7 Node Cluster with 3 dedicated master and 4 data nodes

Hi All,

I am setting up 7 node cluster with 3 dedicated master only and 4 data only nodes, my hosts are not bare metal hosts but JVM's . What should be my memory setting for each host ( master and data nodes) .
What should be my Xmx and Xmx setting ideally. is 8GB enough ? on each host

For now I will be having 3 indexes ( default 5 shards and 1 replica) only and data size for first 2 indexes would be around 10GB and 3rd index is less than 2 GB.

Later on we will be adding time series index which can grow over 40 GB , keeping these in mind, how should i design my memory requirements.

thanks

you should experiment, but still, try to get as close as you can to the recommended settings: balf the ram to eladtic, and no more than 30.5GB. as for the size of the indices you have, 5*1 sounds too much for the overall size of data you have. try to stick to 30gb - 50gb per shard, do the math accordingly.

You don't need huge heaps for master nodes, maybe 3-4GB each. You can also not follow the 50%:50% rule with them as they don't need to do any filesystem caching for Elasticsearch. 8GB for the data nodes sounds ok though.

They should be the same, never set them separately but use ES_HEAP_SIZE.

So what i got is , Basically for master nodes 4GB RAM ( on 3 nodes) should be fine and however data nodes i need 8GB on 4 nodes , Is 5 shards and 1 replica fine for now? or should i reduce this as well ? considering later on my time series index will be about 40-50GB and growing.

For that size I'd go with 1, maybe 2 shards and a replica set.

If i use 1 shards and 1 replica , wouldn't my data nodes be less as per the formula
number of shards * (number of replicas + 1) >= number of data nodes
1 * ( 1 + 1) > = 4 ( data nodes) ??
and in future if i wanted to increase more data nodes, don't i have to honor above formula to have either more shards or replica's

If you add more nodes then just change the template to increase the shard count for the next day's index.

At the volumes you are working at I wouldn't be too worried about 1 (primary) shard per node :slight_smile:

1 Like

Thanks i am little confused , having 1 shard and 1 replica does not satisfy the formula i was referring to
1(shard) * ( 1+1) > = 4 ? , you mean to say i should reduce number of nodes as well to be inline with this formula?

What I meant to say there is;

At the volumes you are working at I wouldn't be too worried about 1 (primary) shard per index :slight_smile:

So you end up with a primary on one node, the replica on the other.

so i can still have 3 master and 4 data nodes with just 1 shard and 1 replica?

Yes. Because once you have more than 1 index each node will have a shard.

1 Like

Perfect Thanks.

This topic was automatically closed 28 days after the last reply. New replies are no longer allowed.