I have a 3 node cluster, all master eligible and all having 4 cores, 16gb memory. I have 2 read/write indexes with none of them above 1 GB/1 million documents. Each index is 1 primary shard and 2 replicas.
I've recently noticed that 2 nodes were swapping. I turned on swap permanently and performed a rolling restart. All my indices and cluster have green health now.
- Do I have to change anything (regarding heap size, etc.) or would Elasticsearch calculate a new heap size after the
swapoff
? Current memory usage is around 9GB per node at the moment. - During rolling restart of the cluster, I noticed a change in
docs.deleted
. I assume that this is related to shards relocating, and nothing to be worried about. Am I correct?
A little related final question;
3. When the docs.deleted
go above 50% of the number of documents in index, I see it reset. (I do not perform a _forcemerge or expunge deletes for that matter. This is the expected behaviour right?
Thanks.