Add /healthz endpoint that checks Alertmanager connection

config/exporter/exporter.yaml

@@ -54,5 +54,12 @@ spec:
           requests:
             cpu: 10m
             memory: 64Mi
+        livenessProbe:
+          httpGet:
+            path: /healthz
+            port: 8080
+          periodSeconds: 20
+          initialDelaySeconds: 15
+          timeoutSeconds: 3
       serviceAccountName: alerts-exporter
       terminationGracePeriodSeconds: 10

internal/healthcheck/healthcheck.go

@@ -0,0 +1,45 @@
+package healthcheck
+
+import (
+	"encoding/json"
+	"net/http"
+
+	"github.com/go-openapi/strfmt"
+	"github.com/prometheus/alertmanager/api/v2/client/general"
+	"github.com/prometheus/alertmanager/api/v2/models"
+)
+
+// HealthCheck is a health check handler for the Alertmanager API.
+type HealthCheck struct {
+	GeneralService general.ClientService
+}
+
+// HandleHealthz handles a health check request.
+// It returns a JSON response with the status of the Alertmanager API or an error if the client returns an error or if receiving a nil response.
+func (h HealthCheck) HandleHealthz(res http.ResponseWriter, req *http.Request) {
+	ams, err := h.GeneralService.GetStatus(general.NewGetStatusParamsWithContext(req.Context()))
+	if err != nil {
+		http.Error(res, err.Error(), http.StatusInternalServerError)
+		return
+	}
+	if ams == nil || ams.Payload == nil {
+		http.Error(res, "Nil response from Alertmanager", http.StatusInternalServerError)
+		return
+	}
+	if err := json.NewEncoder(res).Encode(response{
+		Status:              "connected",
+		AlertmanagerCluster: ams.Payload.Cluster,
+		AlertmanagerVersion: ams.Payload.VersionInfo,
+		AlertmanagerUptime:  ams.Payload.Uptime,
+	}); err != nil {
+		http.Error(res, "Encoding error: "+err.Error(), http.StatusInternalServerError)
+		return
+	}
+}
+
+type response struct {
+	Status              string                `json:"status"`
+	AlertmanagerCluster *models.ClusterStatus `json:"alertmanager_cluster"`
+	AlertmanagerVersion *models.VersionInfo   `json:"alertmanager_version"`
+	AlertmanagerUptime  *strfmt.DateTime      `json:"alertmanager_uptime"`
+}

internal/healthcheck/healthcheck_test.go

@@ -0,0 +1,85 @@
+package healthcheck_test
+
+import (
+	"errors"
+	"net/http"
+	"net/http/httptest"
+	"testing"
+
+	"github.com/appuio/alerts_exporter/internal/healthcheck"
+	"github.com/go-openapi/runtime"
+	"github.com/prometheus/alertmanager/api/v2/client/general"
+	"github.com/prometheus/alertmanager/api/v2/models"
+	"github.com/stretchr/testify/require"
+)
+
+func TestOk(t *testing.T) {
+	t.Parallel()
+
+	hc := &healthcheck.HealthCheck{
+		GeneralService: &mockClientService{
+			OkResponse: &general.GetStatusOK{
+				Payload: &models.AlertmanagerStatus{
+					VersionInfo: &models.VersionInfo{
+						Version: ptr("v0.22.2"),
+					},
+				},
+			},
+		},
+	}
+
+	req := httptest.NewRecorder()
+	hc.HandleHealthz(req, httptest.NewRequest("GET", "/healthz", nil))
+	res := req.Result()
+	require.Equal(t, http.StatusOK, res.StatusCode)
+	require.Contains(t, req.Body.String(), `"status":"connected"`)
+	require.Contains(t, req.Body.String(), `v0.22.2`)
+}
+
+func TestErrResponse(t *testing.T) {
+	t.Parallel()
+
+	hc := &healthcheck.HealthCheck{
+		GeneralService: &mockClientService{
+			Err: errors.New("some error"),
+		},
+	}
+
+	req := httptest.NewRecorder()
+	hc.HandleHealthz(req, httptest.NewRequest("GET", "/healthz", nil))
+	res := req.Result()
+	require.Equal(t, http.StatusInternalServerError, res.StatusCode)
+	require.Contains(t, req.Body.String(), "some error")
+}
+
+func TestNilResponse(t *testing.T) {
+	t.Parallel()
+
+	hc := &healthcheck.HealthCheck{
+		GeneralService: &mockClientService{},
+	}
+
+	req := httptest.NewRecorder()
+	hc.HandleHealthz(req, httptest.NewRequest("GET", "/healthz", nil))
+	res := req.Result()
+	require.Equal(t, http.StatusInternalServerError, res.StatusCode)
+	require.Contains(t, req.Body.String(), "Nil response")
+}
+
+type mockClientService struct {
+	OkResponse *general.GetStatusOK
+	Err        error
+}
+
+var _ general.ClientService = (*mockClientService)(nil)
+
+func (m *mockClientService) GetStatus(*general.GetStatusParams, ...general.ClientOption) (*general.GetStatusOK, error) {
+	if m.Err != nil {
+		return nil, m.Err
+	}
+	return m.OkResponse, nil
+}
+
+func (m *mockClientService) SetTransport(runtime.ClientTransport) {}
+
+func ptr[T any](t T) *T { return &t }

internal/saauth/saauth_test.go

@@ -27,8 +27,8 @@ func Test_ServiceAccountAuthInfoWriter_AuthenticateRequest(t *testing.T) {
 	require.NoError(t, os.WriteFile(tokenFile, []byte("new-token"), 0644))
 	require.EventuallyWithT(t, func(t *assert.CollectT) {
 		r := new(runtime.TestClientRequest)
-		require.NoError(t, subject.AuthenticateRequest(r, nil))
-		require.Equal(t, "Bearer new-token", r.GetHeaderParams().Get("Authorization"))
+		assert.NoError(t, subject.AuthenticateRequest(r, nil))
+		assert.Equal(t, "Bearer new-token", r.GetHeaderParams().Get("Authorization"))
 	}, 5*time.Second, time.Millisecond)
 }
 

main.go

@@ -7,6 +7,7 @@ import (
 	"net/http"
 
 	alertscollector "github.com/appuio/alerts_exporter/internal/alerts_collector"
+	"github.com/appuio/alerts_exporter/internal/healthcheck"
 	"github.com/appuio/alerts_exporter/internal/saauth"
 	openapiclient "github.com/go-openapi/runtime/client"
 	alertmanagerclient "github.com/prometheus/alertmanager/api/v2/client"
@@ -100,6 +101,7 @@ func main() {
 	// Expose metrics and custom registry via an HTTP server
 	// using the HandleFor function. "/metrics" is the usual endpoint for that.
 	http.Handle("/metrics", promhttp.HandlerFor(reg, promhttp.HandlerOpts{Registry: reg}))
+	http.HandleFunc("/healthz", healthcheck.HealthCheck{GeneralService: ac.General}.HandleHealthz)
 	log.Printf("Listening on `%s`", listenAddr)
 	log.Fatal(http.ListenAndServe(listenAddr, nil))
 }