Multiple index routing allocation with require and include

[nopma (what)]

I am trying to set up index routing with require and include but is not working.

I have 3 nodes, and I want one shard to be on node-1 and the second shard to be on node-2 or 3.
But when I try with this index template one of the shards is UNASSIGNED

"settings": {
     "number_of_shards": "1",
     "number_of_replicas": "1",
     "routing" : {
        "allocation" : {
            "include" : {
               "_name" : "es1,es2,es3"
            },
            "require": {
               "_name" : "es3"
           }
      }
   }
}

If i use only include or require is ok. Any ideas ?

[dadoonet (David Pilato)]

I believe that because of the require it can not assign the replica if only one node has the tag es3.

Could you describe why do you want to do this?

[nopma (what)]

the idea is to manually allocate the index which is expected to have high read rates on a separate node so the queries do not affect the performance of other nodes.
e.g. we have 10 indices: 1st is 200GB, others are 20-30GB each, replication will balance it between nodes. Cross-cluster search is an option, but it requires to have multiple clusters.
generally we just need to replicate one single index (or pattern) to another node and keep a copy in the main cluster at the same time.

[dadoonet (David Pilato)]

@nopma

the idea is to manually allocate the index which is expected to have high read rates on a separate node so the queries do not affect the performance of other nodes.

Good idea.

So I'd tag some nodes with a tag like "hot-read" and allocate the index to those nodes (the s is important). With that, you will be able to allocate all primaries and all replicas on all the nodes which has this tag.

Then I guess require is good to use.

And the exclude the other indices from those nodes.

Makes sense?

[nopma (what)]

Yes, thanks