Github user mxm commented on the pull request: https://github.com/apache/flink/pull/934#issuecomment-126752412 Hi @kl0u , Thanks for addressing my comments. I've made some new ones :) but looks much better. As for merging the different types of snapshots, the only solution I see now, is to let all the accumulator results go through the BlobCache. That might cause some extra latency at the end of the job. I'm not sure whether we should proceed on that path but the current solution has a lot of added overhead because we need to deal with two different kind of update mechanisms. I would like to hear some other opinions on whether we should have both update mechanisms (BlobCache and Akka message-based) or just let everything go through the BlobCache. I'm leaning more towards the latter if it is not too slow for slow accumulators. I think you have to rebase to the latest master again because of some minor changes.
--- If your project is set up for it, you can reply to this email and have your reply appear on GitHub as well. If your project does not have this feature enabled and wishes so, or if the feature is enabled but not working, please contact infrastructure at infrastruct...@apache.org or file a JIRA ticket with INFRA. ---