Selenium Grid Scaler should consider "maxSession"

[d-patch]

Proposal

The graphql API of Selenium Grid allows to query the "maxSession" value per grid and/or node:
https://www.selenium.dev/documentation/grid/advanced_features/graphql_support/#querying-the-number-of-maxsession-and-sessioncount-in-the-grid-
That is, if maxSession=5 each node can hold up to 5 concurrent sessions. Therefore, if there are 10 requests in the queue, only 2 nodes are neccessary to fulfill the requirements. Currently, the Selenium Grid Scaler of KEDA will scale up to 10 nodes in this case, ignoring the maxSessions per node.

Use-Case

No response

Anything else?

No response

[zroubalik]

Seems reasonable, are you willing to contribute this?

[d-patch]

Sorry, but unfortunately I'm merely a user and not a developer.

[adborroto]

I can take this one.

[JorTurFer]

Thanks a lot! ❤️

[amit337496]

@adborroto Any update on this issue

[adborroto]

@adborroto Any update on this issue

PR open

[amit337496]

@JorTurFer / @adborroto I see this issue is closed... Is there any version we need to get this

[JorTurFer]

Hi @amit337496
Next release will be during the first half of May. If you need to test this feature, you could use main tag, but we don't recommend it for prod environments because main is built every commit and could be not stable

[amit337496]

@JorTurFer Thank you.. I am able to see changes in 2.7 version.

How this scale down will happen in this scenario.. If node is still running will it kill that?

[JorTurFer]

hi @amit337496
I'm not sure if I get your question. Do you mean how the scale in will be done?
If that's your question, KEDA doesn't take care about that because it's the HPA Controller who manages it. The HPA kills pods with task in progress, so your job could be interrupted. If you have use cases with long-running tasks or when your system doesn't retry non-acknowledge messages, you should use ScaledJob instead of ScaledObject.

[amit337496]

My scenario is I have 10 test cases and I set max nodes to 5. So it scaled up 5 servers and execution started... But for period of time scale down happened even if they are active sessions

[JorTurFer]

Did you check ScaledJob? It, s the solution for long term processes