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         "strconv"
  25         "strings"
  26         "sync"
  27         "syscall"
  28         "time"
  29
  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"
  37 )
  38
  39 type Supervisor struct {
  40         SourcePath           string // e.g., /home/username/src/arvados
  41         SourceVersion        string // e.g., acbd1324...
  42         ClusterType          string // e.g., production
  43         ListenHost           string // e.g., localhost
  44         ControllerAddr       string // e.g., 127.0.0.1:8000
  45         Workbench2Source     string // e.g., /home/username/src/arvados-workbench2
  46         NoWorkbench1         bool
  47         NoWorkbench2         bool
  48         OwnTemporaryDatabase bool
  49         Stderr               io.Writer
  50
  51         logger  logrus.FieldLogger
  52         cluster *arvados.Cluster
  53
  54         ctx           context.Context
  55         cancel        context.CancelFunc
  56         done          chan struct{} // closed when child procs/services have shut down
  57         err           error         // error that caused shutdown (valid when done is closed)
  58         healthChecker *health.Aggregator
  59         tasksReady    map[string]chan bool
  60         waitShutdown  sync.WaitGroup
  61
  62         bindir     string
  63         tempdir    string // in production mode, this is accessible only to root
  64         wwwtempdir string // in production mode, this is accessible only to www-data
  65         configfile string
  66         environ    []string // for child processes
  67 }
  68
  69 func (super *Supervisor) Cluster() *arvados.Cluster { return super.cluster }
  70
  71 func (super *Supervisor) Start(ctx context.Context, cfg *arvados.Config, cfgPath string) {
  72         super.ctx, super.cancel = context.WithCancel(ctx)
  73         super.done = make(chan struct{})
  74
  75         go func() {
  76                 defer close(super.done)
  77
  78                 sigch := make(chan os.Signal)
  79                 signal.Notify(sigch, syscall.SIGINT, syscall.SIGTERM)
  80                 defer signal.Stop(sigch)
  81                 go func() {
  82                         for sig := range sigch {
  83                                 super.logger.WithField("signal", sig).Info("caught signal")
  84                                 if super.err == nil {
  85                                         super.err = fmt.Errorf("caught signal %s", sig)
  86                                 }
  87                                 super.cancel()
  88                         }
  89                 }()
  90
  91                 hupch := make(chan os.Signal)
  92                 signal.Notify(hupch, syscall.SIGHUP)
  93                 defer signal.Stop(hupch)
  94                 go func() {
  95                         for sig := range hupch {
  96                                 super.logger.WithField("signal", sig).Info("caught signal")
  97                                 if super.err == nil {
  98                                         super.err = errNeedConfigReload
  99                                 }
 100                                 super.cancel()
 101                         }
 102                 }()
 103
 104                 if cfgPath != "" && cfgPath != "-" && cfg.AutoReloadConfig {
 105                         go watchConfig(super.ctx, super.logger, cfgPath, copyConfig(cfg), func() {
 106                                 if super.err == nil {
 107                                         super.err = errNeedConfigReload
 108                                 }
 109                                 super.cancel()
 110                         })
 111                 }
 112
 113                 err := super.run(cfg)
 114                 if err != nil {
 115                         super.logger.WithError(err).Warn("supervisor shut down")
 116                         if super.err == nil {
 117                                 super.err = err
 118                         }
 119                 }
 120         }()
 121 }
 122
 123 func (super *Supervisor) Wait() error {
 124         <-super.done
 125         return super.err
 126 }
 127
 128 func (super *Supervisor) run(cfg *arvados.Config) error {
 129         defer super.cancel()
 130
 131         cwd, err := os.Getwd()
 132         if err != nil {
 133                 return err
 134         }
 135         if !strings.HasPrefix(super.SourcePath, "/") {
 136                 super.SourcePath = filepath.Join(cwd, super.SourcePath)
 137         }
 138         super.SourcePath, err = filepath.EvalSymlinks(super.SourcePath)
 139         if err != nil {
 140                 return err
 141         }
 142
 143         // Choose bin and temp dirs: /var/lib/arvados/... in
 144         // production, transient tempdir otherwise.
 145         if super.ClusterType == "production" {
 146                 // These dirs have already been created by
 147                 // "arvados-server install" (or by extracting a
 148                 // package).
 149                 super.tempdir = "/var/lib/arvados/tmp"
 150                 super.wwwtempdir = "/var/lib/arvados/wwwtmp"
 151                 super.bindir = "/var/lib/arvados/bin"
 152         } else {
 153                 super.tempdir, err = ioutil.TempDir("", "arvados-server-boot-")
 154                 if err != nil {
 155                         return err
 156                 }
 157                 defer os.RemoveAll(super.tempdir)
 158                 super.wwwtempdir = super.tempdir
 159                 super.bindir = filepath.Join(super.tempdir, "bin")
 160                 if err := os.Mkdir(super.bindir, 0755); err != nil {
 161                         return err
 162                 }
 163         }
 164
 165         // Fill in any missing config keys, and write the resulting
 166         // config in the temp dir for child services to use.
 167         err = super.autofillConfig(cfg)
 168         if err != nil {
 169                 return err
 170         }
 171         conffile, err := os.OpenFile(filepath.Join(super.wwwtempdir, "config.yml"), os.O_CREATE|os.O_WRONLY, 0644)
 172         if err != nil {
 173                 return err
 174         }
 175         defer conffile.Close()
 176         err = json.NewEncoder(conffile).Encode(cfg)
 177         if err != nil {
 178                 return err
 179         }
 180         err = conffile.Close()
 181         if err != nil {
 182                 return err
 183         }
 184         super.configfile = conffile.Name()
 185
 186         super.environ = os.Environ()
 187         super.cleanEnv([]string{"ARVADOS_"})
 188         super.setEnv("ARVADOS_CONFIG", super.configfile)
 189         super.setEnv("RAILS_ENV", super.ClusterType)
 190         super.setEnv("TMPDIR", super.tempdir)
 191         super.prependEnv("PATH", "/var/lib/arvados/bin:")
 192         if super.ClusterType != "production" {
 193                 super.prependEnv("PATH", super.tempdir+"/bin:")
 194         }
 195
 196         super.cluster, err = cfg.GetCluster("")
 197         if err != nil {
 198                 return err
 199         }
 200         // Now that we have the config, replace the bootstrap logger
 201         // with a new one according to the logging config.
 202         loglevel := super.cluster.SystemLogs.LogLevel
 203         if s := os.Getenv("ARVADOS_DEBUG"); s != "" && s != "0" {
 204                 loglevel = "debug"
 205         }
 206         super.logger = ctxlog.New(super.Stderr, super.cluster.SystemLogs.Format, loglevel).WithFields(logrus.Fields{
 207                 "PID": os.Getpid(),
 208         })
 209
 210         if super.SourceVersion == "" && super.ClusterType == "production" {
 211                 // don't need SourceVersion
 212         } else if super.SourceVersion == "" {
 213                 // Find current source tree version.
 214                 var buf bytes.Buffer
 215                 err = super.RunProgram(super.ctx, ".", runOptions{output: &buf}, "git", "diff", "--shortstat")
 216                 if err != nil {
 217                         return err
 218                 }
 219                 dirty := buf.Len() > 0
 220                 buf.Reset()
 221                 err = super.RunProgram(super.ctx, ".", runOptions{output: &buf}, "git", "log", "-n1", "--format=%H")
 222                 if err != nil {
 223                         return err
 224                 }
 225                 super.SourceVersion = strings.TrimSpace(buf.String())
 226                 if dirty {
 227                         super.SourceVersion += "+uncommitted"
 228                 }
 229         } else {
 230                 return errors.New("specifying a version to run is not yet supported")
 231         }
 232
 233         _, err = super.installGoProgram(super.ctx, "cmd/arvados-server")
 234         if err != nil {
 235                 return err
 236         }
 237         err = super.setupRubyEnv()
 238         if err != nil {
 239                 return err
 240         }
 241
 242         tasks := []supervisedTask{
 243                 createCertificates{},
 244                 runPostgreSQL{},
 245                 runNginx{},
 246                 runServiceCommand{name: "controller", svc: super.cluster.Services.Controller, depends: []supervisedTask{seedDatabase{}}},
 247                 runGoProgram{src: "services/arv-git-httpd", svc: super.cluster.Services.GitHTTP},
 248                 runGoProgram{src: "services/health", svc: super.cluster.Services.Health},
 249                 runGoProgram{src: "services/keepproxy", svc: super.cluster.Services.Keepproxy, depends: []supervisedTask{runPassenger{src: "services/api"}}},
 250                 runServiceCommand{name: "keepstore", svc: super.cluster.Services.Keepstore},
 251                 runGoProgram{src: "services/keep-web", svc: super.cluster.Services.WebDAV},
 252                 runServiceCommand{name: "ws", svc: super.cluster.Services.Websocket, depends: []supervisedTask{seedDatabase{}}},
 253                 installPassenger{src: "services/api"},
 254                 runPassenger{src: "services/api", varlibdir: "railsapi", svc: super.cluster.Services.RailsAPI, depends: []supervisedTask{createCertificates{}, seedDatabase{}, installPassenger{src: "services/api"}}},
 255                 seedDatabase{},
 256         }
 257         if !super.NoWorkbench1 {
 258                 tasks = append(tasks,
 259                         installPassenger{src: "apps/workbench", depends: []supervisedTask{seedDatabase{}}}, // dependency ensures workbench doesn't delay api install/startup
 260                         runPassenger{src: "apps/workbench", varlibdir: "workbench1", svc: super.cluster.Services.Workbench1, depends: []supervisedTask{installPassenger{src: "apps/workbench"}}},
 261                 )
 262         }
 263         if !super.NoWorkbench2 {
 264                 tasks = append(tasks,
 265                         runWorkbench2{svc: super.cluster.Services.Workbench2},
 266                 )
 267         }
 268         if super.ClusterType != "test" {
 269                 tasks = append(tasks,
 270                         runServiceCommand{name: "dispatch-cloud", svc: super.cluster.Services.DispatchCloud},
 271                         runGoProgram{src: "services/keep-balance", svc: super.cluster.Services.Keepbalance},
 272                 )
 273         }
 274         super.tasksReady = map[string]chan bool{}
 275         for _, task := range tasks {
 276                 super.tasksReady[task.String()] = make(chan bool)
 277         }
 278         for _, task := range tasks {
 279                 task := task
 280                 fail := func(err error) {
 281                         if super.ctx.Err() != nil {
 282                                 return
 283                         }
 284                         super.cancel()
 285                         super.logger.WithField("task", task.String()).WithError(err).Error("task failed")
 286                 }
 287                 go func() {
 288                         super.logger.WithField("task", task.String()).Info("starting")
 289                         err := task.Run(super.ctx, fail, super)
 290                         if err != nil {
 291                                 fail(err)
 292                                 return
 293                         }
 294                         close(super.tasksReady[task.String()])
 295                 }()
 296         }
 297         err = super.wait(super.ctx, tasks...)
 298         if err != nil {
 299                 return err
 300         }
 301         super.logger.Info("all startup tasks are complete; starting health checks")
 302         super.healthChecker = &health.Aggregator{Cluster: super.cluster}
 303         <-super.ctx.Done()
 304         super.logger.Info("shutting down")
 305         super.waitShutdown.Wait()
 306         return super.ctx.Err()
 307 }
 308
 309 func (super *Supervisor) wait(ctx context.Context, tasks ...supervisedTask) error {
 310         for _, task := range tasks {
 311                 ch, ok := super.tasksReady[task.String()]
 312                 if !ok {
 313                         return fmt.Errorf("no such task: %s", task)
 314                 }
 315                 super.logger.WithField("task", task.String()).Info("waiting")
 316                 select {
 317                 case <-ch:
 318                         super.logger.WithField("task", task.String()).Info("ready")
 319                 case <-ctx.Done():
 320                         super.logger.WithField("task", task.String()).Info("task was never ready")
 321                         return ctx.Err()
 322                 }
 323         }
 324         return nil
 325 }
 326
 327 func (super *Supervisor) Stop() {
 328         super.cancel()
 329         <-super.done
 330 }
 331
 332 func (super *Supervisor) WaitReady() (*arvados.URL, bool) {
 333         ticker := time.NewTicker(time.Second)
 334         defer ticker.Stop()
 335         for waiting := "all"; waiting != ""; {
 336                 select {
 337                 case <-ticker.C:
 338                 case <-super.ctx.Done():
 339                         return nil, false
 340                 }
 341                 if super.healthChecker == nil {
 342                         // not set up yet
 343                         continue
 344                 }
 345                 resp := super.healthChecker.ClusterHealth()
 346                 // The overall health check (resp.Health=="OK") might
 347                 // never pass due to missing components (like
 348                 // arvados-dispatch-cloud in a test cluster), so
 349                 // instead we wait for all configured components to
 350                 // pass.
 351                 waiting = ""
 352                 for target, check := range resp.Checks {
 353                         if check.Health != "OK" {
 354                                 waiting += " " + target
 355                         }
 356                 }
 357                 if waiting != "" {
 358                         super.logger.WithField("targets", waiting[1:]).Info("waiting")
 359                 }
 360         }
 361         u := super.cluster.Services.Controller.ExternalURL
 362         return &u, true
 363 }
 364
 365 func (super *Supervisor) prependEnv(key, prepend string) {
 366         for i, s := range super.environ {
 367                 if strings.HasPrefix(s, key+"=") {
 368                         super.environ[i] = key + "=" + prepend + s[len(key)+1:]
 369                         return
 370                 }
 371         }
 372         super.environ = append(super.environ, key+"="+prepend)
 373 }
 374
 375 func (super *Supervisor) cleanEnv(prefixes []string) {
 376         var cleaned []string
 377         for _, s := range super.environ {
 378                 drop := false
 379                 for _, p := range prefixes {
 380                         if strings.HasPrefix(s, p) {
 381                                 drop = true
 382                                 break
 383                         }
 384                 }
 385                 if !drop {
 386                         cleaned = append(cleaned, s)
 387                 }
 388         }
 389         super.environ = cleaned
 390 }
 391
 392 func (super *Supervisor) setEnv(key, val string) {
 393         for i, s := range super.environ {
 394                 if strings.HasPrefix(s, key+"=") {
 395                         super.environ[i] = key + "=" + val
 396                         return
 397                 }
 398         }
 399         super.environ = append(super.environ, key+"="+val)
 400 }
 401
 402 // Remove all but the first occurrence of each env var.
 403 func dedupEnv(in []string) []string {
 404         saw := map[string]bool{}
 405         var out []string
 406         for _, kv := range in {
 407                 if split := strings.Index(kv, "="); split < 1 {
 408                         panic("invalid environment var: " + kv)
 409                 } else if saw[kv[:split]] {
 410                         continue
 411                 } else {
 412                         saw[kv[:split]] = true
 413                         out = append(out, kv)
 414                 }
 415         }
 416         return out
 417 }
 418
 419 func (super *Supervisor) installGoProgram(ctx context.Context, srcpath string) (string, error) {
 420         _, basename := filepath.Split(srcpath)
 421         binfile := filepath.Join(super.bindir, basename)
 422         if super.ClusterType == "production" {
 423                 return binfile, nil
 424         }
 425         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)
 426         return binfile, err
 427 }
 428
 429 func (super *Supervisor) usingRVM() bool {
 430         return os.Getenv("rvm_path") != ""
 431 }
 432
 433 func (super *Supervisor) setupRubyEnv() error {
 434         if !super.usingRVM() {
 435                 // (If rvm is in use, assume the caller has everything
 436                 // set up as desired)
 437                 super.cleanEnv([]string{
 438                         "GEM_HOME=",
 439                         "GEM_PATH=",
 440                 })
 441                 gem := "gem"
 442                 if _, err := os.Stat("/var/lib/arvados/bin/gem"); err == nil || super.ClusterType == "production" {
 443                         gem = "/var/lib/arvados/bin/gem"
 444                 }
 445                 cmd := exec.Command(gem, "env", "gempath")
 446                 if super.ClusterType == "production" {
 447                         cmd.Args = append([]string{"sudo", "-u", "www-data", "-E", "HOME=/var/www"}, cmd.Args...)
 448                         path, err := exec.LookPath("sudo")
 449                         if err != nil {
 450                                 return fmt.Errorf("LookPath(\"sudo\"): %w", err)
 451                         }
 452                         cmd.Path = path
 453                 }
 454                 cmd.Stderr = super.Stderr
 455                 cmd.Env = super.environ
 456                 buf, err := cmd.Output() // /var/lib/arvados/.gem/ruby/2.5.0/bin:...
 457                 if err != nil || len(buf) == 0 {
 458                         return fmt.Errorf("gem env gempath: %w", err)
 459                 }
 460                 gempath := string(bytes.Split(buf, []byte{':'})[0])
 461                 super.prependEnv("PATH", gempath+"/bin:")
 462                 super.setEnv("GEM_HOME", gempath)
 463                 super.setEnv("GEM_PATH", gempath)
 464         }
 465         // Passenger install doesn't work unless $HOME is ~user
 466         u, err := user.Current()
 467         if err != nil {
 468                 return err
 469         }
 470         super.setEnv("HOME", u.HomeDir)
 471         return nil
 472 }
 473
 474 func (super *Supervisor) lookPath(prog string) string {
 475         for _, val := range super.environ {
 476                 if strings.HasPrefix(val, "PATH=") {
 477                         for _, dir := range filepath.SplitList(val[5:]) {
 478                                 path := filepath.Join(dir, prog)
 479                                 if fi, err := os.Stat(path); err == nil && fi.Mode()&0111 != 0 {
 480                                         return path
 481                                 }
 482                         }
 483                 }
 484         }
 485         return prog
 486 }
 487
 488 type runOptions struct {
 489         output io.Writer // attach stdout
 490         env    []string  // add/replace environment variables
 491         user   string    // run as specified user
 492         stdin  io.Reader
 493 }
 494
 495 // RunProgram runs prog with args, using dir as working directory. If ctx is
 496 // cancelled while the child is running, RunProgram terminates the child, waits
 497 // for it to exit, then returns.
 498 //
 499 // Child's environment will have our env vars, plus any given in env.
 500 //
 501 // Child's stdout will be written to output if non-nil, otherwise the
 502 // boot command's stderr.
 503 func (super *Supervisor) RunProgram(ctx context.Context, dir string, opts runOptions, prog string, args ...string) error {
 504         cmdline := fmt.Sprintf("%s", append([]string{prog}, args...))
 505         super.logger.WithField("command", cmdline).WithField("dir", dir).Info("executing")
 506
 507         logprefix := prog
 508         {
 509                 innerargs := args
 510                 if logprefix == "sudo" {
 511                         for i := 0; i < len(args); i++ {
 512                                 if args[i] == "-u" {
 513                                         i++
 514                                 } else if args[i] == "-E" || strings.Contains(args[i], "=") {
 515                                 } else {
 516                                         logprefix = args[i]
 517                                         innerargs = args[i+1:]
 518                                         break
 519                                 }
 520                         }
 521                 }
 522                 logprefix = strings.TrimPrefix(logprefix, "/var/lib/arvados/bin/")
 523                 logprefix = strings.TrimPrefix(logprefix, super.tempdir+"/bin/")
 524                 if logprefix == "bundle" && len(innerargs) > 2 && innerargs[0] == "exec" {
 525                         _, dirbase := filepath.Split(dir)
 526                         logprefix = innerargs[1] + "@" + dirbase
 527                 } else if logprefix == "arvados-server" && len(args) > 1 {
 528                         logprefix = args[0]
 529                 }
 530                 if !strings.HasPrefix(dir, "/") {
 531                         logprefix = dir + ": " + logprefix
 532                 }
 533         }
 534
 535         cmd := exec.Command(super.lookPath(prog), args...)
 536         cmd.Stdin = opts.stdin
 537         stdout, err := cmd.StdoutPipe()
 538         if err != nil {
 539                 return err
 540         }
 541         stderr, err := cmd.StderrPipe()
 542         if err != nil {
 543                 return err
 544         }
 545         logwriter := &service.LogPrefixer{Writer: super.Stderr, Prefix: []byte("[" + logprefix + "] ")}
 546         var copiers sync.WaitGroup
 547         copiers.Add(1)
 548         go func() {
 549                 io.Copy(logwriter, stderr)
 550                 copiers.Done()
 551         }()
 552         copiers.Add(1)
 553         go func() {
 554                 if opts.output == nil {
 555                         io.Copy(logwriter, stdout)
 556                 } else {
 557                         io.Copy(opts.output, stdout)
 558                 }
 559                 copiers.Done()
 560         }()
 561
 562         if strings.HasPrefix(dir, "/") {
 563                 cmd.Dir = dir
 564         } else {
 565                 cmd.Dir = filepath.Join(super.SourcePath, dir)
 566         }
 567         env := append([]string(nil), opts.env...)
 568         env = append(env, super.environ...)
 569         cmd.Env = dedupEnv(env)
 570
 571         if opts.user != "" {
 572                 // Note: We use this approach instead of "sudo"
 573                 // because in certain circumstances (we are pid 1 in a
 574                 // docker container, and our passenger child process
 575                 // changes to pgid 1) the intermediate sudo process
 576                 // notices we have the same pgid as our child and
 577                 // refuses to propagate signals from us to our child,
 578                 // so we can't signal/shutdown our passenger/rails
 579                 // apps. "chpst" or "setuidgid" would work, but these
 580                 // few lines avoid depending on runit/daemontools.
 581                 u, err := user.Lookup(opts.user)
 582                 if err != nil {
 583                         return fmt.Errorf("user.Lookup(%q): %w", opts.user, err)
 584                 }
 585                 uid, _ := strconv.Atoi(u.Uid)
 586                 gid, _ := strconv.Atoi(u.Gid)
 587                 cmd.SysProcAttr = &syscall.SysProcAttr{
 588                         Credential: &syscall.Credential{
 589                                 Uid: uint32(uid),
 590                                 Gid: uint32(gid),
 591                         },
 592                 }
 593         }
 594
 595         exited := false
 596         defer func() { exited = true }()
 597         go func() {
 598                 <-ctx.Done()
 599                 log := ctxlog.FromContext(ctx).WithFields(logrus.Fields{"dir": dir, "cmdline": cmdline})
 600                 for !exited {
 601                         if cmd.Process == nil {
 602                                 log.Debug("waiting for child process to start")
 603                                 time.Sleep(time.Second / 2)
 604                         } else {
 605                                 log.WithField("PID", cmd.Process.Pid).Debug("sending SIGTERM")
 606                                 cmd.Process.Signal(syscall.SIGTERM)
 607                                 time.Sleep(5 * time.Second)
 608                                 if !exited {
 609                                         stdout.Close()
 610                                         stderr.Close()
 611                                         log.WithField("PID", cmd.Process.Pid).Warn("still waiting for child process to exit 5s after SIGTERM")
 612                                 }
 613                         }
 614                 }
 615         }()
 616
 617         err = cmd.Start()
 618         if err != nil {
 619                 return err
 620         }
 621         copiers.Wait()
 622         err = cmd.Wait()
 623         if ctx.Err() != nil {
 624                 // Return "context canceled", instead of the "killed"
 625                 // error that was probably caused by the context being
 626                 // canceled.
 627                 return ctx.Err()
 628         } else if err != nil {
 629                 return fmt.Errorf("%s: error: %v", cmdline, err)
 630         }
 631         return nil
 632 }
 633
 634 func (super *Supervisor) autofillConfig(cfg *arvados.Config) error {
 635         cluster, err := cfg.GetCluster("")
 636         if err != nil {
 637                 return err
 638         }
 639         usedPort := map[string]bool{}
 640         nextPort := func(host string) (string, error) {
 641                 for {
 642                         port, err := availablePort(host)
 643                         if err != nil {
 644                                 port, err = availablePort(super.ListenHost)
 645                         }
 646                         if err != nil {
 647                                 return "", err
 648                         }
 649                         if usedPort[port] {
 650                                 continue
 651                         }
 652                         usedPort[port] = true
 653                         return port, nil
 654                 }
 655         }
 656         if cluster.Services.Controller.ExternalURL.Host == "" {
 657                 h, p, err := net.SplitHostPort(super.ControllerAddr)
 658                 if err != nil {
 659                         return fmt.Errorf("SplitHostPort(ControllerAddr): %w", err)
 660                 }
 661                 if h == "" {
 662                         h = super.ListenHost
 663                 }
 664                 if p == "0" {
 665                         p, err = nextPort(h)
 666                         if err != nil {
 667                                 return err
 668                         }
 669                 }
 670                 cluster.Services.Controller.ExternalURL = arvados.URL{Scheme: "https", Host: net.JoinHostPort(h, p), Path: "/"}
 671         }
 672         u := url.URL(cluster.Services.Controller.ExternalURL)
 673         defaultExtHost := u.Hostname()
 674         for _, svc := range []*arvados.Service{
 675                 &cluster.Services.Controller,
 676                 &cluster.Services.DispatchCloud,
 677                 &cluster.Services.GitHTTP,
 678                 &cluster.Services.Health,
 679                 &cluster.Services.Keepproxy,
 680                 &cluster.Services.Keepstore,
 681                 &cluster.Services.RailsAPI,
 682                 &cluster.Services.WebDAV,
 683                 &cluster.Services.WebDAVDownload,
 684                 &cluster.Services.Websocket,
 685                 &cluster.Services.Workbench1,
 686                 &cluster.Services.Workbench2,
 687         } {
 688                 if svc == &cluster.Services.DispatchCloud && super.ClusterType == "test" {
 689                         continue
 690                 }
 691                 if svc.ExternalURL.Host == "" {
 692                         port, err := nextPort(defaultExtHost)
 693                         if err != nil {
 694                                 return err
 695                         }
 696                         host := net.JoinHostPort(defaultExtHost, port)
 697                         if svc == &cluster.Services.Controller ||
 698                                 svc == &cluster.Services.GitHTTP ||
 699                                 svc == &cluster.Services.Health ||
 700                                 svc == &cluster.Services.Keepproxy ||
 701                                 svc == &cluster.Services.WebDAV ||
 702                                 svc == &cluster.Services.WebDAVDownload ||
 703                                 svc == &cluster.Services.Workbench1 ||
 704                                 svc == &cluster.Services.Workbench2 {
 705                                 svc.ExternalURL = arvados.URL{Scheme: "https", Host: host, Path: "/"}
 706                         } else if svc == &cluster.Services.Websocket {
 707                                 svc.ExternalURL = arvados.URL{Scheme: "wss", Host: host, Path: "/websocket"}
 708                         }
 709                 }
 710                 if super.NoWorkbench1 && svc == &cluster.Services.Workbench1 ||
 711                         super.NoWorkbench2 && svc == &cluster.Services.Workbench2 {
 712                         // When workbench1 is disabled, it gets an
 713                         // ExternalURL (so we have a valid listening
 714                         // port to write in our Nginx config) but no
 715                         // InternalURLs (so health checker doesn't
 716                         // complain).
 717                         continue
 718                 }
 719                 if len(svc.InternalURLs) == 0 {
 720                         port, err := nextPort(super.ListenHost)
 721                         if err != nil {
 722                                 return err
 723                         }
 724                         host := net.JoinHostPort(super.ListenHost, port)
 725                         svc.InternalURLs = map[arvados.URL]arvados.ServiceInstance{
 726                                 {Scheme: "http", Host: host, Path: "/"}: {},
 727                         }
 728                 }
 729         }
 730         if super.ClusterType != "production" {
 731                 if cluster.SystemRootToken == "" {
 732                         cluster.SystemRootToken = randomHexString(64)
 733                 }
 734                 if cluster.ManagementToken == "" {
 735                         cluster.ManagementToken = randomHexString(64)
 736                 }
 737                 if cluster.Collections.BlobSigningKey == "" {
 738                         cluster.Collections.BlobSigningKey = randomHexString(64)
 739                 }
 740                 if cluster.Users.AnonymousUserToken == "" {
 741                         cluster.Users.AnonymousUserToken = randomHexString(64)
 742                 }
 743                 if cluster.Containers.DispatchPrivateKey == "" {
 744                         buf, err := ioutil.ReadFile(filepath.Join(super.SourcePath, "lib", "dispatchcloud", "test", "sshkey_dispatch"))
 745                         if err != nil {
 746                                 return err
 747                         }
 748                         cluster.Containers.DispatchPrivateKey = string(buf)
 749                 }
 750                 cluster.TLS.Insecure = true
 751         }
 752         if super.ClusterType == "test" {
 753                 // Add a second keepstore process.
 754                 port, err := nextPort(super.ListenHost)
 755                 if err != nil {
 756                         return err
 757                 }
 758                 host := net.JoinHostPort(super.ListenHost, port)
 759                 cluster.Services.Keepstore.InternalURLs[arvados.URL{Scheme: "http", Host: host, Path: "/"}] = arvados.ServiceInstance{}
 760
 761                 // Create a directory-backed volume for each keepstore
 762                 // process.
 763                 cluster.Volumes = map[string]arvados.Volume{}
 764                 for url := range cluster.Services.Keepstore.InternalURLs {
 765                         volnum := len(cluster.Volumes)
 766                         datadir := fmt.Sprintf("%s/keep%d.data", super.tempdir, volnum)
 767                         if _, err = os.Stat(datadir + "/."); err == nil {
 768                         } else if !os.IsNotExist(err) {
 769                                 return err
 770                         } else if err = os.Mkdir(datadir, 0755); err != nil {
 771                                 return err
 772                         }
 773                         cluster.Volumes[fmt.Sprintf(cluster.ClusterID+"-nyw5e-%015d", volnum)] = arvados.Volume{
 774                                 Driver:           "Directory",
 775                                 DriverParameters: json.RawMessage(fmt.Sprintf(`{"Root":%q}`, datadir)),
 776                                 AccessViaHosts: map[arvados.URL]arvados.VolumeAccess{
 777                                         url: {},
 778                                 },
 779                                 StorageClasses: map[string]bool{
 780                                         "default": true,
 781                                         "foo":     true,
 782                                         "bar":     true,
 783                                 },
 784                         }
 785                 }
 786                 cluster.StorageClasses = map[string]arvados.StorageClassConfig{
 787                         "default": {Default: true},
 788                         "foo":     {},
 789                         "bar":     {},
 790                 }
 791         }
 792         if super.OwnTemporaryDatabase {
 793                 port, err := nextPort("localhost")
 794                 if err != nil {
 795                         return err
 796                 }
 797                 cluster.PostgreSQL.Connection = arvados.PostgreSQLConnection{
 798                         "client_encoding": "utf8",
 799                         "host":            "localhost",
 800                         "port":            port,
 801                         "dbname":          "arvados_test",
 802                         "user":            "arvados",
 803                         "password":        "insecure_arvados_test",
 804                 }
 805         }
 806
 807         cfg.Clusters[cluster.ClusterID] = *cluster
 808         return nil
 809 }
 810
 811 func addrIsLocal(addr string) (bool, error) {
 812         return true, nil
 813         listener, err := net.Listen("tcp", addr)
 814         if err == nil {
 815                 listener.Close()
 816                 return true, nil
 817         } else if strings.Contains(err.Error(), "cannot assign requested address") {
 818                 return false, nil
 819         } else {
 820                 return false, err
 821         }
 822 }
 823
 824 func randomHexString(chars int) string {
 825         b := make([]byte, chars/2)
 826         _, err := rand.Read(b)
 827         if err != nil {
 828                 panic(err)
 829         }
 830         return fmt.Sprintf("%x", b)
 831 }
 832
 833 func internalPort(svc arvados.Service) (host, port string, err error) {
 834         if len(svc.InternalURLs) > 1 {
 835                 return "", "", errors.New("internalPort() doesn't work with multiple InternalURLs")
 836         }
 837         for u := range svc.InternalURLs {
 838                 u := url.URL(u)
 839                 host, port = u.Hostname(), u.Port()
 840                 switch {
 841                 case port != "":
 842                 case u.Scheme == "https", u.Scheme == "ws":
 843                         port = "443"
 844                 default:
 845                         port = "80"
 846                 }
 847                 return
 848         }
 849         return "", "", fmt.Errorf("service has no InternalURLs")
 850 }
 851
 852 func externalPort(svc arvados.Service) (string, error) {
 853         u := url.URL(svc.ExternalURL)
 854         if p := u.Port(); p != "" {
 855                 return p, nil
 856         } else if u.Scheme == "https" || u.Scheme == "wss" {
 857                 return "443", nil
 858         } else {
 859                 return "80", nil
 860         }
 861 }
 862
 863 func availablePort(host string) (string, error) {
 864         ln, err := net.Listen("tcp", net.JoinHostPort(host, "0"))
 865         if err != nil {
 866                 return "", err
 867         }
 868         defer ln.Close()
 869         _, port, err := net.SplitHostPort(ln.Addr().String())
 870         if err != nil {
 871                 return "", err
 872         }
 873         return port, nil
 874 }
 875
 876 // Try to connect to addr until it works, then close ch. Give up if
 877 // ctx cancels.
 878 func waitForConnect(ctx context.Context, addr string) error {
 879         dialer := net.Dialer{Timeout: time.Second}
 880         for ctx.Err() == nil {
 881                 conn, err := dialer.DialContext(ctx, "tcp", addr)
 882                 if err != nil {
 883                         time.Sleep(time.Second / 10)
 884                         continue
 885                 }
 886                 conn.Close()
 887                 return nil
 888         }
 889         return ctx.Err()
 890 }
 891
 892 func copyConfig(cfg *arvados.Config) *arvados.Config {
 893         pr, pw := io.Pipe()
 894         go func() {
 895                 err := json.NewEncoder(pw).Encode(cfg)
 896                 if err != nil {
 897                         panic(err)
 898                 }
 899                 pw.Close()
 900         }()
 901         cfg2 := new(arvados.Config)
 902         err := json.NewDecoder(pr).Decode(cfg2)
 903         if err != nil {
 904                 panic(err)
 905         }
 906         return cfg2
 907 }
 908
 909 func watchConfig(ctx context.Context, logger logrus.FieldLogger, cfgPath string, prevcfg *arvados.Config, fn func()) {
 910         watcher, err := fsnotify.NewWatcher()
 911         if err != nil {
 912                 logger.WithError(err).Error("fsnotify setup failed")
 913                 return
 914         }
 915         defer watcher.Close()
 916
 917         err = watcher.Add(cfgPath)
 918         if err != nil {
 919                 logger.WithError(err).Error("fsnotify watcher failed")
 920                 return
 921         }
 922
 923         for {
 924                 select {
 925                 case <-ctx.Done():
 926                         return
 927                 case err, ok := <-watcher.Errors:
 928                         if !ok {
 929                                 return
 930                         }
 931                         logger.WithError(err).Warn("fsnotify watcher reported error")
 932                 case _, ok := <-watcher.Events:
 933                         if !ok {
 934                                 return
 935                         }
 936                         for len(watcher.Events) > 0 {
 937                                 <-watcher.Events
 938                         }
 939                         loader := config.NewLoader(&bytes.Buffer{}, &logrus.Logger{Out: ioutil.Discard})
 940                         loader.Path = cfgPath
 941                         loader.SkipAPICalls = true
 942                         cfg, err := loader.Load()
 943                         if err != nil {
 944                                 logger.WithError(err).Warn("error reloading config file after change detected; ignoring new config for now")
 945                         } else if reflect.DeepEqual(cfg, prevcfg) {
 946                                 logger.Debug("config file changed but is still DeepEqual to the existing config")
 947                         } else {
 948                                 logger.Debug("config changed, notifying supervisor")
 949                                 fn()
 950                                 prevcfg = cfg
 951                         }
 952                 }
 953         }
 954 }