refactor: unify agent communication with Transport interface

- Introduce `Transport` interface to abstract WebSocket and SSH communication - Add generic `Data` field to `AgentResponse` for streamlined future endpoints - Maintain backward compatibility with legacy hubs and agents using typed fields - Unify fetch operations (SMART, systemd, containers) under a single `request` method - Improve `RequestManager` with deadline awareness and legacy response support - Refactor agent response routing into dedicated `agent/response.go` - Update version to 0.18.0-beta.2
2026-04-26 14:31:50 +02:00 · 2026-01-05 13:13:55 -05:00
parent e7b4be3dc5
commit 4547ff7b5d
19 changed files with 623 additions and 457 deletions
--- a/internal/hub/transport/ssh.go
+++ b/internal/hub/transport/ssh.go
@@ -0,0 +1,227 @@
+package transport
+
+import (
+	"context"
+	"errors"
+	"fmt"
+	"io"
+	"net"
+	"strings"
+	"time"
+
+	"github.com/blang/semver"
+	"github.com/fxamacker/cbor/v2"
+	"github.com/henrygd/beszel/internal/common"
+	"golang.org/x/crypto/ssh"
+)
+
+// SSHTransport implements Transport over SSH connections.
+type SSHTransport struct {
+	client       *ssh.Client
+	config       *ssh.ClientConfig
+	host         string
+	port         string
+	agentVersion semver.Version
+	timeout      time.Duration
+}
+
+// SSHTransportConfig holds configuration for creating an SSH transport.
+type SSHTransportConfig struct {
+	Host         string
+	Port         string
+	Config       *ssh.ClientConfig
+	AgentVersion semver.Version
+	Timeout      time.Duration
+}
+
+// NewSSHTransport creates a new SSH transport with the given configuration.
+func NewSSHTransport(cfg SSHTransportConfig) *SSHTransport {
+	timeout := cfg.Timeout
+	if timeout == 0 {
+		timeout = 4 * time.Second
+	}
+	return &SSHTransport{
+		config:       cfg.Config,
+		host:         cfg.Host,
+		port:         cfg.Port,
+		agentVersion: cfg.AgentVersion,
+		timeout:      timeout,
+	}
+}
+
+// SetClient sets the SSH client for reuse across requests.
+func (t *SSHTransport) SetClient(client *ssh.Client) {
+	t.client = client
+}
+
+// SetAgentVersion sets the agent version (extracted from SSH handshake).
+func (t *SSHTransport) SetAgentVersion(version semver.Version) {
+	t.agentVersion = version
+}
+
+// GetClient returns the current SSH client (for connection management).
+func (t *SSHTransport) GetClient() *ssh.Client {
+	return t.client
+}
+
+// GetAgentVersion returns the agent version.
+func (t *SSHTransport) GetAgentVersion() semver.Version {
+	return t.agentVersion
+}
+
+// Request sends a request to the agent via SSH and unmarshals the response.
+func (t *SSHTransport) Request(ctx context.Context, action common.WebSocketAction, req any, dest any) error {
+	if t.client == nil {
+		if err := t.connect(); err != nil {
+			return err
+		}
+	}
+
+	session, err := t.createSessionWithTimeout(ctx)
+	if err != nil {
+		return err
+	}
+	defer session.Close()
+
+	stdout, err := session.StdoutPipe()
+	if err != nil {
+		return err
+	}
+	stdin, err := session.StdinPipe()
+	if err != nil {
+		return err
+	}
+	if err := session.Shell(); err != nil {
+		return err
+	}
+
+	// Send request
+	hubReq := common.HubRequest[any]{Action: action, Data: req}
+	if err := cbor.NewEncoder(stdin).Encode(hubReq); err != nil {
+		return fmt.Errorf("failed to encode request: %w", err)
+	}
+	stdin.Close()
+
+	// Read response
+	var resp common.AgentResponse
+	if err := cbor.NewDecoder(stdout).Decode(&resp); err != nil {
+		return fmt.Errorf("failed to decode response: %w", err)
+	}
+
+	if resp.Error != "" {
+		return errors.New(resp.Error)
+	}
+
+	if err := session.Wait(); err != nil {
+		return err
+	}
+
+	return UnmarshalResponse(resp, action, dest)
+}
+
+// IsConnected returns true if the SSH connection is active.
+func (t *SSHTransport) IsConnected() bool {
+	return t.client != nil
+}
+
+// Close terminates the SSH connection.
+func (t *SSHTransport) Close() {
+	if t.client != nil {
+		t.client.Close()
+		t.client = nil
+	}
+}
+
+// connect establishes a new SSH connection.
+func (t *SSHTransport) connect() error {
+	if t.config == nil {
+		return errors.New("SSH config not set")
+	}
+
+	network := "tcp"
+	host := t.host
+	if strings.HasPrefix(host, "/") {
+		network = "unix"
+	} else {
+		host = net.JoinHostPort(host, t.port)
+	}
+
+	client, err := ssh.Dial(network, host, t.config)
+	if err != nil {
+		return err
+	}
+	t.client = client
+
+	// Extract agent version from server version string
+	t.agentVersion, _ = extractAgentVersion(string(client.Conn.ServerVersion()))
+	return nil
+}
+
+// createSessionWithTimeout creates a new SSH session with a timeout.
+func (t *SSHTransport) createSessionWithTimeout(ctx context.Context) (*ssh.Session, error) {
+	if t.client == nil {
+		return nil, errors.New("client not initialized")
+	}
+
+	ctx, cancel := context.WithTimeout(ctx, t.timeout)
+	defer cancel()
+
+	sessionChan := make(chan *ssh.Session, 1)
+	errChan := make(chan error, 1)
+
+	go func() {
+		session, err := t.client.NewSession()
+		if err != nil {
+			errChan <- err
+		} else {
+			sessionChan <- session
+		}
+	}()
+
+	select {
+	case session := <-sessionChan:
+		return session, nil
+	case err := <-errChan:
+		return nil, err
+	case <-ctx.Done():
+		return nil, errors.New("timeout creating session")
+	}
+}
+
+// extractAgentVersion extracts the beszel version from SSH server version string.
+func extractAgentVersion(versionString string) (semver.Version, error) {
+	_, after, _ := strings.Cut(versionString, "_")
+	return semver.Parse(after)
+}
+
+// RequestWithRetry sends a request with automatic retry on connection failures.
+func (t *SSHTransport) RequestWithRetry(ctx context.Context, action common.WebSocketAction, req any, dest any, retries int) error {
+	var lastErr error
+	for attempt := 0; attempt <= retries; attempt++ {
+		err := t.Request(ctx, action, req, dest)
+		if err == nil {
+			return nil
+		}
+		lastErr = err
+
+		// Check if it's a connection error that warrants a retry
+		if isConnectionError(err) && attempt < retries {
+			t.Close()
+			continue
+		}
+		return err
+	}
+	return lastErr
+}
+
+// isConnectionError checks if an error indicates a connection problem.
+func isConnectionError(err error) bool {
+	if err == nil {
+		return false
+	}
+	errStr := err.Error()
+	return strings.Contains(errStr, "connection") ||
+		strings.Contains(errStr, "EOF") ||
+		strings.Contains(errStr, "closed") ||
+		errors.Is(err, io.EOF)
+}
--- a/internal/hub/transport/transport.go
+++ b/internal/hub/transport/transport.go
@@ -0,0 +1,112 @@
+// Package transport provides a unified abstraction for hub-agent communication
+// over different transports (WebSocket, SSH).
+package transport
+
+import (
+	"context"
+	"errors"
+	"fmt"
+
+	"github.com/fxamacker/cbor/v2"
+	"github.com/henrygd/beszel/internal/common"
+	"github.com/henrygd/beszel/internal/entities/smart"
+	"github.com/henrygd/beszel/internal/entities/system"
+	"github.com/henrygd/beszel/internal/entities/systemd"
+)
+
+// Transport defines the interface for hub-agent communication.
+// Both WebSocket and SSH transports implement this interface.
+type Transport interface {
+	// Request sends a request to the agent and unmarshals the response into dest.
+	// The dest parameter should be a pointer to the expected response type.
+	Request(ctx context.Context, action common.WebSocketAction, req any, dest any) error
+	// IsConnected returns true if the transport connection is active.
+	IsConnected() bool
+	// Close terminates the transport connection.
+	Close()
+}
+
+// UnmarshalResponse unmarshals an AgentResponse into the destination type.
+// It first checks the generic Data field (0.19+ agents), then falls back
+// to legacy typed fields for backward compatibility with 0.18.0 agents.
+func UnmarshalResponse(resp common.AgentResponse, action common.WebSocketAction, dest any) error {
+	if dest == nil {
+		return errors.New("nil destination")
+	}
+	// Try generic Data field first (0.19+)
+	if len(resp.Data) > 0 {
+		if err := cbor.Unmarshal(resp.Data, dest); err != nil {
+			return fmt.Errorf("failed to unmarshal generic response data: %w", err)
+		}
+		return nil
+	}
+	// Fall back to legacy typed fields for older agents/hubs.
+	return unmarshalLegacyResponse(resp, action, dest)
+}
+
+// unmarshalLegacyResponse handles legacy responses that use typed fields.
+func unmarshalLegacyResponse(resp common.AgentResponse, action common.WebSocketAction, dest any) error {
+	switch action {
+	case common.GetData:
+		d, ok := dest.(*system.CombinedData)
+		if !ok {
+			return fmt.Errorf("unexpected dest type for GetData: %T", dest)
+		}
+		if resp.SystemData == nil {
+			return errors.New("no system data in response")
+		}
+		*d = *resp.SystemData
+		return nil
+	case common.CheckFingerprint:
+		d, ok := dest.(*common.FingerprintResponse)
+		if !ok {
+			return fmt.Errorf("unexpected dest type for CheckFingerprint: %T", dest)
+		}
+		if resp.Fingerprint == nil {
+			return errors.New("no fingerprint in response")
+		}
+		*d = *resp.Fingerprint
+		return nil
+	case common.GetContainerLogs:
+		d, ok := dest.(*string)
+		if !ok {
+			return fmt.Errorf("unexpected dest type for GetContainerLogs: %T", dest)
+		}
+		if resp.String == nil {
+			return errors.New("no logs in response")
+		}
+		*d = *resp.String
+		return nil
+	case common.GetContainerInfo:
+		d, ok := dest.(*string)
+		if !ok {
+			return fmt.Errorf("unexpected dest type for GetContainerInfo: %T", dest)
+		}
+		if resp.String == nil {
+			return errors.New("no info in response")
+		}
+		*d = *resp.String
+		return nil
+	case common.GetSmartData:
+		d, ok := dest.(*map[string]smart.SmartData)
+		if !ok {
+			return fmt.Errorf("unexpected dest type for GetSmartData: %T", dest)
+		}
+		if resp.SmartData == nil {
+			return errors.New("no SMART data in response")
+		}
+		*d = resp.SmartData
+		return nil
+	case common.GetSystemdInfo:
+		d, ok := dest.(*systemd.ServiceDetails)
+		if !ok {
+			return fmt.Errorf("unexpected dest type for GetSystemdInfo: %T", dest)
+		}
+		if resp.ServiceInfo == nil {
+			return errors.New("no systemd info in response")
+		}
+		*d = resp.ServiceInfo
+		return nil
+	}
+	return fmt.Errorf("unsupported action: %d", action)
+}
--- a/internal/hub/transport/websocket.go
+++ b/internal/hub/transport/websocket.go
@@ -0,0 +1,74 @@
+package transport
+
+import (
+	"context"
+	"errors"
+
+	"github.com/fxamacker/cbor/v2"
+	"github.com/henrygd/beszel"
+	"github.com/henrygd/beszel/internal/common"
+	"github.com/henrygd/beszel/internal/hub/ws"
+)
+
+// ErrWebSocketNotConnected indicates a WebSocket transport is not currently connected.
+var ErrWebSocketNotConnected = errors.New("websocket not connected")
+
+// WebSocketTransport implements Transport over WebSocket connections.
+type WebSocketTransport struct {
+	wsConn *ws.WsConn
+}
+
+// NewWebSocketTransport creates a new WebSocket transport wrapper.
+func NewWebSocketTransport(wsConn *ws.WsConn) *WebSocketTransport {
+	return &WebSocketTransport{wsConn: wsConn}
+}
+
+// Request sends a request to the agent via WebSocket and unmarshals the response.
+func (t *WebSocketTransport) Request(ctx context.Context, action common.WebSocketAction, req any, dest any) error {
+	if !t.IsConnected() {
+		return ErrWebSocketNotConnected
+	}
+
+	pendingReq, err := t.wsConn.SendRequest(ctx, action, req)
+	if err != nil {
+		return err
+	}
+
+	// Wait for response
+	select {
+	case message := <-pendingReq.ResponseCh:
+		defer message.Close()
+		defer pendingReq.Cancel()
+
+		// Legacy agents (< MinVersionAgentResponse) respond with a raw payload instead of an AgentResponse wrapper.
+		if t.wsConn.AgentVersion().LT(beszel.MinVersionAgentResponse) {
+			return cbor.Unmarshal(message.Data.Bytes(), dest)
+		}
+
+		var agentResponse common.AgentResponse
+		if err := cbor.Unmarshal(message.Data.Bytes(), &agentResponse); err != nil {
+			return err
+		}
+
+		if agentResponse.Error != "" {
+			return errors.New(agentResponse.Error)
+		}
+
+		return UnmarshalResponse(agentResponse, action, dest)
+
+	case <-pendingReq.Context.Done():
+		return pendingReq.Context.Err()
+	}
+}
+
+// IsConnected returns true if the WebSocket connection is active.
+func (t *WebSocketTransport) IsConnected() bool {
+	return t.wsConn != nil && t.wsConn.IsConnected()
+}
+
+// Close terminates the WebSocket connection.
+func (t *WebSocketTransport) Close() {
+	if t.wsConn != nil {
+		t.wsConn.Close(nil)
+	}
+}