Major Description
Today when we suspend a task we commit and along with the commit we always write checkpoint file even if we are eosEnabled (since the state is already SUSPENDED). But the suspended task may later be resumed and in that case the checkpoint file should be deleted since it should only be written when it is cleanly closed.
With our latest rebalance protocol in KIP-429, resume would not be called since all suspended tasks would be closed, but with the old eager protocol it may still be called — I think that may be the reason we did not get it often.
Attachments
Issue Links
- links to