Risks of cluster-init to increase server quota in a 2.2 cluster

Hi,
I have a 13 node cluster that I want to increase ramsize (http://www.couchbase.com/docs/couchbase-manual-2.0/couchbase-admin-tasks-quotas.html) to use more ram per server.
The reason is I am rebalancing 4 nodes out( currently 17 in ) and those 4 had significantly less RAM than the rest.

I want to go from a 52G to 100G, this nodes have 141G of RAM.

First question is, how risky is it? will the memcached process just allocate the extra ram inmediately? or will restarts happen? Should I expect data issues or performance issues? How long does it take?

Second, is 100G a reasonable number on 141G servers?

Thanks

Hi there, you should be totally fine to go ahead and increase that quota. The change will happen immediately, and there’s no impact to perform or behavior.

All the change is really doing is “allowing” the memcached process to allocate more RAM as needed.

Increasing is totally fine. Even decreasing is fine, but you have to be more careful if decreasing the quota will lead to data being ejected from RAM and therefore cause more reads to come from disk.

And yes, 100G is find for a server that has 141. I would say you could go as high as 112 being safe.

Hope that helps

Perry

Great news!!
Can I do this during a rebalance? Or is it discouraged? By how you described it seems no problem.
Just to make sure, the value of currently reserved data I am extracting it from ep_max_size on a single node.

I would say not to do it during a rebalance. No matter when you do it, you need to make sure that all nodes have enough RAM to support the quota increase. So if you’re still doing the rebalance, that likely means that at least one of the nodes in the cluster doesn’t have enough RAM yet.

“ep_max_size” is not actually reserved data, it’s the maximum amount of RAM that can be allocated for that bucket.

“mem_used” is the amount of RAM actually in use if that’s what you’re looking for.

If you haven’t seen this page (http://docs.couchbase.com/couchbase-manual-2.2/#ram-quotas) it’s a good read on the quotas. There are actually two types. When the cluster is first created, that’s the “server” quota and then out of that is allocated the different “bucket” quotas. ep_max_size is specific to each bucket.

Yes, you should be able to raise that quota during a rebalance without any disruption.

so, ep_max_size should be == to ramsize when the cluster was inited I assume?
I am still doing the rebalance, but I can grow them all by 5G and I was thinking on doing it during rebalance if it was not dangerous.