Performance reduced to the slowest node?

I have been reading this forum for a few weeks as we are getting ready to test crdb as an alternative for our app and so far I saw that having servers with spinning disk causes terrible performance, which is fine (for us) as we’ll be using ssd drives, but the other day I saw a comment that was something along the lines of, overall cluster performance depends on the slowest node, is that the case?

In short, if I have 7 node cluster, all same specs, but all of the sudden one node is slower (we are seeing this on rackspace), will the whole cluster become slower?

My main concern right now is write performance.



@fmpwizard In the short term the answer is “yes”. A slow node will affect the performance of the entire cluster. This is clearly undesirable and we’ll continue making improvements until the answer is “no”. See

It is worthwhile to monitor per node performance metrics so that you can catch a node that has started having problems. For example, if you use prometheus you can monitor per node disk stats (using node_exporter) and you can get per node latency stats from cockroach (e.g. exec.{success,error,latency}). We’re very aware that this is an unsatisfactory answer.

@peter Thanks for the honest answer. I have subscribed to that ticket and we’ll continue to test crdb to see if it fits our use case.