etcd tuning for high latency?

[justyns]

Hello!

Does Talos do any special tuning to etcd to handle high latency? AFAIK the usual recommendations to not let k8s clusters span multiple regions/datacenters is due to etcd not liking high latency by default.

I've had luck in the past setting the parameters mentioned on https://etcd.io/docs/v3.4/tuning/ (heartbeat-interval and election-timeout).

Is this something Talos already handles, or is there a way to set it for etcd?

[rsmitty]

It's not something we do by default, but you can certainly add those tuning parameters yourself with extraArgs in the etcd section of the machine config: https://www.talos.dev/v1.7/reference/configuration/v1alpha1/config/#Config.cluster.etcd

[justyns]

Thanks! I looked for something similar but must have missed it. I created a new patch that does what I want:

cluster:
  etcd:
    extraArgs:
      election-timeout: "50000"
      heartbeat-interval: "5000"