[
https://issues.apache.org/jira/browse/IGNITE-8021?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=16417788#comment-16417788
]
Ilya Kasnacheev commented on IGNITE-8021:
-----------------------------------------
There's a blocking consideration about this patch:
When bringing node up after incomplete checkpoint, we might need information
about cache, but it is already deleted at this point.
Simply removing the file is not safe, it can be only done if cache is empty
before that operation & all checkpoints passed. I think we should be marking
the file in some way that will still keep it for operations, but will prevent
resurrecting it. Then it should be resurrected if there are nodes in cluster on
which it is alive. We should also consider keeping old configurations for
caches that are re-created until they're no longer needed.
Please see the attached patch-for-patch where we try killing node in the middle
of checkpoint. It will sometimes hang with assertion error about empty group
descriptor. However, if files are not deleted (these methods commented out) it
will never hang (but sometimes fail spuriously)
> Destroyed caches can be return to life by restart grid
> ------------------------------------------------------
>
> Key: IGNITE-8021
> URL: https://issues.apache.org/jira/browse/IGNITE-8021
> Project: Ignite
> Issue Type: Bug
> Reporter: Vladislav Pyatkov
> Assignee: Ivan Daschinskiy
> Priority: Major
> Attachments: DstroedCacheReturnTest.java
>
>
> Cache configuration files stay stored on file system after invoke {{destroy}}
> method.
> By the reason after restart grid all removed caches are start.
>
> Look at the test [^DstroedCacheReturnTest.java]
--
This message was sent by Atlassian JIRA
(v7.6.3#76005)
