[
https://issues.apache.org/jira/browse/IGNITE-10348?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=16709739#comment-16709739
]
Alexey Goncharuk commented on IGNITE-10348:
-------------------------------------------
[~ivan.glukos], [~astelmak], I have the following concerns:
I would like to see a test where a node crashes in the process of metastore
migration, right after all values have been put to the new store, but before a
next checkpoint begins.
In the suggested patch you destroy partition files right away after all values
have been written to the new metastore. This can be done only if there was a
successful checkpoint after the last write to the old metastore - we should
enforce this invariant somehow or prove that checkpoint is already enforced.
> Safely recreate metastore to mitigate IGNITE-8735
> -------------------------------------------------
>
> Key: IGNITE-10348
> URL: https://issues.apache.org/jira/browse/IGNITE-10348
> Project: Ignite
> Issue Type: Improvement
> Affects Versions: 2.4
> Reporter: Alexey Goncharuk
> Assignee: Alexey Stelmak
> Priority: Critical
> Fix For: 2.8
>
>
> We've fixed the issue IGNITE-8735, so new Ignite deployments are not affected
> by this issue, but old deployments still may fail if a wrong page was already
> put to the free list. We need to find out a way to repair this situation.
--
This message was sent by Atlassian JIRA
(v7.6.3#76005)
