collector dashboard: query the service.name the collectors emit, lint dashboard names

scripts/check-dashboard-service-names.sh

@@ -0,0 +1,67 @@
+#!/bin/sh
+# CI lint: every literal `"service.name"="<x>"` selector in a shipped grafana
+# dashboard must name a service.name that something emits. A dashboard
+# querying a name nobody sets renders empty panels with no error.
+#
+# `known` lists the emitted names, one `name|file|anchor` per line. The
+# anchor is a fixed string that must still appear in that file, so an entry
+# cannot outlive the code that sets the name. Regex selectors
+# (`"service.name"=~"$service"`) are template variables and are not checked.
+#
+set -eu
+
+dashboards_dir="nix/host-modules/swarm-grafana/dashboards"
+
+# bao: `collectorServiceName` upserted on the store's collector.
+# authelia, forgejo: `serviceName` of the journal forwarder in each module.
+# swarm-controller: the resource of the controller's own metrics.
+# otelcol-contrib: the unset default of the collector binary, which the hive
+#   collector and swarm-otel both run (`pkgs.opentelemetry-collector-contrib`)
+#   and which stamps their self-metrics.
+known='bao|nix/host-modules/swarm-bao.nix|collectorServiceName = "bao"
+authelia|nix/host-modules/swarm-authelia.nix|serviceName = "authelia"
+forgejo|nix/host-modules/hive-forge/default.nix|serviceName = "forgejo"
+swarm-controller|swarm-controller/src/vcs_metrics.rs|with_service_name("swarm-controller")
+otelcol-contrib|nix/host-modules/otel.nix|package = pkgs.opentelemetry-collector-contrib'
+
+violations=0
+
+while IFS='|' read -r name file anchor; do
+  if ! grep -qF -- "$anchor" "$file" 2>/dev/null; then
+    violations=$((violations + 1))
+    printf '::error file=%s::service.name "%s" is listed as emitted but %s no longer contains: %s\n' \
+      "$file" "$name" "$file" "$anchor"
+  fi
+done <<EOF2
+$known
+EOF2
+
+names="$(printf '%s\n' "$known" | cut -d'|' -f1)"
+checked=0
+
+for f in "$dashboards_dir"/*.json; do
+  # The expr strings escape the quotes, so the raw file holds
+  # service.name backslash-quote-equals-backslash-quote name.
+  used_names="$(grep -oE '\\"service\.name\\"=\\"[^\\]+\\"' "$f" | sed -E 's/.*=\\"([^\\]+)\\"$/\1/' | sort -u || true)"
+  while IFS= read -r used; do
+    [ -n "$used" ] || continue
+    checked=$((checked + 1))
+    if ! printf '%s\n' "$names" | grep -qxF -- "$used"; then
+      violations=$((violations + 1))
+      printf '::error file=%s::queries service.name "%s", which nothing emits (known: %s)\n' \
+        "$f" "$used" "$(printf '%s' "$names" | tr '\n' ' ')"
+    fi
+  done <<EOF3
+$used_names
+EOF3
+done
+
+if [ "$checked" -eq 0 ]; then
+  echo 'check-dashboard-service-names: found zero service.name selectors — extractor or tree changed' >&2
+  exit 1
+fi
+if [ "$violations" -gt 0 ]; then
+  printf 'check-dashboard-service-names: %s violation(s)\n' "$violations" >&2
+  exit 1
+fi
+echo "check-dashboard-service-names: $checked selector name(s) checked, clean" >&2