Merge pull request #20413 from taosdata/FIX/TD-22564-main

enh: not allow to propose if fsm applying progress lagging behind too far

include/libs/sync/sync.h

@@ -143,10 +143,11 @@ typedef struct SSyncFSM {
   void* data;
 
   int32_t (*FpCommitCb)(const struct SSyncFSM* pFsm, SRpcMsg* pMsg, const SFsmCbMeta* pMeta);
+  SyncIndex (*FpAppliedIndexCb)(const struct SSyncFSM* pFsm);
   int32_t (*FpPreCommitCb)(const struct SSyncFSM* pFsm, SRpcMsg* pMsg, const SFsmCbMeta* pMeta);
   void (*FpRollBackCb)(const struct SSyncFSM* pFsm, SRpcMsg* pMsg, const SFsmCbMeta* pMeta);
 
-  void (*FpRestoreFinishCb)(const struct SSyncFSM* pFsm);
+  void (*FpRestoreFinishCb)(const struct SSyncFSM* pFsm, const SyncIndex commitIdx);
   void (*FpReConfigCb)(const struct SSyncFSM* pFsm, SRpcMsg* pMsg, const SReConfigCbMeta* pMeta);
   void (*FpLeaderTransferCb)(const struct SSyncFSM* pFsm, SRpcMsg* pMsg, const SFsmCbMeta* pMeta);
   bool (*FpApplyQueueEmptyCb)(const struct SSyncFSM* pFsm);

include/util/taoserror.h

@@ -542,7 +542,8 @@ int32_t* taosGetErrno();
 #define TSDB_CODE_SYN_BATCH_ERROR               TAOS_DEF_ERROR_CODE(0, 0x0913)
 #define TSDB_CODE_SYN_RESTORING                 TAOS_DEF_ERROR_CODE(0, 0x0914)
 #define TSDB_CODE_SYN_INVALID_SNAPSHOT_MSG      TAOS_DEF_ERROR_CODE(0, 0x0915)  // internal
-#define TSDB_CODE_SYN_BUFFER_FULL               TAOS_DEF_ERROR_CODE(0, 0x0916)  //
+#define TSDB_CODE_SYN_BUFFER_FULL               TAOS_DEF_ERROR_CODE(0, 0x0916)
+#define TSDB_CODE_SYN_WRITE_STALL               TAOS_DEF_ERROR_CODE(0, 0x0917)
 #define TSDB_CODE_SYN_INTERNAL_ERROR            TAOS_DEF_ERROR_CODE(0, 0x09FF)
 
 // tq

source/dnode/mnode/impl/inc/mndInt.h

@@ -113,6 +113,7 @@ typedef struct SMnode {
   bool           deploy;
   char          *path;
   int64_t        checkTime;
+  SyncIndex      applied;
   SSdb          *pSdb;
   SArray        *pSteps;
   SQHandle      *pQuery;

source/dnode/mnode/impl/src/mndMain.c

@@ -380,11 +380,13 @@ static int32_t mndInitSdb(SMnode *pMnode) {
 }
 
 static int32_t mndOpenSdb(SMnode *pMnode) {
+  int32_t code = 0;
   if (!pMnode->deploy) {
-    return sdbReadFile(pMnode->pSdb);
+    code = sdbReadFile(pMnode->pSdb);
-  } else {
-    return 0;
   }
+
+  atomic_store_64(&pMnode->applied, pMnode->pSdb->commitIndex);
+  return code;
 }
 
 static void mndCleanupSdb(SMnode *pMnode) {

source/dnode/mnode/impl/src/mndSync.c

@@ -129,6 +129,14 @@ int32_t mndProcessWriteMsg(const SSyncFSM *pFsm, SRpcMsg *pMsg, const SFsmCbMeta
 
 int32_t mndSyncCommitMsg(const SSyncFSM *pFsm, SRpcMsg *pMsg, const SFsmCbMeta *pMeta) {
   int32_t code = 0;
+  pMsg->info.conn.applyIndex = pMeta->index;
+  pMsg->info.conn.applyTerm = pMeta->term;
+
+  if (pMsg->code == 0) {
+    SMnode *pMnode = pFsm->data;
+    atomic_store_64(&pMnode->applied, pMsg->info.conn.applyIndex);
+  }
+
   if (!syncUtilUserCommit(pMsg->msgType)) {
     goto _out;
   }
@@ -140,6 +148,11 @@ _out:
   return code;
 }
 
+SyncIndex mndSyncAppliedIndex(const SSyncFSM *pFSM) {
+  SMnode *pMnode = pFSM->data;
+  return atomic_load_64(&pMnode->applied);
+}
+
 int32_t mndSyncGetSnapshot(const SSyncFSM *pFsm, SSnapshot *pSnapshot, void *pReaderParam, void **ppReader) {
   mInfo("start to read snapshot from sdb in atomic way");
   SMnode *pMnode = pFsm->data;
@@ -153,7 +166,7 @@ static void mndSyncGetSnapshotInfo(const SSyncFSM *pFsm, SSnapshot *pSnapshot) {
   sdbGetCommitInfo(pMnode->pSdb, &pSnapshot->lastApplyIndex, &pSnapshot->lastApplyTerm, &pSnapshot->lastConfigIndex);
 }
 
-void mndRestoreFinish(const SSyncFSM *pFsm) {
+void mndRestoreFinish(const SSyncFSM *pFsm, const SyncIndex commitIdx) {
   SMnode *pMnode = pFsm->data;
 
   if (!pMnode->deploy) {
@@ -167,6 +180,8 @@ void mndRestoreFinish(const SSyncFSM *pFsm) {
   } else {
     mInfo("vgId:1, sync restore finished");
   }
+
+  ASSERT(commitIdx == mndSyncAppliedIndex(pFsm));
 }
 
 int32_t mndSnapshotStartRead(const SSyncFSM *pFsm, void *pParam, void **ppReader) {
@@ -253,6 +268,7 @@ SSyncFSM *mndSyncMakeFsm(SMnode *pMnode) {
   SSyncFSM *pFsm = taosMemoryCalloc(1, sizeof(SSyncFSM));
   pFsm->data = pMnode;
   pFsm->FpCommitCb = mndSyncCommitMsg;
+  pFsm->FpAppliedIndexCb = mndSyncAppliedIndex;
   pFsm->FpPreCommitCb = NULL;
   pFsm->FpRollBackCb = NULL;
   pFsm->FpRestoreFinishCb = mndRestoreFinish;

source/dnode/vnode/src/vnd/vnodeSvr.c

@@ -306,13 +306,7 @@ int32_t vnodeProcessWriteMsg(SVnode *pVnode, SRpcMsg *pMsg, int64_t version, SRp
   void   *pReq;
   int32_t len;
   int32_t ret;
-  /*
+
-  if (!pVnode->inUse) {
-    terrno = TSDB_CODE_VND_NO_AVAIL_BUFPOOL;
-    vError("vgId:%d, not ready to write since %s", TD_VID(pVnode), terrstr());
-    return -1;
-  }
-  */
   if (version <= pVnode->state.applied) {
     vError("vgId:%d, duplicate write request. version: %" PRId64 ", applied: %" PRId64 "", TD_VID(pVnode), version,
            pVnode->state.applied);
@@ -326,8 +320,8 @@ int32_t vnodeProcessWriteMsg(SVnode *pVnode, SRpcMsg *pMsg, int64_t version, SRp
   ASSERT(pVnode->state.applyTerm <= pMsg->info.conn.applyTerm);
   ASSERT(pVnode->state.applied + 1 == version);
 
-  pVnode->state.applied = version;
+  atomic_store_64(&pVnode->state.applied, version);
-  pVnode->state.applyTerm = pMsg->info.conn.applyTerm;
+  atomic_store_64(&pVnode->state.applyTerm, pMsg->info.conn.applyTerm);
 
   if (!syncUtilUserCommit(pMsg->msgType)) goto _exit;
 

source/dnode/vnode/src/vnd/vnodeSync.c

@@ -433,7 +433,23 @@ static int32_t vnodeSyncApplyMsg(const SSyncFSM *pFsm, SRpcMsg *pMsg, const SFsm
 }
 
 static int32_t vnodeSyncCommitMsg(const SSyncFSM *pFsm, SRpcMsg *pMsg, const SFsmCbMeta *pMeta) {
-  return vnodeSyncApplyMsg(pFsm, pMsg, pMeta);
+  if (pMsg->code == 0) {
+    return vnodeSyncApplyMsg(pFsm, pMsg, pMeta);
+  }
+
+  const STraceId *trace = &pMsg->info.traceId;
+  SVnode         *pVnode = pFsm->data;
+  vnodePostBlockMsg(pVnode, pMsg);
+
+  SRpcMsg rsp = {.code = pMsg->code, .info = pMsg->info};
+  if (rsp.info.handle != NULL) {
+    tmsgSendRsp(&rsp);
+  }
+
+  vGTrace("vgId:%d, msg:%p is freed, code:0x%x index:%" PRId64, TD_VID(pVnode), pMsg, rsp.code, pMeta->index);
+  rpcFreeCont(pMsg->pCont);
+  pMsg->pCont = NULL;
+  return 0;
 }
 
 static int32_t vnodeSyncPreCommitMsg(const SSyncFSM *pFsm, SRpcMsg *pMsg, const SFsmCbMeta *pMeta) {
@@ -443,6 +459,11 @@ static int32_t vnodeSyncPreCommitMsg(const SSyncFSM *pFsm, SRpcMsg *pMsg, const
   return 0;
 }
 
+static SyncIndex vnodeSyncAppliedIndex(const SSyncFSM *pFSM) {
+  SVnode *pVnode = pFSM->data;
+  return atomic_load_64(&pVnode->state.applied);
+}
+
 static void vnodeSyncRollBackMsg(const SSyncFSM *pFsm, SRpcMsg *pMsg, const SFsmCbMeta *pMeta) {
   SVnode *pVnode = pFsm->data;
   vTrace("vgId:%d, rollback-cb is excuted, fsm:%p, index:%" PRId64 ", weak:%d, code:%d, state:%d %s, type:%s",
@@ -505,21 +526,26 @@ static int32_t vnodeSnapshotDoWrite(const SSyncFSM *pFsm, void *pWriter, void *p
   return code;
 }
 
-static void vnodeRestoreFinish(const SSyncFSM *pFsm) {
+static void vnodeRestoreFinish(const SSyncFSM *pFsm, const SyncIndex commitIdx) {
   SVnode *pVnode = pFsm->data;
+  SyncIndex appliedIdx = -1;
 
   do {
-    int32_t itemSize = tmsgGetQueueSize(&pVnode->msgCb, pVnode->config.vgId, APPLY_QUEUE);
+    appliedIdx = vnodeSyncAppliedIndex(pFsm);
-    if (itemSize == 0) {
+    ASSERT(appliedIdx <= commitIdx);
-      vInfo("vgId:%d, apply queue is empty, restore finish", pVnode->config.vgId);
+    if (appliedIdx == commitIdx) {
+      vInfo("vgId:%d, no items to be applied, restore finish", pVnode->config.vgId);
       break;
     } else {
-      vInfo("vgId:%d, restore not finish since %d items in apply queue", pVnode->config.vgId, itemSize);
+      vInfo("vgId:%d, restore not finish since %" PRId64 " items to be applied. commit-index:%" PRId64
+            ", applied-index:%" PRId64,
+            pVnode->config.vgId, commitIdx - appliedIdx, commitIdx, appliedIdx);
       taosMsleep(10);
     }
   } while (true);
 
-  walApplyVer(pVnode->pWal, pVnode->state.applied);
+  ASSERT(commitIdx == vnodeSyncAppliedIndex(pFsm));
+  walApplyVer(pVnode->pWal, commitIdx);
 
   pVnode->restored = true;
   vInfo("vgId:%d, sync restore finished", pVnode->config.vgId);
@@ -569,6 +595,7 @@ static SSyncFSM *vnodeSyncMakeFsm(SVnode *pVnode) {
   SSyncFSM *pFsm = taosMemoryCalloc(1, sizeof(SSyncFSM));
   pFsm->data = pVnode;
   pFsm->FpCommitCb = vnodeSyncCommitMsg;
+  pFsm->FpAppliedIndexCb = vnodeSyncAppliedIndex;
   pFsm->FpPreCommitCb = vnodeSyncPreCommitMsg;
   pFsm->FpRollBackCb = vnodeSyncRollBackMsg;
   pFsm->FpGetSnapshotInfo = vnodeSyncGetSnapshotInfo;

source/libs/sync/src/syncMain.c

@@ -2160,8 +2160,8 @@ int32_t syncNodeAppend(SSyncNode* ths, SSyncRaftEntry* pEntry) {
   // append to log buffer
   if (syncLogBufferAppend(ths->pLogBuf, ths, pEntry) < 0) {
     sError("vgId:%d, failed to enqueue sync log buffer, index:%" PRId64, ths->vgId, pEntry->index);
-    terrno = TSDB_CODE_SYN_BUFFER_FULL;
+    ASSERT(terrno != 0);
-    (void)syncLogFsmExecute(ths, ths->pFsm, ths->state, raftStoreGetTerm(ths), pEntry, TSDB_CODE_SYN_BUFFER_FULL);
+    (void)syncLogFsmExecute(ths, ths->pFsm, ths->state, raftStoreGetTerm(ths), pEntry, terrno);
     syncEntryDestroy(pEntry);
     return -1;
   }

source/libs/sync/src/syncPipeline.c

@@ -48,7 +48,16 @@ int32_t syncLogBufferAppend(SSyncLogBuffer* pBuf, SSyncNode* pNode, SSyncRaftEnt
   SyncIndex index = pEntry->index;
 
   if (index - pBuf->startIndex >= pBuf->size) {
-    sError("vgId:%d, failed to append due to sync log buffer full. index:%" PRId64 "", pNode->vgId, index);
+    terrno = TSDB_CODE_SYN_BUFFER_FULL;
+    sError("vgId:%d, failed to append since %s. index:%" PRId64 "", pNode->vgId, terrstr(), index);
+    goto _err;
+  }
+
+  SyncIndex appliedIndex = pNode->pFsm->FpAppliedIndexCb(pNode->pFsm);
+  if (pNode->restoreFinish && pBuf->commitIndex - appliedIndex >= pBuf->size) {
+    terrno = TSDB_CODE_SYN_WRITE_STALL;
+    sError("vgId:%d, failed to append since %s. index:%" PRId64 ", commit-index:%" PRId64 ", applied-index:%" PRId64,
+           pNode->vgId, terrstr(), index, pBuf->commitIndex, appliedIndex);
     goto _err;
   }
 
@@ -475,7 +484,7 @@ _out:
 
 int32_t syncLogFsmExecute(SSyncNode* pNode, SSyncFSM* pFsm, ESyncState role, SyncTerm term, SSyncRaftEntry* pEntry,
                           int32_t applyCode) {
-  if ((pNode->replicaNum == 1) && pNode->restoreFinish && pNode->vgId != 1) {
+  if (pNode->replicaNum == 1 && pNode->restoreFinish && pNode->vgId != 1) {
     return 0;
   }
 
@@ -587,10 +596,10 @@ _out:
   // mark as restored if needed
   if (!pNode->restoreFinish && pBuf->commitIndex >= pNode->commitIndex && pEntry != NULL &&
       currentTerm <= pEntry->term) {
-    pNode->pFsm->FpRestoreFinishCb(pNode->pFsm);
+    pNode->pFsm->FpRestoreFinishCb(pNode->pFsm, pBuf->commitIndex);
     pNode->restoreFinish = true;
-    sInfo("vgId:%d, restore finished. log buffer: [%" PRId64 " %" PRId64 " %" PRId64 ", %" PRId64 ")", pNode->vgId,
+    sInfo("vgId:%d, restore finished. term:%" PRId64 ", log buffer: [%" PRId64 " %" PRId64 " %" PRId64 ", %" PRId64 ")",
-          pBuf->startIndex, pBuf->commitIndex, pBuf->matchIndex, pBuf->endIndex);
+          pNode->vgId, currentTerm, pBuf->startIndex, pBuf->commitIndex, pBuf->matchIndex, pBuf->endIndex);
   }
 
   if (!inBuf) {

source/util/src/terror.c

@@ -423,6 +423,7 @@ TAOS_DEFINE_ERROR(TSDB_CODE_SYN_BATCH_ERROR,              "Sync batch error")
 TAOS_DEFINE_ERROR(TSDB_CODE_SYN_RESTORING,                "Sync leader is restoring")
 TAOS_DEFINE_ERROR(TSDB_CODE_SYN_INVALID_SNAPSHOT_MSG,     "Sync invalid snapshot msg")
 TAOS_DEFINE_ERROR(TSDB_CODE_SYN_BUFFER_FULL,              "Sync buffer is full")
+TAOS_DEFINE_ERROR(TSDB_CODE_SYN_WRITE_STALL,              "Sync write stall")
 TAOS_DEFINE_ERROR(TSDB_CODE_SYN_INTERNAL_ERROR,           "Sync internal error")
 
 //tq