Merge pull request #126 from bnema/fix/daemon-double-fork
fix(app): detach daemon with double fork
3 files · +210 -6
M internal/app/run.go +5 -0
| @@ -61,6 +61,7 @@ const ( | |||
| 61 | 61 | kindKill | |
| 62 | 62 | kindCmd | |
| 63 | 63 | kindDaemon | |
| 64 | + | kindDaemonLauncher | |
| 64 | 65 | kindStdio | |
| 65 | 66 | kindUDPBootstrap | |
| 66 | 67 | kindUDPProxy | |
| @@ -141,6 +142,8 @@ func parseArgs(args []string) (command, error) { | |||
| 141 | 142 | switch args[0] { | |
| 142 | 143 | case "--daemon": | |
| 143 | 144 | return command{kind: kindDaemon}, nil | |
| 145 | + | case "--daemon-launcher": | |
| 146 | + | return command{kind: kindDaemonLauncher}, nil | |
| 144 | 147 | case "_stdio": | |
| 145 | 148 | if len(args) > 2 { | |
| 146 | 149 | return command{}, usagef("`_stdio` accepts at most one session name") | |
| @@ -244,6 +247,8 @@ func dispatch(ctx context.Context, cmd command) error { | |||
| 244 | 247 | return nil | |
| 245 | 248 | case kindDaemon: | |
| 246 | 249 | return runDaemon() | |
| 250 | + | case kindDaemonLauncher: | |
| 251 | + | return runDaemonLauncher() | |
| 247 | 252 | case kindStdio: | |
| 248 | 253 | return runStdio(ctx) | |
| 249 | 254 | case kindUDPBootstrap: | |
| @@ -61,6 +61,7 @@ const ( | |||||
| 61 | kindKill | 61 | kindKill | ||
| 62 | kindCmd | 62 | kindCmd | ||
| 63 | kindDaemon | 63 | kindDaemon | ||
| 64 | + | kindDaemonLauncher | |||
| 64 | kindStdio | 65 | kindStdio | ||
| 65 | kindUDPBootstrap | 66 | kindUDPBootstrap | ||
| 66 | kindUDPProxy | 67 | kindUDPProxy | ||
| @@ -141,6 +142,8 @@ func parseArgs(args []string) (command, error) { | |||||
| 141 | switch args[0] { | 142 | switch args[0] { | ||
| 142 | case "--daemon": | 143 | case "--daemon": | ||
| 143 | return command{kind: kindDaemon}, nil | 144 | return command{kind: kindDaemon}, nil | ||
| 145 | + | case "--daemon-launcher": | |||
| 146 | + | return command{kind: kindDaemonLauncher}, nil | |||
| 144 | case "_stdio": | 147 | case "_stdio": | ||
| 145 | if len(args) > 2 { | 148 | if len(args) > 2 { | ||
| 146 | return command{}, usagef("`_stdio` accepts at most one session name") | 149 | return command{}, usagef("`_stdio` accepts at most one session name") | ||
| @@ -244,6 +247,8 @@ func dispatch(ctx context.Context, cmd command) error { | |||||
| 244 | return nil | 247 | return nil | ||
| 245 | case kindDaemon: | 248 | case kindDaemon: | ||
| 246 | return runDaemon() | 249 | return runDaemon() | ||
| 250 | + | case kindDaemonLauncher: | |||
| 251 | + | return runDaemonLauncher() | |||
| 247 | case kindStdio: | 252 | case kindStdio: | ||
| 248 | return runStdio(ctx) | 253 | return runStdio(ctx) | ||
| 249 | case kindUDPBootstrap: | 254 | case kindUDPBootstrap: | ||
M internal/app/spawn.go +35 -6
| @@ -1,6 +1,7 @@ | |||
| 1 | 1 | package app | |
| 2 | 2 | ||
| 3 | 3 | import ( | |
| 4 | + | "bytes" | |
| 4 | 5 | "context" | |
| 5 | 6 | "errors" | |
| 6 | 7 | "fmt" | |
| @@ -152,10 +153,9 @@ func realDial(ctx context.Context, dir string) (ports.Transport, error) { | |||
| 152 | 153 | return ipc.DialContext(ctx, dir) | |
| 153 | 154 | } | |
| 154 | 155 | ||
| 155 | - | // realSpawn re-execs this binary as a detached daemon: a new session | |
| 156 | - | // (Setsid) with stdio bound to /dev/null so it survives the client's exit | |
| 157 | - | // and never writes to the client's terminal. It logs via slog to the shared | |
| 158 | - | // log file instead. | |
| 156 | + | // realSpawn re-execs this binary through a short-lived launcher. Waiting for | |
| 157 | + | // that launcher to exit ensures the daemon has been re-parented before the | |
| 158 | + | // client continues, keeping process-tree termination from reaching it. | |
| 159 | 159 | func realSpawn() error { | |
| 160 | 160 | exePath, err := selfExePath() | |
| 161 | 161 | if err != nil { | |
| @@ -168,6 +168,37 @@ func realSpawn() error { | |||
| 168 | 168 | } | |
| 169 | 169 | defer func() { _ = devNull.Close() }() | |
| 170 | 170 | ||
| 171 | + | var stderr bytes.Buffer | |
| 172 | + | cmd := exec.Command(exePath, "--daemon-launcher") | |
| 173 | + | cmd.Dir = platform.DirOrHome("") | |
| 174 | + | cmd.Stdin = devNull | |
| 175 | + | cmd.Stdout = devNull | |
| 176 | + | cmd.Stderr = &stderr | |
| 177 | + | if err := cmd.Run(); err != nil { | |
| 178 | + | if detail := bytes.TrimSpace(stderr.Bytes()); len(detail) > 0 { | |
| 179 | + | return fmt.Errorf("%w: %s", err, detail) | |
| 180 | + | } | |
| 181 | + | return err | |
| 182 | + | } | |
| 183 | + | slog.Info("daemon process detached") | |
| 184 | + | return nil | |
| 185 | + | } | |
| 186 | + | ||
| 187 | + | // runDaemonLauncher is the intermediate half of the double-fork. It starts | |
| 188 | + | // the long-lived daemon in a new session and exits immediately; realSpawn | |
| 189 | + | // waits for this exit so the daemon is no longer descended from the client. | |
| 190 | + | func runDaemonLauncher() error { | |
| 191 | + | exePath, err := selfExePath() | |
| 192 | + | if err != nil { | |
| 193 | + | return fmt.Errorf("resolving executable path: %w", err) | |
| 194 | + | } | |
| 195 | + | ||
| 196 | + | devNull, err := os.OpenFile(os.DevNull, os.O_RDWR, 0) | |
| 197 | + | if err != nil { | |
| 198 | + | return fmt.Errorf("opening %s: %w", os.DevNull, err) | |
| 199 | + | } | |
| 200 | + | defer func() { _ = devNull.Close() }() | |
| 201 | + | ||
| 171 | 202 | cmd := exec.Command(exePath, "--daemon") | |
| 172 | 203 | cmd.Dir = platform.DirOrHome("") | |
| 173 | 204 | cmd.Stdin = devNull | |
| @@ -177,7 +208,5 @@ func realSpawn() error { | |||
| 177 | 208 | if err := cmd.Start(); err != nil { | |
| 178 | 209 | return err | |
| 179 | 210 | } | |
| 180 | - | slog.Info("daemon process started", "pid", cmd.Process.Pid) | |
| 181 | - | // Detach: we neither wait for nor signal the daemon after launch. | |
| 182 | 211 | return cmd.Process.Release() | |
| 183 | 212 | } | |
| @@ -1,6 +1,7 @@ | |||||
| 1 | package app | 1 | package app | ||
| 2 | 2 | ||||
| 3 | import ( | 3 | import ( | ||
| 4 | + | "bytes" | |||
| 4 | "context" | 5 | "context" | ||
| 5 | "errors" | 6 | "errors" | ||
| 6 | "fmt" | 7 | "fmt" | ||
| @@ -152,10 +153,9 @@ func realDial(ctx context.Context, dir string) (ports.Transport, error) { | |||||
| 152 | return ipc.DialContext(ctx, dir) | 153 | return ipc.DialContext(ctx, dir) | ||
| 153 | } | 154 | } | ||
| 154 | 155 | ||||
| 155 | - | // realSpawn re-execs this binary as a detached daemon: a new session | 156 | + | // realSpawn re-execs this binary through a short-lived launcher. Waiting for |
| 156 | - | // (Setsid) with stdio bound to /dev/null so it survives the client's exit | 157 | + | // that launcher to exit ensures the daemon has been re-parented before the |
| 157 | - | // and never writes to the client's terminal. It logs via slog to the shared | 158 | + | // client continues, keeping process-tree termination from reaching it. |
| 158 | - | // log file instead. | |||
| 159 | func realSpawn() error { | 159 | func realSpawn() error { | ||
| 160 | exePath, err := selfExePath() | 160 | exePath, err := selfExePath() | ||
| 161 | if err != nil { | 161 | if err != nil { | ||
| @@ -168,6 +168,37 @@ func realSpawn() error { | |||||
| 168 | } | 168 | } | ||
| 169 | defer func() { _ = devNull.Close() }() | 169 | defer func() { _ = devNull.Close() }() | ||
| 170 | 170 | ||||
| 171 | + | var stderr bytes.Buffer | |||
| 172 | + | cmd := exec.Command(exePath, "--daemon-launcher") | |||
| 173 | + | cmd.Dir = platform.DirOrHome("") | |||
| 174 | + | cmd.Stdin = devNull | |||
| 175 | + | cmd.Stdout = devNull | |||
| 176 | + | cmd.Stderr = &stderr | |||
| 177 | + | if err := cmd.Run(); err != nil { | |||
| 178 | + | if detail := bytes.TrimSpace(stderr.Bytes()); len(detail) > 0 { | |||
| 179 | + | return fmt.Errorf("%w: %s", err, detail) | |||
| 180 | + | } | |||
| 181 | + | return err | |||
| 182 | + | } | |||
| 183 | + | slog.Info("daemon process detached") | |||
| 184 | + | return nil | |||
| 185 | + | } | |||
| 186 | + | ||||
| 187 | + | // runDaemonLauncher is the intermediate half of the double-fork. It starts | |||
| 188 | + | // the long-lived daemon in a new session and exits immediately; realSpawn | |||
| 189 | + | // waits for this exit so the daemon is no longer descended from the client. | |||
| 190 | + | func runDaemonLauncher() error { | |||
| 191 | + | exePath, err := selfExePath() | |||
| 192 | + | if err != nil { | |||
| 193 | + | return fmt.Errorf("resolving executable path: %w", err) | |||
| 194 | + | } | |||
| 195 | + | ||||
| 196 | + | devNull, err := os.OpenFile(os.DevNull, os.O_RDWR, 0) | |||
| 197 | + | if err != nil { | |||
| 198 | + | return fmt.Errorf("opening %s: %w", os.DevNull, err) | |||
| 199 | + | } | |||
| 200 | + | defer func() { _ = devNull.Close() }() | |||
| 201 | + | ||||
| 171 | cmd := exec.Command(exePath, "--daemon") | 202 | cmd := exec.Command(exePath, "--daemon") | ||
| 172 | cmd.Dir = platform.DirOrHome("") | 203 | cmd.Dir = platform.DirOrHome("") | ||
| 173 | cmd.Stdin = devNull | 204 | cmd.Stdin = devNull | ||
| @@ -177,7 +208,5 @@ func realSpawn() error { | |||||
| 177 | if err := cmd.Start(); err != nil { | 208 | if err := cmd.Start(); err != nil { | ||
| 178 | return err | 209 | return err | ||
| 179 | } | 210 | } | ||
| 180 | - | slog.Info("daemon process started", "pid", cmd.Process.Pid) | |||
| 181 | - | // Detach: we neither wait for nor signal the daemon after launch. | |||
| 182 | return cmd.Process.Release() | 211 | return cmd.Process.Release() | ||
| 183 | } | 212 | } | ||
M internal/app/spawn_test.go +170 -0
| @@ -3,10 +3,13 @@ package app | |||
| 3 | 3 | import ( | |
| 4 | 4 | "context" | |
| 5 | 5 | "errors" | |
| 6 | + | "fmt" | |
| 6 | 7 | "os" | |
| 7 | 8 | "path/filepath" | |
| 9 | + | "strconv" | |
| 8 | 10 | "sync" | |
| 9 | 11 | "sync/atomic" | |
| 12 | + | "syscall" | |
| 10 | 13 | "testing" | |
| 11 | 14 | "time" | |
| 12 | 15 | ||
| @@ -16,6 +19,173 @@ import ( | |||
| 16 | 19 | ||
| 17 | 20 | var errDialFailed = errors.New("dial failed") | |
| 18 | 21 | ||
| 22 | + | const ( | |
| 23 | + | spawnTestChildFileEnv = "VEV_SPAWN_TEST_CHILD_FILE" | |
| 24 | + | spawnTestLauncherFileEnv = "VEV_SPAWN_TEST_LAUNCHER_FILE" | |
| 25 | + | spawnTestReleaseFileEnv = "VEV_SPAWN_TEST_RELEASE_FILE" | |
| 26 | + | ) | |
| 27 | + | ||
| 28 | + | func TestMain(m *testing.M) { | |
| 29 | + | if len(os.Args) == 2 { | |
| 30 | + | switch os.Args[1] { | |
| 31 | + | case "--daemon-launcher": | |
| 32 | + | launcherFile := os.Getenv(spawnTestLauncherFileEnv) | |
| 33 | + | if launcherFile != "" { | |
| 34 | + | if err := writeProcessRecord(launcherFile); err != nil { | |
| 35 | + | os.Exit(2) | |
| 36 | + | } | |
| 37 | + | } | |
| 38 | + | if err := Run(os.Args[1:]); err != nil { | |
| 39 | + | os.Exit(2) | |
| 40 | + | } | |
| 41 | + | if releaseFile := os.Getenv(spawnTestReleaseFileEnv); releaseFile != "" { | |
| 42 | + | deadline := time.Now().Add(5 * time.Second) | |
| 43 | + | for { | |
| 44 | + | if _, err := os.Stat(releaseFile); err == nil { | |
| 45 | + | break | |
| 46 | + | } else if !errors.Is(err, os.ErrNotExist) || !time.Now().Before(deadline) { | |
| 47 | + | os.Exit(2) | |
| 48 | + | } | |
| 49 | + | time.Sleep(10 * time.Millisecond) | |
| 50 | + | } | |
| 51 | + | } | |
| 52 | + | os.Exit(0) | |
| 53 | + | case "--daemon": | |
| 54 | + | path := os.Getenv(spawnTestChildFileEnv) | |
| 55 | + | if path == "" { | |
| 56 | + | os.Exit(2) | |
| 57 | + | } | |
| 58 | + | if err := writeProcessRecord(path); err != nil { | |
| 59 | + | os.Exit(2) | |
| 60 | + | } | |
| 61 | + | for { | |
| 62 | + | time.Sleep(time.Hour) | |
| 63 | + | } | |
| 64 | + | } | |
| 65 | + | } | |
| 66 | + | os.Exit(m.Run()) | |
| 67 | + | } | |
| 68 | + | ||
| 69 | + | func TestRealSpawnWaitsForLauncherExit(t *testing.T) { | |
| 70 | + | dir := t.TempDir() | |
| 71 | + | childFile := filepath.Join(dir, "daemon.pid") | |
| 72 | + | launcherFile := filepath.Join(dir, "launcher.pid") | |
| 73 | + | releaseFile := filepath.Join(dir, "release-launcher") | |
| 74 | + | t.Setenv(spawnTestChildFileEnv, childFile) | |
| 75 | + | t.Setenv(spawnTestLauncherFileEnv, launcherFile) | |
| 76 | + | t.Setenv(spawnTestReleaseFileEnv, releaseFile) | |
| 77 | + | ||
| 78 | + | // Register cleanup before starting either subprocess. The release marker | |
| 79 | + | // unblocks a launcher even when setup or an assertion fails, and PID files | |
| 80 | + | // let cleanup terminate every helper that reached its acknowledgement. | |
| 81 | + | t.Cleanup(func() { | |
| 82 | + | _ = os.WriteFile(releaseFile, nil, 0o600) | |
| 83 | + | for _, path := range []string{launcherFile, childFile} { | |
| 84 | + | if err := terminateProcessFromFile(path); err != nil { | |
| 85 | + | t.Errorf("clean up %s: %v", filepath.Base(path), err) | |
| 86 | + | } | |
| 87 | + | } | |
| 88 | + | }) | |
| 89 | + | ||
| 90 | + | spawnResult := make(chan error, 1) | |
| 91 | + | go func() { | |
| 92 | + | spawnResult <- realSpawn() | |
| 93 | + | }() | |
| 94 | + | ||
| 95 | + | launcherPID := waitForProcessRecord(t, launcherFile) | |
| 96 | + | daemonPID := waitForProcessRecord(t, childFile) | |
| 97 | + | ||
| 98 | + | select { | |
| 99 | + | case err := <-spawnResult: | |
| 100 | + | t.Fatalf("realSpawn returned while launcher %d was still running: %v", launcherPID, err) | |
| 101 | + | default: | |
| 102 | + | } | |
| 103 | + | if err := syscall.Kill(launcherPID, 0); err != nil { | |
| 104 | + | t.Fatalf("launcher exited before release acknowledgement: %v", err) | |
| 105 | + | } | |
| 106 | + | ||
| 107 | + | if err := os.WriteFile(releaseFile, nil, 0o600); err != nil { | |
| 108 | + | t.Fatalf("release launcher: %v", err) | |
| 109 | + | } | |
| 110 | + | select { | |
| 111 | + | case err := <-spawnResult: | |
| 112 | + | if err != nil { | |
| 113 | + | t.Fatalf("realSpawn: %v", err) | |
| 114 | + | } | |
| 115 | + | case <-time.After(2 * time.Second): | |
| 116 | + | t.Fatal("realSpawn did not return after launcher release") | |
| 117 | + | } | |
| 118 | + | ||
| 119 | + | if err := waitForProcessExit(launcherPID, 2*time.Second); err != nil { | |
| 120 | + | t.Fatalf("launcher still exists after realSpawn returned: %v", err) | |
| 121 | + | } | |
| 122 | + | if err := syscall.Kill(daemonPID, 0); err != nil { | |
| 123 | + | t.Fatalf("daemon exited with its launcher: %v", err) | |
| 124 | + | } | |
| 125 | + | } | |
| 126 | + | ||
| 127 | + | func waitForProcessRecord(t *testing.T, path string) int { | |
| 128 | + | t.Helper() | |
| 129 | + | deadline := time.Now().Add(2 * time.Second) | |
| 130 | + | for time.Now().Before(deadline) { | |
| 131 | + | data, err := os.ReadFile(path) | |
| 132 | + | if err == nil { | |
| 133 | + | pid, err := strconv.Atoi(string(data)) | |
| 134 | + | if err != nil { | |
| 135 | + | t.Fatalf("parse process record %s: %v", filepath.Base(path), err) | |
| 136 | + | } | |
| 137 | + | return pid | |
| 138 | + | } | |
| 139 | + | if !errors.Is(err, os.ErrNotExist) { | |
| 140 | + | t.Fatalf("read process record %s: %v", filepath.Base(path), err) | |
| 141 | + | } | |
| 142 | + | time.Sleep(10 * time.Millisecond) | |
| 143 | + | } | |
| 144 | + | t.Fatalf("process did not write %s", filepath.Base(path)) | |
| 145 | + | return 0 | |
| 146 | + | } | |
| 147 | + | ||
| 148 | + | func waitForProcessExit(pid int, timeout time.Duration) error { | |
| 149 | + | deadline := time.Now().Add(timeout) | |
| 150 | + | for time.Now().Before(deadline) { | |
| 151 | + | err := syscall.Kill(pid, 0) | |
| 152 | + | if errors.Is(err, syscall.ESRCH) { | |
| 153 | + | return nil | |
| 154 | + | } | |
| 155 | + | if err != nil && !errors.Is(err, syscall.EPERM) { | |
| 156 | + | return err | |
| 157 | + | } | |
| 158 | + | time.Sleep(10 * time.Millisecond) | |
| 159 | + | } | |
| 160 | + | return fmt.Errorf("process %d did not exit within %s", pid, timeout) | |
| 161 | + | } | |
| 162 | + | ||
| 163 | + | func writeProcessRecord(path string) error { | |
| 164 | + | tmp := path + ".tmp" | |
| 165 | + | if err := os.WriteFile(tmp, []byte(strconv.Itoa(os.Getpid())), 0o600); err != nil { | |
| 166 | + | return err | |
| 167 | + | } | |
| 168 | + | return os.Rename(tmp, path) | |
| 169 | + | } | |
| 170 | + | ||
| 171 | + | func terminateProcessFromFile(path string) error { | |
| 172 | + | data, err := os.ReadFile(path) | |
| 173 | + | if errors.Is(err, os.ErrNotExist) { | |
| 174 | + | return nil | |
| 175 | + | } | |
| 176 | + | if err != nil { | |
| 177 | + | return err | |
| 178 | + | } | |
| 179 | + | pid, err := strconv.Atoi(string(data)) | |
| 180 | + | if err != nil { | |
| 181 | + | return err | |
| 182 | + | } | |
| 183 | + | if err := syscall.Kill(pid, syscall.SIGKILL); err != nil && !errors.Is(err, syscall.ESRCH) { | |
| 184 | + | return err | |
| 185 | + | } | |
| 186 | + | return waitForProcessExit(pid, 2*time.Second) | |
| 187 | + | } | |
| 188 | + | ||
| 19 | 189 | // fastBackoff keeps the retry loop snappy for tests. | |
| 20 | 190 | var fastBackoff = backoffConfig{initial: time.Millisecond, max: 5 * time.Millisecond, total: 100 * time.Millisecond} | |
| 21 | 191 | ||
| @@ -3,10 +3,13 @@ package app | |||||
| 3 | import ( | 3 | import ( | ||
| 4 | "context" | 4 | "context" | ||
| 5 | "errors" | 5 | "errors" | ||
| 6 | + | "fmt" | |||
| 6 | "os" | 7 | "os" | ||
| 7 | "path/filepath" | 8 | "path/filepath" | ||
| 9 | + | "strconv" | |||
| 8 | "sync" | 10 | "sync" | ||
| 9 | "sync/atomic" | 11 | "sync/atomic" | ||
| 12 | + | "syscall" | |||
| 10 | "testing" | 13 | "testing" | ||
| 11 | "time" | 14 | "time" | ||
| 12 | 15 | ||||
| @@ -16,6 +19,173 @@ import ( | |||||
| 16 | 19 | ||||
| 17 | var errDialFailed = errors.New("dial failed") | 20 | var errDialFailed = errors.New("dial failed") | ||
| 18 | 21 | ||||
| 22 | + | const ( | |||
| 23 | + | spawnTestChildFileEnv = "VEV_SPAWN_TEST_CHILD_FILE" | |||
| 24 | + | spawnTestLauncherFileEnv = "VEV_SPAWN_TEST_LAUNCHER_FILE" | |||
| 25 | + | spawnTestReleaseFileEnv = "VEV_SPAWN_TEST_RELEASE_FILE" | |||
| 26 | + | ) | |||
| 27 | + | ||||
| 28 | + | func TestMain(m *testing.M) { | |||
| 29 | + | if len(os.Args) == 2 { | |||
| 30 | + | switch os.Args[1] { | |||
| 31 | + | case "--daemon-launcher": | |||
| 32 | + | launcherFile := os.Getenv(spawnTestLauncherFileEnv) | |||
| 33 | + | if launcherFile != "" { | |||
| 34 | + | if err := writeProcessRecord(launcherFile); err != nil { | |||
| 35 | + | os.Exit(2) | |||
| 36 | + | } | |||
| 37 | + | } | |||
| 38 | + | if err := Run(os.Args[1:]); err != nil { | |||
| 39 | + | os.Exit(2) | |||
| 40 | + | } | |||
| 41 | + | if releaseFile := os.Getenv(spawnTestReleaseFileEnv); releaseFile != "" { | |||
| 42 | + | deadline := time.Now().Add(5 * time.Second) | |||
| 43 | + | for { | |||
| 44 | + | if _, err := os.Stat(releaseFile); err == nil { | |||
| 45 | + | break | |||
| 46 | + | } else if !errors.Is(err, os.ErrNotExist) || !time.Now().Before(deadline) { | |||
| 47 | + | os.Exit(2) | |||
| 48 | + | } | |||
| 49 | + | time.Sleep(10 * time.Millisecond) | |||
| 50 | + | } | |||
| 51 | + | } | |||
| 52 | + | os.Exit(0) | |||
| 53 | + | case "--daemon": | |||
| 54 | + | path := os.Getenv(spawnTestChildFileEnv) | |||
| 55 | + | if path == "" { | |||
| 56 | + | os.Exit(2) | |||
| 57 | + | } | |||
| 58 | + | if err := writeProcessRecord(path); err != nil { | |||
| 59 | + | os.Exit(2) | |||
| 60 | + | } | |||
| 61 | + | for { | |||
| 62 | + | time.Sleep(time.Hour) | |||
| 63 | + | } | |||
| 64 | + | } | |||
| 65 | + | } | |||
| 66 | + | os.Exit(m.Run()) | |||
| 67 | + | } | |||
| 68 | + | ||||
| 69 | + | func TestRealSpawnWaitsForLauncherExit(t *testing.T) { | |||
| 70 | + | dir := t.TempDir() | |||
| 71 | + | childFile := filepath.Join(dir, "daemon.pid") | |||
| 72 | + | launcherFile := filepath.Join(dir, "launcher.pid") | |||
| 73 | + | releaseFile := filepath.Join(dir, "release-launcher") | |||
| 74 | + | t.Setenv(spawnTestChildFileEnv, childFile) | |||
| 75 | + | t.Setenv(spawnTestLauncherFileEnv, launcherFile) | |||
| 76 | + | t.Setenv(spawnTestReleaseFileEnv, releaseFile) | |||
| 77 | + | ||||
| 78 | + | // Register cleanup before starting either subprocess. The release marker | |||
| 79 | + | // unblocks a launcher even when setup or an assertion fails, and PID files | |||
| 80 | + | // let cleanup terminate every helper that reached its acknowledgement. | |||
| 81 | + | t.Cleanup(func() { | |||
| 82 | + | _ = os.WriteFile(releaseFile, nil, 0o600) | |||
| 83 | + | for _, path := range []string{launcherFile, childFile} { | |||
| 84 | + | if err := terminateProcessFromFile(path); err != nil { | |||
| 85 | + | t.Errorf("clean up %s: %v", filepath.Base(path), err) | |||
| 86 | + | } | |||
| 87 | + | } | |||
| 88 | + | }) | |||
| 89 | + | ||||
| 90 | + | spawnResult := make(chan error, 1) | |||
| 91 | + | go func() { | |||
| 92 | + | spawnResult <- realSpawn() | |||
| 93 | + | }() | |||
| 94 | + | ||||
| 95 | + | launcherPID := waitForProcessRecord(t, launcherFile) | |||
| 96 | + | daemonPID := waitForProcessRecord(t, childFile) | |||
| 97 | + | ||||
| 98 | + | select { | |||
| 99 | + | case err := <-spawnResult: | |||
| 100 | + | t.Fatalf("realSpawn returned while launcher %d was still running: %v", launcherPID, err) | |||
| 101 | + | default: | |||
| 102 | + | } | |||
| 103 | + | if err := syscall.Kill(launcherPID, 0); err != nil { | |||
| 104 | + | t.Fatalf("launcher exited before release acknowledgement: %v", err) | |||
| 105 | + | } | |||
| 106 | + | ||||
| 107 | + | if err := os.WriteFile(releaseFile, nil, 0o600); err != nil { | |||
| 108 | + | t.Fatalf("release launcher: %v", err) | |||
| 109 | + | } | |||
| 110 | + | select { | |||
| 111 | + | case err := <-spawnResult: | |||
| 112 | + | if err != nil { | |||
| 113 | + | t.Fatalf("realSpawn: %v", err) | |||
| 114 | + | } | |||
| 115 | + | case <-time.After(2 * time.Second): | |||
| 116 | + | t.Fatal("realSpawn did not return after launcher release") | |||
| 117 | + | } | |||
| 118 | + | ||||
| 119 | + | if err := waitForProcessExit(launcherPID, 2*time.Second); err != nil { | |||
| 120 | + | t.Fatalf("launcher still exists after realSpawn returned: %v", err) | |||
| 121 | + | } | |||
| 122 | + | if err := syscall.Kill(daemonPID, 0); err != nil { | |||
| 123 | + | t.Fatalf("daemon exited with its launcher: %v", err) | |||
| 124 | + | } | |||
| 125 | + | } | |||
| 126 | + | ||||
| 127 | + | func waitForProcessRecord(t *testing.T, path string) int { | |||
| 128 | + | t.Helper() | |||
| 129 | + | deadline := time.Now().Add(2 * time.Second) | |||
| 130 | + | for time.Now().Before(deadline) { | |||
| 131 | + | data, err := os.ReadFile(path) | |||
| 132 | + | if err == nil { | |||
| 133 | + | pid, err := strconv.Atoi(string(data)) | |||
| 134 | + | if err != nil { | |||
| 135 | + | t.Fatalf("parse process record %s: %v", filepath.Base(path), err) | |||
| 136 | + | } | |||
| 137 | + | return pid | |||
| 138 | + | } | |||
| 139 | + | if !errors.Is(err, os.ErrNotExist) { | |||
| 140 | + | t.Fatalf("read process record %s: %v", filepath.Base(path), err) | |||
| 141 | + | } | |||
| 142 | + | time.Sleep(10 * time.Millisecond) | |||
| 143 | + | } | |||
| 144 | + | t.Fatalf("process did not write %s", filepath.Base(path)) | |||
| 145 | + | return 0 | |||
| 146 | + | } | |||
| 147 | + | ||||
| 148 | + | func waitForProcessExit(pid int, timeout time.Duration) error { | |||
| 149 | + | deadline := time.Now().Add(timeout) | |||
| 150 | + | for time.Now().Before(deadline) { | |||
| 151 | + | err := syscall.Kill(pid, 0) | |||
| 152 | + | if errors.Is(err, syscall.ESRCH) { | |||
| 153 | + | return nil | |||
| 154 | + | } | |||
| 155 | + | if err != nil && !errors.Is(err, syscall.EPERM) { | |||
| 156 | + | return err | |||
| 157 | + | } | |||
| 158 | + | time.Sleep(10 * time.Millisecond) | |||
| 159 | + | } | |||
| 160 | + | return fmt.Errorf("process %d did not exit within %s", pid, timeout) | |||
| 161 | + | } | |||
| 162 | + | ||||
| 163 | + | func writeProcessRecord(path string) error { | |||
| 164 | + | tmp := path + ".tmp" | |||
| 165 | + | if err := os.WriteFile(tmp, []byte(strconv.Itoa(os.Getpid())), 0o600); err != nil { | |||
| 166 | + | return err | |||
| 167 | + | } | |||
| 168 | + | return os.Rename(tmp, path) | |||
| 169 | + | } | |||
| 170 | + | ||||
| 171 | + | func terminateProcessFromFile(path string) error { | |||
| 172 | + | data, err := os.ReadFile(path) | |||
| 173 | + | if errors.Is(err, os.ErrNotExist) { | |||
| 174 | + | return nil | |||
| 175 | + | } | |||
| 176 | + | if err != nil { | |||
| 177 | + | return err | |||
| 178 | + | } | |||
| 179 | + | pid, err := strconv.Atoi(string(data)) | |||
| 180 | + | if err != nil { | |||
| 181 | + | return err | |||
| 182 | + | } | |||
| 183 | + | if err := syscall.Kill(pid, syscall.SIGKILL); err != nil && !errors.Is(err, syscall.ESRCH) { | |||
| 184 | + | return err | |||
| 185 | + | } | |||
| 186 | + | return waitForProcessExit(pid, 2*time.Second) | |||
| 187 | + | } | |||
| 188 | + | ||||
| 19 | // fastBackoff keeps the retry loop snappy for tests. | 189 | // fastBackoff keeps the retry loop snappy for tests. | ||
| 20 | var fastBackoff = backoffConfig{initial: time.Millisecond, max: 5 * time.Millisecond, total: 100 * time.Millisecond} | 190 | var fastBackoff = backoffConfig{initial: time.Millisecond, max: 5 * time.Millisecond, total: 100 * time.Millisecond} | ||
| 21 | 191 | ||||