[
https://issues.apache.org/jira/browse/FLINK-2867?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=14967541#comment-14967541
]
Stephan Ewen commented on FLINK-2867:
-------------------------------------
The Checkpoint/Restore and the {{invoke()}} methods are actually guaranteed to
be exclusive by the runtime - there is outside synchronization.
> Access to pendingFilesPerCheckpoint should be synchronized
> ----------------------------------------------------------
>
> Key: FLINK-2867
> URL: https://issues.apache.org/jira/browse/FLINK-2867
> Project: Flink
> Issue Type: Bug
> Reporter: Ted Yu
>
> In RollingSink#restoreState():
> {code}
> Set<Long> pastCheckpointIds =
> bucketState.pendingFilesPerCheckpoint.keySet();
> LOG.debug("Moving pending files to final location on restore.");
> for (Long pastCheckpointId : pastCheckpointIds) {
> // All the pending files are buckets that have been completed but are
> waiting to be renamed
> // to their final name
> for (String filename :
> bucketState.pendingFilesPerCheckpoint.get(pastCheckpointId)) {
> {code}
> Access to pendingFilesPerCheckpoint should be covered under synchronized
> keyword.
> The other methods in this class use synchronized access.
--
This message was sent by Atlassian JIRA
(v6.3.4#6332)
