cassandra-commits mailing list archives

Site index · List index
Message view « Date » · « Thread »
Top « Date » · « Thread »
From "Francois Richard (JIRA)" <j...@apache.org>
Subject [jira] [Commented] (CASSANDRA-11504) Slow inter-node network growth & gc issues with uptime
Date Tue, 05 Apr 2016 18:15:25 GMT

    [ https://issues.apache.org/jira/browse/CASSANDRA-11504?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=15226821#comment-15226821
] 

Francois Richard commented on CASSANDRA-11504:
----------------------------------------------

Please note that for specific nodes, we have enabled all the Cassandra metrics (via the CSVReporter)
and at this point nothing seems to correlate with the increase of network traffic between
nodes.

> Slow inter-node network growth & gc issues with uptime
> ------------------------------------------------------
>
>                 Key: CASSANDRA-11504
>                 URL: https://issues.apache.org/jira/browse/CASSANDRA-11504
>             Project: Cassandra
>          Issue Type: Bug
>         Environment: Cassandra 2.1.13
>            Reporter: Jeff Griffith
>         Attachments: InterNodeTraffic.jpg
>
>
> We are looking for help troubleshooting our production environment where we are experiencing
GC problems. After much experimentation and troubleshooting with various settings, the only
correlation that we can find with a slow growth in GC is a slow growth in network traffic
BETWEEN cassandra nodes in our cluster. As an example, I have attached an example where in
a cluster of 24 nodes, i restarted 23 of them. Note that the outgoing rate for that 24th node
remains high while all others drop after the restart. Also note that this graph is ONLY traffic
between cassandra nodes. Traffic from the clients remains FLAT throughout. Analyzing column
family stats shows they are flat throughout. Cache hit rates are also consistent across nodes.
GC is of course its own can of worms so we are hoping this considerable increase in traffic
(more than double over the course of 6rs) between nodes explains it. We would greatly appreciate
any ideas as to why this extra network output correlates to uptime or ideas on what to "diff"
between the nodes.



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

Mime
View raw message