improved docker reconnect mechanism, removed redundant checkings, refactor

internal/docker/client.go

@@ -1,8 +1,10 @@
 package docker
 
 import (
+	"context"
 	"errors"
 	"fmt"
+	"net"
 	"net/http"
 	"sync"
 	"time"
@@ -23,11 +25,14 @@ type (
 		key      string
 		refCount uint32
 		closedOn int64
+
+		addr string
+		dial func(ctx context.Context) (net.Conn, error)
 	}
 )
 
 var (
-	clientMap   = make(map[string]*SharedClient, 5)
+	clientMap   = make(map[string]*SharedClient, 10)
 	clientMapMu sync.RWMutex
 
 	clientOptEnvHost = []client.Opt{
@@ -64,9 +69,7 @@ func init() {
 
 		for _, c := range clientMap {
 			delete(clientMap, c.key)
-			if c.Connected() {
-				c.Client.Close()
-			}
+			c.Client.Close()
 		}
 	})
 }
@@ -78,10 +81,6 @@ func closeTimedOutClients() {
 	now := time.Now().Unix()
 
 	for _, c := range clientMap {
-		if !c.Connected() {
-			delete(clientMap, c.key)
-			continue
-		}
 		if c.closedOn == 0 {
 			continue
 		}
@@ -93,8 +92,17 @@ func closeTimedOutClients() {
 	}
 }
 
-func (c *SharedClient) Connected() bool {
-	return c != nil && c.Client != nil
+func (c *SharedClient) Address() string {
+	return c.addr
+}
+
+func (c *SharedClient) CheckConnection(ctx context.Context) error {
+	conn, err := c.dial(ctx)
+	if err != nil {
+		return err
+	}
+	conn.Close()
+	return nil
 }
 
 // if the client is still referenced, this is no-op.
@@ -103,7 +111,7 @@ func (c *SharedClient) Close() {
 	c.refCount--
 }
 
-// ConnectClient creates a new Docker client connection to the specified host.
+// NewClient creates a new Docker client connection to the specified host.
 //
 // Returns existing client if available.
 //
@@ -113,7 +121,7 @@ func (c *SharedClient) Close() {
 // Returns:
 //   - Client: the Docker client connection.
 //   - error: an error if the connection failed.
-func ConnectClient(host string) (*SharedClient, error) {
+func NewClient(host string) (*SharedClient, error) {
 	clientMapMu.Lock()
 	defer clientMapMu.Unlock()
 
@@ -125,6 +133,8 @@ func ConnectClient(host string) (*SharedClient, error) {
 
 	// create client
 	var opt []client.Opt
+	var addr string
+	var dial func(ctx context.Context) (net.Conn, error)
 
 	if agent.IsDockerHostAgent(host) {
 		cfg, ok := config.GetInstance().GetAgent(host)
@@ -136,6 +146,8 @@ func ConnectClient(host string) (*SharedClient, error) {
 			client.WithHTTPClient(cfg.NewHTTPClient()),
 			client.WithAPIVersionNegotiation(),
 		}
+		addr = "tcp://" + cfg.Addr
+		dial = cfg.DialContext
 	} else {
 		switch host {
 		case "":
@@ -177,9 +189,16 @@ func ConnectClient(host string) (*SharedClient, error) {
 		Client:   client,
 		key:      host,
 		refCount: 1,
+		addr:     addr,
+		dial:     dial,
 	}
 
-	defer logging.Debug().Str("host", host).Msg("docker client connected")
+	// non-agent client
+	if c.dial == nil {
+		c.dial = client.Dialer()
+	}
+
+	defer logging.Debug().Str("host", host).Msg("docker client initialized")
 
 	clientMap[c.key] = c
 	return c, nil

internal/docker/idlewatcher/watcher.go

@@ -8,7 +8,7 @@ import (
 
 	"github.com/docker/docker/api/types/container"
 	"github.com/rs/zerolog"
-	D "github.com/yusing/go-proxy/internal/docker"
+	"github.com/yusing/go-proxy/internal/docker"
 	idlewatcher "github.com/yusing/go-proxy/internal/docker/idlewatcher/types"
 	"github.com/yusing/go-proxy/internal/gperr"
 	"github.com/yusing/go-proxy/internal/logging"
@@ -29,7 +29,7 @@ type (
 		*idlewatcher.Config
 		*waker
 
-		client       *D.SharedClient
+		client       *docker.SharedClient
 		stopByMethod StopCallback // send a docker command w.r.t. `stop_method`
 		ticker       *time.Ticker
 		lastReset    time.Time
@@ -70,7 +70,7 @@ func registerWatcher(watcherTask *task.Task, route route.Route, waker *waker) (*
 		return w, nil
 	}
 
-	client, err := D.ConnectClient(cfg.DockerHost)
+	client, err := docker.NewClient(cfg.DockerHost)
 	if err != nil {
 		return nil, err
 	}
@@ -146,9 +146,6 @@ func (w *Watcher) containerStart(ctx context.Context) error {
 }
 
 func (w *Watcher) containerStatus() (string, error) {
-	if !w.client.Connected() {
-		return "", errors.New("docker client not connected")
-	}
 	ctx, cancel := context.WithTimeoutCause(w.task.Context(), dockerReqTimeout, errors.New("docker request timeout"))
 	defer cancel()
 	json, err := w.client.ContainerInspect(ctx, w.ContainerID)
@@ -242,7 +239,7 @@ func (w *Watcher) getEventCh(dockerWatcher *watcher.DockerWatcher) (eventCh <-ch
 // it exits only if the context is canceled, the container is destroyed,
 // errors occurred on docker client, or route provider died (mainly caused by config reload).
 func (w *Watcher) watchUntilDestroy() (returnCause error) {
-	dockerWatcher := watcher.NewDockerWatcherWithClient(w.client)
+	dockerWatcher := watcher.NewDockerWatcher(w.Config.DockerHost)
 	dockerEventCh, dockerEventErrCh := w.getEventCh(dockerWatcher)
 
 	for {

internal/docker/inspect.go

@@ -7,7 +7,7 @@ import (
 )
 
 func Inspect(dockerHost string, containerID string) (*Container, error) {
-	client, err := ConnectClient(dockerHost)
+	client, err := NewClient(dockerHost)
 	if err != nil {
 		return nil, err
 	}

internal/docker/list_containers.go

@@ -5,7 +5,6 @@ import (
 	"errors"
 	"time"
 
-	"github.com/docker/docker/api/types"
 	"github.com/docker/docker/api/types/container"
 	"github.com/docker/docker/client"
 )
@@ -22,8 +21,8 @@ var listOptions = container.ListOptions{
 	All: true,
 }
 
-func ListContainers(clientHost string) ([]types.Container, error) {
-	dockerClient, err := ConnectClient(clientHost)
+func ListContainers(clientHost string) ([]container.Summary, error) {
+	dockerClient, err := NewClient(clientHost)
 	if err != nil {
 		return nil, err
 	}