[
https://issues.apache.org/jira/browse/OAK-2140?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=14194416#comment-14194416
]
Alex Parvulescu commented on OAK-2140:
--------------------------------------
I believe the outcome of this is quite hard to assess, so I'm proposing
exposing this via a flag that controls if a #clone call should by a shallow
clone or a deep clone. this is only called via the compaction bits anyway, and
it will allow the user to choose between the 2 options, but we'll probably have
to document the impact of each choice:
- shallow clone: a more reduced repository growth during compaction itself,
but a less effective cleanup after
ex: using a 1G repo: first compaction, repo size is at 1.1G, second 1.2G,
third 1.3G, forth back to 1Gb.
the example is highly subjective, and cannot be used as a reference for bigger
repositories, and because of the interweaved references between segments, the
size growth has been known to be a lot bigger for larger repos, and the
stabilization to the original size can occur after a larger number of runs.
- deep clone: the repo size effectively doubles during compaction, but goes
back to the original size after cleanup
ex: 1G initial repo, during the first compaction repo size goes up to 2G,
after the cleanup is done, back to 1G, the process repeats itself for the
following compaction runs.
> Segment Compactor will not compact binaries > 16k
> -------------------------------------------------
>
> Key: OAK-2140
> URL: https://issues.apache.org/jira/browse/OAK-2140
> Project: Jackrabbit Oak
> Issue Type: Bug
> Components: core, segmentmk
> Reporter: Alex Parvulescu
> Assignee: Alex Parvulescu
> Fix For: 1.1.3
>
> Attachments: OAK-2140.patch
>
>
> The compaction bit rely on the SegmentBlob#clone method in the case a binary
> is being processed but it looks like the #clone contract is not fully
> enforced for streams that are qualified as 'long values' (>16k if I read the
> code correctly).
> What happens is the stream is initially persisted as chunks in a ListRecord.
> When compaction calls #clone it will get back the original list of record
> ids, which will get referenced from the compacted node state [0], making
> compaction on large binaries ineffective as the bulk segments will never move
> from the original location where they were created, unless the reference node
> gets deleted.
> I think the original design was setup to prevent large binaries from being
> copied over but looking at the size problem we have now it might be a good
> time to reconsider this approach.
> [0]
> https://github.com/apache/jackrabbit-oak/blob/trunk/oak-core/src/main/java/org/apache/jackrabbit/oak/plugins/segment/SegmentBlob.java#L75
--
This message was sent by Atlassian JIRA
(v6.3.4#6332)
