feat(argo-cd): support tpl in env var values

[DrFaust92]

Checklist:

• I have bumped the chart version according to versioning
• I have updated the documentation according to documentation
• I have updated the chart changelog with all the changes that come with this pull request according to changelog.
• Any new values are backwards compatible and/or have sensible default.
• I have signed off all my commits as required by DCO.
• I have created a separate pull request for each chart according to pull requests
• My build is green (troubleshooting builds).

Summary

Switch env var rendering in all component deployments/statefulsets from toYaml to tpl (toYaml .) $, enabling Helm template expressions inside env[].value fields.

This is a backwards-compatible change — plain string values continue to work as before.

Note: when using template expressions, reference root-level values with $.Values.x (not .Values.x) since the dot context inside the env list is the list element.

Example use case — auto-setting GOMEMLIMIT from the container memory limit in a parent chart/values overlay:

server:
  env:
    - name: GOMEMLIMIT
      value: "{{ printf \"%dMiB\" (mulf ((.Values.server.resources.limits.memory | trimSuffix \"Gi\" | float64) | mulf 1024) 0.85 | int) }}"

[mkilchhofer]

Why you don't use resource field refs?

server:
  env:
    - name: GOMEMLIMIT
      valueFrom:
        resourceFieldRef:
          resource: limits.memory
          divisor: "1"

In my opinion this would be the exact use case for this:

$ kubectl  explain pod.spec.containers.env.valueFrom.resourceFieldRef
KIND:       Pod
VERSION:    v1

FIELD: resourceFieldRef <ResourceFieldSelector>


DESCRIPTION:
    Selects a resource of the container: only resources limits and requests
    (limits.cpu, limits.memory, limits.ephemeral-storage, requests.cpu,
    requests.memory and requests.ephemeral-storage) are currently supported.
    ResourceFieldSelector represents container resources (cpu, memory) and their
    output format

FIELDS:
  containerName	<string>
    Container name: required for volumes, optional for env vars

  divisor	<Quantity>
    Specifies the output format of the exposed resources, defaults to "1"

  resource	<string> -required-
    Required: resource to select

[DrFaust92]

Thanks for the suggestion — I looked into resourceFieldRef but it's not viable for this use case.

The goal of setting GOMEMLIMIT is to give Go some headroom below the cgroup memory limit (typically ~90%) so GC kicks in before the kernel OOM-kills the pod. resourceFieldRef has no multiplier — only divisor, which does integer unit conversion. To get ~90% you'd want something like divisor: "1.1", but the API server rejects it:

spec.containers[0].env[0].valueFrom.resourceFieldRef.divisor: Invalid value:
only divisor's values 1, 1k, 1M, 1G, 1T, 1P, 1E, 1Ki, 1Mi, 1Gi, 1Ti, 1Pi, 1Ei
are supported with the memory resource

So resourceFieldRef can only express 100% of limits.memory, which defeats the purpose — at 100% Go thinks it has the entire cgroup budget for the heap, but real RSS includes stack, mmap, runtime overhead, etc., so you're back to OOM risk.

That leaves either hardcoding (drift-prone but explicit) or an entrypoint script that reads the cgroup limit and computes 90% — which is exactly the kind of thing chart-level templating with tpl would let users do cleanly. So this PR is still the right fix.

[mkilchhofer]

@DrFaust92 Perfect answer, thanks for these details! I now also found this issue over there:

• kubernetes/kubernetes#91514

Looks like you already made the right proposal for what is possible today.

[DrFaust92]

Hi @mkilchhofer — just rebased onto current main (resolved a trivial Chart.yaml version conflict, bumped to 9.5.12). The PR has been sitting through several Renovate-driven rebases now and the actual change is small and contained: just wrapping env var values with tpl in the argo-cd templates.

Would you be willing to merge this when you have a moment? Happy to address anything further if you'd like more changes — otherwise I'd love to stop chasing the rebase treadmill. Thanks again for the thoughtful review on the resourceFieldRef discussion.