OwnCord/Server/ws/livekit_process.go

// Package ws provides the LiveKit companion process manager.
//
// LiveKitProcess manages the lifecycle of a livekit-server binary running
// alongside chatserver. It auto-generates a minimal livekit.yaml config,
// starts the process, monitors health, and restarts on crash.
package ws

import (
	"context"
	"fmt"
	"log/slog"
	"net/http"
	"os"
	"os/exec"
	"path/filepath"
	"sync"
	"time"

	"github.com/owncord/server/config"
)

// LiveKitProcess manages the companion livekit-server binary.
type LiveKitProcess struct {
	cfg        *config.VoiceConfig
	tlsCfg     *config.TLSConfig
	dataDir    string
	httpClient *http.Client // for health checks — no redirect following

	mu       sync.Mutex
	cmd      *exec.Cmd
	cancel   context.CancelFunc
	stopped  bool
	runDone  chan struct{} // closed by runLoop when cmd.Run() returns
	loopDone chan struct{} // closed when runLoop exits entirely
}

// NewLiveKitProcess creates a new process manager. It does not start the
// process — call Start() to launch the LiveKit server.
// The tlsCfg is used to configure TLS on the LiveKit server using the same
// certs as OwnCord (avoids mixed-content blocks in WebView2).
func NewLiveKitProcess(cfg *config.VoiceConfig, tlsCfg *config.TLSConfig, dataDir string) *LiveKitProcess {
	return &LiveKitProcess{
		cfg:     cfg,
		tlsCfg:  tlsCfg,
		dataDir: dataDir,
		httpClient: &http.Client{
			CheckRedirect: func(*http.Request, []*http.Request) error {
				return http.ErrUseLastResponse
			},
		},
	}
}

// generateConfig writes a minimal livekit.yaml for the companion process.
func (p *LiveKitProcess) generateConfig() (string, error) {
	cfgPath := filepath.Join(p.dataDir, "livekit.yaml")

	// No TURN TLS config — LiveKit signaling is proxied through OwnCord's
	// HTTPS server at /livekit/*, so no separate TLS is needed on LiveKit.

	// Sanitize credentials for safe YAML interpolation: reject strings
	// containing characters that could break YAML structure.
	for _, cred := range []string{p.cfg.LiveKitAPIKey, p.cfg.LiveKitAPISecret} {
		for _, ch := range cred {
			if ch == ':' || ch == '#' || ch == '{' || ch == '}' || ch == '\n' || ch == '\r' || ch == '"' || ch == '\\' {
				return "", fmt.Errorf("LiveKit credential contains unsafe YAML character %q", string(ch))
			}
		}
	}
	// Build node_ip line only when configured (required for remote users behind NAT).
	// Validate: must be a plain IP address (no YAML-breaking chars).
	nodeIPLine := ""
	if p.cfg.NodeIP != "" {
		for _, ch := range p.cfg.NodeIP {
			if ch == '"' || ch == '\\' || ch == '\n' || ch == '\r' || ch == '#' || ch == '{' || ch == '}' {
				return "", fmt.Errorf("node_ip contains unsafe character %q", string(ch))
			}
		}
		nodeIPLine = fmt.Sprintf("\n  node_ip: \"%s\"", p.cfg.NodeIP)
	}

	content := fmt.Sprintf(`# Auto-generated by OwnCord — do not edit manually.
port: 7880

rtc:
  port_range_start: 50000
  port_range_end: 60000
  use_external_ip: true%s
  pli_throttle:
    low_quality: 500ms
    mid_quality: 1s
    high_quality: 1s

keys:
  "%s": "%s"

logging:
  level: info
`, nodeIPLine, p.cfg.LiveKitAPIKey, p.cfg.LiveKitAPISecret)

	if err := os.MkdirAll(p.dataDir, 0o755); err != nil {
		return "", fmt.Errorf("creating data dir: %w", err)
	}
	if err := os.WriteFile(cfgPath, []byte(content), 0o600); err != nil {
		return "", fmt.Errorf("writing livekit config: %w", err)
	}

	return cfgPath, nil
}

// Start launches the livekit-server binary. If LiveKitBinaryPath is empty,
// this is a no-op (assumes LiveKit is managed externally).
func (p *LiveKitProcess) Start() error {
	if p.cfg.LiveKitBinaryPath == "" {
		slog.Info("livekit: no binary path configured, assuming externally managed")
		return nil
	}

	p.mu.Lock()
	defer p.mu.Unlock()

	if p.cmd != nil {
		return fmt.Errorf("livekit process already running")
	}

	cfgPath, err := p.generateConfig()
	if err != nil {
		return fmt.Errorf("generating livekit config: %w", err)
	}

	ctx, cancel := context.WithCancel(context.Background())
	p.cancel = cancel
	p.loopDone = make(chan struct{})

	go p.runLoop(ctx, cfgPath)

	return nil
}

// runLoop starts and restarts the process until stopped or context cancelled.
// Uses exponential backoff (3s → 6s → 12s … up to 60s) and stops after 10
// consecutive rapid failures (process exits within 30 seconds).
func (p *LiveKitProcess) runLoop(ctx context.Context, cfgPath string) {
	defer func() {
		p.mu.Lock()
		if p.loopDone != nil {
			close(p.loopDone)
		}
		p.mu.Unlock()
	}()

	const (
		baseDelay      = 3 * time.Second
		maxDelay       = 60 * time.Second
		maxRetries     = 10
		stableAfter    = 30 * time.Second // reset counter if process runs longer than this
	)

	rapidFailures := 0
	delay := baseDelay

	for {
		if ctx.Err() != nil {
			return
		}

		cmd := exec.CommandContext(ctx, p.cfg.LiveKitBinaryPath, "--config", cfgPath)
		cmd.Stdout = os.Stdout
		cmd.Stderr = os.Stderr
		cmd.WaitDelay = 6 * time.Second // bound Wait to prevent goroutine leak on Windows

		p.mu.Lock()
		if p.stopped {
			p.mu.Unlock()
			return
		}
		p.cmd = cmd
		p.runDone = make(chan struct{})
		p.mu.Unlock()

		slog.Info("livekit: starting process",
			"binary", p.cfg.LiveKitBinaryPath,
			"config", cfgPath,
			"rapid_failures", rapidFailures)

		startTime := time.Now()
		err := cmd.Run()

		p.mu.Lock()
		p.cmd = nil
		if p.runDone != nil {
			close(p.runDone)
			p.runDone = nil
		}
		stopped := p.stopped
		p.mu.Unlock()

		if stopped || ctx.Err() != nil {
			slog.Info("livekit: process stopped")
			return
		}

		// If the process ran for a while, it was stable — reset backoff.
		if time.Since(startTime) > stableAfter {
			rapidFailures = 0
			delay = baseDelay
		} else {
			rapidFailures++
		}

		if err != nil {
			slog.Error("livekit: process exited unexpectedly",
				"error", err,
				"rapid_failures", rapidFailures,
				"restart_delay", delay)
		}

		if rapidFailures >= maxRetries {
			slog.Error("livekit: too many rapid failures, giving up",
				"rapid_failures", rapidFailures)
			return
		}

		select {
		case <-time.After(delay):
			slog.Info("livekit: restarting process")
		case <-ctx.Done():
			return
		}

		// Exponential backoff capped at maxDelay.
		delay *= 2
		if delay > maxDelay {
			delay = maxDelay
		}
	}
}

// IsRunning returns true if the companion process is currently running.
func (p *LiveKitProcess) IsRunning() bool {
	p.mu.Lock()
	defer p.mu.Unlock()
	return p.cmd != nil && p.cmd.Process != nil
}

// HealthCheck probes the LiveKit HTTP endpoint to verify it is accepting
// connections. Returns true if the server responds (any status code).
func (p *LiveKitProcess) HealthCheck() (bool, error) {
	httpURL := wsToHTTP(p.cfg.LiveKitURL)

	ctx, cancel := context.WithTimeout(context.Background(), 3*time.Second)
	defer cancel()

	req, err := http.NewRequestWithContext(ctx, http.MethodGet, httpURL, nil)
	if err != nil {
		return false, fmt.Errorf("creating health check request: %w", err)
	}

	resp, err := p.httpClient.Do(req)
	if err != nil {
		return false, fmt.Errorf("livekit health check failed: %w", err)
	}
	_ = resp.Body.Close()

	return true, nil
}

// Stop gracefully stops the companion process.
// It cancels the context (which signals runLoop) and waits up to 5 seconds
// for the process to exit. The actual cmd.Wait() is done by runLoop via
// cmd.Run() — we only monitor the process via cmd.Process.Wait() here to
// avoid calling exec.Cmd.Wait() twice (which has undefined behavior).
func (p *LiveKitProcess) Stop() {
	p.mu.Lock()
	p.stopped = true
	cancel := p.cancel
	cmd := p.cmd
	done := p.runDone
	loopDone := p.loopDone
	p.mu.Unlock()

	if cancel != nil {
		cancel()
	}

	// Wait for runLoop's cmd.Run() to return (which closes runDone).
	// This avoids calling cmd.Wait() or cmd.Process.Wait() from a second
	// goroutine, which is unsafe on Windows.
	if done != nil {
		select {
		case <-done:
			slog.Info("livekit: process exited cleanly")
		case <-time.After(5 * time.Second):
			slog.Warn("livekit: process did not exit in time, killing")
			if cmd != nil && cmd.Process != nil {
				_ = cmd.Process.Kill()
			}
		}
	}

	// Wait for the entire runLoop goroutine to finish, ensuring no
	// new iteration can start after Stop returns.
	if loopDone != nil {
		select {
		case <-loopDone:
		case <-time.After(5 * time.Second):
		}
	}
}