[
https://issues.apache.org/jira/browse/CASSANDRA-8685?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=14296699#comment-14296699
]
Thomas Steinmaurer commented on CASSANDRA-8685:
-----------------------------------------------
As a DataStax customer and currently/still in a Astyanax/Thrift-based scenario,
I would like to sincerely ask the Cassandra project to investigate including
the newer library in Cassandra 2.0x already. I guess 3.0 is a long-road ahead
until it is ready for production. As far as I can see is that Cassandra 2.0.11
resp. at least DSE 4.5.3 is using 0.9.1 so an upgrade to 0.9.2 might not be
that tricky. I must admit that I haven't investigated the code or library
dependency changes in 0.9.2, but I can say that a simple drop-in replacement of
0.9.2 in our DSE 4.5.3 based loadtest environment works without troubles.
Attached a screen of the Cassandra heap improvement in our tests simply by
replacing thrift 0.9.1 with 0.9.2.
This is in an environment with 3 of our servers hitting a 6 node EC2
(m1.xlarge) based Cassandra cluster with an Astyanax max connection pool
setting of 50 per Cassandra node, thus max. 150 (3 * 50) connections per
Cassandra nodes, where each connection is represented by a TFramedTransport
instance on the Cassandra side which can grow dramatically depending on the
Client/Astyanax single read request size.
> Consider upgrade to thrift 0.9.2 (or later)
> -------------------------------------------
>
> Key: CASSANDRA-8685
> URL: https://issues.apache.org/jira/browse/CASSANDRA-8685
> Project: Cassandra
> Issue Type: Bug
> Reporter: Adam Hattrell
> Assignee: T Jake Luciani
>
> Folks using Astyanax and the like are subject to
> https://issues.apache.org/jira/browse/THRIFT-1457 and may run into heap
> pressure on the Cassandra side for larger read request, as thrift doesn't
> reset its internal buffer. This can lead to larger TFramedTransport
> instances will be kept on the heap.
> I've seen at least one situation where this has saved around 1Gb of heap
> space on average.
--
This message was sent by Atlassian JIRA
(v6.3.4#6332)
