[HUDI-5041] lock metric register confict error

• Type: Bug
• Status: Closed
• Priority: Major
• Resolution: Fixed
• Fix Version/s: 0.12.2
• Component/s: flink
• Labels: pull-request-available
• Assignee: Unassigned
• Reporter: HBG
• Created: 17/Oct/22 10:04
• Updated: 29/Nov/22 13:19
• Resolved: 31/Oct/22 19:39

Description

There is a chance to see this error when write data to hudi with flink, and slot number per TM is larger than 1.

java.lang.IllegalArgumentException: A metric named hudi_table1.lock.acquire.duration already exists

The cause is in constructor of HoodieLockMetrics, when different thread from different slot of a Flink TM new HoodieLockMetrics, timer of identical metricName may registered to 

MetricRegistry multiple time, causing this error.

we can add synchronized statement in function 'createTimerForMetrics' to make the test and set action atomic to avoid adding a timer multiple times.

 

Issue Links

• relates to: HUDI-5280 Fix taskmanager concurrent registry
• links to: GitHub Pull Request #6968