[jtslear]

Adds global.telemetry.metrics.remoteWriteURL. When set, Alloy's remote_write sends there instead of the bundled Prometheus. When empty (the default), nothing changes: compiled/resources.yaml is byte-identical.

Previously the URL was built from the prometheus subchart's service name, so the chart failed to render with prometheus.enabled: false (no template "prometheus.server.fullname"). The lookup now only happens when remoteWriteURL is empty. I checked a render with prometheus.enabled: false and remoteWriteURL set: it succeeds and writes to the given URL.

Use case: workspace clusters run only the collector (Alloy, kube-state-metrics, node-exporter) and remote-write to the Prometheus in the primary cluster. Disabling just the Prometheus server also needs #101, which makes kube-state-metrics and node-exporter independent of the prometheus subchart.

• The value is trimmed, and the chart fails with a clear message if prometheus.enabled is false and no URL is set.
• Caveats now documented in the values comment, README and CHANGELOG: Loki's ruler and Grafana's datasource still target the bundled Prometheus when it is disabled, the receiver needs --web.enable-remote-write-receiver (and native histograms if enabled), each cluster needs a distinct externalLabels, and there is no auth or TLS option (restrict at the network level).
• Ships together with refactor: make kube-state-metrics and node-exporter top-level subcharts #101; either alone does not give a collector-only install.

🤖 Generated with Claude Code