ApplicationSet List Generator in ArgoCD v2.8: Unclear Prune Propagation When Multiple Generators Overlap

Answer to the Core Questions

1. Does a prune flag on one ApplicationSet affect another?

No. The prune field is evaluated on a per‑ApplicationSet basis. If prune: true is set on ApplicationSet A, only the resources that ApplicationSet A generated will be considered for deletion during its sync. ApplicationSet B, even if it produces identical manifests, will not be impacted by A’s prune state unless B itself has prune: true.

When both ApplicationSets generate the same resource, the one with prune: true will delete its own copy after sync. The other ApplicationSet will retain the resource if it has prune: false, leading to a “stale” duplicate that can persist until the first ApplicationSet’s prune runs again.

2. Is there a deterministic pruning mechanism?

ArgoCD v2.8 does not expose a built‑in deterministic policy for overlapping generators. The controller processes syncs sequentially per ApplicationSet, but the overall order is governed by the controller’s reconcile loop and can vary between runs. Therefore, you cannot rely on a guaranteed deletion order without explicit coordination.

3. What logs show a race condition?

• Multiple Prune started events with nearly identical timestamps for different ApplicationSets.
• Interleaved deleted events for the same resource, each prefixed by the ApplicationSet name.
• Repository server errors such as repo server timeout or failed to fetch file occurring during concurrent syncs.
• Kubernetes events showing Normal Pruned from the same resource but with different InvolvedObject references.

Practical Steps for Your Cluster

1. Enable verbose sync logging for both ApplicationSets:

   kubectl patch applicationset myset -n argocd -p '{"spec":{"syncPolicy":{"automated":{"prune":true,"selfHeal":true,"syncOptions":["PrunePropagationPolicy=foreground"]}}}}'
   

2. Trigger a sync on each ApplicationSet (or wait for the next automated sync) and capture the controller logs:

   kubectl logs -n argocd -l app.kubernetes.io/component=application-controller -c application-controller --since=1h
   

3. Sort the logs by timestamp and look for overlapping Prune started entries. Note the resource deletion lines that follow.
4. Verify the final state of the shared resource with:

   kubectl get deployment myapp -o yaml
   

5. If you see that the resource was deleted by one ApplicationSet and recreated by another, you have a race condition. To avoid it, consider:
• Setting prune: false on one of the overlapping ApplicationSets.
• Using a single ApplicationSet that aggregates all desired generators.
• Adding a syncPolicy.prunePropagationPolicy=foreground to enforce a specific delete order.

Missing Diagnostic Detail

To refine the recommendation, could you share the exact timestamps of the Prune started events for both ApplicationSets? This will confirm whether the syncs truly overlapped or were simply back‑to‑back.