feat(observability): service discovery, health cards, alertmanager widget

Slice 3 of observability-service-registry (frontend). The Observability
page discovers Grafana from the service registry instead of env vars,
adds Grafana + Prometheus health cards, and ships an alertmanager
active_alerts dashboard widget.

- types: added GrafanaStatus + PrometheusStatus; added optional
  service_id/error to AlertmanagerStatus.
- api/client.ts + hooks/useObservability.ts: fetchGrafanaStatus,
  fetchPrometheusStatus, useGrafanaStatus, usePrometheusStatus.
- widgets/AlertmanagerAlertsWidget.tsx (new): presentational widget
  consuming the active_alerts summary shape (total/by_severity/alerts);
  exported from widgets/index.ts.
- integrations/registry.ts: alertmanager binding (active_alerts kind,
  30s refresh, optional severity_filter); registry.test.ts updated to
  6 service types incl alertmanager + a resolve test.
- components/ObservabilityPage.tsx: removed
  import.meta.env.VITE_GRAFANA_URL; derive GRAFANA_BASE_URL from the
  first enabled grafana service via useServiceInstances("grafana");
  added Grafana + Prometheus HealthCards (up/not-configured/unreachable)
  with QueryError retry blocks; machine dashboard shows a "No Grafana
  service configured" empty-state linking to /services when none is set.

npm run build (tsc -b + vite) clean; 0 lint errors; 72 frontend tests
pass. Reviewed fresh-context (read-only): no blockers.
This commit is contained in:
Developer
2026-06-24 08:23:23 +00:00
parent 14771ae990
commit 0c5698c903
10 changed files with 314 additions and 19 deletions
@@ -56,9 +56,7 @@ def _auth_headers(service: ServiceRecord) -> dict[str, str]:
return {"Authorization": f"Bearer {api_key}"} if api_key else {}
def _status_response(
service: ServiceRecord | None, *, version: str = "", error: str | None = None
) -> dict[str, Any]:
def _status_response(service: ServiceRecord | None, *, version: str = "", error: str | None = None) -> dict[str, Any]:
return {
"up": error is None,
"version": version or "",
@@ -232,9 +230,7 @@ def get_prometheus_status(
try:
health = requests.get(f"{base}/-/healthy", headers=headers, timeout=timeout)
health.raise_for_status()
build_info = requests.get(
f"{base}/api/v1/status/buildinfo", headers=headers, timeout=timeout
)
build_info = requests.get(f"{base}/api/v1/status/buildinfo", headers=headers, timeout=timeout)
build_info.raise_for_status()
version = build_info.json().get("data", {}).get("version", "")
except Exception:
+13 -4
View File
@@ -636,6 +636,7 @@ class TestResolveServiceRecord:
def _store(self, rows):
store = MagicMock()
store.get_service = lambda sid: next((r for r in rows if r["id"] == sid), None)
def list_filtered(service_type=None):
return [r for r in rows if r["service_type"] == (service_type or r["service_type"])]
@@ -882,7 +883,9 @@ class TestGrafanaStatus:
assert data["error"] == "no_service_configured"
def test_grafana_status_when_unreachable(self, test_client):
service = ServiceRecord(id="g1", service_type="grafana", name="Grafana", config={"base_url": "http://grafana:3000"})
service = ServiceRecord(
id="g1", service_type="grafana", name="Grafana", config={"base_url": "http://grafana:3000"}
)
with (
patch(f"{_MON}._resolve_service_record", return_value=service),
patch(f"{_MON}.requests.get", side_effect=Exception("refused")),
@@ -895,7 +898,9 @@ class TestGrafanaStatus:
assert data["name"] == "Grafana"
def test_grafana_status_returns_version(self, test_client):
service = ServiceRecord(id="g1", service_type="grafana", name="Grafana", config={"base_url": "http://grafana:3000"})
service = ServiceRecord(
id="g1", service_type="grafana", name="Grafana", config={"base_url": "http://grafana:3000"}
)
resp = MagicMock()
resp.json.return_value = {"version": "11.3.1", "database": "ok"}
resp.raise_for_status = MagicMock()
@@ -920,7 +925,9 @@ class TestPrometheusStatus:
assert data["error"] == "no_service_configured"
def test_prometheus_status_when_unreachable(self, test_client):
service = ServiceRecord(id="p1", service_type="prometheus", name="Prometheus", config={"base_url": "http://prometheus:9090"})
service = ServiceRecord(
id="p1", service_type="prometheus", name="Prometheus", config={"base_url": "http://prometheus:9090"}
)
with (
patch(f"{_MON}._resolve_service_record", return_value=service),
patch(f"{_MON}.requests.get", side_effect=Exception("refused")),
@@ -932,7 +939,9 @@ class TestPrometheusStatus:
assert data["error"] == "prometheus_unreachable"
def test_prometheus_status_returns_version(self, test_client):
service = ServiceRecord(id="p1", service_type="prometheus", name="Prometheus", config={"base_url": "http://prometheus:9090"})
service = ServiceRecord(
id="p1", service_type="prometheus", name="Prometheus", config={"base_url": "http://prometheus:9090"}
)
health = MagicMock()
health.raise_for_status = MagicMock()
build_info = MagicMock()