Remove run node UI and expire registrations

platform/repo/resources.go

@@ -60,6 +60,7 @@ type RunEndpointRepository interface {
 	Get(id string) (domain.RunEndpoint, error)
 	List(domain.RunEndpointFilter) ([]domain.RunEndpoint, error)
 	Update(domain.RunEndpoint) error
+	Delete(id string) error
 }
 
 type JobRepository interface {

platform/service/auth_sessions_test.go

@@ -168,7 +168,7 @@ func TestRunSessionPersistsAndSignedEnvelopeRejectsReplay(t *testing.T) {
 	if err != nil {
 		t.Fatalf("reload store: %v", err)
 	}
-	reloaded := newCoreService(reloadedStore, func() time.Time { return now.Add(time.Minute) })
+	reloaded := newCoreService(reloadedStore, func() time.Time { return now.Add(runHeartbeatStaleAfter / 2) })
 	result, err := reloaded.AcceptRunHeartbeat(domain.RunControlHeartbeat{
 		RunEndpointID: "run-local", SessionToken: hello.SessionToken, Version: "0.1.1",
 		Status: domain.RunEndpointStatusOnline, CapabilityFingerprint: "cap-jobs",

platform/service/control.go

@@ -73,6 +73,9 @@ func (svc *CoreService) RegisterRunHello(hello domain.RunControlHello) (domain.R
 	svc.controlMu.Lock()
 	defer svc.controlMu.Unlock()
 
+	if err := svc.sweepExpiredRunRegistrationsLocked(stamp); err != nil {
+		return domain.RunControlHelloResult{}, err
+	}
 	if err := svc.upsertRunEndpoint(endpoint); err != nil {
 		return domain.RunControlHelloResult{}, err
 	}
@@ -183,6 +186,9 @@ func (svc *CoreService) AcceptRunHeartbeat(heartbeat domain.RunControlHeartbeat)
 	svc.controlMu.Lock()
 	defer svc.controlMu.Unlock()
 
+	if err := svc.sweepExpiredRunRegistrationsLocked(stamp); err != nil {
+		return domain.RunControlHeartbeatResult{}, err
+	}
 	session, err := svc.currentRunSession(heartbeat.RunEndpointID, heartbeat.SessionToken)
 	if err != nil {
 		return domain.RunControlHeartbeatResult{}, err
@@ -230,6 +236,43 @@ func (svc *CoreService) upsertRunEndpoint(endpoint domain.RunEndpoint) error {
 	return svc.store.RunEndpoints().Update(endpoint)
 }
 
+func (svc *CoreService) sweepExpiredRunRegistrationsLocked(stamp time.Time) error {
+	endpoints, err := svc.store.RunEndpoints().List(domain.RunEndpointFilter{})
+	if err != nil {
+		return err
+	}
+	for _, endpoint := range endpoints {
+		if runEndpointRegistrationCurrentAt(endpoint, stamp) {
+			continue
+		}
+		if err := svc.store.RunEndpoints().Delete(endpoint.ID); err != nil && !errors.Is(err, repo.ErrNotFound) {
+			return err
+		}
+		delete(svc.runSessions, endpoint.ID)
+		session, err := svc.store.RunControlSessions().Get(endpoint.ID)
+		if err != nil {
+			if errors.Is(err, repo.ErrNotFound) {
+				continue
+			}
+			return err
+		}
+		if session.Status != domain.AuthSessionStatusActive {
+			continue
+		}
+		session.Status = domain.AuthSessionStatusRevoked
+		session.RevokedAt = stamp
+		session.UpdatedAt = stamp
+		if err := svc.store.RunControlSessions().Update(session); err != nil {
+			return err
+		}
+	}
+	return nil
+}
+
+func runEndpointRegistrationCurrentAt(endpoint domain.RunEndpoint, stamp time.Time) bool {
+	return !endpoint.LastHeartbeatAt.IsZero() && !stamp.After(endpoint.LastHeartbeatAt.Add(runHeartbeatStaleAfter))
+}
+
 func (svc *CoreService) nextSessionToken() (string, error) {
 	return randomToken()
 }

platform/service/control_test.go

@@ -118,6 +118,40 @@ func TestCoreServiceAcceptsRunHeartbeat(t *testing.T) {
 	}
 }
 
+func TestCoreServiceClearsRunRegistrationAfterHeartbeatTTL(t *testing.T) {
+	svc := newTestCoreService()
+	hello, err := svc.RegisterRunHello(validRunControlHello())
+	if err != nil {
+		t.Fatalf("register hello: %v", err)
+	}
+
+	svc.now = func() time.Time { return fixedTime.Add(runHeartbeatStaleAfter - time.Nanosecond) }
+	active, err := svc.ListRunEndpoints(domain.RunEndpointFilter{Status: domain.RunEndpointStatusOnline})
+	if err != nil || len(active) != 1 || active[0].ID != "run-local" {
+		t.Fatalf("expected registration inside TTL, endpoints=%+v err=%v", active, err)
+	}
+
+	svc.now = func() time.Time { return fixedTime.Add(runHeartbeatStaleAfter + time.Second) }
+	if _, err := svc.GetRunEndpoint("run-local"); !errors.Is(err, repo.ErrNotFound) {
+		t.Fatalf("expected expired registration to be cleared, got %v", err)
+	}
+	cleared, err := svc.ListRunEndpoints(domain.RunEndpointFilter{})
+	if err != nil || len(cleared) != 0 {
+		t.Fatalf("expected no stale registrations in list, endpoints=%+v err=%v", cleared, err)
+	}
+	_, err = svc.AcceptRunHeartbeat(domain.RunControlHeartbeat{
+		RunEndpointID:         "run-local",
+		SessionToken:          hello.SessionToken,
+		Version:               "0.1.1",
+		Status:                domain.RunEndpointStatusOnline,
+		CapabilityFingerprint: "cap-v1",
+		Capacity:              domain.RunCapacity{MaxJobs: 4},
+	})
+	if err == nil || !strings.Contains(err.Error(), "sessionToken is invalid") {
+		t.Fatalf("expected stale session to be rejected after registration cleanup, got %v", err)
+	}
+}
+
 func TestCoreServiceRejectsInvalidRunHeartbeatToken(t *testing.T) {
 	svc := newTestCoreService()
 	if _, err := svc.RegisterRunHello(validRunControlHello()); err != nil {

platform/service/resources.go

@@ -25,7 +25,7 @@ var (
 
 const (
 	ServerDeletionForceConfirmation = "FORCE DELETE"
-	runHeartbeatStaleAfter          = 2 * time.Minute
+	runHeartbeatStaleAfter          = 30 * time.Second
 )
 
 type ForbiddenError struct {
@@ -1609,10 +1609,20 @@ func (svc *CoreService) CreateRunEndpoint(endpoint domain.RunEndpoint) (domain.R
 }
 
 func (svc *CoreService) GetRunEndpoint(id string) (domain.RunEndpoint, error) {
+	svc.controlMu.Lock()
+	defer svc.controlMu.Unlock()
+	if err := svc.sweepExpiredRunRegistrationsLocked(svc.now()); err != nil {
+		return domain.RunEndpoint{}, err
+	}
 	return svc.store.RunEndpoints().Get(id)
 }
 
 func (svc *CoreService) ListRunEndpoints(filter domain.RunEndpointFilter) ([]domain.RunEndpoint, error) {
+	svc.controlMu.Lock()
+	defer svc.controlMu.Unlock()
+	if err := svc.sweepExpiredRunRegistrationsLocked(svc.now()); err != nil {
+		return nil, err
+	}
 	return svc.store.RunEndpoints().List(filter)
 }
 
@@ -2645,10 +2655,7 @@ func (svc *CoreService) validateRunnableEndpoint(endpoint domain.RunEndpoint, ca
 }
 
 func (svc *CoreService) runEndpointHeartbeatCurrent(endpoint domain.RunEndpoint) bool {
-	if endpoint.LastHeartbeatAt.IsZero() {
-		return false
-	}
-	return !svc.now().After(endpoint.LastHeartbeatAt.Add(runHeartbeatStaleAfter))
+	return runEndpointRegistrationCurrentAt(endpoint, svc.now())
 }
 
 func maxInt(a, b int) int {