cassandra-commits mailing list archives

Site index · List index
Message view « Date » · « Thread »
Top « Date » · « Thread »
From "Robert Stupp (JIRA)" <j...@apache.org>
Subject [jira] [Commented] (CASSANDRA-7845) Negative load of C* nodes
Date Thu, 28 Aug 2014 18:38:08 GMT

    [ https://issues.apache.org/jira/browse/CASSANDRA-7845?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=14114133#comment-14114133
] 

Robert Stupp commented on CASSANDRA-7845:
-----------------------------------------

Will do the upgrade again - but with data/cl/sc/cfg backed up before and after the upgrade
to be able to reproduce it. Maybe it's reproducible on a single node with data from one of
the "failing" nodes. We did not see any data loss - just these strange load numbers.

> Negative load of C* nodes
> -------------------------
>
>                 Key: CASSANDRA-7845
>                 URL: https://issues.apache.org/jira/browse/CASSANDRA-7845
>             Project: Cassandra
>          Issue Type: Bug
>            Reporter: Robert Stupp
>
> I've completed two C* workshops. Both groups also did an upgrade of C* 2.0.9 to 2.1.0rc6
in a 6 node multi-DC cluster.
> Both groups encountered the same phenomenon that "nodetool status" and OpsCenter report
a negative load (data size) of most (not all) nodes. I did not take the phenomenon seriously
for the first group, because there were only operations guys that "did their best to crash
the cluster". But the second groups did nothing seriously wrong.
> 2.0.9 configuration was the default one with just changed directories (data, cl, caches)
and cluster name. Configurations of 2.1.0rc6 nodes matched the config of 2.0.9 - they just
removed 5 config parameters that were removed in 2.1. They did not run any repair or forced
a compaction.
> After a rolling restart both "nodetool status" and OpsCenter report the correct load.
> I was not able to reproduce this locally.
> I have a third group tomorrow and hope to have some time to do the upgrade again. Anything
that I can check? I think it would be possible to grab the data files from at least one node
for further analysis. Anything else I can do to check that?



--
This message was sent by Atlassian JIRA
(v6.2#6252)

Mime
View raw message