I think this jira is helpful for your question: https://issues.apache.org/jira/browse/FLINK-6328 <https://issues.apache.org/jira/browse/FLINK-6328>
> Am 08.05.2017 um 19:33 schrieb Cliff Resnick <cre...@gmail.com>: > > When a job cancel-with-savepoint finishes a successful Savepoint, the > preceding last successful Checkpoint is removed. Is this the intended > behavior? I thought that checkpoints and savepoints were separate entities > and, as such, savepoints should not infringe on checkpoints. This is actually > an issue for us because we have seen occurrences of false-positive successful > savepoints, perhaps due to S3 latency. Bottom line, we'd like to treat > savepoints as insurance rather than the critical path and would rather they > be oblivious to checkpoint management. > > We are using externalized checkpoints, which may be confusing things. Also I > know checkpoint management is undergoing some changes in Flink 1.3 (we are on > Flink 1.2.0). Any insight is greatly appreciated. > >