Upgrade from keda-add-ons-http 0.7.0 → 0.12.0 fails due to immutable Deployment selectors (even with CRD-first + force_update)

[tppalani]

Summary

Upgrading the KEDA HTTP Add-On from chart v0.7.0 to v0.12.0 fails with immutable
Deployment selector errors for all 3 components (interceptor, operator, external-scaler).

This occurs even after:

• installing the updated HTTPScaledObject CRD first (as required in v0.12.0 release notes)
• using Helm --force via Terraform (force_update=true)
• verifying the new CRD schema (metav1.Condition)
• fully deleting HTTPSO resources
• correct sequencing and dependency ordering in Terraform

Error

spec.selector: Invalid value ... field is immutable

Root cause

Chart v0.12.0 appears to introduce new spec.selector.matchLabels for the three Deployments,
which Kubernetes cannot patch. Helm force-update does not seem to replace these resources as expected.

What I expected

A supported upgrade path that allows migrating from older add-on installs to v0.12.0,
especially given that the release notes require CRD-first update for .status.conditions schema.

What happened

• CRD installs successfully
• Helm/Terraform upgrade still attempts to patch selector fields
• Upgrade fails immediately with immutable field errors

[linkvt]

Hi @tppalani ,

I understand that you expect the upgrade to work but can only say that there is nothing we can do right now to fix the behavior.
The helm chart is in alpha state and this change was introduced in 0.11.0, see kedacore/charts#740 in early 2025.

The problem is that deployment selectors are immutable and the only way I know of to resolve this is to remove the deployments before deploying the upgrade via helm.

I have this in mind for future changes which is why I'm tracking all breaking changes for a future v1 release, see e.g. kedacore/charts#816 . Until then we will try to not introduce any more breaking changes.
Chart changes are also not tracked yet in a changelog (also probably unexpected from a user perspective), this is also something we should change.

Let me know if you have more questions.

[tppalani]

Hi @linkvt

thanks for the response can you please guide me which version I should use to avoid this issue?

and also I used terraform helm chart module to install keda and keda-add-one-http.

[linkvt]

0.12.1 is the correct version but you can't avoid the issue, you just have to delete the deployments manually before using terraform to upgrade the chart. Unless you have a lot of clusters it isn't worth automating this deletion via terraform IMO.