wsrep_slave_threads - how high is high ?


I’ve got a 3 node cluster running on Amazon i2.2xlarge instances (SSD Drives + 60GB Ram + 8 E5-2670 v2 CPU’s).

its working very nice the only problem is that I’m getting a lot of flow controls per day from each node mainly due to our application which does a lot of small writes almost simultaneously.

I’ve been playing around with the amount of slave threads used for parallel replication and i’ve been wondering whats considered a high/good value ? i’ve started with 8 then bounced to 32 and now I’m up to 100 for test purposes, so far the flow-controls seems to have stopped, is this normal ? can i maybe go even higher ? the cpu usage on the nodes is almost idle…

i’ve read somewhere that a good indication is wsrep_cert_deps_distance but this value is always showing 1.000000 on all nodes.


I’d say there is no golden rule, all depends on workload characteristics, hardware, etc. But I think there is no point in setting more then recommended here: