Nodetool Repair Best Practices

[Daniel Subak <da...@klaviyo.com>]

Hey everyone,

We've just migrated to a new Cassandra cluster running 3.7 and wanted to
get some information on best practices when running nodetool repair; our
last cluster was 1.2 and per the documentation it seems that a lot of
behavior has changed between those versions.

From a read of the documentation, it looks like running nodetool repair
without any additional arguments will perform a parallel, incremental on a
specific node. For our first run of a repair, do we need to simply run this
across all nodes in our cluster? When should we specify full vs incremental
when doing repairs? Additionally, what are the implications of running
repairs sequentially vs in parallel with regards to cluster load or
completion time? Are there any best practices around running cluster wide
manual repairs in 3.7? Any advice around how others have managed repairs
would be greatly appreciated.

Thanks!
Dan