itiyama

@anasalkouz Please re-open this issue. This is an ongoing issue in one of our clusters. If you know of a fix, I would like to backport it to older versions...

One of the ways that I can think of solving this: 1. Primary should never fail the replica when primary itself is cancelling replication operations for its own failure. Currently...

@nknize What are your thoughts?

> So what is the duration of rest_actions numbers? Total since node start? It will become extremely difficult to make those useful across a distributed cluster. For example after running...

This is a good idea and can improve network performance depending on the cluster topology. Think of this example: N1: P1, R2,R3 N2: P2, R1,R3 N3: P3, R1, R2 With...

Yes, we should cache it.

@dblock @nknize What are your thoughts?

Bloom filter is probabilistic, so just using bloom filter will not be enough for determinism. We will use it in addition to some other data structure to optimize read performance,...

@muralikpbhat What do you think?

These improvements will prove useful even for customers using auto generated ids. The ids are unique per document and are stored as terms in FST. This causes the FST index...