All,
Setting up an index cluster of 3 nodes soon and sizing some disks. Feels like you would always want a
replication factor of 2
searchable copies of 2.
But I see that I can in theory set
replication factor of 2
searchable copies of 1, which leaves out tsidx/bloom etc.
Docs really seem to gloss over this. What benefit does this have? What would recovery look like in a RF=2/SC=1 situation with a lost indexer? How would I bring that replicated copy online if I lost an indexer for good and didn't have a second searchable copy?
While this approach can save you some disk space, whereas it compromises some basic advantages for Clustering. For example if the node with the search copy goes down, the process of converting non searchable buckets to searchable and converting them to primary may/would be a time intensive process. which may delay your searches for some time. Or may produce partial results as it may search only over the available primary buckets until its recovery. So its always a best practice to go with a minimum of 2 SF.