[region migration] Log optimization (apache#14536) (apache#14545)

* save log

* done?

(cherry picked from commit 3ae2355)

iotdb-core/confignode/src/main/java/org/apache/iotdb/confignode/procedure/env/RegionMaintainHandler.java

@@ -96,6 +96,10 @@ public static String getIdWithRpcEndpoint(TDataNodeLocation location) {
         location.getDataNodeId(), location.getClientRpcEndPoint());
   }
 
+  public String simplifiedLocation(TDataNodeLocation dataNodeLocation) {
+    return dataNodeLocation.getDataNodeId() + "@" + dataNodeLocation.getInternalEndPoint().getIp();
+  }
+
   /**
    * Find dest data node.
    *
@@ -397,17 +401,32 @@ public void removeRegionLocation(
    * @return DataNode locations
    */
   public List<TDataNodeLocation> findRegionLocations(TConsensusGroupId regionId) {
-    Optional<TRegionReplicaSet> regionReplicaSet =
-        configManager.getPartitionManager().getAllReplicaSets().stream()
-            .filter(rg -> rg.regionId.equals(regionId))
-            .findAny();
+    Optional<TRegionReplicaSet> regionReplicaSet = getRegionReplicaSet(regionId);
     if (regionReplicaSet.isPresent()) {
       return regionReplicaSet.get().getDataNodeLocations();
     }
-
     return Collections.emptyList();
   }
 
+  public Optional<TRegionReplicaSet> getRegionReplicaSet(TConsensusGroupId regionId) {
+    return configManager.getPartitionManager().getAllReplicaSets().stream()
+        .filter(rg -> rg.regionId.equals(regionId))
+        .findAny();
+  }
+
+  public String getRegionReplicaSetString(TConsensusGroupId regionId) {
+    Optional<TRegionReplicaSet> regionReplicaSet = getRegionReplicaSet(regionId);
+    if (!regionReplicaSet.isPresent()) {
+      return "UNKNOWN!";
+    }
+    StringBuilder result = new StringBuilder(regionReplicaSet.get().getRegionId() + ": {");
+    for (TDataNodeLocation dataNodeLocation : regionReplicaSet.get().getDataNodeLocations()) {
+      result.append(simplifiedLocation(dataNodeLocation)).append(", ");
+    }
+    result.append("}");
+    return result.toString();
+  }
+
   private Optional<TDataNodeLocation> pickNewReplicaNodeForRegion(
       List<TDataNodeLocation> regionReplicaNodes) {
     List<TDataNodeConfiguration> dataNodeConfigurations =

iotdb-core/confignode/src/main/java/org/apache/iotdb/confignode/procedure/impl/region/AddRegionPeerProcedure.java

@@ -88,10 +88,10 @@ protected Flow executeFromState(ConfigNodeProcedureEnv env, AddRegionPeerState s
       switch (state) {
         case CREATE_NEW_REGION_PEER:
           LOGGER.info(
-              "[pid{}][AddRegion] started, region {} will be added to DataNode {}.",
+              "[pid{}][AddRegion] started, {} will be added to DataNode {}.",
               getProcId(),
-              consensusGroupId.getId(),
-              destDataNode.getDataNodeId());
+              consensusGroupId,
+              handler.simplifiedLocation(destDataNode));
           handler.addRegionLocation(consensusGroupId, destDataNode);
           handler.forceUpdateRegionCache(consensusGroupId, destDataNode, RegionStatus.Adding);
           TSStatus status = handler.createNewRegionPeer(consensusGroupId, destDataNode);
@@ -138,10 +138,10 @@ protected Flow executeFromState(ConfigNodeProcedureEnv env, AddRegionPeerState s
           setKillPoint(state);
           LOGGER.info("[pid{}][AddRegion] state {} complete", getProcId(), state);
           LOGGER.info(
-              "[pid{}][AddRegion] success, region {} has been added to DataNode {}. Procedure took {} (start at {}).",
+              "[pid{}][AddRegion] success, {} has been added to DataNode {}. Procedure took {} (start at {}).",
               getProcId(),
-              consensusGroupId.getId(),
-              destDataNode.getDataNodeId(),
+              consensusGroupId,
+              handler.simplifiedLocation(destDataNode),
               CommonDateTimeUtils.convertMillisecondToDurationStr(
                   System.currentTimeMillis() - getSubmittedTime()),
               DateTimeUtils.convertLongToDate(getSubmittedTime(), "ms"));

iotdb-core/confignode/src/main/java/org/apache/iotdb/confignode/procedure/impl/region/RegionMigrateProcedure.java

@@ -83,11 +83,11 @@ protected Flow executeFromState(ConfigNodeProcedureEnv env, RegionTransitionStat
       switch (state) {
         case REGION_MIGRATE_PREPARE:
           LOGGER.info(
-              "[pid{}][MigrateRegion] started, region {} will be migrated from DataNode {} to {}.",
+              "[pid{}][MigrateRegion] started, {} will be migrated from DataNode {} to {}.",
               getProcId(),
-              consensusGroupId.getId(),
-              originalDataNode.getDataNodeId(),
-              destDataNode.getDataNodeId());
+              consensusGroupId,
+              handler.simplifiedLocation(originalDataNode),
+              handler.simplifiedLocation(destDataNode));
           setNextState(RegionTransitionState.ADD_REGION_PEER);
           break;
         case ADD_REGION_PEER:
@@ -113,23 +113,23 @@ protected Flow executeFromState(ConfigNodeProcedureEnv env, RegionTransitionStat
           setNextState(RegionTransitionState.CHECK_REMOVE_REGION_PEER);
           break;
         case CHECK_REMOVE_REGION_PEER:
+          String cleanHint = "";
           if (env.getConfigManager()
               .getPartitionManager()
               .isDataNodeContainsRegion(originalDataNode.getDataNodeId(), consensusGroupId)) {
-            LOGGER.warn(
-                "[pid{}][MigrateRegion] success, but you may need to manually clean the old region to make everything works fine",
-                getProcId());
-          } else {
-            LOGGER.info(
-                "[pid{}][MigrateRegion] success, region {} has been migrated from DataNode {} to {}. Procedure took {} (started at {})",
-                getProcId(),
-                consensusGroupId.getId(),
-                originalDataNode.getDataNodeId(),
-                destDataNode.getDataNodeId(),
-                CommonDateTimeUtils.convertMillisecondToDurationStr(
-                    System.currentTimeMillis() - getSubmittedTime()),
-                DateTimeUtils.convertLongToDate(getSubmittedTime(), "ms"));
+            cleanHint =
+                "but you may need to restart the related DataNode to make sure everything is cleaned up. ";
           }
+          LOGGER.info(
+              "[pid{}][MigrateRegion] success,{} {} has been migrated from DataNode {} to {}. Procedure took {} (started at {}).",
+              getProcId(),
+              cleanHint,
+              consensusGroupId,
+              handler.simplifiedLocation(originalDataNode),
+              handler.simplifiedLocation(destDataNode),
+              CommonDateTimeUtils.convertMillisecondToDurationStr(
+                  System.currentTimeMillis() - getSubmittedTime()),
+              DateTimeUtils.convertLongToDate(getSubmittedTime(), "ms"));
           return Flow.NO_MORE_STATE;
         default:
           throw new ProcedureException("Unsupported state: " + state.name());

iotdb-core/confignode/src/main/java/org/apache/iotdb/confignode/procedure/impl/region/RemoveRegionPeerProcedure.java

@@ -106,19 +106,23 @@ protected Flow executeFromState(ConfigNodeProcedureEnv env, RemoveRegionPeerStat
           setKillPoint(state);
           if (tsStatus.getCode() != SUCCESS_STATUS.getStatusCode()) {
             LOGGER.warn(
-                "[pid{}][RemoveRegion] {} task submitted failed, procedure will continue. You should manually clear peer list.",
+                "[pid{}][RemoveRegion] {} task submitted failed, ConfigNode believe current peer list of {} is {}. Procedure will continue. You should manually clear peer list.",
                 getProcId(),
-                state);
+                state,
+                consensusGroupId,
+                handler.getRegionReplicaSetString(consensusGroupId));
             setNextState(DELETE_OLD_REGION_PEER);
             return Flow.HAS_MORE_STATE;
           }
           TRegionMigrateResult removeRegionPeerResult =
               handler.waitTaskFinish(this.getProcId(), coordinator);
           if (removeRegionPeerResult.getTaskStatus() != TRegionMaintainTaskStatus.SUCCESS) {
             LOGGER.warn(
-                "[pid{}][RemoveRegion] {} executed failed, procedure will continue. You should manually clear peer list.",
+                "[pid{}][RemoveRegion] {} executed failed, ConfigNode believe current peer list of {} is {}. Procedure will continue. You should manually clear peer list.",
                 getProcId(),
-                state);
+                state,
+                consensusGroupId,
+                handler.getRegionReplicaSetString(consensusGroupId));
             setNextState(DELETE_OLD_REGION_PEER);
             return Flow.HAS_MORE_STATE;
           }
@@ -132,17 +136,19 @@ protected Flow executeFromState(ConfigNodeProcedureEnv env, RemoveRegionPeerStat
           setKillPoint(state);
           if (tsStatus.getCode() != SUCCESS_STATUS.getStatusCode()) {
             LOGGER.warn(
-                "[pid{}][RemoveRegion] DELETE_OLD_REGION_PEER task submitted failed, procedure will continue. You should manually delete region file.",
-                getProcId());
+                "[pid{}][RemoveRegion] DELETE_OLD_REGION_PEER task submitted failed, procedure will continue. You should manually delete region file. {}",
+                getProcId(),
+                consensusGroupId);
             setNextState(REMOVE_REGION_LOCATION_CACHE);
             return Flow.HAS_MORE_STATE;
           }
           TRegionMigrateResult deleteOldRegionPeerResult =
               handler.waitTaskFinish(this.getProcId(), targetDataNode);
           if (deleteOldRegionPeerResult.getTaskStatus() != TRegionMaintainTaskStatus.SUCCESS) {
             LOGGER.warn(
-                "[pid{}][RemoveRegion] DELETE_OLD_REGION_PEER executed failed, procedure will continue. You should manually delete region file.",
-                getProcId());
+                "[pid{}][RemoveRegion] DELETE_OLD_REGION_PEER executed failed, procedure will continue. You should manually delete region file. {}",
+                getProcId(),
+                consensusGroupId);
             setNextState(REMOVE_REGION_LOCATION_CACHE);
             return Flow.HAS_MORE_STATE;
           }