I see some keys from apiserver that are a couple of hundreds of mb

Thanks guys

Some way more k8s native like a job with a base image and the exported task

Ok found the problem

Thanks for the reply 👍

Ok thanks, I’m searching a way to not be dependent on the aws auto scaler script but rather use a k8s job which will run and die after it’s finished thus saving resources

One moment

Reqs to the api

One of the keys:
b”<class ‘apiserver.database.model.base.GetMixin.GetManyScrollState’>/f4010ba7df0f45dbbea10a71fe568a94"

simple helm install, didn’t change anything in values

Returned 400 for users.get_current_user in 3ms, msg=Invalid user (failed loading user)

Hey I have another question:
How to I use gunicorn in api server?
I see that during the build in the entrypoint.sh you have an env called “CLEARML_USE_GUNICORN”
How do I use it in the helm charts or do I need to build a custom image?

Hi SuccessfulKoala55 ,

On the agent pod the tasks logs are written locally to /tmp/.clearml_agent_out.<random num>.txt

I wan’t those logs to also bind to the hosts stdout, with a console handler for example

SuccessfulKoala55 Seems like apiserver uses redis as cache for ui pagination with a long ttl

Thanks! SuccessfulKoala55 I’ll take a look

Yes

I had an old cookie from an old deployment that’s why it didn’t recognize my user…

The cookie has a ttl of 3000+ years…

Awesome guys! thanks a lot 🙌

cloned the repo “clearml-helm-charts”

🙏 Thanks guys!

Helm charts

Also added the “CLEARML_SUPPRESS_UPDATE_MESSAGE” env which didn’t work…

maybe change the session cookie to 24 hours?

And I checked, both envs are set in the agents pod

With