sync refactor

source/libs/sync/src/syncMain.c

@@ -37,15 +37,16 @@ static int32_t tsNodeRefId = -1;
 
 // ------ local funciton ---------
 // enqueue message ----
-static void syncNodeEqPingTimer(void* param, void* tmrId);
-static void syncNodeEqElectTimer(void* param, void* tmrId);
-static void syncNodeEqHeartbeatTimer(void* param, void* tmrId);
+static void    syncNodeEqPingTimer(void* param, void* tmrId);
+static void    syncNodeEqElectTimer(void* param, void* tmrId);
+static void    syncNodeEqHeartbeatTimer(void* param, void* tmrId);
+static int32_t syncNodeEqNoop(SSyncNode* ths);
+static int32_t syncNodeAppendNoop(SSyncNode* ths);
 
 // process message ----
 static int32_t syncNodeOnPingCb(SSyncNode* ths, SyncPing* pMsg);
 static int32_t syncNodeOnPingReplyCb(SSyncNode* ths, SyncPingReply* pMsg);
 static int32_t syncNodeOnClientRequestCb(SSyncNode* ths, SyncClientRequest* pMsg);
-static int32_t syncNodeAppendNoop(SSyncNode* ths);
 
 // life cycle
 static void syncFreeNode(void* param);
@@ -673,6 +674,7 @@ void syncNodeCandidate2Leader(SSyncNode* pSyncNode) {
 
   // Raft 3.6.2 Committing entries from previous terms
   syncNodeAppendNoop(pSyncNode);
+  // syncNodeEqNoop(pSyncNode);
 }
 
 void syncNodeFollower2Candidate(SSyncNode* pSyncNode) {
@@ -807,6 +809,47 @@ static void syncNodeEqHeartbeatTimer(void* param, void* tmrId) {
   }
 }
 
+static int32_t syncNodeEqNoop(SSyncNode* ths) {
+  int32_t ret = 0;
+  assert(ths->state == TAOS_SYNC_STATE_LEADER);
+
+  SyncIndex       index = ths->pLogStore->getLastIndex(ths->pLogStore) + 1;
+  SyncTerm        term = ths->pRaftStore->currentTerm;
+  SSyncRaftEntry* pEntry = syncEntryBuildNoop(term, index);
+  assert(pEntry != NULL);
+
+  uint32_t           entryLen;
+  char*              serialized = syncEntrySerialize(pEntry, &entryLen);
+  SyncClientRequest* pSyncMsg = syncClientRequestBuild(entryLen);
+  assert(pSyncMsg->dataLen == entryLen);
+  memcpy(pSyncMsg->data, serialized, entryLen);
+
+  SRpcMsg rpcMsg;
+  syncClientRequest2RpcMsg(pSyncMsg, &rpcMsg);
+  ths->FpEqMsg(ths->queue, &rpcMsg);
+
+  free(serialized);
+  syncClientRequestDestroy(pSyncMsg);
+
+  return ret;
+}
+
+static int32_t syncNodeAppendNoop(SSyncNode* ths) {
+  int32_t ret = 0;
+
+  SyncIndex       index = ths->pLogStore->getLastIndex(ths->pLogStore) + 1;
+  SyncTerm        term = ths->pRaftStore->currentTerm;
+  SSyncRaftEntry* pEntry = syncEntryBuildNoop(term, index);
+  assert(pEntry != NULL);
+
+  if (ths->state == TAOS_SYNC_STATE_LEADER) {
+    ths->pLogStore->appendEntry(ths->pLogStore, pEntry);
+    syncNodeReplicate(ths);
+  }
+
+  return ret;
+}
+
 // on message ----
 static int32_t syncNodeOnPingCb(SSyncNode* ths, SyncPing* pMsg) {
   int32_t ret = 0;
@@ -881,22 +924,6 @@ static int32_t syncNodeOnClientRequestCb(SSyncNode* ths, SyncClientRequest* pMsg
   return ret;
 }
 
-static int32_t syncNodeAppendNoop(SSyncNode* ths) {
-  int32_t ret = 0;
-
-  SyncIndex       index = ths->pLogStore->getLastIndex(ths->pLogStore) + 1;
-  SyncTerm        term = ths->pRaftStore->currentTerm;
-  SSyncRaftEntry* pEntry = syncEntryBuildNoop(term, index);
-  assert(pEntry != NULL);
-
-  if (ths->state == TAOS_SYNC_STATE_LEADER) {
-    ths->pLogStore->appendEntry(ths->pLogStore, pEntry);
-    syncNodeReplicate(ths);
-  }
-
-  return ret;
-}
-
 static void syncFreeNode(void* param) {
   SSyncNode* pNode = param;
   syncNodePrint2((char*)"==syncFreeNode==", pNode);

source/libs/sync/src/syncRaftLog.c

@@ -48,14 +48,18 @@ int32_t logStoreAppendEntry(SSyncLogStore* pLogStore, SSyncRaftEntry* pEntry) {
   SSyncLogStoreData* pData = pLogStore->data;
   SWal*              pWal = pData->pWal;
 
-  assert(pEntry->index == logStoreLastIndex(pLogStore) + 1);
+  SyncIndex lastIndex = logStoreLastIndex(pLogStore);
+  assert(pEntry->index == lastIndex + 1);
   uint32_t len;
   char*    serialized = syncEntrySerialize(pEntry, &len);
   assert(serialized != NULL);
 
-  int code;
-  code = walWrite(pWal, pEntry->index, pEntry->entryType, serialized, len);
-  assert(code == 0);
+  int code = 0;
+  /*
+    code = walWrite(pWal, pEntry->index, pEntry->entryType, serialized, len);
+    assert(code == 0);
+  */
+  assert(walWrite(pWal, pEntry->index, pEntry->entryType, serialized, len) == 0);
 
   walFsync(pWal, true);
   free(serialized);
@@ -69,7 +73,7 @@ SSyncRaftEntry* logStoreGetEntry(SSyncLogStore* pLogStore, SyncIndex index) {
 
   if (index >= SYNC_INDEX_BEGIN && index <= logStoreLastIndex(pLogStore)) {
     SWalReadHandle* pWalHandle = walOpenReadHandle(pWal);
-    walReadWithHandle(pWalHandle, index);
+    assert(walReadWithHandle(pWalHandle, index) == 0);
     pEntry = syncEntryDeserialize(pWalHandle->pHead->head.body, pWalHandle->pHead->head.len);
     assert(pEntry != NULL);
 

source/libs/sync/test/CMakeLists.txt

@@ -34,6 +34,7 @@ add_executable(syncEncodeTest "")
 add_executable(syncWriteTest "")
 add_executable(syncReplicateTest "")
 add_executable(syncReplicateTest2 "")
+add_executable(syncReplicateTest3 "")
 add_executable(syncReplicateLoadTest "")
 add_executable(syncRefTest "")
 add_executable(syncLogStoreCheck "")
@@ -183,6 +184,10 @@ target_sources(syncReplicateTest2
     PRIVATE
     "syncReplicateTest2.cpp"
 )
+target_sources(syncReplicateTest3
+    PRIVATE
+    "syncReplicateTest3.cpp"
+)
 target_sources(syncReplicateLoadTest
     PRIVATE
     "syncReplicateLoadTest.cpp"
@@ -377,6 +382,11 @@ target_include_directories(syncReplicateTest2
     "${CMAKE_SOURCE_DIR}/include/libs/sync"
     "${CMAKE_CURRENT_SOURCE_DIR}/../inc"
 )
+target_include_directories(syncReplicateTest3
+    PUBLIC
+    "${CMAKE_SOURCE_DIR}/include/libs/sync"
+    "${CMAKE_CURRENT_SOURCE_DIR}/../inc"
+)
 target_include_directories(syncReplicateLoadTest
     PUBLIC
     "${CMAKE_SOURCE_DIR}/include/libs/sync"
@@ -538,6 +548,10 @@ target_link_libraries(syncReplicateTest2
     sync
     gtest_main
 )
+target_link_libraries(syncReplicateTest3
+    sync
+    gtest_main
+)
 target_link_libraries(syncReplicateLoadTest
     sync
     gtest_main

source/libs/sync/test/syncEntryTest.cpp

@@ -90,7 +90,7 @@ int main(int argc, char** argv) {
 
   test1();
   test2();
-  test3();  
+  test3();
   test4();
 
   return 0;

source/libs/sync/test/syncReplicateLoadTest.cpp

@@ -178,9 +178,10 @@ int main(int argc, char **argv) {
   while (1) {
     sTrace(
         "replicate sleep, state: %d, %s, term:%lu electTimerLogicClock:%lu, electTimerLogicClockUser:%lu, "
-        "electTimerMS:%d",
+        "electTimerMS:%d, commitIndex:%ld",
         pSyncNode->state, syncUtilState2String(pSyncNode->state), pSyncNode->pRaftStore->currentTerm,
-        pSyncNode->electTimerLogicClock, pSyncNode->electTimerLogicClockUser, pSyncNode->electTimerMS);
+        pSyncNode->electTimerLogicClock, pSyncNode->electTimerLogicClockUser, pSyncNode->electTimerMS,
+        pSyncNode->commitIndex);
     taosMsleep(1000);
   }
 

source/libs/sync/test/syncReplicateTest.cpp

@@ -183,18 +183,20 @@ int main(int argc, char **argv) {
 
     taosMsleep(1000);
     sTrace(
-        "replicate sleep, state: %d, %s, term:%lu electTimerLogicClock:%lu, electTimerLogicClockUser:%lu, "
-        "electTimerMS:%d",
+        "syncPropose sleep, state: %d, %s, term:%lu electTimerLogicClock:%lu, electTimerLogicClockUser:%lu, "
+        "electTimerMS:%d, commitIndex:%ld",
         gSyncNode->state, syncUtilState2String(gSyncNode->state), gSyncNode->pRaftStore->currentTerm,
-        gSyncNode->electTimerLogicClock, gSyncNode->electTimerLogicClockUser, gSyncNode->electTimerMS);
+        gSyncNode->electTimerLogicClock, gSyncNode->electTimerLogicClockUser, gSyncNode->electTimerMS,
+        gSyncNode->commitIndex);
   }
 
   while (1) {
     sTrace(
         "replicate sleep, state: %d, %s, term:%lu electTimerLogicClock:%lu, electTimerLogicClockUser:%lu, "
-        "electTimerMS:%d",
+        "electTimerMS:%d, commitIndex:%ld",
         gSyncNode->state, syncUtilState2String(gSyncNode->state), gSyncNode->pRaftStore->currentTerm,
-        gSyncNode->electTimerLogicClock, gSyncNode->electTimerLogicClockUser, gSyncNode->electTimerMS);
+        gSyncNode->electTimerLogicClock, gSyncNode->electTimerLogicClockUser, gSyncNode->electTimerMS,
+        gSyncNode->commitIndex);
     taosMsleep(1000);
   }
 

source/libs/sync/test/syncReplicateTest2.cpp

@@ -185,17 +185,19 @@ int main(int argc, char **argv) {
 
     sTrace(
         "syncPropose sleep, state: %d, %s, term:%lu electTimerLogicClock:%lu, electTimerLogicClockUser:%lu, "
-        "electTimerMS:%d",
+        "electTimerMS:%d, commitIndex:%ld",
         pSyncNode->state, syncUtilState2String(pSyncNode->state), pSyncNode->pRaftStore->currentTerm,
-        pSyncNode->electTimerLogicClock, pSyncNode->electTimerLogicClockUser, pSyncNode->electTimerMS);
+        pSyncNode->electTimerLogicClock, pSyncNode->electTimerLogicClockUser, pSyncNode->electTimerMS,
+        pSyncNode->commitIndex);
   }
 
   while (1) {
     sTrace(
         "replicate sleep, state: %d, %s, term:%lu electTimerLogicClock:%lu, electTimerLogicClockUser:%lu, "
-        "electTimerMS:%d",
+        "electTimerMS:%d, commitIndex:%ld",
         pSyncNode->state, syncUtilState2String(pSyncNode->state), pSyncNode->pRaftStore->currentTerm,
-        pSyncNode->electTimerLogicClock, pSyncNode->electTimerLogicClockUser, pSyncNode->electTimerMS);
+        pSyncNode->electTimerLogicClock, pSyncNode->electTimerLogicClockUser, pSyncNode->electTimerMS,
+        pSyncNode->commitIndex);
     taosMsleep(1000);
   }
 

source/libs/sync/test/syncReplicateTest3.cpp

@@ -0,0 +1,217 @@
+#define ALLOW_FORBID_FUNC
+
+#include <gtest/gtest.h>
+#include <stdio.h>
+#include "syncEnv.h"
+#include "syncIO.h"
+#include "syncInt.h"
+#include "syncMessage.h"
+#include "syncRaftEntry.h"
+#include "syncRaftLog.h"
+#include "syncRaftStore.h"
+#include "syncUtil.h"
+
+void logTest() {
+  sTrace("--- sync log test: trace");
+  sDebug("--- sync log test: debug");
+  sInfo("--- sync log test: info");
+  sWarn("--- sync log test: warn");
+  sError("--- sync log test: error");
+  sFatal("--- sync log test: fatal");
+}
+
+uint16_t ports[] = {7010, 7110, 7210, 7310, 7410};
+int32_t  replicaNum = 3;
+int32_t  myIndex = 0;
+
+SRaftId   ids[TSDB_MAX_REPLICA];
+SSyncInfo syncInfo;
+SSyncFSM *pFsm;
+SWal *    pWal;
+
+void CommitCb(struct SSyncFSM *pFsm, const SRpcMsg *pMsg, SyncIndex index, bool isWeak, int32_t code,
+              ESyncState state) {
+  char logBuf[256];
+  snprintf(logBuf, sizeof(logBuf), "==callback== ==CommitCb== pFsm:%p, index:%ld, isWeak:%d, code:%d, state:%d %s \n",
+           pFsm, index, isWeak, code, state, syncUtilState2String(state));
+  syncRpcMsgPrint2(logBuf, (SRpcMsg *)pMsg);
+}
+
+void PreCommitCb(struct SSyncFSM *pFsm, const SRpcMsg *pMsg, SyncIndex index, bool isWeak, int32_t code,
+                 ESyncState state) {
+  char logBuf[256];
+  snprintf(logBuf, sizeof(logBuf),
+           "==callback== ==PreCommitCb== pFsm:%p, index:%ld, isWeak:%d, code:%d, state:%d %s \n", pFsm, index, isWeak,
+           code, state, syncUtilState2String(state));
+  syncRpcMsgPrint2(logBuf, (SRpcMsg *)pMsg);
+}
+
+void RollBackCb(struct SSyncFSM *pFsm, const SRpcMsg *pMsg, SyncIndex index, bool isWeak, int32_t code,
+                ESyncState state) {
+  char logBuf[256];
+  snprintf(logBuf, sizeof(logBuf), "==callback== ==RollBackCb== pFsm:%p, index:%ld, isWeak:%d, code:%d, state:%d %s \n",
+           pFsm, index, isWeak, code, state, syncUtilState2String(state));
+  syncRpcMsgPrint2(logBuf, (SRpcMsg *)pMsg);
+}
+
+void initFsm() {
+  pFsm = (SSyncFSM *)malloc(sizeof(SSyncFSM));
+  pFsm->FpCommitCb = CommitCb;
+  pFsm->FpPreCommitCb = PreCommitCb;
+  pFsm->FpRollBackCb = RollBackCb;
+}
+
+int64_t syncNodeInit() {
+  syncInfo.vgId = 1234;
+  syncInfo.rpcClient = gSyncIO->clientRpc;
+  syncInfo.FpSendMsg = syncIOSendMsg;
+  syncInfo.queue = gSyncIO->pMsgQ;
+  syncInfo.FpEqMsg = syncIOEqMsg;
+  syncInfo.pFsm = pFsm;
+  snprintf(syncInfo.path, sizeof(syncInfo.path), "./replicate2_test_%d", myIndex);
+
+  int code = walInit();
+  assert(code == 0);
+  SWalCfg walCfg;
+  memset(&walCfg, 0, sizeof(SWalCfg));
+  walCfg.vgId = syncInfo.vgId;
+  walCfg.fsyncPeriod = 1000;
+  walCfg.retentionPeriod = 1000;
+  walCfg.rollPeriod = 1000;
+  walCfg.retentionSize = 1000;
+  walCfg.segSize = 1000;
+  walCfg.level = TAOS_WAL_FSYNC;
+
+  char tmpdir[128];
+  snprintf(tmpdir, sizeof(tmpdir), "./replicate2_test_wal_%d", myIndex);
+  pWal = walOpen(tmpdir, &walCfg);
+  assert(pWal != NULL);
+
+  syncInfo.pWal = pWal;
+
+  SSyncCfg *pCfg = &syncInfo.syncCfg;
+  pCfg->myIndex = myIndex;
+  pCfg->replicaNum = replicaNum;
+
+  for (int i = 0; i < replicaNum; ++i) {
+    pCfg->nodeInfo[i].nodePort = ports[i];
+    snprintf(pCfg->nodeInfo[i].nodeFqdn, sizeof(pCfg->nodeInfo[i].nodeFqdn), "%s", "127.0.0.1");
+    // taosGetFqdn(pCfg->nodeInfo[0].nodeFqdn);
+  }
+
+  int64_t rid = syncStart(&syncInfo);
+  assert(rid > 0);
+
+  SSyncNode *pSyncNode = (SSyncNode *)syncNodeAcquire(rid);
+  assert(pSyncNode != NULL);
+
+  // pSyncNode->hbBaseLine = 500;
+  // pSyncNode->electBaseLine = 1500;
+
+  gSyncIO->FpOnSyncPing = pSyncNode->FpOnPing;
+  gSyncIO->FpOnSyncPingReply = pSyncNode->FpOnPingReply;
+  gSyncIO->FpOnSyncRequestVote = pSyncNode->FpOnRequestVote;
+  gSyncIO->FpOnSyncRequestVoteReply = pSyncNode->FpOnRequestVoteReply;
+  gSyncIO->FpOnSyncAppendEntries = pSyncNode->FpOnAppendEntries;
+  gSyncIO->FpOnSyncAppendEntriesReply = pSyncNode->FpOnAppendEntriesReply;
+  gSyncIO->FpOnSyncTimeout = pSyncNode->FpOnTimeout;
+  gSyncIO->FpOnSyncClientRequest = pSyncNode->FpOnClientRequest;
+  gSyncIO->pSyncNode = pSyncNode;
+
+  syncNodeRelease(pSyncNode);
+
+  return rid;
+}
+
+void initRaftId(SSyncNode *pSyncNode) {
+  for (int i = 0; i < replicaNum; ++i) {
+    ids[i] = pSyncNode->replicasId[i];
+    char *s = syncUtilRaftId2Str(&ids[i]);
+    printf("raftId[%d] : %s\n", i, s);
+    free(s);
+  }
+}
+
+SRpcMsg *step0(int i) {
+  SRpcMsg *pMsg = (SRpcMsg *)malloc(sizeof(SRpcMsg));
+  memset(pMsg, 0, sizeof(SRpcMsg));
+  pMsg->msgType = 9999;
+  pMsg->contLen = 128;
+  pMsg->pCont = malloc(pMsg->contLen);
+  snprintf((char *)(pMsg->pCont), pMsg->contLen, "value-%u-%d", ports[myIndex], i);
+  return pMsg;
+}
+
+SyncClientRequest *step1(const SRpcMsg *pMsg) {
+  SyncClientRequest *pRetMsg = syncClientRequestBuild2(pMsg, 123, true);
+  return pRetMsg;
+}
+
+int main(int argc, char **argv) {
+  // taosInitLog((char *)"syncTest.log", 100000, 10);
+  tsAsyncLog = 0;
+  sDebugFlag = 143 + 64;
+  void logTest();
+
+  myIndex = 0;
+  if (argc >= 2) {
+    myIndex = atoi(argv[1]);
+  }
+
+  int recordCount = 100;
+  if (argc >= 3) {
+    recordCount = atoi(argv[2]);
+  }
+
+  int sleepMS = 10;
+  if (argc >= 4) {
+    sleepMS = atoi(argv[3]);
+  }
+
+  int32_t ret = syncIOStart((char *)"127.0.0.1", ports[myIndex]);
+  assert(ret == 0);
+
+  initFsm();
+
+  ret = syncInit();
+  assert(ret == 0);
+
+  int64_t rid = syncNodeInit();
+  assert(rid > 0);
+
+  SSyncNode *pSyncNode = (SSyncNode *)syncNodeAcquire(rid);
+  assert(pSyncNode != NULL);
+
+  syncNodePrint2((char *)"", pSyncNode);
+  initRaftId(pSyncNode);
+
+  for (int i = 0; i < recordCount; ++i) {
+    // step0
+    SRpcMsg *pMsg0 = step0(i);
+    syncRpcMsgPrint2((char *)"==step0==", pMsg0);
+
+    syncPropose(rid, pMsg0, true);
+    taosMsleep(sleepMS);
+
+    free(pMsg0);
+
+    sTrace(
+        "syncPropose sleep, state: %d, %s, term:%lu electTimerLogicClock:%lu, electTimerLogicClockUser:%lu, "
+        "electTimerMS:%d, commitIndex:%ld",
+        pSyncNode->state, syncUtilState2String(pSyncNode->state), pSyncNode->pRaftStore->currentTerm,
+        pSyncNode->electTimerLogicClock, pSyncNode->electTimerLogicClockUser, pSyncNode->electTimerMS,
+        pSyncNode->commitIndex);
+  }
+
+  while (1) {
+    sTrace(
+        "replicate sleep, state: %d, %s, term:%lu electTimerLogicClock:%lu, electTimerLogicClockUser:%lu, "
+        "electTimerMS:%d, commitIndex:%ld",
+        pSyncNode->state, syncUtilState2String(pSyncNode->state), pSyncNode->pRaftStore->currentTerm,
+        pSyncNode->electTimerLogicClock, pSyncNode->electTimerLogicClockUser, pSyncNode->electTimerMS,
+        pSyncNode->commitIndex);
+    taosMsleep(1000);
+  }
+
+  return 0;
+}