From bd283c895f6e81f97608c0f12fec22f6fcf8e89d Mon Sep 17 00:00:00 2001
From: Cao Manh Dat <datcm@apache.org>
Date: Thu, 20 Jul 2017 14:39:30 +0700
Subject: [PATCH] SOLR-11124: MoveReplicaCmd should skip deleting old replica
 in case of its node is not live

---
 solr/CHANGES.txt                              |  2 +
 .../org/apache/solr/cloud/MoveReplicaCmd.java | 64 ++++++++++---------
 2 files changed, 37 insertions(+), 29 deletions(-)

diff --git a/solr/CHANGES.txt b/solr/CHANGES.txt
index dfbc01074c5..90ab2924721 100644
--- a/solr/CHANGES.txt
+++ b/solr/CHANGES.txt
@@ -79,6 +79,8 @@ Optimizations
 * SOLR-10985: Remove unnecessary toString() calls in solr-core's search package's debug logging.
   (Michael Braun via Christine Poerschke)
 
+* SOLR-11124: MoveReplicaCmd should skip deleting old replica in case of its node is not live (Cao Manh Dat)
+
 Other Changes
 ----------------------
 
diff --git a/solr/core/src/java/org/apache/solr/cloud/MoveReplicaCmd.java b/solr/core/src/java/org/apache/solr/cloud/MoveReplicaCmd.java
index 8c4e9d76ee5..3f29fc5f69c 100644
--- a/solr/core/src/java/org/apache/solr/cloud/MoveReplicaCmd.java
+++ b/solr/core/src/java/org/apache/solr/cloud/MoveReplicaCmd.java
@@ -119,36 +119,41 @@ public class MoveReplicaCmd implements Cmd{
 
   private void moveHdfsReplica(ClusterState clusterState, NamedList results, String dataDir, String targetNode, String async,
                                  DocCollection coll, Replica replica, Slice slice, int timeout) throws Exception {
-    ZkNodeProps removeReplicasProps = new ZkNodeProps(
-        COLLECTION_PROP, coll.getName(),
-        SHARD_ID_PROP, slice.getName(),
-        REPLICA_PROP, replica.getName()
-        );
-    removeReplicasProps.getProperties().put(CoreAdminParams.DELETE_DATA_DIR, false);
-    removeReplicasProps.getProperties().put(CoreAdminParams.DELETE_INDEX, false);
-    if(async!=null) removeReplicasProps.getProperties().put(ASYNC, async);
-    NamedList deleteResult = new NamedList();
-    ocmh.deleteReplica(clusterState, removeReplicasProps, deleteResult, null);
-    if (deleteResult.get("failure") != null) {
-      String errorString = String.format(Locale.ROOT, "Failed to cleanup replica collection=%s shard=%s name=%s",
-          coll.getName(), slice.getName(), replica.getName());
-      log.warn(errorString);
-      results.add("failure", errorString + ", because of : " + deleteResult.get("failure"));
-      return;
-    }
-
-    TimeOut timeOut = new TimeOut(20L, TimeUnit.SECONDS);
-    while (!timeOut.hasTimedOut()) {
-      coll = ocmh.zkStateReader.getClusterState().getCollection(coll.getName());
-      if (coll.getReplica(replica.getName()) != null) {
-        Thread.sleep(100);
-      } else {
-        break;
+    String skipCreateReplicaInClusterState = "true";
+    if (clusterState.getLiveNodes().contains(replica.getNodeName())) {
+      skipCreateReplicaInClusterState = "false";
+      ZkNodeProps removeReplicasProps = new ZkNodeProps(
+          COLLECTION_PROP, coll.getName(),
+          SHARD_ID_PROP, slice.getName(),
+          REPLICA_PROP, replica.getName()
+      );
+      removeReplicasProps.getProperties().put(CoreAdminParams.DELETE_DATA_DIR, false);
+      removeReplicasProps.getProperties().put(CoreAdminParams.DELETE_INDEX, false);
+      if(async!=null) removeReplicasProps.getProperties().put(ASYNC, async);
+      NamedList deleteResult = new NamedList();
+      ocmh.deleteReplica(clusterState, removeReplicasProps, deleteResult, null);
+      if (deleteResult.get("failure") != null) {
+        String errorString = String.format(Locale.ROOT, "Failed to cleanup replica collection=%s shard=%s name=%s",
+            coll.getName(), slice.getName(), replica.getName());
+        log.warn(errorString);
+        results.add("failure", errorString + ", because of : " + deleteResult.get("failure"));
+        return;
       }
-    }
-    if (timeOut.hasTimedOut()) {
-      results.add("failure", "Still see deleted replica in clusterstate!");
-      return;
+
+      TimeOut timeOut = new TimeOut(20L, TimeUnit.SECONDS);
+      while (!timeOut.hasTimedOut()) {
+        coll = ocmh.zkStateReader.getClusterState().getCollection(coll.getName());
+        if (coll.getReplica(replica.getName()) != null) {
+          Thread.sleep(100);
+        } else {
+          break;
+        }
+      }
+      if (timeOut.hasTimedOut()) {
+        results.add("failure", "Still see deleted replica in clusterstate!");
+        return;
+      }
+
     }
 
     String ulogDir = replica.getStr(CoreAdminParams.ULOG_DIR);
@@ -158,6 +163,7 @@ public class MoveReplicaCmd implements Cmd{
         CoreAdminParams.NODE, targetNode,
         CoreAdminParams.CORE_NODE_NAME, replica.getName(),
         CoreAdminParams.NAME, replica.getCoreName(),
+        SKIP_CREATE_REPLICA_IN_CLUSTER_STATE, skipCreateReplicaInClusterState,
         CoreAdminParams.ULOG_DIR, ulogDir.substring(0, ulogDir.lastIndexOf(UpdateLog.TLOG_NAME)),
         CoreAdminParams.DATA_DIR, dataDir);
     if(async!=null) addReplicasProps.getProperties().put(ASYNC, async);