fix: persist service host port; demo-data scope for telemetry

Public URLs were ephemeral: every restart/redeploy rebound a random host
port, so endpoint links rotted. Persist published_port on the service and
prefer it for replica 0 on create. If the port is taken, fall back to
ephemeral and repersist the new assignment. Post-start the binding is
verified — Docker Desktop accepts a conflicting publish silently, which
previously left services running with no endpoint at all.

demo-data.ts gains the required scope field for HostMonitoring (CI fix).

Generated with [Devin](https://devin.ai)

Co-Authored-By: Devin <158243242+devin-ai-integration[bot]@users.noreply.github.com>

app/backend/internal/api/deployments.go

@@ -356,6 +356,12 @@ func runDeploymentAndSyncWithImage(
 		replicas = 1
 	}
 
+	var publishedPort int32
+	// Best effort: reuse the host port from the last live deployment so public
+	// URLs survive redeploys. Column exists post-migration; older DBs get
+	// ephemeral ports.
+	_ = db.QueryRow(`SELECT published_port FROM services WHERE id = $1`, service.ID).Scan(&publishedPort)
+
 	deployReq := &deployment.DeploymentRequest{
 		ProjectID:   service.ProjectID.String(),
 		ServiceID:   service.ID.String(),
@@ -367,6 +373,7 @@ func runDeploymentAndSyncWithImage(
 			Environment:   env,
 			Replicas:      replicas,
 			PublicPort:    int32(service.Port),
+			PublishedPort: publishedPort,
 			Domain:        service.Domain,
 			HealthPath:    service.HealthCheckPath,
 			RestartPolicy: service.RestartPolicy,

app/backend/internal/api/runtime.go

@@ -36,6 +36,10 @@ func serviceRuntimeSpec(db *database.DB, service Service) (deployment.RuntimeSpe
 		HealthPath:    service.HealthCheckPath,
 		RestartPolicy: service.RestartPolicy,
 	}
+	// Best effort: reuse the host port from the last live deployment so the
+	// public URL survives restarts. Column exists post-migration; older DBs
+	// silently get ephemeral ports.
+	_ = db.QueryRow(`SELECT published_port FROM services WHERE id = $1`, service.ID).Scan(&spec.PublishedPort)
 	if cmd := strings.TrimSpace(service.Command); cmd != "" {
 		spec.Command = strings.Fields(cmd)
 	}
@@ -284,6 +288,7 @@ func handleGetServiceRuntime(c *gin.Context) {
 		return
 	}
 	engine.ProbeServiceHealth(c.Request.Context(), state, service.ProjectID.String(), service.Name, service.Port, service.HealthCheckPath)
+	persistPublishedPort(db, service.ID, state.Ports)
 	if host := requestHostname(c); host != "" {
 		for i, u := range state.URLs {
 			state.URLs[i] = rewriteLoopbackURL(u, host)
@@ -308,6 +313,7 @@ func reconcileNow(c *gin.Context, engine *deployment.DeploymentEngine, db *datab
 	status := "stopped"
 	if state != nil {
 		status = state.Status
+		persistPublishedPort(db, service.ID, state.Ports)
 	}
 	_, _ = db.Exec(`UPDATE services SET status = $1, updated_at = $2 WHERE id = $3`, status, time.Now(), service.ID)
 	c.JSON(http.StatusOK, gin.H{"runtime": state})
@@ -403,6 +409,18 @@ func handleServiceRedeploy(c *gin.Context) {
 	reconcileNow(c, engine, db, service)
 }
 
+// persistPublishedPort records the host port Docker actually bound so later
+// redeploys can reuse it. Idempotent — only writes when the value changed.
+func persistPublishedPort(db *database.DB, serviceID uuid.UUID, ports []uint16) {
+	if len(ports) == 0 {
+		return
+	}
+	_, _ = db.Exec(
+		`UPDATE services SET published_port = $1 WHERE id = $2 AND published_port <> $1`,
+		int(ports[0]), serviceID,
+	)
+}
+
 // liveServiceStatus reconciles the stored status with real container state.
 // No-op when Docker is unavailable — the stored status is returned as-is.
 func liveServiceStatus(c *gin.Context, db *database.DB, service *Service) {
@@ -421,6 +439,7 @@ func liveServiceStatus(c *gin.Context, db *database.DB, service *Service) {
 		service.Status = state.Status
 		_, _ = db.Exec(`UPDATE services SET status = $1, updated_at = $2 WHERE id = $3`, state.Status, time.Now(), service.ID)
 	}
+	persistPublishedPort(db, service.ID, state.Ports)
 	if service.Domain != "" {
 		service.PublicURL = "https://" + service.Domain
 	} else if len(state.URLs) > 0 {
@@ -481,9 +500,13 @@ func runtimeScaleService(c *gin.Context, serviceID string, replicas int) error {
 
 	ctx, cancel := context.WithTimeout(context.Background(), 2*time.Minute)
 	defer cancel()
-	if _, err := engine.ReconcileService(ctx, spec); err != nil {
+	state, err := engine.ReconcileService(ctx, spec)
+	if err != nil {
 		return err
 	}
+	if state != nil {
+		persistPublishedPort(db, service.ID, state.Ports)
+	}
 	_, _ = db.Exec(`UPDATE services SET replicas = $1, updated_at = $2 WHERE id = $3`, replicas, time.Now(), service.ID)
 	return nil
 }

app/backend/internal/api/services.go

@@ -520,7 +520,9 @@ func handleUpdateService(c *gin.Context) {
 				}
 				ctx, cancel := context.WithTimeout(context.Background(), 2*time.Minute)
 				defer cancel()
-				_, _ = engine.ReconcileService(ctx, spec)
+				if state, err := engine.ReconcileService(ctx, spec); err == nil && state != nil {
+					persistPublishedPort(db.(*database.DB), serviceCopy.ID, state.Ports)
+				}
 			}()
 		}
 	}

app/backend/internal/deployment/engine.go

@@ -59,6 +59,9 @@ type ServiceConfig struct {
 	Domain string `json:"domain,omitempty"`
 	// HealthPath is an HTTP path probed on PublicPort for container health.
 	HealthPath string `json:"health_path,omitempty"`
+	// PublishedPort is the host port bound by the previous deployment; replica 0
+	// reuses it so public URLs survive restarts. 0 = pick ephemeral.
+	PublishedPort int32 `json:"published_port,omitempty"`
 }
 
 type PortMapping struct {
@@ -315,6 +318,7 @@ func (de *DeploymentEngine) deployService(ctx context.Context, deployment *Deplo
 		Env:           cfg.Environment,
 		Replicas:      cfg.Replicas,
 		Port:          cfg.PublicPort,
+		PublishedPort: cfg.PublishedPort,
 		Domain:        cfg.Domain,
 		HealthPath:    cfg.HealthPath,
 		RestartPolicy: cfg.RestartPolicy,

app/backend/internal/deployment/runtime.go

@@ -5,9 +5,11 @@ import (
 	"crypto/sha256"
 	"encoding/hex"
 	"encoding/json"
+	"errors"
 	"fmt"
 	"net/http"
 	"sort"
+	"strconv"
 	"strings"
 	"time"
 
@@ -39,6 +41,7 @@ type RuntimeSpec struct {
 	Env           map[string]string // already resolved
 	Replicas      int
 	Port          int32  // container port to expose publicly; 0 = none
+	PublishedPort int32  // preferred host port for replica 0; 0 = pick ephemeral
 	Domain        string // public hostname routed via Traefik; empty = none
 	HealthPath    string // http path probed on Port for container healthcheck
 	RestartPolicy string
@@ -240,22 +243,70 @@ func (de *DeploymentEngine) createReplica(ctx context.Context, spec RuntimeSpec,
 		Networks:      endpoints,
 	}
 
+	var wantPort nat.Port
 	if spec.Port > 0 {
-		port := nat.Port(fmt.Sprintf("%d/tcp", spec.Port))
+		wantPort = nat.Port(fmt.Sprintf("%d/tcp", spec.Port))
-		cfg.ExposedPorts = nat.PortSet{port: struct{}{}}
+		cfg.ExposedPorts = nat.PortSet{wantPort: struct{}{}}
-		// Always publish an ephemeral host port for direct access; a configured
-		// domain is additionally routed via Traefik when the edge network exists.
-		cfg.PortBindings = nat.PortMap{port: []nat.PortBinding{{HostIP: "0.0.0.0", HostPort: ""}}}
 	}
 
+	// tryBind creates and starts a container publishing wantPort on hostPort
+	// ("" = ephemeral). A requested port is verified after start: Docker
+	// Desktop accepts a conflicting binding at create/start but publishes
+	// nothing, which would leave the service running with no endpoint.
+	tryBind := func(hostPort string) (string, error) {
+		if spec.Port > 0 {
+			cfg.PortBindings = nat.PortMap{wantPort: []nat.PortBinding{{HostIP: "0.0.0.0", HostPort: hostPort}}}
+		}
 		id, err := de.dockerClient.CreateContainer(ctx, cfg)
 		if err != nil {
 			return "", err
 		}
 		if err := de.dockerClient.StartContainer(ctx, id); err != nil {
+			_ = de.dockerClient.RemoveContainer(ctx, id, true)
 			return "", fmt.Errorf("start: %w", err)
 		}
+		if hostPort != "" {
+			info, err := de.dockerClient.GetContainer(ctx, id)
+			if err != nil {
+				_ = de.dockerClient.RemoveContainer(ctx, id, true)
+				return "", fmt.Errorf("inspect: %w", err)
+			}
+			if info.NetworkSettings == nil || len(info.NetworkSettings.Ports[wantPort]) == 0 {
+				_ = de.dockerClient.RemoveContainer(ctx, id, true)
+				return "", errPortNotBound
+			}
+		}
 		return id, nil
+	}
+
+	preferred := ""
+	if index == 0 && spec.PublishedPort > 0 {
+		preferred = strconv.Itoa(int(spec.PublishedPort))
+	}
+	id, err := tryBind(preferred)
+	if err != nil && preferred != "" && (isPortConflict(err) || errors.Is(err, errPortNotBound)) {
+		// Stored port got taken since last deploy — fall back to ephemeral; the
+		// new assignment is persisted on the next runtime read.
+		id, err = tryBind("")
+	}
+	if err != nil {
+		return "", err
+	}
+	return id, nil
+}
+
+// errPortNotBound marks a container that started but failed to publish the
+// requested host port (seen on Docker Desktop, which reports success while
+// vpnkit drops a conflicting binding).
+var errPortNotBound = errors.New("requested host port not bound")
+
+// isPortConflict reports whether a container create failed because the
+// requested host port was already bound.
+func isPortConflict(err error) bool {
+	msg := strings.ToLower(err.Error())
+	return strings.Contains(msg, "port is already allocated") ||
+		strings.Contains(msg, "address already in use") ||
+		strings.Contains(msg, "cannot assign requested address")
 }
 
 // RuntimeState reads live container state for a service.

app/backend/migrations_goose/20261003120000_service_published_port.sql

@@ -0,0 +1,8 @@
+-- +goose Up
+-- Stable public host port for service replica 0. 0 means "not yet assigned";
+-- the runtime writes the ephemeral port it picked back here so restarts and
+-- redeploys keep the same public endpoint instead of churning.
+ALTER TABLE public.services ADD COLUMN IF NOT EXISTS published_port integer NOT NULL DEFAULT 0;
+
+-- +goose Down
+ALTER TABLE public.services DROP COLUMN IF EXISTS published_port;

app/frontend/src/lib/demo-data.ts

@@ -456,6 +456,7 @@ export const demoHostMonitoring: HostMonitoring = {
   uptimeSeconds: 1_828_800,
   dockerAvailable: true,
   docker: { containers: 11, images: 34, driver: 'overlay2', server: '27.3.1' },
+  scope: 'host',
   collectedAt: new Date().toISOString(),
 };