SmartScaler : about behaviour and triggers

[paullebovic]

Hello everyone,

I'm managing around 20 opensearch cluster using the operator.
To ease our updates duty, I'm looking into the confMgmt.smartScaler: true option.

Since the documentation is really vague about what it exactly does and how it is triggered, does anyone have further details on this feature .

I tried to test it on the following configuration (preemptively drained the warm pod to prevent data loss) :

• confMgmt.smartScaler: true (kubectl describe opensearch shows it)
• 2 warm nodes hosted on my-host-01 (cluster-warm-0 & cluster-warm-1)
• 1 warm node hoster on my-host-02 (cluster-warm-2)

With an OS Update to run on my-host-02, I'm trying to trigger smartScaler to :

• Mark the node as excluded (move data to other warm nodes)
• Once cluster-warm-2 is emptied, pod is destroyed, pvc too
• cluster-warm-2 (pod & pvc) scheduled on another host.

First of all :

1. Is my understanding of the smartScaler feature right ?
2. I was hoping kubectl drain my-host-02 would trigger the above behaviour, but it doesnt.

Hope I was clear in my explanations (english ain't my native langage).
Only content i've found about that topic is an unanswered opensearch forum question
Edit : Also found discussion avout auto scaling but I'm not quite sure if the answer also covers the topics I'm talking about.

Best regards,
Paul

[swoehrl-mw]

Hi @paullebovic.
The SmartScaler only kicks in when scaling down a nodepool (decreasing the number of replicas of a nodepool in the cluster spec). If active it will exclude the pod in Opensearch, trigger a drain, then remove the pod and remove the exclusion to clean up.
It does not deal with evictions/drains on the Kubernetes side for host replacements/restarts. That is handled by Kubernetes logic alone. For that I recommend you set a Pod Disruption Budget to limit the pods being restarted.