Matthias J. Sax created KAFKA-19682:
---------------------------------------
Summary: Improve logging about task readiness
Key: KAFKA-19682
URL: https://issues.apache.org/jira/browse/KAFKA-19682
Project: Kafka
Issue Type: Bug
Components: streams
Reporter: Matthias J. Sax
Kafka Streams processes its assigned tasks in a round robin fashion, however,
if a task is considered not ready for processing, it might get skipped. We have
observed cases, in which Kafka Streams seems to get stuck on a partition, and
restarting the application instance resolves this issue. We suspect, it's
related to considering the task for the stuck partition as not ready for
processing. (As the ready/not-ready decision is based on in-memory state of KS
runtime, bouncing the instance would reset KS runtime into a clean state,
unblocking the stuck task.)
However, we currently don't have sufficient logging to reason about this case,
and to understand if and why a task might be skipped. We should add more log
statement (DEBUG and/or TRACE) to get better visibility. There might be a
lurking bug in this logic that we cannot narrow down w/o the corresponding
information logging could provide.
--
This message was sent by Atlassian Jira
(v8.20.10#820010)
