You are viewing a plain text version of this content. The canonical link for it is here.
Posted to issues@flink.apache.org by "Lennon Yu (Jira)" <ji...@apache.org> on 2024/05/09 12:13:00 UTC
[jira] [Created] (FLINK-35321) CheckpointCommittableManagerImpl re-registers pendingCommittables gauge on every commit operation
Lennon Yu created FLINK-35321:
---------------------------------
Summary: CheckpointCommittableManagerImpl re-registers pendingCommittables gauge on every commit operation
Key: FLINK-35321
URL: https://issues.apache.org/jira/browse/FLINK-35321
Project: Flink
Issue Type: Bug
Components: API / Core, Runtime / Checkpointing, Runtime / Metrics
Affects Versions: 1.19.0
Reporter: Lennon Yu
Found while testing a home-made Sink implementation that implements SupportCommitter. We observed that starting from the *second* checkpoint, every committer commit will be accompanied by the warning log:
{quote}
Name collision: Group already contains a Metric with the name 'pendingCommittables'.
{quote}
Enabling the debugger and tracing the origin of this log took us to {{org.apache.flink.streaming.runtime.operators.sink.committables.CheckpointCommittableManagerImpl}} at line 137 of the commit() method:
{code:java}
metricGroup.setCurrentPendingCommittablesGauge(() -> getPendingRequests(false).size());
{code}
It looks like that instead of modifying the value of the gauge, the manager class is *re-setting with a different guage* on every commit operation, which explains the appearance of the warning log shown above.
--
This message was sent by Atlassian Jira
(v8.20.10#820010)