cassandra-commits mailing list archives

Site index · List index
Message view « Date » · « Thread »
Top « Date » · « Thread »
From "Dominic Letz (JIRA)" <j...@apache.org>
Subject [jira] [Updated] (CASSANDRA-8546) RangeTombstoneList becoming bottleneck on tombstone heavy tasks
Date Mon, 29 Dec 2014 17:59:13 GMT

     [ https://issues.apache.org/jira/browse/CASSANDRA-8546?page=com.atlassian.jira.plugin.system.issuetabpanels:all-tabpanel
]

Dominic Letz updated CASSANDRA-8546:
------------------------------------
    Attachment: cassandra-2.0.11-8546.txt

Attaching a patch that replaces the plain arrays with a GapList from the brownies java collections
library (apache 2.0 license).

Passes ant test
Improves performance in the worse case scenarios like this on my notebook:

50k deletes: 7 seconds without patch 700ms with patch.
100k deletes: 3 minutes without patch 3 seconds with patch.

> RangeTombstoneList becoming bottleneck on tombstone heavy tasks
> ---------------------------------------------------------------
>
>                 Key: CASSANDRA-8546
>                 URL: https://issues.apache.org/jira/browse/CASSANDRA-8546
>             Project: Cassandra
>          Issue Type: Improvement
>          Components: Core
>         Environment: 2.0.11 / 2.1
>            Reporter: Dominic Letz
>            Priority: Critical
>             Fix For: 2.0.12
>
>         Attachments: cassandra-2.0.11-8546.txt, tombstone_test.tgz
>
>
> I would like to propose a change of the data structure used in the RangeTombstoneList
to store and insert tombstone ranges to something with at least O(log N) insert in the middle
and at near O(1) and start AND end. Here is why:
> When having tombstone heavy work-loads the current implementation of RangeTombstoneList
becomes a bottleneck with slice queries.
> Scanning the number of tombstones up to the default maximum (100k) can take up to 3 minutes
of how addInternal() scales on insertion of middle and start elements.
> The attached test shows that with 50k deletes from both sides of a range.
> INSERT 1...110000
> flush()
> DELETE 1...50000
> DELETE 110000...60000
> While one direction performs ok (~400ms on my notebook):
> {code}
> SELECT * FROM timeseries WHERE name = 'a' ORDER BY timestamp DESC LIMIT 1
> {code}
> The other direction underperforms (~7seconds on my notebook)
> {code}
> SELECT * FROM timeseries WHERE name = 'a' ORDER BY timestamp ASC LIMIT 1
> {code}



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

Mime
View raw message