[
https://issues.apache.org/jira/browse/FLINK-20872?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=17260942#comment-17260942
]
Yun Tang commented on FLINK-20872:
----------------------------------
As we all know, savepoint/checkpoint is used for fault tolerance. If you choose
to start the job from a previous savepoint/checkpoint, the correct mechanism is
to enable job could recover from previous savepoint/checkpoint to avoid data
lost during failover. I don't know why we need to add specific warning to
describe such expected behavior. Moreover, I think current documentation should
already give enough description over this: ["Checkpoints allow Flink to recover
state and positions in the streams to give the application the same semantics
as a failure-free
execution."|https://ci.apache.org/projects/flink/flink-docs-release-1.12/dev/stream/state/checkpointing.html]
I will close this ticket as this conflicts with the basic idea of fault
tolerance in Apache Flink.
> Job resume from history savepoint when failover if checkpoint is disabled
> -------------------------------------------------------------------------
>
> Key: FLINK-20872
> URL: https://issues.apache.org/jira/browse/FLINK-20872
> Project: Flink
> Issue Type: Improvement
> Affects Versions: 1.10.0, 1.12.0
> Reporter: Liu
> Priority: Minor
>
> I have a long running job. Its checkpoint is disabled and restartStrategy is
> set. One time I upgrade the job through savepoint. One day later, the job is
> failed and restart automatically. But it is resumed from the previous
> savepoint so that the job is heavily lagged.
>
> I have checked the code and find that the job will first try to resume from
> checkpoint and then savepoint.
> {code:java}
> if (checkpointCoordinator != null) {
> // check whether we find a valid checkpoint
> if (!checkpointCoordinator.restoreInitialCheckpointIfPresent(
> new HashSet<>(newExecutionGraph.getAllVertices().values()))) {
> // check whether we can restore from a savepoint
> tryRestoreExecutionGraphFromSavepoint(
> newExecutionGraph, jobGraph.getSavepointRestoreSettings());
> }
> }
> {code}
> For job which checkpoint is disabled, internal failover should not resume
> from previous savepoint, especially the savepoint is done long long ago. In
> this situation, state loss is acceptable but lag is not acceptable.
--
This message was sent by Atlassian Jira
(v8.3.4#803005)
