bugfix:check slave partition state after dbslaveof no one (OpenAtomFo…

…undation#841)
sharper · Mar 20, 2020 · 51a5826 · 51a5826
1 parent 2a6b1f0
commit 51a5826
Show file tree

Hide file tree

Showing 3 changed files with 5 additions and 4 deletions.
diff --git a/include/pika_rm.h b/include/pika_rm.h
@@ -213,7 +213,7 @@ class PikaReplicaManager {
 
   void RmStatus(std::string* debug_info);
 
-  bool CheckSlaveDBConnect();
+  static bool CheckSlavePartitionState(const std::string& ip, const int port);
 
   Status LostConnection(const std::string& ip, int port);
 

diff --git a/src/pika_repl_client_thread.cc b/src/pika_repl_client_thread.cc
@@ -45,7 +45,7 @@ void PikaReplClientThread::ReplClientHandle::FdTimeoutHandle(int fd, const std::
   if (ip == g_pika_server->master_ip()
     && port == g_pika_server->master_port() + kPortShiftReplServer
     && PIKA_REPL_ERROR != g_pika_server->repl_state()
-    && g_pika_rm->CheckSlaveDBConnect()) {  // if state machine in error state, no retry
+    && g_pika_rm->CheckSlavePartitionState(ip, port)) {  // if state machine in error state, no retry
     LOG(WARNING) << "Master conn timeout : " << ip_port << " try reconnect";
     g_pika_server->ResetMetaSyncStatus();
     g_pika_server->UpdateMetaSyncTimestamp();

diff --git a/src/pika_rm.cc b/src/pika_rm.cc
@@ -797,11 +797,12 @@ Status PikaReplicaManager::UpdateSyncBinlogStatus(const RmNode& slave, const Bin
   return Status::OK();
 }
 
-bool PikaReplicaManager::CheckSlaveDBConnect() {
+bool PikaReplicaManager::CheckSlavePartitionState(const std::string& ip, const int port) {
   std::shared_ptr<SyncSlavePartition> partition = nullptr;
   for (auto iter : g_pika_rm->sync_slave_partitions_) {
     partition = iter.second;
-    if (partition->State() == ReplState::kDBNoConnect) { 
+    if (partition->State() == ReplState::kDBNoConnect &&
+      partition->MasterIp() == ip && partition->MasterPort() + kPortShiftReplServer == port) {
       LOG(INFO) << "DB: " << partition->SyncPartitionInfo().ToString()
         << " has been dbslaveof no one, then will not try reconnect.";
       return false;