fix: prevent race condition with Docker network creation (#89)

internal/machine/docker/server.go

@@ -50,15 +50,42 @@ type Server struct {
 	// internalDNSIP is a function that returns the IP address of the internal DNS server. It may return an empty
 	// address if the address is unknown (e.g. when the machine is not initialised yet).
 	internalDNSIP func() netip.Addr
+	// networkReady is a function that returns true if the Docker network is ready for containers.
+	networkReady func() bool
+	// waitForNetworkReady is a function that waits for the Docker network to be ready for containers.
+	waitForNetworkReady func(ctx context.Context) error
+}
+
+// ServerOption configures the Docker server.
+type ServerOption func(*Server)
+
+// WithNetworkReady sets the network readiness check function.
+func WithNetworkReady(networkReady func() bool) ServerOption {
+	return func(s *Server) {
+		s.networkReady = networkReady
+	}
+}
+
+// WithWaitForNetworkReady sets the network readiness wait function.
+func WithWaitForNetworkReady(waitForNetworkReady func(ctx context.Context) error) ServerOption {
+	return func(s *Server) {
+		s.waitForNetworkReady = waitForNetworkReady
+	}
 }
 
 // NewServer creates a new Docker gRPC server with the provided Docker client.
-func NewServer(cli *client.Client, db *sqlx.DB, internalDNSIP func() netip.Addr) *Server {
+func NewServer(cli *client.Client, db *sqlx.DB, internalDNSIP func() netip.Addr, opts ...ServerOption) *Server {
-	return &Server{
+	s := &Server{
 		client:        cli,
 		db:            db,
 		internalDNSIP: internalDNSIP,
 	}
+	
+	for _, opt := range opts {
+		opt(s)
+	}
+	
+	return s
 }
 
 // CreateContainer creates a new container based on the given configuration.
@@ -118,6 +145,15 @@ func (s *Server) InspectContainer(ctx context.Context, req *pb.InspectContainerR
 
 // StartContainer starts a container with the given ID and options.
 func (s *Server) StartContainer(ctx context.Context, req *pb.StartContainerRequest) (*emptypb.Empty, error) {
+	// Wait for Docker network to be ready before starting the container
+	if s.waitForNetworkReady != nil {
+		if err := s.waitForNetworkReady(ctx); err != nil {
+			return nil, status.Errorf(codes.Unavailable, "Docker network not ready: %v", err)
+		}
+	} else if s.networkReady != nil && !s.networkReady() {
+		return nil, status.Errorf(codes.Unavailable, "Docker network not ready")
+	}
+
 	var opts container.StartOptions
 	if len(req.Options) > 0 {
 		if err := json.Unmarshal(req.Options, &opts); err != nil {

internal/machine/machine.go

@@ -13,6 +13,7 @@ import (
 	"path/filepath"
 	"slices"
 	"strconv"
+	"sync"
 
 	"github.com/docker/docker/client"
 	"github.com/docker/go-connections/sockets"
@@ -149,6 +150,10 @@ type Machine struct {
 	started chan struct{}
 	// initialised is signalled when the machine is configured as a member of a cluster.
 	initialised chan struct{}
+	// networkReady is signalled when the Docker network is configured and ready for containers.
+	networkReady chan struct{}
+	// networkReadyMu protects networkReady channel operations
+	networkReadyMu sync.RWMutex
 
 	// store is the cluster store backed by a distributed Corrosion database.
 	store   *store.Store
@@ -235,6 +240,7 @@ func NewMachine(config *Config) (*Machine, error) {
 		state:            state,
 		started:          make(chan struct{}),
 		initialised:      make(chan struct{}, 1),
+		networkReady:     make(chan struct{}),
 		store:            corroStore,
 		cluster:          c,
 		localProxyServer: localProxyServer,
@@ -245,11 +251,17 @@ func NewMachine(config *Config) (*Machine, error) {
 	internalDNSIP := func() netip.Addr {
 		return m.IP()
 	}
-	m.docker = machinedocker.NewServer(dockerCli, db, internalDNSIP)
+	m.docker = machinedocker.NewServer(dockerCli, db, internalDNSIP, 
+		machinedocker.WithNetworkReady(m.IsNetworkReady),
+		machinedocker.WithWaitForNetworkReady(m.WaitForNetworkReady))
 	m.localMachineServer = newGRPCServer(m, c, m.docker)
 
 	if m.Initialised() {
 		m.initialised <- struct{}{}
+	} else {
+		// For non-initialized machines, signal network is ready immediately
+		// since there's no cluster network to set up
+		close(m.networkReady)
 	}
 
 	return m, nil
@@ -362,6 +374,11 @@ func (m *Machine) Run(ctx context.Context) error {
 				case <-m.initialised:
 					var err error
 					
+					// Reset networkReady channel for the new cluster configuration
+					m.networkReadyMu.Lock()
+					m.networkReady = make(chan struct{})
+					m.networkReadyMu.Unlock()
+
 					m.cluster.UpdateMachineID(m.state.ID)
 
 					// Ensure the corrosion config is up to date, including a new gossip address if the machine
@@ -404,6 +421,7 @@ func (m *Machine) Run(ctx context.Context) error {
 						caddyfileCtrl,
 						dnsServer,
 						dnsResolver,
+						m.networkReady,
 					)
 					if err != nil {
 						return fmt.Errorf("initialise network controller: %w", err)
@@ -776,6 +794,47 @@ func (m *Machine) Inspect(_ context.Context, _ *emptypb.Empty) (*pb.MachineInfo,
 	}, nil
 }
 
+// IsNetworkReady returns true if the Docker network is ready for containers.
+func (m *Machine) IsNetworkReady() bool {
+	if !m.Initialised() {
+		// If machine is not initialized, there's no network to check
+		return true
+	}
+	
+	// Check if network is ready by checking if the networkReady channel has been closed
+	m.networkReadyMu.RLock()
+	defer m.networkReadyMu.RUnlock()
+	
+	select {
+	case <-m.networkReady:
+		return true
+	default:
+		return false
+	}
+}
+
+// WaitForNetworkReady waits for the Docker network to be ready for containers.
+// It returns nil when the network is ready or an error if the context is cancelled.
+func (m *Machine) WaitForNetworkReady(ctx context.Context) error {
+	if !m.Initialised() {
+		// If machine is not initialized, there's no network to wait for
+		return nil
+	}
+	
+	// Get a copy of the channel to wait on
+	m.networkReadyMu.RLock()
+	networkReady := m.networkReady
+	m.networkReadyMu.RUnlock()
+	
+	// Wait for network to be ready or context to be cancelled
+	select {
+	case <-networkReady:
+		return nil
+	case <-ctx.Done():
+		return ctx.Err()
+	}
+}
+
 // Reset restores the machine to a clean state, removing all cluster-related сonfiguration and data and scheduling
 // a graceful shutdown. The uncloud daemon will restart the machine if managed by systemd.
 func (m *Machine) Reset(ctx context.Context, _ *pb.ResetRequest) (*emptypb.Empty, error) {

internal/machine/network.go

@@ -44,6 +44,9 @@ type networkController struct {
 	// dnsServer is the embedded internal DNS server for the cluster listening on the machine IP.
 	dnsServer   *dns.Server
 	dnsResolver *dns.ClusterResolver
+
+	// networkReady is signalled when the Docker network is configured and ready for containers.
+	networkReady chan<- struct{}
 }
 
 func newNetworkController(
@@ -55,6 +58,7 @@ func newNetworkController(
 	caddyfileCtrl *caddyconfig.Controller,
 	dnsServer *dns.Server,
 	dnsResolver *dns.ClusterResolver,
+	networkReady chan<- struct{},
 ) (
 	*networkController, error,
 ) {
@@ -76,6 +80,7 @@ func newNetworkController(
 		caddyfileCtrl:   caddyfileCtrl,
 		dnsServer:       dnsServer,
 		dnsResolver:     dnsResolver,
+		networkReady:    networkReady,
 	}, nil
 }
 
@@ -225,6 +230,9 @@ func (nc *networkController) prepareAndWatchDocker(ctx context.Context) error {
 	}
 	slog.Info("Docker network configured.")
 
+	// Signal that the Docker network is ready for containers
+	close(nc.networkReady)
+
 	slog.Info("Watching Docker containers and syncing them to cluster store.")
 	// Retry to watch and sync containers until the context is done.
 	boff := backoff.WithContext(backoff.NewExponentialBackOff(