ApplicationSet controller sets requeueAfter to 0 when using cluster and nested generators

[rumstead]

Checklist:

• I've searched in the docs and FAQ for my answer: https://bit.ly/argocd-faq.
• I've included steps to reproduce the bug.
• I've pasted the output of argocd version.

Describe the bug

ApplicationSets are not being queued.

There is quite a bit of waterfall logic to get the requeue time but I am seeing this line log a time of 0 for both 2.4.14 and 2.5.10. I am not sure why the reconciliation time is much worse on 2.5.10 yet. The controller is saying 0 here. You can see in the controller runtime that 0 means never requeue, or at least their docs state that.

Reproduced on HEAD as well.

To Reproduce

Here are unit tests that reproduce the issue.

rumstead@11146a9#diff-1bc0b86ed20d2c14e5c2d4efb5deed697369c8ed49fe91a2c84fcb48c412739d

Expected behavior

ApplicationSets are requeued with the default time of 3m regardless of their generators.

Screenshots

Version
2.4.14 and 2.5.10

Logs

2.4.10

time="2023-02-09T22:35:54Z" level=info msg="end reconcile" requeueAfter=0s

2.5.10

time="2023-02-09T18:59:44Z" level=info msg="end reconcile" applicationset=argocd/prometheus-appset requeueAfter=0s

[rumstead]

@rishabh625 do you know why List and Cluster have no default requeue time? That seems to be the issue. If we add the DefaultRequeueAfterSeconds, which is 3 minutes, that feels like a good solution.

EDIT: I believe the reason is because of the clusterSecretEventHandler. Which requeues the appset if a cluster secret changes. However, that breaks down when you use the matrix/merge generators with something like a GIT generator.

[rumstead]

When a merge or matrix generator is mixed NoRequeueAfter generator, the code doesn't traverse the nested generators.
Ie,

generators:
    - merge # traverses
           - cluster # returns 0 requeue
           -  matrix # doesn't traverse
               - cluster
               - git