[HDFS-5014] BPOfferService#processCommandFromActor() synchronization on namenode RPC call delays IBR to Active NN, if Stanby NN is unstable

• Type: Bug
• Status: Closed
• Priority: Major
• Resolution: Fixed
• Affects Version/s: 2.0.4-alpha, 3.0.0-alpha1
• Fix Version/s: 2.3.0
• Component/s: datanode, ha
• Hadoop Flags: Reviewed
• Assignee: Vinayakumar B
• Reporter: Vinayakumar B
• Created: 20/Jul/13 15:18
• Updated: 12/May/16 18:11
• Resolved: 20/Nov/13 16:42

Description

In one of our cluster, following has happened which failed HDFS write.

1. Standby NN was unstable and continously restarting due to some errors. But Active NN was stable.
2. MR Job was writing files.
3. At some point SNN went down again while datanode processing the REGISTER command for SNN.
4. Datanodes started retrying to connect to SNN to register at the following code in BPServiceActor#retrieveNamespaceInfo() which will be called under synchronization.

      try {
        nsInfo = bpNamenode.versionRequest();
        LOG.debug(this + " received versionRequest response: " + nsInfo);
        break;

Unfortunately in all datanodes at same point this happened.
5. For next 7-8 min standby was down, and no blocks were reported to active NN at this point and writes have failed.

So culprit is BPOfferService#processCommandFromActor() is completely synchronized which is not required.

Issue Links

• is duplicated by: HDFS-6179 Synchronized BPOfferService - datanode locks for slow namenode reply.
• relates to: HDFS-2627 HA: determine DN's view of which NN is active based on heartbeat responses