Currently atenet-dns has no metrics, no tracing, and no exporter and the only evidence it is working is an unwatched log line.
No metrics, no tracing, no exporter. The binary never calls internal/serverboot; there is noStartMetricsServer, no meter provider, no tracer provider. manifests/ate-install/atenet-dns.yamlhas no OTEL_* environment variables and no prometheus.io/scrape annotation — consistent, since there is nothing to export or scrape.
Logs carry no trace ID. cmd/atenet/internal/dns.go:52-63 installs its own handler:
slog.SetDefault(slog.New(slog.NewJSONHandler(os.Stdout, &slog.HandlerOptions{Level: level})))This bypasses contextlogging.NewHandler, which is what stamps ate.dev/trace-id onto records (internal/contextlogging/contextlogging.go:37-44). Note the reason it was bypassed:serverboot.InitLogger hardcodes nil handler options and so offers no level control, while this command supports a --log-level flag. That shared root cause is worth fixing in serverboot rather than per-binary.
Suggested minimum instrument set
The point is detecting a wedged or silently-skipping loop:
- reconcile attempts and failures (counter, split by outcome)
- seconds since last successful reconcile (gauge) — the single most useful signal here, because the failure mode is "quietly does nothing," not "crashes"
- a counter for the skip branches, labelled by reason (
router-service-missing,no-clusterip, kube-dns-configmap-missing), so a stuck precondition is visible
- CoreDNS reload signal successes/failures (
dns.go:205-219)
Bounded label sets throughout — the skip reasons are a small fixed enum.
Currently atenet-dns has no metrics, no tracing, and no exporter and the only evidence it is working is an unwatched log line.
No metrics, no tracing, no exporter. The binary never calls
internal/serverboot; there is noStartMetricsServer, no meter provider, no tracer provider.manifests/ate-install/atenet-dns.yamlhas noOTEL_*environment variables and noprometheus.io/scrapeannotation — consistent, since there is nothing to export or scrape.Logs carry no trace ID.
cmd/atenet/internal/dns.go:52-63installs its own handler:This bypasses
contextlogging.NewHandler, which is what stampsate.dev/trace-idonto records (internal/contextlogging/contextlogging.go:37-44). Note the reason it was bypassed:serverboot.InitLoggerhardcodesnilhandler options and so offers no level control, while this command supports a--log-levelflag. That shared root cause is worth fixing inserverbootrather than per-binary.Suggested minimum instrument set
The point is detecting a wedged or silently-skipping loop:
router-service-missing,no-clusterip,kube-dns-configmap-missing), so a stuck precondition is visibledns.go:205-219)Bounded label sets throughout — the skip reasons are a small fixed enum.