Github user StephanEwen commented on the issue:
https://github.com/apache/flink/pull/5399
I am unsure about this change, so let's discuss the pros and cons a bit.
So far, this purposefully logs on *INFO* so far, because from the
JobManager's perspective, a task failing and recovering if not an erroneous
situation. It conveys the assumption that failures and recoveries are perfectly
expected as part of the Job life cycle.
The assumption Something that is logged on "ERROR" is something where a
user may want to dig into and see whether they should do something about it.
Flink does not follow this perspective perfectly in all parts, but that was
the reasoning behind the fact to use *info* for these state changes.
Happy to hear other thoughts on this.
---
