HDFS-10477. Stop decommission a rack of DataNodes caused NameNode fail over to standby. Contributed by yunjiong zhao and Wei-Chiu Chuang.

hadoop-hdfs-project/hadoop-hdfs/src/main/java/org/apache/hadoop/hdfs/server/blockmanagement/BlockManager.java

@@ -4235,8 +4235,15 @@ void processExtraRedundancyBlocksOnInService(
     if (!isPopulatingReplQueues()) {
       return;
     }
-    final Iterator<BlockInfo> it = srcNode.getBlockIterator();
+
     int numExtraRedundancy = 0;
+    for (DatanodeStorageInfo datanodeStorageInfo : srcNode.getStorageInfos()) {
+      // the namesystem lock is released between iterations. Make sure the
+      // storage is not removed before continuing.
+      if (srcNode.getStorageInfo(datanodeStorageInfo.getStorageID()) == null) {
+        continue;
+      }
+      final Iterator<BlockInfo> it = datanodeStorageInfo.getBlockIterator();
       while(it.hasNext()) {
         final BlockInfo block = it.next();
         if (block.isDeleted()) {
@@ -4252,6 +4259,19 @@ void processExtraRedundancyBlocksOnInService(
           numExtraRedundancy++;
         }
       }
+      // When called by tests like TestDefaultBlockPlacementPolicy.
+      // testPlacementWithLocalRackNodesDecommissioned, it is not protected by
+      // lock, only when called by DatanodeManager.refreshNodes have writeLock
+      if (namesystem.hasWriteLock()) {
+        namesystem.writeUnlock();
+        try {
+          Thread.sleep(1);
+        } catch (InterruptedException e) {
+          Thread.currentThread().interrupt();
+        }
+        namesystem.writeLock();
+      }
+    }
     LOG.info("Invalidated {} extra redundancy blocks on {} after "
              + "it is in service", numExtraRedundancy, srcNode);
   }