Cassandra 2.2 - "nodetool repair" VS "nodetool repair -dc dc1,dc2,dc3"

Question

I want to completely repair a Cassandra 2.2 cluster composed of 3 datacenters (dc1, dc2, dc3) by running only one single command line on a single node. Do I need to run?

nodetool repair

or

nodetool repair -dc dc1,dc2,dc3

or I'm totally wrong? are those 2 commands repairing all token ranges of all cluster nodes?

score 1 · Answer 1 · answered Aug 15 '16 at 22:10

1

For complete repair of any Cassandra cluster, you need to run 'nodetool repair' on all nodes, one by one, at least once before reaching the time specified in the gc_grace_seconds setting.

You can't repair whole cluster running repair command on one node because that node doesn't contain all data.

answered Aug 15 '16 at 22:10

deniszh

142
7

thanks for the clarification. but I never update or detete any row so I think I can forget about gc_grace_seconds setting (defaulting to 10 days) – Greg M. Aug 16 '16 at 08:39
Yes, then you can just run a repair on all nodes in a rolling fashion. – deniszh Aug 16 '16 at 09:15
yes I think I will adopt this strategy to start with, thank you – Greg M. Aug 16 '16 at 17:17

Cassandra 2.2 - "nodetool repair" VS "nodetool repair -dc dc1,dc2,dc3"

1 Answers1