feat: wire ARI (RFC 9702) into renewal scheduler

CheckExpiringCertificates() now queries each issuer's ARI endpoint before creating renewal jobs. If the CA says "not yet" (suggested window hasn't opened), renewal is deferred. ARI errors fall back gracefully to threshold-based logic. Audit trail records renewal_trigger=ari when ARI drives the decision. 4 new unit tests: ShouldRenewNow, NotYet, NilFallback, ErrorFallback. 3 new smoke tests in testing-guide.md Part 35. Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>
2026-06-10 06:49:00 +00:00 · 2026-03-30 12:00:22 -04:00
parent d265087661
commit 2caf03c543
4 changed files with 387 additions and 16 deletions
@@ -163,10 +163,39 @@ func (s *RenewalService) CheckExpiringCertificates(ctx context.Context) error {
 		s.sendThresholdAlerts(ctx, cert, int(daysUntil), thresholds)

 		// Only create renewal job if an issuer connector is registered for this cert's issuer
-		if _, hasIssuer := s.issuerRegistry[cert.IssuerID]; !hasIssuer {
+		connector, hasIssuer := s.issuerRegistry[cert.IssuerID]
+		if !hasIssuer {
 			continue
 		}

+		// ARI check (RFC 9702): if the issuer supports ARI, let the CA direct renewal timing.
+		// Fetch the latest cert version to get the PEM chain for the ARI query.
+		ariChecked := false
+		if version, vErr := s.certRepo.GetLatestVersion(ctx, cert.ID); vErr == nil && version != nil && version.PEMChain != "" {
+			if ariResult, ariErr := connector.GetRenewalInfo(ctx, version.PEMChain); ariErr != nil {
+				// ARI error is non-fatal — log and fall through to threshold-based renewal
+				slog.Warn("ARI check failed, falling back to threshold-based renewal",
+					"cert_id", cert.ID, "issuer_id", cert.IssuerID, "error", ariErr)
+			} else if ariResult != nil {
+				ariChecked = true
+				now := time.Now()
+				if now.Before(ariResult.SuggestedWindowStart) {
+					// CA says it's too early to renew — skip this cert
+					slog.Debug("ARI: renewal not yet suggested by CA",
+						"cert_id", cert.ID,
+						"suggested_start", ariResult.SuggestedWindowStart,
+						"suggested_end", ariResult.SuggestedWindowEnd)
+					continue
+				}
+				slog.Info("ARI: CA suggests renewal now",
+					"cert_id", cert.ID,
+					"suggested_start", ariResult.SuggestedWindowStart,
+					"suggested_end", ariResult.SuggestedWindowEnd)
+			}
+			// ariResult == nil means issuer doesn't support ARI — fall through to threshold logic
+		}
+		_ = ariChecked // used for audit metadata below
+
 		// Check for existing pending/running renewal jobs to avoid duplicates
 		existingJobs, err := s.jobRepo.ListByCertificate(ctx, cert.ID)
 		if err == nil {
@@ -206,9 +235,12 @@ func (s *RenewalService) CheckExpiringCertificates(ctx context.Context) error {
 		}

 		// Record audit event
+		auditMeta := map[string]interface{}{"days_until_expiry": daysUntil, "job_id": job.ID}
+		if ariChecked {
+			auditMeta["renewal_trigger"] = "ari"
+		}
 		if auditErr := s.auditService.RecordEvent(ctx, "system", domain.ActorTypeSystem,
-			"renewal_job_created", "certificate", cert.ID,
-			map[string]interface{}{"days_until_expiry": daysUntil, "job_id": job.ID}); auditErr != nil {
+			"renewal_job_created", "certificate", cert.ID, auditMeta); auditErr != nil {
 			slog.Error("failed to record audit event", "error", auditErr)
 		}
 	}
@@ -863,4 +863,283 @@ func TestProcessRenewalJob_NoCertificate(t *testing.T) {
 	}
 }

+// --- ARI (RFC 9702) Scheduler Integration Tests ---
+
+func TestCheckExpiringCertificates_ARI_ShouldRenewNow(t *testing.T) {
+	t.Helper()
+	ctx := context.Background()
+
+	certRepo := newMockCertificateRepository()
+	jobRepo := newMockJobRepository()
+	policyRepo := newMockRenewalPolicyRepository()
+	auditRepo := newMockAuditRepository()
+	notifRepo := newMockNotificationRepository()
+
+	auditSvc := NewAuditService(auditRepo)
+	notifSvc := NewNotificationService(notifRepo, map[string]Notifier{})
+
+	// ARI says renew now: window started in the past
+	ariConnector := &mockIssuerConnector{
+		getRenewalInfoResult: &RenewalInfoResult{
+			SuggestedWindowStart: time.Now().Add(-24 * time.Hour),
+			SuggestedWindowEnd:   time.Now().Add(48 * time.Hour),
+		},
+	}
+	issuerRegistry := map[string]IssuerConnector{
+		"iss-acme": ariConnector,
+	}
+
+	svc := NewRenewalService(certRepo, jobRepo, policyRepo, nil, auditSvc, notifSvc, issuerRegistry, "server")
+
+	// Create cert expiring in 20 days with a cert version (needed for ARI lookup)
+	cert := &domain.ManagedCertificate{
+		ID:              "mc-ari-renew",
+		Name:            "ARI Cert",
+		CommonName:      "ari.example.com",
+		SANs:            []string{},
+		OwnerID:         "owner-1",
+		TeamID:          "team-1",
+		IssuerID:        "iss-acme",
+		RenewalPolicyID: "rp-standard",
+		Status:          domain.CertificateStatusActive,
+		ExpiresAt:       time.Now().AddDate(0, 0, 20),
+		Tags:            make(map[string]string),
+		CreatedAt:       time.Now(),
+		UpdatedAt:       time.Now(),
+	}
+	certRepo.AddCert(cert)
+	certRepo.Versions[cert.ID] = []*domain.CertificateVersion{
+		{ID: "cv-1", CertificateID: cert.ID, PEMChain: "-----BEGIN CERTIFICATE-----\ntest\n-----END CERTIFICATE-----"},
+	}
+
+	policy := &domain.RenewalPolicy{
+		ID: "rp-standard", Name: "Standard", RenewalWindowDays: 30,
+		AutoRenew: true, MaxRetries: 3, RetryInterval: 300,
+		AlertThresholdsDays: []int{30, 14, 7, 0},
+		CreatedAt: time.Now(), UpdatedAt: time.Now(),
+	}
+	policyRepo.AddPolicy(policy)
+
+	err := svc.CheckExpiringCertificates(ctx)
+	if err != nil {
+		t.Fatalf("CheckExpiringCertificates failed: %v", err)
+	}
+
+	// ARI says renew now, so a renewal job should be created
+	hasRenewalJob := false
+	for _, job := range jobRepo.Jobs {
+		if job.Type == domain.JobTypeRenewal {
+			hasRenewalJob = true
+			break
+		}
+	}
+	if !hasRenewalJob {
+		t.Errorf("expected renewal job when ARI ShouldRenewNow is true")
+	}
+}
+
+func TestCheckExpiringCertificates_ARI_NotYet(t *testing.T) {
+	t.Helper()
+	ctx := context.Background()
+
+	certRepo := newMockCertificateRepository()
+	jobRepo := newMockJobRepository()
+	policyRepo := newMockRenewalPolicyRepository()
+	auditRepo := newMockAuditRepository()
+	notifRepo := newMockNotificationRepository()
+
+	auditSvc := NewAuditService(auditRepo)
+	notifSvc := NewNotificationService(notifRepo, map[string]Notifier{})
+
+	// ARI says NOT yet: window starts in the future
+	ariConnector := &mockIssuerConnector{
+		getRenewalInfoResult: &RenewalInfoResult{
+			SuggestedWindowStart: time.Now().Add(72 * time.Hour),
+			SuggestedWindowEnd:   time.Now().Add(96 * time.Hour),
+		},
+	}
+	issuerRegistry := map[string]IssuerConnector{
+		"iss-acme": ariConnector,
+	}
+
+	svc := NewRenewalService(certRepo, jobRepo, policyRepo, nil, auditSvc, notifSvc, issuerRegistry, "server")
+
+	// Cert is within the 30-day threshold window (would normally trigger renewal),
+	// but ARI says "not yet"
+	cert := &domain.ManagedCertificate{
+		ID:              "mc-ari-wait",
+		Name:            "ARI Wait Cert",
+		CommonName:      "ari-wait.example.com",
+		SANs:            []string{},
+		OwnerID:         "owner-1",
+		TeamID:          "team-1",
+		IssuerID:        "iss-acme",
+		RenewalPolicyID: "rp-standard",
+		Status:          domain.CertificateStatusActive,
+		ExpiresAt:       time.Now().AddDate(0, 0, 10),
+		Tags:            make(map[string]string),
+		CreatedAt:       time.Now(),
+		UpdatedAt:       time.Now(),
+	}
+	certRepo.AddCert(cert)
+	certRepo.Versions[cert.ID] = []*domain.CertificateVersion{
+		{ID: "cv-2", CertificateID: cert.ID, PEMChain: "-----BEGIN CERTIFICATE-----\ntest\n-----END CERTIFICATE-----"},
+	}
+
+	policy := &domain.RenewalPolicy{
+		ID: "rp-standard", Name: "Standard", RenewalWindowDays: 30,
+		AutoRenew: true, MaxRetries: 3, RetryInterval: 300,
+		AlertThresholdsDays: []int{30, 14, 7, 0},
+		CreatedAt: time.Now(), UpdatedAt: time.Now(),
+	}
+	policyRepo.AddPolicy(policy)
+
+	err := svc.CheckExpiringCertificates(ctx)
+	if err != nil {
+		t.Fatalf("CheckExpiringCertificates failed: %v", err)
+	}
+
+	// ARI says not yet, so NO renewal job should be created
+	for _, job := range jobRepo.Jobs {
+		if job.Type == domain.JobTypeRenewal {
+			t.Errorf("expected no renewal job when ARI says not yet, but found one")
+		}
+	}
+}
+
+func TestCheckExpiringCertificates_ARI_NilResult_FallsThrough(t *testing.T) {
+	t.Helper()
+	ctx := context.Background()
+
+	certRepo := newMockCertificateRepository()
+	jobRepo := newMockJobRepository()
+	policyRepo := newMockRenewalPolicyRepository()
+	auditRepo := newMockAuditRepository()
+	notifRepo := newMockNotificationRepository()
+
+	auditSvc := NewAuditService(auditRepo)
+	notifSvc := NewNotificationService(notifRepo, map[string]Notifier{})
+
+	// ARI returns nil (issuer doesn't support ARI) — default mock behavior
+	issuerRegistry := map[string]IssuerConnector{
+		"iss-local": &mockIssuerConnector{},
+	}
+
+	svc := NewRenewalService(certRepo, jobRepo, policyRepo, nil, auditSvc, notifSvc, issuerRegistry, "server")
+
+	cert := &domain.ManagedCertificate{
+		ID:              "mc-ari-nil",
+		Name:            "No ARI Cert",
+		CommonName:      "no-ari.example.com",
+		SANs:            []string{},
+		OwnerID:         "owner-1",
+		TeamID:          "team-1",
+		IssuerID:        "iss-local",
+		RenewalPolicyID: "rp-standard",
+		Status:          domain.CertificateStatusActive,
+		ExpiresAt:       time.Now().AddDate(0, 0, 20),
+		Tags:            make(map[string]string),
+		CreatedAt:       time.Now(),
+		UpdatedAt:       time.Now(),
+	}
+	certRepo.AddCert(cert)
+	certRepo.Versions[cert.ID] = []*domain.CertificateVersion{
+		{ID: "cv-3", CertificateID: cert.ID, PEMChain: "-----BEGIN CERTIFICATE-----\ntest\n-----END CERTIFICATE-----"},
+	}
+
+	policy := &domain.RenewalPolicy{
+		ID: "rp-standard", Name: "Standard", RenewalWindowDays: 30,
+		AutoRenew: true, MaxRetries: 3, RetryInterval: 300,
+		AlertThresholdsDays: []int{30, 14, 7, 0},
+		CreatedAt: time.Now(), UpdatedAt: time.Now(),
+	}
+	policyRepo.AddPolicy(policy)
+
+	err := svc.CheckExpiringCertificates(ctx)
+	if err != nil {
+		t.Fatalf("CheckExpiringCertificates failed: %v", err)
+	}
+
+	// ARI is nil (not supported), so threshold-based logic applies; cert is within 30-day window
+	hasRenewalJob := false
+	for _, job := range jobRepo.Jobs {
+		if job.Type == domain.JobTypeRenewal {
+			hasRenewalJob = true
+			break
+		}
+	}
+	if !hasRenewalJob {
+		t.Errorf("expected renewal job via threshold fallback when ARI returns nil")
+	}
+}
+
+func TestCheckExpiringCertificates_ARI_Error_FallsThrough(t *testing.T) {
+	t.Helper()
+	ctx := context.Background()
+
+	certRepo := newMockCertificateRepository()
+	jobRepo := newMockJobRepository()
+	policyRepo := newMockRenewalPolicyRepository()
+	auditRepo := newMockAuditRepository()
+	notifRepo := newMockNotificationRepository()
+
+	auditSvc := NewAuditService(auditRepo)
+	notifSvc := NewNotificationService(notifRepo, map[string]Notifier{})
+
+	// ARI returns an error — should fall through to threshold-based renewal
+	ariConnector := &mockIssuerConnector{
+		getRenewalInfoErr: fmt.Errorf("ARI endpoint unreachable"),
+	}
+	issuerRegistry := map[string]IssuerConnector{
+		"iss-acme": ariConnector,
+	}
+
+	svc := NewRenewalService(certRepo, jobRepo, policyRepo, nil, auditSvc, notifSvc, issuerRegistry, "server")
+
+	cert := &domain.ManagedCertificate{
+		ID:              "mc-ari-err",
+		Name:            "ARI Error Cert",
+		CommonName:      "ari-err.example.com",
+		SANs:            []string{},
+		OwnerID:         "owner-1",
+		TeamID:          "team-1",
+		IssuerID:        "iss-acme",
+		RenewalPolicyID: "rp-standard",
+		Status:          domain.CertificateStatusActive,
+		ExpiresAt:       time.Now().AddDate(0, 0, 15),
+		Tags:            make(map[string]string),
+		CreatedAt:       time.Now(),
+		UpdatedAt:       time.Now(),
+	}
+	certRepo.AddCert(cert)
+	certRepo.Versions[cert.ID] = []*domain.CertificateVersion{
+		{ID: "cv-4", CertificateID: cert.ID, PEMChain: "-----BEGIN CERTIFICATE-----\ntest\n-----END CERTIFICATE-----"},
+	}
+
+	policy := &domain.RenewalPolicy{
+		ID: "rp-standard", Name: "Standard", RenewalWindowDays: 30,
+		AutoRenew: true, MaxRetries: 3, RetryInterval: 300,
+		AlertThresholdsDays: []int{30, 14, 7, 0},
+		CreatedAt: time.Now(), UpdatedAt: time.Now(),
+	}
+	policyRepo.AddPolicy(policy)
+
+	err := svc.CheckExpiringCertificates(ctx)
+	if err != nil {
+		t.Fatalf("CheckExpiringCertificates failed: %v", err)
+	}
+
+	// ARI failed but renewal should still happen via threshold fallback
+	hasRenewalJob := false
+	for _, job := range jobRepo.Jobs {
+		if job.Type == domain.JobTypeRenewal {
+			hasRenewalJob = true
+			break
+		}
+	}
+	if !hasRenewalJob {
+		t.Errorf("expected renewal job via threshold fallback when ARI errors")
+	}
+}
+
 // stringPtr is defined in notification_test.go
@@ -660,8 +660,10 @@ func (m *mockTargetRepo) AddTarget(target *domain.DeploymentTarget) {

 // mockIssuerConnector is a test implementation of IssuerConnector
 type mockIssuerConnector struct {
-	Result *IssuanceResult
-	Err    error
+	Result             *IssuanceResult
+	Err                error
+	getRenewalInfoResult *RenewalInfoResult
+	getRenewalInfoErr    error
 }

 func (m *mockIssuerConnector) IssueCertificate(ctx context.Context, commonName string, sans []string, csrPEM string, ekus []string) (*IssuanceResult, error) {
@@ -717,14 +719,14 @@ func (m *mockIssuerConnector) GetCACertPEM(ctx context.Context) (string, error)
 }

 func (m *mockIssuerConnector) GetRenewalInfo(ctx context.Context, certPEM string) (*RenewalInfoResult, error) {
-	if m.Err != nil {
-		return nil, m.Err
+	if m.getRenewalInfoErr != nil {
+		return nil, m.getRenewalInfoErr
 	}
-	now := time.Now()
-	return &RenewalInfoResult{
-		SuggestedWindowStart: now,
-		SuggestedWindowEnd:   now.Add(7 * 24 * time.Hour),
-	}, nil
+	if m.getRenewalInfoResult != nil {
+		return m.getRenewalInfoResult, nil
+	}
+	// Default: return nil, nil (issuer does not support ARI)
+	return nil, nil
 }

 // Constructor functions for mocks