[HDFS-16061] DFTestUtil.waitReplication can produce false positives

• Type: Bug
• Status: Resolved
• Priority: Major
• Resolution: Fixed
• Fix Version/s: 3.4.0, 3.2.3, 3.3.2
• Component/s: hdfs
• Labels: pull-request-available
• Hadoop Flags: Reviewed
• Assignee: Ahmed Hussein
• Reporter: Ahmed Hussein
• Created: 09/Jun/21 17:37
• Updated: 20/Jun/21 05:41
• Resolved: 20/Jun/21 05:41

Description

While checking the intermittent failure in TestBalancerRPCDelay#testBalancerRPCDelayQpsDefault described in HDFS-15146, I found that the implementation of waitReplication is incorrect.
In the last iteration, when correctReplFactor is false, the thread sleeps for 1 second, then a TimeoutException is thrown without check whether the replication was complete in the last second.

Issue Links

• Discovered while testing: HDFS-15671 TestBalancerRPCDelay#testBalancerRPCDelayQpsDefault fails on Trunk
• links to: GitHub Pull Request #3095