Storage utilization in the RESTBase Cassandra cluster is quite high (some nodes faring worse than others):
Description
Event Timeline
I did some nodetool cleanups on some of the worst of the nodes, and lowered utilization by a few percent.
The RESTBase mobile-sections tables (corresponding to Cassandra keyspaces {commons,enwiki,others,wikipedia}_T_mobileoZCBVtILw5eSrwi0VIGaFVSr2jY) seem to be no longer in use (having been superseded by PCS?). See:
Combined that's almost 15TB of storage:
I propose that we TRUNCATE these tables (leaving the schema in place for now). Any objections @Jgiannelos? @akosiaris ?
Ok, here (for posterity sake) is what I will apply:
TRUNCATE "commons_T_mobileoZCBVtILw5eSrwi0VIGaFVSr2jY".data; TRUNCATE "enwiki_T_mobileoZCBVtILw5eSrwi0VIGaFVSr2jY".data; TRUNCATE "others_T_mobileoZCBVtILw5eSrwi0VIGaFVSr2jY".data; TRUNCATE "wikipedia_T_mobileoZCBVtILw5eSrwi0VIGaFVSr2jY".data;
Mentioned in SAL (#wikimedia-operations) [2024-08-28T15:52:41Z] <urandom> TRUNCATE-ing RESTBase tables ({commons,enwiki,others,wikipedia}_T_mobileoZCBVtILw5eSrwi0VIGaFVSr2jY) — T342148
That done, I'll give it a couple of hours before clearing the snapshots and actually reclaiming the space (out of an abundance of caution).