Restart Expectations
Fri, 08/26/2011 - 11:52
I've got a quiet system: no input, no client activity, freshly restated, then rebalanced. 15 machines, plenty of memory, 500M keys, about 1TB of data, replication factor of 1 for all buckets, 90% of data in one of the 4 buckets.
Assume a completely utiopian, spherical cow setup: all nodes are *identical*, and all are working perfectly, etc, etc.
Given that I've just restarted, rebalanced, and made sure that everything is quiet, if I restart the entire cluster again, should I be able to expect that the startup and key-loading times should be roughly equivalent for all nodes?
Thanks!
Paul
Short answer, yes.
Long answer, yes-ish. The main determining factor on warmup speed is the fragmentation of the sqlite database. This is going away with Couchbase Server 2.0...but I'm referring to what's available in Membase right now.
The longer each node is taking writes, the more fragmented the DB is going to be. You can use sqlite's vacuum capability, but unfortunately that can't really be done online. If you're going to restart all the nodes, it would probably be worth vacuuming before restarting.
You can also monitor the time it took for each node to warmup the last time, and the time it's taking currently, by looking at the 'ep_warmup_time' stat.
Perry
Forum support is great for free but sometimes you need a guaranteed response time and dedicated resources for your questions or issues.
Consider purchasing enterprise-level support from Couchbase: http://www.couchbase.com/products-and-services/overview
Call or email "sales -at- couchbase-dot- com" today!