1 // Copyright (C) The Arvados Authors. All rights reserved.
3 // SPDX-License-Identifier: AGPL-3.0
29 "git.arvados.org/arvados.git/lib/config"
30 "git.arvados.org/arvados.git/lib/service"
31 "git.arvados.org/arvados.git/sdk/go/arvados"
32 "git.arvados.org/arvados.git/sdk/go/ctxlog"
33 "git.arvados.org/arvados.git/sdk/go/health"
34 "github.com/fsnotify/fsnotify"
35 "github.com/sirupsen/logrus"
38 type Supervisor struct {
39 SourcePath string // e.g., /home/username/src/arvados
40 SourceVersion string // e.g., acbd1324...
41 ClusterType string // e.g., production
42 ListenHost string // e.g., localhost
43 ControllerAddr string // e.g., 127.0.0.1:8000
44 OwnTemporaryDatabase bool
47 logger logrus.FieldLogger
48 cluster *arvados.Cluster
51 cancel context.CancelFunc
52 done chan struct{} // closed when child procs/services have shut down
53 err error // error that caused shutdown (valid when done is closed)
54 healthChecker *health.Aggregator
55 tasksReady map[string]chan bool
56 waitShutdown sync.WaitGroup
62 environ []string // for child processes
65 func (super *Supervisor) Start(ctx context.Context, cfg *arvados.Config, cfgPath string) {
66 super.ctx, super.cancel = context.WithCancel(ctx)
67 super.done = make(chan struct{})
70 defer close(super.done)
72 sigch := make(chan os.Signal)
73 signal.Notify(sigch, syscall.SIGINT, syscall.SIGTERM)
74 defer signal.Stop(sigch)
76 for sig := range sigch {
77 super.logger.WithField("signal", sig).Info("caught signal")
79 super.err = fmt.Errorf("caught signal %s", sig)
85 hupch := make(chan os.Signal)
86 signal.Notify(hupch, syscall.SIGHUP)
87 defer signal.Stop(hupch)
89 for sig := range hupch {
90 super.logger.WithField("signal", sig).Info("caught signal")
92 super.err = errNeedConfigReload
98 if cfgPath != "" && cfgPath != "-" && cfg.AutoReloadConfig {
99 go watchConfig(super.ctx, super.logger, cfgPath, copyConfig(cfg), func() {
100 if super.err == nil {
101 super.err = errNeedConfigReload
107 err := super.run(cfg)
109 super.logger.WithError(err).Warn("supervisor shut down")
110 if super.err == nil {
117 func (super *Supervisor) Wait() error {
122 func (super *Supervisor) run(cfg *arvados.Config) error {
125 cwd, err := os.Getwd()
129 if !strings.HasPrefix(super.SourcePath, "/") {
130 super.SourcePath = filepath.Join(cwd, super.SourcePath)
132 super.SourcePath, err = filepath.EvalSymlinks(super.SourcePath)
137 // Choose bin and temp dirs: /var/lib/arvados/... in
138 // production, transient tempdir otherwise.
139 if super.ClusterType == "production" {
140 // These dirs have already been created by
141 // "arvados-server install" (or by extracting a
143 super.tempdir = "/var/lib/arvados/tmp"
144 super.wwwtempdir = "/var/lib/arvados/wwwtmp"
145 super.bindir = "/var/lib/arvados/bin"
146 super.configfile = "/etc/arvados/config.yml"
148 super.tempdir, err = ioutil.TempDir("", "arvados-server-boot-")
152 defer os.RemoveAll(super.tempdir)
153 super.wwwtempdir = super.tempdir
154 super.bindir = filepath.Join(super.tempdir, "bin")
155 if err := os.Mkdir(super.bindir, 0755); err != nil {
159 // Fill in any missing config keys, and write the resulting
160 // config in the temp dir for child services to use.
161 err = super.autofillConfig(cfg)
165 conffile, err := os.OpenFile(filepath.Join(super.tempdir, "config.yml"), os.O_CREATE|os.O_WRONLY, 0644)
169 defer conffile.Close()
170 err = json.NewEncoder(conffile).Encode(cfg)
174 err = conffile.Close()
178 super.configfile = conffile.Name()
181 super.environ = os.Environ()
182 super.cleanEnv([]string{"ARVADOS_"})
183 super.setEnv("ARVADOS_CONFIG", super.configfile)
184 super.setEnv("RAILS_ENV", super.ClusterType)
185 super.setEnv("TMPDIR", super.tempdir)
186 super.prependEnv("PATH", "/var/lib/arvados/bin:")
187 if super.ClusterType != "production" {
188 super.prependEnv("PATH", super.tempdir+"/bin:")
191 super.cluster, err = cfg.GetCluster("")
195 // Now that we have the config, replace the bootstrap logger
196 // with a new one according to the logging config.
197 loglevel := super.cluster.SystemLogs.LogLevel
198 if s := os.Getenv("ARVADOS_DEBUG"); s != "" && s != "0" {
201 super.logger = ctxlog.New(super.Stderr, super.cluster.SystemLogs.Format, loglevel).WithFields(logrus.Fields{
205 if super.SourceVersion == "" && super.ClusterType == "production" {
206 // don't need SourceVersion
207 } else if super.SourceVersion == "" {
208 // Find current source tree version.
210 err = super.RunProgram(super.ctx, ".", &buf, nil, "git", "diff", "--shortstat")
214 dirty := buf.Len() > 0
216 err = super.RunProgram(super.ctx, ".", &buf, nil, "git", "log", "-n1", "--format=%H")
220 super.SourceVersion = strings.TrimSpace(buf.String())
222 super.SourceVersion += "+uncommitted"
225 return errors.New("specifying a version to run is not yet supported")
228 _, err = super.installGoProgram(super.ctx, "cmd/arvados-server")
232 err = super.setupRubyEnv()
237 tasks := []supervisedTask{
238 createCertificates{},
241 runServiceCommand{name: "controller", svc: super.cluster.Services.Controller, depends: []supervisedTask{runPostgreSQL{}}},
242 runGoProgram{src: "services/arv-git-httpd", svc: super.cluster.Services.GitHTTP},
243 runGoProgram{src: "services/health", svc: super.cluster.Services.Health},
244 runGoProgram{src: "services/keepproxy", svc: super.cluster.Services.Keepproxy, depends: []supervisedTask{runPassenger{src: "services/api"}}},
245 runGoProgram{src: "services/keepstore", svc: super.cluster.Services.Keepstore},
246 runGoProgram{src: "services/keep-web", svc: super.cluster.Services.WebDAV},
247 runServiceCommand{name: "ws", svc: super.cluster.Services.Websocket, depends: []supervisedTask{runPostgreSQL{}}},
248 installPassenger{src: "services/api"},
249 runPassenger{src: "services/api", varlibdir: "railsapi", svc: super.cluster.Services.RailsAPI, depends: []supervisedTask{createCertificates{}, runPostgreSQL{}, installPassenger{src: "services/api"}}},
250 installPassenger{src: "apps/workbench", depends: []supervisedTask{installPassenger{src: "services/api"}}}, // dependency ensures workbench doesn't delay api startup
251 runPassenger{src: "apps/workbench", varlibdir: "workbench1", svc: super.cluster.Services.Workbench1, depends: []supervisedTask{installPassenger{src: "apps/workbench"}}},
254 if super.ClusterType != "test" {
255 tasks = append(tasks,
256 runServiceCommand{name: "dispatch-cloud", svc: super.cluster.Services.DispatchCloud},
257 runGoProgram{src: "services/keep-balance", svc: super.cluster.Services.Keepbalance},
260 super.tasksReady = map[string]chan bool{}
261 for _, task := range tasks {
262 super.tasksReady[task.String()] = make(chan bool)
264 for _, task := range tasks {
266 fail := func(err error) {
267 if super.ctx.Err() != nil {
271 super.logger.WithField("task", task.String()).WithError(err).Error("task failed")
274 super.logger.WithField("task", task.String()).Info("starting")
275 err := task.Run(super.ctx, fail, super)
280 close(super.tasksReady[task.String()])
283 err = super.wait(super.ctx, tasks...)
287 super.logger.Info("all startup tasks are complete; starting health checks")
288 super.healthChecker = &health.Aggregator{Cluster: super.cluster}
290 super.logger.Info("shutting down")
291 super.waitShutdown.Wait()
292 return super.ctx.Err()
295 func (super *Supervisor) wait(ctx context.Context, tasks ...supervisedTask) error {
296 for _, task := range tasks {
297 ch, ok := super.tasksReady[task.String()]
299 return fmt.Errorf("no such task: %s", task)
301 super.logger.WithField("task", task.String()).Info("waiting")
304 super.logger.WithField("task", task.String()).Info("ready")
306 super.logger.WithField("task", task.String()).Info("task was never ready")
313 func (super *Supervisor) Stop() {
318 func (super *Supervisor) WaitReady() (*arvados.URL, bool) {
319 ticker := time.NewTicker(time.Second)
321 for waiting := "all"; waiting != ""; {
324 case <-super.ctx.Done():
327 if super.healthChecker == nil {
331 resp := super.healthChecker.ClusterHealth()
332 // The overall health check (resp.Health=="OK") might
333 // never pass due to missing components (like
334 // arvados-dispatch-cloud in a test cluster), so
335 // instead we wait for all configured components to
338 for target, check := range resp.Checks {
339 if check.Health != "OK" {
340 waiting += " " + target
344 super.logger.WithField("targets", waiting[1:]).Info("waiting")
347 u := super.cluster.Services.Controller.ExternalURL
351 func (super *Supervisor) prependEnv(key, prepend string) {
352 for i, s := range super.environ {
353 if strings.HasPrefix(s, key+"=") {
354 super.environ[i] = key + "=" + prepend + s[len(key)+1:]
358 super.environ = append(super.environ, key+"="+prepend)
361 func (super *Supervisor) cleanEnv(prefixes []string) {
363 for _, s := range super.environ {
365 for _, p := range prefixes {
366 if strings.HasPrefix(s, p) {
372 cleaned = append(cleaned, s)
375 super.environ = cleaned
378 func (super *Supervisor) setEnv(key, val string) {
379 for i, s := range super.environ {
380 if strings.HasPrefix(s, key+"=") {
381 super.environ[i] = key + "=" + val
385 super.environ = append(super.environ, key+"="+val)
388 // Remove all but the first occurrence of each env var.
389 func dedupEnv(in []string) []string {
390 saw := map[string]bool{}
392 for _, kv := range in {
393 if split := strings.Index(kv, "="); split < 1 {
394 panic("invalid environment var: " + kv)
395 } else if saw[kv[:split]] {
398 saw[kv[:split]] = true
399 out = append(out, kv)
405 func (super *Supervisor) installGoProgram(ctx context.Context, srcpath string) (string, error) {
406 _, basename := filepath.Split(srcpath)
407 binfile := filepath.Join(super.bindir, basename)
408 if super.ClusterType == "production" {
411 err := super.RunProgram(ctx, filepath.Join(super.SourcePath, srcpath), nil, []string{"GOBIN=" + super.bindir}, "go", "install", "-ldflags", "-X git.arvados.org/arvados.git/lib/cmd.version="+super.SourceVersion+" -X main.version="+super.SourceVersion)
415 func (super *Supervisor) usingRVM() bool {
416 return os.Getenv("rvm_path") != ""
419 func (super *Supervisor) setupRubyEnv() error {
420 if !super.usingRVM() {
421 // (If rvm is in use, assume the caller has everything
422 // set up as desired)
423 super.cleanEnv([]string{
428 if _, err := os.Stat("/var/lib/arvados/bin/gem"); err == nil || super.ClusterType == "production" {
429 gem = "/var/lib/arvados/bin/gem"
431 cmd := exec.Command(gem, "env", "gempath")
432 if super.ClusterType == "production" {
433 cmd.Args = append([]string{"sudo", "-u", "www-data", "-E", "HOME=/var/www"}, cmd.Args...)
434 path, err := exec.LookPath("sudo")
436 return fmt.Errorf("LookPath(\"sudo\"): %w", err)
440 cmd.Stderr = super.Stderr
441 cmd.Env = super.environ
442 buf, err := cmd.Output() // /var/lib/arvados/.gem/ruby/2.5.0/bin:...
443 if err != nil || len(buf) == 0 {
444 return fmt.Errorf("gem env gempath: %v", err)
446 gempath := string(bytes.Split(buf, []byte{':'})[0])
447 super.prependEnv("PATH", gempath+"/bin:")
448 super.setEnv("GEM_HOME", gempath)
449 super.setEnv("GEM_PATH", gempath)
451 // Passenger install doesn't work unless $HOME is ~user
452 u, err := user.Current()
456 super.setEnv("HOME", u.HomeDir)
460 func (super *Supervisor) lookPath(prog string) string {
461 for _, val := range super.environ {
462 if strings.HasPrefix(val, "PATH=") {
463 for _, dir := range filepath.SplitList(val[5:]) {
464 path := filepath.Join(dir, prog)
465 if fi, err := os.Stat(path); err == nil && fi.Mode()&0111 != 0 {
474 // Run prog with args, using dir as working directory. If ctx is
475 // cancelled while the child is running, RunProgram terminates the
476 // child, waits for it to exit, then returns.
478 // Child's environment will have our env vars, plus any given in env.
480 // Child's stdout will be written to output if non-nil, otherwise the
481 // boot command's stderr.
482 func (super *Supervisor) RunProgram(ctx context.Context, dir string, output io.Writer, env []string, prog string, args ...string) error {
483 cmdline := fmt.Sprintf("%s", append([]string{prog}, args...))
484 super.logger.WithField("command", cmdline).WithField("dir", dir).Info("executing")
488 if logprefix == "setuidgid" && len(args) >= 3 {
492 if logprefix == "sudo" {
493 for i := 0; i < len(args); i++ {
496 } else if args[i] == "-E" || strings.Contains(args[i], "=") {
499 innerargs = args[i+1:]
504 logprefix = strings.TrimPrefix(logprefix, "/var/lib/arvados/bin/")
505 logprefix = strings.TrimPrefix(logprefix, super.tempdir+"/bin/")
506 if logprefix == "bundle" && len(innerargs) > 2 && innerargs[0] == "exec" {
507 _, dirbase := filepath.Split(dir)
508 logprefix = innerargs[1] + "@" + dirbase
509 } else if logprefix == "arvados-server" && len(args) > 1 {
512 if !strings.HasPrefix(dir, "/") {
513 logprefix = dir + ": " + logprefix
517 cmd := exec.Command(super.lookPath(prog), args...)
518 stdout, err := cmd.StdoutPipe()
522 stderr, err := cmd.StderrPipe()
526 logwriter := &service.LogPrefixer{Writer: super.Stderr, Prefix: []byte("[" + logprefix + "] ")}
527 var copiers sync.WaitGroup
530 io.Copy(logwriter, stderr)
536 io.Copy(logwriter, stdout)
538 io.Copy(output, stdout)
543 if strings.HasPrefix(dir, "/") {
546 cmd.Dir = filepath.Join(super.SourcePath, dir)
548 env = append([]string(nil), env...)
549 env = append(env, super.environ...)
550 cmd.Env = dedupEnv(env)
553 defer func() { exited = true }()
556 log := ctxlog.FromContext(ctx).WithFields(logrus.Fields{"dir": dir, "cmdline": cmdline})
558 if cmd.Process == nil {
559 log.Debug("waiting for child process to start")
560 time.Sleep(time.Second / 2)
562 log.WithField("PID", cmd.Process.Pid).Debug("sending SIGTERM")
563 cmd.Process.Signal(syscall.SIGTERM)
564 time.Sleep(5 * time.Second)
568 log.WithField("PID", cmd.Process.Pid).Warn("still waiting for child process to exit 5s after SIGTERM")
580 if ctx.Err() != nil {
581 // Return "context canceled", instead of the "killed"
582 // error that was probably caused by the context being
585 } else if err != nil {
586 return fmt.Errorf("%s: error: %v", cmdline, err)
591 func (super *Supervisor) autofillConfig(cfg *arvados.Config) error {
592 cluster, err := cfg.GetCluster("")
596 usedPort := map[string]bool{}
597 nextPort := func(host string) string {
599 port, err := availablePort(host)
606 usedPort[port] = true
610 if cluster.Services.Controller.ExternalURL.Host == "" {
611 h, p, err := net.SplitHostPort(super.ControllerAddr)
621 cluster.Services.Controller.ExternalURL = arvados.URL{Scheme: "https", Host: net.JoinHostPort(h, p), Path: "/"}
623 for _, svc := range []*arvados.Service{
624 &cluster.Services.Controller,
625 &cluster.Services.DispatchCloud,
626 &cluster.Services.GitHTTP,
627 &cluster.Services.Health,
628 &cluster.Services.Keepproxy,
629 &cluster.Services.Keepstore,
630 &cluster.Services.RailsAPI,
631 &cluster.Services.WebDAV,
632 &cluster.Services.WebDAVDownload,
633 &cluster.Services.Websocket,
634 &cluster.Services.Workbench1,
636 if svc == &cluster.Services.DispatchCloud && super.ClusterType == "test" {
639 if svc.ExternalURL.Host == "" {
640 if svc == &cluster.Services.Controller ||
641 svc == &cluster.Services.GitHTTP ||
642 svc == &cluster.Services.Health ||
643 svc == &cluster.Services.Keepproxy ||
644 svc == &cluster.Services.WebDAV ||
645 svc == &cluster.Services.WebDAVDownload ||
646 svc == &cluster.Services.Workbench1 {
647 svc.ExternalURL = arvados.URL{Scheme: "https", Host: fmt.Sprintf("%s:%s", super.ListenHost, nextPort(super.ListenHost)), Path: "/"}
648 } else if svc == &cluster.Services.Websocket {
649 svc.ExternalURL = arvados.URL{Scheme: "wss", Host: fmt.Sprintf("%s:%s", super.ListenHost, nextPort(super.ListenHost)), Path: "/websocket"}
652 if len(svc.InternalURLs) == 0 {
653 svc.InternalURLs = map[arvados.URL]arvados.ServiceInstance{
654 arvados.URL{Scheme: "http", Host: fmt.Sprintf("%s:%s", super.ListenHost, nextPort(super.ListenHost)), Path: "/"}: arvados.ServiceInstance{},
658 if cluster.SystemRootToken == "" {
659 cluster.SystemRootToken = randomHexString(64)
661 if cluster.ManagementToken == "" {
662 cluster.ManagementToken = randomHexString(64)
664 if cluster.API.RailsSessionSecretToken == "" {
665 cluster.API.RailsSessionSecretToken = randomHexString(64)
667 if cluster.Collections.BlobSigningKey == "" {
668 cluster.Collections.BlobSigningKey = randomHexString(64)
670 if super.ClusterType != "production" && cluster.Containers.DispatchPrivateKey == "" {
671 buf, err := ioutil.ReadFile(filepath.Join(super.SourcePath, "lib", "dispatchcloud", "test", "sshkey_dispatch"))
675 cluster.Containers.DispatchPrivateKey = string(buf)
677 if super.ClusterType != "production" {
678 cluster.TLS.Insecure = true
680 if super.ClusterType == "test" {
681 // Add a second keepstore process.
682 cluster.Services.Keepstore.InternalURLs[arvados.URL{Scheme: "http", Host: fmt.Sprintf("%s:%s", super.ListenHost, nextPort(super.ListenHost)), Path: "/"}] = arvados.ServiceInstance{}
684 // Create a directory-backed volume for each keepstore
686 cluster.Volumes = map[string]arvados.Volume{}
687 for url := range cluster.Services.Keepstore.InternalURLs {
688 volnum := len(cluster.Volumes)
689 datadir := fmt.Sprintf("%s/keep%d.data", super.tempdir, volnum)
690 if _, err = os.Stat(datadir + "/."); err == nil {
691 } else if !os.IsNotExist(err) {
693 } else if err = os.Mkdir(datadir, 0755); err != nil {
696 cluster.Volumes[fmt.Sprintf(cluster.ClusterID+"-nyw5e-%015d", volnum)] = arvados.Volume{
698 DriverParameters: json.RawMessage(fmt.Sprintf(`{"Root":%q}`, datadir)),
699 AccessViaHosts: map[arvados.URL]arvados.VolumeAccess{
705 if super.OwnTemporaryDatabase {
706 cluster.PostgreSQL.Connection = arvados.PostgreSQLConnection{
707 "client_encoding": "utf8",
709 "port": nextPort(super.ListenHost),
710 "dbname": "arvados_test",
712 "password": "insecure_arvados_test",
716 cfg.Clusters[cluster.ClusterID] = *cluster
720 func addrIsLocal(addr string) (bool, error) {
722 listener, err := net.Listen("tcp", addr)
726 } else if strings.Contains(err.Error(), "cannot assign requested address") {
733 func randomHexString(chars int) string {
734 b := make([]byte, chars/2)
735 _, err := rand.Read(b)
739 return fmt.Sprintf("%x", b)
742 func internalPort(svc arvados.Service) (string, error) {
743 if len(svc.InternalURLs) > 1 {
744 return "", errors.New("internalPort() doesn't work with multiple InternalURLs")
746 for u := range svc.InternalURLs {
748 if p := u.Port(); p != "" {
750 } else if u.Scheme == "https" || u.Scheme == "ws" {
756 return "", fmt.Errorf("service has no InternalURLs")
759 func externalPort(svc arvados.Service) (string, error) {
760 u := url.URL(svc.ExternalURL)
761 if p := u.Port(); p != "" {
763 } else if u.Scheme == "https" || u.Scheme == "wss" {
770 func availablePort(host string) (string, error) {
771 ln, err := net.Listen("tcp", net.JoinHostPort(host, "0"))
776 _, port, err := net.SplitHostPort(ln.Addr().String())
783 // Try to connect to addr until it works, then close ch. Give up if
785 func waitForConnect(ctx context.Context, addr string) error {
786 dialer := net.Dialer{Timeout: time.Second}
787 for ctx.Err() == nil {
788 conn, err := dialer.DialContext(ctx, "tcp", addr)
790 time.Sleep(time.Second / 10)
799 func copyConfig(cfg *arvados.Config) *arvados.Config {
802 err := json.NewEncoder(pw).Encode(cfg)
808 cfg2 := new(arvados.Config)
809 err := json.NewDecoder(pr).Decode(cfg2)
816 func watchConfig(ctx context.Context, logger logrus.FieldLogger, cfgPath string, prevcfg *arvados.Config, fn func()) {
817 watcher, err := fsnotify.NewWatcher()
819 logger.WithError(err).Error("fsnotify setup failed")
822 defer watcher.Close()
824 err = watcher.Add(cfgPath)
826 logger.WithError(err).Error("fsnotify watcher failed")
834 case err, ok := <-watcher.Errors:
838 logger.WithError(err).Warn("fsnotify watcher reported error")
839 case _, ok := <-watcher.Events:
843 for len(watcher.Events) > 0 {
846 loader := config.NewLoader(&bytes.Buffer{}, &logrus.Logger{Out: ioutil.Discard})
847 loader.Path = cfgPath
848 loader.SkipAPICalls = true
849 cfg, err := loader.Load()
851 logger.WithError(err).Warn("error reloading config file after change detected; ignoring new config for now")
852 } else if reflect.DeepEqual(cfg, prevcfg) {
853 logger.Debug("config file changed but is still DeepEqual to the existing config")
855 logger.Debug("config changed, notifying supervisor")