ApplicationSet tokenRef strict mode changes only apply after the controller is restarted

Ouverte
#1,554 1 commentaire 0 réactions 1 personne assignée Voir sur GitHub

Les mainteneurs répondent en général sous 1 jour

@Gabrielandre02 y travaille déjà.

Depuis le 13/11/2026.

  • #1561 par @Gabrielandre02 — ouverte

Évaluation

Cette issue n'a pas encore été évaluée.

Description

Describe the bug

While working on #1551 I noticed that changing the ApplicationSet tokenRef strict mode on an existing instance has no effect until the ApplicationSet controller Pod is restarted, and the operator doesn't restart it.

The operator writes applicationsetcontroller.enable.tokenref.strict.mode to argocd-cmd-params-cm, and the controller Deployment reads it through a configMapKeyRef (argocd-operator/controllers/argocd/applicationset.go). The controller only reads that variable when it starts, and the Deployment spec is the same for both values, so a change in the ConfigMap never rolls it out.

Two things follow from that:

  1. The opt-out described in docs/usage/appsets-in-any-namespace.md (.spec.cmdParams set to "false") does nothing until someone restarts the Pod. The docs don't mention a restart.
  2. The other direction is the one that worries me. After the opt-out is removed, the CR and the ConfigMap say strict mode is on, but the controller keeps reading unlabeled Secrets through tokenRef until its next restart.

To Reproduce

I ran this on a kind cluster, with the operator running from master (8f72660b) and the default Argo CD image (v3.5.6).

  1. Create a cluster-scoped instance with ApplicationSets in any namespace:

    apiVersion: argoproj.io/v1beta1
    kind: ArgoCD
    metadata:
      name: argocd
      namespace: openshift-gitops
    spec:
      sourceNamespaces: [tr-bar]
      applicationSet:
        sourceNamespaces: [tr-bar]
        scmProviders: ["https://gitlab.invalid"]
    
  2. In tr-bar, create a Secret scm-token without the argocd.argoproj.io/secret-type: scm-creds label, and an ApplicationSet with a Pull Request generator that uses it:

    generators:
      - pullRequest:
          gitlab:
            project: "1"
            api: https://gitlab.invalid
            tokenRef:
              secretName: scm-token
              key: token
    

    The ApplicationSet reports secret tr-bar/scm-token is not a valid SCM creds secret, as it should.

  3. Set applicationsetcontroller.enable.tokenref.strict.mode: "false" in .spec.cmdParams and wait for the ConfigMap to change. Recreate the ApplicationSet: the Secret is still refused.

  4. Run kubectl rollout restart deploy/argocd-applicationset-controller. Now the opt-out is live: the error becomes dial tcp: lookup gitlab.invalid ... no such host, so the token was read and the provider was called with it.

  5. Remove the key from .spec.cmdParams and wait for the ConfigMap to go back to true. Recreate the ApplicationSet: the unlabeled Secret is still read.

This is what I saw at each step. The Pod was the same one from step 4 until I restarted it again:

Step .spec.cmdParams ConfigMap Env in the running Pod Unlabeled Secret
2 not set true true refused
3, 60s after the ConfigMap changed false false true refused
4, after the manual restart false false false read
5, 60s after the ConfigMap changed not set true false read
5, three minutes later not set true false read
after another manual restart not set true true refused

Expected behavior

A change in the effective strict mode value reaches the ApplicationSet controller without a manual restart, in both directions.

Additional context

  • Adding sourceNamespaces to an existing instance works. The Deployment command changes in the same reconcile, so the Pod is replaced, and it came up with strict mode on the three times I tried.
  • The E2E tests for this feature (1-037) check the value in the ConfigMap, not what the controller does, so they pass.
  • I didn't test on OpenShift or with an OLM install, only with the operator running from source against kind.

A possible fix is to set the value on the container from the same logic that fills the ConfigMap, so that a change in the value changes the Deployment. That's what #1551 does for server.disable.swagger.ui. I'm happy to send a PR with unit and E2E tests if that sounds right to you.

Langage dominant
Go
Étoiles
190
Forks
367
Merge moyen
4 j 2 h
PR mergées (30 j)
63

Préparer son environnement

Par où commencer

  1. Lisez l'issue en entier, puis le guide de contribution du projet.
  2. Signalez en commentaire que vous la prenez — cela évite que deux personnes fassent le même travail.
  3. Forkez le dépôt et travaillez sur une branche.
  4. Ouvrez une pull request qui référence le numéro de l'issue.

Autres issues de redhat-developer/gitops-operator

Toutes les issues de redhat-developer/gitops-operator

Issues similaires

Plus d'issues Go

Recevez les nouvelles issues par e-mail

Un résumé court des issues GitHub adaptées aux débutants.