Multiple replicas of metric server

[yuvalweber]

I had question regarding metricServer high-availability.
In your operational doc you said that metric-server supports multiple replicas and you suggest that but you said that the number of replicas is only 1.
Can you explain if there is a support in multiple metricsServer and if there is something to take in consideration when doing so, and if so what is the number of replicas recommended?

[tomkerkhove]

You can run it with multiple replicas but it's more of an active passive scenario and not a load-balancing solution

[zroubalik]

See this issue: kedacore/keda#2282

[yuvalweber]

I read this issue and I have one final question cause I'm not sure I totally understand.
Regarding pollingInterval you wrote in this issue that it queries the external scaler every time the k8s-server asks him to, but isn't the k8s-server asks him to query the external scaler only based on the pollingInteval.
Like what other cases does the kubernetes-server would ask him for metric? @zroubalik

[zroubalik]

pollingInterval is only for 0<->1 scaling. See this for reference: https://keda.sh/docs/2.9/concepts/scaling-deployments/#caching-metrics-experimental

[yuvalweber]

@zroubalik but regarding what I read from your answer in the issue it seems like regarding this caching mechanism the pollingInterval would work also for scaling as it is. No?

[zroubalik]

If you enable the new feature then yes.