Herman Kolstad Jakobsen created KAFKA-20508:
-----------------------------------------------
Summary: Do not log/throw error/exception for corrupted tasks due
to compacted changelog topic
Key: KAFKA-20508
URL: https://issues.apache.org/jira/browse/KAFKA-20508
Project: Kafka
Issue Type: Improvement
Reporter: Herman Kolstad Jakobsen
The current state store position is tracked in the changelog locally (writing
offsets into local {{.checkpoint}} file. If a task is moved to a pod (active or
standby), the store offset is used as start offset to restore state. If the
offset is too old, ie, smaller than beginning-offset, an error triggered.
Because the store content aged out, and we need to ensure we don't miss any
tombstone (which might have cleaned up, too), a restoration based on the full
changelog is needed. So we wipe the state store entirely before restoring.
This scenario is nothing to be worried about as everything is gracefully
handled, but when the scenario happens an ERROR is logged together with a
TaskCorruptedException.
The ERROR log is noisy for the users and can confuse the user into taking
action when indeed no action should be taken.
A more sutiable log level could then be WARN or INFO.
--
This message was sent by Atlassian Jira
(v8.20.10#820010)
