[
https://issues.apache.org/jira/browse/IGNITE-7278?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=16332039#comment-16332039
]
Ilya Kasnacheev commented on IGNITE-7278:
-----------------------------------------
We have located at least one reason for "Failed to get page IO instance" errors:
{code:java}
Failed to get page store for the given cache ID (cache has not been started):
-244919260
at
org.apache.ignite.internal.processors.cache.persistence.file.FilePageStoreManager.getStore(FilePageStoreManager.java:619)
at
org.apache.ignite.internal.processors.cache.persistence.file.FilePageStoreManager.read(FilePageStoreManager.java:272)
at
org.apache.ignite.internal.processors.cache.persistence.file.FilePageStoreManager.read(FilePageStoreManager.java:259)
at
org.apache.ignite.internal.processors.cache.persistence.pagemem.PageMemoryImpl.acquirePage(PageMemoryImpl.java:613){code}
There looks to be 2 thing wrong:
* Insufficient locking on cache delete.
* Exceptions in acquirePage leading to data corruption on disk.
> Node failed to recover partition from WAL on unstable topology
> --------------------------------------------------------------
>
> Key: IGNITE-7278
> URL: https://issues.apache.org/jira/browse/IGNITE-7278
> Project: Ignite
> Issue Type: Bug
> Components: persistence
> Reporter: Andrew Mashenkov
> Assignee: Dmitriy Govorukhin
> Priority: Blocker
> Fix For: 2.4
>
> Attachments: page_corrupted2.tar.gz
>
>
> The use case is:
> -Grid with partitioned cache with 2 backups (or replicated)
> -Node-1 is killed in the middle of checkpoint and started again.
> -Node-1 detects unfinished checkpoint and tries to recover it.
> -At this point Node-2 is killed while node-1 recovering is in progress.
> -Node-1 fails with AssertionError.
> PFA logs, parsed WAL, reproducer.
> Can be reproduced with IgnitePdsContinuousRestartTest with minor changes,
> we have to have 2 nodes flapping and kill nodes ungracefully.
--
This message was sent by Atlassian JIRA
(v7.6.3#76005)
