The reverted changes looks good by itself, but I personally don't feel too comfortable giving a +1 here. Perhaps Till should give the final approval here.
If we do plan to revert this, we should at least have a big warning in the savepoints document page about how things could break. Also as a side note, in the original JIRA discussion, it was mentioned that if we had FLINK-4815 merged, then it can be considered that we readd savepoints to the CompletedCheckpointsStore. How would that relate to the long-term solution that @dawidwys mentioned? [ Full content available at: https://github.com/apache/flink/pull/6704 ] This message was relayed via gitbox.apache.org for [email protected]
