[
https://issues.apache.org/jira/browse/OAK-6046?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=15959083#comment-15959083
]
Michael Dürig commented on OAK-6046:
------------------------------------
I think these references are missing as the set of bulk references passed to
{{TarFiles.cleanup()}} is not necessarily consistent with the set of readers
the latter cleans later on: if a new {{TarReader}} is created after collecting
the bulk references in {{GarbageCollector.cleanup()}} that {{TarReader}} is
also cleaned in {{TarFiles.cleanup()}} but its references are missing from the
set of bulk references and thus erroneously gc'ed.
As a fix we could move collecting the bulk references into the lock inside
{{TarFiles.cleanup()}}.
Pre OAK-6002 revisions are not affected as there the bulk references and the
tar readers for cleaning are collected within the file store lock.
> Include references from unpersisted segments when running cleanup
> -----------------------------------------------------------------
>
> Key: OAK-6046
> URL: https://issues.apache.org/jira/browse/OAK-6046
> Project: Jackrabbit Oak
> Issue Type: Bug
> Components: segment-tar
> Affects Versions: 1.8
> Reporter: Francesco Mari
> Assignee: Francesco Mari
> Attachments: OAK-6046-01.patch, OAK-6046-02.patch
>
>
> If a data segment A that is not persisted references a bulk segment B that is
> already persisted, and if A is the only segment that references B, B is
> marked for removal during cleanup.
> To solve this problem, references from A should be included in the initial
> set of references that is fed as input to the cleanup operation.
--
This message was sent by Atlassian JIRA
(v6.3.15#6346)
