cassandra-user mailing list archives

Site index · List index
Message view « Date » · « Thread »
Top « Date » · « Thread »
From Alain RODRIGUEZ <>
Subject Re: Test repair command
Date Tue, 05 Sep 2017 18:49:23 GMT

To check that repairs are running, you should be able to see VALIDATION
compactions (using  'nodetool compactionstats -H'). You could also see some
streaming if you had some entropy (using 'nodetool netstats -H | grep -v
100%') during the repair.

To know roughly the state of your data, you can try to use the metric
( The metric
is exposed in 'nodetool cfstats', according to the previous comment.

A more accurate way I used in the past is to grep logs. You should have a
known number of segments to repair per table depending on the number of
vnodes and the number of nodes.

The thing is you should be able to check that a consistent number of
segments is being repaired for each repair you run, as long as you run it
the same way.

You can try to find "repair" "Success" in your logs, grep this and then
count for each table and node. Comparing to the expected number or the
number from the previous run, you should manage to know if *every* segment
was repaired or not. This works even on the older Cassandra versions.

Finally, you can have a look at that Spotify
created and TLP is now maintaining. This tool automatically manages repairs
and it handles segment repair failures among other things.

Alain Rodriguez - @arodream -

The Last Pickle - Apache Cassandra Consulting

2017-09-04 13:02 GMT+01:00 kurt greaves <>:

> Try checking the Percent Repaired reported in nodetool cfstats​

View raw message