[
https://issues.apache.org/jira/browse/IGNITE-26952?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=18036778#comment-18036778
]
Roman Puchkovskiy commented on IGNITE-26952:
--------------------------------------------
The patch looks good to me
> Fix race between starting and cancelling IncomingSnapshotCopier
> ---------------------------------------------------------------
>
> Key: IGNITE-26952
> URL: https://issues.apache.org/jira/browse/IGNITE-26952
> Project: Ignite
> Issue Type: Bug
> Reporter: Filipp Shergalis
> Assignee: Filipp Shergalis
> Priority: Major
> Labels: ignite-3
> Time Spent: 0.5h
> Remaining Estimate: 0h
>
> When cancellling incoming snapshot, we cancel rebalance future:
>
> {code:java}
> busyLock.block();
> LOG.info("Copier is canceled for partition [{}]", createPartitionInfo());
> // Cancel all futures that might be upstream wrt joinFuture.
> List<CompletableFuture<?>> futuresToCancel = Stream.of(snapshotMetaFuture,
> rebalanceFuture)
> .filter(Objects::nonNull)
> .collect(toList());
> futuresToCancel.forEach(future -> future.cancel(false)); {code}
> completeRebalance is a next stage after rebalanceFuture:
>
> {code:java}
> return rebalanceFuture
> .handleAsync((v, throwable) -> completeRebalance(snapshotContext,
> throwable), executor) {code}
> So when future is cancelled, handleAsync starts completing rebalance
> immediately, not after rebalanceFuture logic actually finished. If it happens
> during startRebalance execution, we have race with startRebalance method and
> completeRebalance
--
This message was sent by Atlassian Jira
(v8.20.10#820010)
