Very slow performance after setting replicas

Elastic Stack Elasticsearch
1

Hi all,
I have a cluster of 4 ec2 nodes with 4 cores and 35gb ram each.

The cluster worked pretty good(with 0 replicas), until I changed all
indexes to have 1 replica, then everything became terribly slow.

The cluster size is about 300gb(with replica 600).

Does adding replicas cripple performance by that much?
I imagined it would actually be beneficial to
search performance, because more nodes can serve the same data(slower
indexing though, makes sense).

Thanks,
Ben

2

By enabling replicas for all indices, you doubled the number of shards in
your cluster. Each shard has memory overhead associated with it and
maintains several caches. So, by doubling the number of shards, your
doubled the amount of memory that these shards are using. It's possible
that your cluster no longer has enough memory to properly allocate all
caches that it used to have.

On Sunday, April 1, 2012 7:22:46 AM UTC-4, Ben Aviram wrote:

Hi all,
I have a cluster of 4 ec2 nodes with 4 cores and 35gb ram each.

The cluster worked pretty good(with 0 replicas), until I changed all
indexes to have 1 replica, then everything became terribly slow.

The cluster size is about 300gb(with replica 600).

Does adding replicas cripple performance by that much?
I imagined it would actually be beneficial to
search performance, because more nodes can serve the same data(slower
indexing though, makes sense).

Thanks,
Ben

3

On Sun, Apr 1, 2012 at 13:22, Ben Aviram benaviram@gmail.com wrote:

Does adding replicas cripple performance by that much?
I imagined it would actually be beneficial to
search performance, because more nodes can serve the same data(slower
indexing though, makes sense).

Your issue is most likely what Igor pointed out, but also indexing
shouldn't slow down with replicas depending on your refresh settings.
So if you're looking to add reliability you can also have fast
indexing.