Shangshu Qian created FLINK-38367:
-------------------------------------
Summary: Potential Feedback Loop Between IterationHeadTask and
SynchronizationSinkTask
Key: FLINK-38367
URL: https://issues.apache.org/jira/browse/FLINK-38367
Project: Flink
Issue Type: Bug
Affects Versions: 1.20.0
Reporter: Shangshu Qian
IterationHeadTask got an IOException or got interrupted, marked as failed
-> The head task failure causes the SynchronizationSinkTask to be cancelled as
well.
-> However, either because of the grace period introduced (FLINK-24182), or the
some other task simply hang because of the cancelling (FLINK-17726). Additional
tasks which is dependent on this got retried, introducing even more load into
the system.
If the first failed is caused by the system being overloaded, a feedback loop
exists.
--
This message was sent by Atlassian Jira
(v8.20.10#820010)
