chore(e2e): increase ucind cluster init timeouts, decrease max backoff delay for TCP connection to retry faster

internal/ucind/cluster.go

@@ -125,7 +125,7 @@ func (p *Provisioner) initCluster(ctx context.Context, machines []Machine) error
 	}
 	defer initClient.Close()
 
-	if err := initClient.WaitMachineReady(ctx, 30*time.Second); err != nil {
+	if err := initClient.WaitMachineReady(ctx, 90*time.Second); err != nil {
 		return fmt.Errorf("wait for machine %q to be ready: %w", initMachine.Name, err)
 	}
 
@@ -140,7 +140,7 @@ func (p *Provisioner) initCluster(ctx context.Context, machines []Machine) error
 
 	fmt.Printf("Cluster %q initialised with machine %q\n", initMachine.ClusterName, initResp.Machine.Name)
 	fmt.Printf("Waiting for cluster to be ready...")
-	if err = initClient.WaitClusterReady(ctx, 30*time.Second); err != nil {
+	if err = initClient.WaitClusterReady(ctx, 90*time.Second); err != nil {
 		return fmt.Errorf("wait for cluster to be ready: %w", err)
 	}
 	fmt.Println(" done.")
@@ -160,7 +160,7 @@ func (p *Provisioner) initCluster(ctx context.Context, machines []Machine) error
 		//goland:noinspection GoDeferInLoop
 		defer cli.Close()
 
-		if err := cli.WaitMachineReady(ctx, 30*time.Second); err != nil {
+		if err := cli.WaitMachineReady(ctx, 90*time.Second); err != nil {
 			return fmt.Errorf("wait for machine %q to be ready: %w", m.Name, err)
 		}
 

pkg/client/connector/tcp.go

@@ -4,10 +4,12 @@ import (
 	"context"
 	"fmt"
 	"net/netip"
+	"time"
 
 	"github.com/psviderski/uncloud/internal/grpcversion"
 	"golang.org/x/net/proxy"
 	"google.golang.org/grpc"
+	"google.golang.org/grpc/backoff"
 	"google.golang.org/grpc/credentials/insecure"
 )
 
@@ -21,10 +23,19 @@ func NewTCPConnector(apiAddr netip.AddrPort) *TCPConnector {
 }
 
 func (c *TCPConnector) Connect(_ context.Context) (*grpc.ClientConn, error) {
+	// Use a faster connection backoff than the default (which grows up to 120s). TCP connections are typically to
+	// local Docker containers (ucind) or LAN machines where long backoff delays are unnecessary.
+	backoffConfig := backoff.DefaultConfig
+	backoffConfig.MaxDelay = 5 * time.Second
+
 	conn, err := grpc.NewClient(
 		c.apiAddr.String(),
 		grpc.WithTransportCredentials(insecure.NewCredentials()),
 		grpc.WithDefaultServiceConfig(defaultServiceConfig),
+		grpc.WithConnectParams(grpc.ConnectParams{
+			Backoff:           backoffConfig,
+			MinConnectTimeout: 5 * time.Second,
+		}),
 		grpc.WithUnaryInterceptor(grpcversion.ClientUnaryInterceptor),
 		grpc.WithStreamInterceptor(grpcversion.ClientStreamInterceptor),
 	)

scripts/docker/entrypoint.sh

@@ -25,7 +25,7 @@ trap cleanup INT TERM EXIT
 
 dind dockerd &
 echo "Waiting for Docker in Docker to be ready..."
-timeout 5s sh -c "until docker info &> /dev/null; do sleep 0.1; done"
+timeout 60s sh -c "until docker info &> /dev/null; do sleep 0.5; done"
 echo "Docker in Docker is ready."
 
 echo "Loading corrosion image from /images/corrosion.tar..."

test/e2e/cluster_test.go

@@ -55,7 +55,7 @@ func createTestCluster(
 	})
 
 	if waitReady {
-		require.NoError(t, p.WaitClusterReady(ctx, c, 60*time.Second))
+		require.NoError(t, p.WaitClusterReady(ctx, c, 90*time.Second))
 	}
 
 	return c, p