Github user tillrohrmann commented on the issue:
https://github.com/apache/flink/pull/5879
@sihuazhou I'm actually no longer sure whether sending exceptions as causes
along with the RPC messages was the best decision. Exceptions signal an
exceptional situation and should not occur in the happy case.
Therefore, I'm a bit hesitant to introduce this special
`normal-life-cycle-exception` because they should not exist in the first place
since it is the normal life cycle. Instead, I think we should log the exception
where they occur and send a `String reason` which can be logged on the remote
component. That way we should get rid of most of the exceptions being printed
on `DEBUG`.
As a first step to get there, I think it would be enough to simply print
the exception message in the log statement. That way we would get rid of the
stack traces and don't have to introduce a special exception type. As a second
step, we can refactor the RPC interfaces which have the `Exception cause`
parameters. What do you think?
FLINK-9277 is related but not directly the same. The scope of this issue is
to decrease the redundant messages which occur when releasing a shared slot
starting from the leafs.
---
