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         OwnTemporaryDatabase bool
  46         Stderr               io.Writer
  47
  48         logger  logrus.FieldLogger
  49         cluster *arvados.Cluster
  50
  51         ctx           context.Context
  52         cancel        context.CancelFunc
  53         done          chan struct{} // closed when child procs/services have shut down
  54         err           error         // error that caused shutdown (valid when done is closed)
  55         healthChecker *health.Aggregator
  56         tasksReady    map[string]chan bool
  57         waitShutdown  sync.WaitGroup
  58
  59         bindir     string
  60         tempdir    string
  61         wwwtempdir string
  62         configfile string
  63         environ    []string // for child processes
  64 }
  65
  66 func (super *Supervisor) Start(ctx context.Context, cfg *arvados.Config, cfgPath string) {
  67         super.ctx, super.cancel = context.WithCancel(ctx)
  68         super.done = make(chan struct{})
  69
  70         go func() {
  71                 defer close(super.done)
  72
  73                 sigch := make(chan os.Signal)
  74                 signal.Notify(sigch, syscall.SIGINT, syscall.SIGTERM)
  75                 defer signal.Stop(sigch)
  76                 go func() {
  77                         for sig := range sigch {
  78                                 super.logger.WithField("signal", sig).Info("caught signal")
  79                                 if super.err == nil {
  80                                         super.err = fmt.Errorf("caught signal %s", sig)
  81                                 }
  82                                 super.cancel()
  83                         }
  84                 }()
  85
  86                 hupch := make(chan os.Signal)
  87                 signal.Notify(hupch, syscall.SIGHUP)
  88                 defer signal.Stop(hupch)
  89                 go func() {
  90                         for sig := range hupch {
  91                                 super.logger.WithField("signal", sig).Info("caught signal")
  92                                 if super.err == nil {
  93                                         super.err = errNeedConfigReload
  94                                 }
  95                                 super.cancel()
  96                         }
  97                 }()
  98
  99                 if cfgPath != "" && cfgPath != "-" && cfg.AutoReloadConfig {
 100                         go watchConfig(super.ctx, super.logger, cfgPath, copyConfig(cfg), func() {
 101                                 if super.err == nil {
 102                                         super.err = errNeedConfigReload
 103                                 }
 104                                 super.cancel()
 105                         })
 106                 }
 107
 108                 err := super.run(cfg)
 109                 if err != nil {
 110                         super.logger.WithError(err).Warn("supervisor shut down")
 111                         if super.err == nil {
 112                                 super.err = err
 113                         }
 114                 }
 115         }()
 116 }
 117
 118 func (super *Supervisor) Wait() error {
 119         <-super.done
 120         return super.err
 121 }
 122
 123 func (super *Supervisor) run(cfg *arvados.Config) error {
 124         defer super.cancel()
 125
 126         cwd, err := os.Getwd()
 127         if err != nil {
 128                 return err
 129         }
 130         if !strings.HasPrefix(super.SourcePath, "/") {
 131                 super.SourcePath = filepath.Join(cwd, super.SourcePath)
 132         }
 133         super.SourcePath, err = filepath.EvalSymlinks(super.SourcePath)
 134         if err != nil {
 135                 return err
 136         }
 137
 138         // Choose bin and temp dirs: /var/lib/arvados/... in
 139         // production, transient tempdir otherwise.
 140         if super.ClusterType == "production" {
 141                 // These dirs have already been created by
 142                 // "arvados-server install" (or by extracting a
 143                 // package).
 144                 super.tempdir = "/var/lib/arvados/tmp"
 145                 super.wwwtempdir = "/var/lib/arvados/wwwtmp"
 146                 super.bindir = "/var/lib/arvados/bin"
 147         } else {
 148                 super.tempdir, err = ioutil.TempDir("", "arvados-server-boot-")
 149                 if err != nil {
 150                         return err
 151                 }
 152                 defer os.RemoveAll(super.tempdir)
 153                 super.wwwtempdir = super.tempdir
 154                 super.bindir = filepath.Join(super.tempdir, "bin")
 155                 if err := os.Mkdir(super.bindir, 0755); err != nil {
 156                         return err
 157                 }
 158         }
 159
 160         // Fill in any missing config keys, and write the resulting
 161         // config in the temp dir for child services to use.
 162         err = super.autofillConfig(cfg)
 163         if err != nil {
 164                 return err
 165         }
 166         conffile, err := os.OpenFile(filepath.Join(super.wwwtempdir, "config.yml"), os.O_CREATE|os.O_WRONLY, 0644)
 167         if err != nil {
 168                 return err
 169         }
 170         defer conffile.Close()
 171         err = json.NewEncoder(conffile).Encode(cfg)
 172         if err != nil {
 173                 return err
 174         }
 175         err = conffile.Close()
 176         if err != nil {
 177                 return err
 178         }
 179         super.configfile = conffile.Name()
 180
 181         super.environ = os.Environ()
 182         super.cleanEnv([]string{"ARVADOS_"})
 183         super.setEnv("ARVADOS_CONFIG", super.configfile)
 184         super.setEnv("RAILS_ENV", super.ClusterType)
 185         super.setEnv("TMPDIR", super.tempdir)
 186         super.prependEnv("PATH", "/var/lib/arvados/bin:")
 187         if super.ClusterType != "production" {
 188                 super.prependEnv("PATH", super.tempdir+"/bin:")
 189         }
 190
 191         super.cluster, err = cfg.GetCluster("")
 192         if err != nil {
 193                 return err
 194         }
 195         // Now that we have the config, replace the bootstrap logger
 196         // with a new one according to the logging config.
 197         loglevel := super.cluster.SystemLogs.LogLevel
 198         if s := os.Getenv("ARVADOS_DEBUG"); s != "" && s != "0" {
 199                 loglevel = "debug"
 200         }
 201         super.logger = ctxlog.New(super.Stderr, super.cluster.SystemLogs.Format, loglevel).WithFields(logrus.Fields{
 202                 "PID": os.Getpid(),
 203         })
 204
 205         if super.SourceVersion == "" && super.ClusterType == "production" {
 206                 // don't need SourceVersion
 207         } else if super.SourceVersion == "" {
 208                 // Find current source tree version.
 209                 var buf bytes.Buffer
 210                 err = super.RunProgram(super.ctx, ".", runOptions{output: &buf}, "git", "diff", "--shortstat")
 211                 if err != nil {
 212                         return err
 213                 }
 214                 dirty := buf.Len() > 0
 215                 buf.Reset()
 216                 err = super.RunProgram(super.ctx, ".", runOptions{output: &buf}, "git", "log", "-n1", "--format=%H")
 217                 if err != nil {
 218                         return err
 219                 }
 220                 super.SourceVersion = strings.TrimSpace(buf.String())
 221                 if dirty {
 222                         super.SourceVersion += "+uncommitted"
 223                 }
 224         } else {
 225                 return errors.New("specifying a version to run is not yet supported")
 226         }
 227
 228         _, err = super.installGoProgram(super.ctx, "cmd/arvados-server")
 229         if err != nil {
 230                 return err
 231         }
 232         err = super.setupRubyEnv()
 233         if err != nil {
 234                 return err
 235         }
 236
 237         tasks := []supervisedTask{
 238                 createCertificates{},
 239                 runPostgreSQL{},
 240                 runNginx{},
 241                 runServiceCommand{name: "controller", svc: super.cluster.Services.Controller, depends: []supervisedTask{runPostgreSQL{}}},
 242                 runGoProgram{src: "services/arv-git-httpd", svc: super.cluster.Services.GitHTTP},
 243                 runGoProgram{src: "services/health", svc: super.cluster.Services.Health},
 244                 runGoProgram{src: "services/keepproxy", svc: super.cluster.Services.Keepproxy, depends: []supervisedTask{runPassenger{src: "services/api"}}},
 245                 runGoProgram{src: "services/keepstore", svc: super.cluster.Services.Keepstore},
 246                 runGoProgram{src: "services/keep-web", svc: super.cluster.Services.WebDAV},
 247                 runServiceCommand{name: "ws", svc: super.cluster.Services.Websocket, depends: []supervisedTask{runPostgreSQL{}}},
 248                 installPassenger{src: "services/api"},
 249                 runPassenger{src: "services/api", varlibdir: "railsapi", svc: super.cluster.Services.RailsAPI, depends: []supervisedTask{createCertificates{}, runPostgreSQL{}, installPassenger{src: "services/api"}}},
 250                 installPassenger{src: "apps/workbench", depends: []supervisedTask{installPassenger{src: "services/api"}}}, // dependency ensures workbench doesn't delay api startup
 251                 runPassenger{src: "apps/workbench", varlibdir: "workbench1", svc: super.cluster.Services.Workbench1, depends: []supervisedTask{installPassenger{src: "apps/workbench"}}},
 252                 seedDatabase{},
 253         }
 254         if super.ClusterType != "test" {
 255                 tasks = append(tasks,
 256                         runServiceCommand{name: "dispatch-cloud", svc: super.cluster.Services.DispatchCloud},
 257                         runGoProgram{src: "services/keep-balance", svc: super.cluster.Services.Keepbalance},
 258                 )
 259         }
 260         super.tasksReady = map[string]chan bool{}
 261         for _, task := range tasks {
 262                 super.tasksReady[task.String()] = make(chan bool)
 263         }
 264         for _, task := range tasks {
 265                 task := task
 266                 fail := func(err error) {
 267                         if super.ctx.Err() != nil {
 268                                 return
 269                         }
 270                         super.cancel()
 271                         super.logger.WithField("task", task.String()).WithError(err).Error("task failed")
 272                 }
 273                 go func() {
 274                         super.logger.WithField("task", task.String()).Info("starting")
 275                         err := task.Run(super.ctx, fail, super)
 276                         if err != nil {
 277                                 fail(err)
 278                                 return
 279                         }
 280                         close(super.tasksReady[task.String()])
 281                 }()
 282         }
 283         err = super.wait(super.ctx, tasks...)
 284         if err != nil {
 285                 return err
 286         }
 287         super.logger.Info("all startup tasks are complete; starting health checks")
 288         super.healthChecker = &health.Aggregator{Cluster: super.cluster}
 289         <-super.ctx.Done()
 290         super.logger.Info("shutting down")
 291         super.waitShutdown.Wait()
 292         return super.ctx.Err()
 293 }
 294
 295 func (super *Supervisor) wait(ctx context.Context, tasks ...supervisedTask) error {
 296         for _, task := range tasks {
 297                 ch, ok := super.tasksReady[task.String()]
 298                 if !ok {
 299                         return fmt.Errorf("no such task: %s", task)
 300                 }
 301                 super.logger.WithField("task", task.String()).Info("waiting")
 302                 select {
 303                 case <-ch:
 304                         super.logger.WithField("task", task.String()).Info("ready")
 305                 case <-ctx.Done():
 306                         super.logger.WithField("task", task.String()).Info("task was never ready")
 307                         return ctx.Err()
 308                 }
 309         }
 310         return nil
 311 }
 312
 313 func (super *Supervisor) Stop() {
 314         super.cancel()
 315         <-super.done
 316 }
 317
 318 func (super *Supervisor) WaitReady() (*arvados.URL, bool) {
 319         ticker := time.NewTicker(time.Second)
 320         defer ticker.Stop()
 321         for waiting := "all"; waiting != ""; {
 322                 select {
 323                 case <-ticker.C:
 324                 case <-super.ctx.Done():
 325                         return nil, false
 326                 }
 327                 if super.healthChecker == nil {
 328                         // not set up yet
 329                         continue
 330                 }
 331                 resp := super.healthChecker.ClusterHealth()
 332                 // The overall health check (resp.Health=="OK") might
 333                 // never pass due to missing components (like
 334                 // arvados-dispatch-cloud in a test cluster), so
 335                 // instead we wait for all configured components to
 336                 // pass.
 337                 waiting = ""
 338                 for target, check := range resp.Checks {
 339                         if check.Health != "OK" {
 340                                 waiting += " " + target
 341                         }
 342                 }
 343                 if waiting != "" {
 344                         super.logger.WithField("targets", waiting[1:]).Info("waiting")
 345                 }
 346         }
 347         u := super.cluster.Services.Controller.ExternalURL
 348         return &u, true
 349 }
 350
 351 func (super *Supervisor) prependEnv(key, prepend string) {
 352         for i, s := range super.environ {
 353                 if strings.HasPrefix(s, key+"=") {
 354                         super.environ[i] = key + "=" + prepend + s[len(key)+1:]
 355                         return
 356                 }
 357         }
 358         super.environ = append(super.environ, key+"="+prepend)
 359 }
 360
 361 func (super *Supervisor) cleanEnv(prefixes []string) {
 362         var cleaned []string
 363         for _, s := range super.environ {
 364                 drop := false
 365                 for _, p := range prefixes {
 366                         if strings.HasPrefix(s, p) {
 367                                 drop = true
 368                                 break
 369                         }
 370                 }
 371                 if !drop {
 372                         cleaned = append(cleaned, s)
 373                 }
 374         }
 375         super.environ = cleaned
 376 }
 377
 378 func (super *Supervisor) setEnv(key, val string) {
 379         for i, s := range super.environ {
 380                 if strings.HasPrefix(s, key+"=") {
 381                         super.environ[i] = key + "=" + val
 382                         return
 383                 }
 384         }
 385         super.environ = append(super.environ, key+"="+val)
 386 }
 387
 388 // Remove all but the first occurrence of each env var.
 389 func dedupEnv(in []string) []string {
 390         saw := map[string]bool{}
 391         var out []string
 392         for _, kv := range in {
 393                 if split := strings.Index(kv, "="); split < 1 {
 394                         panic("invalid environment var: " + kv)
 395                 } else if saw[kv[:split]] {
 396                         continue
 397                 } else {
 398                         saw[kv[:split]] = true
 399                         out = append(out, kv)
 400                 }
 401         }
 402         return out
 403 }
 404
 405 func (super *Supervisor) installGoProgram(ctx context.Context, srcpath string) (string, error) {
 406         _, basename := filepath.Split(srcpath)
 407         binfile := filepath.Join(super.bindir, basename)
 408         if super.ClusterType == "production" {
 409                 return binfile, nil
 410         }
 411         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)
 412         return binfile, err
 413 }
 414
 415 func (super *Supervisor) usingRVM() bool {
 416         return os.Getenv("rvm_path") != ""
 417 }
 418
 419 func (super *Supervisor) setupRubyEnv() error {
 420         if !super.usingRVM() {
 421                 // (If rvm is in use, assume the caller has everything
 422                 // set up as desired)
 423                 super.cleanEnv([]string{
 424                         "GEM_HOME=",
 425                         "GEM_PATH=",
 426                 })
 427                 gem := "gem"
 428                 if _, err := os.Stat("/var/lib/arvados/bin/gem"); err == nil || super.ClusterType == "production" {
 429                         gem = "/var/lib/arvados/bin/gem"
 430                 }
 431                 cmd := exec.Command(gem, "env", "gempath")
 432                 if super.ClusterType == "production" {
 433                         cmd.Args = append([]string{"sudo", "-u", "www-data", "-E", "HOME=/var/www"}, cmd.Args...)
 434                         path, err := exec.LookPath("sudo")
 435                         if err != nil {
 436                                 return fmt.Errorf("LookPath(\"sudo\"): %w", err)
 437                         }
 438                         cmd.Path = path
 439                 }
 440                 cmd.Stderr = super.Stderr
 441                 cmd.Env = super.environ
 442                 buf, err := cmd.Output() // /var/lib/arvados/.gem/ruby/2.5.0/bin:...
 443                 if err != nil || len(buf) == 0 {
 444                         return fmt.Errorf("gem env gempath: %v", err)
 445                 }
 446                 gempath := string(bytes.Split(buf, []byte{':'})[0])
 447                 super.prependEnv("PATH", gempath+"/bin:")
 448                 super.setEnv("GEM_HOME", gempath)
 449                 super.setEnv("GEM_PATH", gempath)
 450         }
 451         // Passenger install doesn't work unless $HOME is ~user
 452         u, err := user.Current()
 453         if err != nil {
 454                 return err
 455         }
 456         super.setEnv("HOME", u.HomeDir)
 457         return nil
 458 }
 459
 460 func (super *Supervisor) lookPath(prog string) string {
 461         for _, val := range super.environ {
 462                 if strings.HasPrefix(val, "PATH=") {
 463                         for _, dir := range filepath.SplitList(val[5:]) {
 464                                 path := filepath.Join(dir, prog)
 465                                 if fi, err := os.Stat(path); err == nil && fi.Mode()&0111 != 0 {
 466                                         return path
 467                                 }
 468                         }
 469                 }
 470         }
 471         return prog
 472 }
 473
 474 type runOptions struct {
 475         output io.Writer // attach stdout
 476         env    []string  // add/replace environment variables
 477         user   string    // run as specified user
 478 }
 479
 480 // RunProgram runs prog with args, using dir as working directory. If ctx is
 481 // cancelled while the child is running, RunProgram terminates the child, waits
 482 // for it to exit, then returns.
 483 //
 484 // Child's environment will have our env vars, plus any given in env.
 485 //
 486 // Child's stdout will be written to output if non-nil, otherwise the
 487 // boot command's stderr.
 488 func (super *Supervisor) RunProgram(ctx context.Context, dir string, opts runOptions, prog string, args ...string) error {
 489         cmdline := fmt.Sprintf("%s", append([]string{prog}, args...))
 490         super.logger.WithField("command", cmdline).WithField("dir", dir).Info("executing")
 491
 492         logprefix := prog
 493         {
 494                 if logprefix == "setuidgid" && len(args) >= 3 {
 495                         logprefix = args[2]
 496                 }
 497                 innerargs := args
 498                 if logprefix == "sudo" {
 499                         for i := 0; i < len(args); i++ {
 500                                 if args[i] == "-u" {
 501                                         i++
 502                                 } else if args[i] == "-E" || strings.Contains(args[i], "=") {
 503                                 } else {
 504                                         logprefix = args[i]
 505                                         innerargs = args[i+1:]
 506                                         break
 507                                 }
 508                         }
 509                 }
 510                 logprefix = strings.TrimPrefix(logprefix, "/var/lib/arvados/bin/")
 511                 logprefix = strings.TrimPrefix(logprefix, super.tempdir+"/bin/")
 512                 if logprefix == "bundle" && len(innerargs) > 2 && innerargs[0] == "exec" {
 513                         _, dirbase := filepath.Split(dir)
 514                         logprefix = innerargs[1] + "@" + dirbase
 515                 } else if logprefix == "arvados-server" && len(args) > 1 {
 516                         logprefix = args[0]
 517                 }
 518                 if !strings.HasPrefix(dir, "/") {
 519                         logprefix = dir + ": " + logprefix
 520                 }
 521         }
 522
 523         cmd := exec.Command(super.lookPath(prog), args...)
 524         stdout, err := cmd.StdoutPipe()
 525         if err != nil {
 526                 return err
 527         }
 528         stderr, err := cmd.StderrPipe()
 529         if err != nil {
 530                 return err
 531         }
 532         logwriter := &service.LogPrefixer{Writer: super.Stderr, Prefix: []byte("[" + logprefix + "] ")}
 533         var copiers sync.WaitGroup
 534         copiers.Add(1)
 535         go func() {
 536                 io.Copy(logwriter, stderr)
 537                 copiers.Done()
 538         }()
 539         copiers.Add(1)
 540         go func() {
 541                 if opts.output == nil {
 542                         io.Copy(logwriter, stdout)
 543                 } else {
 544                         io.Copy(opts.output, stdout)
 545                 }
 546                 copiers.Done()
 547         }()
 548
 549         if strings.HasPrefix(dir, "/") {
 550                 cmd.Dir = dir
 551         } else {
 552                 cmd.Dir = filepath.Join(super.SourcePath, dir)
 553         }
 554         env := append([]string(nil), opts.env...)
 555         env = append(env, super.environ...)
 556         cmd.Env = dedupEnv(env)
 557
 558         if opts.user != "" {
 559                 u, err := user.Lookup(opts.user)
 560                 if err != nil {
 561                         return fmt.Errorf("user.Lookup(%q): %w", opts.user, err)
 562                 }
 563                 uid, _ := strconv.Atoi(u.Uid)
 564                 gid, _ := strconv.Atoi(u.Gid)
 565                 cmd.SysProcAttr = &syscall.SysProcAttr{
 566                         Credential: &syscall.Credential{
 567                                 Uid: uint32(uid),
 568                                 Gid: uint32(gid),
 569                         },
 570                 }
 571         }
 572
 573         exited := false
 574         defer func() { exited = true }()
 575         go func() {
 576                 <-ctx.Done()
 577                 log := ctxlog.FromContext(ctx).WithFields(logrus.Fields{"dir": dir, "cmdline": cmdline})
 578                 for !exited {
 579                         if cmd.Process == nil {
 580                                 log.Debug("waiting for child process to start")
 581                                 time.Sleep(time.Second / 2)
 582                         } else {
 583                                 log.WithField("PID", cmd.Process.Pid).Debug("sending SIGTERM")
 584                                 cmd.Process.Signal(syscall.SIGTERM)
 585                                 time.Sleep(5 * time.Second)
 586                                 if !exited {
 587                                         stdout.Close()
 588                                         stderr.Close()
 589                                         log.WithField("PID", cmd.Process.Pid).Warn("still waiting for child process to exit 5s after SIGTERM")
 590                                 }
 591                         }
 592                 }
 593         }()
 594
 595         err = cmd.Start()
 596         if err != nil {
 597                 return err
 598         }
 599         copiers.Wait()
 600         err = cmd.Wait()
 601         if ctx.Err() != nil {
 602                 // Return "context canceled", instead of the "killed"
 603                 // error that was probably caused by the context being
 604                 // canceled.
 605                 return ctx.Err()
 606         } else if err != nil {
 607                 return fmt.Errorf("%s: error: %v", cmdline, err)
 608         }
 609         return nil
 610 }
 611
 612 func (super *Supervisor) autofillConfig(cfg *arvados.Config) error {
 613         cluster, err := cfg.GetCluster("")
 614         if err != nil {
 615                 return err
 616         }
 617         usedPort := map[string]bool{}
 618         nextPort := func(host string) string {
 619                 for {
 620                         port, err := availablePort(host)
 621                         if err != nil {
 622                                 panic(err)
 623                         }
 624                         if usedPort[port] {
 625                                 continue
 626                         }
 627                         usedPort[port] = true
 628                         return port
 629                 }
 630         }
 631         if cluster.Services.Controller.ExternalURL.Host == "" {
 632                 h, p, err := net.SplitHostPort(super.ControllerAddr)
 633                 if err != nil {
 634                         return err
 635                 }
 636                 if h == "" {
 637                         h = super.ListenHost
 638                 }
 639                 if p == "0" {
 640                         p = nextPort(h)
 641                 }
 642                 cluster.Services.Controller.ExternalURL = arvados.URL{Scheme: "https", Host: net.JoinHostPort(h, p), Path: "/"}
 643         }
 644         for _, svc := range []*arvados.Service{
 645                 &cluster.Services.Controller,
 646                 &cluster.Services.DispatchCloud,
 647                 &cluster.Services.GitHTTP,
 648                 &cluster.Services.Health,
 649                 &cluster.Services.Keepproxy,
 650                 &cluster.Services.Keepstore,
 651                 &cluster.Services.RailsAPI,
 652                 &cluster.Services.WebDAV,
 653                 &cluster.Services.WebDAVDownload,
 654                 &cluster.Services.Websocket,
 655                 &cluster.Services.Workbench1,
 656         } {
 657                 if svc == &cluster.Services.DispatchCloud && super.ClusterType == "test" {
 658                         continue
 659                 }
 660                 if svc.ExternalURL.Host == "" {
 661                         if svc == &cluster.Services.Controller ||
 662                                 svc == &cluster.Services.GitHTTP ||
 663                                 svc == &cluster.Services.Health ||
 664                                 svc == &cluster.Services.Keepproxy ||
 665                                 svc == &cluster.Services.WebDAV ||
 666                                 svc == &cluster.Services.WebDAVDownload ||
 667                                 svc == &cluster.Services.Workbench1 {
 668                                 svc.ExternalURL = arvados.URL{Scheme: "https", Host: fmt.Sprintf("%s:%s", super.ListenHost, nextPort(super.ListenHost)), Path: "/"}
 669                         } else if svc == &cluster.Services.Websocket {
 670                                 svc.ExternalURL = arvados.URL{Scheme: "wss", Host: fmt.Sprintf("%s:%s", super.ListenHost, nextPort(super.ListenHost)), Path: "/websocket"}
 671                         }
 672                 }
 673                 if len(svc.InternalURLs) == 0 {
 674                         svc.InternalURLs = map[arvados.URL]arvados.ServiceInstance{
 675                                 {Scheme: "http", Host: fmt.Sprintf("%s:%s", super.ListenHost, nextPort(super.ListenHost)), Path: "/"}: {},
 676                         }
 677                 }
 678         }
 679         if super.ClusterType != "production" {
 680                 if cluster.SystemRootToken == "" {
 681                         cluster.SystemRootToken = randomHexString(64)
 682                 }
 683                 if cluster.ManagementToken == "" {
 684                         cluster.ManagementToken = randomHexString(64)
 685                 }
 686                 if cluster.API.RailsSessionSecretToken == "" {
 687                         cluster.API.RailsSessionSecretToken = randomHexString(64)
 688                 }
 689                 if cluster.Collections.BlobSigningKey == "" {
 690                         cluster.Collections.BlobSigningKey = randomHexString(64)
 691                 }
 692                 if cluster.Containers.DispatchPrivateKey == "" {
 693                         buf, err := ioutil.ReadFile(filepath.Join(super.SourcePath, "lib", "dispatchcloud", "test", "sshkey_dispatch"))
 694                         if err != nil {
 695                                 return err
 696                         }
 697                         cluster.Containers.DispatchPrivateKey = string(buf)
 698                 }
 699                 cluster.TLS.Insecure = true
 700         }
 701         if super.ClusterType == "test" {
 702                 // Add a second keepstore process.
 703                 cluster.Services.Keepstore.InternalURLs[arvados.URL{Scheme: "http", Host: fmt.Sprintf("%s:%s", super.ListenHost, nextPort(super.ListenHost)), Path: "/"}] = arvados.ServiceInstance{}
 704
 705                 // Create a directory-backed volume for each keepstore
 706                 // process.
 707                 cluster.Volumes = map[string]arvados.Volume{}
 708                 for url := range cluster.Services.Keepstore.InternalURLs {
 709                         volnum := len(cluster.Volumes)
 710                         datadir := fmt.Sprintf("%s/keep%d.data", super.tempdir, volnum)
 711                         if _, err = os.Stat(datadir + "/."); err == nil {
 712                         } else if !os.IsNotExist(err) {
 713                                 return err
 714                         } else if err = os.Mkdir(datadir, 0755); err != nil {
 715                                 return err
 716                         }
 717                         cluster.Volumes[fmt.Sprintf(cluster.ClusterID+"-nyw5e-%015d", volnum)] = arvados.Volume{
 718                                 Driver:           "Directory",
 719                                 DriverParameters: json.RawMessage(fmt.Sprintf(`{"Root":%q}`, datadir)),
 720                                 AccessViaHosts: map[arvados.URL]arvados.VolumeAccess{
 721                                         url: {},
 722                                 },
 723                         }
 724                 }
 725         }
 726         if super.OwnTemporaryDatabase {
 727                 cluster.PostgreSQL.Connection = arvados.PostgreSQLConnection{
 728                         "client_encoding": "utf8",
 729                         "host":            "localhost",
 730                         "port":            nextPort(super.ListenHost),
 731                         "dbname":          "arvados_test",
 732                         "user":            "arvados",
 733                         "password":        "insecure_arvados_test",
 734                 }
 735         }
 736
 737         cfg.Clusters[cluster.ClusterID] = *cluster
 738         return nil
 739 }
 740
 741 func addrIsLocal(addr string) (bool, error) {
 742         return true, nil
 743         listener, err := net.Listen("tcp", addr)
 744         if err == nil {
 745                 listener.Close()
 746                 return true, nil
 747         } else if strings.Contains(err.Error(), "cannot assign requested address") {
 748                 return false, nil
 749         } else {
 750                 return false, err
 751         }
 752 }
 753
 754 func randomHexString(chars int) string {
 755         b := make([]byte, chars/2)
 756         _, err := rand.Read(b)
 757         if err != nil {
 758                 panic(err)
 759         }
 760         return fmt.Sprintf("%x", b)
 761 }
 762
 763 func internalPort(svc arvados.Service) (string, error) {
 764         if len(svc.InternalURLs) > 1 {
 765                 return "", errors.New("internalPort() doesn't work with multiple InternalURLs")
 766         }
 767         for u := range svc.InternalURLs {
 768                 u := url.URL(u)
 769                 if p := u.Port(); p != "" {
 770                         return p, nil
 771                 } else if u.Scheme == "https" || u.Scheme == "ws" {
 772                         return "443", nil
 773                 } else {
 774                         return "80", nil
 775                 }
 776         }
 777         return "", fmt.Errorf("service has no InternalURLs")
 778 }
 779
 780 func externalPort(svc arvados.Service) (string, error) {
 781         u := url.URL(svc.ExternalURL)
 782         if p := u.Port(); p != "" {
 783                 return p, nil
 784         } else if u.Scheme == "https" || u.Scheme == "wss" {
 785                 return "443", nil
 786         } else {
 787                 return "80", nil
 788         }
 789 }
 790
 791 func availablePort(host string) (string, error) {
 792         ln, err := net.Listen("tcp", net.JoinHostPort(host, "0"))
 793         if err != nil {
 794                 return "", err
 795         }
 796         defer ln.Close()
 797         _, port, err := net.SplitHostPort(ln.Addr().String())
 798         if err != nil {
 799                 return "", err
 800         }
 801         return port, nil
 802 }
 803
 804 // Try to connect to addr until it works, then close ch. Give up if
 805 // ctx cancels.
 806 func waitForConnect(ctx context.Context, addr string) error {
 807         dialer := net.Dialer{Timeout: time.Second}
 808         for ctx.Err() == nil {
 809                 conn, err := dialer.DialContext(ctx, "tcp", addr)
 810                 if err != nil {
 811                         time.Sleep(time.Second / 10)
 812                         continue
 813                 }
 814                 conn.Close()
 815                 return nil
 816         }
 817         return ctx.Err()
 818 }
 819
 820 func copyConfig(cfg *arvados.Config) *arvados.Config {
 821         pr, pw := io.Pipe()
 822         go func() {
 823                 err := json.NewEncoder(pw).Encode(cfg)
 824                 if err != nil {
 825                         panic(err)
 826                 }
 827                 pw.Close()
 828         }()
 829         cfg2 := new(arvados.Config)
 830         err := json.NewDecoder(pr).Decode(cfg2)
 831         if err != nil {
 832                 panic(err)
 833         }
 834         return cfg2
 835 }
 836
 837 func watchConfig(ctx context.Context, logger logrus.FieldLogger, cfgPath string, prevcfg *arvados.Config, fn func()) {
 838         watcher, err := fsnotify.NewWatcher()
 839         if err != nil {
 840                 logger.WithError(err).Error("fsnotify setup failed")
 841                 return
 842         }
 843         defer watcher.Close()
 844
 845         err = watcher.Add(cfgPath)
 846         if err != nil {
 847                 logger.WithError(err).Error("fsnotify watcher failed")
 848                 return
 849         }
 850
 851         for {
 852                 select {
 853                 case <-ctx.Done():
 854                         return
 855                 case err, ok := <-watcher.Errors:
 856                         if !ok {
 857                                 return
 858                         }
 859                         logger.WithError(err).Warn("fsnotify watcher reported error")
 860                 case _, ok := <-watcher.Events:
 861                         if !ok {
 862                                 return
 863                         }
 864                         for len(watcher.Events) > 0 {
 865                                 <-watcher.Events
 866                         }
 867                         loader := config.NewLoader(&bytes.Buffer{}, &logrus.Logger{Out: ioutil.Discard})
 868                         loader.Path = cfgPath
 869                         loader.SkipAPICalls = true
 870                         cfg, err := loader.Load()
 871                         if err != nil {
 872                                 logger.WithError(err).Warn("error reloading config file after change detected; ignoring new config for now")
 873                         } else if reflect.DeepEqual(cfg, prevcfg) {
 874                                 logger.Debug("config file changed but is still DeepEqual to the existing config")
 875                         } else {
 876                                 logger.Debug("config changed, notifying supervisor")
 877                                 fn()
 878                                 prevcfg = cfg
 879                         }
 880                 }
 881         }
 882 }