lib/boot/supervisor.go

   1 // Copyright (C) The Arvados Authors. All rights reserved.
   2 //
   3 // SPDX-License-Identifier: AGPL-3.0
   4
   5 package boot
   6
   7 import (
   8         "bytes"
   9         "context"
  10         "crypto/rand"
  11         "encoding/json"
  12         "errors"
  13         "fmt"
  14         "io"
  15         "io/ioutil"
  16         "net"
  17         "net/url"
  18         "os"
  19         "os/exec"
  20         "os/signal"
  21         "os/user"
  22         "path/filepath"
  23         "reflect"
  24         "strings"
  25         "sync"
  26         "syscall"
  27         "time"
  28
  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"
  36 )
  37
  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
  45         Stderr               io.Writer
  46
  47         logger  logrus.FieldLogger
  48         cluster *arvados.Cluster
  49
  50         ctx           context.Context
  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
  57
  58         bindir     string
  59         tempdir    string
  60         wwwtempdir string
  61         configfile string
  62         environ    []string // for child processes
  63 }
  64
  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{})
  68
  69         go func() {
  70                 defer close(super.done)
  71
  72                 sigch := make(chan os.Signal)
  73                 signal.Notify(sigch, syscall.SIGINT, syscall.SIGTERM)
  74                 defer signal.Stop(sigch)
  75                 go func() {
  76                         for sig := range sigch {
  77                                 super.logger.WithField("signal", sig).Info("caught signal")
  78                                 if super.err == nil {
  79                                         super.err = fmt.Errorf("caught signal %s", sig)
  80                                 }
  81                                 super.cancel()
  82                         }
  83                 }()
  84
  85                 hupch := make(chan os.Signal)
  86                 signal.Notify(hupch, syscall.SIGHUP)
  87                 defer signal.Stop(hupch)
  88                 go func() {
  89                         for sig := range hupch {
  90                                 super.logger.WithField("signal", sig).Info("caught signal")
  91                                 if super.err == nil {
  92                                         super.err = errNeedConfigReload
  93                                 }
  94                                 super.cancel()
  95                         }
  96                 }()
  97
  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
 102                                 }
 103                                 super.cancel()
 104                         })
 105                 }
 106
 107                 err := super.run(cfg)
 108                 if err != nil {
 109                         super.logger.WithError(err).Warn("supervisor shut down")
 110                         if super.err == nil {
 111                                 super.err = err
 112                         }
 113                 }
 114         }()
 115 }
 116
 117 func (super *Supervisor) Wait() error {
 118         <-super.done
 119         return super.err
 120 }
 121
 122 func (super *Supervisor) run(cfg *arvados.Config) error {
 123         defer super.cancel()
 124
 125         cwd, err := os.Getwd()
 126         if err != nil {
 127                 return err
 128         }
 129         if !strings.HasPrefix(super.SourcePath, "/") {
 130                 super.SourcePath = filepath.Join(cwd, super.SourcePath)
 131         }
 132         super.SourcePath, err = filepath.EvalSymlinks(super.SourcePath)
 133         if err != nil {
 134                 return err
 135         }
 136
 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
 142                 // package).
 143                 super.tempdir = "/var/lib/arvados/tmp"
 144                 super.wwwtempdir = "/var/lib/arvados/wwwtmp"
 145                 super.bindir = "/var/lib/arvados/bin"
 146         } else {
 147                 super.tempdir, err = ioutil.TempDir("", "arvados-server-boot-")
 148                 if err != nil {
 149                         return err
 150                 }
 151                 defer os.RemoveAll(super.tempdir)
 152                 super.wwwtempdir = super.tempdir
 153                 super.bindir = filepath.Join(super.tempdir, "bin")
 154                 if err := os.Mkdir(super.bindir, 0755); err != nil {
 155                         return err
 156                 }
 157         }
 158
 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)
 162         if err != nil {
 163                 return err
 164         }
 165         conffile, err := os.OpenFile(filepath.Join(super.tempdir, "config.yml"), os.O_CREATE|os.O_WRONLY, 0644)
 166         if err != nil {
 167                 return err
 168         }
 169         defer conffile.Close()
 170         err = json.NewEncoder(conffile).Encode(cfg)
 171         if err != nil {
 172                 return err
 173         }
 174         err = conffile.Close()
 175         if err != nil {
 176                 return err
 177         }
 178         super.configfile = conffile.Name()
 179
 180         super.environ = os.Environ()
 181         super.cleanEnv([]string{"ARVADOS_"})
 182         super.setEnv("ARVADOS_CONFIG", super.configfile)
 183         super.setEnv("RAILS_ENV", super.ClusterType)
 184         super.setEnv("TMPDIR", super.tempdir)
 185         super.prependEnv("PATH", "/var/lib/arvados/bin:")
 186         if super.ClusterType != "production" {
 187                 super.prependEnv("PATH", super.tempdir+"/bin:")
 188         }
 189
 190         super.cluster, err = cfg.GetCluster("")
 191         if err != nil {
 192                 return err
 193         }
 194         // Now that we have the config, replace the bootstrap logger
 195         // with a new one according to the logging config.
 196         loglevel := super.cluster.SystemLogs.LogLevel
 197         if s := os.Getenv("ARVADOS_DEBUG"); s != "" && s != "0" {
 198                 loglevel = "debug"
 199         }
 200         super.logger = ctxlog.New(super.Stderr, super.cluster.SystemLogs.Format, loglevel).WithFields(logrus.Fields{
 201                 "PID": os.Getpid(),
 202         })
 203
 204         if super.SourceVersion == "" && super.ClusterType == "production" {
 205                 // don't need SourceVersion
 206         } else if super.SourceVersion == "" {
 207                 // Find current source tree version.
 208                 var buf bytes.Buffer
 209                 err = super.RunProgram(super.ctx, ".", &buf, nil, "git", "diff", "--shortstat")
 210                 if err != nil {
 211                         return err
 212                 }
 213                 dirty := buf.Len() > 0
 214                 buf.Reset()
 215                 err = super.RunProgram(super.ctx, ".", &buf, nil, "git", "log", "-n1", "--format=%H")
 216                 if err != nil {
 217                         return err
 218                 }
 219                 super.SourceVersion = strings.TrimSpace(buf.String())
 220                 if dirty {
 221                         super.SourceVersion += "+uncommitted"
 222                 }
 223         } else {
 224                 return errors.New("specifying a version to run is not yet supported")
 225         }
 226
 227         _, err = super.installGoProgram(super.ctx, "cmd/arvados-server")
 228         if err != nil {
 229                 return err
 230         }
 231         err = super.setupRubyEnv()
 232         if err != nil {
 233                 return err
 234         }
 235
 236         tasks := []supervisedTask{
 237                 createCertificates{},
 238                 runPostgreSQL{},
 239                 runNginx{},
 240                 runServiceCommand{name: "controller", svc: super.cluster.Services.Controller, depends: []supervisedTask{runPostgreSQL{}}},
 241                 runGoProgram{src: "services/arv-git-httpd", svc: super.cluster.Services.GitHTTP},
 242                 runGoProgram{src: "services/health", svc: super.cluster.Services.Health},
 243                 runGoProgram{src: "services/keepproxy", svc: super.cluster.Services.Keepproxy, depends: []supervisedTask{runPassenger{src: "services/api"}}},
 244                 runGoProgram{src: "services/keepstore", svc: super.cluster.Services.Keepstore},
 245                 runGoProgram{src: "services/keep-web", svc: super.cluster.Services.WebDAV},
 246                 runServiceCommand{name: "ws", svc: super.cluster.Services.Websocket, depends: []supervisedTask{runPostgreSQL{}}},
 247                 installPassenger{src: "services/api"},
 248                 runPassenger{src: "services/api", varlibdir: "railsapi", svc: super.cluster.Services.RailsAPI, depends: []supervisedTask{createCertificates{}, runPostgreSQL{}, installPassenger{src: "services/api"}}},
 249                 installPassenger{src: "apps/workbench", depends: []supervisedTask{installPassenger{src: "services/api"}}}, // dependency ensures workbench doesn't delay api startup
 250                 runPassenger{src: "apps/workbench", varlibdir: "workbench1", svc: super.cluster.Services.Workbench1, depends: []supervisedTask{installPassenger{src: "apps/workbench"}}},
 251                 seedDatabase{},
 252         }
 253         if super.ClusterType != "test" {
 254                 tasks = append(tasks,
 255                         runServiceCommand{name: "dispatch-cloud", svc: super.cluster.Services.Controller},
 256                         runGoProgram{src: "services/keep-balance", svc: super.cluster.Services.Keepbalance},
 257                 )
 258         }
 259         super.tasksReady = map[string]chan bool{}
 260         for _, task := range tasks {
 261                 super.tasksReady[task.String()] = make(chan bool)
 262         }
 263         for _, task := range tasks {
 264                 task := task
 265                 fail := func(err error) {
 266                         if super.ctx.Err() != nil {
 267                                 return
 268                         }
 269                         super.cancel()
 270                         super.logger.WithField("task", task.String()).WithError(err).Error("task failed")
 271                 }
 272                 go func() {
 273                         super.logger.WithField("task", task.String()).Info("starting")
 274                         err := task.Run(super.ctx, fail, super)
 275                         if err != nil {
 276                                 fail(err)
 277                                 return
 278                         }
 279                         close(super.tasksReady[task.String()])
 280                 }()
 281         }
 282         err = super.wait(super.ctx, tasks...)
 283         if err != nil {
 284                 return err
 285         }
 286         super.logger.Info("all startup tasks are complete; starting health checks")
 287         super.healthChecker = &health.Aggregator{Cluster: super.cluster}
 288         <-super.ctx.Done()
 289         super.logger.Info("shutting down")
 290         super.waitShutdown.Wait()
 291         return super.ctx.Err()
 292 }
 293
 294 func (super *Supervisor) wait(ctx context.Context, tasks ...supervisedTask) error {
 295         for _, task := range tasks {
 296                 ch, ok := super.tasksReady[task.String()]
 297                 if !ok {
 298                         return fmt.Errorf("no such task: %s", task)
 299                 }
 300                 super.logger.WithField("task", task.String()).Info("waiting")
 301                 select {
 302                 case <-ch:
 303                         super.logger.WithField("task", task.String()).Info("ready")
 304                 case <-ctx.Done():
 305                         super.logger.WithField("task", task.String()).Info("task was never ready")
 306                         return ctx.Err()
 307                 }
 308         }
 309         return nil
 310 }
 311
 312 func (super *Supervisor) Stop() {
 313         super.cancel()
 314         <-super.done
 315 }
 316
 317 func (super *Supervisor) WaitReady() (*arvados.URL, bool) {
 318         ticker := time.NewTicker(time.Second)
 319         defer ticker.Stop()
 320         for waiting := "all"; waiting != ""; {
 321                 select {
 322                 case <-ticker.C:
 323                 case <-super.ctx.Done():
 324                         return nil, false
 325                 }
 326                 if super.healthChecker == nil {
 327                         // not set up yet
 328                         continue
 329                 }
 330                 resp := super.healthChecker.ClusterHealth()
 331                 // The overall health check (resp.Health=="OK") might
 332                 // never pass due to missing components (like
 333                 // arvados-dispatch-cloud in a test cluster), so
 334                 // instead we wait for all configured components to
 335                 // pass.
 336                 waiting = ""
 337                 for target, check := range resp.Checks {
 338                         if check.Health != "OK" {
 339                                 waiting += " " + target
 340                         }
 341                 }
 342                 if waiting != "" {
 343                         super.logger.WithField("targets", waiting[1:]).Info("waiting")
 344                 }
 345         }
 346         u := super.cluster.Services.Controller.ExternalURL
 347         return &u, true
 348 }
 349
 350 func (super *Supervisor) prependEnv(key, prepend string) {
 351         for i, s := range super.environ {
 352                 if strings.HasPrefix(s, key+"=") {
 353                         super.environ[i] = key + "=" + prepend + s[len(key)+1:]
 354                         return
 355                 }
 356         }
 357         super.environ = append(super.environ, key+"="+prepend)
 358 }
 359
 360 func (super *Supervisor) cleanEnv(prefixes []string) {
 361         var cleaned []string
 362         for _, s := range super.environ {
 363                 drop := false
 364                 for _, p := range prefixes {
 365                         if strings.HasPrefix(s, p) {
 366                                 drop = true
 367                                 break
 368                         }
 369                 }
 370                 if !drop {
 371                         cleaned = append(cleaned, s)
 372                 }
 373         }
 374         super.environ = cleaned
 375 }
 376
 377 func (super *Supervisor) setEnv(key, val string) {
 378         for i, s := range super.environ {
 379                 if strings.HasPrefix(s, key+"=") {
 380                         super.environ[i] = key + "=" + val
 381                         return
 382                 }
 383         }
 384         super.environ = append(super.environ, key+"="+val)
 385 }
 386
 387 // Remove all but the first occurrence of each env var.
 388 func dedupEnv(in []string) []string {
 389         saw := map[string]bool{}
 390         var out []string
 391         for _, kv := range in {
 392                 if split := strings.Index(kv, "="); split < 1 {
 393                         panic("invalid environment var: " + kv)
 394                 } else if saw[kv[:split]] {
 395                         continue
 396                 } else {
 397                         saw[kv[:split]] = true
 398                         out = append(out, kv)
 399                 }
 400         }
 401         return out
 402 }
 403
 404 func (super *Supervisor) installGoProgram(ctx context.Context, srcpath string) (string, error) {
 405         _, basename := filepath.Split(srcpath)
 406         binfile := filepath.Join(super.bindir, basename)
 407         if super.ClusterType == "production" {
 408                 return binfile, nil
 409         }
 410         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)
 411         return binfile, err
 412 }
 413
 414 func (super *Supervisor) usingRVM() bool {
 415         return os.Getenv("rvm_path") != ""
 416 }
 417
 418 func (super *Supervisor) setupRubyEnv() error {
 419         if !super.usingRVM() {
 420                 // (If rvm is in use, assume the caller has everything
 421                 // set up as desired)
 422                 super.cleanEnv([]string{
 423                         "GEM_HOME=",
 424                         "GEM_PATH=",
 425                 })
 426                 gem := "gem"
 427                 if _, err := os.Stat("/var/lib/arvados/bin/gem"); err == nil || super.ClusterType == "production" {
 428                         gem = "/var/lib/arvados/bin/gem"
 429                 }
 430                 cmd := exec.Command(gem, "env", "gempath")
 431                 if super.ClusterType == "production" {
 432                         cmd.Args = append([]string{"sudo", "-u", "www-data", "-E", "HOME=/var/www"}, cmd.Args...)
 433                         path, err := exec.LookPath("sudo")
 434                         if err != nil {
 435                                 return fmt.Errorf("LookPath(\"sudo\"): %w", err)
 436                         }
 437                         cmd.Path = path
 438                 }
 439                 cmd.Stderr = super.Stderr
 440                 cmd.Env = super.environ
 441                 buf, err := cmd.Output() // /var/lib/arvados/.gem/ruby/2.5.0/bin:...
 442                 if err != nil || len(buf) == 0 {
 443                         return fmt.Errorf("gem env gempath: %v", err)
 444                 }
 445                 gempath := string(bytes.Split(buf, []byte{':'})[0])
 446                 super.prependEnv("PATH", gempath+"/bin:")
 447                 super.setEnv("GEM_HOME", gempath)
 448                 super.setEnv("GEM_PATH", gempath)
 449         }
 450         // Passenger install doesn't work unless $HOME is ~user
 451         u, err := user.Current()
 452         if err != nil {
 453                 return err
 454         }
 455         super.setEnv("HOME", u.HomeDir)
 456         return nil
 457 }
 458
 459 func (super *Supervisor) lookPath(prog string) string {
 460         for _, val := range super.environ {
 461                 if strings.HasPrefix(val, "PATH=") {
 462                         for _, dir := range filepath.SplitList(val[5:]) {
 463                                 path := filepath.Join(dir, prog)
 464                                 if fi, err := os.Stat(path); err == nil && fi.Mode()&0111 != 0 {
 465                                         return path
 466                                 }
 467                         }
 468                 }
 469         }
 470         return prog
 471 }
 472
 473 // Run prog with args, using dir as working directory. If ctx is
 474 // cancelled while the child is running, RunProgram terminates the
 475 // child, waits for it to exit, then returns.
 476 //
 477 // Child's environment will have our env vars, plus any given in env.
 478 //
 479 // Child's stdout will be written to output if non-nil, otherwise the
 480 // boot command's stderr.
 481 func (super *Supervisor) RunProgram(ctx context.Context, dir string, output io.Writer, env []string, prog string, args ...string) error {
 482         cmdline := fmt.Sprintf("%s", append([]string{prog}, args...))
 483         super.logger.WithField("command", cmdline).WithField("dir", dir).Info("executing")
 484
 485         logprefix := prog
 486         if logprefix == "setuidgid" && len(args) >= 3 {
 487                 logprefix = args[2]
 488         }
 489         logprefix = strings.TrimPrefix(logprefix, super.tempdir+"/bin/")
 490         if logprefix == "bundle" && len(args) > 2 && args[0] == "exec" {
 491                 logprefix = args[1]
 492         } else if logprefix == "arvados-server" && len(args) > 1 {
 493                 logprefix = args[0]
 494         }
 495         if !strings.HasPrefix(dir, "/") {
 496                 logprefix = dir + ": " + logprefix
 497         }
 498
 499         cmd := exec.Command(super.lookPath(prog), args...)
 500         stdout, err := cmd.StdoutPipe()
 501         if err != nil {
 502                 return err
 503         }
 504         stderr, err := cmd.StderrPipe()
 505         if err != nil {
 506                 return err
 507         }
 508         logwriter := &service.LogPrefixer{Writer: super.Stderr, Prefix: []byte("[" + logprefix + "] ")}
 509         var copiers sync.WaitGroup
 510         copiers.Add(1)
 511         go func() {
 512                 io.Copy(logwriter, stderr)
 513                 copiers.Done()
 514         }()
 515         copiers.Add(1)
 516         go func() {
 517                 if output == nil {
 518                         io.Copy(logwriter, stdout)
 519                 } else {
 520                         io.Copy(output, stdout)
 521                 }
 522                 copiers.Done()
 523         }()
 524
 525         if strings.HasPrefix(dir, "/") {
 526                 cmd.Dir = dir
 527         } else {
 528                 cmd.Dir = filepath.Join(super.SourcePath, dir)
 529         }
 530         env = append([]string(nil), env...)
 531         env = append(env, super.environ...)
 532         cmd.Env = dedupEnv(env)
 533
 534         exited := false
 535         defer func() { exited = true }()
 536         go func() {
 537                 <-ctx.Done()
 538                 log := ctxlog.FromContext(ctx).WithFields(logrus.Fields{"dir": dir, "cmdline": cmdline})
 539                 for !exited {
 540                         if cmd.Process == nil {
 541                                 log.Debug("waiting for child process to start")
 542                                 time.Sleep(time.Second / 2)
 543                         } else {
 544                                 log.WithField("PID", cmd.Process.Pid).Debug("sending SIGTERM")
 545                                 cmd.Process.Signal(syscall.SIGTERM)
 546                                 time.Sleep(5 * time.Second)
 547                                 if !exited {
 548                                         stdout.Close()
 549                                         stderr.Close()
 550                                         log.WithField("PID", cmd.Process.Pid).Warn("still waiting for child process to exit 5s after SIGTERM")
 551                                 }
 552                         }
 553                 }
 554         }()
 555
 556         err = cmd.Start()
 557         if err != nil {
 558                 return err
 559         }
 560         copiers.Wait()
 561         err = cmd.Wait()
 562         if ctx.Err() != nil {
 563                 // Return "context canceled", instead of the "killed"
 564                 // error that was probably caused by the context being
 565                 // canceled.
 566                 return ctx.Err()
 567         } else if err != nil {
 568                 return fmt.Errorf("%s: error: %v", cmdline, err)
 569         }
 570         return nil
 571 }
 572
 573 func (super *Supervisor) autofillConfig(cfg *arvados.Config) error {
 574         cluster, err := cfg.GetCluster("")
 575         if err != nil {
 576                 return err
 577         }
 578         usedPort := map[string]bool{}
 579         nextPort := func(host string) string {
 580                 for {
 581                         port, err := availablePort(host)
 582                         if err != nil {
 583                                 panic(err)
 584                         }
 585                         if usedPort[port] {
 586                                 continue
 587                         }
 588                         usedPort[port] = true
 589                         return port
 590                 }
 591         }
 592         if cluster.Services.Controller.ExternalURL.Host == "" {
 593                 h, p, err := net.SplitHostPort(super.ControllerAddr)
 594                 if err != nil {
 595                         return err
 596                 }
 597                 if h == "" {
 598                         h = super.ListenHost
 599                 }
 600                 if p == "0" {
 601                         p = nextPort(h)
 602                 }
 603                 cluster.Services.Controller.ExternalURL = arvados.URL{Scheme: "https", Host: net.JoinHostPort(h, p), Path: "/"}
 604         }
 605         for _, svc := range []*arvados.Service{
 606                 &cluster.Services.Controller,
 607                 &cluster.Services.DispatchCloud,
 608                 &cluster.Services.GitHTTP,
 609                 &cluster.Services.Health,
 610                 &cluster.Services.Keepproxy,
 611                 &cluster.Services.Keepstore,
 612                 &cluster.Services.RailsAPI,
 613                 &cluster.Services.WebDAV,
 614                 &cluster.Services.WebDAVDownload,
 615                 &cluster.Services.Websocket,
 616                 &cluster.Services.Workbench1,
 617         } {
 618                 if svc == &cluster.Services.DispatchCloud && super.ClusterType == "test" {
 619                         continue
 620                 }
 621                 if svc.ExternalURL.Host == "" {
 622                         if svc == &cluster.Services.Controller ||
 623                                 svc == &cluster.Services.GitHTTP ||
 624                                 svc == &cluster.Services.Health ||
 625                                 svc == &cluster.Services.Keepproxy ||
 626                                 svc == &cluster.Services.WebDAV ||
 627                                 svc == &cluster.Services.WebDAVDownload ||
 628                                 svc == &cluster.Services.Workbench1 {
 629                                 svc.ExternalURL = arvados.URL{Scheme: "https", Host: fmt.Sprintf("%s:%s", super.ListenHost, nextPort(super.ListenHost)), Path: "/"}
 630                         } else if svc == &cluster.Services.Websocket {
 631                                 svc.ExternalURL = arvados.URL{Scheme: "wss", Host: fmt.Sprintf("%s:%s", super.ListenHost, nextPort(super.ListenHost)), Path: "/websocket"}
 632                         }
 633                 }
 634                 if len(svc.InternalURLs) == 0 {
 635                         svc.InternalURLs = map[arvados.URL]arvados.ServiceInstance{
 636                                 arvados.URL{Scheme: "http", Host: fmt.Sprintf("%s:%s", super.ListenHost, nextPort(super.ListenHost)), Path: "/"}: arvados.ServiceInstance{},
 637                         }
 638                 }
 639         }
 640         if cluster.SystemRootToken == "" {
 641                 cluster.SystemRootToken = randomHexString(64)
 642         }
 643         if cluster.ManagementToken == "" {
 644                 cluster.ManagementToken = randomHexString(64)
 645         }
 646         if cluster.API.RailsSessionSecretToken == "" {
 647                 cluster.API.RailsSessionSecretToken = randomHexString(64)
 648         }
 649         if cluster.Collections.BlobSigningKey == "" {
 650                 cluster.Collections.BlobSigningKey = randomHexString(64)
 651         }
 652         if super.ClusterType != "production" && cluster.Containers.DispatchPrivateKey == "" {
 653                 buf, err := ioutil.ReadFile(filepath.Join(super.SourcePath, "lib", "dispatchcloud", "test", "sshkey_dispatch"))
 654                 if err != nil {
 655                         return err
 656                 }
 657                 cluster.Containers.DispatchPrivateKey = string(buf)
 658         }
 659         if super.ClusterType != "production" {
 660                 cluster.TLS.Insecure = true
 661         }
 662         if super.ClusterType == "test" {
 663                 // Add a second keepstore process.
 664                 cluster.Services.Keepstore.InternalURLs[arvados.URL{Scheme: "http", Host: fmt.Sprintf("%s:%s", super.ListenHost, nextPort(super.ListenHost)), Path: "/"}] = arvados.ServiceInstance{}
 665
 666                 // Create a directory-backed volume for each keepstore
 667                 // process.
 668                 cluster.Volumes = map[string]arvados.Volume{}
 669                 for url := range cluster.Services.Keepstore.InternalURLs {
 670                         volnum := len(cluster.Volumes)
 671                         datadir := fmt.Sprintf("%s/keep%d.data", super.tempdir, volnum)
 672                         if _, err = os.Stat(datadir + "/."); err == nil {
 673                         } else if !os.IsNotExist(err) {
 674                                 return err
 675                         } else if err = os.Mkdir(datadir, 0755); err != nil {
 676                                 return err
 677                         }
 678                         cluster.Volumes[fmt.Sprintf(cluster.ClusterID+"-nyw5e-%015d", volnum)] = arvados.Volume{
 679                                 Driver:           "Directory",
 680                                 DriverParameters: json.RawMessage(fmt.Sprintf(`{"Root":%q}`, datadir)),
 681                                 AccessViaHosts: map[arvados.URL]arvados.VolumeAccess{
 682                                         url: {},
 683                                 },
 684                         }
 685                 }
 686         }
 687         if super.OwnTemporaryDatabase {
 688                 cluster.PostgreSQL.Connection = arvados.PostgreSQLConnection{
 689                         "client_encoding": "utf8",
 690                         "host":            "localhost",
 691                         "port":            nextPort(super.ListenHost),
 692                         "dbname":          "arvados_test",
 693                         "user":            "arvados",
 694                         "password":        "insecure_arvados_test",
 695                 }
 696         }
 697
 698         cfg.Clusters[cluster.ClusterID] = *cluster
 699         return nil
 700 }
 701
 702 func addrIsLocal(addr string) (bool, error) {
 703         return true, nil
 704         listener, err := net.Listen("tcp", addr)
 705         if err == nil {
 706                 listener.Close()
 707                 return true, nil
 708         } else if strings.Contains(err.Error(), "cannot assign requested address") {
 709                 return false, nil
 710         } else {
 711                 return false, err
 712         }
 713 }
 714
 715 func randomHexString(chars int) string {
 716         b := make([]byte, chars/2)
 717         _, err := rand.Read(b)
 718         if err != nil {
 719                 panic(err)
 720         }
 721         return fmt.Sprintf("%x", b)
 722 }
 723
 724 func internalPort(svc arvados.Service) (string, error) {
 725         if len(svc.InternalURLs) > 1 {
 726                 return "", errors.New("internalPort() doesn't work with multiple InternalURLs")
 727         }
 728         for u := range svc.InternalURLs {
 729                 u := url.URL(u)
 730                 if p := u.Port(); p != "" {
 731                         return p, nil
 732                 } else if u.Scheme == "https" || u.Scheme == "ws" {
 733                         return "443", nil
 734                 } else {
 735                         return "80", nil
 736                 }
 737         }
 738         return "", fmt.Errorf("service has no InternalURLs")
 739 }
 740
 741 func externalPort(svc arvados.Service) (string, error) {
 742         u := url.URL(svc.ExternalURL)
 743         if p := u.Port(); p != "" {
 744                 return p, nil
 745         } else if u.Scheme == "https" || u.Scheme == "wss" {
 746                 return "443", nil
 747         } else {
 748                 return "80", nil
 749         }
 750 }
 751
 752 func availablePort(host string) (string, error) {
 753         ln, err := net.Listen("tcp", net.JoinHostPort(host, "0"))
 754         if err != nil {
 755                 return "", err
 756         }
 757         defer ln.Close()
 758         _, port, err := net.SplitHostPort(ln.Addr().String())
 759         if err != nil {
 760                 return "", err
 761         }
 762         return port, nil
 763 }
 764
 765 // Try to connect to addr until it works, then close ch. Give up if
 766 // ctx cancels.
 767 func waitForConnect(ctx context.Context, addr string) error {
 768         dialer := net.Dialer{Timeout: time.Second}
 769         for ctx.Err() == nil {
 770                 conn, err := dialer.DialContext(ctx, "tcp", addr)
 771                 if err != nil {
 772                         time.Sleep(time.Second / 10)
 773                         continue
 774                 }
 775                 conn.Close()
 776                 return nil
 777         }
 778         return ctx.Err()
 779 }
 780
 781 func copyConfig(cfg *arvados.Config) *arvados.Config {
 782         pr, pw := io.Pipe()
 783         go func() {
 784                 err := json.NewEncoder(pw).Encode(cfg)
 785                 if err != nil {
 786                         panic(err)
 787                 }
 788                 pw.Close()
 789         }()
 790         cfg2 := new(arvados.Config)
 791         err := json.NewDecoder(pr).Decode(cfg2)
 792         if err != nil {
 793                 panic(err)
 794         }
 795         return cfg2
 796 }
 797
 798 func watchConfig(ctx context.Context, logger logrus.FieldLogger, cfgPath string, prevcfg *arvados.Config, fn func()) {
 799         watcher, err := fsnotify.NewWatcher()
 800         if err != nil {
 801                 logger.WithError(err).Error("fsnotify setup failed")
 802                 return
 803         }
 804         defer watcher.Close()
 805
 806         err = watcher.Add(cfgPath)
 807         if err != nil {
 808                 logger.WithError(err).Error("fsnotify watcher failed")
 809                 return
 810         }
 811
 812         for {
 813                 select {
 814                 case <-ctx.Done():
 815                         return
 816                 case err, ok := <-watcher.Errors:
 817                         if !ok {
 818                                 return
 819                         }
 820                         logger.WithError(err).Warn("fsnotify watcher reported error")
 821                 case _, ok := <-watcher.Events:
 822                         if !ok {
 823                                 return
 824                         }
 825                         for len(watcher.Events) > 0 {
 826                                 <-watcher.Events
 827                         }
 828                         loader := config.NewLoader(&bytes.Buffer{}, &logrus.Logger{Out: ioutil.Discard})
 829                         loader.Path = cfgPath
 830                         loader.SkipAPICalls = true
 831                         cfg, err := loader.Load()
 832                         if err != nil {
 833                                 logger.WithError(err).Warn("error reloading config file after change detected; ignoring new config for now")
 834                         } else if reflect.DeepEqual(cfg, prevcfg) {
 835                                 logger.Debug("config file changed but is still DeepEqual to the existing config")
 836                         } else {
 837                                 logger.Debug("config changed, notifying supervisor")
 838                                 fn()
 839                                 prevcfg = cfg
 840                         }
 841                 }
 842         }
 843 }