feat: agents connect by WebSocket, falling back to long polling and trying again each minute; auto, ws or lp per machine (0.1.0)

internal/agent/wslink.go

@@ -0,0 +1,172 @@
+package agent
+
+import (
+	"context"
+	"errors"
+	"fmt"
+	"log"
+	"net/http"
+	"strings"
+	"sync"
+	"time"
+
+	"github.com/coder/websocket"
+	"github.com/coder/websocket/wsjson"
+
+	"github.com/x-grk/longpolling-remote-access/internal/proto"
+)
+
+// The WebSocket to the server: control messages arrive on it as soon as they
+// are queued, and terminal and http-forward sessions send and take their
+// bytes over it instead of a request each. When it drops, sessions carry on
+// over plain requests.
+
+const wsPing = 20 * time.Second
+
+type wsLink struct {
+	c     *websocket.Conn
+	ctx   context.Context
+	mu    sync.Mutex
+	subs  map[string]chan wsIn
+	asked map[string]bool // sessions whose input was asked for
+	done  chan struct{}
+}
+
+// wsIn is what the server sent for a session: its input, or that it is over.
+type wsIn struct {
+	frames []proto.InFrame
+	gone   bool
+}
+
+func (l *wsLink) send(m proto.WSMsg) error {
+	ctx, cancel := context.WithTimeout(l.ctx, 30*time.Second)
+	defer cancel()
+	return wsjson.Write(ctx, l.c, m)
+}
+
+// subscribe asks the server for a session's input on this link, once, and
+// returns where it arrives; nil when the link cannot take it.
+func (l *wsLink) subscribe(sid string) chan wsIn {
+	l.mu.Lock()
+	ch := l.subs[sid]
+	if ch == nil {
+		ch = make(chan wsIn, 64)
+		l.subs[sid] = ch
+	}
+	fresh := !l.subscribed(sid)
+	l.mu.Unlock()
+	if fresh && l.send(proto.WSMsg{T: "sub", SID: sid}) != nil {
+		return nil
+	}
+	return ch
+}
+
+// subscribed marks a session as asked for; l.mu is held.
+func (l *wsLink) subscribed(sid string) bool {
+	if l.asked == nil {
+		l.asked = map[string]bool{}
+	}
+	if l.asked[sid] {
+		return true
+	}
+	l.asked[sid] = true
+	return false
+}
+
+func (l *wsLink) unsubscribe(sid string) {
+	l.mu.Lock()
+	delete(l.subs, sid)
+	delete(l.asked, sid)
+	l.mu.Unlock()
+}
+
+func (l *wsLink) deliver(sid string, in wsIn) {
+	l.mu.Lock()
+	ch := l.subs[sid]
+	l.mu.Unlock()
+	if ch == nil {
+		return
+	}
+	select {
+	case ch <- in:
+	case <-l.done:
+	}
+}
+
+// runWS holds the WebSocket until it ends, dispatching what comes on it.
+func (a *Agent) runWS(ctx context.Context) error {
+	u := a.cfg.Host + "/api/agent/ws"
+	u = "ws" + strings.TrimPrefix(u, "http")
+	hdr := http.Header{}
+	hdr.Set("X-Agent-Id", a.state.AgentID)
+	hdr.Set("Authorization", "Bearer "+a.state.Token)
+	hdr.Set("X-Agent-Version", Version)
+	hdr.Set("X-Agent-User", currentUser())
+	dctx, dcancel := context.WithTimeout(ctx, 15*time.Second)
+	c, resp, err := websocket.Dial(dctx, u, &websocket.DialOptions{
+		HTTPClient: a.stream, HTTPHeader: hdr, CompressionMode: websocket.CompressionDisabled,
+	})
+	dcancel()
+	if err != nil {
+		if resp != nil {
+			a.applyTransport(resp.Header.Get(proto.TransportHeader))
+			return fmt.Errorf("server answered %s", resp.Status)
+		}
+		return err
+	}
+	c.SetReadLimit(8 << 20)
+	lctx, lcancel := context.WithCancel(ctx)
+	l := &wsLink{c: c, ctx: lctx, subs: map[string]chan wsIn{}, done: make(chan struct{})}
+	a.link.Store(l)
+	defer func() {
+		a.link.CompareAndSwap(l, nil)
+		close(l.done)
+		lcancel()
+		c.CloseNow()
+	}()
+	log.Printf("connected by websocket")
+	go func() {
+		t := time.NewTicker(wsPing)
+		defer t.Stop()
+		for {
+			select {
+			case <-lctx.Done():
+				return
+			case <-t.C:
+				pctx, pcancel := context.WithTimeout(lctx, 15*time.Second)
+				err := c.Ping(pctx)
+				pcancel()
+				if err != nil {
+					lcancel()
+					return
+				}
+			}
+		}
+	}()
+	for {
+		var m proto.WSMsg
+		if err := wsjson.Read(lctx, c, &m); err != nil {
+			if errors.Is(err, context.Canceled) && ctx.Err() == nil {
+				return errors.New("no answer to ping")
+			}
+			return err
+		}
+		switch m.T {
+		case "ctrl":
+			if m.Ctrl != nil {
+				// Sessions live on the agent's context: they outlast the link.
+				a.dispatch(ctx, *m.Ctrl)
+			}
+		case "cfg":
+			a.applyTunnelAddr(ctx, m.Tunnel)
+			a.applyTransport(m.Transport)
+			if a.mode() == proto.TransportLP {
+				return errors.New("set to long polling")
+			}
+		case "in":
+			l.deliver(m.SID, wsIn{frames: m.Frames})
+		case "gone":
+			l.deliver(m.SID, wsIn{gone: true})
+		}
+	}
+}