1 // Copyright (C) The Arvados Authors. All rights reserved.
3 // SPDX-License-Identifier: AGPL-3.0
30 "git.arvados.org/arvados.git/lib/config"
31 "git.arvados.org/arvados.git/lib/service"
32 "git.arvados.org/arvados.git/sdk/go/arvados"
33 "git.arvados.org/arvados.git/sdk/go/ctxlog"
34 "git.arvados.org/arvados.git/sdk/go/health"
35 "github.com/fsnotify/fsnotify"
36 "github.com/sirupsen/logrus"
39 type Supervisor struct {
40 // Config file location like "/etc/arvados/config.yml", or "-"
41 // to read from Stdin (see below).
43 // Literal config file (useful for test suites). If non-empty,
44 // this is used instead of ConfigPath.
46 // Path to arvados source tree. Only used for dev/test
49 // Version number to build into binaries. Only used for
52 // "production", "development", or "test".
54 // Listening address for external services, and internal
55 // services whose InternalURLs are not explicitly configured.
56 // If blank, listen on the configured controller ExternalURL
57 // host; if that is also blank, listen on all addresses
60 // Default host:port for controller ExternalURL if not
61 // explicitly configured in config file. If blank, use a
62 // random port on ListenHost.
67 OwnTemporaryDatabase bool
71 logger logrus.FieldLogger
72 cluster *arvados.Cluster // nil if this is a multi-cluster supervisor
73 children map[string]*Supervisor // nil if this is a single-cluster supervisor
76 cancel context.CancelFunc
77 done chan struct{} // closed when child procs/services have shut down
78 err error // error that caused shutdown (valid when done is closed)
79 healthChecker *health.Aggregator // nil if this is a multi-cluster supervisor, or still booting
80 tasksReady map[string]chan bool
81 waitShutdown sync.WaitGroup
84 tempdir string // in production mode, this is accessible only to root
85 wwwtempdir string // in production mode, this is accessible only to www-data
87 environ []string // for child processes
90 func (super *Supervisor) Clusters() map[string]*arvados.Cluster {
91 m := map[string]*arvados.Cluster{}
92 if super.cluster != nil {
93 m[super.cluster.ClusterID] = super.cluster
95 for id, super2 := range super.children {
96 m[id] = super2.Cluster("")
101 func (super *Supervisor) Cluster(id string) *arvados.Cluster {
102 if super.children != nil {
103 return super.children[id].Cluster(id)
109 func (super *Supervisor) Start(ctx context.Context) {
110 super.logger = ctxlog.FromContext(ctx)
111 super.ctx, super.cancel = context.WithCancel(ctx)
112 super.done = make(chan struct{})
114 sigch := make(chan os.Signal, 1)
115 signal.Notify(sigch, syscall.SIGINT, syscall.SIGTERM, syscall.SIGHUP)
117 defer signal.Stop(sigch)
123 super.logger.WithField("signal", sig).Info("caught signal")
124 if super.err == nil {
125 if sig == syscall.SIGHUP {
126 super.err = errNeedConfigReload
128 super.err = fmt.Errorf("caught signal %s", sig)
136 loaderStdin := super.Stdin
137 if super.ConfigYAML != "" {
138 loaderStdin = bytes.NewBufferString(super.ConfigYAML)
140 loader := config.NewLoader(loaderStdin, super.logger)
141 loader.SkipLegacy = true
142 loader.SkipAPICalls = true
143 loader.Path = super.ConfigPath
144 if super.ConfigYAML != "" {
147 cfg, err := loader.Load()
155 if super.ConfigPath != "" && super.ConfigPath != "-" && cfg.AutoReloadConfig {
156 go watchConfig(super.ctx, super.logger, super.ConfigPath, copyConfig(cfg), func() {
157 if super.err == nil {
158 super.err = errNeedConfigReload
164 if len(cfg.Clusters) > 1 {
165 super.startFederation(cfg)
168 defer close(super.done)
169 for _, super2 := range super.children {
171 if super.err == nil {
179 defer close(super.done)
180 super.cluster, super.err = cfg.GetCluster("")
181 if super.err != nil {
184 err := super.runCluster()
186 super.logger.WithError(err).Info("supervisor shut down")
187 if super.err == nil {
195 // Wait returns when all child processes and goroutines have exited.
196 func (super *Supervisor) Wait() error {
201 // startFederation starts a child Supervisor for each cluster in the
202 // given config. Each is a copy of the original/parent with the
203 // original config reduced to a single cluster.
204 func (super *Supervisor) startFederation(cfg *arvados.Config) {
205 super.children = map[string]*Supervisor{}
206 for id, cc := range cfg.Clusters {
207 yaml, err := json.Marshal(arvados.Config{Clusters: map[string]arvados.Cluster{id: cc}})
209 panic(fmt.Sprintf("json.Marshal partial config: %s", err))
211 super2 := &Supervisor{
213 ConfigYAML: string(yaml),
214 SourcePath: super.SourcePath,
215 SourceVersion: super.SourceVersion,
216 ClusterType: super.ClusterType,
217 ListenHost: super.ListenHost,
218 ControllerAddr: super.ControllerAddr,
219 NoWorkbench1: super.NoWorkbench1,
220 NoWorkbench2: super.NoWorkbench2,
221 OwnTemporaryDatabase: super.OwnTemporaryDatabase,
223 Stderr: super.Stderr,
225 if super2.ClusterType == "test" {
226 super2.Stderr = &service.LogPrefixer{
227 Writer: super.Stderr,
228 Prefix: []byte("[" + id + "] "),
231 super2.Start(super.ctx)
232 super.children[id] = super2
236 func (super *Supervisor) runCluster() error {
237 cwd, err := os.Getwd()
241 if super.ClusterType == "test" && super.SourcePath == "" {
242 // When invoked by test suite, default to current
244 buf, err := exec.Command("git", "rev-parse", "--show-toplevel").CombinedOutput()
246 return fmt.Errorf("git rev-parse: %w", err)
248 super.SourcePath = strings.TrimSuffix(string(buf), "\n")
249 } else if !strings.HasPrefix(super.SourcePath, "/") {
250 super.SourcePath = filepath.Join(cwd, super.SourcePath)
252 super.SourcePath, err = filepath.EvalSymlinks(super.SourcePath)
257 if super.ListenHost == "" {
258 u := url.URL(super.cluster.Services.Controller.ExternalURL)
259 super.ListenHost = u.Hostname()
260 if super.ListenHost == "" {
261 super.ListenHost = "0.0.0.0"
265 // Choose bin and temp dirs: /var/lib/arvados/... in
266 // production, transient tempdir otherwise.
267 if super.ClusterType == "production" {
268 // These dirs have already been created by
269 // "arvados-server install" (or by extracting a
271 super.tempdir = "/var/lib/arvados/tmp"
272 super.wwwtempdir = "/var/lib/arvados/wwwtmp"
273 super.bindir = "/var/lib/arvados/bin"
275 super.tempdir, err = ioutil.TempDir("", "arvados-server-boot-")
279 defer os.RemoveAll(super.tempdir)
280 super.wwwtempdir = super.tempdir
281 super.bindir = filepath.Join(super.tempdir, "bin")
282 if err := os.Mkdir(super.bindir, 0755); err != nil {
287 // Fill in any missing config keys, and write the resulting
288 // config in the temp dir for child services to use.
289 err = super.autofillConfig()
293 conffile, err := os.OpenFile(filepath.Join(super.wwwtempdir, "config.yml"), os.O_CREATE|os.O_WRONLY|os.O_TRUNC, 0644)
297 defer conffile.Close()
298 err = json.NewEncoder(conffile).Encode(arvados.Config{
299 Clusters: map[string]arvados.Cluster{
300 super.cluster.ClusterID: *super.cluster}})
304 err = conffile.Close()
308 super.configfile = conffile.Name()
310 super.environ = os.Environ()
311 super.cleanEnv([]string{"ARVADOS_"})
312 super.setEnv("ARVADOS_CONFIG", super.configfile)
313 super.setEnv("RAILS_ENV", super.ClusterType)
314 super.setEnv("TMPDIR", super.tempdir)
315 super.prependEnv("PATH", "/var/lib/arvados/bin:")
316 if super.ClusterType != "production" {
317 super.prependEnv("PATH", super.tempdir+"/bin:")
319 super.setEnv("ARVADOS_SERVER_ADDRESS", super.ListenHost)
321 // Now that we have the config, replace the bootstrap logger
322 // with a new one according to the logging config.
323 loglevel := super.cluster.SystemLogs.LogLevel
324 if s := os.Getenv("ARVADOS_DEBUG"); s != "" && s != "0" {
327 super.logger = ctxlog.New(super.Stderr, super.cluster.SystemLogs.Format, loglevel).WithFields(logrus.Fields{
331 if super.SourceVersion == "" && super.ClusterType == "production" {
332 // don't need SourceVersion
333 } else if super.SourceVersion == "" {
334 // Find current source tree version.
336 err = super.RunProgram(super.ctx, super.SourcePath, runOptions{output: &buf}, "git", "diff", "--shortstat")
340 dirty := buf.Len() > 0
342 err = super.RunProgram(super.ctx, super.SourcePath, runOptions{output: &buf}, "git", "log", "-n1", "--format=%H")
346 super.SourceVersion = strings.TrimSpace(buf.String())
348 super.SourceVersion += "+uncommitted"
351 return errors.New("specifying a version to run is not yet supported")
354 _, err = super.installGoProgram(super.ctx, "cmd/arvados-server")
358 err = super.setupRubyEnv()
363 tasks := []supervisedTask{
364 createCertificates{},
368 runServiceCommand{name: "controller", svc: super.cluster.Services.Controller, depends: []supervisedTask{railsDatabase{}}},
369 runServiceCommand{name: "health", svc: super.cluster.Services.Health},
370 runServiceCommand{name: "keepproxy", svc: super.cluster.Services.Keepproxy, depends: []supervisedTask{runPassenger{src: "services/api"}}},
371 runServiceCommand{name: "keepstore", svc: super.cluster.Services.Keepstore},
372 runServiceCommand{name: "keep-web", svc: super.cluster.Services.WebDAV},
373 runServiceCommand{name: "ws", svc: super.cluster.Services.Websocket, depends: []supervisedTask{railsDatabase{}}},
374 installPassenger{src: "services/api", varlibdir: "railsapi"},
375 runPassenger{src: "services/api", varlibdir: "railsapi", svc: super.cluster.Services.RailsAPI, depends: []supervisedTask{
376 createCertificates{},
377 installPassenger{src: "services/api", varlibdir: "railsapi"},
381 if !super.NoWorkbench1 {
382 return errors.New("workbench1 is no longer supported")
384 if !super.NoWorkbench2 {
385 tasks = append(tasks,
386 runWorkbench2{svc: super.cluster.Services.Workbench2},
389 if super.ClusterType != "test" {
390 tasks = append(tasks,
391 runServiceCommand{name: "keep-balance", svc: super.cluster.Services.Keepbalance},
394 if super.cluster.Containers.CloudVMs.Enable {
395 tasks = append(tasks,
396 runServiceCommand{name: "dispatch-cloud", svc: super.cluster.Services.DispatchCloud},
399 super.tasksReady = map[string]chan bool{}
400 for _, task := range tasks {
401 super.tasksReady[task.String()] = make(chan bool)
403 for _, task := range tasks {
405 fail := func(err error) {
406 if super.ctx.Err() != nil {
410 super.logger.WithField("task", task.String()).WithError(err).Error("task failed")
413 super.logger.WithField("task", task.String()).Info("starting")
414 err := task.Run(super.ctx, fail, super)
419 close(super.tasksReady[task.String()])
422 err = super.wait(super.ctx, tasks...)
426 super.logger.Info("all startup tasks are complete; starting health checks")
427 super.healthChecker = &health.Aggregator{Cluster: super.cluster}
429 super.logger.Info("shutting down")
430 super.waitShutdown.Wait()
431 return super.ctx.Err()
434 func (super *Supervisor) wait(ctx context.Context, tasks ...supervisedTask) error {
435 ticker := time.NewTicker(15 * time.Second)
437 for _, task := range tasks {
438 ch, ok := super.tasksReady[task.String()]
440 return fmt.Errorf("no such task: %s", task)
442 super.logger.WithField("task", task.String()).Info("waiting")
446 super.logger.WithField("task", task.String()).Info("ready")
448 super.logger.WithField("task", task.String()).Info("task was never ready")
451 super.logger.WithField("task", task.String()).Info("still waiting...")
460 // Stop shuts down all child processes and goroutines, and returns
461 // when all of them have exited.
462 func (super *Supervisor) Stop() {
467 // WaitReady waits for the cluster(s) to be ready to handle requests,
468 // then returns true. If startup fails, it returns false.
469 func (super *Supervisor) WaitReady() bool {
470 if super.children != nil {
471 for id, super2 := range super.children {
472 super.logger.Infof("waiting for %s to be ready", id)
473 if !super2.WaitReady() {
474 super.logger.Infof("%s startup failed", id)
478 super.logger.Infof("%s is ready", id)
480 super.logger.Info("all clusters are ready")
483 ticker := time.NewTicker(time.Second)
485 for waiting := "all"; waiting != ""; {
488 case <-super.ctx.Done():
492 if super.healthChecker == nil {
496 resp := super.healthChecker.ClusterHealth()
497 // The overall health check (resp.Health=="OK") might
498 // never pass due to missing components (like
499 // arvados-dispatch-cloud in a test cluster), so
500 // instead we wait for all configured components to
503 for target, check := range resp.Checks {
504 if check.Health != "OK" {
505 waiting += " " + target
509 super.logger.WithField("targets", waiting[1:]).Info("waiting")
515 func (super *Supervisor) prependEnv(key, prepend string) {
516 for i, s := range super.environ {
517 if strings.HasPrefix(s, key+"=") {
518 super.environ[i] = key + "=" + prepend + s[len(key)+1:]
522 super.environ = append(super.environ, key+"="+prepend)
525 func (super *Supervisor) cleanEnv(prefixes []string) {
527 for _, s := range super.environ {
529 for _, p := range prefixes {
530 if strings.HasPrefix(s, p) {
536 cleaned = append(cleaned, s)
539 super.environ = cleaned
542 func (super *Supervisor) setEnv(key, val string) {
543 for i, s := range super.environ {
544 if strings.HasPrefix(s, key+"=") {
545 super.environ[i] = key + "=" + val
549 super.environ = append(super.environ, key+"="+val)
552 // Remove all but the first occurrence of each env var.
553 func dedupEnv(in []string) []string {
554 saw := map[string]bool{}
556 for _, kv := range in {
557 if split := strings.Index(kv, "="); split < 1 {
558 panic("invalid environment var: " + kv)
559 } else if saw[kv[:split]] {
562 saw[kv[:split]] = true
563 out = append(out, kv)
569 func (super *Supervisor) installGoProgram(ctx context.Context, srcpath string) (string, error) {
570 _, basename := filepath.Split(srcpath)
571 binfile := filepath.Join(super.bindir, basename)
572 if super.ClusterType == "production" {
575 err := super.RunProgram(ctx, filepath.Join(super.SourcePath, srcpath), runOptions{env: []string{"GOBIN=" + super.bindir}}, "go", "install", "-ldflags", "-X git.arvados.org/arvados.git/lib/cmd.version="+super.SourceVersion+" -X main.version="+super.SourceVersion)
579 func (super *Supervisor) usingRVM() bool {
580 return os.Getenv("rvm_path") != ""
583 func (super *Supervisor) setupRubyEnv() error {
584 if !super.usingRVM() {
585 // (If rvm is in use, assume the caller has everything
586 // set up as desired)
587 super.cleanEnv([]string{
592 if _, err := os.Stat("/var/lib/arvados/bin/gem"); err == nil || super.ClusterType == "production" {
593 gem = "/var/lib/arvados/bin/gem"
595 cmd := exec.Command(gem, "env", "gempath")
596 if super.ClusterType == "production" {
597 cmd.Args = append([]string{"sudo", "-u", "www-data", "-E", "HOME=/var/www"}, cmd.Args...)
598 path, err := exec.LookPath("sudo")
600 return fmt.Errorf("LookPath(\"sudo\"): %w", err)
604 cmd.Stderr = super.Stderr
605 cmd.Env = super.environ
606 buf, err := cmd.Output() // /var/lib/arvados/.gem/ruby/2.5.0/bin:...
607 if err != nil || len(buf) == 0 {
608 return fmt.Errorf("gem env gempath: %w", err)
610 gempath := string(bytes.Split(buf, []byte{':'})[0])
611 super.prependEnv("PATH", gempath+"/bin:")
612 super.setEnv("GEM_HOME", gempath)
613 super.setEnv("GEM_PATH", gempath)
615 // Passenger install doesn't work unless $HOME is ~user
616 u, err := user.Current()
620 super.setEnv("HOME", u.HomeDir)
624 func (super *Supervisor) lookPath(prog string) string {
625 for _, val := range super.environ {
626 if strings.HasPrefix(val, "PATH=") {
627 for _, dir := range filepath.SplitList(val[5:]) {
628 path := filepath.Join(dir, prog)
629 if fi, err := os.Stat(path); err == nil && fi.Mode()&0111 != 0 {
638 type runOptions struct {
639 output io.Writer // attach stdout
640 env []string // add/replace environment variables
641 user string // run as specified user
645 // RunProgram runs prog with args, using dir as working directory. If ctx is
646 // cancelled while the child is running, RunProgram terminates the child, waits
647 // for it to exit, then returns.
649 // Child's environment will have our env vars, plus any given in env.
651 // Child's stdout will be written to output if non-nil, otherwise the
652 // boot command's stderr.
653 func (super *Supervisor) RunProgram(ctx context.Context, dir string, opts runOptions, prog string, args ...string) error {
654 cmdline := fmt.Sprintf("%s", append([]string{prog}, args...))
655 super.logger.WithField("command", cmdline).WithField("dir", dir).Info("executing")
660 if logprefix == "sudo" {
661 for i := 0; i < len(args); i++ {
664 } else if args[i] == "-E" || strings.Contains(args[i], "=") {
667 innerargs = args[i+1:]
672 logprefix = strings.TrimPrefix(logprefix, "/var/lib/arvados/bin/")
673 logprefix = strings.TrimPrefix(logprefix, super.tempdir+"/bin/")
674 if logprefix == "bundle" && len(innerargs) > 2 && innerargs[0] == "exec" {
675 _, dirbase := filepath.Split(dir)
676 logprefix = innerargs[1] + "@" + dirbase
677 } else if logprefix == "arvados-server" && len(args) > 1 {
680 if !strings.HasPrefix(dir, "/") {
681 logprefix = dir + ": " + logprefix
685 cmd := exec.Command(super.lookPath(prog), args...)
686 cmd.Stdin = opts.stdin
687 stdout, err := cmd.StdoutPipe()
691 stderr, err := cmd.StderrPipe()
695 logwriter := &service.LogPrefixer{Writer: super.Stderr, Prefix: []byte("[" + logprefix + "] ")}
696 var copiers sync.WaitGroup
699 io.Copy(logwriter, stderr)
704 if opts.output == nil {
705 io.Copy(logwriter, stdout)
707 io.Copy(opts.output, stdout)
712 if strings.HasPrefix(dir, "/") {
715 cmd.Dir = filepath.Join(super.SourcePath, dir)
717 env := append([]string(nil), opts.env...)
718 env = append(env, super.environ...)
719 cmd.Env = dedupEnv(env)
722 // Note: We use this approach instead of "sudo"
723 // because in certain circumstances (we are pid 1 in a
724 // docker container, and our passenger child process
725 // changes to pgid 1) the intermediate sudo process
726 // notices we have the same pgid as our child and
727 // refuses to propagate signals from us to our child,
728 // so we can't signal/shutdown our passenger/rails
729 // apps. "chpst" or "setuidgid" would work, but these
730 // few lines avoid depending on runit/daemontools.
731 u, err := user.Lookup(opts.user)
733 return fmt.Errorf("user.Lookup(%q): %w", opts.user, err)
735 uid, _ := strconv.Atoi(u.Uid)
736 gid, _ := strconv.Atoi(u.Gid)
737 cmd.SysProcAttr = &syscall.SysProcAttr{
738 Credential: &syscall.Credential{
746 defer func() { exited = true }()
749 log := ctxlog.FromContext(ctx).WithFields(logrus.Fields{"dir": dir, "cmdline": cmdline})
751 if cmd.Process == nil {
752 log.Debug("waiting for child process to start")
753 time.Sleep(time.Second / 2)
755 log.WithField("PID", cmd.Process.Pid).Debug("sending SIGTERM")
756 cmd.Process.Signal(syscall.SIGTERM)
757 time.Sleep(5 * time.Second)
761 log.WithField("PID", cmd.Process.Pid).Warn("still waiting for child process to exit 5s after SIGTERM")
773 if ctx.Err() != nil {
774 // Return "context canceled", instead of the "killed"
775 // error that was probably caused by the context being
778 } else if err != nil {
779 return fmt.Errorf("%s: error: %v", cmdline, err)
784 func (super *Supervisor) autofillConfig() error {
785 usedPort := map[string]bool{}
786 nextPort := func(host string) (string, error) {
788 port, err := availablePort(host)
790 port, err = availablePort(super.ListenHost)
798 usedPort[port] = true
802 if super.cluster.Services.Controller.ExternalURL.Host == "" {
803 h, p, err := net.SplitHostPort(super.ControllerAddr)
804 if err != nil && super.ControllerAddr != "" {
805 return fmt.Errorf("SplitHostPort(ControllerAddr %q): %w", super.ControllerAddr, err)
810 if p == "0" || p == "" {
816 super.cluster.Services.Controller.ExternalURL = arvados.URL{Scheme: "https", Host: net.JoinHostPort(h, p), Path: "/"}
818 u := url.URL(super.cluster.Services.Controller.ExternalURL)
819 defaultExtHost := u.Hostname()
820 for _, svc := range []*arvados.Service{
821 &super.cluster.Services.Controller,
822 &super.cluster.Services.DispatchCloud,
823 &super.cluster.Services.Health,
824 &super.cluster.Services.Keepproxy,
825 &super.cluster.Services.Keepstore,
826 &super.cluster.Services.RailsAPI,
827 &super.cluster.Services.WebDAV,
828 &super.cluster.Services.WebDAVDownload,
829 &super.cluster.Services.Websocket,
830 &super.cluster.Services.Workbench1,
831 &super.cluster.Services.Workbench2,
833 if svc.ExternalURL.Host == "" {
834 port, err := nextPort(defaultExtHost)
838 host := net.JoinHostPort(defaultExtHost, port)
839 if svc == &super.cluster.Services.Controller ||
840 svc == &super.cluster.Services.Health ||
841 svc == &super.cluster.Services.Keepproxy ||
842 svc == &super.cluster.Services.WebDAV ||
843 svc == &super.cluster.Services.WebDAVDownload ||
844 svc == &super.cluster.Services.Workbench1 ||
845 svc == &super.cluster.Services.Workbench2 {
846 svc.ExternalURL = arvados.URL{Scheme: "https", Host: host, Path: "/"}
847 } else if svc == &super.cluster.Services.Websocket {
848 svc.ExternalURL = arvados.URL{Scheme: "wss", Host: host, Path: "/websocket"}
851 if super.NoWorkbench1 && svc == &super.cluster.Services.Workbench1 ||
852 super.NoWorkbench2 && svc == &super.cluster.Services.Workbench2 ||
853 !super.cluster.Containers.CloudVMs.Enable && svc == &super.cluster.Services.DispatchCloud {
854 // When Workbench is disabled, it gets an
855 // ExternalURL (so we have a valid listening
856 // port to write in our Nginx config) but no
857 // InternalURLs (so health checker doesn't
861 if len(svc.InternalURLs) == 0 {
862 port, err := nextPort(super.ListenHost)
866 host := net.JoinHostPort(super.ListenHost, port)
867 svc.InternalURLs = map[arvados.URL]arvados.ServiceInstance{
868 {Scheme: "http", Host: host, Path: "/"}: {},
872 if super.ClusterType != "production" {
873 if super.cluster.SystemRootToken == "" {
874 super.cluster.SystemRootToken = randomHexString(64)
876 if super.cluster.ManagementToken == "" {
877 super.cluster.ManagementToken = randomHexString(64)
879 if super.cluster.Collections.BlobSigningKey == "" {
880 super.cluster.Collections.BlobSigningKey = randomHexString(64)
882 if super.cluster.Users.AnonymousUserToken == "" {
883 super.cluster.Users.AnonymousUserToken = randomHexString(64)
885 if super.cluster.Containers.DispatchPrivateKey == "" {
886 buf, err := ioutil.ReadFile(filepath.Join(super.SourcePath, "lib", "dispatchcloud", "test", "sshkey_dispatch"))
890 super.cluster.Containers.DispatchPrivateKey = string(buf)
892 super.cluster.TLS.Insecure = true
894 if super.ClusterType == "test" {
895 // Add a second keepstore process.
896 port, err := nextPort(super.ListenHost)
900 host := net.JoinHostPort(super.ListenHost, port)
901 super.cluster.Services.Keepstore.InternalURLs[arvados.URL{Scheme: "http", Host: host, Path: "/"}] = arvados.ServiceInstance{}
903 // Create a directory-backed volume for each keepstore
905 super.cluster.Volumes = map[string]arvados.Volume{}
906 for url := range super.cluster.Services.Keepstore.InternalURLs {
907 volnum := len(super.cluster.Volumes)
908 datadir := fmt.Sprintf("%s/keep%d.data", super.tempdir, volnum)
909 if _, err = os.Stat(datadir + "/."); err == nil {
910 } else if !os.IsNotExist(err) {
912 } else if err = os.Mkdir(datadir, 0755); err != nil {
915 super.cluster.Volumes[fmt.Sprintf(super.cluster.ClusterID+"-nyw5e-%015d", volnum)] = arvados.Volume{
917 DriverParameters: json.RawMessage(fmt.Sprintf(`{"Root":%q}`, datadir)),
918 AccessViaHosts: map[arvados.URL]arvados.VolumeAccess{
921 StorageClasses: map[string]bool{
928 super.cluster.StorageClasses = map[string]arvados.StorageClassConfig{
929 "default": {Default: true},
934 if super.OwnTemporaryDatabase {
935 port, err := nextPort("localhost")
939 super.cluster.PostgreSQL.Connection = arvados.PostgreSQLConnection{
940 "client_encoding": "utf8",
943 "dbname": "arvados_test",
945 "password": "insecure_arvados_test",
951 func addrIsLocal(addr string) (bool, error) {
952 if h, _, err := net.SplitHostPort(addr); err != nil {
955 addr = net.JoinHostPort(h, "0")
957 listener, err := net.Listen("tcp", addr)
961 } else if strings.Contains(err.Error(), "cannot assign requested address") {
968 func randomHexString(chars int) string {
969 b := make([]byte, chars/2)
970 _, err := rand.Read(b)
974 return fmt.Sprintf("%x", b)
977 func internalPort(svc arvados.Service) (host, port string, err error) {
978 if len(svc.InternalURLs) > 1 {
979 return "", "", errors.New("internalPort() doesn't work with multiple InternalURLs")
981 for u := range svc.InternalURLs {
983 host, port = u.Hostname(), u.Port()
986 case u.Scheme == "https", u.Scheme == "ws":
993 return "", "", fmt.Errorf("service has no InternalURLs")
996 func externalPort(svc arvados.Service) (string, error) {
997 u := url.URL(svc.ExternalURL)
998 if p := u.Port(); p != "" {
1000 } else if u.Scheme == "https" || u.Scheme == "wss" {
1007 func availablePort(host string) (string, error) {
1008 ln, err := net.Listen("tcp", net.JoinHostPort(host, "0"))
1013 _, port, err := net.SplitHostPort(ln.Addr().String())
1020 // Try to connect to addr until it works, then close ch. Give up if
1022 func waitForConnect(ctx context.Context, addr string) error {
1023 ctxlog.FromContext(ctx).WithField("addr", addr).Info("waitForConnect")
1024 dialer := net.Dialer{Timeout: time.Second}
1025 for ctx.Err() == nil {
1026 conn, err := dialer.DialContext(ctx, "tcp", addr)
1028 time.Sleep(time.Second / 10)
1037 func copyConfig(cfg *arvados.Config) *arvados.Config {
1040 err := json.NewEncoder(pw).Encode(cfg)
1046 cfg2 := new(arvados.Config)
1047 err := json.NewDecoder(pr).Decode(cfg2)
1054 func watchConfig(ctx context.Context, logger logrus.FieldLogger, cfgPath string, prevcfg *arvados.Config, fn func()) {
1055 watcher, err := fsnotify.NewWatcher()
1057 logger.WithError(err).Error("fsnotify setup failed")
1060 defer watcher.Close()
1062 err = watcher.Add(cfgPath)
1064 logger.WithError(err).Error("fsnotify watcher failed")
1072 case err, ok := <-watcher.Errors:
1076 logger.WithError(err).Warn("fsnotify watcher reported error")
1077 case _, ok := <-watcher.Events:
1081 for len(watcher.Events) > 0 {
1084 loader := config.NewLoader(&bytes.Buffer{}, &logrus.Logger{Out: ioutil.Discard})
1085 loader.Path = cfgPath
1086 loader.SkipAPICalls = true
1087 cfg, err := loader.Load()
1089 logger.WithError(err).Warn("error reloading config file after change detected; ignoring new config for now")
1090 } else if reflect.DeepEqual(cfg, prevcfg) {
1091 logger.Debug("config file changed but is still DeepEqual to the existing config")
1093 logger.Debug("config changed, notifying supervisor")