Github user StefanRRichter commented on the issue:
https://github.com/apache/flink/pull/6019
@sihuazhou my plan was to integrate the timer service more closely with the
keyed state backends, starting from the point that we are merging the PR for
timers in RocksDB. I think timers should just be considered as keyed state and
eventually become part of the keyed state backend's snapshot.
@makeyang with the above comment, that essentially means that with the work
on the RocksDB timer service we are planning to have a larger rewrite of how
snapshotting for timer works anyways. I have also concrete plans and work for
doing this for the heap timer service, as well as including a efficient way to
delete timers. All this unfortunately does not match too well with this PR. I
suggest that this can be closed because we will address this issue for the next
release in somewhat more holistic way.
---
