[
https://issues.apache.org/jira/browse/IGNITE-12591?page=com.atlassian.jira.plugin.system.issuetabpanels:all-tabpanel
]
Ivan Pavlukhin updated IGNITE-12591:
------------------------------------
Description:
Apparently scan query {{ScanQuery}} (without specifying explicit partition)
does not reserve partitions. Consequently when there is a partition movement
(unstable topology, e.g. new node joins the cluster) query can read from an
invalid (e.g. already destroyed) partition.
It is easy to catch "java.lang.IllegalStateException: Tree is being
concurrently destroyed". As it is a consequence of a race condition other
exceptions might be possible. See attached reproducer
[^ScanQueryTopologyMappingTest.java].
One possible solution here is to reserve partitions properly. Other approach is
to consider some optimistic strategy.
was:
Apparently scan query {{ScanQuery}} (without specifying explicit partition)
does not reserve partitions. Consequently when there is a partition movement
(unstable topology, e.g. new node joins the cluster) query can read from an
invalid (e.g. already destroyed) partition.
It is easy to catch "java.lang.IllegalStateException: Tree is being
concurrently destroyed". As it is a consequence of a race condition other
exceptions might be possible. See attached reproducer.
One possible solution here is to reserve partitions properly. Other approach is
to consider some optimistic strategy.
> Scan query scans invalid partitions on unstable topology
> --------------------------------------------------------
>
> Key: IGNITE-12591
> URL: https://issues.apache.org/jira/browse/IGNITE-12591
> Project: Ignite
> Issue Type: Bug
> Components: cache
> Affects Versions: 2.8
> Reporter: Ivan Pavlukhin
> Priority: Major
> Attachments: ScanQueryTopologyMappingTest.java
>
>
> Apparently scan query {{ScanQuery}} (without specifying explicit partition)
> does not reserve partitions. Consequently when there is a partition movement
> (unstable topology, e.g. new node joins the cluster) query can read from an
> invalid (e.g. already destroyed) partition.
> It is easy to catch "java.lang.IllegalStateException: Tree is being
> concurrently destroyed". As it is a consequence of a race condition other
> exceptions might be possible. See attached reproducer
> [^ScanQueryTopologyMappingTest.java].
> One possible solution here is to reserve partitions properly. Other approach
> is to consider some optimistic strategy.
--
This message was sent by Atlassian Jira
(v8.3.4#803005)
