Description
After the backport patch of HDFS-4975 was committed, TestAzureBlockPlacementPolicy and TestReplicationPolicyWithNodeGroup failed.
The cause for the failure of TestReplicationPolicyWithNodeGroup is that some part in the patch of HDFS-3941 is missing. Our patch for HADOOP-495 makes methods in super class to be called incorrectly. More specifically, HDFS-4975 backported HDFS-4350, HDFS-4351, and HDFS-3912 to enable the method parameter "boolean avoidStaleNodes", and updated the APIs in BlockPlacementPolicyDefault. However, the override methods in ReplicationPolicyWithNodeGroup weren't updated.
The cause for the failure of TestAzureBlockPlacementPolicy is similar.
In addition, TestAzureBlockPlacementPolicy has an error. Here is the error info.
Testcase: testPolicyWithDefaultRacks took 0.005 sec
Caused an ERROR
Invalid network topology. You cannot have a rack and a non-rack node at the same level of the network topology.
org.apache.hadoop.net.NetworkTopology$InvalidTopologyException: Invalid network topology. You cannot have a rack and a non-rack node at the same level of the network topology.
at org.apache.hadoop.net.NetworkTopology.add(NetworkTopology.java:396)
at org.apache.hadoop.hdfs.server.namenode.TestAzureBlockPlacementPolicy.testPolicyWithDefaultRacks(TestAzureBlockPlacementPolicy.java:779)
The error is caused by a check in NetworkTopology#add(Node node)
if (depthOfAllLeaves != node.getLevel()) { LOG.error("Error: can't add leaf node at depth " + node.getLevel() + " to topology:\n" + oldTopoStr); throw new InvalidTopologyException("Invalid network topology. " + "You cannot have a rack and a non-rack node at the same " + "level of the network topology."); }
The problem of this check is that when we use NetworkTopology#remove(Node node) to remove a node from the cluster, depthOfAllLeaves won't change. As a result, we can't reset the value of NetworkTopology#depathOfAllLeaves of the old topology of a cluster by just removing all its dataNode. See TestAzureBlockPlacementPolicy#testPolicyWithDefaultRacks()
// clear the old topology for (Node node : dataNodes) { cluster.remove(node); }
Attachments
Attachments
Issue Links
- is related to
-
HDFS-4975 Branch-1-win TestReplicationPolicy failed caused by stale data node handling
- Resolved