Monitoring - Mastering Apache Cassandra - page 288

• Periodic repair : Periodic repair should be executed on every node periodically
within gc_grace_seconds , normally within 10 days. Running the repair
weekly, timed at a relatively low-traffic zone, is a good idea. It is suggested that
periodic repair should be executed with the -pr option.
• Outage : When a node goes down long enough to get hinted handoff, since the
node may have been deleted, repair should be executed. Note that you should
not use the -p r option in this case.
A typical complete node repair looks as follows:
$bin/nodetool -h cassandra01.naishe.in repair
[2014-11-29 15:07:49,923] Nothing to repair for keyspace
'system'
[ -- snip -- ]
[2014-11-29 15:24:39,178] Repair command #1 finished
[2014-11-29 15:24:39,264] Nothing to repair for keyspace
'system_auth'
[2014-11-29 15:24:39,270] Starting repair command #2,
repairing 507 ranges for keyspace system_traces (seq=true,
full=true)
[2014-11-29 15:25:03,035] Repair session
d6152290-77db-11e4-b488-d74579598740 for range
(38800545708018974,96096257197984437] finished
[2014-11-29 15:25:03,035] Repair session
d61c9ca0-77db-11e4-b488-d74579598740 for range
(-6999654802673786063,-6991737081345190369] finished
[2014-11-29 15:25:03,035] Repair session
d6261280-77db-11e4-b488-d74579598740 for range
(8899547794114173830,8919386680314068240] finished
[ -- snip -- ]