fix: don't render PodMonitor and HPA for disabled components

[mouchar]

Fixes #716

Motivation

The PodMonitor (or VMPodScrape, when victoria-metrics-k8s-stack is enabled) for the proxy, broker, bookkeeper and autorecovery components was guarded only by <component>.podMonitor.enabled, which defaults to true. As a result it was rendered even when the component itself was disabled via components.*=false, leaving scrape configs whose selectors match no pods — harmless to Prometheus, but permanent drift in GitOps setups and empty targets/series in dashboards. Disabling a component required disabling it twice.

proxy-hpa.yaml and broker-hpa.yaml had the same missing guard. That case is rarer, since it also requires autoscaling.enabled=true, but the consequence is worse: the HPA targets a StatefulSet that was never created, so the HPA controller sits at ScalingActive=False / FailedGetScale and emits warning events continuously.

The zookeeper, function-worker and oxia pod monitors already check components.*, as does every other template belonging to these four components (statefulset, service, configmap, pdb, service-account) — these six were the outliers.

Modifications

Added the missing components.* condition to six templates, keeping the existing standalone.enabled checks where present:

Template	New guard
proxy-podmonitor.yaml	and $.Values.components.proxy $.Values.proxy.podMonitor.enabled
broker-podmonitor.yaml	and $.Values.components.broker $.Values.broker.podMonitor.enabled (not $.Values.standalone.enabled)
bookkeeper-podmonitor.yaml	and $.Values.components.bookkeeper $.Values.bookkeeper.podMonitor.enabled (not $.Values.standalone.enabled)
autorecovery-podmonitor.yaml	and $.Values.components.autorecovery $.Values.autorecovery.podMonitor.enabled (not $.Values.standalone.enabled)
proxy-hpa.yaml	and .Values.components.proxy .Values.proxy.autoscaling.enabled
broker-hpa.yaml	and .Values.components.broker .Values.broker.autoscaling.enabled (not .Values.standalone.enabled)

The now-stale # deploy <component> PodMonitor only when ... is true comment above each pod monitor guard was updated to match.

No new tests: the chart has no harness that can assert a resource is absent (the template checks in CI are kubeconform schema validation over rendered output), and adding one would be a larger change than the fix itself.

Verifying this change

Before the change, both of these rendered resources for components that were switched off; after it, both render nothing:

$ helm template t charts/pulsar \
    --set components.proxy=false --set components.broker=false \
    --set components.bookkeeper=false --set components.autorecovery=false \
    --set components.zookeeper=false \
    --set proxy.autoscaling.enabled=true --set broker.autoscaling.enabled=true \
  | awk '/^# Source:/{s=$3} /^kind: (PodMonitor|VMPodScrape|HorizontalPodAutoscaler)/{print $2, s}'

No behavior change with default values — all five pod monitors still render, and both HPAs still render with autoscaling.enabled=true.

Also verified locally: helm lint clean; the CI "all features enabled" renders (.ci/templates-all-values.yaml, and with the patch1 overlay, at --kube-version 1.36.0) and the .ci/clusters/values-standalone.yaml scenario all render successfully.

Note this is technically a behavior change for anyone who sets components.<component>=false and relies on the stray PodMonitor/HPA being present — unlikely, but may be worth a line in the release notes.

• Make sure that the change passes the CI checks.