1 // Copyright (C) The Arvados Authors. All rights reserved.
3 // SPDX-License-Identifier: AGPL-3.0
22 lockprefix = "crunch-run-"
26 // procinfo is saved in each process's lockfile.
27 type procinfo struct {
34 // Detach acquires a lock for the given uuid, and starts the current
35 // program as a child process (with -nodetach prepended to the given
36 // arguments so the child knows not to detach again). The lock is
37 // passed along to the child process.
38 func Detach(uuid string, args []string, stdout, stderr io.Writer) int {
39 return exitcode(stderr, detach(uuid, args, stdout, stderr))
41 func detach(uuid string, args []string, stdout, stderr io.Writer) error {
42 lockfile, err := os.OpenFile(filepath.Join(lockdir, lockprefix+uuid+locksuffix), os.O_CREATE|os.O_RDWR, 0700)
46 defer lockfile.Close()
47 err = syscall.Flock(int(lockfile.Fd()), syscall.LOCK_EX|syscall.LOCK_NB)
53 outfile, err := ioutil.TempFile("", "crunch-run-"+uuid+"-stdout-")
58 errfile, err := ioutil.TempFile("", "crunch-run-"+uuid+"-stderr-")
60 os.Remove(outfile.Name())
65 cmd := exec.Command(args[0], append([]string{"-nodetach"}, args[1:]...)...)
68 cmd.ExtraFiles = []*os.File{lockfile}
71 os.Remove(outfile.Name())
72 os.Remove(errfile.Name())
76 w := io.MultiWriter(stdout, lockfile)
77 err = json.NewEncoder(w).Encode(procinfo{
79 Stdout: outfile.Name(),
80 Stderr: errfile.Name(),
83 os.Remove(outfile.Name())
84 os.Remove(errfile.Name())
90 // KillProcess finds the crunch-run process corresponding to the given
91 // uuid, and sends the given signal to it. It then waits up to 1
92 // second for the process to die. It returns 0 if the process is
93 // successfully killed or didn't exist in the first place.
94 func KillProcess(uuid string, signal syscall.Signal, stdout, stderr io.Writer) int {
95 return exitcode(stderr, kill(uuid, signal, stdout, stderr))
98 func kill(uuid string, signal syscall.Signal, stdout, stderr io.Writer) error {
99 path := filepath.Join(lockdir, lockprefix+uuid+locksuffix)
100 f, err := os.Open(path)
101 if os.IsNotExist(err) {
103 } else if err != nil {
109 err = json.NewDecoder(f).Decode(&pi)
111 return fmt.Errorf("%s: %s\n", path, err)
114 if pi.UUID != uuid || pi.PID == 0 {
115 return fmt.Errorf("%s: bogus procinfo: %+v", path, pi)
118 proc, err := os.FindProcess(pi.PID)
123 err = proc.Signal(signal)
124 for deadline := time.Now().Add(time.Second); err == nil && time.Now().Before(deadline); time.Sleep(time.Second / 100) {
125 err = proc.Signal(syscall.Signal(0))
128 return fmt.Errorf("pid %d: sent signal %d (%s) but process is still alive", pi.PID, signal, signal)
130 fmt.Fprintf(stderr, "pid %d: %s\n", pi.PID, err)
134 // List UUIDs of active crunch-run processes.
135 func ListProcesses(stdout, stderr io.Writer) int {
136 return exitcode(stderr, filepath.Walk(lockdir, func(path string, info os.FileInfo, err error) error {
138 return filepath.SkipDir
140 if name := info.Name(); !strings.HasPrefix(name, lockprefix) || !strings.HasSuffix(name, locksuffix) {
143 if info.Size() == 0 {
144 // race: process has opened/locked but hasn't yet written pid/uuid
148 f, err := os.Open(path)
154 // TODO: Do this check without risk of disrupting lock
155 // acquisition during races, e.g., by connecting to a
156 // unix socket or checking /proc/$pid/fd/$n ->
158 err = syscall.Flock(int(f.Fd()), syscall.LOCK_SH)
161 err := os.Remove(path)
163 fmt.Fprintln(stderr, err)
169 err = json.NewDecoder(f).Decode(&pi)
171 fmt.Fprintf(stderr, "%s: %s\n", path, err)
174 if pi.UUID == "" || pi.PID == 0 {
175 fmt.Fprintf(stderr, "%s: bogus procinfo: %+v", path, pi)
179 fmt.Fprintln(stdout, pi.UUID)
184 // If err is nil, return 0 ("success"); otherwise, print err to stderr
186 func exitcode(stderr io.Writer, err error) int {
188 fmt.Fprintln(stderr, err)