Open dahankzter opened 4 years ago
any update on that ?
There is an external autoscale project that works, you can try it out now
On Wed, Aug 11, 2021 at 10:26 AM Meydan Ozeri @.***> wrote:
any update on that ?
— You are receiving this because you are subscribed to this thread. Reply to this email directly, view it on GitHub https://github.com/scylladb/scylla-operator/issues/80#issuecomment-897012053, or unsubscribe https://github.com/notifications/unsubscribe-auth/AANHURPMWFRRTYCOBAII34DT4KXERANCNFSM4MZQUVYA . Triage notifications on the go with GitHub Mobile for iOS https://apps.apple.com/app/apple-store/id1477376905?ct=notification-email&mt=8&pt=524675 or Android https://play.google.com/store/apps/details?id=com.github.android&utm_campaign=notification-email .
We had some discussions about autoscaling direction this week (centered around upstream HPA), @rzetelskik is looking into options and will be working a proposal that will land in https://github.com/scylladb/scylla-operator/tree/master/enhancements/proposals
(the student project Dor is referencing is https://github.com/scylladb/scylla-cluster-autoscaler)
Hello - is there any update on that?
I've got one proposition - maybe instead of 1 CRD which is ScyllaCluster
we can have 2 or 3 CRDs:
ScyllaCluster
ScyllaClusterRack
ScyllaClusterDatacenter
ScyllaCluster
will have then definition like:
...
spec:
datacenter:
name: "simple-datacenter" # it can be also get through labels/selectors
...
ScyllaClusterDatacenter
will be:
metadata:
name: simple-datacenter
spec:
racks:
- name: "us-east-1a" # it can be also get through labels/selectors
ScyllaClusterRack
will be:
metadata:
name: "us-east-1a"
spec:
members: 3
...
Thx to that we will be able to use k8s scale subresource on the ScyllaClusterRack
and instead of editing the CR we will be able to do kubectl scale --replicas=4 scyllaclusterrack/us-east-1a
Scale subresource can look like this in the
subresources:
scale:
specReplicasPath: .spec.members
statusReplicasPath: .status.members
Thanks to this for autoscaling later can be use some external tool like KEDA
What do you think about it?
The Scylla Operator project currently lacks enough contributors to adequately respond to all issues.
This bot triages un-triaged issues according to the following rules:
lifecycle/stale
is appliedlifecycle/stale
was applied, lifecycle/rotten
is appliedlifecycle/rotten
was applied, the issue is closedYou can:
/remove-lifecycle stale
/close
/lifecycle stale
We should enable and use the "Horizontal Pod Autoscaler" (HPA) to support automatic scale up and downs. We have to make sure there is a latency in the behavior so there can be a smooth and controlled scale up and down. We don't want it to jump up and down...