Hey folks 👋 I have one (last🤞) question regarding the k8s deployment — is the Helm chart provided by Dagster expected to force a new deployment of the user code containers (in

userDeployments

) when running a

helm upgrade

command? That hasn’t worked for me so far, I’ve only been able to get the user code to update by completely re-installing through

helm uninstall

/

install

🤔 I read about the

--recreate-pods

flag, but this has been deprecated in Helm 3. Do we need to generate a unique tag (instead of the default

:latest

) in

values.yaml

to ensure that the new image is pulled when running an upgrade?

Helm upgrade does a diff between the generated kubernetes manifests to determines whether to update the kubernetes object

hey Charles, yes I think a unique tag for each new image version you want to deploy is what you want. Otherwise AFAIK Helm doesn’t know that the tag

latest

now resolves to a different image hash

We could introduce some dynamic metadata in the deployment template spec to replicate the behavior of `--recreate-pods`: https://stackoverflow.com/questions/53450759/helm-upgrade-install-isnt-picking-up-new-changes

Ok awesome, thanks for confirming 👍 I had also read about annotating deployments with a checksum (i.e. Automatically Roll Deployments), but wasn’t sure if that was already set up in the Dagster-provided Helm chart and simply not working on my end for some reason. I will generate a unique image tag in the meantime. Thanks! 🙏