cassandra-commits mailing list archives

Site index · List index
Message view « Date » · « Thread »
Top « Date » · « Thread »
From "Ariel Weisberg (JIRA)" <j...@apache.org>
Subject [jira] [Commented] (CASSANDRA-9932) Make all partitions btree backed
Date Mon, 10 Aug 2015 23:17:45 GMT

    [ https://issues.apache.org/jira/browse/CASSANDRA-9932?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=14680938#comment-14680938
] 

Ariel Weisberg commented on CASSANDRA-9932:
-------------------------------------------

OK, I can be +1 if we plan on following up by adding unit tests for the derived classes of
AbstractBTreePartition and BTree. I am definitely not persuaded by the test situation that
they are done though, and it would be pretty bad for us to release that way. If there is an
opportunity to constrain scope somewhere else and put more effort into this I would be in
favor of that.

I ran the long btree test and and testOversizedMiddleInsert failed for me. I couldn't find
a job running the burn tests. I think we decided against it because this stuff never changes
(oops).

The combine coverage for BTreeTest and LongBTreeTest shows some stuff uncovered. I will try
and figure out a way to post it that format well.


> Make all partitions btree backed
> --------------------------------
>
>                 Key: CASSANDRA-9932
>                 URL: https://issues.apache.org/jira/browse/CASSANDRA-9932
>             Project: Cassandra
>          Issue Type: Improvement
>          Components: Core
>            Reporter: Benedict
>            Assignee: Benedict
>             Fix For: 3.0.0 rc1
>
>
> Following on from the other btree related refactors, this patch makes all partition (and
partition-like) objects backed by the same basic structure: {{AbstractBTreePartition}}. With
two main offshoots: {{ImmutableBTreePartition}} and {{AtomicBTreePartition}}
> The main upshot is a 30% net code reduction, meaning better exercise of btree code paths
and fewer new code paths to go wrong. A secondary upshort is that, by funnelling all our comparisons
through a btree, there is a higher likelihood of icache occupancy and we have only one area
to focus delivery of improvements for their enjoyment by all.



--
This message was sent by Atlassian JIRA
(v6.3.4#6332)

Mime
View raw message