refactor(repo): consolidate openflare-server to root and move subprojects to internal/apps

- Merge all files inside openflare-server to the repository root directory. - Relocate agent, relay, and flared subprojects from internal/ to internal/apps/. - Combine docker-compose files and update build context paths to root. - Update GitHub workflows and Dockerfiles to refer to new directories and package names. - Rewrite Go package imports across all files. - Resolve database renew test race condition and clean up docs.
2026-10-05 23:26:38 +08:00 · 2026-06-19 14:23:29 +08:00
parent 19d476ed7f
commit 63cd906cfc
1064 changed files with 366 additions and 1397 deletions
@@ -0,0 +1,350 @@
+package frpc
+
+import (
+	"bytes"
+	"context"
+	"encoding/json"
+	"fmt"
+	"log/slog"
+	"net"
+	"os"
+	"os/exec"
+	"path/filepath"
+	"strings"
+	"sync"
+	"time"
+
+	"github.com/Rain-kl/Wavelet/internal/apps/flared/config"
+	service "github.com/Rain-kl/Wavelet/pkg/protocol"
+)
+
+type Manager struct {
+	cfg       *config.Config
+	processes map[string]*Process
+	mu        sync.RWMutex
+
+	currentVersion  string
+	currentChecksum string
+}
+
+type Process struct {
+	RelayID   string
+	Cmd       *exec.Cmd
+	Cancel    context.CancelFunc
+	Status    string
+	StartTime time.Time
+	LastError string
+}
+
+func NewManager(cfg *config.Config) *Manager {
+	return &Manager{
+		cfg:       cfg,
+		processes: make(map[string]*Process),
+	}
+}
+
+func (m *Manager) GetVersion() string {
+	cmd := exec.Command(m.cfg.FrpcPath, "-v")
+	out, err := cmd.Output()
+	if err != nil {
+		return "unknown"
+	}
+	return strings.TrimSpace(string(out))
+}
+
+func (m *Manager) GetConnectedRelays() []service.FlaredConnectedRelay {
+	m.mu.RLock()
+	defer m.mu.RUnlock()
+
+	result := make([]service.FlaredConnectedRelay, 0, len(m.processes))
+	for relayID, proc := range m.processes {
+		result = append(result, service.FlaredConnectedRelay{
+			RelayNodeID: relayID,
+			Status:      proc.Status,
+		})
+	}
+	return result
+}
+
+func (m *Manager) GetCurrentConfigVersion() string {
+	m.mu.RLock()
+	defer m.mu.RUnlock()
+	return m.currentVersion
+}
+
+func (m *Manager) GetCurrentConfigChecksum() string {
+	m.mu.RLock()
+	defer m.mu.RUnlock()
+	return m.currentChecksum
+}
+
+func (m *Manager) UpdateConfig(ctx context.Context, newConfig *service.FlaredTunnelConfigResponse) error {
+	m.mu.Lock()
+	defer m.mu.Unlock()
+
+	if newConfig == nil {
+		return nil
+	}
+
+	versionChanged := newConfig.Version != m.currentVersion || newConfig.Checksum != m.currentChecksum
+	if versionChanged {
+		slog.Info("applying new tunnel config", "version", newConfig.Version)
+	} else {
+		slog.Debug("tunnel config version unchanged, ensuring processes are running", "version", newConfig.Version)
+	}
+
+	if err := os.MkdirAll(m.cfg.DataDir, 0o755); err != nil {
+		return fmt.Errorf("create data dir failed: %w", err)
+	}
+
+	activeRelays := make(map[string]struct{})
+
+	for _, relay := range newConfig.Relays {
+		activeRelays[relay.RelayNodeID] = struct{}{}
+		tomlContent := buildFrpcToml(relay, newConfig.Proxies)
+		configPath := filepath.Join(m.cfg.DataDir, fmt.Sprintf("frpc_%s.toml", relay.RelayNodeID))
+
+		needsRestart := false
+		existingData, err := os.ReadFile(configPath)
+		if err != nil || string(existingData) != tomlContent {
+			// 配置文件不存在或内容有变化，需要写入并重启
+			needsRestart = true
+		}
+
+		if needsRestart {
+			if err := os.WriteFile(configPath, []byte(tomlContent), 0o644); err != nil {
+				slog.Error("failed to write frpc config", "relay_id", relay.RelayNodeID, "error", err)
+				continue
+			}
+			m.restartProcess(ctx, relay.RelayNodeID, configPath)
+		} else if _, ok := m.processes[relay.RelayNodeID]; !ok {
+			// 配置未变但进程不存在（如重启后），直接启动进程
+			slog.Info("frpc process missing, starting", "relay_id", relay.RelayNodeID)
+			m.restartProcess(ctx, relay.RelayNodeID, configPath)
+		}
+	}
+
+	// Stop obsolete processes
+	for relayID, proc := range m.processes {
+		if _, ok := activeRelays[relayID]; !ok {
+			slog.Info("stopping obsolete frpc process", "relay_id", relayID)
+			proc.Cancel()
+			pidPath := filepath.Join(m.cfg.DataDir, fmt.Sprintf("frpc_%s.pid", relayID))
+			_ = os.Remove(pidPath)
+			delete(m.processes, relayID)
+		}
+	}
+
+	if versionChanged {
+		m.currentVersion = newConfig.Version
+		m.currentChecksum = newConfig.Checksum
+		return m.saveState()
+	}
+	return nil
+}
+
+func (m *Manager) restartProcess(ctx context.Context, relayID string, configPath string) {
+	pidPath := filepath.Join(m.cfg.DataDir, fmt.Sprintf("frpc_%s.pid", relayID))
+	if proc, ok := m.processes[relayID]; ok {
+		proc.Cancel()
+		_ = os.Remove(pidPath)
+	}
+
+	if ctx == nil {
+		ctx = context.Background()
+	}
+	procCtx, cancel := context.WithCancel(ctx)
+	proc := &Process{
+		RelayID:   relayID,
+		Cancel:    cancel,
+		Status:    "starting",
+		StartTime: time.Now(),
+	}
+	m.processes[relayID] = proc
+
+	go func() {
+		backoff := 1 * time.Second
+		const maxBackoff = 60 * time.Second
+
+		for {
+			m.mu.Lock()
+			if procCtx.Err() != nil {
+				m.mu.Unlock()
+				return
+			}
+			m.mu.Unlock()
+
+			ensureNoOrphanProcess(pidPath)
+
+			cmd := exec.CommandContext(procCtx, m.cfg.FrpcPath, "-c", configPath)
+
+			m.mu.Lock()
+			proc.Cmd = cmd
+			proc.Status = "running"
+			m.mu.Unlock()
+
+			startedAt := time.Now()
+			err := cmd.Start()
+			if err == nil {
+				_ = os.WriteFile(pidPath, []byte(fmt.Sprintf("%d", cmd.Process.Pid)), 0o644)
+				err = cmd.Wait()
+			}
+			_ = os.Remove(pidPath)
+
+			m.mu.Lock()
+			if procCtx.Err() != nil {
+				proc.Status = "stopped"
+				m.mu.Unlock()
+				return
+			}
+
+			if err != nil {
+				proc.LastError = err.Error()
+				proc.Status = "error"
+				slog.Error("frpc process exited unexpectedly", "relay_id", relayID, "error", err)
+			} else {
+				proc.Status = "stopped"
+				proc.LastError = "exited unexpectedly with code 0"
+				slog.Warn("frpc process exited unexpectedly with code 0", "relay_id", relayID)
+			}
+			m.mu.Unlock()
+
+			if time.Since(startedAt) >= 10*time.Second {
+				backoff = 1 * time.Second
+			}
+
+			select {
+			case <-procCtx.Done():
+				return
+			case <-time.After(backoff):
+				backoff = backoff * 2
+				if backoff > maxBackoff {
+					backoff = maxBackoff
+				}
+			}
+		}
+	}()
+}
+
+func (m *Manager) Stop() {
+	m.mu.Lock()
+	defer m.mu.Unlock()
+
+	for relayID, proc := range m.processes {
+		if proc != nil && proc.Cancel != nil {
+			proc.Cancel()
+		}
+		pidPath := filepath.Join(m.cfg.DataDir, fmt.Sprintf("frpc_%s.pid", relayID))
+		_ = os.Remove(pidPath)
+		delete(m.processes, relayID)
+	}
+}
+
+func buildFrpcToml(relay service.FlaredRelayInfo, proxies []service.FlaredProxyEntry) string {
+	var buf bytes.Buffer
+
+	host, port := parseAddr(relay.Address)
+
+	buf.WriteString(fmt.Sprintf(`serverAddr = "%s"
+serverPort = %s
+`, host, port))
+
+	if relay.AuthToken != "" {
+		buf.WriteString(fmt.Sprintf(`auth.method = "token"
+auth.token = "%s"
+`, relay.AuthToken))
+	}
+
+	if relay.ProxyURL != "" {
+		buf.WriteString(fmt.Sprintf(`transport.proxyURL = "%s"
+`, relay.ProxyURL))
+	}
+
+	buf.WriteString("\n")
+
+	for _, proxy := range proxies {
+		buf.WriteString(fmt.Sprintf("[[proxies]]\nname = \"%s\"\ntype = \"%s\"\nlocalIP = \"%s\"\nlocalPort = %d\n",
+			proxy.Name, proxy.Type, proxy.LocalAddr, proxy.LocalPort))
+		if len(proxy.CustomDomains) > 0 {
+			buf.WriteString(fmt.Sprintf("customDomains = [\"%s\"]\n", strings.Join(proxy.CustomDomains, "\", \"")))
+		}
+		buf.WriteString("\n")
+	}
+
+	return buf.String()
+}
+
+func parseAddr(addr string) (string, string) {
+	addr = strings.TrimSpace(addr)
+	if addr == "" {
+		return "127.0.0.1", "7000"
+	}
+	host, port, err := net.SplitHostPort(addr)
+	if err == nil {
+		return strings.Trim(host, "[]"), port
+	}
+	lastColon := strings.LastIndex(addr, ":")
+	if lastColon > 0 && strings.Count(addr, ":") == 1 {
+		return addr[:lastColon], addr[lastColon+1:]
+	}
+	return addr, "7000"
+}
+
+// State persistence
+type ManagerState struct {
+	Version  string
+	Checksum string
+}
+
+func (m *Manager) saveState() error {
+	state := ManagerState{
+		Version:  m.currentVersion,
+		Checksum: m.currentChecksum,
+	}
+	data, err := json.MarshalIndent(state, "", "  ")
+	if err != nil {
+		return err
+	}
+	return os.WriteFile(m.cfg.StatePath, data, 0o644)
+}
+
+func (m *Manager) LoadState() error {
+	data, err := os.ReadFile(m.cfg.StatePath)
+	if err != nil {
+		if os.IsNotExist(err) {
+			return nil
+		}
+		return err
+	}
+	var state ManagerState
+	if err := json.Unmarshal(data, &state); err != nil {
+		return err
+	}
+	m.mu.Lock()
+	m.currentVersion = state.Version
+	m.currentChecksum = state.Checksum
+	m.mu.Unlock()
+	return nil
+}
+
+func ensureNoOrphanProcess(pidPath string) {
+	data, err := os.ReadFile(pidPath)
+	if err != nil {
+		return
+	}
+	var pid int
+	if _, err := fmt.Sscanf(string(data), "%d", &pid); err != nil {
+		return
+	}
+	if pid <= 0 {
+		return
+	}
+	process, err := os.FindProcess(pid)
+	if err == nil && process != nil {
+		slog.Warn("attempting to kill potentially orphan process", "pid", pid, "pid_path", pidPath)
+		_ = process.Kill()
+		// Wait a little bit to ensure the OS has reclaimed ports
+		time.Sleep(500 * time.Millisecond)
+	}
+	_ = os.Remove(pidPath)
+}
@@ -0,0 +1,387 @@
+package frpc
+
+import (
+	"context"
+	"fmt"
+	"os"
+	"os/exec"
+	"path/filepath"
+	"strings"
+	"testing"
+	"time"
+
+	"github.com/Rain-kl/Wavelet/internal/apps/flared/config"
+	service "github.com/Rain-kl/Wavelet/pkg/protocol"
+)
+
+// Helper to write control file for the dummy script
+func writeControl(t *testing.T, dir string, exitCode int, delaySeconds int) {
+	controlPath := filepath.Join(dir, "control.txt")
+	content := fmt.Sprintf("%d %d\n", exitCode, delaySeconds)
+	err := os.WriteFile(controlPath, []byte(content), 0644)
+	if err != nil {
+		t.Fatalf("failed to write control file: %v", err)
+	}
+}
+
+// Setup a dummy executable script that reads control.txt to decide exit code and sleep duration
+func setupDummyScript(t *testing.T) (string, string) {
+	dir := t.TempDir()
+	scriptPath := filepath.Join(dir, "dummy_frpc")
+
+	// On macOS/Linux, we write a shell script
+	scriptContent := fmt.Sprintf(`#!/bin/sh
+control_file="%s/control.txt"
+EXIT_CODE=0
+DELAY=0
+if [ -f "$control_file" ]; then
+    read -r EXIT_CODE DELAY < "$control_file"
+fi
+if [ -n "$DELAY" ] && [ "$DELAY" -gt 0 ] 2>/dev/null; then
+    sleep "$DELAY"
+fi
+exit "${EXIT_CODE:-0}"
+`, dir)
+
+	err := os.WriteFile(scriptPath, []byte(scriptContent), 0755)
+	if err != nil {
+		t.Fatalf("failed to write dummy script: %v", err)
+	}
+
+	return scriptPath, dir
+}
+
+// Helper to poll for status to eliminate timing flakiness in tests
+func assertStatusEventually(t *testing.T, m *Manager, relayID string, expectedStatus string, timeout time.Duration) {
+	deadline := time.Now().Add(timeout)
+	for time.Now().Before(deadline) {
+		m.mu.RLock()
+		proc, ok := m.processes[relayID]
+		m.mu.RUnlock()
+		if ok && proc.Status == expectedStatus {
+			return
+		}
+		time.Sleep(50 * time.Millisecond)
+	}
+	m.mu.RLock()
+	proc, ok := m.processes[relayID]
+	var got string
+	var errStr string
+	if ok {
+		got = proc.Status
+		errStr = proc.LastError
+	} else {
+		got = "not_found"
+	}
+	m.mu.RUnlock()
+	t.Fatalf("expected status eventually %s, got %s (err: %s)", expectedStatus, got, errStr)
+}
+
+func assertCommandExitedEventually(t *testing.T, cmd *exec.Cmd, timeout time.Duration) {
+	t.Helper()
+
+	done := make(chan error, 1)
+	go func() {
+		done <- cmd.Wait()
+	}()
+
+	select {
+	case <-time.After(timeout):
+		t.Fatalf("expected process pid=%d to exit within %s", cmd.Process.Pid, timeout)
+	case <-done:
+	}
+}
+
+func TestStartProcessSuccess(t *testing.T) {
+	scriptPath, dir := setupDummyScript(t)
+	writeControl(t, dir, 0, 5) // exit code 0, sleep 5s
+
+	cfg := &config.Config{
+		ServerURL:   "http://localhost:8080",
+		TunnelToken: "test-token",
+		FrpcPath:    scriptPath,
+		DataDir:     dir,
+		StatePath:   filepath.Join(dir, "flared-state.json"),
+	}
+
+	m := NewManager(cfg)
+
+	newConfig := &service.FlaredTunnelConfigResponse{
+		Version:  "1",
+		Checksum: "sum1",
+		Relays: []service.FlaredRelayInfo{
+			{
+				RelayNodeID: "relay-1",
+				Address:     "127.0.0.1:7000",
+				AuthToken:   "auth-1",
+			},
+		},
+		Proxies: nil,
+	}
+
+	err := m.UpdateConfig(context.Background(), newConfig)
+	if err != nil {
+		t.Fatalf("failed to UpdateConfig: %v", err)
+	}
+
+	assertStatusEventually(t, m, "relay-1", "running", 4*time.Second)
+
+	m.mu.RLock()
+	proc := m.processes["relay-1"]
+	m.mu.RUnlock()
+
+	proc.Cancel()
+	assertStatusEventually(t, m, "relay-1", "stopped", 4*time.Second) // wait for clean stop
+}
+
+func TestStartProcessFailureAndBackoff(t *testing.T) {
+	dir := t.TempDir()
+	invalidScriptPath := filepath.Join(dir, "non_existent_frpc")
+
+	cfg := &config.Config{
+		ServerURL:   "http://localhost:8080",
+		TunnelToken: "test-token",
+		FrpcPath:    invalidScriptPath,
+		DataDir:     dir,
+		StatePath:   filepath.Join(dir, "flared-state.json"),
+	}
+
+	m := NewManager(cfg)
+	newConfig := &service.FlaredTunnelConfigResponse{
+		Version:  "1",
+		Checksum: "sum1",
+		Relays: []service.FlaredRelayInfo{
+			{
+				RelayNodeID: "relay-1",
+				Address:     "127.0.0.1:7000",
+				AuthToken:   "auth-1",
+			},
+		},
+		Proxies: nil,
+	}
+
+	_ = m.UpdateConfig(context.Background(), newConfig)
+
+	assertStatusEventually(t, m, "relay-1", "error", 4*time.Second)
+
+	// Correct the path to dummy script
+	scriptPath, _ := setupDummyScript(t)
+	writeControl(t, filepath.Dir(scriptPath), 0, 5)
+
+	m.mu.Lock()
+	m.cfg.FrpcPath = scriptPath
+	m.mu.Unlock()
+
+	// Wait for backoff retry (1s backoff)
+	assertStatusEventually(t, m, "relay-1", "running", 4*time.Second)
+
+	m.mu.RLock()
+	proc := m.processes["relay-1"]
+	m.mu.RUnlock()
+	proc.Cancel()
+}
+
+func TestUnexpectedExit0CPUProtection(t *testing.T) {
+	scriptPath, dir := setupDummyScript(t)
+	// Start with immediate exit code 0
+	writeControl(t, dir, 0, 0)
+
+	cfg := &config.Config{
+		ServerURL:   "http://localhost:8080",
+		TunnelToken: "test-token",
+		FrpcPath:    scriptPath,
+		DataDir:     dir,
+		StatePath:   filepath.Join(dir, "flared-state.json"),
+	}
+
+	m := NewManager(cfg)
+	newConfig := &service.FlaredTunnelConfigResponse{
+		Version:  "1",
+		Checksum: "sum1",
+		Relays: []service.FlaredRelayInfo{
+			{
+				RelayNodeID: "relay-1",
+				Address:     "127.0.0.1:7000",
+				AuthToken:   "auth-1",
+			},
+		},
+		Proxies: nil,
+	}
+
+	_ = m.UpdateConfig(context.Background(), newConfig)
+
+	assertStatusEventually(t, m, "relay-1", "stopped", 4*time.Second)
+
+	m.mu.RLock()
+	proc := m.processes["relay-1"]
+	if !strings.Contains(proc.LastError, "exited unexpectedly with code 0") {
+		t.Errorf("expected LastError to record exit status 0 warning, got %s", proc.LastError)
+	}
+	m.mu.RUnlock()
+
+	proc.Cancel()
+}
+
+func TestBackoffReset(t *testing.T) {
+	scriptPath, dir := setupDummyScript(t)
+	// Rapid exit code 1 to increase backoff
+	writeControl(t, dir, 1, 0)
+
+	cfg := &config.Config{
+		ServerURL:   "http://localhost:8080",
+		TunnelToken: "test-token",
+		FrpcPath:    scriptPath,
+		DataDir:     dir,
+		StatePath:   filepath.Join(dir, "flared-state.json"),
+	}
+
+	m := NewManager(cfg)
+	newConfig := &service.FlaredTunnelConfigResponse{
+		Version:  "1",
+		Checksum: "sum1",
+		Relays: []service.FlaredRelayInfo{
+			{
+				RelayNodeID: "relay-1",
+				Address:     "127.0.0.1:7000",
+				AuthToken:   "auth-1",
+			},
+		},
+		Proxies: nil,
+	}
+
+	_ = m.UpdateConfig(context.Background(), newConfig)
+
+	// Wait to crash
+	assertStatusEventually(t, m, "relay-1", "error", 4*time.Second)
+
+	// Now make it run successfully for 11 seconds (exit code 0, sleep 11s)
+	writeControl(t, dir, 0, 11)
+
+	// Wait for next retry to start running
+	assertStatusEventually(t, m, "relay-1", "running", 4*time.Second)
+
+	// Wait for process to run for 10.5 seconds to trigger backoff reset
+	time.Sleep(10500 * time.Millisecond)
+
+	// Now make it crash again (exit code 1, sleep 0s)
+	writeControl(t, dir, 1, 0)
+
+	// Wait for it to finish and crash
+	assertStatusEventually(t, m, "relay-1", "error", 4*time.Second)
+
+	// It crashed. Since it ran for > 10s, backoff should have been reset to 1s.
+	// We make it healthy again (exit code 0, sleep 5)
+	writeControl(t, dir, 0, 5)
+
+	// Wait 1.5 seconds. If backoff was reset to 1s, it should be running now.
+	assertStatusEventually(t, m, "relay-1", "running", 4*time.Second)
+
+	m.mu.RLock()
+	proc := m.processes["relay-1"]
+	m.mu.RUnlock()
+	proc.Cancel()
+}
+
+func TestUpdateConfigKillsOrphanProcessBeforeRestart(t *testing.T) {
+	scriptPath, dir := setupDummyScript(t)
+	writeControl(t, dir, 0, 5)
+
+	cfg := &config.Config{
+		ServerURL:   "http://localhost:8080",
+		TunnelToken: "test-token",
+		FrpcPath:    scriptPath,
+		DataDir:     dir,
+		StatePath:   filepath.Join(dir, "flared-state.json"),
+	}
+
+	m := NewManager(cfg)
+
+	orphan := exec.Command("sh", "-c", "sleep 30")
+	if err := orphan.Start(); err != nil {
+		t.Fatalf("failed to start orphan process: %v", err)
+	}
+	t.Cleanup(func() {
+		if orphan.Process != nil {
+			_ = orphan.Process.Kill()
+		}
+	})
+
+	pidPath := filepath.Join(dir, "frpc_relay-1.pid")
+	if err := os.WriteFile(pidPath, []byte(fmt.Sprintf("%d", orphan.Process.Pid)), 0o644); err != nil {
+		t.Fatalf("failed to seed orphan pid file: %v", err)
+	}
+
+	newConfig := &service.FlaredTunnelConfigResponse{
+		Version:  "1",
+		Checksum: "sum1",
+		Relays: []service.FlaredRelayInfo{
+			{
+				RelayNodeID: "relay-1",
+				Address:     "127.0.0.1:7000",
+				AuthToken:   "auth-1",
+			},
+		},
+	}
+
+	if err := m.UpdateConfig(context.Background(), newConfig); err != nil {
+		t.Fatalf("failed to UpdateConfig: %v", err)
+	}
+
+	assertCommandExitedEventually(t, orphan, 2*time.Second)
+	assertStatusEventually(t, m, "relay-1", "running", 4*time.Second)
+
+	m.mu.RLock()
+	proc := m.processes["relay-1"]
+	m.mu.RUnlock()
+	proc.Cancel()
+}
+
+func TestStopCancelsRunningProcesses(t *testing.T) {
+	scriptPath, dir := setupDummyScript(t)
+	writeControl(t, dir, 0, 30)
+
+	cfg := &config.Config{
+		ServerURL:   "http://localhost:8080",
+		TunnelToken: "test-token",
+		FrpcPath:    scriptPath,
+		DataDir:     dir,
+		StatePath:   filepath.Join(dir, "flared-state.json"),
+	}
+
+	m := NewManager(cfg)
+	newConfig := &service.FlaredTunnelConfigResponse{
+		Version:  "1",
+		Checksum: "sum1",
+		Relays: []service.FlaredRelayInfo{
+			{
+				RelayNodeID: "relay-1",
+				Address:     "127.0.0.1:7000",
+				AuthToken:   "auth-1",
+			},
+		},
+	}
+
+	if err := m.UpdateConfig(context.Background(), newConfig); err != nil {
+		t.Fatalf("failed to UpdateConfig: %v", err)
+	}
+
+	assertStatusEventually(t, m, "relay-1", "running", 4*time.Second)
+
+	m.mu.RLock()
+	proc := m.processes["relay-1"]
+	if proc == nil || proc.Cmd == nil {
+		m.mu.RUnlock()
+		t.Fatal("expected running process to have a command handle")
+	}
+	cmd := proc.Cmd
+	m.mu.RUnlock()
+
+	m.Stop()
+	assertCommandExitedEventually(t, cmd, 2*time.Second)
+
+	m.mu.RLock()
+	defer m.mu.RUnlock()
+	if len(m.processes) != 0 {
+		t.Fatalf("expected no managed processes after stop, got %d", len(m.processes))
+	}
+}