[ https://issues.apache.org/jira/browse/FLINK-15253?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=17008805#comment-17008805 ]
Maximilian Michels commented on FLINK-15253: -------------------------------------------- It is fine with me to make them checkpointed. I would also suggest to decouple them from the heartbeat, to avoid heartbeat failures when they get too large. > Accumulators are not checkpointed > --------------------------------- > > Key: FLINK-15253 > URL: https://issues.apache.org/jira/browse/FLINK-15253 > Project: Flink > Issue Type: Bug > Components: Runtime / Checkpointing > Reporter: Maximilian Michels > Priority: Major > > Accumulators are not checkpointed which make them relatively useless for > streaming applications. They are also tied to the heartbeat (FLINK-15252), > which makes them fragile. > We could consider deactivating accumulators in streaming mode since they seem > to originally be a batch feature. -- This message was sent by Atlassian Jira (v8.3.4#803005)