[
https://issues.apache.org/jira/browse/HBASE-1521?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=12719281#action_12719281
]
Jonathan Gray commented on HBASE-1521:
--------------------------------------
My opinion on the delete matter is basically that if we are trading off between
making all compactions faster but making Gets in the presence of deletes
slower, then we should do it. Compactions are very important to get fast and
with optimizations in this issue we could potentially get rid of almost all KV
disassembly.
> Optimize codepath for minor compactions
> ---------------------------------------
>
> Key: HBASE-1521
> URL: https://issues.apache.org/jira/browse/HBASE-1521
> Project: Hadoop HBase
> Issue Type: Improvement
> Affects Versions: 0.20.0
> Reporter: Jonathan Gray
> Fix For: 0.20.1
>
>
> There are some additional optimizations in the specialized StoreScanner and
> also in HFile for minor compactions. For example, there is some KV
> disassembling and sanity checking in HFile even though in a minor compaction
> these checks have already been run during the flush.
> Another area to discuss is whether we should actually process deletes during
> minor compactions. It's not especially expensive (ScanDeleteTracker is quite
> simple) but it requires looking at both the row and the qualifier value of
> every single KV. Removing this would drop our axiom that "Deletes only apply
> to later storefiles", which is used during Get processing to have more
> efficient delete handling.
--
This message is automatically generated by JIRA.
-
You can reply to this email to add a comment to the issue online.
