cassandra-user mailing list archives

Site index · List index
Message view « Date » · « Thread »
Top « Date » · « Thread »
From Meg Mara <>
Subject RE: Does "nodetool repair" need to be run on each node for a given table?
Date Tue, 14 Mar 2017 19:08:56 GMT
Yes, whichever node initiates “nodetool repair” becomes the co-ordinator node for that
repair operation and only the token ranges it owns are repaired. Other nodes which own data
in that “repaired token range” will also be involved in this repair operation, create
their merkle trees and send it to coordinator for comparison.

There is a lot to explain when it comes to repair and its many options (-pr, -inc, -par etc).
I recommend the following two links which provide the answers to all your questions.


-          Meg Mara

From: Thakrar, Jayesh []
Sent: Tuesday, March 14, 2017 1:26 PM
To: daemeon reiydelle <>;
Subject: Re: Does "nodetool repair" need to be run on each node for a given table?

Thanks Daemon - it’s the documentation that prompted me to ask the question.

The doc points to further info on incremental, and would I be correct to interpret that nodetool
initiates a repair on a specific node (i.e. the node specified with -h or the current node)?


From: daemeon reiydelle <<>>
Date: Monday, March 13, 2017 at 5:02 PM
To: <<>>
Subject: Re: Does "nodetool repair" need to be run on each node for a given table?

​ find it helpful to read the manual first. After review, I would be happy to answer specific


Daemeon C.M. Reiydelle
USA (+1) 415.501.0198
London (+44) (0) 20 8144 9872

On Mon, Mar 13, 2017 at 1:30 PM, Thakrar, Jayesh <<>>
I understand that the nodetool command connects to a specific server and for many of the commands,
e.g. "info", "compactionstats", etc, the information is for that specific node.
While for some other commands like "status", the info is for the whole cluster.

So is "nodetool repair" that operates at a single node level (i.e. repairs the partitions
contained on the target node?).
If so, what is the recommended approach to doing repairs?

E.g. we have a large number of tables (20+), large amount of data (40+ TB) and a number of
nodes (40+).
Do I need to iterate through each server AND each table?


View raw message