[ 
https://issues.apache.org/jira/browse/FLINK-6964?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=16082389#comment-16082389
 ] 

ASF GitHub Bot commented on FLINK-6964:
---------------------------------------

Github user StefanRRichter commented on a diff in the pull request:

    https://github.com/apache/flink/pull/4192#discussion_r126725439
  
    --- Diff: 
flink-runtime/src/test/java/org/apache/flink/runtime/checkpoint/ZooKeeperCompletedCheckpointStoreTest.java
 ---
    @@ -160,9 +162,12 @@ public Void answer(InvocationOnMock invocation) throws 
Throwable {
                        stateStorage,
                        Executors.directExecutor());
     
    -           SharedStateRegistry sharedStateRegistry = new 
SharedStateRegistry();
    +           SharedStateRegistry sharedStateRegistry = spy(new 
SharedStateRegistry());
                zooKeeperCompletedCheckpointStore.recover(sharedStateRegistry);
     
    +           verify(retrievableStateHandle1.retrieveState(), 
times(1)).registerSharedStatesAfterRestored(sharedStateRegistry);
    --- End diff --
    
    The behaviour that is tested here is not strongly connected to the bug, I 
just noticed it might be also worth checking to avoid future problems. 
{{StandaloneCompletedCheckpointStore}} is not expected to do this, so there is 
no need for a this particular test.
    
    The question for a general test is valid. However, I noticed that 
externalized checkpoints in general are not tested at all, so this is not just 
a small addition but a whole new test (including some required test methods to 
trigger and restore from externalized checkpoints in the JM) that would almost 
justify a new planning task and PR.


> Fix recovery for incremental checkpoints in StandaloneCompletedCheckpointStore
> ------------------------------------------------------------------------------
>
>                 Key: FLINK-6964
>                 URL: https://issues.apache.org/jira/browse/FLINK-6964
>             Project: Flink
>          Issue Type: Bug
>          Components: State Backends, Checkpointing
>            Reporter: Stefan Richter
>            Assignee: Stefan Richter
>            Priority: Blocker
>             Fix For: 1.3.2
>
>
> {{StandaloneCompletedCheckpointStore}} does not register shared states ion 
> resume. However, for externalized checkpoints, it register the checkpoint 
> from which it resumed. This checkpoint gets added to the completed checkpoint 
> store as part of resume.



--
This message was sent by Atlassian JIRA
(v6.4.14#64029)

Reply via email to