VMOperator sometimes fails to convert pre-existing Prometheus Operator CRs on first install

[ringerc]

During testing I've observed that the VictoriaMetrics Operator sometimes fails to convert Prometheus Operator CRs like PodMonitor and ServiceMonitor to their VictoriaMetrics equivalents when it is first installed in a kube cluster in which the Prometheus Operator resources already exist.

Restarting the operator works, causing the objects to be created as expected.

To reproduce:

• Install Prometheus Operator
• Create a PodMonitor, ServiceMonitor, etc
• Install VM Operator (I used the Helm chart)
• Observe that VM Operator CRs were not auto created
• kubectl rollout restart the vm operator
• Observe that CM operator CRs are now created

I used this Helm values (.txt extension because of GitHub's awful filename based attachment filters):

vmoperator-values-yaml.txt

[vrutkovs]

sometimes fails to convert

iiuc it doesn't fail to convert them - convertion succeeds every time, but it sometimes doesn't get notified about the objects at all, am I correct?

[ringerc]

@vrutkovs Correct - my wording was imprecise. "Fails to notice that objects need conversion" would be more accurate.

[vrutkovs]

This should hopefully be fixed by #1889