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         // Config file location like "/etc/arvados/config.yml", or "-"
  41         // to read from Stdin (see below).
  42         ConfigPath string
  43         // Literal config file (useful for test suites). If non-empty,
  44         // this is used instead of ConfigPath.
  45         ConfigYAML string
  46         // Path to arvados source tree. Only used for dev/test
  47         // clusters.
  48         SourcePath string
  49         // Version number to build into binaries. Only used for
  50         // dev/test clusters.
  51         SourceVersion string
  52         // "production", "development", or "test".
  53         ClusterType string
  54         // Listening address for external services, and internal
  55         // services whose InternalURLs are not explicitly configured.
  56         // If blank, listen on the configured controller ExternalURL
  57         // host; if that is also blank, listen on all addresses
  58         // (0.0.0.0).
  59         ListenHost string
  60         // Default host:port for controller ExternalURL if not
  61         // explicitly configured in config file. If blank, use a
  62         // random port on ListenHost.
  63         ControllerAddr string
  64         // Path to arvados-workbench2 source tree checkout.
  65         Workbench2Source     string
  66         NoWorkbench1         bool
  67         NoWorkbench2         bool
  68         OwnTemporaryDatabase bool
  69         Stdin                io.Reader
  70         Stderr               io.Writer
  71
  72         logger   logrus.FieldLogger
  73         cluster  *arvados.Cluster       // nil if this is a multi-cluster supervisor
  74         children map[string]*Supervisor // nil if this is a single-cluster supervisor
  75
  76         ctx           context.Context
  77         cancel        context.CancelFunc
  78         done          chan struct{}      // closed when child procs/services have shut down
  79         err           error              // error that caused shutdown (valid when done is closed)
  80         healthChecker *health.Aggregator // nil if this is a multi-cluster supervisor, or still booting
  81         tasksReady    map[string]chan bool
  82         waitShutdown  sync.WaitGroup
  83
  84         bindir     string
  85         tempdir    string // in production mode, this is accessible only to root
  86         wwwtempdir string // in production mode, this is accessible only to www-data
  87         configfile string
  88         environ    []string // for child processes
  89 }
  90
  91 func (super *Supervisor) Clusters() map[string]*arvados.Cluster {
  92         m := map[string]*arvados.Cluster{}
  93         if super.cluster != nil {
  94                 m[super.cluster.ClusterID] = super.cluster
  95         }
  96         for id, super2 := range super.children {
  97                 m[id] = super2.Cluster("")
  98         }
  99         return m
 100 }
 101
 102 func (super *Supervisor) Cluster(id string) *arvados.Cluster {
 103         if super.children != nil {
 104                 return super.children[id].Cluster(id)
 105         } else {
 106                 return super.cluster
 107         }
 108 }
 109
 110 func (super *Supervisor) Start(ctx context.Context) {
 111         super.logger = ctxlog.FromContext(ctx)
 112         super.ctx, super.cancel = context.WithCancel(ctx)
 113         super.done = make(chan struct{})
 114
 115         sigch := make(chan os.Signal)
 116         signal.Notify(sigch, syscall.SIGINT, syscall.SIGTERM, syscall.SIGHUP)
 117         go func() {
 118                 defer signal.Stop(sigch)
 119                 for {
 120                         select {
 121                         case <-ctx.Done():
 122                                 return
 123                         case sig := <-sigch:
 124                                 super.logger.WithField("signal", sig).Info("caught signal")
 125                                 if super.err == nil {
 126                                         if sig == syscall.SIGHUP {
 127                                                 super.err = errNeedConfigReload
 128                                         } else {
 129                                                 super.err = fmt.Errorf("caught signal %s", sig)
 130                                         }
 131                                 }
 132                                 super.cancel()
 133                         }
 134                 }
 135         }()
 136
 137         loaderStdin := super.Stdin
 138         if super.ConfigYAML != "" {
 139                 loaderStdin = bytes.NewBufferString(super.ConfigYAML)
 140         }
 141         loader := config.NewLoader(loaderStdin, super.logger)
 142         loader.SkipLegacy = true
 143         loader.SkipAPICalls = true
 144         loader.Path = super.ConfigPath
 145         if super.ConfigYAML != "" {
 146                 loader.Path = "-"
 147         }
 148         cfg, err := loader.Load()
 149         if err != nil {
 150                 super.err = err
 151                 close(super.done)
 152                 super.cancel()
 153                 return
 154         }
 155
 156         if super.ConfigPath != "" && super.ConfigPath != "-" && cfg.AutoReloadConfig {
 157                 go watchConfig(super.ctx, super.logger, super.ConfigPath, copyConfig(cfg), func() {
 158                         if super.err == nil {
 159                                 super.err = errNeedConfigReload
 160                         }
 161                         super.cancel()
 162                 })
 163         }
 164
 165         if len(cfg.Clusters) > 1 {
 166                 super.startFederation(cfg)
 167                 go func() {
 168                         defer super.cancel()
 169                         defer close(super.done)
 170                         for _, super2 := range super.children {
 171                                 err := super2.Wait()
 172                                 if super.err == nil {
 173                                         super.err = err
 174                                 }
 175                         }
 176                 }()
 177         } else {
 178                 go func() {
 179                         defer super.cancel()
 180                         defer close(super.done)
 181                         super.cluster, super.err = cfg.GetCluster("")
 182                         if super.err != nil {
 183                                 return
 184                         }
 185                         err := super.runCluster()
 186                         if err != nil {
 187                                 super.logger.WithError(err).Info("supervisor shut down")
 188                                 if super.err == nil {
 189                                         super.err = err
 190                                 }
 191                         }
 192                 }()
 193         }
 194 }
 195
 196 // Wait returns when all child processes and goroutines have exited.
 197 func (super *Supervisor) Wait() error {
 198         <-super.done
 199         return super.err
 200 }
 201
 202 // startFederation starts a child Supervisor for each cluster in the
 203 // given config. Each is a copy of the original/parent with the
 204 // original config reduced to a single cluster.
 205 func (super *Supervisor) startFederation(cfg *arvados.Config) {
 206         super.children = map[string]*Supervisor{}
 207         for id, cc := range cfg.Clusters {
 208                 super2 := *super
 209                 yaml, err := json.Marshal(arvados.Config{Clusters: map[string]arvados.Cluster{id: cc}})
 210                 if err != nil {
 211                         panic(fmt.Sprintf("json.Marshal partial config: %s", err))
 212                 }
 213                 super2.ConfigYAML = string(yaml)
 214                 super2.ConfigPath = "-"
 215                 super2.children = nil
 216
 217                 if super2.ClusterType == "test" {
 218                         super2.Stderr = &service.LogPrefixer{
 219                                 Writer: super.Stderr,
 220                                 Prefix: []byte("[" + id + "] "),
 221                         }
 222                 }
 223                 super2.Start(super.ctx)
 224                 super.children[id] = &super2
 225         }
 226 }
 227
 228 func (super *Supervisor) runCluster() error {
 229         cwd, err := os.Getwd()
 230         if err != nil {
 231                 return err
 232         }
 233         if super.ClusterType == "test" && super.SourcePath == "" {
 234                 // When invoked by test suite, default to current
 235                 // source tree
 236                 buf, err := exec.Command("git", "rev-parse", "--show-toplevel").CombinedOutput()
 237                 if err != nil {
 238                         return fmt.Errorf("git rev-parse: %w", err)
 239                 }
 240                 super.SourcePath = strings.TrimSuffix(string(buf), "\n")
 241         } else if !strings.HasPrefix(super.SourcePath, "/") {
 242                 super.SourcePath = filepath.Join(cwd, super.SourcePath)
 243         }
 244         super.SourcePath, err = filepath.EvalSymlinks(super.SourcePath)
 245         if err != nil {
 246                 return err
 247         }
 248
 249         if super.ListenHost == "" {
 250                 u := url.URL(super.cluster.Services.Controller.ExternalURL)
 251                 super.ListenHost = u.Hostname()
 252                 if super.ListenHost == "" {
 253                         super.ListenHost = "0.0.0.0"
 254                 }
 255         }
 256
 257         // Choose bin and temp dirs: /var/lib/arvados/... in
 258         // production, transient tempdir otherwise.
 259         if super.ClusterType == "production" {
 260                 // These dirs have already been created by
 261                 // "arvados-server install" (or by extracting a
 262                 // package).
 263                 super.tempdir = "/var/lib/arvados/tmp"
 264                 super.wwwtempdir = "/var/lib/arvados/wwwtmp"
 265                 super.bindir = "/var/lib/arvados/bin"
 266         } else {
 267                 super.tempdir, err = ioutil.TempDir("", "arvados-server-boot-")
 268                 if err != nil {
 269                         return err
 270                 }
 271                 defer os.RemoveAll(super.tempdir)
 272                 super.wwwtempdir = super.tempdir
 273                 super.bindir = filepath.Join(super.tempdir, "bin")
 274                 if err := os.Mkdir(super.bindir, 0755); err != nil {
 275                         return err
 276                 }
 277         }
 278
 279         // Fill in any missing config keys, and write the resulting
 280         // config in the temp dir for child services to use.
 281         err = super.autofillConfig()
 282         if err != nil {
 283                 return err
 284         }
 285         conffile, err := os.OpenFile(filepath.Join(super.wwwtempdir, "config.yml"), os.O_CREATE|os.O_WRONLY|os.O_TRUNC, 0644)
 286         if err != nil {
 287                 return err
 288         }
 289         defer conffile.Close()
 290         err = json.NewEncoder(conffile).Encode(arvados.Config{
 291                 Clusters: map[string]arvados.Cluster{
 292                         super.cluster.ClusterID: *super.cluster}})
 293         if err != nil {
 294                 return err
 295         }
 296         err = conffile.Close()
 297         if err != nil {
 298                 return err
 299         }
 300         super.configfile = conffile.Name()
 301
 302         super.environ = os.Environ()
 303         super.cleanEnv([]string{"ARVADOS_"})
 304         super.setEnv("ARVADOS_CONFIG", super.configfile)
 305         super.setEnv("RAILS_ENV", super.ClusterType)
 306         super.setEnv("TMPDIR", super.tempdir)
 307         super.prependEnv("PATH", "/var/lib/arvados/bin:")
 308         if super.ClusterType != "production" {
 309                 super.prependEnv("PATH", super.tempdir+"/bin:")
 310         }
 311         super.setEnv("ARVADOS_SERVER_ADDRESS", super.ListenHost)
 312
 313         // Now that we have the config, replace the bootstrap logger
 314         // with a new one according to the logging config.
 315         loglevel := super.cluster.SystemLogs.LogLevel
 316         if s := os.Getenv("ARVADOS_DEBUG"); s != "" && s != "0" {
 317                 loglevel = "debug"
 318         }
 319         super.logger = ctxlog.New(super.Stderr, super.cluster.SystemLogs.Format, loglevel).WithFields(logrus.Fields{
 320                 "PID": os.Getpid(),
 321         })
 322
 323         if super.SourceVersion == "" && super.ClusterType == "production" {
 324                 // don't need SourceVersion
 325         } else if super.SourceVersion == "" {
 326                 // Find current source tree version.
 327                 var buf bytes.Buffer
 328                 err = super.RunProgram(super.ctx, ".", runOptions{output: &buf}, "git", "diff", "--shortstat")
 329                 if err != nil {
 330                         return err
 331                 }
 332                 dirty := buf.Len() > 0
 333                 buf.Reset()
 334                 err = super.RunProgram(super.ctx, ".", runOptions{output: &buf}, "git", "log", "-n1", "--format=%H")
 335                 if err != nil {
 336                         return err
 337                 }
 338                 super.SourceVersion = strings.TrimSpace(buf.String())
 339                 if dirty {
 340                         super.SourceVersion += "+uncommitted"
 341                 }
 342         } else {
 343                 return errors.New("specifying a version to run is not yet supported")
 344         }
 345
 346         _, err = super.installGoProgram(super.ctx, "cmd/arvados-server")
 347         if err != nil {
 348                 return err
 349         }
 350         err = super.setupRubyEnv()
 351         if err != nil {
 352                 return err
 353         }
 354
 355         tasks := []supervisedTask{
 356                 createCertificates{},
 357                 runPostgreSQL{},
 358                 runNginx{},
 359                 runServiceCommand{name: "controller", svc: super.cluster.Services.Controller, depends: []supervisedTask{seedDatabase{}}},
 360                 runServiceCommand{name: "git-httpd", svc: super.cluster.Services.GitHTTP},
 361                 runServiceCommand{name: "health", svc: super.cluster.Services.Health},
 362                 runServiceCommand{name: "keepproxy", svc: super.cluster.Services.Keepproxy, depends: []supervisedTask{runPassenger{src: "services/api"}}},
 363                 runServiceCommand{name: "keepstore", svc: super.cluster.Services.Keepstore},
 364                 runServiceCommand{name: "keep-web", svc: super.cluster.Services.WebDAV},
 365                 runServiceCommand{name: "ws", svc: super.cluster.Services.Websocket, depends: []supervisedTask{seedDatabase{}}},
 366                 installPassenger{src: "services/api", varlibdir: "railsapi"},
 367                 runPassenger{src: "services/api", varlibdir: "railsapi", svc: super.cluster.Services.RailsAPI, depends: []supervisedTask{createCertificates{}, seedDatabase{}, installPassenger{src: "services/api", varlibdir: "railsapi"}}},
 368                 seedDatabase{},
 369         }
 370         if !super.NoWorkbench1 {
 371                 tasks = append(tasks,
 372                         installPassenger{src: "apps/workbench", varlibdir: "workbench1", depends: []supervisedTask{seedDatabase{}}}, // dependency ensures workbench doesn't delay api install/startup
 373                         runPassenger{src: "apps/workbench", varlibdir: "workbench1", svc: super.cluster.Services.Workbench1, depends: []supervisedTask{installPassenger{src: "apps/workbench", varlibdir: "workbench1"}}},
 374                 )
 375         }
 376         if !super.NoWorkbench2 {
 377                 tasks = append(tasks,
 378                         runWorkbench2{svc: super.cluster.Services.Workbench2},
 379                 )
 380         }
 381         if super.ClusterType != "test" {
 382                 tasks = append(tasks,
 383                         runServiceCommand{name: "keep-balance", svc: super.cluster.Services.Keepbalance},
 384                 )
 385         }
 386         if super.cluster.Containers.CloudVMs.Enable {
 387                 tasks = append(tasks,
 388                         runServiceCommand{name: "dispatch-cloud", svc: super.cluster.Services.DispatchCloud},
 389                 )
 390         }
 391         super.tasksReady = map[string]chan bool{}
 392         for _, task := range tasks {
 393                 super.tasksReady[task.String()] = make(chan bool)
 394         }
 395         for _, task := range tasks {
 396                 task := task
 397                 fail := func(err error) {
 398                         if super.ctx.Err() != nil {
 399                                 return
 400                         }
 401                         super.cancel()
 402                         super.logger.WithField("task", task.String()).WithError(err).Error("task failed")
 403                 }
 404                 go func() {
 405                         super.logger.WithField("task", task.String()).Info("starting")
 406                         err := task.Run(super.ctx, fail, super)
 407                         if err != nil {
 408                                 fail(err)
 409                                 return
 410                         }
 411                         close(super.tasksReady[task.String()])
 412                 }()
 413         }
 414         err = super.wait(super.ctx, tasks...)
 415         if err != nil {
 416                 return err
 417         }
 418         super.logger.Info("all startup tasks are complete; starting health checks")
 419         super.healthChecker = &health.Aggregator{Cluster: super.cluster}
 420         <-super.ctx.Done()
 421         super.logger.Info("shutting down")
 422         super.waitShutdown.Wait()
 423         return super.ctx.Err()
 424 }
 425
 426 func (super *Supervisor) wait(ctx context.Context, tasks ...supervisedTask) error {
 427         ticker := time.NewTicker(15 * time.Second)
 428         defer ticker.Stop()
 429         for _, task := range tasks {
 430                 ch, ok := super.tasksReady[task.String()]
 431                 if !ok {
 432                         return fmt.Errorf("no such task: %s", task)
 433                 }
 434                 super.logger.WithField("task", task.String()).Info("waiting")
 435                 for {
 436                         select {
 437                         case <-ch:
 438                                 super.logger.WithField("task", task.String()).Info("ready")
 439                         case <-ctx.Done():
 440                                 super.logger.WithField("task", task.String()).Info("task was never ready")
 441                                 return ctx.Err()
 442                         case <-ticker.C:
 443                                 super.logger.WithField("task", task.String()).Info("still waiting...")
 444                                 continue
 445                         }
 446                         break
 447                 }
 448         }
 449         return nil
 450 }
 451
 452 // Stop shuts down all child processes and goroutines, and returns
 453 // when all of them have exited.
 454 func (super *Supervisor) Stop() {
 455         super.cancel()
 456         <-super.done
 457 }
 458
 459 // WaitReady waits for the cluster(s) to be ready to handle requests,
 460 // then returns true. If startup fails, it returns false.
 461 func (super *Supervisor) WaitReady() bool {
 462         if super.children != nil {
 463                 for id, super2 := range super.children {
 464                         super.logger.Infof("waiting for %s to be ready", id)
 465                         if !super2.WaitReady() {
 466                                 super.logger.Infof("%s startup failed", id)
 467                                 super.Stop()
 468                                 return false
 469                         }
 470                         super.logger.Infof("%s is ready", id)
 471                 }
 472                 super.logger.Info("all clusters are ready")
 473                 return true
 474         }
 475         ticker := time.NewTicker(time.Second)
 476         defer ticker.Stop()
 477         for waiting := "all"; waiting != ""; {
 478                 select {
 479                 case <-ticker.C:
 480                 case <-super.ctx.Done():
 481                         super.Stop()
 482                         return false
 483                 }
 484                 if super.healthChecker == nil {
 485                         // not set up yet
 486                         continue
 487                 }
 488                 resp := super.healthChecker.ClusterHealth()
 489                 // The overall health check (resp.Health=="OK") might
 490                 // never pass due to missing components (like
 491                 // arvados-dispatch-cloud in a test cluster), so
 492                 // instead we wait for all configured components to
 493                 // pass.
 494                 waiting = ""
 495                 for target, check := range resp.Checks {
 496                         if check.Health != "OK" {
 497                                 waiting += " " + target
 498                         }
 499                 }
 500                 if waiting != "" {
 501                         super.logger.WithField("targets", waiting[1:]).Info("waiting")
 502                 }
 503         }
 504         return true
 505 }
 506
 507 func (super *Supervisor) prependEnv(key, prepend string) {
 508         for i, s := range super.environ {
 509                 if strings.HasPrefix(s, key+"=") {
 510                         super.environ[i] = key + "=" + prepend + s[len(key)+1:]
 511                         return
 512                 }
 513         }
 514         super.environ = append(super.environ, key+"="+prepend)
 515 }
 516
 517 func (super *Supervisor) cleanEnv(prefixes []string) {
 518         var cleaned []string
 519         for _, s := range super.environ {
 520                 drop := false
 521                 for _, p := range prefixes {
 522                         if strings.HasPrefix(s, p) {
 523                                 drop = true
 524                                 break
 525                         }
 526                 }
 527                 if !drop {
 528                         cleaned = append(cleaned, s)
 529                 }
 530         }
 531         super.environ = cleaned
 532 }
 533
 534 func (super *Supervisor) setEnv(key, val string) {
 535         for i, s := range super.environ {
 536                 if strings.HasPrefix(s, key+"=") {
 537                         super.environ[i] = key + "=" + val
 538                         return
 539                 }
 540         }
 541         super.environ = append(super.environ, key+"="+val)
 542 }
 543
 544 // Remove all but the first occurrence of each env var.
 545 func dedupEnv(in []string) []string {
 546         saw := map[string]bool{}
 547         var out []string
 548         for _, kv := range in {
 549                 if split := strings.Index(kv, "="); split < 1 {
 550                         panic("invalid environment var: " + kv)
 551                 } else if saw[kv[:split]] {
 552                         continue
 553                 } else {
 554                         saw[kv[:split]] = true
 555                         out = append(out, kv)
 556                 }
 557         }
 558         return out
 559 }
 560
 561 func (super *Supervisor) installGoProgram(ctx context.Context, srcpath string) (string, error) {
 562         _, basename := filepath.Split(srcpath)
 563         binfile := filepath.Join(super.bindir, basename)
 564         if super.ClusterType == "production" {
 565                 return binfile, nil
 566         }
 567         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)
 568         return binfile, err
 569 }
 570
 571 func (super *Supervisor) usingRVM() bool {
 572         return os.Getenv("rvm_path") != ""
 573 }
 574
 575 func (super *Supervisor) setupRubyEnv() error {
 576         if !super.usingRVM() {
 577                 // (If rvm is in use, assume the caller has everything
 578                 // set up as desired)
 579                 super.cleanEnv([]string{
 580                         "GEM_HOME=",
 581                         "GEM_PATH=",
 582                 })
 583                 gem := "gem"
 584                 if _, err := os.Stat("/var/lib/arvados/bin/gem"); err == nil || super.ClusterType == "production" {
 585                         gem = "/var/lib/arvados/bin/gem"
 586                 }
 587                 cmd := exec.Command(gem, "env", "gempath")
 588                 if super.ClusterType == "production" {
 589                         cmd.Args = append([]string{"sudo", "-u", "www-data", "-E", "HOME=/var/www"}, cmd.Args...)
 590                         path, err := exec.LookPath("sudo")
 591                         if err != nil {
 592                                 return fmt.Errorf("LookPath(\"sudo\"): %w", err)
 593                         }
 594                         cmd.Path = path
 595                 }
 596                 cmd.Stderr = super.Stderr
 597                 cmd.Env = super.environ
 598                 buf, err := cmd.Output() // /var/lib/arvados/.gem/ruby/2.5.0/bin:...
 599                 if err != nil || len(buf) == 0 {
 600                         return fmt.Errorf("gem env gempath: %w", err)
 601                 }
 602                 gempath := string(bytes.Split(buf, []byte{':'})[0])
 603                 super.prependEnv("PATH", gempath+"/bin:")
 604                 super.setEnv("GEM_HOME", gempath)
 605                 super.setEnv("GEM_PATH", gempath)
 606         }
 607         // Passenger install doesn't work unless $HOME is ~user
 608         u, err := user.Current()
 609         if err != nil {
 610                 return err
 611         }
 612         super.setEnv("HOME", u.HomeDir)
 613         return nil
 614 }
 615
 616 func (super *Supervisor) lookPath(prog string) string {
 617         for _, val := range super.environ {
 618                 if strings.HasPrefix(val, "PATH=") {
 619                         for _, dir := range filepath.SplitList(val[5:]) {
 620                                 path := filepath.Join(dir, prog)
 621                                 if fi, err := os.Stat(path); err == nil && fi.Mode()&0111 != 0 {
 622                                         return path
 623                                 }
 624                         }
 625                 }
 626         }
 627         return prog
 628 }
 629
 630 type runOptions struct {
 631         output io.Writer // attach stdout
 632         env    []string  // add/replace environment variables
 633         user   string    // run as specified user
 634         stdin  io.Reader
 635 }
 636
 637 // RunProgram runs prog with args, using dir as working directory. If ctx is
 638 // cancelled while the child is running, RunProgram terminates the child, waits
 639 // for it to exit, then returns.
 640 //
 641 // Child's environment will have our env vars, plus any given in env.
 642 //
 643 // Child's stdout will be written to output if non-nil, otherwise the
 644 // boot command's stderr.
 645 func (super *Supervisor) RunProgram(ctx context.Context, dir string, opts runOptions, prog string, args ...string) error {
 646         cmdline := fmt.Sprintf("%s", append([]string{prog}, args...))
 647         super.logger.WithField("command", cmdline).WithField("dir", dir).Info("executing")
 648
 649         logprefix := prog
 650         {
 651                 innerargs := args
 652                 if logprefix == "sudo" {
 653                         for i := 0; i < len(args); i++ {
 654                                 if args[i] == "-u" {
 655                                         i++
 656                                 } else if args[i] == "-E" || strings.Contains(args[i], "=") {
 657                                 } else {
 658                                         logprefix = args[i]
 659                                         innerargs = args[i+1:]
 660                                         break
 661                                 }
 662                         }
 663                 }
 664                 logprefix = strings.TrimPrefix(logprefix, "/var/lib/arvados/bin/")
 665                 logprefix = strings.TrimPrefix(logprefix, super.tempdir+"/bin/")
 666                 if logprefix == "bundle" && len(innerargs) > 2 && innerargs[0] == "exec" {
 667                         _, dirbase := filepath.Split(dir)
 668                         logprefix = innerargs[1] + "@" + dirbase
 669                 } else if logprefix == "arvados-server" && len(args) > 1 {
 670                         logprefix = args[0]
 671                 }
 672                 if !strings.HasPrefix(dir, "/") {
 673                         logprefix = dir + ": " + logprefix
 674                 }
 675         }
 676
 677         cmd := exec.Command(super.lookPath(prog), args...)
 678         cmd.Stdin = opts.stdin
 679         stdout, err := cmd.StdoutPipe()
 680         if err != nil {
 681                 return err
 682         }
 683         stderr, err := cmd.StderrPipe()
 684         if err != nil {
 685                 return err
 686         }
 687         logwriter := &service.LogPrefixer{Writer: super.Stderr, Prefix: []byte("[" + logprefix + "] ")}
 688         var copiers sync.WaitGroup
 689         copiers.Add(1)
 690         go func() {
 691                 io.Copy(logwriter, stderr)
 692                 copiers.Done()
 693         }()
 694         copiers.Add(1)
 695         go func() {
 696                 if opts.output == nil {
 697                         io.Copy(logwriter, stdout)
 698                 } else {
 699                         io.Copy(opts.output, stdout)
 700                 }
 701                 copiers.Done()
 702         }()
 703
 704         if strings.HasPrefix(dir, "/") {
 705                 cmd.Dir = dir
 706         } else {
 707                 cmd.Dir = filepath.Join(super.SourcePath, dir)
 708         }
 709         env := append([]string(nil), opts.env...)
 710         env = append(env, super.environ...)
 711         cmd.Env = dedupEnv(env)
 712
 713         if opts.user != "" {
 714                 // Note: We use this approach instead of "sudo"
 715                 // because in certain circumstances (we are pid 1 in a
 716                 // docker container, and our passenger child process
 717                 // changes to pgid 1) the intermediate sudo process
 718                 // notices we have the same pgid as our child and
 719                 // refuses to propagate signals from us to our child,
 720                 // so we can't signal/shutdown our passenger/rails
 721                 // apps. "chpst" or "setuidgid" would work, but these
 722                 // few lines avoid depending on runit/daemontools.
 723                 u, err := user.Lookup(opts.user)
 724                 if err != nil {
 725                         return fmt.Errorf("user.Lookup(%q): %w", opts.user, err)
 726                 }
 727                 uid, _ := strconv.Atoi(u.Uid)
 728                 gid, _ := strconv.Atoi(u.Gid)
 729                 cmd.SysProcAttr = &syscall.SysProcAttr{
 730                         Credential: &syscall.Credential{
 731                                 Uid: uint32(uid),
 732                                 Gid: uint32(gid),
 733                         },
 734                 }
 735         }
 736
 737         exited := false
 738         defer func() { exited = true }()
 739         go func() {
 740                 <-ctx.Done()
 741                 log := ctxlog.FromContext(ctx).WithFields(logrus.Fields{"dir": dir, "cmdline": cmdline})
 742                 for !exited {
 743                         if cmd.Process == nil {
 744                                 log.Debug("waiting for child process to start")
 745                                 time.Sleep(time.Second / 2)
 746                         } else {
 747                                 log.WithField("PID", cmd.Process.Pid).Debug("sending SIGTERM")
 748                                 cmd.Process.Signal(syscall.SIGTERM)
 749                                 time.Sleep(5 * time.Second)
 750                                 if !exited {
 751                                         stdout.Close()
 752                                         stderr.Close()
 753                                         log.WithField("PID", cmd.Process.Pid).Warn("still waiting for child process to exit 5s after SIGTERM")
 754                                 }
 755                         }
 756                 }
 757         }()
 758
 759         err = cmd.Start()
 760         if err != nil {
 761                 return err
 762         }
 763         copiers.Wait()
 764         err = cmd.Wait()
 765         if ctx.Err() != nil {
 766                 // Return "context canceled", instead of the "killed"
 767                 // error that was probably caused by the context being
 768                 // canceled.
 769                 return ctx.Err()
 770         } else if err != nil {
 771                 return fmt.Errorf("%s: error: %v", cmdline, err)
 772         }
 773         return nil
 774 }
 775
 776 func (super *Supervisor) autofillConfig() error {
 777         usedPort := map[string]bool{}
 778         nextPort := func(host string) (string, error) {
 779                 for {
 780                         port, err := availablePort(host)
 781                         if err != nil {
 782                                 port, err = availablePort(super.ListenHost)
 783                         }
 784                         if err != nil {
 785                                 return "", err
 786                         }
 787                         if usedPort[port] {
 788                                 continue
 789                         }
 790                         usedPort[port] = true
 791                         return port, nil
 792                 }
 793         }
 794         if super.cluster.Services.Controller.ExternalURL.Host == "" {
 795                 h, p, err := net.SplitHostPort(super.ControllerAddr)
 796                 if err != nil && super.ControllerAddr != "" {
 797                         return fmt.Errorf("SplitHostPort(ControllerAddr %q): %w", super.ControllerAddr, err)
 798                 }
 799                 if h == "" {
 800                         h = super.ListenHost
 801                 }
 802                 if p == "0" || p == "" {
 803                         p, err = nextPort(h)
 804                         if err != nil {
 805                                 return err
 806                         }
 807                 }
 808                 super.cluster.Services.Controller.ExternalURL = arvados.URL{Scheme: "https", Host: net.JoinHostPort(h, p), Path: "/"}
 809         }
 810         u := url.URL(super.cluster.Services.Controller.ExternalURL)
 811         defaultExtHost := u.Hostname()
 812         for _, svc := range []*arvados.Service{
 813                 &super.cluster.Services.Controller,
 814                 &super.cluster.Services.DispatchCloud,
 815                 &super.cluster.Services.GitHTTP,
 816                 &super.cluster.Services.Health,
 817                 &super.cluster.Services.Keepproxy,
 818                 &super.cluster.Services.Keepstore,
 819                 &super.cluster.Services.RailsAPI,
 820                 &super.cluster.Services.WebDAV,
 821                 &super.cluster.Services.WebDAVDownload,
 822                 &super.cluster.Services.Websocket,
 823                 &super.cluster.Services.Workbench1,
 824                 &super.cluster.Services.Workbench2,
 825         } {
 826                 if svc.ExternalURL.Host == "" {
 827                         port, err := nextPort(defaultExtHost)
 828                         if err != nil {
 829                                 return err
 830                         }
 831                         host := net.JoinHostPort(defaultExtHost, port)
 832                         if svc == &super.cluster.Services.Controller ||
 833                                 svc == &super.cluster.Services.GitHTTP ||
 834                                 svc == &super.cluster.Services.Health ||
 835                                 svc == &super.cluster.Services.Keepproxy ||
 836                                 svc == &super.cluster.Services.WebDAV ||
 837                                 svc == &super.cluster.Services.WebDAVDownload ||
 838                                 svc == &super.cluster.Services.Workbench1 ||
 839                                 svc == &super.cluster.Services.Workbench2 {
 840                                 svc.ExternalURL = arvados.URL{Scheme: "https", Host: host, Path: "/"}
 841                         } else if svc == &super.cluster.Services.Websocket {
 842                                 svc.ExternalURL = arvados.URL{Scheme: "wss", Host: host, Path: "/websocket"}
 843                         }
 844                 }
 845                 if super.NoWorkbench1 && svc == &super.cluster.Services.Workbench1 ||
 846                         super.NoWorkbench2 && svc == &super.cluster.Services.Workbench2 ||
 847                         !super.cluster.Containers.CloudVMs.Enable && svc == &super.cluster.Services.DispatchCloud {
 848                         // When workbench1 is disabled, it gets an
 849                         // ExternalURL (so we have a valid listening
 850                         // port to write in our Nginx config) but no
 851                         // InternalURLs (so health checker doesn't
 852                         // complain).
 853                         continue
 854                 }
 855                 if len(svc.InternalURLs) == 0 {
 856                         port, err := nextPort(super.ListenHost)
 857                         if err != nil {
 858                                 return err
 859                         }
 860                         host := net.JoinHostPort(super.ListenHost, port)
 861                         svc.InternalURLs = map[arvados.URL]arvados.ServiceInstance{
 862                                 {Scheme: "http", Host: host, Path: "/"}: {},
 863                         }
 864                 }
 865         }
 866         if super.ClusterType != "production" {
 867                 if super.cluster.SystemRootToken == "" {
 868                         super.cluster.SystemRootToken = randomHexString(64)
 869                 }
 870                 if super.cluster.ManagementToken == "" {
 871                         super.cluster.ManagementToken = randomHexString(64)
 872                 }
 873                 if super.cluster.Collections.BlobSigningKey == "" {
 874                         super.cluster.Collections.BlobSigningKey = randomHexString(64)
 875                 }
 876                 if super.cluster.Users.AnonymousUserToken == "" {
 877                         super.cluster.Users.AnonymousUserToken = randomHexString(64)
 878                 }
 879                 if super.cluster.Containers.DispatchPrivateKey == "" {
 880                         buf, err := ioutil.ReadFile(filepath.Join(super.SourcePath, "lib", "dispatchcloud", "test", "sshkey_dispatch"))
 881                         if err != nil {
 882                                 return err
 883                         }
 884                         super.cluster.Containers.DispatchPrivateKey = string(buf)
 885                 }
 886                 super.cluster.TLS.Insecure = true
 887         }
 888         if super.ClusterType == "test" {
 889                 // Add a second keepstore process.
 890                 port, err := nextPort(super.ListenHost)
 891                 if err != nil {
 892                         return err
 893                 }
 894                 host := net.JoinHostPort(super.ListenHost, port)
 895                 super.cluster.Services.Keepstore.InternalURLs[arvados.URL{Scheme: "http", Host: host, Path: "/"}] = arvados.ServiceInstance{}
 896
 897                 // Create a directory-backed volume for each keepstore
 898                 // process.
 899                 super.cluster.Volumes = map[string]arvados.Volume{}
 900                 for url := range super.cluster.Services.Keepstore.InternalURLs {
 901                         volnum := len(super.cluster.Volumes)
 902                         datadir := fmt.Sprintf("%s/keep%d.data", super.tempdir, volnum)
 903                         if _, err = os.Stat(datadir + "/."); err == nil {
 904                         } else if !os.IsNotExist(err) {
 905                                 return err
 906                         } else if err = os.Mkdir(datadir, 0755); err != nil {
 907                                 return err
 908                         }
 909                         super.cluster.Volumes[fmt.Sprintf(super.cluster.ClusterID+"-nyw5e-%015d", volnum)] = arvados.Volume{
 910                                 Driver:           "Directory",
 911                                 DriverParameters: json.RawMessage(fmt.Sprintf(`{"Root":%q}`, datadir)),
 912                                 AccessViaHosts: map[arvados.URL]arvados.VolumeAccess{
 913                                         url: {},
 914                                 },
 915                                 StorageClasses: map[string]bool{
 916                                         "default": true,
 917                                         "foo":     true,
 918                                         "bar":     true,
 919                                 },
 920                         }
 921                 }
 922                 super.cluster.StorageClasses = map[string]arvados.StorageClassConfig{
 923                         "default": {Default: true},
 924                         "foo":     {},
 925                         "bar":     {},
 926                 }
 927         }
 928         if super.OwnTemporaryDatabase {
 929                 port, err := nextPort("localhost")
 930                 if err != nil {
 931                         return err
 932                 }
 933                 super.cluster.PostgreSQL.Connection = arvados.PostgreSQLConnection{
 934                         "client_encoding": "utf8",
 935                         "host":            "localhost",
 936                         "port":            port,
 937                         "dbname":          "arvados_test",
 938                         "user":            "arvados",
 939                         "password":        "insecure_arvados_test",
 940                 }
 941         }
 942         return nil
 943 }
 944
 945 func addrIsLocal(addr string) (bool, error) {
 946         if h, _, err := net.SplitHostPort(addr); err != nil {
 947                 return false, err
 948         } else {
 949                 addr = net.JoinHostPort(h, "0")
 950         }
 951         listener, err := net.Listen("tcp", addr)
 952         if err == nil {
 953                 listener.Close()
 954                 return true, nil
 955         } else if strings.Contains(err.Error(), "cannot assign requested address") {
 956                 return false, nil
 957         } else {
 958                 return false, err
 959         }
 960 }
 961
 962 func randomHexString(chars int) string {
 963         b := make([]byte, chars/2)
 964         _, err := rand.Read(b)
 965         if err != nil {
 966                 panic(err)
 967         }
 968         return fmt.Sprintf("%x", b)
 969 }
 970
 971 func internalPort(svc arvados.Service) (host, port string, err error) {
 972         if len(svc.InternalURLs) > 1 {
 973                 return "", "", errors.New("internalPort() doesn't work with multiple InternalURLs")
 974         }
 975         for u := range svc.InternalURLs {
 976                 u := url.URL(u)
 977                 host, port = u.Hostname(), u.Port()
 978                 switch {
 979                 case port != "":
 980                 case u.Scheme == "https", u.Scheme == "ws":
 981                         port = "443"
 982                 default:
 983                         port = "80"
 984                 }
 985                 return
 986         }
 987         return "", "", fmt.Errorf("service has no InternalURLs")
 988 }
 989
 990 func externalPort(svc arvados.Service) (string, error) {
 991         u := url.URL(svc.ExternalURL)
 992         if p := u.Port(); p != "" {
 993                 return p, nil
 994         } else if u.Scheme == "https" || u.Scheme == "wss" {
 995                 return "443", nil
 996         } else {
 997                 return "80", nil
 998         }
 999 }
1000
1001 func availablePort(host string) (string, error) {
1002         ln, err := net.Listen("tcp", net.JoinHostPort(host, "0"))
1003         if err != nil {
1004                 return "", err
1005         }
1006         defer ln.Close()
1007         _, port, err := net.SplitHostPort(ln.Addr().String())
1008         if err != nil {
1009                 return "", err
1010         }
1011         return port, nil
1012 }
1013
1014 // Try to connect to addr until it works, then close ch. Give up if
1015 // ctx cancels.
1016 func waitForConnect(ctx context.Context, addr string) error {
1017         ctxlog.FromContext(ctx).WithField("addr", addr).Info("waitForConnect")
1018         dialer := net.Dialer{Timeout: time.Second}
1019         for ctx.Err() == nil {
1020                 conn, err := dialer.DialContext(ctx, "tcp", addr)
1021                 if err != nil {
1022                         time.Sleep(time.Second / 10)
1023                         continue
1024                 }
1025                 conn.Close()
1026                 return nil
1027         }
1028         return ctx.Err()
1029 }
1030
1031 func copyConfig(cfg *arvados.Config) *arvados.Config {
1032         pr, pw := io.Pipe()
1033         go func() {
1034                 err := json.NewEncoder(pw).Encode(cfg)
1035                 if err != nil {
1036                         panic(err)
1037                 }
1038                 pw.Close()
1039         }()
1040         cfg2 := new(arvados.Config)
1041         err := json.NewDecoder(pr).Decode(cfg2)
1042         if err != nil {
1043                 panic(err)
1044         }
1045         return cfg2
1046 }
1047
1048 func watchConfig(ctx context.Context, logger logrus.FieldLogger, cfgPath string, prevcfg *arvados.Config, fn func()) {
1049         watcher, err := fsnotify.NewWatcher()
1050         if err != nil {
1051                 logger.WithError(err).Error("fsnotify setup failed")
1052                 return
1053         }
1054         defer watcher.Close()
1055
1056         err = watcher.Add(cfgPath)
1057         if err != nil {
1058                 logger.WithError(err).Error("fsnotify watcher failed")
1059                 return
1060         }
1061
1062         for {
1063                 select {
1064                 case <-ctx.Done():
1065                         return
1066                 case err, ok := <-watcher.Errors:
1067                         if !ok {
1068                                 return
1069                         }
1070                         logger.WithError(err).Warn("fsnotify watcher reported error")
1071                 case _, ok := <-watcher.Events:
1072                         if !ok {
1073                                 return
1074                         }
1075                         for len(watcher.Events) > 0 {
1076                                 <-watcher.Events
1077                         }
1078                         loader := config.NewLoader(&bytes.Buffer{}, &logrus.Logger{Out: ioutil.Discard})
1079                         loader.Path = cfgPath
1080                         loader.SkipAPICalls = true
1081                         cfg, err := loader.Load()
1082                         if err != nil {
1083                                 logger.WithError(err).Warn("error reloading config file after change detected; ignoring new config for now")
1084                         } else if reflect.DeepEqual(cfg, prevcfg) {
1085                                 logger.Debug("config file changed but is still DeepEqual to the existing config")
1086                         } else {
1087                                 logger.Debug("config changed, notifying supervisor")
1088                                 fn()
1089                                 prevcfg = cfg
1090                         }
1091                 }
1092         }
1093 }