Index replication disabled...remove replicated buckets?

[ckurtz]

I had Index Clustering and Replication (Rep Factor 3) enabled on my Splunk 5.0.4 indexer cluster, but had to disable replication because my cold storage (NFS) was too slow -- replication was causing the cluster to panic as the storage became farther and farther behind.

I now have several dozen GBs of replicated buckets (starting with rb) on the NFS storage that the indexers are no longer referencing, since replication is disabled (replication factor = 1).

To be clear, I am still Clustering, just have replication and search factors set to 1.

Can I safely delete the buckets in cold (labeled as rb*) to reclaim space? Or will these buckets age to frozen properly?

I plan on re-enabling Replication once we upgrade to 6 (in several months) and can replicate hot->hot.

[ckurtz]

Official answer from support is to NOT remove any replicated buckets even with clustering disabled, as they may be marked as the Primary Bucket. It is best to let them age out.

[sowings]

Good to know; thanks for following up.

[sowings · Accepted solution]

My interpretation of this is that yes, you can remove the rb* from cold, but that you shouldn't do so with a live index. Thawed is the only bucket lifecycle intended to be tolerant of "new bucket", "delete a bucket" behavior while online, to the best of my knowledge.

So, shut down, remove (or simply relocate the directory) with the rb_ colds, and then bring Splunk back up.

Note that if you set repFactor = 0, the bucket names will no longer have the rb_GUID form.