[
https://issues.apache.org/jira/browse/OAK-5004?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=15608381#comment-15608381
]
Michael Dürig commented on OAK-5004:
------------------------------------
bq. no more offline compaction predicate?
This is now covered by the cost function, which determines the priority an item
is given in the cache. Items with low priority have a higher chance of being
evicted.
bq. splitting the old cache into 2: node cache and binary cache.
I did this because there is no easy way to align the cost functions for nodes
and binaries. I think there is no way to decide whether a binary of say length
100000 is more expensive than a node with 100 child nodes. Having two caches
means nodes and binaries cannot cause each other to be evicted.
bq. no more caching of recordids for external binaries at compactor level, this
may be taken care of at the SegmentWriter level, but not sure
I think these are just filling up the cache and don't make sense as duplicating
just the ids shouldn't take up much space. It might be more important to save
the cache slots for the expensive stuff.
> Offline compaction explodes checkpoints
> ----------------------------------------
>
> Key: OAK-5004
> URL: https://issues.apache.org/jira/browse/OAK-5004
> Project: Jackrabbit Oak
> Issue Type: Improvement
> Components: segment-tar
> Reporter: Michael Dürig
> Assignee: Michael Dürig
> Labels: compaction, gc, tooling
> Fix For: 1.6, 1.5.14
>
> Attachments: OAK_5004.patch
>
>
> Running offline compaction on a repository with checkpoints will explode
> those into full copies. Observed e.g. with OAK-5001.
> I think we should consider improving this by compacting checkpoints on top of
> each other in the proper order ({{oak-upgrade}} does this successfully).
> [~alex.parvulescu], WDYT? What was our take on this in the previous Oak
> versions?
--
This message was sent by Atlassian JIRA
(v6.3.4#6332)
