup) {

×

271

  SSdb *pSdb = pMnode->pSdb;

×

272

  sdbRelease(pSdb, pVgroup);

×

273

×

275

void *mndBuildCreateVnodeReq(SMnode *pMnode, SDnodeObj *pDnode, SDbObj *pDb, SVgObj *pVgroup, int32_t *pContLen) {

×

276

  SCreateVnodeReq createReq = {0};

×

277

  createReq.vgId = pVgroup->vgId;

×

278

  memcpy(createReq.db, pDb->name, TSDB_DB_FNAME_LEN);

×

279

  createReq.dbUid = pDb->uid;

×

280

  createReq.vgVersion = pVgroup->version;

×

281

  createReq.numOfStables = pDb->cfg.numOfStables;

×

282

  createReq.buffer = pDb->cfg.buffer;

×

283

  createReq.pageSize = pDb->cfg.pageSize;

×

284

  createReq.pages = pDb->cfg.pages;

×

285

  createReq.cacheLastSize = pDb->cfg.cacheLastSize;

×

286

  createReq.daysPerFile = pDb->cfg.daysPerFile;

×

287

  createReq.daysToKeep0 = pDb->cfg.daysToKeep0;

×

288

  createReq.daysToKeep1 = pDb->cfg.daysToKeep1;

×

289

  createReq.daysToKeep2 = pDb->cfg.daysToKeep2;

×

290

  createReq.keepTimeOffset = pDb->cfg.keepTimeOffset;

×

291

  createReq.s3ChunkSize = pDb->cfg.s3ChunkSize;

×

292

  createReq.s3KeepLocal = pDb->cfg.s3KeepLocal;

×

293

  createReq.s3Compact = pDb->cfg.s3Compact;

×

294

  createReq.minRows = pDb->cfg.minRows;

×

295

  createReq.maxRows = pDb->cfg.maxRows;

×

296

  createReq.walFsyncPeriod = pDb->cfg.walFsyncPeriod;

×

297

  createReq.walLevel = pDb->cfg.walLevel;

×

298

  createReq.precision = pDb->cfg.precision;

×

299

  createReq.compression = pDb->cfg.compression;

×

300

  createReq.strict = pDb->cfg.strict;

×

301

  createReq.cacheLast = pDb->cfg.cacheLast;

×

302

  createReq.replica = 0;

×

303

  createReq.learnerReplica = 0;

×

304

  createReq.selfIndex = -1;

×

305

  createReq.learnerSelfIndex = -1;

×

306

  createReq.hashBegin = pVgroup->hashBegin;

×

307

  createReq.hashEnd = pVgroup->hashEnd;

×

308

  createReq.hashMethod = pDb->cfg.hashMethod;

×

309

  createReq.numOfRetensions = pDb->cfg.numOfRetensions;

×

310

  createReq.pRetensions = pDb->cfg.pRetensions;

×

311

  createReq.isTsma = pVgroup->isTsma;

×

312

  createReq.pTsma = pVgroup->pTsma;

×

313

  createReq.walRetentionPeriod = pDb->cfg.walRetentionPeriod;

×

314

  createReq.walRetentionSize = pDb->cfg.walRetentionSize;

×

315

  createReq.walRollPeriod = pDb->cfg.walRollPeriod;

×

316

  createReq.walSegmentSize = pDb->cfg.walSegmentSize;

×

317

  createReq.sstTrigger = pDb->cfg.sstTrigger;

×

318

  createReq.hashPrefix = pDb->cfg.hashPrefix;

×

319

  createReq.hashSuffix = pDb->cfg.hashSuffix;

×

320

  createReq.tsdbPageSize = pDb->cfg.tsdbPageSize;

×

321

  createReq.changeVersion = ++(pVgroup->syncConfChangeVer);

×

322

  createReq.encryptAlgorithm = pDb->cfg.encryptAlgorithm;

×

323

  int32_t code = 0;

×

325

  for (int32_t v = 0; v < pVgroup->replica; ++v) {

×

326

    SReplica *pReplica = NULL;

×

328

    if (pVgroup->vnodeGid[v].nodeRole == TAOS_SYNC_ROLE_VOTER) {

×

329

      pReplica = &createReq.replicas[createReq.replica];

×

331

      pReplica = &createReq.learnerReplicas[createReq.learnerReplica];

×

334

    SVnodeGid *pVgid = &pVgroup->vnodeGid[v];

×

335

    SDnodeObj *pVgidDnode = mndAcquireDnode(pMnode, pVgid->dnodeId);

×

336

    if (pVgidDnode == NULL) {

×

337

      return NULL;

×

340

    pReplica->id = pVgidDnode->id;

×

341

    pReplica->port = pVgidDnode->port;

×

342

    memcpy(pReplica->fqdn, pVgidDnode->fqdn, TSDB_FQDN_LEN);

×

343

    mndReleaseDnode(pMnode, pVgidDnode);

×

345

    if (pVgroup->vnodeGid[v].nodeRole == TAOS_SYNC_ROLE_VOTER) {

×

346

      if (pDnode->id == pVgid->dnodeId) {

×

347

        createReq.selfIndex = createReq.replica;

×

350

      if (pDnode->id == pVgid->dnodeId) {

×

351

        createReq.learnerSelfIndex = createReq.learnerReplica;

×

355

    if (pVgroup->vnodeGid[v].nodeRole == TAOS_SYNC_ROLE_VOTER) {

×

356

      createReq.replica++;

×

358

      createReq.learnerReplica++;

×

362

  if (createReq.selfIndex == -1 && createReq.learnerSelfIndex == -1) {

×

363

    terrno = TSDB_CODE_APP_ERROR;

×

364

    return NULL;

×

367

  createReq.changeVersion = pVgroup->syncConfChangeVer;

×

369

  mInfo(

×

374

  for (int32_t i = 0; i < createReq.replica; ++i) {

×

375

    mInfo("vgId:%d, replica:%d ep:%s:%u", createReq.vgId, i, createReq.replicas[i].fqdn, createReq.replicas[i].port);

×

377

  for (int32_t i = 0; i < createReq.learnerReplica; ++i) {

×

378

    mInfo("vgId:%d, replica:%d ep:%s:%u", createReq.vgId, i, createReq.learnerReplicas[i].fqdn,

×

382

  int32_t contLen = tSerializeSCreateVnodeReq(NULL, 0, &createReq);

×

383

  if (contLen < 0) {

×

384

    terrno = TSDB_CODE_OUT_OF_MEMORY;

×

385

    return NULL;

×

388

  void *pReq = taosMemoryMalloc(contLen);

×

389

  if (pReq == NULL) {

×

390

    terrno = TSDB_CODE_OUT_OF_MEMORY;

×

391

    return NULL;

×

394

  code = tSerializeSCreateVnodeReq(pReq, contLen, &createReq);

×

395

  if (code < 0) {

×

396

    terrno = TSDB_CODE_APP_ERROR;

×

397

    taosMemoryFree(pReq);

×

398

    mError("vgId:%d, failed to serialize create vnode req,since %s", createReq.vgId, terrstr());

×

399

    return NULL;

×

401

  *pContLen = contLen;

×

402

  return pReq;

×

405

static void *mndBuildAlterVnodeConfigReq(SMnode *pMnode, SDbObj *pDb, SVgObj *pVgroup, int32_t *pContLen) {

×

406

  SAlterVnodeConfigReq alterReq = {0};

×

407

  alterReq.vgVersion = pVgroup->version;

×

408

  alterReq.buffer = pDb->cfg.buffer;

×

409

  alterReq.pageSize = pDb->cfg.pageSize;

×

410

  alterReq.pages = pDb->cfg.pages;

×

411

  alterReq.cacheLastSize = pDb->cfg.cacheLastSize;

×

412

  alterReq.daysPerFile = pDb->cfg.daysPerFile;

×

413

  alterReq.daysToKeep0 = pDb->cfg.daysToKeep0;

×

414

  alterReq.daysToKeep1 = pDb->cfg.daysToKeep1;

×

415

  alterReq.daysToKeep2 = pDb->cfg.daysToKeep2;

×

416

  alterReq.keepTimeOffset = pDb->cfg.keepTimeOffset;

×

417

  alterReq.walFsyncPeriod = pDb->cfg.walFsyncPeriod;

×

418

  alterReq.walLevel = pDb->cfg.walLevel;

×

419

  alterReq.strict = pDb->cfg.strict;

×

420

  alterReq.cacheLast = pDb->cfg.cacheLast;

×

421

  alterReq.sttTrigger = pDb->cfg.sstTrigger;

×

422

  alterReq.minRows = pDb->cfg.minRows;

×

423

  alterReq.walRetentionPeriod = pDb->cfg.walRetentionPeriod;

×

424

  alterReq.walRetentionSize = pDb->cfg.walRetentionSize;

×

425

  alterReq.s3KeepLocal = pDb->cfg.s3KeepLocal;

×

426

  alterReq.s3Compact = pDb->cfg.s3Compact;

×

428

  mInfo("vgId:%d, build alter vnode config req", pVgroup->vgId);

×

429

  int32_t contLen = tSerializeSAlterVnodeConfigReq(NULL, 0, &alterReq);

×

430

  if (contLen < 0) {

×

431

    terrno = TSDB_CODE_OUT_OF_MEMORY;

×

432

    return NULL;

×

434

  contLen += sizeof(SMsgHead);

×

436

  void *pReq = taosMemoryMalloc(contLen);

×

437

  if (pReq == NULL) {

×

438

    terrno = TSDB_CODE_OUT_OF_MEMORY;

×

439

    return NULL;

×

442

  SMsgHead *pHead = pReq;

×

443

  pHead->contLen = htonl(contLen);

×

444

  pHead->vgId = htonl(pVgroup->vgId);

×

446

  if (tSerializeSAlterVnodeConfigReq((char *)pReq + sizeof(SMsgHead), contLen, &alterReq) < 0) {

×

447

    taosMemoryFree(pReq);

×

448

    mError("vgId:%d, failed to serialize alter vnode config req,since %s", pVgroup->vgId, terrstr());

×

449

    terrno = TSDB_CODE_OUT_OF_MEMORY;

×

450

    return NULL;

×

452

  *pContLen = contLen;

×

453

  return pReq;

×

456

static void *mndBuildAlterVnodeReplicaReq(SMnode *pMnode, SDbObj *pDb, SVgObj *pVgroup, int32_t dnodeId,

×

458

  SAlterVnodeReplicaReq alterReq = {

×

459

      .vgId = pVgroup->vgId,

×

460

      .strict = pDb->cfg.strict,

×

465

      .changeVersion = ++(pVgroup->syncConfChangeVer),

×

468

  for (int32_t v = 0; v < pVgroup->replica; ++v) {

×

469

    SReplica *pReplica = NULL;

×

471

    if (pVgroup->vnodeGid[v].nodeRole == TAOS_SYNC_ROLE_VOTER) {

×

472

      pReplica = &alterReq.replicas[alterReq.replica];

×

473

      alterReq.replica++;

×

475

      pReplica = &alterReq.learnerReplicas[alterReq.learnerReplica];

×

476

      alterReq.learnerReplica++;

×

479

    SVnodeGid *pVgid = &pVgroup->vnodeGid[v];

×

480

    SDnodeObj *pVgidDnode = mndAcquireDnode(pMnode, pVgid->dnodeId);

×

481

    if (pVgidDnode == NULL) return NULL;

×

483

    pReplica->id = pVgidDnode->id;

×

484

    pReplica->port = pVgidDnode->port;

×

485

    memcpy(pReplica->fqdn, pVgidDnode->fqdn, TSDB_FQDN_LEN);

×

486

    mndReleaseDnode(pMnode, pVgidDnode);

×

488

    if (pVgroup->vnodeGid[v].nodeRole == TAOS_SYNC_ROLE_VOTER) {

×

489

      if (dnodeId == pVgid->dnodeId) {

×

490

        alterReq.selfIndex = v;

×

493

      if (dnodeId == pVgid->dnodeId) {

×

494

        alterReq.learnerSelfIndex = v;

×

499

  mInfo(

×

504

  for (int32_t i = 0; i < alterReq.replica; ++i) {

×

505

    mInfo("vgId:%d, replica:%d ep:%s:%u", alterReq.vgId, i, alterReq.replicas[i].fqdn, alterReq.replicas[i].port);

×

507

  for (int32_t i = 0; i < alterReq.learnerReplica; ++i) {

×

508

    mInfo("vgId:%d, learnerReplica:%d ep:%s:%u", alterReq.vgId, i, alterReq.learnerReplicas[i].fqdn,

×

512

  if (alterReq.selfIndex == -1 && alterReq.learnerSelfIndex == -1) {

×

513

    terrno = TSDB_CODE_APP_ERROR;

×

514

    return NULL;

×

517

  int32_t contLen = tSerializeSAlterVnodeReplicaReq(NULL, 0, &alterReq);

×

518

  if (contLen < 0) {

×

519

    terrno = TSDB_CODE_OUT_OF_MEMORY;

×

520

    return NULL;

×

523

  void *pReq = taosMemoryMalloc(contLen);

×

524

  if (pReq == NULL) {

×

525

    terrno = TSDB_CODE_OUT_OF_MEMORY;

×

526

    return NULL;

×

529

  if (tSerializeSAlterVnodeReplicaReq(pReq, contLen, &alterReq) < 0) {

×

530

    mError("vgId:%d, failed to serialize alter vnode req,since %s", alterReq.vgId, terrstr());

×

531

    taosMemoryFree(pReq);

×

532

    terrno = TSDB_CODE_OUT_OF_MEMORY;

×

533

    return NULL;

×

535

  *pContLen = contLen;

×

536

  return pReq;

×

539

static void *mndBuildCheckLearnCatchupReq(SMnode *pMnode, SDbObj *pDb, SVgObj *pVgroup, int32_t dnodeId,

×

541

  SCheckLearnCatchupReq req = {

×

542

      .vgId = pVgroup->vgId,

×

543

      .strict = pDb->cfg.strict,

×

550

  for (int32_t v = 0; v < pVgroup->replica; ++v) {

×

551

    SReplica *pReplica = NULL;

×

553

    if (pVgroup->vnodeGid[v].nodeRole == TAOS_SYNC_ROLE_VOTER) {

×

554

      pReplica = &req.replicas[req.replica];

×

555

      req.replica++;

×

557

      pReplica = &req.learnerReplicas[req.learnerReplica];

×

558

      req.learnerReplica++;

×

561

    SVnodeGid *pVgid = &pVgroup->vnodeGid[v];

×

562

    SDnodeObj *pVgidDnode = mndAcquireDnode(pMnode, pVgid->dnodeId);

×

563

    if (pVgidDnode == NULL) return NULL;

×

565

    pReplica->id = pVgidDnode->id;

×

566

    pReplica->port = pVgidDnode->port;

×

567

    memcpy(pReplica->fqdn, pVgidDnode->fqdn, TSDB_FQDN_LEN);

×

568

    mndReleaseDnode(pMnode, pVgidDnode);

×

570

    if (pVgroup->vnodeGid[v].nodeRole == TAOS_SYNC_ROLE_VOTER) {

×

571

      if (dnodeId == pVgid->dnodeId) {

×

572

        req.selfIndex = v;

×

575

      if (dnodeId == pVgid->dnodeId) {

×

576

        req.learnerSelfIndex = v;

×

581

  mInfo("vgId:%d, build alter vnode req, replica:%d selfIndex:%d learnerReplica:%d learnerSelfIndex:%d strict:%d",

×

583

  for (int32_t i = 0; i < req.replica; ++i) {

×

584

    mInfo("vgId:%d, replica:%d ep:%s:%u", req.vgId, i, req.replicas[i].fqdn, req.replicas[i].port);

×

586

  for (int32_t i = 0; i < req.learnerReplica; ++i) {

×

587

    mInfo("vgId:%d, learnerReplica:%d ep:%s:%u", req.vgId, i, req.learnerReplicas[i].fqdn, req.learnerReplicas[i].port);

×

590

  if (req.selfIndex == -1 && req.learnerSelfIndex == -1) {

×

591

    terrno = TSDB_CODE_APP_ERROR;

×

592

    return NULL;

×

595

  int32_t contLen = tSerializeSAlterVnodeReplicaReq(NULL, 0, &req);

×

596

  if (contLen < 0) {

×

597

    terrno = TSDB_CODE_OUT_OF_MEMORY;

×

598

    return NULL;

×

601

  void *pReq = taosMemoryMalloc(contLen);

×

602

  if (pReq == NULL) {

×

603

    terrno = TSDB_CODE_OUT_OF_MEMORY;

×

604

    return NULL;

×

607

  if (tSerializeSAlterVnodeReplicaReq(pReq, contLen, &req) < 0) {

×

608

    mError("vgId:%d, failed to serialize alter vnode req,since %s", req.vgId, terrstr());

×

609

    taosMemoryFree(pReq);

×

610

    terrno = TSDB_CODE_OUT_OF_MEMORY;

×

611

    return NULL;

×

613

  *pContLen = contLen;

×

614

  return pReq;

×

617

static void *mndBuildDisableVnodeWriteReq(SMnode *pMnode, SDbObj *pDb, int32_t vgId, int32_t *pContLen) {

×

618

  SDisableVnodeWriteReq disableReq = {

×

623

  mInfo("vgId:%d, build disable vnode write req", vgId);

×

624

  int32_t contLen = tSerializeSDisableVnodeWriteReq(NULL, 0, &disableReq);

×

625

  if (contLen < 0) {

×

626

    terrno = TSDB_CODE_OUT_OF_MEMORY;

×

627

    return NULL;

×

630

  void *pReq = taosMemoryMalloc(contLen);

×

631

  if (pReq == NULL) {

×

632

    terrno = TSDB_CODE_OUT_OF_MEMORY;

×

633

    return NULL;

×

636

  if (tSerializeSDisableVnodeWriteReq(pReq, contLen, &disableReq) < 0) {

×

637

    mError("vgId:%d, failed to serialize disable vnode write req,since %s", vgId, terrstr());

×

638

    taosMemoryFree(pReq);

×

639

    terrno = TSDB_CODE_OUT_OF_MEMORY;

×

640

    return NULL;

×

642

  *pContLen = contLen;

×

643

  return pReq;

×

646

static void *mndBuildAlterVnodeHashRangeReq(SMnode *pMnode, int32_t srcVgId, SVgObj *pVgroup, int32_t *pContLen) {

×

647

  SAlterVnodeHashRangeReq alterReq = {

×

649

      .dstVgId = pVgroup->vgId,

×

650

      .hashBegin = pVgroup->hashBegin,

×

651

      .hashEnd = pVgroup->hashEnd,

×

652

      .changeVersion = ++(pVgroup->syncConfChangeVer),

×

655

  mInfo("vgId:%d, build alter vnode hashrange req, dstVgId:%d, hashrange:[%u, %u]", srcVgId, pVgroup->vgId,

×

657

  int32_t contLen = tSerializeSAlterVnodeHashRangeReq(NULL, 0, &alterReq);

×

658

  if (contLen < 0) {

×

659

    terrno = TSDB_CODE_OUT_OF_MEMORY;

×

660

    return NULL;

×

663

  void *pReq = taosMemoryMalloc(contLen);

×

664

  if (pReq == NULL) {

×

665

    terrno = TSDB_CODE_OUT_OF_MEMORY;

×

666

    return NULL;

×

669

  if (tSerializeSAlterVnodeHashRangeReq(pReq, contLen, &alterReq) < 0) {

×

670

    mError("vgId:%d, failed to serialize alter vnode hashrange req,since %s", srcVgId, terrstr());

×

671

    taosMemoryFree(pReq);

×

672

    terrno = TSDB_CODE_OUT_OF_MEMORY;

×

673

    return NULL;

×

675

  *pContLen = contLen;

×

676

  return pReq;

×

679

void *mndBuildDropVnodeReq(SMnode *pMnode, SDnodeObj *pDnode, SDbObj *pDb, SVgObj *pVgroup, int32_t *pContLen) {

×

680

  SDropVnodeReq dropReq = {0};

×

681

  dropReq.dnodeId = pDnode->id;

×

682

  dropReq.vgId = pVgroup->vgId;

×

683

  memcpy(dropReq.db, pDb->name, TSDB_DB_FNAME_LEN);

×

684

  dropReq.dbUid = pDb->uid;

×

686

  mInfo("vgId:%d, build drop vnode req", dropReq.vgId);

×

687

  int32_t contLen = tSerializeSDropVnodeReq(NULL, 0, &dropReq);

×

688

  if (contLen < 0) {

×

689

    terrno = TSDB_CODE_OUT_OF_MEMORY;

×

690

    return NULL;

×

693

  void *pReq = taosMemoryMalloc(contLen);

×

694

  if (pReq == NULL) {

×

695

    terrno = TSDB_CODE_OUT_OF_MEMORY;

×

696

    return NULL;

×

699

  if (tSerializeSDropVnodeReq(pReq, contLen, &dropReq) < 0) {

×

700

    mError("vgId:%d, failed to serialize drop vnode req,since %s", dropReq.vgId, terrstr());

×

701

    taosMemoryFree(pReq);

×

702

    terrno = TSDB_CODE_OUT_OF_MEMORY;

×

703

    return NULL;

×

705

  *pContLen = contLen;

×

706

  return pReq;

×

709

static bool mndResetDnodesArrayFp(SMnode *pMnode, void *pObj, void *p1, void *p2, void *p3) {

×

710

  SDnodeObj *pDnode = pObj;

×

711

  pDnode->numOfVnodes = 0;

×

712

  pDnode->numOfOtherNodes = 0;

×

713

  return true;

×

716

static bool mndBuildDnodesArrayFp(SMnode *pMnode, void *pObj, void *p1, void *p2, void *p3) {

×

717

  SDnodeObj *pDnode = pObj;

×

718

  SArray    *pArray = p1;

×

719

  int32_t    exceptDnodeId = *(int32_t *)p2;

×

720

  SArray    *dnodeList = p3;

×

722

  if (exceptDnodeId == pDnode->id) {

×

723

    return true;

×

726

  if (dnodeList != NULL) {

×

727

    int32_t dnodeListSize = taosArrayGetSize(dnodeList);

×

728

    if (dnodeListSize > 0) {

×

729

      bool inDnodeList = false;

×

730

      for (int32_t index = 0; index < dnodeListSize; ++index) {

×

731

        int32_t dnodeId = *(int32_t *)taosArrayGet(dnodeList, index);

×

732

        if (pDnode->id == dnodeId) {

×

733

          inDnodeList = true;

×

736

      if (!inDnodeList) {

×

737

        return true;

×

742

  int64_t curMs = taosGetTimestampMs();

×

743

  bool    online = mndIsDnodeOnline(pDnode, curMs);

×

744

  bool    isMnode = mndIsMnode(pMnode, pDnode->id);

×

745

  pDnode->numOfVnodes = mndGetVnodesNum(pMnode, pDnode->id);

×

746

  pDnode->memUsed = mndGetVnodesMemory(pMnode, pDnode->id);

×

748

  mInfo("dnode:%d, vnodes:%d supportVnodes:%d isMnode:%d online:%d memory avail:%" PRId64 " used:%" PRId64, pDnode->id,

×

751

  if (isMnode) {

×

752

    pDnode->numOfOtherNodes++;

×

755

  if (online && pDnode->numOfSupportVnodes > 0) {

×

756

    if (taosArrayPush(pArray, pDnode) == NULL) return false;

×

758

  return true;

×

761

SArray *mndBuildDnodesArray(SMnode *pMnode, int32_t exceptDnodeId, SArray *dnodeList) {

×

762

  SSdb   *pSdb = pMnode->pSdb;

×

763

  int32_t numOfDnodes = mndGetDnodeSize(pMnode);

×

765

  SArray *pArray = taosArrayInit(numOfDnodes, sizeof(SDnodeObj));

×

766

  if (pArray == NULL) {

×

767

    terrno = TSDB_CODE_OUT_OF_MEMORY;

×

768

    return NULL;

×

771

  sdbTraverse(pSdb, SDB_DNODE, mndResetDnodesArrayFp, NULL, NULL, NULL);

×

772

  sdbTraverse(pSdb, SDB_DNODE, mndBuildDnodesArrayFp, pArray, &exceptDnodeId, dnodeList);

×

774

  mDebug("build %d dnodes array", (int32_t)taosArrayGetSize(pArray));

×

775

  for (int32_t i = 0; i < (int32_t)taosArrayGetSize(pArray); ++i) {

×

776

    SDnodeObj *pDnode = taosArrayGet(pArray, i);

×

777

    mDebug("dnode:%d, vnodes:%d others:%d", pDnode->id, pDnode->numOfVnodes, pDnode->numOfOtherNodes);

×

779

  return pArray;

×

782

static int32_t mndCompareDnodeId(int32_t *dnode1Id, int32_t *dnode2Id) {

×

783

  if (*dnode1Id == *dnode2Id) {

×

784

    return 0;

×

786

  return *dnode1Id > *dnode2Id ? 1 : -1;

×

789

static float mndGetDnodeScore(SDnodeObj *pDnode, int32_t additionDnodes, float ratio) {

×

790

  float totalDnodes = pDnode->numOfVnodes + (float)pDnode->numOfOtherNodes * ratio + additionDnodes;

×

791

  return totalDnodes / pDnode->numOfSupportVnodes;

×

794

static int32_t mndCompareDnodeVnodes(SDnodeObj *pDnode1, SDnodeObj *pDnode2) {

×

795

  float d1Score = mndGetDnodeScore(pDnode1, 0, 0.9);

×

796

  float d2Score = mndGetDnodeScore(pDnode2, 0, 0.9);

×

797

  if (d1Score == d2Score) {

×

798

    return 0;

×

800

  return d1Score > d2Score ? 1 : -1;

×

803

void mndSortVnodeGid(SVgObj *pVgroup) {

×

804

  for (int32_t i = 0; i < pVgroup->replica; ++i) {

×

805

    for (int32_t j = 0; j < pVgroup->replica - 1 - i; ++j) {

×

806

      if (pVgroup->vnodeGid[j].dnodeId > pVgroup->vnodeGid[j + 1].dnodeId) {

×

807

        TSWAP(pVgroup->vnodeGid[j], pVgroup->vnodeGid[j + 1]);

×

811

×

813

static int32_t mndGetAvailableDnode(SMnode *pMnode, SDbObj *pDb, SVgObj *pVgroup, SArray *pArray) {

×

814

  mDebug("start to sort %d dnodes", (int32_t)taosArrayGetSize(pArray));

×

815

  taosArraySort(pArray, (__compar_fn_t)mndCompareDnodeVnodes);

×

816

  for (int32_t i = 0; i < (int32_t)taosArrayGetSize(pArray); ++i) {

×

817

    SDnodeObj *pDnode = taosArrayGet(pArray, i);

×

818

    mDebug("dnode:%d, score:%f", pDnode->id, mndGetDnodeScore(pDnode, 0, 0.9));

×

821

  int32_t size = taosArrayGetSize(pArray);

×

822

  if (size < pVgroup->replica) {

×

823

    mError("db:%s, vgId:%d, no enough online dnodes:%d to alloc %d replica", pVgroup->dbName, pVgroup->vgId, size,

×

825

    TAOS_RETURN(TSDB_CODE_MND_NO_ENOUGH_DNODES);

×

828

  for (int32_t v = 0; v < pVgroup->replica; ++v) {

×

829

    SVnodeGid *pVgid = &pVgroup->vnodeGid[v];

×

830

    SDnodeObj *pDnode = taosArrayGet(pArray, v);

×

831

    if (pDnode == NULL) {

×

832

      TAOS_RETURN(TSDB_CODE_MND_NO_ENOUGH_DNODES);

×

834

    if (pDnode->numOfVnodes >= pDnode->numOfSupportVnodes) {

×

835

      TAOS_RETURN(TSDB_CODE_MND_NO_ENOUGH_VNODES);

×

838

    int64_t vgMem = mndGetVgroupMemory(pMnode, pDb, pVgroup);

×

839

    if (pDnode->memAvail - vgMem - pDnode->memUsed <= 0) {

×

840

      mError("db:%s, vgId:%d, no enough memory:%" PRId64 " in dnode:%d, avail:%" PRId64 " used:%" PRId64,

×

842

      TAOS_RETURN(TSDB_CODE_MND_NO_ENOUGH_MEM_IN_DNODE);

×

844

      pDnode->memUsed += vgMem;

×

847

    pVgid->dnodeId = pDnode->id;

×

848

    if (pVgroup->replica == 1) {

×

849

      pVgid->syncState = TAOS_SYNC_STATE_LEADER;

×

851

      pVgid->syncState = TAOS_SYNC_STATE_FOLLOWER;

×

854

    mInfo("db:%s, vgId:%d, vn:%d is alloced, memory:%" PRId64 ", dnode:%d avail:%" PRId64 " used:%" PRId64,

×

856

    pDnode->numOfVnodes++;

×

859

  mndSortVnodeGid(pVgroup);

×

860

  return 0;

×

863

int32_t mndAllocSmaVgroup(SMnode *pMnode, SDbObj *pDb, SVgObj *pVgroup) {

×

864

  int32_t code = 0;

×

865

  SArray *pArray = mndBuildDnodesArray(pMnode, 0, NULL);

×

866

  if (pArray == NULL) {

×

867

    code = TSDB_CODE_MND_RETURN_VALUE_NULL;

×

868

    if (terrno != 0) code = terrno;

×

869

    TAOS_RETURN(code);

×

872

  pVgroup->vgId = sdbGetMaxId(pMnode->pSdb, SDB_VGROUP);

×

873

  pVgroup->isTsma = 1;

×

874

  pVgroup->createdTime = taosGetTimestampMs();

×

875

  pVgroup->updateTime = pVgroup->createdTime;

×

876

  pVgroup->version = 1;

×

877

  memcpy(pVgroup->dbName, pDb->name, TSDB_DB_FNAME_LEN);

×

878

  pVgroup->dbUid = pDb->uid;

×

879

  pVgroup->replica = 1;

×

881

  if (mndGetAvailableDnode(pMnode, pDb, pVgroup, pArray) != 0) return -1;

×

882

  taosArrayDestroy(pArray);

×

884

  mInfo("db:%s, sma vgId:%d is alloced", pDb->name, pVgroup->vgId);

×

885

  return 0;

×

888

int32_t mndAllocVgroup(SMnode *pMnode, SDbObj *pDb, SVgObj **ppVgroups, SArray *dnodeList) {

×

889

  int32_t code = -1;

×

890

  SArray *pArray = NULL;

×

891

  SVgObj *pVgroups = NULL;

×

893

  pVgroups = taosMemoryCalloc(pDb->cfg.numOfVgroups, sizeof(SVgObj));

×

894

  if (pVgroups == NULL) {

×

895

    code = terrno;

×

896

    goto _OVER;

×

899

  pArray = mndBuildDnodesArray(pMnode, 0, dnodeList);

×

900

  if (pArray == NULL) {

×

901

    code = TSDB_CODE_MND_RETURN_VALUE_NULL;

×

902

    if (terrno != 0) code = terrno;

×

903

    goto _OVER;

×

906

  mInfo("db:%s, total %d dnodes used to create %d vgroups (%d vnodes)", pDb->name, (int32_t)taosArrayGetSize(pArray),

×

909

  int32_t  allocedVgroups = 0;

×

910

  int32_t  maxVgId = sdbGetMaxId(pMnode->pSdb, SDB_VGROUP);

×

911

  uint32_t hashMin = 0;

×

912

  uint32_t hashMax = UINT32_MAX;

×

913

  uint32_t hashInterval = (hashMax - hashMin) / pDb->cfg.numOfVgroups;

×

915

  if (maxVgId < 2) maxVgId = 2;

×

917

  for (uint32_t v = 0; v < pDb->cfg.numOfVgroups; v++) {

×

918

    SVgObj *pVgroup = &pVgroups[v];

×

919

    pVgroup->vgId = maxVgId++;

×

920

    pVgroup->createdTime = taosGetTimestampMs();

×

921

    pVgroup->updateTime = pVgroups->createdTime;

×

922

    pVgroup->version = 1;

×

923

    pVgroup->hashBegin = hashMin + hashInterval * v;

×

924

    if (v == pDb->cfg.numOfVgroups - 1) {

×

925

      pVgroup->hashEnd = hashMax;

×

927

      pVgroup->hashEnd = hashMin + hashInterval * (v + 1) - 1;

×

930

    memcpy(pVgroup->dbName, pDb->name, TSDB_DB_FNAME_LEN);

×

931

    pVgroup->dbUid = pDb->uid;

×

932

    pVgroup->replica = pDb->cfg.replications;

×

934

    if ((code = mndGetAvailableDnode(pMnode, pDb, pVgroup, pArray)) != 0) {

×

935

      goto _OVER;

×

938

    allocedVgroups++;

×

941

  *ppVgroups = pVgroups;

×

942

  code = 0;

×

944

  mInfo("db:%s, total %d vgroups is alloced, replica:%d", pDb->name, pDb->cfg.numOfVgroups, pDb->cfg.replications);

×

946

_OVER:

×

947

  if (code != 0) taosMemoryFree(pVgroups);

×

948

  taosArrayDestroy(pArray);

×

949

  TAOS_RETURN(code);

×

952

SEpSet mndGetVgroupEpset(SMnode *pMnode, const SVgObj *pVgroup) {

×

953

  SEpSet epset = {0};

×

955

  for (int32_t v = 0; v < pVgroup->replica; ++v) {

×

956

    const SVnodeGid *pVgid = &pVgroup->vnodeGid[v];

×

957

    SDnodeObj       *pDnode = mndAcquireDnode(pMnode, pVgid->dnodeId);

×

958

    if (pDnode == NULL) continue;

×

960

    if (pVgid->syncState == TAOS_SYNC_STATE_LEADER || pVgid->syncState == TAOS_SYNC_STATE_ASSIGNED_LEADER) {

×

961

      epset.inUse = epset.numOfEps;

×

964

    if (addEpIntoEpSet(&epset, pDnode->fqdn, pDnode->port) != 0) {

×

965

      mWarn("vgId:%d, failed to add ep:%s:%d into epset", pVgroup->vgId, pDnode->fqdn, pDnode->port);

×

967

    mndReleaseDnode(pMnode, pDnode);

×

969

  epsetSort(&epset);

×

971

  return epset;

×

974

SEpSet mndGetVgroupEpsetById(SMnode *pMnode, int32_t vgId) {

×

975

  SEpSet epset = {0};

×

977

  SVgObj *pVgroup = mndAcquireVgroup(pMnode, vgId);

×

978

  if (!pVgroup) return epset;

×

980

  for (int32_t v = 0; v < pVgroup->replica; ++v) {

×

981

    const SVnodeGid *pVgid = &pVgroup->vnodeGid[v];

×

982

    SDnodeObj       *pDnode = mndAcquireDnode(pMnode, pVgid->dnodeId);

×

983

    if (pDnode == NULL) continue;

×

985

    if (pVgid->syncState == TAOS_SYNC_STATE_LEADER || pVgid->syncState == TAOS_SYNC_STATE_ASSIGNED_LEADER) {

×

986

      epset.inUse = epset.numOfEps;

×

989

    if (addEpIntoEpSet(&epset, pDnode->fqdn, pDnode->port) != 0) {

×

990

      mWarn("vgId:%d, failed to add ep:%s:%d into epset", pVgroup->vgId, pDnode->fqdn, pDnode->port);

×

992

    mndReleaseDnode(pMnode, pDnode);

×

995

  mndReleaseVgroup(pMnode, pVgroup);

×

996

  return epset;

×

999

static int32_t mndRetrieveVgroups(SRpcMsg *pReq, SShowObj *pShow, SSDataBlock *pBlock, int32_t rows) {

×

1000

  SMnode *pMnode = pReq->info.node;

×

1001

  SSdb   *pSdb = pMnode->pSdb;

×

1002

  int32_t numOfRows = 0;

×

1003

  SVgObj *pVgroup = NULL;

×

1004

  int32_t cols = 0;

×

1005

  int64_t curMs = taosGetTimestampMs();

×

1006

  int32_t code = 0;

×

1008

  SDbObj *pDb = NULL;

×

1009

  if (strlen(pShow->db) > 0) {

×

1010

    pDb = mndAcquireDb(pMnode, pShow->db);

×

1011

    if (pDb == NULL) {

×

1012

      return 0;

×

1016

  while (numOfRows < rows) {

×

1017

    pShow->pIter = sdbFetch(pSdb, SDB_VGROUP, pShow->pIter, (void **)&pVgroup);

×

1018

    if (pShow->pIter == NULL) break;

×

1020

    if (pDb != NULL && pVgroup->dbUid != pDb->uid) {

×

1021

      sdbRelease(pSdb, pVgroup);

×

1022

      continue;

×

1025

    cols = 0;

×

1026

    SColumnInfoData *pColInfo = taosArrayGet(pBlock->pDataBlock, cols++);

×

1027

    code = colDataSetVal(pColInfo, numOfRows, (const char *)&pVgroup->vgId, false);

×

1028

    if (code != 0) {

×

1029

      mError("vgId:%d, failed to set vgId, since %s", pVgroup->vgId, tstrerror(code));

×

1030

      return code;

×

1033

    SName name = {0};

×

1034

    char  db[TSDB_DB_NAME_LEN + VARSTR_HEADER_SIZE] = {0};

×

1035

    code = tNameFromString(&name, pVgroup->dbName, T_NAME_ACCT | T_NAME_DB);

×

1036

    if (code != 0) {

×

1037

      mError("vgId:%d, failed to set dbName, since %s", pVgroup->vgId, tstrerror(code));

×

1038

      return code;

×

1040

    (void)tNameGetDbName(&name, varDataVal(db));

×

1041

    varDataSetLen(db, strlen(varDataVal(db)));

×

1043

    pColInfo = taosArrayGet(pBlock->pDataBlock, cols++);

×

1044

    code = colDataSetVal(pColInfo, numOfRows, (const char *)db, false);

×

1045

    if (code != 0) {

×

1046

      mError("vgId:%d, failed to set dbName, since %s", pVgroup->vgId, tstrerror(code));

×

1047

      return code;

×

1050

    pColInfo = taosArrayGet(pBlock->pDataBlock, cols++);

×

1051

    code = colDataSetVal(pColInfo, numOfRows, (const char *)&pVgroup->numOfTables, false);

×

1052

    if (code != 0) {

×

1053

      mError("vgId:%d, failed to set numOfTables, since %s", pVgroup->vgId, tstrerror(code));

×

1054

      return code;

×

1058

    for (int32_t i = 0; i < 4; ++i) {

×

1059

      pColInfo = taosArrayGet(pBlock->pDataBlock, cols++);

×

1060

      if (i < pVgroup->replica) {

×

1061

        int16_t dnodeId = (int16_t)pVgroup->vnodeGid[i].dnodeId;

×

1062

        code = colDataSetVal(pColInfo, numOfRows, (const char *)&dnodeId, false);

×

1063

        if (code != 0) {

×

1064

          mError("vgId:%d, failed to set dnodeId, since %s", pVgroup->vgId, tstrerror(code));

×

1065

          return code;

×

1068

        bool       exist = false;

×

1069

        bool       online = false;

×

1070

        SDnodeObj *pDnode = mndAcquireDnode(pMnode, pVgroup->vnodeGid[i].dnodeId);

×

1071

        if (pDnode != NULL) {

×

1072

          exist = true;

×

1073

          online = mndIsDnodeOnline(pDnode, curMs);

×

1074

          mndReleaseDnode(pMnode, pDnode);

×

1077

        char buf1[20] = {0};

×

1078

        char role[20] = "offline";

×

1079

        if (!exist) {

×

1080

          tstrncpy(role, "dropping", sizeof(role));

×

1081

        } else if (online) {

×

1082

          char *star = "";

×

1083

          if (pVgroup->vnodeGid[i].syncState == TAOS_SYNC_STATE_LEADER ||

×

1084

              pVgroup->vnodeGid[i].syncState == TAOS_SYNC_STATE_ASSIGNED_LEADER) {

×

1085

            if (!pVgroup->vnodeGid[i].syncRestore && !pVgroup->vnodeGid[i].syncCanRead) {

×

1086

              star = "**";

×

1087

            } else if (!pVgroup->vnodeGid[i].syncRestore && pVgroup->vnodeGid[i].syncCanRead) {

×

1088

              star = "*";

×

1092

          snprintf(role, sizeof(role), "%s%s", syncStr(pVgroup->vnodeGid[i].syncState), star);

×

1117

        STR_WITH_MAXSIZE_TO_VARSTR(buf1, role, pShow->pMeta->pSchemas[cols].bytes);

×

1119

        pColInfo = taosArrayGet(pBlock->pDataBlock, cols++);

×

1120

        code = colDataSetVal(pColInfo, numOfRows, (const char *)buf1, false);

×

1121

        if (code != 0) {

×

1122

          mError("vgId:%d, failed to set role, since %s", pVgroup->vgId, tstrerror(code));

×

1123

          return code;

×

1126

        colDataSetNULL(pColInfo, numOfRows);

×

1127

        pColInfo = taosArrayGet(pBlock->pDataBlock, cols++);

×

1128

        colDataSetNULL(pColInfo, numOfRows);

×

1132

    pColInfo = taosArrayGet(pBlock->pDataBlock, cols++);

×

1133

    int32_t cacheUsage = (int32_t)pVgroup->cacheUsage;

×

1134

    code = colDataSetVal(pColInfo, numOfRows, (const char *)&cacheUsage, false);

×

1135

    if (code != 0) {

×

1136

      mError("vgId:%d, failed to set cacheUsage, since %s", pVgroup->vgId, tstrerror(code));

×

1137

      return code;

×

1140

    pColInfo = taosArrayGet(pBlock->pDataBlock, cols++);

×

1141

    code = colDataSetVal(pColInfo, numOfRows, (const char *)&pVgroup->numOfCachedTables, false);

×

1142

    if (code != 0) {

×

1143

      mError("vgId:%d, failed to set numOfCachedTables, since %s", pVgroup->vgId, tstrerror(code));

×

1144

      return code;

×

1147

    pColInfo = taosArrayGet(pBlock->pDataBlock, cols++);

×

1148

    code = colDataSetVal(pColInfo, numOfRows, (const char *)&pVgroup->isTsma, false);

×

1149

    if (code != 0) {

×

1150

      mError("vgId:%d, failed to set isTsma, since %s", pVgroup->vgId, tstrerror(code));

×

1151

      return code;

×

1153

    numOfRows++;

×

1154

    sdbRelease(pSdb, pVgroup);

×

1157

  if (pDb != NULL) {

×

1158

    mndReleaseDb(pMnode, pDb);

×

1161

  pShow->numOfRows += numOfRows;

×

1162

  return numOfRows;

×

1165

static void mndCancelGetNextVgroup(SMnode *pMnode, void *pIter) {

×

1166

  SSdb *pSdb = pMnode->pSdb;

×

1167

  sdbCancelFetchByType(pSdb, pIter, SDB_VGROUP);

×

1168

×

1170

static bool mndGetVnodesNumFp(SMnode *pMnode, void *pObj, void *p1, void *p2, void *p3) {

×

1171

  SVgObj  *pVgroup = pObj;

×

1172

  int32_t  dnodeId = *(int32_t *)p1;

×

1173

  int32_t *pNumOfVnodes = (int32_t *)p2;

×

1175

  for (int32_t v = 0; v < pVgroup->replica; ++v) {

×

1176

    if (pVgroup->vnodeGid[v].dnodeId == dnodeId) {

×

1177

      (*pNumOfVnodes)++;

×

1181

  return true;

×

1184

int32_t mndGetVnodesNum(SMnode *pMnode, int32_t dnodeId) {

×

1185

  int32_t numOfVnodes = 0;

×

1186

  sdbTraverse(pMnode->pSdb, SDB_VGROUP, mndGetVnodesNumFp, &dnodeId, &numOfVnodes, NULL);

×

1187

  return numOfVnodes;

×

1190

int64_t mndGetVgroupMemory(SMnode *pMnode, SDbObj *pDbInput, SVgObj *pVgroup) {

×

1191

  SDbObj *pDb = pDbInput;

×

1192

  if (pDbInput == NULL) {

×

1193

    pDb = mndAcquireDb(pMnode, pVgroup->dbName);

×

1196

  int64_t vgroupMemroy = 0;

×

1197

  if (pDb != NULL) {

×

1198

    int64_t buffer = (int64_t)pDb->cfg.buffer * 1024 * 1024;

×

1199

    int64_t cache = (int64_t)pDb->cfg.pages * pDb->cfg.pageSize * 1024;

×

1200

    vgroupMemroy = buffer + cache;

×

1201

    int64_t cacheLast = (int64_t)pDb->cfg.cacheLastSize * 1024 * 1024;

×

1202

    if (pDb->cfg.cacheLast > 0) {

×

1203

      vgroupMemroy += cacheLast;

×

1205

    mDebug("db:%s, vgroup:%d, buffer:%" PRId64 " cache:%" PRId64 " cacheLast:%" PRId64, pDb->name, pVgroup->vgId,

×

1209

  if (pDbInput == NULL) {

×

1210

    mndReleaseDb(pMnode, pDb);

×

1212

  return vgroupMemroy;

×

1215

static bool mndGetVnodeMemroyFp(SMnode *pMnode, void *pObj, void *p1, void *p2, void *p3) {

×

1216

  SVgObj  *pVgroup = pObj;

×

1217

  int32_t  dnodeId = *(int32_t *)p1;

×

1218

  int64_t *pVnodeMemory = (int64_t *)p2;

×

1220

  for (int32_t v = 0; v < pVgroup->replica; ++v) {

×

1221

    if (pVgroup->vnodeGid[v].dnodeId == dnodeId) {

×

1222

      *pVnodeMemory += mndGetVgroupMemory(pMnode, NULL, pVgroup);

×

1226

  return true;

×

1229

int64_t mndGetVnodesMemory(SMnode *pMnode, int32_t dnodeId) {

×

1230

  int64_t vnodeMemory = 0;

×

1231

  sdbTraverse(pMnode->pSdb, SDB_VGROUP, mndGetVnodeMemroyFp, &dnodeId, &vnodeMemory, NULL);

×

1232

  return vnodeMemory;

×

1235

static int32_t mndRetrieveVnodes(SRpcMsg *pReq, SShowObj *pShow, SSDataBlock *pBlock, int32_t rows) {

×

1236

  SMnode *pMnode = pReq->info.node;

×

1237

  SSdb   *pSdb = pMnode->pSdb;

×

1238

  int32_t numOfRows = 0;

×

1239

  SVgObj *pVgroup = NULL;

×

1240

  int32_t cols = 0;

×

1241

  int64_t curMs = taosGetTimestampMs();

×

1242

  int32_t code = 0;

×

1244

  while (numOfRows < rows - TSDB_MAX_REPLICA) {

×

1245

    pShow->pIter = sdbFetch(pSdb, SDB_VGROUP, pShow->pIter, (void **)&pVgroup);

×

1246

    if (pShow->pIter == NULL) break;

×

1248

    for (int32_t i = 0; i < pVgroup->replica && numOfRows < rows; ++i) {

×

1249

      SVnodeGid       *pGid = &pVgroup->vnodeGid[i];

×

1250

      SColumnInfoData *pColInfo = NULL;

×

1251

      cols = 0;

×

1253

      pColInfo = taosArrayGet(pBlock->pDataBlock, cols++);

×

1254

      code = colDataSetVal(pColInfo, numOfRows, (const char *)&pGid->dnodeId, false);

×

1255

      if (code != 0) {

×

1256

        mError("vgId:%d, failed to set dnodeId, since %s", pVgroup->vgId, tstrerror(code));

×

1257

        return code;

×

1259

      pColInfo = taosArrayGet(pBlock->pDataBlock, cols++);

×

1260

      code = colDataSetVal(pColInfo, numOfRows, (const char *)&pVgroup->vgId, false);

×

1261

      if (code != 0) {

×

1262

        mError("vgId:%d, failed to set vgId, since %s", pVgroup->vgId, tstrerror(code));

×

1263

        return code;

×

1267

      const char *dbname = mndGetDbStr(pVgroup->dbName);

×

1268

      char        b1[TSDB_DB_NAME_LEN + VARSTR_HEADER_SIZE] = {0};

×

1269

      if (dbname != NULL) {

×

1270

        STR_WITH_MAXSIZE_TO_VARSTR(b1, dbname, TSDB_DB_NAME_LEN + VARSTR_HEADER_SIZE);

×

1272

        STR_WITH_MAXSIZE_TO_VARSTR(b1, "NULL", TSDB_DB_NAME_LEN + VARSTR_HEADER_SIZE);

×

1274

      pColInfo = taosArrayGet(pBlock->pDataBlock, cols++);

×

1275

      code = colDataSetVal(pColInfo, numOfRows, (const char *)b1, false);

×

1276

      if (code != 0) {

×

1277

        mError("vgId:%d, failed to set dbName, since %s", pVgroup->vgId, tstrerror(code));

×

1278

        return code;

×

1282

      SDnodeObj *pDnode = mndAcquireDnode(pMnode, pGid->dnodeId);

×

1283

      if (pDnode == NULL) {

×

1284

        mError("failed to acquire dnode. dnodeId:%d", pGid->dnodeId);

×

1285

        break;

×

1287

      bool isDnodeOnline = mndIsDnodeOnline(pDnode, curMs);

×

1289

      char       buf[20] = {0};

×

1290

      ESyncState syncState = (isDnodeOnline) ? pGid->syncState : TAOS_SYNC_STATE_OFFLINE;

×

1291

      STR_TO_VARSTR(buf, syncStr(syncState));

×

1292

      pColInfo = taosArrayGet(pBlock->pDataBlock, cols++);

×

1293

      code = colDataSetVal(pColInfo, numOfRows, (const char *)buf, false);

×

1294

      if (code != 0) {

×

1295

        mError("vgId:%d, failed to set syncState, since %s", pVgroup->vgId, tstrerror(code));

×

1296

        return code;

×

1299

      int64_t roleTimeMs = (isDnodeOnline) ? pGid->roleTimeMs : 0;

×

1300

      pColInfo = taosArrayGet(pBlock->pDataBlock, cols++);

×

1301

      code = colDataSetVal(pColInfo, numOfRows, (const char *)&roleTimeMs, false);

×

1302

      if (code != 0) {

×

1303

        mError("vgId:%d, failed to set roleTimeMs, since %s", pVgroup->vgId, tstrerror(code));

×

1304

        return code;

×

1307

      int64_t startTimeMs = (isDnodeOnline) ? pGid->startTimeMs : 0;

×

1308

      pColInfo = taosArrayGet(pBlock->pDataBlock, cols++);

×

1309

      code = colDataSetVal(pColInfo, numOfRows, (const char *)&startTimeMs, false);

×

1310

      if (code != 0) {

×

1311

        mError("vgId:%d, failed to set startTimeMs, since %s", pVgroup->vgId, tstrerror(code));

×

1312

        return code;

×

1315

      pColInfo = taosArrayGet(pBlock->pDataBlock, cols++);

×

1316

      code = colDataSetVal(pColInfo, numOfRows, (const char *)&pGid->syncRestore, false);

×

1317

      if (code != 0) {

×

1318

        mError("vgId:%d, failed to set syncRestore, since %s", pVgroup->vgId, tstrerror(code));

×

1319

        return code;

×

1322

      numOfRows++;

×

1323

      sdbRelease(pSdb, pDnode);

×

1326

    sdbRelease(pSdb, pVgroup);

×

1329

  pShow->numOfRows += numOfRows;

×

1330

  return numOfRows;

×

1333

static void mndCancelGetNextVnode(SMnode *pMnode, void *pIter) {

×

1334

  SSdb *pSdb = pMnode->pSdb;

×

1335

  sdbCancelFetchByType(pSdb, pIter, SDB_VGROUP);

×

1336

×

1338

static int32_t mndAddVnodeToVgroup(SMnode *pMnode, STrans *pTrans, SVgObj *pVgroup, SArray *pArray) {

×

1339

  int32_t code = 0;

×

1340

  taosArraySort(pArray, (__compar_fn_t)mndCompareDnodeVnodes);

×

1341

  for (int32_t i = 0; i < taosArrayGetSize(pArray); ++i) {

×

1342

    SDnodeObj *pDnode = taosArrayGet(pArray, i);

×

1343

    mInfo("dnode:%d, equivalent vnodes:%d others:%d", pDnode->id, pDnode->numOfVnodes, pDnode->numOfOtherNodes);

×

1346

  SVnodeGid *pVgid = &pVgroup->vnodeGid[pVgroup->replica];

×

1347

  for (int32_t d = 0; d < taosArrayGetSize(pArray); ++d) {

×

1348

    SDnodeObj *pDnode = taosArrayGet(pArray, d);

×

1350

    bool used = false;

×

1351

    for (int32_t vn = 0; vn < pVgroup->replica; ++vn) {

×

1352

      if (pDnode->id == pVgroup->vnodeGid[vn].dnodeId) {

×

1353

        used = true;

×

1354

        break;

×

1357

    if (used) continue;

×

1359

    if (pDnode == NULL) {

×

1360

      TAOS_RETURN(TSDB_CODE_MND_NO_ENOUGH_DNODES);

×

1362

    if (pDnode->numOfVnodes >= pDnode->numOfSupportVnodes) {

×

1363

      TAOS_RETURN(TSDB_CODE_MND_NO_ENOUGH_VNODES);

×

1366

    int64_t vgMem = mndGetVgroupMemory(pMnode, NULL, pVgroup);

×

1367

    if (pDnode->memAvail - vgMem - pDnode->memUsed <= 0) {

×

1368

      mError("db:%s, vgId:%d, no enough memory:%" PRId64 " in dnode:%d avail:%" PRId64 " used:%" PRId64,

×

1370

      TAOS_RETURN(TSDB_CODE_MND_NO_ENOUGH_MEM_IN_DNODE);

×

1372

      pDnode->memUsed += vgMem;

×

1375

    pVgid->dnodeId = pDnode->id;

×

1376

    pVgid->syncState = TAOS_SYNC_STATE_OFFLINE;

×

1377

    mInfo("db:%s, vgId:%d, vn:%d is added, memory:%" PRId64 ", dnode:%d avail:%" PRId64 " used:%" PRId64,

×

1380

    pVgroup->replica++;

×

1381

    pDnode->numOfVnodes++;

×

1383

    SSdbRaw *pVgRaw = mndVgroupActionEncode(pVgroup);

×

1384

    if (pVgRaw == NULL) {

×

1385

      code = TSDB_CODE_MND_RETURN_VALUE_NULL;

×

1386

      if (terrno != 0) code = terrno;

×

1387

      TAOS_RETURN(code);

×

1389

    if ((code = mndTransAppendRedolog(pTrans, pVgRaw)) != 0) {

×

1390

      sdbFreeRaw(pVgRaw);

×

1391

      TAOS_RETURN(code);

×

1393

    code = sdbSetRawStatus(pVgRaw, SDB_STATUS_READY);

×

1394

    if (code != 0) {

×

1395

      mError("vgId:%d, failed to set raw status since %s at line:%d", pVgroup->vgId, tstrerror(code), __LINE__);

×

1397

    TAOS_RETURN(code);

×

1400

  code = TSDB_CODE_MND_NO_ENOUGH_DNODES;

×

1401

  mError("db:%s, failed to add vnode to vgId:%d since %s", pVgroup->dbName, pVgroup->vgId, tstrerror(code));

×

1402

  TAOS_RETURN(code);

×

1405

static int32_t mndRemoveVnodeFromVgroup(SMnode *pMnode, STrans *pTrans, SVgObj *pVgroup, SArray *pArray,

×

1407

  taosArraySort(pArray, (__compar_fn_t)mndCompareDnodeVnodes);

×

1408

  for (int32_t i = 0; i < taosArrayGetSize(pArray); ++i) {

×

1409

    SDnodeObj *pDnode = taosArrayGet(pArray, i);

×

1410

    mInfo("dnode:%d, equivalent vnodes:%d others:%d", pDnode->id, pDnode->numOfVnodes, pDnode->numOfOtherNodes);

×

1413

  int32_t code = -1;

×

1414

  for (int32_t d = taosArrayGetSize(pArray) - 1; d >= 0; --d) {

×

1415

    SDnodeObj *pDnode = taosArrayGet(pArray, d);

×

1417

    for (int32_t vn = 0; vn < pVgroup->replica; ++vn) {

×

1418

      SVnodeGid *pVgid = &pVgroup->vnodeGid[vn];

×

1419

      if (pVgid->dnodeId == pDnode->id) {

×

1420

        int64_t vgMem = mndGetVgroupMemory(pMnode, NULL, pVgroup);

×

1421

        pDnode->memUsed -= vgMem;

×

1422

        mInfo("db:%s, vgId:%d, vn:%d is removed, memory:%" PRId64 ", dnode:%d avail:%" PRId64 " used:%" PRId64,

×

1424

        pDnode->numOfVnodes--;

×

1425

        pVgroup->replica--;

×

1426

        *pDelVgid = *pVgid;

×

1427

        *pVgid = pVgroup->vnodeGid[pVgroup->replica];

×

1428

        memset(&pVgroup->vnodeGid[pVgroup->replica], 0, sizeof(SVnodeGid));

×

1429

        code = 0;

×

1430

        goto _OVER;

×

1435

_OVER:

×

1436

  if (code != 0) {

×

1437

    code = TSDB_CODE_APP_ERROR;

×

1438

    mError("db:%s, failed to remove vnode from vgId:%d since %s", pVgroup->dbName, pVgroup->vgId, tstrerror(code));

×

1439

    TAOS_RETURN(code);

×

1442

  for (int32_t vn = 0; vn < pVgroup->replica; ++vn) {

×

1443

    SVnodeGid *pVgid = &pVgroup->vnodeGid[vn];

×

1444

    mInfo("db:%s, vgId:%d, vn:%d dnode:%d is reserved", pVgroup->dbName, pVgroup->vgId, vn, pVgid->dnodeId);

×

1447

  SSdbRaw *pVgRaw = mndVgroupActionEncode(pVgroup);

×

1448

  if (pVgRaw == NULL) {

×

1449

    code = TSDB_CODE_MND_RETURN_VALUE_NULL;

×

1450

    if (terrno != 0) code = terrno;

×

1451

    TAOS_RETURN(code);

×

1453

  if (mndTransAppendRedolog(pTrans, pVgRaw) != 0) {

×

1454

    sdbFreeRaw(pVgRaw);

×

1455

    TAOS_RETURN(code);

×

1457

  code = sdbSetRawStatus(pVgRaw, SDB_STATUS_READY);

×

1458

  if (code != 0) {

×

1459

    mError("vgId:%d, failed to set raw status since %s at line:%d", pVgroup->vgId, tstrerror(code), __LINE__);

×

1462

  TAOS_RETURN(code);

×

1465

static int32_t mndRemoveVnodeFromVgroupWithoutSave(SMnode *pMnode, STrans *pTrans, SVgObj *pVgroup, SArray *pArray,

×

1467

  taosArraySort(pArray, (__compar_fn_t)mndCompareDnodeVnodes);

×

1468

  for (int32_t i = 0; i < taosArrayGetSize(pArray); ++i) {

×

1469

    SDnodeObj *pDnode = taosArrayGet(pArray, i);

×

1470

    mInfo("dnode:%d, equivalent vnodes:%d others:%d", pDnode->id, pDnode->numOfVnodes, pDnode->numOfOtherNodes);

×

1473

  int32_t code = -1;

×

1474

  for (int32_t d = taosArrayGetSize(pArray) - 1; d >= 0; --d) {

×

1475

    SDnodeObj *pDnode = taosArrayGet(pArray, d);

×

1477

    for (int32_t vn = 0; vn < pVgroup->replica; ++vn) {

×

1478

      SVnodeGid *pVgid = &pVgroup->vnodeGid[vn];

×

1479

      if (pVgid->dnodeId == pDnode->id) {

×

1480

        int64_t vgMem = mndGetVgroupMemory(pMnode, NULL, pVgroup);

×

1481

        pDnode->memUsed -= vgMem;

×

1482

        mInfo("db:%s, vgId:%d, vn:%d is removed, memory:%" PRId64 ", dnode:%d avail:%" PRId64 " used:%" PRId64,

×

1484

        pDnode->numOfVnodes--;

×

1485

        pVgroup->replica--;

×

1486

        *pDelVgid = *pVgid;

×

1487

        *pVgid = pVgroup->vnodeGid[pVgroup->replica];

×

1488

        memset(&pVgroup->vnodeGid[pVgroup->replica], 0, sizeof(SVnodeGid));

×

1489

        code = 0;

×

1490

        goto _OVER;

×

1495

_OVER:

×

1496

  if (code != 0) {

×

1497

    code = TSDB_CODE_APP_ERROR;

×

1498

    mError("db:%s, failed to remove vnode from vgId:%d since %s", pVgroup->dbName, pVgroup->vgId, tstrerror(code));

×

1499

    TAOS_RETURN(code);

×

1502

  for (int32_t vn = 0; vn < pVgroup->replica; ++vn) {

×

1503

    SVnodeGid *pVgid = &pVgroup->vnodeGid[vn];

×

1504

    mInfo("db:%s, vgId:%d, vn:%d dnode:%d is reserved", pVgroup->dbName, pVgroup->vgId, vn, pVgid->dnodeId);

×

1507

  TAOS_RETURN(code);

×

1510

int32_t mndAddCreateVnodeAction(SMnode *pMnode, STrans *pTrans, SDbObj *pDb, SVgObj *pVgroup, SVnodeGid *pVgid) {

×

1511

  int32_t      code = 0;

×

1512

  STransAction action = {0};

×

1514

  SDnodeObj *pDnode = mndAcquireDnode(pMnode, pVgid->dnodeId);

×

1515

  if (pDnode == NULL) return -1;

×

1516

  action.epSet = mndGetDnodeEpset(pDnode);

×

1517

  mndReleaseDnode(pMnode, pDnode);

×

1519

  int32_t contLen = 0;

×

1520

  void   *pReq = mndBuildCreateVnodeReq(pMnode, pDnode, pDb, pVgroup, &contLen);

×

1521

  if (pReq == NULL) return -1;

×

1523

  action.pCont = pReq;

×

1524

  action.contLen = contLen;

×

1525

  action.msgType = TDMT_DND_CREATE_VNODE;

×

1526

  action.acceptableCode = TSDB_CODE_VND_ALREADY_EXIST;

×

1528

  if ((code = mndTransAppendRedoAction(pTrans, &action)) != 0) {

×

1529

    taosMemoryFree(pReq);

×

1530

    TAOS_RETURN(code);

×

1533

  TAOS_RETURN(code);

×

1536

int32_t mndRestoreAddCreateVnodeAction(SMnode *pMnode, STrans *pTrans, SDbObj *pDb, SVgObj *pVgroup,

×

1538

  int32_t      code = 0;

×

1539

  STransAction action = {0};

×

1541

  action.epSet = mndGetDnodeEpset(pDnode);

×

1543

  int32_t contLen = 0;

×

1544

  void   *pReq = mndBuildCreateVnodeReq(pMnode, pDnode, pDb, pVgroup, &contLen);

×

1545

  if (pReq == NULL) {

×

1546

    code = TSDB_CODE_MND_RETURN_VALUE_NULL;

×

1547

    if (terrno != 0) code = terrno;

×

1548

    TAOS_RETURN(code);

×

1551

  action.pCont = pReq;

×

1552

  action.contLen = contLen;

×

1553

  action.msgType = TDMT_DND_CREATE_VNODE;

×

1554

  action.acceptableCode = TSDB_CODE_VND_ALREADY_EXIST;

×

1556

  if ((code = mndTransAppendRedoAction(pTrans, &action)) != 0) {

×

1557

    taosMemoryFree(pReq);

×

1558

    TAOS_RETURN(code);

×

1561

  TAOS_RETURN(code);

×

1564

int32_t mndAddAlterVnodeConfirmAction(SMnode *pMnode, STrans *pTrans, SDbObj *pDb, SVgObj *pVgroup) {

×

1565

  int32_t      code = 0;

×

1566

  STransAction action = {0};

×

1567

  action.epSet = mndGetVgroupEpset(pMnode, pVgroup);

×

1569

  mInfo("vgId:%d, build alter vnode confirm req", pVgroup->vgId);

×

1570

  int32_t   contLen = sizeof(SMsgHead);

×

1571

  SMsgHead *pHead = taosMemoryMalloc(contLen);

×

1572

  if (pHead == NULL) {

×

1573

    TAOS_RETURN(terrno);

×

1576

  pHead->contLen = htonl(contLen);

×

1577

  pHead->vgId = htonl(pVgroup->vgId);

×

1579

  action.pCont = pHead;

×

1580

  action.contLen = contLen;

×

1581

  action.msgType = TDMT_VND_ALTER_CONFIRM;

×

1583

  action.retryCode = TSDB_CODE_VND_INVALID_VGROUP_ID;

×

1585

  if ((code = mndTransAppendRedoAction(pTrans, &action)) != 0) {

×

1586

    taosMemoryFree(pHead);

×

1587

    TAOS_RETURN(code);

×

1590

  TAOS_RETURN(code);

×

1593

int32_t mndAddChangeConfigAction(SMnode *pMnode, STrans *pTrans, SDbObj *pDb, SVgObj *pOldVgroup, SVgObj *pNewVgroup,

×

1595

  int32_t      code = 0;

×

1596

  STransAction action = {0};

×

1597

  action.epSet = mndGetVgroupEpset(pMnode, pNewVgroup);

×

1599

  int32_t contLen = 0;

×

1600

  void   *pReq = mndBuildAlterVnodeReplicaReq(pMnode, pDb, pNewVgroup, dnodeId, &contLen);

×

1601

  if (pReq == NULL) {

×

1602

    code = TSDB_CODE_MND_RETURN_VALUE_NULL;

×

1603

    if (terrno != 0) code = terrno;

×

1604

    TAOS_RETURN(code);

×

1607

  int32_t totallen = contLen + sizeof(SMsgHead);

×

1609

  SMsgHead *pHead = taosMemoryMalloc(totallen);

×

1610

  if (pHead == NULL) {

×

1611

    taosMemoryFree(pReq);

×

1612

    TAOS_RETURN(terrno);

×

1615

  pHead->contLen = htonl(totallen);

×

1616

  pHead->vgId = htonl(pNewVgroup->vgId);

×

1618

  memcpy((void *)(pHead + 1), pReq, contLen);

×

1619

  taosMemoryFree(pReq);

×

1621

  action.pCont = pHead;

×

1622

  action.contLen = totallen;

×

1623

  action.msgType = TDMT_SYNC_CONFIG_CHANGE;

×

1625

  if ((code = mndTransAppendRedoAction(pTrans, &action)) != 0) {

×

1626

    taosMemoryFree(pHead);

×

1627

    TAOS_RETURN(code);

×

1630

  TAOS_RETURN(code);

×

1633

static int32_t mndAddAlterVnodeHashRangeAction(SMnode *pMnode, STrans *pTrans, int32_t srcVgId, SVgObj *pVgroup) {

×

1634

  int32_t      code = 0;

×

1635

  STransAction action = {0};

×

1636

  action.epSet = mndGetVgroupEpset(pMnode, pVgroup);

×

1638

  int32_t contLen = 0;

×

1639

  void   *pReq = mndBuildAlterVnodeHashRangeReq(pMnode, srcVgId, pVgroup, &contLen);

×

1640

  if (pReq == NULL) {

×

1641

    code = TSDB_CODE_MND_RETURN_VALUE_NULL;

×

1642

    if (terrno != 0) code = terrno;

×

1643

    TAOS_RETURN(code);

×

1646

  action.pCont = pReq;

×

1647

  action.contLen = contLen;

×

1648

  action.msgType = TDMT_VND_ALTER_HASHRANGE;

×

1649

  action.acceptableCode = TSDB_CODE_VND_ALREADY_EXIST;

×

1651

  if ((code = mndTransAppendRedoAction(pTrans, &action)) != 0) {

×

1652

    taosMemoryFree(pReq);

×

1653

    TAOS_RETURN(code);

×

1656

  mInfo("trans:%d, add alter vnode hash range action for from vgId:%d to vgId:%d", pTrans->id, srcVgId, pVgroup->vgId);

×

1657

  TAOS_RETURN(code);

×

1660

int32_t mndAddAlterVnodeConfigAction(SMnode *pMnode, STrans *pTrans, SDbObj *pDb, SVgObj *pVgroup) {

×

1661

  int32_t      code = 0;

×

1662

  STransAction action = {0};

×

1663

  action.epSet = mndGetVgroupEpset(pMnode, pVgroup);

×

1665

  int32_t contLen = 0;

×

1666

  void   *pReq = mndBuildAlterVnodeConfigReq(pMnode, pDb, pVgroup, &contLen);

×

1667

  if (pReq == NULL) {

×

1668

    code = TSDB_CODE_MND_RETURN_VALUE_NULL;

×

1669

    if (terrno != 0) code = terrno;

×

1670

    TAOS_RETURN(code);

×

1673

  action.pCont = pReq;

×

1674

  action.contLen = contLen;

×

1675

  action.msgType = TDMT_VND_ALTER_CONFIG;

×

1677

  if ((code = mndTransAppendRedoAction(pTrans, &action)) != 0) {

×

1678

    taosMemoryFree(pReq);

×

1679

    TAOS_RETURN(code);

×

1682

  TAOS_RETURN(code);

×

1685

int32_t mndAddNewVgPrepareAction(SMnode *pMnode, STrans *pTrans, SVgObj *pVg) {

×

1686

  int32_t  code = 0;

×

1687

  SSdbRaw *pRaw = mndVgroupActionEncode(pVg);

×

1688

  if (pRaw == NULL) {

×

1689

    code = TSDB_CODE_MND_RETURN_VALUE_NULL;

×

1690

    if (terrno != 0) code = terrno;

×

1691

    goto _err;

×

1694

  TAOS_CHECK_GOTO(mndTransAppendPrepareLog(pTrans, pRaw), NULL, _err);

×

1695

  if (sdbSetRawStatus(pRaw, SDB_STATUS_CREATING) != 0) {

×

1696

    mError("vgId:%d, failed to set raw status at line:%d", pVg->vgId, __LINE__);

×

1698

  if (code != 0) {

×

1699

    mError("vgId:%d, failed to set raw status since %s at line:%d", pVg->vgId, tstrerror(code), __LINE__);

×

1700

    TAOS_RETURN(code);

×

1702

  pRaw = NULL;

×

1703

  TAOS_RETURN(code);

×

1705

_err:

×

1706

  sdbFreeRaw(pRaw);

×

1707

  TAOS_RETURN(code);

×

1710

int32_t mndAddAlterVnodeReplicaAction(SMnode *pMnode, STrans *pTrans, SDbObj *pDb, SVgObj *pVgroup, int32_t dnodeId) {

×

1711

  int32_t    code = 0;

×

1712

  SDnodeObj *pDnode = mndAcquireDnode(pMnode, dnodeId);

×

1713

  if (pDnode == NULL) {

×

1714

    code = TSDB_CODE_MND_RETURN_VALUE_NULL;

×

1715

    if (terrno != 0) code = terrno;

×

1716

    TAOS_RETURN(code);

×

1719

  STransAction action = {0};

×

1720

  action.epSet = mndGetDnodeEpset(pDnode);

×

1721

  mndReleaseDnode(pMnode, pDnode);

×

1723

  int32_t contLen = 0;

×

1724

  void   *pReq = mndBuildAlterVnodeReplicaReq(pMnode, pDb, pVgroup, dnodeId, &contLen);

×

1725

  if (pReq == NULL) {

×

1726

    code = TSDB_CODE_MND_RETURN_VALUE_NULL;

×

1727

    if (terrno != 0) code = terrno;

×

1728

    TAOS_RETURN(code);

×

1731

  action.pCont = pReq;

×

1732

  action.contLen = contLen;

×

1733

  action.msgType = TDMT_VND_ALTER_REPLICA;

×

1735

  if ((code = mndTransAppendRedoAction(pTrans, &action)) != 0) {

×

1736

    taosMemoryFree(pReq);

×

1737

    TAOS_RETURN(code);

×

1740

  TAOS_RETURN(code);

×

1743

int32_t mndAddCheckLearnerCatchupAction(SMnode *pMnode, STrans *pTrans, SDbObj *pDb, SVgObj *pVgroup, int32_t dnodeId) {

×

1744

  int32_t    code = 0;

×

1745

  SDnodeObj *pDnode = mndAcquireDnode(pMnode, dnodeId);

×

1746

  if (pDnode == NULL) {

×

1747

    code = TSDB_CODE_MND_RETURN_VALUE_NULL;

×

1748

    if (terrno != 0) code = terrno;

×

1749

    TAOS_RETURN(code);

×

1752

  STransAction action = {0};

×

1753

  action.epSet = mndGetDnodeEpset(pDnode);

×

1754

  mndReleaseDnode(pMnode, pDnode);

×

1756

  int32_t contLen = 0;

×

1757

  void   *pReq = mndBuildCheckLearnCatchupReq(pMnode, pDb, pVgroup, dnodeId, &contLen);

×

1758

  if (pReq == NULL) {

×

1759

    code = TSDB_CODE_MND_RETURN_VALUE_NULL;

×

1760

    if (terrno != 0) code = terrno;

×

1761

    TAOS_RETURN(code);

×

1764

  action.pCont = pReq;

×

1765

  action.contLen = contLen;

×

1766

  action.msgType = TDMT_DND_CHECK_VNODE_LEARNER_CATCHUP;

×

1767

  action.acceptableCode = TSDB_CODE_VND_ALREADY_IS_VOTER;

×

1768

  action.retryCode = TSDB_CODE_VND_NOT_CATCH_UP;

×

1770

  if ((code = mndTransAppendRedoAction(pTrans, &action)) != 0) {

×

1771

    taosMemoryFree(pReq);

×

1772

    TAOS_RETURN(code);

×

1775

  TAOS_RETURN(code);

×

1778

int32_t mndAddAlterVnodeTypeAction(SMnode *pMnode, STrans *pTrans, SDbObj *pDb, SVgObj *pVgroup, int32_t dnodeId) {

×

1779

  int32_t    code = 0;

×

1780

  SDnodeObj *pDnode = mndAcquireDnode(pMnode, dnodeId);

×

1781

  if (pDnode == NULL) {

×

1782

    code = TSDB_CODE_MND_RETURN_VALUE_NULL;

×

1783

    if (terrno != 0) code = terrno;

×

1784

    TAOS_RETURN(code);

×

1787

  STransAction action = {0};

×

1788

  action.epSet = mndGetDnodeEpset(pDnode);

×

1789

  mndReleaseDnode(pMnode, pDnode);

×

1791

  int32_t contLen = 0;

×

1792

  void   *pReq = mndBuildAlterVnodeReplicaReq(pMnode, pDb, pVgroup, dnodeId, &contLen);

×

1793

  if (pReq == NULL) {

×

1794

    code = TSDB_CODE_MND_RETURN_VALUE_NULL;

×

1795

    if (terrno != 0) code = terrno;

×

1796

    TAOS_RETURN(code);

×

1799

  action.pCont = pReq;

×

1800

  action.contLen = contLen;

×

1801

  action.msgType = TDMT_DND_ALTER_VNODE_TYPE;

×

1802

  action.acceptableCode = TSDB_CODE_VND_ALREADY_IS_VOTER;

×

1803

  action.retryCode = TSDB_CODE_VND_NOT_CATCH_UP;

×

1805

  if ((code = mndTransAppendRedoAction(pTrans, &action)) != 0) {

×

1806

    taosMemoryFree(pReq);

×

1807

    TAOS_RETURN(code);

×

1810

  TAOS_RETURN(code);

×

1813

int32_t mndRestoreAddAlterVnodeTypeAction(SMnode *pMnode, STrans *pTrans, SDbObj *pDb, SVgObj *pVgroup,

×

1815

  int32_t      code = 0;

×

1816

  STransAction action = {0};

×

1817

  action.epSet = mndGetDnodeEpset(pDnode);

×

1819

  int32_t contLen = 0;

×

1820

  void   *pReq = mndBuildAlterVnodeReplicaReq(pMnode, pDb, pVgroup, pDnode->id, &contLen);

×

1821

  if (pReq == NULL) {

×

1822

    code = TSDB_CODE_MND_RETURN_VALUE_NULL;

×

1823

    if (terrno != 0) code = terrno;

×

1824

    TAOS_RETURN(code);

×

1827

  action.pCont = pReq;

×

1828

  action.contLen = contLen;

×

1829

  action.msgType = TDMT_DND_ALTER_VNODE_TYPE;

×

1830

  action.acceptableCode = TSDB_CODE_VND_ALREADY_IS_VOTER;

×

1831

  action.retryCode = TSDB_CODE_VND_NOT_CATCH_UP;

×

1833

  if ((code = mndTransAppendRedoAction(pTrans, &action)) != 0) {

×

1834

    taosMemoryFree(pReq);

×

1835

    TAOS_RETURN(code);

×

1838

  TAOS_RETURN(code);

×

1841

static int32_t mndAddDisableVnodeWriteAction(SMnode *pMnode, STrans *pTrans, SDbObj *pDb, SVgObj *pVgroup,

×

1843

  int32_t    code = 0;

×

1844

  SDnodeObj *pDnode = mndAcquireDnode(pMnode, dnodeId);

×

1845

  if (pDnode == NULL) {

×

1846

    code = TSDB_CODE_MND_RETURN_VALUE_NULL;

×

1847

    if (terrno != 0) code = terrno;

×

1848

    TAOS_RETURN(code);

×

1851

  STransAction action = {0};

×

1852

  action.epSet = mndGetDnodeEpset(pDnode);

×

1853

  mndReleaseDnode(pMnode, pDnode);

×

1855

  int32_t contLen = 0;

×

1856

  void   *pReq = mndBuildDisableVnodeWriteReq(pMnode, pDb, pVgroup->vgId, &contLen);

×

1857

  if (pReq == NULL) {

×

1858

    code = TSDB_CODE_MND_RETURN_VALUE_NULL;

×

1859

    if (terrno != 0) code = terrno;

×

1860

    TAOS_RETURN(code);

×

1863

  action.pCont = pReq;

×

1864

  action.contLen = contLen;

×

1865

  action.msgType = TDMT_VND_DISABLE_WRITE;

×

1867

  if ((code = mndTransAppendRedoAction(pTrans, &action)) != 0) {

×

1868

    taosMemoryFree(pReq);

×

1869

    TAOS_RETURN(code);

×

1872

  TAOS_RETURN(code);

×

1875

int32_t mndAddDropVnodeAction(SMnode *pMnode, STrans *pTrans, SDbObj *pDb, SVgObj *pVgroup, SVnodeGid *pVgid,

×

1877

  int32_t      code = 0;

×

1878

  STransAction action = {0};

×

1880

  SDnodeObj *pDnode = mndAcquireDnode(pMnode, pVgid->dnodeId);

×

1881

  if (pDnode == NULL) {

×

1882

    code = TSDB_CODE_MND_RETURN_VALUE_NULL;

×

1883

    if (terrno != 0) code = terrno;

×

1884

    TAOS_RETURN(code);

×

1886

  action.epSet = mndGetDnodeEpset(pDnode);

×

1887

  mndReleaseDnode(pMnode, pDnode);

×

1889

  int32_t contLen = 0;

×

1890

  void   *pReq = mndBuildDropVnodeReq(pMnode, pDnode, pDb, pVgroup, &contLen);

×

1891

  if (pReq == NULL) {

×

1892

    code = TSDB_CODE_MND_RETURN_VALUE_NULL;

×

1893

    if (terrno != 0) code = terrno;

×

1894

    TAOS_RETURN(code);

×

1897

  action.pCont = pReq;

×

1898

  action.contLen = contLen;

×

1899

  action.msgType = TDMT_DND_DROP_VNODE;

×

1900

  action.acceptableCode = TSDB_CODE_VND_NOT_EXIST;

×

1902

  if (isRedo) {

×

1903

    if ((code = mndTransAppendRedoAction(pTrans, &action)) != 0) {

×

1904

      taosMemoryFree(pReq);

×

1905

      TAOS_RETURN(code);

×

1908

    if ((code = mndTransAppendUndoAction(pTrans, &action)) != 0) {

×

1909

      taosMemoryFree(pReq);

×

1910

      TAOS_RETURN(code);

×

1914

  TAOS_RETURN(code);

×

1917

int32_t mndSetMoveVgroupInfoToTrans(SMnode *pMnode, STrans *pTrans, SDbObj *pDb, SVgObj *pVgroup, int32_t vnIndex,

×

1919

  int32_t code = 0;

×

1920

  SVgObj  newVg = {0};

×

1921

  memcpy(&newVg, pVgroup, sizeof(SVgObj));

×

1923

  mInfo("vgId:%d, vgroup info before move, replica:%d", newVg.vgId, newVg.replica);

×

1924

  for (int32_t i = 0; i < newVg.replica; ++i) {

×

1925

    mInfo("vgId:%d, vnode:%d dnode:%d", newVg.vgId, i, newVg.vnodeGid[i].dnodeId);

×

1928

  if (!force) {

×

1934

      mInfo("vgId:%d, will add 1 vnode, replca:%d", pVgroup->vgId, newVg.replica);

×

1935

      TAOS_CHECK_RETURN(mndAddVnodeToVgroup(pMnode, pTrans, &newVg, pArray));

×

1936

      for (int32_t i = 0; i < newVg.replica - 1; ++i) {

×

1937

        TAOS_CHECK_RETURN(mndAddAlterVnodeReplicaAction(pMnode, pTrans, pDb, &newVg, newVg.vnodeGid[i].dnodeId));

×

1939

      TAOS_CHECK_RETURN(mndAddCreateVnodeAction(pMnode, pTrans, pDb, &newVg, &newVg.vnodeGid[newVg.replica - 1]));

×

1940

      TAOS_CHECK_RETURN(mndAddAlterVnodeConfirmAction(pMnode, pTrans, pDb, &newVg));

×

1942

      mInfo("vgId:%d, will remove 1 vnode, replca:2", pVgroup->vgId);

×

1943

      newVg.replica--;

×

1944

      SVnodeGid del = newVg.vnodeGid[vnIndex];

×

1945

      newVg.vnodeGid[vnIndex] = newVg.vnodeGid[newVg.replica];

×

1946

      memset(&newVg.vnodeGid[newVg.replica], 0, sizeof(SVnodeGid));

×

1948

        SSdbRaw *pRaw = mndVgroupActionEncode(&newVg);

×

1949

        if (pRaw == NULL) {

×

1950

          code = TSDB_CODE_MND_RETURN_VALUE_NULL;

×

1951

          if (terrno != 0) code = terrno;

×

1952

          TAOS_RETURN(code);

×

1954

        if ((code = mndTransAppendRedolog(pTrans, pRaw)) != 0) {

×

1955

          sdbFreeRaw(pRaw);

×

1956

          TAOS_RETURN(code);

×

1958

        code = sdbSetRawStatus(pRaw, SDB_STATUS_READY);

×

1959

        if (code != 0) {

×

1960

          mError("vgId:%d, failed to set raw status since %s at line:%d", newVg.vgId, tstrerror(code), __LINE__);

×

1961

          return code;

×

1965

      TAOS_CHECK_RETURN(mndAddDropVnodeAction(pMnode, pTrans, pDb, &newVg, &del, true));

×

1966

      for (int32_t i = 0; i < newVg.replica; ++i) {

×

1967

        TAOS_CHECK_RETURN(mndAddAlterVnodeReplicaAction(pMnode, pTrans, pDb, &newVg, newVg.vnodeGid[i].dnodeId));

×

1969

      TAOS_CHECK_RETURN(mndAddAlterVnodeConfirmAction(pMnode, pTrans, pDb, &newVg));

×

2000

    mInfo("vgId:%d, will add 1 vnode and force remove 1 vnode", pVgroup->vgId);

×

2001

    TAOS_CHECK_RETURN(mndAddVnodeToVgroup(pMnode, pTrans, &newVg, pArray));

×

2002

    newVg.replica--;

×

2004

    newVg.vnodeGid[vnIndex] = newVg.vnodeGid[newVg.replica];

×

2005

    memset(&newVg.vnodeGid[newVg.replica], 0, sizeof(SVnodeGid));

×

2007

      SSdbRaw *pRaw = mndVgroupActionEncode(&newVg);

×

2008

      if (pRaw == NULL) {

×

2009

        code = TSDB_CODE_MND_RETURN_VALUE_NULL;

×

2010

        if (terrno != 0) code = terrno;

×

2011

        TAOS_RETURN(code);

×

2013

      if ((code = mndTransAppendRedolog(pTrans, pRaw)) != 0) {

×

2014

        sdbFreeRaw(pRaw);

×

2015

        TAOS_RETURN(code);

×

2017

      code = sdbSetRawStatus(pRaw, SDB_STATUS_READY);

×

2018

      if (code != 0) {

×

2019

        mError("vgId:%d, failed to set raw status since %s at line:%d", newVg.vgId, tstrerror(code), __LINE__);

×

2020

        return code;

×

2024

    for (int32_t i = 0; i < newVg.replica; ++i) {

×

2025

      if (i != vnIndex) {

×

2026

        TAOS_CHECK_RETURN(mndAddAlterVnodeReplicaAction(pMnode, pTrans, pDb, &newVg, newVg.vnodeGid[i].dnodeId));

×

2029

    TAOS_CHECK_RETURN(mndAddCreateVnodeAction(pMnode, pTrans, pDb, &newVg, &newVg.vnodeGid[vnIndex]));

×

2030

    TAOS_CHECK_RETURN(mndAddAlterVnodeConfirmAction(pMnode, pTrans, pDb, &newVg));

×

2032

    if (newVg.replica == 1) {

×

2033

      if (force && !unsafe) {

×

2034

        TAOS_RETURN(TSDB_CODE_VND_META_DATA_UNSAFE_DELETE);

×

2037

      SSdb *pSdb = pMnode->pSdb;

×

2038

      void *pIter = NULL;

×

2040

      while (1) {

×

2041

        SStbObj *pStb = NULL;

×

2042

        pIter = sdbFetch(pSdb, SDB_STB, pIter, (void **)&pStb);

×

2043

        if (pIter == NULL) break;

×

2045

        if (strcmp(pStb->db, pDb->name) == 0) {

×

2046

          if ((code = mndSetForceDropCreateStbRedoActions(pMnode, pTrans, &newVg, pStb)) != 0) {

×

2047

            sdbCancelFetch(pSdb, pIter);

×

2048

            sdbRelease(pSdb, pStb);

×

2049

            TAOS_RETURN(code);

×

2053

        sdbRelease(pSdb, pStb);

×

2056

      mInfo("vgId:%d, all data is dropped since replica=1", pVgroup->vgId);

×

2061

    SSdbRaw *pRaw = mndVgroupActionEncode(&newVg);

×

2062

    if (pRaw == NULL) {

×

2063

      code = TSDB_CODE_MND_RETURN_VALUE_NULL;

×

2064

      if (terrno != 0) code = terrno;

×

2065

      TAOS_RETURN(code);

×

2067

    if ((code = mndTransAppendCommitlog(pTrans, pRaw)) != 0) {

×

2068

      sdbFreeRaw(pRaw);

×

2069

      TAOS_RETURN(code);

×

2071

    code = sdbSetRawStatus(pRaw, SDB_STATUS_READY);

×

2072

    if (code != 0) {

×

2073

      mError("vgId:%d, failed to set raw status since %s at line:%d", newVg.vgId, tstrerror(code), __LINE__);

×

2074

      return code;

×

2078

  mInfo("vgId:%d, vgroup info after move, replica:%d", newVg.vgId, newVg.replica);

×

2079

  for (int32_t i = 0; i < newVg.replica; ++i) {

×

2080

    mInfo("vgId:%d, vnode:%d dnode:%d", newVg.vgId, i, newVg.vnodeGid[i].dnodeId);

×

2082

  TAOS_RETURN(code);

×

2085

int32_t mndSetMoveVgroupsInfoToTrans(SMnode *pMnode, STrans *pTrans, int32_t delDnodeId, bool force, bool unsafe) {

×

2086

  int32_t code = 0;

×

2087

  SArray *pArray = mndBuildDnodesArray(pMnode, delDnodeId, NULL);

×

2088

  if (pArray == NULL) {

×

2089

    code = TSDB_CODE_MND_RETURN_VALUE_NULL;

×

2090

    if (terrno != 0) code = terrno;

×

2091

    TAOS_RETURN(code);

×

2094

  void *pIter = NULL;

×

2095

  while (1) {

×

2096

    SVgObj *pVgroup = NULL;

×

2097

    pIter = sdbFetch(pMnode->pSdb, SDB_VGROUP, pIter, (void **)&pVgroup);

×

2098

    if (pIter == NULL) break;

×

2100

    int32_t vnIndex = -1;

×

2101

    for (int32_t i = 0; i < pVgroup->replica; ++i) {

×

2102

      if (pVgroup->vnodeGid[i].dnodeId == delDnodeId) {

×

2103

        vnIndex = i;

×

2104

        break;

×

2108

    code = 0;

×

2109

    if (vnIndex != -1) {

×

2110

      mInfo("vgId:%d, vnode:%d will be removed from dnode:%d, force:%d", pVgroup->vgId, vnIndex, delDnodeId, force);

×

2111

      SDbObj *pDb = mndAcquireDb(pMnode, pVgroup->dbName);

×

2112

      code = mndSetMoveVgroupInfoToTrans(pMnode, pTrans, pDb, pVgroup, vnIndex, pArray, force, unsafe);

×

2113

      mndReleaseDb(pMnode, pDb);

×

2116

    sdbRelease(pMnode->pSdb, pVgroup);

×

2118

    if (code != 0) {

×

2119

      sdbCancelFetch(pMnode->pSdb, pIter);

×

2120

      break;

×

2124

  taosArrayDestroy(pArray);

×

2125

  TAOS_RETURN(code);

×

2128

static int32_t mndAddIncVgroupReplicaToTrans(SMnode *pMnode, STrans *pTrans, SDbObj *pDb, SVgObj *pVgroup,

×

2130

  int32_t code = 0;

×

2131

  mInfo("vgId:%d, will add 1 vnode, replica:%d dnode:%d", pVgroup->vgId, pVgroup->replica, newDnodeId);

×

2134

  SVnodeGid *pGid = &pVgroup->vnodeGid[pVgroup->replica];

×

2135

  pVgroup->replica++;

×

2136

  pGid->dnodeId = newDnodeId;

×

2137

  pGid->syncState = TAOS_SYNC_STATE_OFFLINE;

×

2138

  pGid->nodeRole = TAOS_SYNC_ROLE_LEARNER;

×

2140

  SSdbRaw *pVgRaw = mndVgroupActionEncode(pVgroup);

×

2141

  if (pVgRaw == NULL) {

×

2142

    code = TSDB_CODE_MND_RETURN_VALUE_NULL;

×

2143

    if (terrno != 0) code = terrno;

×

2144

    TAOS_RETURN(code);

×

2146

  if ((code = mndTransAppendRedolog(pTrans, pVgRaw)) != 0) {

×

2147

    sdbFreeRaw(pVgRaw);

×

2148

    TAOS_RETURN(code);

×

2150

  code = sdbSetRawStatus(pVgRaw, SDB_STATUS_READY);

×

2151

  if (code != 0) {

×

2152

    mError("vgId:%d, failed to set raw status since %s at line:%d", pVgroup->vgId, tstrerror(code), __LINE__);

×

2153

    TAOS_RETURN(code);

×

2157

  for (int32_t i = 0; i < pVgroup->replica - 1; ++i) {

×

2158

    TAOS_CHECK_RETURN(mndAddAlterVnodeReplicaAction(pMnode, pTrans, pDb, pVgroup, pVgroup->vnodeGid[i].dnodeId));

×

2160

  TAOS_CHECK_RETURN(mndAddCreateVnodeAction(pMnode, pTrans, pDb, pVgroup, pGid));

×

2163

  pGid->nodeRole = TAOS_SYNC_ROLE_VOTER;

×

2164

  TAOS_CHECK_RETURN(mndAddAlterVnodeTypeAction(pMnode, pTrans, pDb, pVgroup, pGid->dnodeId));

×

2165

  for (int32_t i = 0; i < pVgroup->replica - 1; ++i) {

×

2166

    TAOS_CHECK_RETURN(mndAddAlterVnodeReplicaAction(pMnode, pTrans, pDb, pVgroup, pVgroup->vnodeGid[i].dnodeId));

×

2170

  TAOS_CHECK_RETURN(mndAddAlterVnodeConfirmAction(pMnode, pTrans, pDb, pVgroup));

×

2172

  TAOS_RETURN(code);

×

2175

static int32_t mndAddDecVgroupReplicaFromTrans(SMnode *pMnode, STrans *pTrans, SDbObj *pDb, SVgObj *pVgroup,

×

2177

  int32_t code = 0;

×

2178

  mInfo("vgId:%d, will remove 1 vnode, replica:%d dnode:%d", pVgroup->vgId, pVgroup->replica, delDnodeId);

×

2180

  SVnodeGid *pGid = NULL;

×

2181

  SVnodeGid  delGid = {0};

×

2182

  for (int32_t i = 0; i < pVgroup->replica; ++i) {

×

2183

    if (pVgroup->vnodeGid[i].dnodeId == delDnodeId) {

×

2184

      pGid = &pVgroup->vnodeGid[i];

×

2185

      break;

×

2189

  if (pGid == NULL) return 0;

×

2191

  pVgroup->replica--;

×

2192

  memcpy(&delGid, pGid, sizeof(SVnodeGid));

×

2193

  memcpy(pGid, &pVgroup->vnodeGid[pVgroup->replica], sizeof(SVnodeGid));

×

2194

  memset(&pVgroup->vnodeGid[pVgroup->replica], 0, sizeof(SVnodeGid));

×

2196

  SSdbRaw *pVgRaw = mndVgroupActionEncode(pVgroup);

×

2197

  if (pVgRaw == NULL) {

×

2198

    code = TSDB_CODE_MND_RETURN_VALUE_NULL;

×

2199

    if (terrno != 0) code = terrno;

×

2200

    TAOS_RETURN(code);

×

2202

  if ((code = mndTransAppendRedolog(pTrans, pVgRaw)) != 0) {

×

2203

    sdbFreeRaw(pVgRaw);

×

2204

    TAOS_RETURN(code);

×

2206

  code = sdbSetRawStatus(pVgRaw, SDB_STATUS_READY);

×

2207

  if (code != 0) {

×

2208

    mError("vgId:%d, failed to set raw status since %s at line:%d", pVgroup->vgId, tstrerror(code), __LINE__);

×

2209

    TAOS_RETURN(code);

×

2212

  TAOS_CHECK_RETURN(mndAddDropVnodeAction(pMnode, pTrans, pDb, pVgroup, &delGid, true));

×

2213

  for (int32_t i = 0; i < pVgroup->replica; ++i) {

×

2214

    TAOS_CHECK_RETURN(mndAddAlterVnodeReplicaAction(pMnode, pTrans, pDb, pVgroup, pVgroup->vnodeGid[i].dnodeId));

×

2216

  TAOS_CHECK_RETURN(mndAddAlterVnodeConfirmAction(pMnode, pTrans, pDb, pVgroup));

×

2218

  TAOS_RETURN(code);

×

2221

static int32_t mndRedistributeVgroup(SMnode *pMnode, SRpcMsg *pReq, SDbObj *pDb, SVgObj *pVgroup, SDnodeObj *pNew1,

×

2224

  int32_t code = -1;

×

2225

  STrans *pTrans = NULL;

×

2227

  pTrans = mndTransCreate(pMnode, TRN_POLICY_RETRY, TRN_CONFLICT_GLOBAL, pReq, "red-vgroup");

×

2228

  if (pTrans == NULL) {

×

2229

    code = TSDB_CODE_MND_RETURN_VALUE_NULL;

×

2230

    if (terrno != 0) code = terrno;

×

2231

    goto _OVER;

×

2234

  mndTransSetDbName(pTrans, pVgroup->dbName, NULL);

×

2235

  TAOS_CHECK_GOTO(mndTransCheckConflictWithCompact(pMnode, pTrans), NULL, _OVER);

×

2237

  mndTransSetSerial(pTrans);

×

2238

  mInfo("trans:%d, used to redistribute vgroup, vgId:%d", pTrans->id, pVgroup->vgId);

×

2240

  SVgObj newVg = {0};

×

2241

  memcpy(&newVg, pVgroup, sizeof(SVgObj));

×

2242

  mInfo("vgId:%d, vgroup info before redistribute, replica:%d", newVg.vgId, newVg.replica);

×

2243

  for (int32_t i = 0; i < newVg.replica; ++i) {

×

2244

    mInfo("vgId:%d, vnode:%d dnode:%d role:%s", newVg.vgId, i, newVg.vnodeGid[i].dnodeId,

×

2248

  if (pNew1 != NULL && pOld1 != NULL) {

×

2249

    int32_t numOfVnodes = mndGetVnodesNum(pMnode, pNew1->id);

×

2250

    if (numOfVnodes >= pNew1->numOfSupportVnodes) {

×

2251

      mError("vgId:%d, no enough vnodes in dnode:%d, numOfVnodes:%d support:%d", newVg.vgId, pNew1->id, numOfVnodes,

×

2253

      code = TSDB_CODE_MND_NO_ENOUGH_VNODES;

×

2254

      goto _OVER;

×

2257

    int64_t vgMem = mndGetVgroupMemory(pMnode, NULL, pVgroup);

×

2258

    if (pNew1->memAvail - vgMem - pNew1->memUsed <= 0) {

×

2259

      mError("db:%s, vgId:%d, no enough memory:%" PRId64 " in dnode:%d avail:%" PRId64 " used:%" PRId64,

×

2261

      code = TSDB_CODE_MND_NO_ENOUGH_MEM_IN_DNODE;

×

2262

      goto _OVER;

×

2264

      pNew1->memUsed += vgMem;

×

2267

    TAOS_CHECK_GOTO(mndAddIncVgroupReplicaToTrans(pMnode, pTrans, pDb, &newVg, pNew1->id), NULL, _OVER);

×

2268

    TAOS_CHECK_GOTO(mndAddDecVgroupReplicaFromTrans(pMnode, pTrans, pDb, &newVg, pOld1->id), NULL, _OVER);

×

2271

  if (pNew2 != NULL && pOld2 != NULL) {

×

2272

    int32_t numOfVnodes = mndGetVnodesNum(pMnode, pNew2->id);

×

2273

    if (numOfVnodes >= pNew2->numOfSupportVnodes) {

×

2274

      mError("vgId:%d, no enough vnodes in dnode:%d, numOfVnodes:%d support:%d", newVg.vgId, pNew2->id, numOfVnodes,

×

2276

      code = TSDB_CODE_MND_NO_ENOUGH_VNODES;

×

2277

      goto _OVER;

×

2279

    int64_t vgMem = mndGetVgroupMemory(pMnode, NULL, pVgroup);

×

2280

    if (pNew2->memAvail - vgMem - pNew2->memUsed <= 0) {

×

2281

      mError("db:%s, vgId:%d, no enough memory:%" PRId64 " in dnode:%d avail:%" PRId64 " used:%" PRId64,

×

2283

      code = TSDB_CODE_MND_NO_ENOUGH_MEM_IN_DNODE;

×

2284

      goto _OVER;

×

2286

      pNew2->memUsed += vgMem;

×

2288

    TAOS_CHECK_GOTO(mndAddIncVgroupReplicaToTrans(pMnode, pTrans, pDb, &newVg, pNew2->id), NULL, _OVER);

×

2289

    TAOS_CHECK_GOTO(mndAddDecVgroupReplicaFromTrans(pMnode, pTrans, pDb, &newVg, pOld2->id), NULL, _OVER);

×

2292

  if (pNew3 != NULL && pOld3 != NULL) {

×

2293

    int32_t numOfVnodes = mndGetVnodesNum(pMnode, pNew3->id);

×

2294

    if (numOfVnodes >= pNew3->numOfSupportVnodes) {

×

2295

      mError("vgId:%d, no enough vnodes in dnode:%d, numOfVnodes:%d support:%d", newVg.vgId, pNew3->id, numOfVnodes,

×

2297

      code = TSDB_CODE_MND_NO_ENOUGH_VNODES;

×

2298

      goto _OVER;

×

2300

    int64_t vgMem = mndGetVgroupMemory(pMnode, NULL, pVgroup);

×

2301

    if (pNew3->memAvail - vgMem - pNew3->memUsed <= 0) {

×

2302

      mError("db:%s, vgId:%d, no enough memory:%" PRId64 " in dnode:%d avail:%" PRId64 " used:%" PRId64,

×

2304

      code = TSDB_CODE_MND_NO_ENOUGH_MEM_IN_DNODE;

×

2305

      goto _OVER;

×

2307

      pNew3->memUsed += vgMem;

×

2309

    TAOS_CHECK_GOTO(mndAddIncVgroupReplicaToTrans(pMnode, pTrans, pDb, &newVg, pNew3->id), NULL, _OVER);

×

2310

    TAOS_CHECK_GOTO(mndAddDecVgroupReplicaFromTrans(pMnode, pTrans, pDb, &newVg, pOld3->id), NULL, _OVER);

×

2314

    SSdbRaw *pRaw = mndVgroupActionEncode(&newVg);

×

2315

    if (pRaw == NULL) {

×

2316

      code = TSDB_CODE_MND_RETURN_VALUE_NULL;

×

2317

      if (terrno != 0) code = terrno;

×

2318

      goto _OVER;

×

2320

    if ((code = mndTransAppendCommitlog(pTrans, pRaw)) != 0) {

×

2321

      sdbFreeRaw(pRaw);

×

2322

      goto _OVER;

×

2324

    code = sdbSetRawStatus(pRaw, SDB_STATUS_READY);

×

2325

    if (code != 0) {

×

2326

      mError("vgId:%d, failed to set raw status since %s at line:%d", newVg.vgId, tstrerror(code), __LINE__);

×

2327

      goto _OVER;

×

2331

  mInfo("vgId:%d, vgroup info after redistribute, replica:%d", newVg.vgId, newVg.replica);

×

2332

  for (int32_t i = 0; i < newVg.replica; ++i) {

×

2333

    mInfo("vgId:%d, vnode:%d dnode:%d", newVg.vgId, i, newVg.vnodeGid[i].dnodeId);

×

2336

  TAOS_CHECK_GOTO(mndTransPrepare(pMnode, pTrans), NULL, _OVER);

×

2337

  code = 0;

×

2339

_OVER:

×

2340

  mndTransDrop(pTrans);

×

2341

  mndReleaseDb(pMnode, pDb);

×

2342

  TAOS_RETURN(code);

×

2345

static int32_t mndProcessRedistributeVgroupMsg(SRpcMsg *pReq) {

×

2346

  SMnode    *pMnode = pReq->info.node;

×

2347

  SDnodeObj *pNew1 = NULL;

×

2348

  SDnodeObj *pNew2 = NULL;

×

2349

  SDnodeObj *pNew3 = NULL;

×

2350

  SDnodeObj *pOld1 = NULL;

×

2351

  SDnodeObj *pOld2 = NULL;

×

2352

  SDnodeObj *pOld3 = NULL;

×

2353

  SVgObj    *pVgroup = NULL;

×

2354

  SDbObj    *pDb = NULL;

×

2355

  int32_t    code = -1;

×

2356

  int64_t    curMs = taosGetTimestampMs();

×

2357

  int32_t    newDnodeId[3] = {0};

×

2358

  int32_t    oldDnodeId[3] = {0};

×

2359

  int32_t    newIndex = -1;

×

2360

  int32_t    oldIndex = -1;

×

2362

  SRedistributeVgroupReq req = {0};

×

2363

  if (tDeserializeSRedistributeVgroupReq(pReq->pCont, pReq->contLen, &req) != 0) {

×

2364

    code = TSDB_CODE_INVALID_MSG;

×

2365

    goto _OVER;

×

2368

  mInfo("vgId:%d, start to redistribute vgroup to dnode %d:%d:%d", req.vgId, req.dnodeId1, req.dnodeId2, req.dnodeId3);

×

2369

  if ((code = mndCheckOperPrivilege(pMnode, pReq->info.conn.user, MND_OPER_REDISTRIBUTE_VGROUP)) != 0) {

×

2370

    goto _OVER;

×

2373

  pVgroup = mndAcquireVgroup(pMnode, req.vgId);

×

2374

  if (pVgroup == NULL) {

×

2375

    code = TSDB_CODE_MND_RETURN_VALUE_NULL;

×

2376

    if (terrno != 0) code = terrno;

×

2377

    goto _OVER;

×

2380

  pDb = mndAcquireDb(pMnode, pVgroup->dbName);

×

2381

  if (pDb == NULL) {

×

2382

    code = TSDB_CODE_MND_RETURN_VALUE_NULL;

×

2383

    if (terrno != 0) code = terrno;

×

2384

    goto _OVER;

×

2387

  if (pVgroup->replica == 1) {

×

2388

    if (req.dnodeId1 <= 0 || req.dnodeId2 > 0 || req.dnodeId3 > 0) {

×

2389

      code = TSDB_CODE_MND_INVALID_REPLICA;

×

2390

      goto _OVER;

×

2393

    if (req.dnodeId1 == pVgroup->vnodeGid[0].dnodeId) {

×

2395

      code = 0;

×

2396

      goto _OVER;

×

2399

    pNew1 = mndAcquireDnode(pMnode, req.dnodeId1);

×

2400

    if (pNew1 == NULL) {

×

2401

      code = TSDB_CODE_MND_RETURN_VALUE_NULL;

×

2402

      if (terrno != 0) code = terrno;

×

2403

      goto _OVER;

×

2405

    if (!mndIsDnodeOnline(pNew1, curMs)) {

×

2406

      code = TSDB_CODE_MND_HAS_OFFLINE_DNODE;

×

2407

      goto _OVER;

×

2410

    pOld1 = mndAcquireDnode(pMnode, pVgroup->vnodeGid[0].dnodeId);

×

2411

    if (pOld1 == NULL) {

×

2412

      code = TSDB_CODE_MND_RETURN_VALUE_NULL;

×

2413

      if (terrno != 0) code = terrno;

×

2414

      goto _OVER;

×

2416

    if (!mndIsDnodeOnline(pOld1, curMs)) {

×

2417

      code = TSDB_CODE_MND_HAS_OFFLINE_DNODE;

×

2418

      goto _OVER;

×

2421

    code = mndRedistributeVgroup(pMnode, pReq, pDb, pVgroup, pNew1, pOld1, NULL, NULL, NULL, NULL);

×

2423

  } else if (pVgroup->replica == 3) {

×

2424

    if (req.dnodeId1 <= 0 || req.dnodeId2 <= 0 || req.dnodeId3 <= 0) {

×

2425

      code = TSDB_CODE_MND_INVALID_REPLICA;

×

2426

      goto _OVER;

×

2429

    if (req.dnodeId1 == req.dnodeId2 || req.dnodeId1 == req.dnodeId3 || req.dnodeId2 == req.dnodeId3) {

×

2430

      code = TSDB_CODE_MND_INVALID_REPLICA;

×

2431

      goto _OVER;

×

2434

    if (req.dnodeId1 != pVgroup->vnodeGid[0].dnodeId && req.dnodeId1 != pVgroup->vnodeGid[1].dnodeId &&

×

2435

        req.dnodeId1 != pVgroup->vnodeGid[2].dnodeId) {

×

2436

      newDnodeId[++newIndex] = req.dnodeId1;

×

2437

      mInfo("vgId:%d, dnode:%d will be added, index:%d", pVgroup->vgId, newDnodeId[newIndex], newIndex);

×

2440

    if (req.dnodeId2 != pVgroup->vnodeGid[0].dnodeId && req.dnodeId2 != pVgroup->vnodeGid[1].dnodeId &&

×

2441

        req.dnodeId2 != pVgroup->vnodeGid[2].dnodeId) {

×

2442

      newDnodeId[++newIndex] = req.dnodeId2;

×

2443

      mInfo("vgId:%d, dnode:%d will be added, index:%d", pVgroup->vgId, newDnodeId[newIndex], newIndex);

×

2446

    if (req.dnodeId3 != pVgroup->vnodeGid[0].dnodeId && req.dnodeId3 != pVgroup->vnodeGid[1].dnodeId &&

×

2447

        req.dnodeId3 != pVgroup->vnodeGid[2].dnodeId) {

×

2448

      newDnodeId[++newIndex] = req.dnodeId3;

×

2449

      mInfo("vgId:%d, dnode:%d will be added, index:%d", pVgroup->vgId, newDnodeId[newIndex], newIndex);

×

2452

    if (req.dnodeId1 != pVgroup->vnodeGid[0].dnodeId && req.dnodeId2 != pVgroup->vnodeGid[0].dnodeId &&

×

2453

        req.dnodeId3 != pVgroup->vnodeGid[0].dnodeId) {

×

2454

      oldDnodeId[++oldIndex] = pVgroup->vnodeGid[0].dnodeId;

×

2455

      mInfo("vgId:%d, dnode:%d will be removed, index:%d", pVgroup->vgId, oldDnodeId[oldIndex], oldIndex);

×

2458

    if (req.dnodeId1 != pVgroup->vnodeGid[1].dnodeId && req.dnodeId2 != pVgroup->vnodeGid[1].dnodeId &&

×

2459

        req.dnodeId3 != pVgroup->vnodeGid[1].dnodeId) {

×

2460

      oldDnodeId[++oldIndex] = pVgroup->vnodeGid[1].dnodeId;

×

2461

      mInfo("vgId:%d, dnode:%d will be removed, index:%d", pVgroup->vgId, oldDnodeId[oldIndex], oldIndex);

×

2464

    if (req.dnodeId1 != pVgroup->vnodeGid[2].dnodeId && req.dnodeId2 != pVgroup->vnodeGid[2].dnodeId &&

×

2465

        req.dnodeId3 != pVgroup->vnodeGid[2].dnodeId) {

×

2466

      oldDnodeId[++oldIndex] = pVgroup->vnodeGid[2].dnodeId;

×

2467

      mInfo("vgId:%d, dnode:%d will be removed, index:%d", pVgroup->vgId, oldDnodeId[oldIndex], oldIndex);

×

2470

    if (newDnodeId[0] != 0) {

×

2471

      pNew1 = mndAcquireDnode(pMnode, newDnodeId[0]);

×

2472

      if (pNew1 == NULL) {

×

2473

        code = TSDB_CODE_MND_RETURN_VALUE_NULL;

×

2474

        if (terrno != 0) code = terrno;

×

2475

        goto _OVER;

×

2477

      if (!mndIsDnodeOnline(pNew1, curMs)) {

×

2478

        code = TSDB_CODE_MND_HAS_OFFLINE_DNODE;

×

2479

        goto _OVER;

×

2483

    if (newDnodeId[1] != 0) {

×

2484

      pNew2 = mndAcquireDnode(pMnode, newDnodeId[1]);

×

2485

      if (pNew2 == NULL) {

×

2486

        code = TSDB_CODE_MND_RETURN_VALUE_NULL;

×

2487

        if (terrno != 0) code = terrno;

×

2488

        goto _OVER;

×

2490

      if (!mndIsDnodeOnline(pNew2, curMs)) {

×

2491

        code = TSDB_CODE_MND_HAS_OFFLINE_DNODE;

×

2492

        goto _OVER;

×

2496

    if (newDnodeId[2] != 0) {

×

2497

      pNew3 = mndAcquireDnode(pMnode, newDnodeId[2]);

×

2498

      if (pNew3 == NULL) {

×

2499

        code = TSDB_CODE_MND_RETURN_VALUE_NULL;

×

2500

        if (terrno != 0) code = terrno;

×

2501

        goto _OVER;

×

2503

      if (!mndIsDnodeOnline(pNew3, curMs)) {

×

2504

        code = TSDB_CODE_MND_HAS_OFFLINE_DNODE;

×

2505

        goto _OVER;

×

2509

    if (oldDnodeId[0] != 0) {

×

2510

      pOld1 = mndAcquireDnode(pMnode, oldDnodeId[0]);

×

2511

      if (pOld1 == NULL) {

×

2512

        code = TSDB_CODE_MND_RETURN_VALUE_NULL;

×

2513

        if (terrno != 0) code = terrno;

×

2514

        goto _OVER;

×

2516

      if (!mndIsDnodeOnline(pOld1, curMs)) {

×

2517

        code = TSDB_CODE_MND_HAS_OFFLINE_DNODE;

×

2518

        goto _OVER;

×

2522

    if (oldDnodeId[1] != 0) {

×

2523

      pOld2 = mndAcquireDnode(pMnode, oldDnodeId[1]);

×

2524

      if (pOld2 == NULL) {

×

2525

        code = TSDB_CODE_MND_RETURN_VALUE_NULL;

×

2526

        if (terrno != 0) code = terrno;

×

2527

        goto _OVER;

×

2529

      if (!mndIsDnodeOnline(pOld2, curMs)) {

×

2530

        code = TSDB_CODE_MND_HAS_OFFLINE_DNODE;

×

2531

        goto _OVER;

×

2535

    if (oldDnodeId[2] != 0) {

×

2536

      pOld3 = mndAcquireDnode(pMnode, oldDnodeId[2]);

×

2537

      if (pOld3 == NULL) {

×

2538

        code = TSDB_CODE_MND_RETURN_VALUE_NULL;

×

2539

        if (terrno != 0) code = terrno;

×

2540

        goto _OVER;

×

2542

      if (!mndIsDnodeOnline(pOld3, curMs)) {

×

2543

        code = TSDB_CODE_MND_HAS_OFFLINE_DNODE;

×

2544

        goto _OVER;

×

2548

    if (pNew1 == NULL && pOld1 == NULL && pNew2 == NULL && pOld2 == NULL && pNew3 == NULL && pOld3 == NULL) {

×

2550

      code = 0;

×

2551

      goto _OVER;

×

2554

    code = mndRedistributeVgroup(pMnode, pReq, pDb, pVgroup, pNew1, pOld1, pNew2, pOld2, pNew3, pOld3);

×

2557

    code = TSDB_CODE_MND_REQ_REJECTED;

×

2558

    goto _OVER;

×

2561

  if (code == 0) code = TSDB_CODE_ACTION_IN_PROGRESS;

×

2563

  char obj[33] = {0};

×

2564

  (void)tsnprintf(obj, sizeof(obj), "%d", req.vgId);

×

2566

  auditRecord(pReq, pMnode->clusterId, "RedistributeVgroup", "", obj, req.sql, req.sqlLen);

×

2568

_OVER:

×

2569

  if (code != 0 && code != TSDB_CODE_ACTION_IN_PROGRESS) {

×

2570

    mError("vgId:%d, failed to redistribute to dnode %d:%d:%d since %s", req.vgId, req.dnodeId1, req.dnodeId2,

×

2574

  mndReleaseDnode(pMnode, pNew1);

×

2575

  mndReleaseDnode(pMnode, pNew2);

×

2576

  mndReleaseDnode(pMnode, pNew3);

×

2577

  mndReleaseDnode(pMnode, pOld1);

×

2578

  mndReleaseDnode(pMnode, pOld2);

×

2579

  mndReleaseDnode(pMnode, pOld3);

×

2580

  mndReleaseVgroup(pMnode, pVgroup);

×

2581

  mndReleaseDb(pMnode, pDb);

×

2582

  tFreeSRedistributeVgroupReq(&req);

×

2584

  TAOS_RETURN(code);

×

2587

static void *mndBuildSForceBecomeFollowerReq(SMnode *pMnode, SVgObj *pVgroup, int32_t dnodeId, int32_t *pContLen) {

×

2588

  SForceBecomeFollowerReq balanceReq = {

×

2589

      .vgId = pVgroup->vgId,

×

2592

  int32_t contLen = tSerializeSForceBecomeFollowerReq(NULL, 0, &balanceReq);

×

2593

  if (contLen < 0) {

×

2594

    terrno = TSDB_CODE_OUT_OF_MEMORY;

×

2595

    return NULL;

×

2597

  contLen += sizeof(SMsgHead);

×

2599

  void *pReq = taosMemoryMalloc(contLen);

×

2600

  if (pReq == NULL) {

×

2601

    terrno = TSDB_CODE_OUT_OF_MEMORY;

×

2602

    return NULL;

×

2605

  SMsgHead *pHead = pReq;

×

2606

  pHead->contLen = htonl(contLen);

×

2607

  pHead->vgId = htonl(pVgroup->vgId);

×

2609

  if (tSerializeSForceBecomeFollowerReq((char *)pReq + sizeof(SMsgHead), contLen, &balanceReq) < 0) {

×

2610

    terrno = TSDB_CODE_OUT_OF_MEMORY;

×

2611

    taosMemoryFree(pReq);

×

2612

    return NULL;

×

2614

  *pContLen = contLen;

×

2615

  return pReq;

×

2618

int32_t mndAddBalanceVgroupLeaderAction(SMnode *pMnode, STrans *pTrans, SVgObj *pVgroup, int32_t dnodeId) {

×

2619

  int32_t    code = 0;

×

2620

  SDnodeObj *pDnode = mndAcquireDnode(pMnode, dnodeId);

×

2621

  if (pDnode == NULL) {

×

2622

    code = TSDB_CODE_MND_RETURN_VALUE_NULL;

×

2623

    if (terrno != 0) code = terrno;

×

2624

    TAOS_RETURN(code);

×

2627

  STransAction action = {0};

×

2628

  action.epSet = mndGetDnodeEpset(pDnode);

×

2629

  mndReleaseDnode(pMnode, pDnode);

×

2631

  int32_t contLen = 0;

×

2632

  void   *pReq = mndBuildSForceBecomeFollowerReq(pMnode, pVgroup, dnodeId, &contLen);

×

2633

  if (pReq == NULL) {

×

2634

    code = TSDB_CODE_MND_RETURN_VALUE_NULL;

×

2635

    if (terrno != 0) code = terrno;

×

2636

    TAOS_RETURN(code);

×

2639

  action.pCont = pReq;

×

2640

  action.contLen = contLen;

×

2641

  action.msgType = TDMT_SYNC_FORCE_FOLLOWER;

×

2643

  if ((code = mndTransAppendRedoAction(pTrans, &action)) != 0) {

×

2644

    taosMemoryFree(pReq);

×

2645

    TAOS_RETURN(code);

×

2648

  TAOS_RETURN(code);

×

2651

int32_t mndAddVgroupBalanceToTrans(SMnode *pMnode, SVgObj *pVgroup, STrans *pTrans) {

×

2652

  int32_t code = 0;

×

2653

  SSdb   *pSdb = pMnode->pSdb;

×

2655

  int32_t vgid = pVgroup->vgId;

×

2656

  int8_t  replica = pVgroup->replica;

×

2658

  if (pVgroup->replica <= 1) {

×

2659

    mInfo("trans:%d, vgid:%d no need to balance, replica:%d", pTrans->id, vgid, replica);

×

2660

    return -1;

×

2663

  int32_t dnodeId = 0;

×

2665

  for (int i = 0; i < replica; i++) {

×

2666

    if (pVgroup->vnodeGid[i].syncState == TAOS_SYNC_STATE_LEADER) {

×

2667

      dnodeId = pVgroup->vnodeGid[i].dnodeId;

×

2668

      break;

×

2672

  bool       exist = false;

×

2673

  bool       online = false;

×

2674

  int64_t    curMs = taosGetTimestampMs();

×

2675

  SDnodeObj *pDnode = mndAcquireDnode(pMnode, dnodeId);

×

2676

  if (pDnode != NULL) {

×

2677

    exist = true;

×

2678

    online = mndIsDnodeOnline(pDnode, curMs);

×

2679

    mndReleaseDnode(pMnode, pDnode);

×

2682

  if (exist && online) {

×

2683

    mInfo("trans:%d, vgid:%d leader to dnode:%d", pTrans->id, vgid, dnodeId);

×

2685

    if ((code = mndAddBalanceVgroupLeaderAction(pMnode, pTrans, pVgroup, dnodeId)) != 0) {

×

2686

      mError("trans:%d, vgid:%d failed to be balanced to dnode:%d", pTrans->id, vgid, dnodeId);

×

2687

      TAOS_RETURN(code);

×

2690

    SDbObj *pDb = mndAcquireDb(pMnode, pVgroup->dbName);

×

2691

    if (pDb == NULL) {

×

2692

      code = TSDB_CODE_MND_RETURN_VALUE_NULL;

×

2693

      if (terrno != 0) code = terrno;

×

2694

      mError("trans:%d, vgid:%d failed to be balanced to dnode:%d, because db not exist", pTrans->id, vgid, dnodeId);

×

2695

      TAOS_RETURN(code);

×

2698

    mndReleaseDb(pMnode, pDb);

×

2700

    mInfo("trans:%d, vgid:%d cant be balanced to dnode:%d, exist:%d, online:%d", pTrans->id, vgid, dnodeId, exist,

×

2704

  TAOS_RETURN(code);

×

2709

int32_t mndProcessVgroupBalanceLeaderMsg(SRpcMsg *pReq) { return mndProcessVgroupBalanceLeaderMsgImp(pReq); }

×

2715

static int32_t mndCheckDnodeMemory(SMnode *pMnode, SDbObj *pOldDb, SDbObj *pNewDb, SVgObj *pOldVgroup,

×

2717

  for (int32_t i = 0; i < (int32_t)taosArrayGetSize(pArray); ++i) {

×

2718

    SDnodeObj *pDnode = taosArrayGet(pArray, i);

×

2719

    bool       inVgroup = false;

×

2720

    int64_t    oldMemUsed = 0;

×

2721

    int64_t    newMemUsed = 0;

×

2722

    mDebug("db:%s, vgId:%d, check dnode:%d, avail:%" PRId64 " used:%" PRId64, pNewVgroup->dbName, pNewVgroup->vgId,

×

2724

    for (int32_t j = 0; j < pOldVgroup->replica; ++j) {

×

2725

      SVnodeGid *pVgId = &pOldVgroup->vnodeGid[j];

×

2726

      if (pDnode->id == pVgId->dnodeId) {

×

2727

        oldMemUsed = mndGetVgroupMemory(pMnode, pOldDb, pOldVgroup);

×

2728

        inVgroup = true;

×

2731

    for (int32_t j = 0; j < pNewVgroup->replica; ++j) {

×

2732

      SVnodeGid *pVgId = &pNewVgroup->vnodeGid[j];

×

2733

      if (pDnode->id == pVgId->dnodeId) {

×

2734

        newMemUsed = mndGetVgroupMemory(pMnode, pNewDb, pNewVgroup);

×

2735

        inVgroup = true;

×

2739

    mDebug("db:%s, vgId:%d, memory in dnode:%d, oldUsed:%" PRId64 ", newUsed:%" PRId64, pNewVgroup->dbName,

×

2742

    pDnode->memUsed = pDnode->memUsed - oldMemUsed + newMemUsed;

×

2743

    if (pDnode->memAvail - pDnode->memUsed <= 0) {

×

2744

      mError("db:%s, vgId:%d, no enough memory in dnode:%d, avail:%" PRId64 " used:%" PRId64, pNewVgroup->dbName,

×

2746

      TAOS_RETURN(TSDB_CODE_MND_NO_ENOUGH_MEM_IN_DNODE);

×

2747

    } else if (inVgroup) {

×

2748

      mInfo("db:%s, vgId:%d, memory in dnode:%d, avail:%" PRId64 " used:%" PRId64, pNewVgroup->dbName, pNewVgroup->vgId,

×

2753

  return 0;

×

2756

int32_t mndBuildAlterVgroupAction(SMnode *pMnode, STrans *pTrans, SDbObj *pOldDb, SDbObj *pNewDb, SVgObj *pVgroup,

×

2758

  int32_t code = 0;

×

2759

  memcpy(pNewVgroup, pVgroup, sizeof(SVgObj));

×

2761

  if (pVgroup->replica <= 0 || pVgroup->replica == pNewDb->cfg.replications) {

×

2762

    TAOS_CHECK_RETURN(mndAddAlterVnodeConfigAction(pMnode, pTrans, pNewDb, pVgroup));

×

2763

    TAOS_CHECK_RETURN(mndCheckDnodeMemory(pMnode, pOldDb, pNewDb, pNewVgroup, pVgroup, pArray));

×

2764

    return 0;

×

2767

  mndTransSetSerial(pTrans);

×

2769

  if (pNewDb->cfg.replications == 3) {

×

2770

    mInfo("db:%s, vgId:%d, will add 2 vnodes, vn:0 dnode:%d", pVgroup->dbName, pVgroup->vgId,

×

2774

    if (pNewVgroup->replica == 1){

×

2775

      TAOS_CHECK_RETURN(mndAddVnodeToVgroup(pMnode, pTrans, pNewVgroup, pArray));

×

2779

    pNewVgroup->vnodeGid[0].nodeRole = TAOS_SYNC_ROLE_VOTER;

×

2780

    pNewVgroup->vnodeGid[1].nodeRole = TAOS_SYNC_ROLE_LEARNER;

×

2781

    TAOS_CHECK_RETURN(

×

2784

    TAOS_CHECK_RETURN(mndAddCreateVnodeAction(pMnode, pTrans, pNewDb, pNewVgroup, &pNewVgroup->vnodeGid[1]));

×

2787

    pNewVgroup->vnodeGid[1].nodeRole = TAOS_SYNC_ROLE_VOTER;

×

2788

    TAOS_CHECK_RETURN(mndAddAlterVnodeTypeAction(pMnode, pTrans, pNewDb, pNewVgroup, pNewVgroup->vnodeGid[1].dnodeId));

×

2789

    TAOS_CHECK_RETURN(

×

2792

    TAOS_CHECK_RETURN(mndAddAlterVnodeConfirmAction(pMnode, pTrans, pNewDb, pNewVgroup));

×

2795

    if (pNewVgroup->replica == 2){

×

2796

      TAOS_CHECK_RETURN (mndAddVnodeToVgroup(pMnode, pTrans, pNewVgroup, pArray));

×

2799

    pNewVgroup->vnodeGid[0].nodeRole = TAOS_SYNC_ROLE_VOTER;

×

2800

    pNewVgroup->vnodeGid[1].nodeRole = TAOS_SYNC_ROLE_VOTER;

×

2801

    pNewVgroup->vnodeGid[2].nodeRole = TAOS_SYNC_ROLE_VOTER;

×

2802

    TAOS_CHECK_RETURN(

×

2804

    TAOS_CHECK_RETURN(

×

2806

    TAOS_CHECK_RETURN(mndAddCreateVnodeAction(pMnode, pTrans, pNewDb, pNewVgroup, &pNewVgroup->vnodeGid[2]));

×

2808

    TAOS_CHECK_RETURN(mndAddAlterVnodeConfirmAction(pMnode, pTrans, pNewDb, pNewVgroup));

×

2809

  } else if (pNewDb->cfg.replications == 1) {

×

2810

    mInfo("db:%s, vgId:%d, will remove 2 vnodes, vn:0 dnode:%d vn:1 dnode:%d vn:2 dnode:%d", pVgroup->dbName,

×

2813

    SVnodeGid del1 = {0};

×

2814

    SVnodeGid del2 = {0};

×

2815

    TAOS_CHECK_RETURN(mndRemoveVnodeFromVgroup(pMnode, pTrans, pNewVgroup, pArray, &del1));

×

2816

    TAOS_CHECK_RETURN(mndAddDropVnodeAction(pMnode, pTrans, pNewDb, pNewVgroup, &del1, true));

×

2817

    TAOS_CHECK_RETURN(

×

2819

    TAOS_CHECK_RETURN(

×

2821

    TAOS_CHECK_RETURN(mndAddAlterVnodeConfirmAction(pMnode, pTrans, pNewDb, pNewVgroup));

×

2823

    TAOS_CHECK_RETURN(mndRemoveVnodeFromVgroup(pMnode, pTrans, pNewVgroup, pArray, &del2));

×

2824

    TAOS_CHECK_RETURN(mndAddDropVnodeAction(pMnode, pTrans, pNewDb, pNewVgroup, &del2, true));

×

2825

    TAOS_CHECK_RETURN(

×

2827

    TAOS_CHECK_RETURN(mndAddAlterVnodeConfirmAction(pMnode, pTrans, pNewDb, pNewVgroup));

×

2828

  } else if (pNewDb->cfg.replications == 2) {

×

2829

    mInfo("db:%s, vgId:%d, will add 1 vnode, vn:0 dnode:%d", pVgroup->dbName, pVgroup->vgId,

×

2833

    TAOS_CHECK_RETURN(mndAddVnodeToVgroup(pMnode, pTrans, pNewVgroup, pArray));

×

2836

    pNewVgroup->vnodeGid[0].nodeRole = TAOS_SYNC_ROLE_VOTER;

×

2837

    pNewVgroup->vnodeGid[1].nodeRole = TAOS_SYNC_ROLE_LEARNER;

×

2838

    TAOS_CHECK_RETURN(

×

2841

    TAOS_CHECK_RETURN(mndAddCreateVnodeAction(pMnode, pTrans, pNewDb, pNewVgroup, &pNewVgroup->vnodeGid[1]));

×

2844

    pNewVgroup->vnodeGid[1].nodeRole = TAOS_SYNC_ROLE_VOTER;

×

2845

    TAOS_CHECK_RETURN(mndAddAlterVnodeTypeAction(pMnode, pTrans, pNewDb, pNewVgroup, pNewVgroup->vnodeGid[1].dnodeId));

×

2846

    TAOS_CHECK_RETURN(

×

2849

    TAOS_CHECK_RETURN(mndAddAlterVnodeConfirmAction(pMnode, pTrans, pNewDb, pNewVgroup));

×

2851

    return -1;

×

2854

  mndSortVnodeGid(pNewVgroup);

×

2857

    SSdbRaw *pVgRaw = mndVgroupActionEncode(pNewVgroup);

×

2858

    if (pVgRaw == NULL) {

×

2859

      code = TSDB_CODE_MND_RETURN_VALUE_NULL;

×

2860

      if (terrno != 0) code = terrno;

×

2861

      TAOS_RETURN(code);

×

2863

    if ((code = mndTransAppendCommitlog(pTrans, pVgRaw)) != 0) {

×

2864

      sdbFreeRaw(pVgRaw);

×

2865

      TAOS_RETURN(code);

×

2867

    code = sdbSetRawStatus(pVgRaw, SDB_STATUS_READY);

×

2868

    if (code != 0) {

×

2869

      mError("vgId:%d, failed to set raw status since %s at line:%d", pNewVgroup->vgId, tstrerror(code), __LINE__);

×

2870

      TAOS_RETURN(code);

×

2874

  TAOS_RETURN(code);

×

2877

int32_t mndBuildRaftAlterVgroupAction(SMnode *pMnode, STrans *pTrans, SDbObj *pOldDb, SDbObj *pNewDb, SVgObj *pVgroup,

×

2879

  int32_t code = 0;

×

2880

  SVgObj  newVgroup = {0};

×

2881

  memcpy(&newVgroup, pVgroup, sizeof(SVgObj));

×

2883

  if (pVgroup->replica <= 0 || pVgroup->replica == pNewDb->cfg.replications) {

×

2884

    TAOS_CHECK_RETURN(mndAddAlterVnodeConfigAction(pMnode, pTrans, pNewDb, pVgroup));

×

2885

    TAOS_CHECK_RETURN(mndCheckDnodeMemory(pMnode, pOldDb, pNewDb, &newVgroup, pVgroup, pArray));

×

2886

    return 0;

×

2889

  mndTransSetSerial(pTrans);

×

2891

  mInfo("trans:%d, vgId:%d, alter vgroup, syncConfChangeVer:%d, version:%d, replica:%d", pTrans->id, pVgroup->vgId,

×

2894

  if (newVgroup.replica == 1 && pNewDb->cfg.replications == 3) {

×

2895

    mInfo("db:%s, vgId:%d, will add 2 vnodes, vn:0 dnode:%d", pVgroup->dbName, pVgroup->vgId,

×

2899

    TAOS_CHECK_RETURN(mndAddVnodeToVgroup(pMnode, pTrans, &newVgroup, pArray));

×

2901

    TAOS_CHECK_RETURN(mndAddVnodeToVgroup(pMnode, pTrans, &newVgroup, pArray));

×

2904

    newVgroup.vnodeGid[0].nodeRole = TAOS_SYNC_ROLE_VOTER;

×

2905

    newVgroup.vnodeGid[1].nodeRole = TAOS_SYNC_ROLE_LEARNER;

×

2906

    newVgroup.vnodeGid[2].nodeRole = TAOS_SYNC_ROLE_LEARNER;

×

2907

    TAOS_CHECK_RETURN(

×

2909

    mInfo("trans:%d, vgId:%d, add change config, syncConfChangeVer:%d, version:%d, replica:%d", pTrans->id,

×

2911

    TAOS_CHECK_RETURN(mndAddCreateVnodeAction(pMnode, pTrans, pNewDb, &newVgroup, &newVgroup.vnodeGid[1]));

×

2912

    mInfo("trans:%d, vgId:%d, create vnode, syncConfChangeVer:%d, version:%d, replica:%d", pTrans->id, pVgroup->vgId,

×

2914

    TAOS_CHECK_RETURN(mndAddCreateVnodeAction(pMnode, pTrans, pNewDb, &newVgroup, &newVgroup.vnodeGid[2]));

×

2915

    mInfo("trans:%d, vgId:%d, create vnode, syncConfChangeVer:%d, version:%d, replica:%d", pTrans->id, pVgroup->vgId,

×

2919

    newVgroup.vnodeGid[0].nodeRole = TAOS_SYNC_ROLE_VOTER;

×

2920

    newVgroup.vnodeGid[1].nodeRole = TAOS_SYNC_ROLE_VOTER;

×

2921

    newVgroup.vnodeGid[2].nodeRole = TAOS_SYNC_ROLE_VOTER;

×

2922

    TAOS_CHECK_RETURN(

×

2924

    TAOS_CHECK_RETURN(

×

2928

    newVgroup.vnodeGid[0].nodeRole = TAOS_SYNC_ROLE_VOTER;

×

2929

    newVgroup.vnodeGid[1].nodeRole = TAOS_SYNC_ROLE_VOTER;

×

2930

    newVgroup.vnodeGid[2].nodeRole = TAOS_SYNC_ROLE_LEARNER;

×

2931

    TAOS_CHECK_RETURN(

×

2934

    TAOS_CHECK_RETURN(mndAddAlterVnodeConfirmAction(pMnode, pTrans, pNewDb, &newVgroup));

×

2936

    newVgroup.vnodeGid[0].nodeRole = TAOS_SYNC_ROLE_VOTER;

×

2937

    newVgroup.vnodeGid[1].nodeRole = TAOS_SYNC_ROLE_VOTER;

×

2938

    newVgroup.vnodeGid[2].nodeRole = TAOS_SYNC_ROLE_VOTER;

×

2939

    TAOS_CHECK_RETURN(

×

2942

    TAOS_CHECK_RETURN(mndAddAlterVnodeConfirmAction(pMnode, pTrans, pNewDb, &newVgroup));

×

2944

    SSdbRaw *pVgRaw = mndVgroupActionEncode(&newVgroup);

×

2945

    if (pVgRaw == NULL) {

×

2946

      code = TSDB_CODE_MND_RETURN_VALUE_NULL;

×

2947

      if (terrno != 0) code = terrno;

×

2948

      TAOS_RETURN(code);

×

2950

    if ((code = mndTransAppendRedolog(pTrans, pVgRaw)) != 0) {

×

2951

      sdbFreeRaw(pVgRaw);

×

2952

      TAOS_RETURN(code);

×

2954

    code = sdbSetRawStatus(pVgRaw, SDB_STATUS_READY);

×

2955

    if (code != 0) {

×

2956

      mError("vgId:%d, failed to set raw status to ready, error:%s, line:%d", newVgroup.vgId, tstrerror(code),

×

2958

      TAOS_RETURN(code);

×

2960

  } else if (newVgroup.replica == 3 && pNewDb->cfg.replications == 1) {

×

2961

    mInfo("db:%s, vgId:%d, will remove 2 vnodes, vn:0 dnode:%d vn:1 dnode:%d vn:2 dnode:%d", pVgroup->dbName,

×

2964

    SVnodeGid del1 = {0};

×

2965

    TAOS_CHECK_RETURN(mndRemoveVnodeFromVgroupWithoutSave(pMnode, pTrans, &newVgroup, pArray, &del1));

×

2967

    TAOS_CHECK_RETURN(

×

2970

    TAOS_CHECK_RETURN(mndAddAlterVnodeConfirmAction(pMnode, pTrans, pNewDb, &newVgroup));

×

2972

    TAOS_CHECK_RETURN(mndAddDropVnodeAction(pMnode, pTrans, pNewDb, &newVgroup, &del1, true));

×

2974

    SSdbRaw *pVgRaw = mndVgroupActionEncode(&newVgroup);

×

2975

    if (pVgRaw == NULL) {

×

2976

      code = TSDB_CODE_MND_RETURN_VALUE_NULL;

×

2977

      if (terrno != 0) code = terrno;

×

2978

      TAOS_RETURN(code);

×

2980

    if ((code = mndTransAppendRedolog(pTrans, pVgRaw)) != 0) {

×

2981

      sdbFreeRaw(pVgRaw);

×

2982

      TAOS_RETURN(code);

×

2984

    code = sdbSetRawStatus(pVgRaw, SDB_STATUS_READY);

×

2985

    if (code != 0) {

×

2986

      mError("vgId:%d, failed to set raw status to ready, error:%s, line:%d", newVgroup.vgId, tstrerror(code),

×

2988

      TAOS_RETURN(code);

×

2991

    SVnodeGid del2 = {0};

×

2992

    TAOS_CHECK_RETURN(mndRemoveVnodeFromVgroupWithoutSave(pMnode, pTrans, &newVgroup, pArray, &del2));

×

2994

    TAOS_CHECK_RETURN(

×

2997

    TAOS_CHECK_RETURN(mndAddAlterVnodeConfirmAction(pMnode, pTrans, pNewDb, &newVgroup));

×

2999

    TAOS_CHECK_RETURN(mndAddDropVnodeAction(pMnode, pTrans, pNewDb, &newVgroup, &del2, true));

×

3001

    pVgRaw = mndVgroupActionEncode(&newVgroup);

×

3002

    if (pVgRaw == NULL) {

×

3003

      code = TSDB_CODE_MND_RETURN_VALUE_NULL;

×

3004

      if (terrno != 0) code = terrno;

×

3005

      TAOS_RETURN(code);

×

3007

    if ((code = mndTransAppendRedolog(pTrans, pVgRaw)) != 0) {

×

3008

      sdbFreeRaw(pVgRaw);

×

3009

      TAOS_RETURN(code);

×

3011

    code = sdbSetRawStatus(pVgRaw, SDB_STATUS_READY);

×

3012

    if (code != 0) {

×

3013

      mError("vgId:%d, failed to set raw status to ready, error:%s, line:%d", newVgroup.vgId, tstrerror(code),

×

3015

      TAOS_RETURN(code);

×

3018

    return -1;

×

3021

  mndSortVnodeGid(&newVgroup);

×

3024

    SSdbRaw *pVgRaw = mndVgroupActionEncode(&newVgroup);

×

3025

    if (pVgRaw == NULL) {

×

3026

      code = TSDB_CODE_MND_RETURN_VALUE_NULL;

×

3027

      if (terrno != 0) code = terrno;

×

3028

      TAOS_RETURN(code);

×

3030

    if ((code = mndTransAppendCommitlog(pTrans, pVgRaw)) != 0) {

×

3031

      sdbFreeRaw(pVgRaw);

×

3032

      TAOS_RETURN(code);

×

3034

    code = sdbSetRawStatus(pVgRaw, SDB_STATUS_READY);

×

3035

    if (code != 0) {

×

3036

      mError("vgId:%d, failed to set raw status to ready, error:%s, line:%d", newVgroup.vgId, tstrerror(code),

×

3038

      TAOS_RETURN(code);

×

3042

  TAOS_RETURN(code);

×

3045

int32_t mndBuildRestoreAlterVgroupAction(SMnode *pMnode, STrans *pTrans, SDbObj *db, SVgObj *pVgroup, SDnodeObj *pDnode,

×

3047

  int32_t code = 0;

×

3048

  SVgObj  newVgroup = {0};

×

3049

  memcpy(&newVgroup, pVgroup, sizeof(SVgObj));

×

3051

  mInfo("db:%s, vgId:%d, restore vnodes, vn:0 dnode:%d", pVgroup->dbName, pVgroup->vgId, pVgroup->vnodeGid[0].dnodeId);

×

3053

  if (newVgroup.replica == 1) {

×

3054

    int selected = 0;

×

3055

    for (int i = 0; i < newVgroup.replica; i++) {

×

3056

      newVgroup.vnodeGid[i].nodeRole = TAOS_SYNC_ROLE_VOTER;

×

3057

      if (newVgroup.vnodeGid[i].dnodeId == pDnode->id) {

×

3058

        selected = i;

×

3061

    TAOS_CHECK_RETURN(mndAddCreateVnodeAction(pMnode, pTrans, db, &newVgroup, &newVgroup.vnodeGid[selected]));

×

3062

  } else if (newVgroup.replica == 2) {

×

3063

    for (int i = 0; i < newVgroup.replica; i++) {

×

3064

      if (newVgroup.vnodeGid[i].dnodeId == pDnode->id) {

×

3065

        newVgroup.vnodeGid[i].nodeRole = TAOS_SYNC_ROLE_LEARNER;

×

3067

        newVgroup.vnodeGid[i].nodeRole = TAOS_SYNC_ROLE_VOTER;

×

3070

    TAOS_CHECK_RETURN(mndRestoreAddAlterVnodeTypeAction(pMnode, pTrans, db, &newVgroup, pAnotherDnode));

×

3072

    for (int i = 0; i < newVgroup.replica; i++) {

×

3073

      if (newVgroup.vnodeGid[i].dnodeId == pDnode->id) {

×

3074

        newVgroup.vnodeGid[i].nodeRole = TAOS_SYNC_ROLE_LEARNER;

×

3076

        newVgroup.vnodeGid[i].nodeRole = TAOS_SYNC_ROLE_VOTER;

×

3079

    TAOS_CHECK_RETURN(mndRestoreAddCreateVnodeAction(pMnode, pTrans, db, &newVgroup, pDnode));

×

3081

    for (int i = 0; i < newVgroup.replica; i++) {

×

3082

      newVgroup.vnodeGid[i].nodeRole = TAOS_SYNC_ROLE_VOTER;

×

3083

      if (newVgroup.vnodeGid[i].dnodeId == pDnode->id) {

×

3086

    TAOS_CHECK_RETURN(mndRestoreAddAlterVnodeTypeAction(pMnode, pTrans, db, &newVgroup, pDnode));

×

3087

    TAOS_CHECK_RETURN(mndRestoreAddAlterVnodeTypeAction(pMnode, pTrans, db, &newVgroup, pAnotherDnode));

×

3088

  } else if (newVgroup.replica == 3) {

×

3089

    for (int i = 0; i < newVgroup.replica; i++) {

×

3090

      if (newVgroup.vnodeGid[i].dnodeId == pDnode->id) {

×

3091

        newVgroup.vnodeGid[i].nodeRole = TAOS_SYNC_ROLE_LEARNER;

×

3093

        newVgroup.vnodeGid[i].nodeRole = TAOS_SYNC_ROLE_VOTER;

×

3096

    TAOS_CHECK_RETURN(mndRestoreAddCreateVnodeAction(pMnode, pTrans, db, &newVgroup, pDnode));

×

3098

    for (int i = 0; i < newVgroup.replica; i++) {

×

3099

      newVgroup.vnodeGid[i].nodeRole = TAOS_SYNC_ROLE_VOTER;

×

3100

      if (newVgroup.vnodeGid[i].dnodeId == pDnode->id) {

×

3103

    TAOS_CHECK_RETURN(mndRestoreAddAlterVnodeTypeAction(pMnode, pTrans, db, &newVgroup, pDnode));

×

3105

  SSdbRaw *pVgRaw = mndVgroupActionEncode(&newVgroup);

×

3106

  if (pVgRaw == NULL) {

×

3107

    code = TSDB_CODE_MND_RETURN_VALUE_NULL;

×

3108

    if (terrno != 0) code = terrno;

×

3109

    TAOS_RETURN(code);

×

3111

  if ((code = mndTransAppendCommitlog(pTrans, pVgRaw)) != 0) {

×

3112

    sdbFreeRaw(pVgRaw);

×

3113

    TAOS_RETURN(code);

×

3115

  code = sdbSetRawStatus(pVgRaw, SDB_STATUS_READY);

×

3116

  if (code != 0) {

×

3117

    mError("vgId:%d, failed to set raw status to ready, error:%s, line:%d", newVgroup.vgId, tstrerror(code), __LINE__);

×

3118

    TAOS_RETURN(code);

×

3121

  TAOS_RETURN(code);

×

3124

static int32_t mndAddAdjustVnodeHashRangeAction(SMnode *pMnode, STrans *pTrans, SDbObj *pDb, SVgObj *pVgroup) {

×

3125

  return 0;

×

3130

static int32_t mndAddVgStatusAction(STrans *pTrans, SVgObj *pVg, ESdbStatus vgStatus, ETrnStage stage) {

×

3131

  int32_t         code = 0;

×

3132

  FpTransActionCb appendActionCb = (stage == TRN_STAGE_COMMIT_ACTION) ? mndTransAppendCommitlog : mndTransAppendRedolog;

×

3133

  SSdbRaw        *pRaw = mndVgroupActionEncode(pVg);

×

3134

  if (pRaw == NULL) {

×

3135

    code = TSDB_CODE_MND_RETURN_VALUE_NULL;

×

3136

    if (terrno != 0) code = terrno;

×

3137

    goto _err;

×

3139

  if ((code = appendActionCb(pTrans, pRaw)) != 0) goto _err;

×

3140

  code = sdbSetRawStatus(pRaw, vgStatus);

×

3141

  if (code != 0) {

×

3142

    mError("vgId:%d, failed to set raw status to ready, error:%s, line:%d", pVg->vgId, tstrerror(code), __LINE__);

×

3143

    goto _err;

×

3145

  pRaw = NULL;

×

3146

  TAOS_RETURN(code);

×

3147

_err:

×

3148

  sdbFreeRaw(pRaw);

×

3149

  TAOS_RETURN(code);

×

3152

static int32_t mndAddDbStatusAction(STrans *pTrans, SDbObj *pDb, ESdbStatus dbStatus, ETrnStage stage) {

×

3153

  int32_t         code = 0;

×

3154

  FpTransActionCb appendActionCb = (stage == TRN_STAGE_COMMIT_ACTION) ? mndTransAppendCommitlog : mndTransAppendRedolog;

×

3155

  SSdbRaw        *pRaw = mndDbActionEncode(pDb);

×

3156

  if (pRaw == NULL) {

×

3157

    code = TSDB_CODE_MND_RETURN_VALUE_NULL;

×

3158

    if (terrno != 0) code = terrno;

×

3159

    goto _err;

×

3161

  if ((code = appendActionCb(pTrans, pRaw)) != 0) goto _err;

×

3162

  code = sdbSetRawStatus(pRaw, dbStatus);

×

3163

  if (code != 0) {

×

3164

    mError("db:%s, failed to set raw status to ready, error:%s, line:%d", pDb->name, tstrerror(code), __LINE__);

×

3165

    goto _err;

×

3167

  pRaw = NULL;

×

3168

  TAOS_RETURN(code);

×

3169

_err:

×

3170

  sdbFreeRaw(pRaw);

×

3171

  TAOS_RETURN(code);

×

3174

int32_t mndSplitVgroup(SMnode *pMnode, SRpcMsg *pReq, SDbObj *pDb, SVgObj *pVgroup) {

×

3175

  int32_t code = -1;

×

3176

  STrans *pTrans = NULL;

×

3177

  SDbObj  dbObj = {0};

×

3178

  SArray *pArray = mndBuildDnodesArray(pMnode, 0, NULL);

×

3180

  int32_t numOfStreams = 0;

×

3181

  if ((code = mndGetNumOfStreams(pMnode, pDb->name, &numOfStreams)) != 0) {

×

3182

    goto _OVER;

×

3184

  if (numOfStreams > 0) {

×

3185

    code = TSDB_CODE_MND_STREAM_MUST_BE_DELETED;

×

3186

    goto _OVER;

×

3191

  if (tsS3Enabled) {

×

3192

    code = TSDB_CODE_OPS_NOT_SUPPORT;

×

3193

    mError("vgId:%d, db:%s, s3 exists, split vgroup not allowed", pVgroup->vgId, pVgroup->dbName);

×

3194

    goto _OVER;

×

3198

  if (pDb->cfg.withArbitrator) {

×

3199

    code = TSDB_CODE_OPS_NOT_SUPPORT;

×

3200

    mError("vgId:%d, db:%s, with arbitrator, split vgroup not allowed", pVgroup->vgId, pVgroup->dbName);

×

3201

    goto _OVER;

×

3204

  pTrans = mndTransCreate(pMnode, TRN_POLICY_RETRY, TRN_CONFLICT_DB, pReq, "split-vgroup");

×

3205

  if (pTrans == NULL) {

×

3206

    code = TSDB_CODE_MND_RETURN_VALUE_NULL;

×

3207

    if (terrno != 0) code = terrno;

×

3208

    goto _OVER;

×

3210

  mndTransSetSerial(pTrans);

×

3211

  mInfo("trans:%d, used to split vgroup, vgId:%d", pTrans->id, pVgroup->vgId);

×

3213

  mndTransSetDbName(pTrans, pDb->name, NULL);

×

3214

  TAOS_CHECK_GOTO(mndTransCheckConflictWithCompact(pMnode, pTrans), NULL, _OVER);

×

3216

  SVgObj newVg1 = {0};

×

3217

  memcpy(&newVg1, pVgroup, sizeof(SVgObj));

×

3218

  mInfo("vgId:%d, vgroup info before split, replica:%d hashBegin:%u hashEnd:%u", newVg1.vgId, newVg1.replica,

×

3220

  for (int32_t i = 0; i < newVg1.replica; ++i) {

×

3221

    mInfo("vgId:%d, vnode:%d dnode:%d", newVg1.vgId, i, newVg1.vnodeGid[i].dnodeId);

×

3224

  if (newVg1.replica == 1) {

×

3225

    TAOS_CHECK_GOTO(mndAddVnodeToVgroup(pMnode, pTrans, &newVg1, pArray), NULL, _OVER);

×

3227

    newVg1.vnodeGid[1].nodeRole = TAOS_SYNC_ROLE_LEARNER;

×

3228

    TAOS_CHECK_GOTO(mndAddAlterVnodeReplicaAction(pMnode, pTrans, pDb, &newVg1, newVg1.vnodeGid[0].dnodeId), NULL,

×

3230

    TAOS_CHECK_GOTO(mndAddCreateVnodeAction(pMnode, pTrans, pDb, &newVg1, &newVg1.vnodeGid[1]), NULL, _OVER);

×

3232

    newVg1.vnodeGid[1].nodeRole = TAOS_SYNC_ROLE_VOTER;

×

3233

    TAOS_CHECK_GOTO(mndAddAlterVnodeTypeAction(pMnode, pTrans, pDb, &newVg1, newVg1.vnodeGid[1].dnodeId), NULL, _OVER);

×

3234

    TAOS_CHECK_GOTO(mndAddAlterVnodeReplicaAction(pMnode, pTrans, pDb, &newVg1, newVg1.vnodeGid[0].dnodeId), NULL,

×

3237

    TAOS_CHECK_GOTO(mndAddAlterVnodeConfirmAction(pMnode, pTrans, pDb, &newVg1), NULL, _OVER);

×

3238

  } else if (newVg1.replica == 3) {

×

3239

    SVnodeGid del1 = {0};

×

3240

    TAOS_CHECK_GOTO(mndRemoveVnodeFromVgroup(pMnode, pTrans, &newVg1, pArray, &del1), NULL, _OVER);

×

3241

    TAOS_CHECK_GOTO(mndAddDropVnodeAction(pMnode, pTrans, pDb, &newVg1, &del1, true), NULL, _OVER);

×

3242

    TAOS_CHECK_GOTO(mndAddAlterVnodeReplicaAction(pMnode, pTrans, pDb, &newVg1, newVg1.vnodeGid[0].dnodeId), NULL,

×

3244

    TAOS_CHECK_GOTO(mndAddAlterVnodeReplicaAction(pMnode, pTrans, pDb, &newVg1, newVg1.vnodeGid[1].dnodeId), NULL,

×

3247

    goto _OVER;

×

3250

  for (int32_t i = 0; i < newVg1.replica; ++i) {

×

3251

    TAOS_CHECK_GOTO(mndAddDisableVnodeWriteAction(pMnode, pTrans, pDb, &newVg1, newVg1.vnodeGid[i].dnodeId), NULL,

×

3254

  TAOS_CHECK_GOTO(mndAddAlterVnodeConfirmAction(pMnode, pTrans, pDb, &newVg1), NULL, _OVER);

×

3256

  SVgObj newVg2 = {0};

×

3257

  memcpy(&newVg2, &newVg1, sizeof(SVgObj));

×

3258

  newVg1.replica = 1;

×

3259

  newVg1.hashEnd = newVg1.hashBegin / 2 + newVg1.hashEnd / 2;

×

3260

  memset(&newVg1.vnodeGid[1], 0, sizeof(SVnodeGid));

×

3262

  newVg2.replica = 1;

×

3263

  newVg2.hashBegin = newVg1.hashEnd + 1;

×

3264

  memcpy(&newVg2.vnodeGid[0], &newVg2.vnodeGid[1], sizeof(SVnodeGid));

×

3265

  memset(&newVg2.vnodeGid[1], 0, sizeof(SVnodeGid));

×

3267

  mInfo("vgId:%d, vgroup info after split, replica:%d hashrange:[%u, %u] vnode:0 dnode:%d", newVg1.vgId, newVg1.replica,

×

3269

  for (int32_t i = 0; i < newVg1.replica; ++i) {

×

3270

    mInfo("vgId:%d, vnode:%d dnode:%d", newVg1.vgId, i, newVg1.vnodeGid[i].dnodeId);

×

3272

  mInfo("vgId:%d, vgroup info after split, replica:%d hashrange:[%u, %u] vnode:0 dnode:%d", newVg2.vgId, newVg2.replica,

×

3274

  for (int32_t i = 0; i < newVg1.replica; ++i) {

×

3275

    mInfo("vgId:%d, vnode:%d dnode:%d", newVg2.vgId, i, newVg2.vnodeGid[i].dnodeId);

×

3279

  int32_t maxVgId = sdbGetMaxId(pMnode->pSdb, SDB_VGROUP);

×

3280

  int32_t srcVgId = newVg1.vgId;

×

3281

  newVg1.vgId = maxVgId;

×

3282

  TAOS_CHECK_GOTO(mndAddNewVgPrepareAction(pMnode, pTrans, &newVg1), NULL, _OVER);

×

3283

  TAOS_CHECK_GOTO(mndAddAlterVnodeHashRangeAction(pMnode, pTrans, srcVgId, &newVg1), NULL, _OVER);

×

3285

  maxVgId++;

×

3286

  srcVgId = newVg2.vgId;

×

3287

  newVg2.vgId = maxVgId;

×

3288

  TAOS_CHECK_GOTO(mndAddNewVgPrepareAction(pMnode, pTrans, &newVg2), NULL, _OVER);

×

3289

  TAOS_CHECK_GOTO(mndAddAlterVnodeHashRangeAction(pMnode, pTrans, srcVgId, &newVg2), NULL, _OVER);

×

3291

  TAOS_CHECK_GOTO(mndAddAlterVnodeConfirmAction(pMnode, pTrans, pDb, &newVg1), NULL, _OVER);

×

3292

  TAOS_CHECK_GOTO(mndAddAlterVnodeConfirmAction(pMnode, pTrans, pDb, &newVg2), NULL, _OVER);

×

3294

  TAOS_CHECK_GOTO(mndAddVgStatusAction(pTrans, &newVg1, SDB_STATUS_READY, TRN_STAGE_REDO_ACTION), NULL, _OVER);

×

3295

  TAOS_CHECK_GOTO(mndAddVgStatusAction(pTrans, &newVg2, SDB_STATUS_READY, TRN_STAGE_REDO_ACTION), NULL, _OVER);

×

3296

  TAOS_CHECK_GOTO(mndAddVgStatusAction(pTrans, pVgroup, SDB_STATUS_DROPPED, TRN_STAGE_REDO_ACTION), NULL, _OVER);

×

3299

  memcpy(&dbObj, pDb, sizeof(SDbObj));

×

3300

  if (dbObj.cfg.pRetensions != NULL) {

×

3301

    dbObj.cfg.pRetensions = taosArrayDup(pDb->cfg.pRetensions, NULL);

×

3302

    if (dbObj.cfg.pRetensions == NULL) {

×

3303

      code = terrno;

×

3304

      goto _OVER;

×

3307

  dbObj.vgVersion++;

×

3308

  dbObj.updateTime = taosGetTimestampMs();

×

3309

  dbObj.cfg.numOfVgroups++;

×

3310

  TAOS_CHECK_GOTO(mndAddDbStatusAction(pTrans, &dbObj, SDB_STATUS_READY, TRN_STAGE_REDO_ACTION), NULL, _OVER);

×

3313

  if (pDb->cfg.replications != newVg1.replica) {

×

3314

    SVgObj tmpGroup = {0};

×

3315

    TAOS_CHECK_GOTO(mndBuildAlterVgroupAction(pMnode, pTrans, pDb, pDb, &newVg1, pArray, &tmpGroup), NULL, _OVER);

×

3317

    TAOS_CHECK_GOTO(mndAddVgStatusAction(pTrans, &newVg1, SDB_STATUS_READY, TRN_STAGE_COMMIT_ACTION), NULL, _OVER);

×

3320

  if (pDb->cfg.replications != newVg2.replica) {

×

3321

    SVgObj tmpGroup = {0};

×

3322

    TAOS_CHECK_GOTO(mndBuildAlterVgroupAction(pMnode, pTrans, pDb, pDb, &newVg2, pArray, &tmpGroup), NULL, _OVER);

×

3324

    TAOS_CHECK_GOTO(mndAddVgStatusAction(pTrans, &newVg2, SDB_STATUS_READY, TRN_STAGE_COMMIT_ACTION), NULL, _OVER);

×

3327

  TAOS_CHECK_GOTO(mndAddVgStatusAction(pTrans, pVgroup, SDB_STATUS_DROPPED, TRN_STAGE_COMMIT_ACTION), NULL, _OVER);

×

3330

  dbObj.vgVersion++;

×

3331

  dbObj.updateTime = taosGetTimestampMs();

×

3332

  TAOS_CHECK_GOTO(mndAddDbStatusAction(pTrans, &dbObj, SDB_STATUS_READY, TRN_STAGE_COMMIT_ACTION), NULL, _OVER);

×

3334

  TAOS_CHECK_GOTO(mndTransPrepare(pMnode, pTrans), NULL, _OVER);

×

3335

  code = 0;

×

3337

_OVER:

×

3338

  taosArrayDestroy(pArray);

×

3339

  mndTransDrop(pTrans);

×

3340

  taosArrayDestroy(dbObj.cfg.pRetensions);

×

3341

  TAOS_RETURN(code);

×

3346

static int32_t mndProcessSplitVgroupMsg(SRpcMsg *pReq) { return mndProcessSplitVgroupMsgImp(pReq); }

×

3352

static int32_t mndSetBalanceVgroupInfoToTrans(SMnode *pMnode, STrans *pTrans, SDbObj *pDb, SVgObj *pVgroup,

×

3354

  int32_t code = 0;

×

3355

  SVgObj  newVg = {0};

×

3356

  memcpy(&newVg, pVgroup, sizeof(SVgObj));

×

3357

  mInfo("vgId:%d, vgroup info before balance, replica:%d", newVg.vgId, newVg.replica);

×

3358

  for (int32_t i = 0; i < newVg.replica; ++i) {

×

3359

    mInfo("vgId:%d, vnode:%d dnode:%d", newVg.vgId, i, newVg.vnodeGid[i].dnodeId);

×

3362

  TAOS_CHECK_RETURN(mndAddIncVgroupReplicaToTrans(pMnode, pTrans, pDb, &newVg, pDst->id));

×

3363

  TAOS_CHECK_RETURN(mndAddDecVgroupReplicaFromTrans(pMnode, pTrans, pDb, &newVg, pSrc->id));

×

3366

    SSdbRaw *pRaw = mndVgroupActionEncode(&newVg);

×

3367

    if (pRaw == NULL) {

×

3368

      code = TSDB_CODE_MND_RETURN_VALUE_NULL;

×

3369

      if (terrno != 0) code = terrno;

×

3370

      TAOS_RETURN(code);

×

3372

    if ((code = mndTransAppendCommitlog(pTrans, pRaw)) != 0) {

×

3373

      sdbFreeRaw(pRaw);

×

3374

      TAOS_RETURN(code);

×

3376

    code = sdbSetRawStatus(pRaw, SDB_STATUS_READY);

×

3377

    if (code != 0) {

×

3378

      mError("vgId:%d, failed to set raw status to ready, error:%s, line:%d", newVg.vgId, tstrerror(code), __LINE__);

×

3379

      TAOS_RETURN(code);

×

3383

  mInfo("vgId:%d, vgroup info after balance, replica:%d", newVg.vgId, newVg.replica);

×

3384

  for (int32_t i = 0; i < newVg.replica; ++i) {

×

3385

    mInfo("vgId:%d, vnode:%d dnode:%d", newVg.vgId, i, newVg.vnodeGid[i].dnodeId);

×

3387

  TAOS_RETURN(code);

×

3390

static int32_t mndBalanceVgroupBetweenDnode(SMnode *pMnode, STrans *pTrans, SDnodeObj *pSrc, SDnodeObj *pDst,

×

3392

  void   *pIter = NULL;

×

3393

  int32_t code = -1;

×

3394

  SSdb   *pSdb = pMnode->pSdb;

×

3396

  while (1) {

×

3397

    SVgObj *pVgroup = NULL;

×

3398

    pIter = sdbFetch(pSdb, SDB_VGROUP, pIter, (void **)&pVgroup);

×

3399

    if (pIter == NULL) break;

×

3400

    if (taosHashGet(pBalancedVgroups, &pVgroup->vgId, sizeof(int32_t)) != NULL) {

×

3401

      sdbRelease(pSdb, pVgroup);

×

3402

      continue;

×

3405

    bool existInSrc = false;

×

3406

    bool existInDst = false;

×

3407

    for (int32_t i = 0; i < pVgroup->replica; ++i) {

×

3408

      SVnodeGid *pGid = &pVgroup->vnodeGid[i];

×

3409

      if (pGid->dnodeId == pSrc->id) existInSrc = true;

×

3410

      if (pGid->dnodeId == pDst->id) existInDst = true;

×

3413

    if (!existInSrc || existInDst) {

×

3414

      sdbRelease(pSdb, pVgroup);

×

3415

      continue;

×

3418

    SDbObj *pDb = mndAcquireDb(pMnode, pVgroup->dbName);

×

3419

    if (pDb == NULL) {

×

3420

      code = TSDB_CODE_MND_RETURN_VALUE_NULL;

×

3421

      if (terrno != 0) code = terrno;

×

3422

      mError("vgId:%d, balance vgroup can't find db obj dbName:%s", pVgroup->vgId, pVgroup->dbName);

×

3423

      goto _OUT;

×

3426

    if (pDb->cfg.withArbitrator) {

×

3427

      mInfo("vgId:%d, db:%s, with arbitrator, balance vgroup not allowed", pVgroup->vgId, pVgroup->dbName);

×

3428

      goto _OUT;

×

3431

    code = mndSetBalanceVgroupInfoToTrans(pMnode, pTrans, pDb, pVgroup, pSrc, pDst);

×

3432

    if (code == 0) {

×

3433

      code = taosHashPut(pBalancedVgroups, &pVgroup->vgId, sizeof(int32_t), &pVgroup->vgId, sizeof(int32_t));

×

3436

  _OUT:

×

3437

    mndReleaseDb(pMnode, pDb);

×

3438

    sdbRelease(pSdb, pVgroup);

×

3439

    sdbCancelFetch(pSdb, pIter);

×

3440

    break;

×

3443

  return code;

×

3446

static int32_t mndBalanceVgroup(SMnode *pMnode, SRpcMsg *pReq, SArray *pArray) {

×

3447

  int32_t   code = -1;

×

3448

  int32_t   numOfVgroups = 0;

×

3449

  STrans   *pTrans = NULL;

×

3450

  SHashObj *pBalancedVgroups = NULL;

×

3452

  pBalancedVgroups = taosHashInit(16, taosGetDefaultHashFunction(TSDB_DATA_TYPE_INT), false, HASH_NO_LOCK);

×

3453

  if (pBalancedVgroups == NULL) goto _OVER;

×

3455

  pTrans = mndTransCreate(pMnode, TRN_POLICY_RETRY, TRN_CONFLICT_GLOBAL, pReq, "balance-vgroup");

×

3456

  if (pTrans == NULL) {

×

3457

    code = TSDB_CODE_MND_RETURN_VALUE_NULL;

×

3458

    if (terrno != 0) code = terrno;

×

3459

    goto _OVER;

×

3461

  mndTransSetSerial(pTrans);

×

3462

  mInfo("trans:%d, used to balance vgroup", pTrans->id);

×

3463

  TAOS_CHECK_GOTO(mndTransCheckConflict(pMnode, pTrans), NULL, _OVER);

×

3464

  TAOS_CHECK_GOTO(mndTransCheckConflictWithCompact(pMnode, pTrans), NULL, _OVER);

×

3466

  while (1) {

×

3467

    taosArraySort(pArray, (__compar_fn_t)mndCompareDnodeVnodes);

×

3468

    for (int32_t i = 0; i < taosArrayGetSize(pArray); ++i) {

×

3469

      SDnodeObj *pDnode = taosArrayGet(pArray, i);

×

3470

      mInfo("dnode:%d, equivalent vnodes:%d others:%d support:%d, score:%f", pDnode->id, pDnode->numOfVnodes,

×

3474

    SDnodeObj *pSrc = taosArrayGet(pArray, taosArrayGetSize(pArray) - 1);

×

3475

    SDnodeObj *pDst = taosArrayGet(pArray, 0);

×

3477

    float srcScore = mndGetDnodeScore(pSrc, -1, 1);

×

3478

    float dstScore = mndGetDnodeScore(pDst, 1, 1);

×

3479

    mInfo("trans:%d, after balance, src dnode:%d score:%f, dst dnode:%d score:%f", pTrans->id, pSrc->id, dstScore,

×

3482

    if (srcScore > dstScore - 0.000001) {

×

3483

      code = mndBalanceVgroupBetweenDnode(pMnode, pTrans, pSrc, pDst, pBalancedVgroups);

×

3484

      if (code == 0) {

×

3485

        pSrc->numOfVnodes--;

×

3486

        pDst->numOfVnodes++;

×

3487

        numOfVgroups++;

×

3488

        continue;

×

3490

        mInfo("trans:%d, no vgroup need to balance from dnode:%d to dnode:%d", pTrans->id, pSrc->id, pDst->id);

×

3491

        break;

×

3494

      mInfo("trans:%d, no vgroup need to balance any more", pTrans->id);

×

3495

      break;

×

3499

  if (numOfVgroups <= 0) {

×

3500

    mInfo("no need to balance vgroup");

×

3501

    code = 0;

×

3503

    mInfo("start to balance vgroup, numOfVgroups:%d", numOfVgroups);

×

3504

    if (mndTransPrepare(pMnode, pTrans) != 0) goto _OVER;

×

3505

    code = TSDB_CODE_ACTION_IN_PROGRESS;

×

3508

_OVER:

×

3509

  taosHashCleanup(pBalancedVgroups);

×

3510

  mndTransDrop(pTrans);

×

3511

  TAOS_RETURN(code);

×

3514

static int32_t mndProcessBalanceVgroupMsg(SRpcMsg *pReq) {

×

3515

  SMnode *pMnode = pReq->info.node;

×

3516

  int32_t code = -1;

×

3517

  SArray *pArray = NULL;

×

3518

  void   *pIter = NULL;

×

3519

  int64_t curMs = taosGetTimestampMs();

×

3521

  SBalanceVgroupReq req = {0};

×

3522

  if (tDeserializeSBalanceVgroupReq(pReq->pCont, pReq->contLen, &req) != 0) {

×

3523

    code = TSDB_CODE_INVALID_MSG;

×

3524

    goto _OVER;

×

3527

  mInfo("start to balance vgroup");

×

3528

  if ((code = mndCheckOperPrivilege(pMnode, pReq->info.conn.user, MND_OPER_BALANCE_VGROUP)) != 0) {

×

3529

    goto _OVER;

×

3532

  while (1) {

×

3533

    SDnodeObj *pDnode = NULL;

×

3534

    pIter = sdbFetch(pMnode->pSdb, SDB_DNODE, pIter, (void **)&pDnode);

×

3535

    if (pIter == NULL) break;

×

3536

    if (!mndIsDnodeOnline(pDnode, curMs)) {

×

3537

      sdbCancelFetch(pMnode->pSdb, pIter);

×

3538

      code = TSDB_CODE_MND_HAS_OFFLINE_DNODE;

×

3539

      mError("failed to balance vgroup since %s, dnode:%d", terrstr(), pDnode->id);

×

3540

      sdbRelease(pMnode->pSdb, pDnode);

×

3541

      goto _OVER;

×

3544

    sdbRelease(pMnode->pSdb, pDnode);

×

3547

  pArray = mndBuildDnodesArray(pMnode, 0, NULL);

×

3548

  if (pArray == NULL) {

×

3549

    code = TSDB_CODE_MND_RETURN_VALUE_NULL;

×

3550

    if (terrno != 0) code = terrno;

×

3551

    goto _OVER;

×

3554

  if (taosArrayGetSize(pArray) < 2) {

×

3555

    mInfo("no need to balance vgroup since dnode num less than 2");

×

3556

    code = 0;

×

3558

    code = mndBalanceVgroup(pMnode, pReq, pArray);

×

3561

  auditRecord(pReq, pMnode->clusterId, "balanceVgroup", "", "", req.sql, req.sqlLen);

×

3563

_OVER:

×

3564

  if (code != 0 && code != TSDB_CODE_ACTION_IN_PROGRESS) {

×

3565

    mError("failed to balance vgroup since %s", tstrerror(code));

×

3568

  taosArrayDestroy(pArray);

×

3569

  tFreeSBalanceVgroupReq(&req);

×

3570

  TAOS_RETURN(code);

×

3573

bool mndVgroupInDb(SVgObj *pVgroup, int64_t dbUid) { return !pVgroup->isTsma && pVgroup->dbUid == dbUid; }

×

3575

bool mndVgroupInDnode(SVgObj *pVgroup, int32_t dnodeId) {

×

3576

  for (int i = 0; i < pVgroup->replica; i++) {

×

3577

    if (pVgroup->vnodeGid[i].dnodeId == dnodeId) return true;

×

3579

  return false;

×

3582

static void *mndBuildCompactVnodeReq(SMnode *pMnode, SDbObj *pDb, SVgObj *pVgroup, int32_t *pContLen, int64_t compactTs,

×

3584

  SCompactVnodeReq compactReq = {0};

×

3585

  compactReq.dbUid = pDb->uid;

×

3586

  compactReq.compactStartTime = compactTs;

×

3587

  compactReq.tw = tw;

×

3588

  tstrncpy(compactReq.db, pDb->name, TSDB_DB_FNAME_LEN);

×

3590

  mInfo("vgId:%d, build compact vnode config req", pVgroup->vgId);

×

3591

  int32_t contLen = tSerializeSCompactVnodeReq(NULL, 0, &compactReq);

×

3592

  if (contLen < 0) {

×

3593

    terrno = TSDB_CODE_OUT_OF_MEMORY;

×

3594

    return NULL;

×

3596

  contLen += sizeof(SMsgHead);

×

3598

  void *pReq = taosMemoryMalloc(contLen);

×

3599

  if (pReq == NULL) {

×

3600

    terrno = TSDB_CODE_OUT_OF_MEMORY;

×

3601

    return NULL;

×

3604

  SMsgHead *pHead = pReq;

×

3605

  pHead->contLen = htonl(contLen);

×

3606

  pHead->vgId = htonl(pVgroup->vgId);

×

3608

  if (tSerializeSCompactVnodeReq((char *)pReq + sizeof(SMsgHead), contLen, &compactReq) < 0) {

×

3609

    taosMemoryFree(pReq);

×

3610

    terrno = TSDB_CODE_OUT_OF_MEMORY;

×

3611

    return NULL;

×

3613

  *pContLen = contLen;

×

3614

  return pReq;

×

3617

static int32_t mndAddCompactVnodeAction(SMnode *pMnode, STrans *pTrans, SDbObj *pDb, SVgObj *pVgroup, int64_t compactTs,

×

3619

  int32_t      code = 0;

×

3620

  STransAction action = {0};

×

3621

  action.epSet = mndGetVgroupEpset(pMnode, pVgroup);

×

3623

  int32_t contLen = 0;

×

3624

  void   *pReq = mndBuildCompactVnodeReq(pMnode, pDb, pVgroup, &contLen, compactTs, tw);

×

3625

  if (pReq == NULL) {

×

3626

    code = TSDB_CODE_MND_RETURN_VALUE_NULL;

×

3627

    if (terrno != 0) code = terrno;

×

3628

    TAOS_RETURN(code);

×

3631

  action.pCont = pReq;

×

3632

  action.contLen = contLen;

×

3633

  action.msgType = TDMT_VND_COMPACT;

×

3635

  if ((code = mndTransAppendRedoAction(pTrans, &action)) != 0) {

×

3636

    taosMemoryFree(pReq);

×

3637

    TAOS_RETURN(code);

×

3640

  TAOS_RETURN(code);

×

3643

int32_t mndBuildCompactVgroupAction(SMnode *pMnode, STrans *pTrans, SDbObj *pDb, SVgObj *pVgroup, int64_t compactTs,

×

3645

  TAOS_CHECK_RETURN(mndAddCompactVnodeAction(pMnode, pTrans, pDb, pVgroup, compactTs, tw));

×

3646

  return 0;

×

taosdata / TDengine / #3599

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

taosdata / TDengine / #3599

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