I’m having a very uneven distribution of 100usec and 1msec query time for following test command:
cbs-pillowfight -h localhost -b default -i 1 -I 10000
It’s a two-node cluster running on i7-2600’s.
Problem shows up ONLY when running in a cluster; single node does not show that degradation.
Configuration: there are 2 writers configured for the test bucket.
probably a 3-node configuration will solve data/replica issues, but I still see no reason for 2-node cluster without replicas to affect write request latency % this much…
it looks like you are running pillowfight from the server that is running couchbase as well. that means your load generation is competing with couchbase server - I think that could cause such unpredictable results. is there a way to separate load generation to a new machine?