[09:24:22] Hi SIG! I was wondering whether someone had already deployed an HPA using a custom prometheus query, here at WMF [09:28:15] o/ [09:29:00] IIRC no since we haven't set up a metrics server, I am very ignorant so not sure if it can be done pointing to prometheus directly [09:29:38] the ML team uses Knative that offers autoscaling based on rps/concurrent-clients but it is implemented by the framework [09:29:54] (there are dedicated pods in the control plane that gather metrics and decide what to do) [09:35:09] So if we were going this route, something a la https://github.com/kubernetes-sigs/prometheus-adapter would need to be used, I'm guessing [09:37:19] I'm probing with the team to see whether CPU increases when the rps increases, to see whether we can simply using the autoscaling/v1 HPA based on CPU metrics [09:41:05] what is the use case that you have in mind? [09:45:00] gmodena reached out asking whether we could autoscale the flink task managers based on the rps metrics [09:52:26] we don't currently have HPA. But that's mostly due to prioritization, nothing more. [09:52:41] although, it seems we could just rely on the flink operator for this https://nightlies.apache.org/flink/flink-kubernetes-operator-docs-main/docs/custom-resource/autoscaler/ [09:52:42] it's been in the low priority queue for serviceops for quite a while now [09:56:45] ack thanks, that's good to know! [10:34:06] brouberol akosiaris ack. Thanks & good to know.