add opptimistic sleep on corrosion service start to wait for schema initialisation

internal/machine/corroservice/service.go

@@ -1,7 +1,9 @@
 package corroservice
 
+import "context"
+
 type Service interface {
-	Start() error
+	Start(ctx context.Context) error
-	Restart() error
+	Restart(ctx context.Context) error
 	Running() bool
 }

internal/machine/corroservice/systemd.go

@@ -1,9 +1,11 @@
 package corroservice
 
 import (
+	"context"
 	"fmt"
 	"log/slog"
 	"os/exec"
+	"time"
 )
 
 const DefaultSystemdUnit = "uncloud-corrosion.service"
@@ -21,23 +23,30 @@ func DefaultSystemdService(dataDir string) *SystemdService {
 	}
 }
 
-func (s *SystemdService) Start() error {
+func (s *SystemdService) Start(ctx context.Context) error {
-	if _, err := exec.Command("systemctl", "start", s.Unit).Output(); err != nil {
+	return s.startOrRestart(ctx, "start")
-		return fmt.Errorf("systemctl start %s: %w", s.Unit, err)
 }
-	slog.Info("Corrosion systemd service started.", "unit", s.Unit)
 
-	// TODO: run a goroutine to check the status of the service and log any errors in the uncloud log.
+func (s *SystemdService) Restart(ctx context.Context) error {
-	s.running = true
+	return s.startOrRestart(ctx, "restart")
+}
+
+func (s *SystemdService) startOrRestart(ctx context.Context, cmd string) error {
+	if _, err := exec.Command("systemctl", cmd, s.Unit).Output(); err != nil {
+		return fmt.Errorf("systemctl %s %s: %w", cmd, s.Unit, err)
+	}
+	slog.Info(fmt.Sprintf("Corrosion systemd service %sed.", cmd), "unit", s.Unit)
+
+	// Optimistically wait for the corrosion service to start and initialise the database schema before proceeding.
+	timer := time.NewTimer(2 * time.Second)
+	defer timer.Stop()
+
+	select {
+	case <-timer.C:
+	case <-ctx.Done():
 		return nil
 	}
 
-func (s *SystemdService) Restart() error {
-	if _, err := exec.Command("systemctl", "restart", s.Unit).Output(); err != nil {
-		return fmt.Errorf("systemctl restart %s: %w", s.Unit, err)
-	}
-	slog.Info("Corrosion systemd service restarted.", "unit", s.Unit)
-
 	// TODO: run a goroutine to check the status of the service and log any errors in the uncloud log.
 	s.running = true
 	return nil

internal/machine/machine.go

@@ -170,7 +170,7 @@ func (m *Machine) Run(ctx context.Context) error {
 		}
 		slog.Info("Configured corrosion service.", "dir", m.config.CorrosionDir)
 
-		if err := m.config.CorrosionService.Start(); err != nil {
+		if err := m.config.CorrosionService.Start(ctx); err != nil {
 			return fmt.Errorf("start corrosion service: %w", err)
 		}
 	}

internal/machine/network.go

@@ -64,11 +64,11 @@ func (nc *networkController) Run(ctx context.Context) error {
 
 	if nc.corroService.Running() {
 		// Corrosion service was running before the WireGuard network was configured so we need to restart it.
-		if err := nc.corroService.Restart(); err != nil {
+		if err := nc.corroService.Restart(ctx); err != nil {
 			return fmt.Errorf("restart corrosion service: %w", err)
 		}
 	} else {
-		if err := nc.corroService.Start(); err != nil {
+		if err := nc.corroService.Start(ctx); err != nil {
 			return fmt.Errorf("start corrosion service: %w", err)
 		}
 	}