[
https://issues.apache.org/jira/browse/CASSANDRA-446?page=com.atlassian.jira.plugin.system.issuetabpanels:all-tabpanel
]
Jonathan Ellis updated CASSANDRA-446:
-------------------------------------
Attachment: CASSANDRA-446.patch
Some modifications:
- was using a decoration comparator on the strings returned by getKeyRange
(which are undecorated). fixed. this was the only place using the old String
DK comparators, so removed those too.
- made code use partitioner.convertToDiskFormat explicitly instead of relying
on DK.toString; the two did not return the same values but needed to (e.g.
bloom filter related code was relying on this). making toString return
something human-readable instead removes the temptation of treating them as
interchangeable.
- made OPP DK (null, key) again; there's no reason (?) to generate unnecessary
garbage
> Use DecoratedKey objects in Memtable, SSTableReader/Writer objects
> ------------------------------------------------------------------
>
> Key: CASSANDRA-446
> URL: https://issues.apache.org/jira/browse/CASSANDRA-446
> Project: Cassandra
> Issue Type: Improvement
> Components: Core
> Reporter: Jonathan Ellis
> Fix For: 0.5
>
> Attachments: CASSANDRA-446.patch, CASSANDRA-446.patch,
> CASSANDRA-446.patch
>
>
> See CASSANDRA-420 for introduction of DecoratedKey.
> IMO we should move Memtable and SSTR objects to using DK objects, too. This
> will help Memtable flushes similar to how this helps BMt, and using DK
> objects in SSTR will avoid a ton of potential confusion about whether a
> String is a key or a decorated key.
> As a side benefit, moving the decoration operation out of memtable flush and
> into the insert path will help make better use of multicore systems, since we
> can have multiple write threads but only one core can flush at a time. (See
> CASSANDRA-445.)
--
This message is automatically generated by JIRA.
-
You can reply to this email to add a comment to the issue online.
