Contending block synchronizations can result in scary log messages
------------------------------------------------------------------
Key: HDFS-2970
URL: https://issues.apache.org/jira/browse/HDFS-2970
Project: Hadoop HDFS
Issue Type: Bug
Components: data-node
Affects Versions: 0.23.0
Reporter: Todd Lipcon
If multiple datanodes are attempting to act as the coordinator for a block
recovery, but one is being particularly slow, it's possible that you see the
following interleaving:
- Primary A receives block recovery command for recovery genstamp = 1, then
starts acting slow
- Primary B receives block recovery command for recovery genstamp = 2
- Primary B calls initReplicaRecovery on other nodes for genstamp = 2
- Primary A calls initReplicaRecovery on other nodes for genstamp = 1
This results in a scary message. For example:
java.io.IOException: java.io.IOException: THIS IS NOT SUPPOSED TO HAPPEN:
replica.getGenerationStamp() >= recoveryId = 4148,
block=blk_6899379920748342698_4136, replica=FinalizedReplica,
blk_6899379920748342698_4176, FINALIZED
BUT, this scenario is properly handled by the recovery protocol. We should tone
down the message a bit.
--
This message is automatically generated by JIRA.
If you think it was sent incorrectly, please contact your JIRA administrators:
https://issues.apache.org/jira/secure/ContactAdministrators!default.jspa
For more information on JIRA, see: http://www.atlassian.com/software/jira
