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
  65         NoWorkbench1         bool
  66         NoWorkbench2         bool
  67         OwnTemporaryDatabase bool
  68         Stdin                io.Reader
  69         Stderr               io.Writer
  70
  71         logger   logrus.FieldLogger
  72         cluster  *arvados.Cluster       // nil if this is a multi-cluster supervisor
  73         children map[string]*Supervisor // nil if this is a single-cluster supervisor
  74
  75         ctx           context.Context
  76         cancel        context.CancelFunc
  77         done          chan struct{}      // closed when child procs/services have shut down
  78         err           error              // error that caused shutdown (valid when done is closed)
  79         healthChecker *health.Aggregator // nil if this is a multi-cluster supervisor, or still booting
  80         tasksReady    map[string]chan bool
  81         waitShutdown  sync.WaitGroup
  82
  83         bindir     string
  84         tempdir    string // in production mode, this is accessible only to root
  85         wwwtempdir string // in production mode, this is accessible only to www-data
  86         configfile string
  87         environ    []string // for child processes
  88 }
  89
  90 func (super *Supervisor) Clusters() map[string]*arvados.Cluster {
  91         m := map[string]*arvados.Cluster{}
  92         if super.cluster != nil {
  93                 m[super.cluster.ClusterID] = super.cluster
  94         }
  95         for id, super2 := range super.children {
  96                 m[id] = super2.Cluster("")
  97         }
  98         return m
  99 }
 100
 101 func (super *Supervisor) Cluster(id string) *arvados.Cluster {
 102         if super.children != nil {
 103                 return super.children[id].Cluster(id)
 104         } else {
 105                 return super.cluster
 106         }
 107 }
 108
 109 func (super *Supervisor) Start(ctx context.Context) {
 110         super.logger = ctxlog.FromContext(ctx)
 111         super.ctx, super.cancel = context.WithCancel(ctx)
 112         super.done = make(chan struct{})
 113
 114         sigch := make(chan os.Signal)
 115         signal.Notify(sigch, syscall.SIGINT, syscall.SIGTERM, syscall.SIGHUP)
 116         go func() {
 117                 defer signal.Stop(sigch)
 118                 for {
 119                         select {
 120                         case <-ctx.Done():
 121                                 return
 122                         case sig := <-sigch:
 123                                 super.logger.WithField("signal", sig).Info("caught signal")
 124                                 if super.err == nil {
 125                                         if sig == syscall.SIGHUP {
 126                                                 super.err = errNeedConfigReload
 127                                         } else {
 128                                                 super.err = fmt.Errorf("caught signal %s", sig)
 129                                         }
 130                                 }
 131                                 super.cancel()
 132                         }
 133                 }
 134         }()
 135
 136         loaderStdin := super.Stdin
 137         if super.ConfigYAML != "" {
 138                 loaderStdin = bytes.NewBufferString(super.ConfigYAML)
 139         }
 140         loader := config.NewLoader(loaderStdin, super.logger)
 141         loader.SkipLegacy = true
 142         loader.SkipAPICalls = true
 143         loader.Path = super.ConfigPath
 144         if super.ConfigYAML != "" {
 145                 loader.Path = "-"
 146         }
 147         cfg, err := loader.Load()
 148         if err != nil {
 149                 super.err = err
 150                 close(super.done)
 151                 super.cancel()
 152                 return
 153         }
 154
 155         if super.ConfigPath != "" && super.ConfigPath != "-" && cfg.AutoReloadConfig {
 156                 go watchConfig(super.ctx, super.logger, super.ConfigPath, copyConfig(cfg), func() {
 157                         if super.err == nil {
 158                                 super.err = errNeedConfigReload
 159                         }
 160                         super.cancel()
 161                 })
 162         }
 163
 164         if len(cfg.Clusters) > 1 {
 165                 super.startFederation(cfg)
 166                 go func() {
 167                         defer super.cancel()
 168                         defer close(super.done)
 169                         for _, super2 := range super.children {
 170                                 err := super2.Wait()
 171                                 if super.err == nil {
 172                                         super.err = err
 173                                 }
 174                         }
 175                 }()
 176         } else {
 177                 go func() {
 178                         defer super.cancel()
 179                         defer close(super.done)
 180                         super.cluster, super.err = cfg.GetCluster("")
 181                         if super.err != nil {
 182                                 return
 183                         }
 184                         err := super.runCluster()
 185                         if err != nil {
 186                                 super.logger.WithError(err).Info("supervisor shut down")
 187                                 if super.err == nil {
 188                                         super.err = err
 189                                 }
 190                         }
 191                 }()
 192         }
 193 }
 194
 195 // Wait returns when all child processes and goroutines have exited.
 196 func (super *Supervisor) Wait() error {
 197         <-super.done
 198         return super.err
 199 }
 200
 201 // startFederation starts a child Supervisor for each cluster in the
 202 // given config. Each is a copy of the original/parent with the
 203 // original config reduced to a single cluster.
 204 func (super *Supervisor) startFederation(cfg *arvados.Config) {
 205         super.children = map[string]*Supervisor{}
 206         for id, cc := range cfg.Clusters {
 207                 super2 := *super
 208                 yaml, err := json.Marshal(arvados.Config{Clusters: map[string]arvados.Cluster{id: cc}})
 209                 if err != nil {
 210                         panic(fmt.Sprintf("json.Marshal partial config: %s", err))
 211                 }
 212                 super2.ConfigYAML = string(yaml)
 213                 super2.ConfigPath = "-"
 214                 super2.children = nil
 215
 216                 if super2.ClusterType == "test" {
 217                         super2.Stderr = &service.LogPrefixer{
 218                                 Writer: super.Stderr,
 219                                 Prefix: []byte("[" + id + "] "),
 220                         }
 221                 }
 222                 super2.Start(super.ctx)
 223                 super.children[id] = &super2
 224         }
 225 }
 226
 227 func (super *Supervisor) runCluster() error {
 228         cwd, err := os.Getwd()
 229         if err != nil {
 230                 return err
 231         }
 232         if super.ClusterType == "test" && super.SourcePath == "" {
 233                 // When invoked by test suite, default to current
 234                 // source tree
 235                 buf, err := exec.Command("git", "rev-parse", "--show-toplevel").CombinedOutput()
 236                 if err != nil {
 237                         return fmt.Errorf("git rev-parse: %w", err)
 238                 }
 239                 super.SourcePath = strings.TrimSuffix(string(buf), "\n")
 240         } else if !strings.HasPrefix(super.SourcePath, "/") {
 241                 super.SourcePath = filepath.Join(cwd, super.SourcePath)
 242         }
 243         super.SourcePath, err = filepath.EvalSymlinks(super.SourcePath)
 244         if err != nil {
 245                 return err
 246         }
 247
 248         if super.ListenHost == "" {
 249                 u := url.URL(super.cluster.Services.Controller.ExternalURL)
 250                 super.ListenHost = u.Hostname()
 251                 if super.ListenHost == "" {
 252                         super.ListenHost = "0.0.0.0"
 253                 }
 254         }
 255
 256         // Choose bin and temp dirs: /var/lib/arvados/... in
 257         // production, transient tempdir otherwise.
 258         if super.ClusterType == "production" {
 259                 // These dirs have already been created by
 260                 // "arvados-server install" (or by extracting a
 261                 // package).
 262                 super.tempdir = "/var/lib/arvados/tmp"
 263                 super.wwwtempdir = "/var/lib/arvados/wwwtmp"
 264                 super.bindir = "/var/lib/arvados/bin"
 265         } else {
 266                 super.tempdir, err = ioutil.TempDir("", "arvados-server-boot-")
 267                 if err != nil {
 268                         return err
 269                 }
 270                 defer os.RemoveAll(super.tempdir)
 271                 super.wwwtempdir = super.tempdir
 272                 super.bindir = filepath.Join(super.tempdir, "bin")
 273                 if err := os.Mkdir(super.bindir, 0755); err != nil {
 274                         return err
 275                 }
 276         }
 277
 278         // Fill in any missing config keys, and write the resulting
 279         // config in the temp dir for child services to use.
 280         err = super.autofillConfig()
 281         if err != nil {
 282                 return err
 283         }
 284         conffile, err := os.OpenFile(filepath.Join(super.wwwtempdir, "config.yml"), os.O_CREATE|os.O_WRONLY|os.O_TRUNC, 0644)
 285         if err != nil {
 286                 return err
 287         }
 288         defer conffile.Close()
 289         err = json.NewEncoder(conffile).Encode(arvados.Config{
 290                 Clusters: map[string]arvados.Cluster{
 291                         super.cluster.ClusterID: *super.cluster}})
 292         if err != nil {
 293                 return err
 294         }
 295         err = conffile.Close()
 296         if err != nil {
 297                 return err
 298         }
 299         super.configfile = conffile.Name()
 300
 301         super.environ = os.Environ()
 302         super.cleanEnv([]string{"ARVADOS_"})
 303         super.setEnv("ARVADOS_CONFIG", super.configfile)
 304         super.setEnv("RAILS_ENV", super.ClusterType)
 305         super.setEnv("TMPDIR", super.tempdir)
 306         super.prependEnv("PATH", "/var/lib/arvados/bin:")
 307         if super.ClusterType != "production" {
 308                 super.prependEnv("PATH", super.tempdir+"/bin:")
 309         }
 310         super.setEnv("ARVADOS_SERVER_ADDRESS", super.ListenHost)
 311
 312         // Now that we have the config, replace the bootstrap logger
 313         // with a new one according to the logging config.
 314         loglevel := super.cluster.SystemLogs.LogLevel
 315         if s := os.Getenv("ARVADOS_DEBUG"); s != "" && s != "0" {
 316                 loglevel = "debug"
 317         }
 318         super.logger = ctxlog.New(super.Stderr, super.cluster.SystemLogs.Format, loglevel).WithFields(logrus.Fields{
 319                 "PID": os.Getpid(),
 320         })
 321
 322         if super.SourceVersion == "" && super.ClusterType == "production" {
 323                 // don't need SourceVersion
 324         } else if super.SourceVersion == "" {
 325                 // Find current source tree version.
 326                 var buf bytes.Buffer
 327                 err = super.RunProgram(super.ctx, super.SourcePath, runOptions{output: &buf}, "git", "diff", "--shortstat")
 328                 if err != nil {
 329                         return err
 330                 }
 331                 dirty := buf.Len() > 0
 332                 buf.Reset()
 333                 err = super.RunProgram(super.ctx, super.SourcePath, runOptions{output: &buf}, "git", "log", "-n1", "--format=%H")
 334                 if err != nil {
 335                         return err
 336                 }
 337                 super.SourceVersion = strings.TrimSpace(buf.String())
 338                 if dirty {
 339                         super.SourceVersion += "+uncommitted"
 340                 }
 341         } else {
 342                 return errors.New("specifying a version to run is not yet supported")
 343         }
 344
 345         _, err = super.installGoProgram(super.ctx, "cmd/arvados-server")
 346         if err != nil {
 347                 return err
 348         }
 349         err = super.setupRubyEnv()
 350         if err != nil {
 351                 return err
 352         }
 353
 354         tasks := []supervisedTask{
 355                 createCertificates{},
 356                 runPostgreSQL{},
 357                 runNginx{},
 358                 railsDatabase{},
 359                 runServiceCommand{name: "controller", svc: super.cluster.Services.Controller, depends: []supervisedTask{railsDatabase{}}},
 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{railsDatabase{}}},
 366                 installPassenger{src: "services/api", varlibdir: "railsapi"},
 367                 runPassenger{src: "services/api", varlibdir: "railsapi", svc: super.cluster.Services.RailsAPI, depends: []supervisedTask{
 368                         createCertificates{},
 369                         installPassenger{src: "services/api", varlibdir: "railsapi"},
 370                         railsDatabase{},
 371                 }},
 372         }
 373         if !super.NoWorkbench1 {
 374                 tasks = append(tasks,
 375                         installPassenger{src: "apps/workbench", varlibdir: "workbench1", depends: []supervisedTask{railsDatabase{}}}, // dependency ensures workbench doesn't delay api install/startup
 376                         runPassenger{src: "apps/workbench", varlibdir: "workbench1", svc: super.cluster.Services.Workbench1, depends: []supervisedTask{installPassenger{src: "apps/workbench", varlibdir: "workbench1"}}},
 377                 )
 378         }
 379         if !super.NoWorkbench2 {
 380                 tasks = append(tasks,
 381                         runWorkbench2{svc: super.cluster.Services.Workbench2},
 382                 )
 383         }
 384         if super.ClusterType != "test" {
 385                 tasks = append(tasks,
 386                         runServiceCommand{name: "keep-balance", svc: super.cluster.Services.Keepbalance},
 387                 )
 388         }
 389         if super.cluster.Containers.CloudVMs.Enable {
 390                 tasks = append(tasks,
 391                         runServiceCommand{name: "dispatch-cloud", svc: super.cluster.Services.DispatchCloud},
 392                 )
 393         }
 394         super.tasksReady = map[string]chan bool{}
 395         for _, task := range tasks {
 396                 super.tasksReady[task.String()] = make(chan bool)
 397         }
 398         for _, task := range tasks {
 399                 task := task
 400                 fail := func(err error) {
 401                         if super.ctx.Err() != nil {
 402                                 return
 403                         }
 404                         super.cancel()
 405                         super.logger.WithField("task", task.String()).WithError(err).Error("task failed")
 406                 }
 407                 go func() {
 408                         super.logger.WithField("task", task.String()).Info("starting")
 409                         err := task.Run(super.ctx, fail, super)
 410                         if err != nil {
 411                                 fail(err)
 412                                 return
 413                         }
 414                         close(super.tasksReady[task.String()])
 415                 }()
 416         }
 417         err = super.wait(super.ctx, tasks...)
 418         if err != nil {
 419                 return err
 420         }
 421         super.logger.Info("all startup tasks are complete; starting health checks")
 422         super.healthChecker = &health.Aggregator{Cluster: super.cluster}
 423         <-super.ctx.Done()
 424         super.logger.Info("shutting down")
 425         super.waitShutdown.Wait()
 426         return super.ctx.Err()
 427 }
 428
 429 func (super *Supervisor) wait(ctx context.Context, tasks ...supervisedTask) error {
 430         ticker := time.NewTicker(15 * time.Second)
 431         defer ticker.Stop()
 432         for _, task := range tasks {
 433                 ch, ok := super.tasksReady[task.String()]
 434                 if !ok {
 435                         return fmt.Errorf("no such task: %s", task)
 436                 }
 437                 super.logger.WithField("task", task.String()).Info("waiting")
 438                 for {
 439                         select {
 440                         case <-ch:
 441                                 super.logger.WithField("task", task.String()).Info("ready")
 442                         case <-ctx.Done():
 443                                 super.logger.WithField("task", task.String()).Info("task was never ready")
 444                                 return ctx.Err()
 445                         case <-ticker.C:
 446                                 super.logger.WithField("task", task.String()).Info("still waiting...")
 447                                 continue
 448                         }
 449                         break
 450                 }
 451         }
 452         return nil
 453 }
 454
 455 // Stop shuts down all child processes and goroutines, and returns
 456 // when all of them have exited.
 457 func (super *Supervisor) Stop() {
 458         super.cancel()
 459         <-super.done
 460 }
 461
 462 // WaitReady waits for the cluster(s) to be ready to handle requests,
 463 // then returns true. If startup fails, it returns false.
 464 func (super *Supervisor) WaitReady() bool {
 465         if super.children != nil {
 466                 for id, super2 := range super.children {
 467                         super.logger.Infof("waiting for %s to be ready", id)
 468                         if !super2.WaitReady() {
 469                                 super.logger.Infof("%s startup failed", id)
 470                                 super.Stop()
 471                                 return false
 472                         }
 473                         super.logger.Infof("%s is ready", id)
 474                 }
 475                 super.logger.Info("all clusters are ready")
 476                 return true
 477         }
 478         ticker := time.NewTicker(time.Second)
 479         defer ticker.Stop()
 480         for waiting := "all"; waiting != ""; {
 481                 select {
 482                 case <-ticker.C:
 483                 case <-super.ctx.Done():
 484                         super.Stop()
 485                         return false
 486                 }
 487                 if super.healthChecker == nil {
 488                         // not set up yet
 489                         continue
 490                 }
 491                 resp := super.healthChecker.ClusterHealth()
 492                 // The overall health check (resp.Health=="OK") might
 493                 // never pass due to missing components (like
 494                 // arvados-dispatch-cloud in a test cluster), so
 495                 // instead we wait for all configured components to
 496                 // pass.
 497                 waiting = ""
 498                 for target, check := range resp.Checks {
 499                         if check.Health != "OK" {
 500                                 waiting += " " + target
 501                         }
 502                 }
 503                 if waiting != "" {
 504                         super.logger.WithField("targets", waiting[1:]).Info("waiting")
 505                 }
 506         }
 507         return true
 508 }
 509
 510 func (super *Supervisor) prependEnv(key, prepend string) {
 511         for i, s := range super.environ {
 512                 if strings.HasPrefix(s, key+"=") {
 513                         super.environ[i] = key + "=" + prepend + s[len(key)+1:]
 514                         return
 515                 }
 516         }
 517         super.environ = append(super.environ, key+"="+prepend)
 518 }
 519
 520 func (super *Supervisor) cleanEnv(prefixes []string) {
 521         var cleaned []string
 522         for _, s := range super.environ {
 523                 drop := false
 524                 for _, p := range prefixes {
 525                         if strings.HasPrefix(s, p) {
 526                                 drop = true
 527                                 break
 528                         }
 529                 }
 530                 if !drop {
 531                         cleaned = append(cleaned, s)
 532                 }
 533         }
 534         super.environ = cleaned
 535 }
 536
 537 func (super *Supervisor) setEnv(key, val string) {
 538         for i, s := range super.environ {
 539                 if strings.HasPrefix(s, key+"=") {
 540                         super.environ[i] = key + "=" + val
 541                         return
 542                 }
 543         }
 544         super.environ = append(super.environ, key+"="+val)
 545 }
 546
 547 // Remove all but the first occurrence of each env var.
 548 func dedupEnv(in []string) []string {
 549         saw := map[string]bool{}
 550         var out []string
 551         for _, kv := range in {
 552                 if split := strings.Index(kv, "="); split < 1 {
 553                         panic("invalid environment var: " + kv)
 554                 } else if saw[kv[:split]] {
 555                         continue
 556                 } else {
 557                         saw[kv[:split]] = true
 558                         out = append(out, kv)
 559                 }
 560         }
 561         return out
 562 }
 563
 564 func (super *Supervisor) installGoProgram(ctx context.Context, srcpath string) (string, error) {
 565         _, basename := filepath.Split(srcpath)
 566         binfile := filepath.Join(super.bindir, basename)
 567         if super.ClusterType == "production" {
 568                 return binfile, nil
 569         }
 570         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)
 571         return binfile, err
 572 }
 573
 574 func (super *Supervisor) usingRVM() bool {
 575         return os.Getenv("rvm_path") != ""
 576 }
 577
 578 func (super *Supervisor) setupRubyEnv() error {
 579         if !super.usingRVM() {
 580                 // (If rvm is in use, assume the caller has everything
 581                 // set up as desired)
 582                 super.cleanEnv([]string{
 583                         "GEM_HOME=",
 584                         "GEM_PATH=",
 585                 })
 586                 gem := "gem"
 587                 if _, err := os.Stat("/var/lib/arvados/bin/gem"); err == nil || super.ClusterType == "production" {
 588                         gem = "/var/lib/arvados/bin/gem"
 589                 }
 590                 cmd := exec.Command(gem, "env", "gempath")
 591                 if super.ClusterType == "production" {
 592                         cmd.Args = append([]string{"sudo", "-u", "www-data", "-E", "HOME=/var/www"}, cmd.Args...)
 593                         path, err := exec.LookPath("sudo")
 594                         if err != nil {
 595                                 return fmt.Errorf("LookPath(\"sudo\"): %w", err)
 596                         }
 597                         cmd.Path = path
 598                 }
 599                 cmd.Stderr = super.Stderr
 600                 cmd.Env = super.environ
 601                 buf, err := cmd.Output() // /var/lib/arvados/.gem/ruby/2.5.0/bin:...
 602                 if err != nil || len(buf) == 0 {
 603                         return fmt.Errorf("gem env gempath: %w", err)
 604                 }
 605                 gempath := string(bytes.Split(buf, []byte{':'})[0])
 606                 super.prependEnv("PATH", gempath+"/bin:")
 607                 super.setEnv("GEM_HOME", gempath)
 608                 super.setEnv("GEM_PATH", gempath)
 609         }
 610         // Passenger install doesn't work unless $HOME is ~user
 611         u, err := user.Current()
 612         if err != nil {
 613                 return err
 614         }
 615         super.setEnv("HOME", u.HomeDir)
 616         return nil
 617 }
 618
 619 func (super *Supervisor) lookPath(prog string) string {
 620         for _, val := range super.environ {
 621                 if strings.HasPrefix(val, "PATH=") {
 622                         for _, dir := range filepath.SplitList(val[5:]) {
 623                                 path := filepath.Join(dir, prog)
 624                                 if fi, err := os.Stat(path); err == nil && fi.Mode()&0111 != 0 {
 625                                         return path
 626                                 }
 627                         }
 628                 }
 629         }
 630         return prog
 631 }
 632
 633 type runOptions struct {
 634         output io.Writer // attach stdout
 635         env    []string  // add/replace environment variables
 636         user   string    // run as specified user
 637         stdin  io.Reader
 638 }
 639
 640 // RunProgram runs prog with args, using dir as working directory. If ctx is
 641 // cancelled while the child is running, RunProgram terminates the child, waits
 642 // for it to exit, then returns.
 643 //
 644 // Child's environment will have our env vars, plus any given in env.
 645 //
 646 // Child's stdout will be written to output if non-nil, otherwise the
 647 // boot command's stderr.
 648 func (super *Supervisor) RunProgram(ctx context.Context, dir string, opts runOptions, prog string, args ...string) error {
 649         cmdline := fmt.Sprintf("%s", append([]string{prog}, args...))
 650         super.logger.WithField("command", cmdline).WithField("dir", dir).Info("executing")
 651
 652         logprefix := prog
 653         {
 654                 innerargs := args
 655                 if logprefix == "sudo" {
 656                         for i := 0; i < len(args); i++ {
 657                                 if args[i] == "-u" {
 658                                         i++
 659                                 } else if args[i] == "-E" || strings.Contains(args[i], "=") {
 660                                 } else {
 661                                         logprefix = args[i]
 662                                         innerargs = args[i+1:]
 663                                         break
 664                                 }
 665                         }
 666                 }
 667                 logprefix = strings.TrimPrefix(logprefix, "/var/lib/arvados/bin/")
 668                 logprefix = strings.TrimPrefix(logprefix, super.tempdir+"/bin/")
 669                 if logprefix == "bundle" && len(innerargs) > 2 && innerargs[0] == "exec" {
 670                         _, dirbase := filepath.Split(dir)
 671                         logprefix = innerargs[1] + "@" + dirbase
 672                 } else if logprefix == "arvados-server" && len(args) > 1 {
 673                         logprefix = args[0]
 674                 }
 675                 if !strings.HasPrefix(dir, "/") {
 676                         logprefix = dir + ": " + logprefix
 677                 }
 678         }
 679
 680         cmd := exec.Command(super.lookPath(prog), args...)
 681         cmd.Stdin = opts.stdin
 682         stdout, err := cmd.StdoutPipe()
 683         if err != nil {
 684                 return err
 685         }
 686         stderr, err := cmd.StderrPipe()
 687         if err != nil {
 688                 return err
 689         }
 690         logwriter := &service.LogPrefixer{Writer: super.Stderr, Prefix: []byte("[" + logprefix + "] ")}
 691         var copiers sync.WaitGroup
 692         copiers.Add(1)
 693         go func() {
 694                 io.Copy(logwriter, stderr)
 695                 copiers.Done()
 696         }()
 697         copiers.Add(1)
 698         go func() {
 699                 if opts.output == nil {
 700                         io.Copy(logwriter, stdout)
 701                 } else {
 702                         io.Copy(opts.output, stdout)
 703                 }
 704                 copiers.Done()
 705         }()
 706
 707         if strings.HasPrefix(dir, "/") {
 708                 cmd.Dir = dir
 709         } else {
 710                 cmd.Dir = filepath.Join(super.SourcePath, dir)
 711         }
 712         env := append([]string(nil), opts.env...)
 713         env = append(env, super.environ...)
 714         cmd.Env = dedupEnv(env)
 715
 716         if opts.user != "" {
 717                 // Note: We use this approach instead of "sudo"
 718                 // because in certain circumstances (we are pid 1 in a
 719                 // docker container, and our passenger child process
 720                 // changes to pgid 1) the intermediate sudo process
 721                 // notices we have the same pgid as our child and
 722                 // refuses to propagate signals from us to our child,
 723                 // so we can't signal/shutdown our passenger/rails
 724                 // apps. "chpst" or "setuidgid" would work, but these
 725                 // few lines avoid depending on runit/daemontools.
 726                 u, err := user.Lookup(opts.user)
 727                 if err != nil {
 728                         return fmt.Errorf("user.Lookup(%q): %w", opts.user, err)
 729                 }
 730                 uid, _ := strconv.Atoi(u.Uid)
 731                 gid, _ := strconv.Atoi(u.Gid)
 732                 cmd.SysProcAttr = &syscall.SysProcAttr{
 733                         Credential: &syscall.Credential{
 734                                 Uid: uint32(uid),
 735                                 Gid: uint32(gid),
 736                         },
 737                 }
 738         }
 739
 740         exited := false
 741         defer func() { exited = true }()
 742         go func() {
 743                 <-ctx.Done()
 744                 log := ctxlog.FromContext(ctx).WithFields(logrus.Fields{"dir": dir, "cmdline": cmdline})
 745                 for !exited {
 746                         if cmd.Process == nil {
 747                                 log.Debug("waiting for child process to start")
 748                                 time.Sleep(time.Second / 2)
 749                         } else {
 750                                 log.WithField("PID", cmd.Process.Pid).Debug("sending SIGTERM")
 751                                 cmd.Process.Signal(syscall.SIGTERM)
 752                                 time.Sleep(5 * time.Second)
 753                                 if !exited {
 754                                         stdout.Close()
 755                                         stderr.Close()
 756                                         log.WithField("PID", cmd.Process.Pid).Warn("still waiting for child process to exit 5s after SIGTERM")
 757                                 }
 758                         }
 759                 }
 760         }()
 761
 762         err = cmd.Start()
 763         if err != nil {
 764                 return err
 765         }
 766         copiers.Wait()
 767         err = cmd.Wait()
 768         if ctx.Err() != nil {
 769                 // Return "context canceled", instead of the "killed"
 770                 // error that was probably caused by the context being
 771                 // canceled.
 772                 return ctx.Err()
 773         } else if err != nil {
 774                 return fmt.Errorf("%s: error: %v", cmdline, err)
 775         }
 776         return nil
 777 }
 778
 779 func (super *Supervisor) autofillConfig() error {
 780         usedPort := map[string]bool{}
 781         nextPort := func(host string) (string, error) {
 782                 for {
 783                         port, err := availablePort(host)
 784                         if err != nil {
 785                                 port, err = availablePort(super.ListenHost)
 786                         }
 787                         if err != nil {
 788                                 return "", err
 789                         }
 790                         if usedPort[port] {
 791                                 continue
 792                         }
 793                         usedPort[port] = true
 794                         return port, nil
 795                 }
 796         }
 797         if super.cluster.Services.Controller.ExternalURL.Host == "" {
 798                 h, p, err := net.SplitHostPort(super.ControllerAddr)
 799                 if err != nil && super.ControllerAddr != "" {
 800                         return fmt.Errorf("SplitHostPort(ControllerAddr %q): %w", super.ControllerAddr, err)
 801                 }
 802                 if h == "" {
 803                         h = super.ListenHost
 804                 }
 805                 if p == "0" || p == "" {
 806                         p, err = nextPort(h)
 807                         if err != nil {
 808                                 return err
 809                         }
 810                 }
 811                 super.cluster.Services.Controller.ExternalURL = arvados.URL{Scheme: "https", Host: net.JoinHostPort(h, p), Path: "/"}
 812         }
 813         u := url.URL(super.cluster.Services.Controller.ExternalURL)
 814         defaultExtHost := u.Hostname()
 815         for _, svc := range []*arvados.Service{
 816                 &super.cluster.Services.Controller,
 817                 &super.cluster.Services.DispatchCloud,
 818                 &super.cluster.Services.GitHTTP,
 819                 &super.cluster.Services.Health,
 820                 &super.cluster.Services.Keepproxy,
 821                 &super.cluster.Services.Keepstore,
 822                 &super.cluster.Services.RailsAPI,
 823                 &super.cluster.Services.WebDAV,
 824                 &super.cluster.Services.WebDAVDownload,
 825                 &super.cluster.Services.Websocket,
 826                 &super.cluster.Services.Workbench1,
 827                 &super.cluster.Services.Workbench2,
 828         } {
 829                 if svc.ExternalURL.Host == "" {
 830                         port, err := nextPort(defaultExtHost)
 831                         if err != nil {
 832                                 return err
 833                         }
 834                         host := net.JoinHostPort(defaultExtHost, port)
 835                         if svc == &super.cluster.Services.Controller ||
 836                                 svc == &super.cluster.Services.GitHTTP ||
 837                                 svc == &super.cluster.Services.Health ||
 838                                 svc == &super.cluster.Services.Keepproxy ||
 839                                 svc == &super.cluster.Services.WebDAV ||
 840                                 svc == &super.cluster.Services.WebDAVDownload ||
 841                                 svc == &super.cluster.Services.Workbench1 ||
 842                                 svc == &super.cluster.Services.Workbench2 {
 843                                 svc.ExternalURL = arvados.URL{Scheme: "https", Host: host, Path: "/"}
 844                         } else if svc == &super.cluster.Services.Websocket {
 845                                 svc.ExternalURL = arvados.URL{Scheme: "wss", Host: host, Path: "/websocket"}
 846                         }
 847                 }
 848                 if super.NoWorkbench1 && svc == &super.cluster.Services.Workbench1 ||
 849                         super.NoWorkbench2 && svc == &super.cluster.Services.Workbench2 ||
 850                         !super.cluster.Containers.CloudVMs.Enable && svc == &super.cluster.Services.DispatchCloud {
 851                         // When workbench1 is disabled, it gets an
 852                         // ExternalURL (so we have a valid listening
 853                         // port to write in our Nginx config) but no
 854                         // InternalURLs (so health checker doesn't
 855                         // complain).
 856                         continue
 857                 }
 858                 if len(svc.InternalURLs) == 0 {
 859                         port, err := nextPort(super.ListenHost)
 860                         if err != nil {
 861                                 return err
 862                         }
 863                         host := net.JoinHostPort(super.ListenHost, port)
 864                         svc.InternalURLs = map[arvados.URL]arvados.ServiceInstance{
 865                                 {Scheme: "http", Host: host, Path: "/"}: {},
 866                         }
 867                 }
 868         }
 869         if super.ClusterType != "production" {
 870                 if super.cluster.SystemRootToken == "" {
 871                         super.cluster.SystemRootToken = randomHexString(64)
 872                 }
 873                 if super.cluster.ManagementToken == "" {
 874                         super.cluster.ManagementToken = randomHexString(64)
 875                 }
 876                 if super.cluster.Collections.BlobSigningKey == "" {
 877                         super.cluster.Collections.BlobSigningKey = randomHexString(64)
 878                 }
 879                 if super.cluster.Users.AnonymousUserToken == "" {
 880                         super.cluster.Users.AnonymousUserToken = randomHexString(64)
 881                 }
 882                 if super.cluster.Containers.DispatchPrivateKey == "" {
 883                         buf, err := ioutil.ReadFile(filepath.Join(super.SourcePath, "lib", "dispatchcloud", "test", "sshkey_dispatch"))
 884                         if err != nil {
 885                                 return err
 886                         }
 887                         super.cluster.Containers.DispatchPrivateKey = string(buf)
 888                 }
 889                 super.cluster.TLS.Insecure = true
 890         }
 891         if super.ClusterType == "test" {
 892                 // Add a second keepstore process.
 893                 port, err := nextPort(super.ListenHost)
 894                 if err != nil {
 895                         return err
 896                 }
 897                 host := net.JoinHostPort(super.ListenHost, port)
 898                 super.cluster.Services.Keepstore.InternalURLs[arvados.URL{Scheme: "http", Host: host, Path: "/"}] = arvados.ServiceInstance{}
 899
 900                 // Create a directory-backed volume for each keepstore
 901                 // process.
 902                 super.cluster.Volumes = map[string]arvados.Volume{}
 903                 for url := range super.cluster.Services.Keepstore.InternalURLs {
 904                         volnum := len(super.cluster.Volumes)
 905                         datadir := fmt.Sprintf("%s/keep%d.data", super.tempdir, volnum)
 906                         if _, err = os.Stat(datadir + "/."); err == nil {
 907                         } else if !os.IsNotExist(err) {
 908                                 return err
 909                         } else if err = os.Mkdir(datadir, 0755); err != nil {
 910                                 return err
 911                         }
 912                         super.cluster.Volumes[fmt.Sprintf(super.cluster.ClusterID+"-nyw5e-%015d", volnum)] = arvados.Volume{
 913                                 Driver:           "Directory",
 914                                 DriverParameters: json.RawMessage(fmt.Sprintf(`{"Root":%q}`, datadir)),
 915                                 AccessViaHosts: map[arvados.URL]arvados.VolumeAccess{
 916                                         url: {},
 917                                 },
 918                                 StorageClasses: map[string]bool{
 919                                         "default": true,
 920                                         "foo":     true,
 921                                         "bar":     true,
 922                                 },
 923                         }
 924                 }
 925                 super.cluster.StorageClasses = map[string]arvados.StorageClassConfig{
 926                         "default": {Default: true},
 927                         "foo":     {},
 928                         "bar":     {},
 929                 }
 930         }
 931         if super.OwnTemporaryDatabase {
 932                 port, err := nextPort("localhost")
 933                 if err != nil {
 934                         return err
 935                 }
 936                 super.cluster.PostgreSQL.Connection = arvados.PostgreSQLConnection{
 937                         "client_encoding": "utf8",
 938                         "host":            "localhost",
 939                         "port":            port,
 940                         "dbname":          "arvados_test",
 941                         "user":            "arvados",
 942                         "password":        "insecure_arvados_test",
 943                 }
 944         }
 945         return nil
 946 }
 947
 948 func addrIsLocal(addr string) (bool, error) {
 949         if h, _, err := net.SplitHostPort(addr); err != nil {
 950                 return false, err
 951         } else {
 952                 addr = net.JoinHostPort(h, "0")
 953         }
 954         listener, err := net.Listen("tcp", addr)
 955         if err == nil {
 956                 listener.Close()
 957                 return true, nil
 958         } else if strings.Contains(err.Error(), "cannot assign requested address") {
 959                 return false, nil
 960         } else {
 961                 return false, err
 962         }
 963 }
 964
 965 func randomHexString(chars int) string {
 966         b := make([]byte, chars/2)
 967         _, err := rand.Read(b)
 968         if err != nil {
 969                 panic(err)
 970         }
 971         return fmt.Sprintf("%x", b)
 972 }
 973
 974 func internalPort(svc arvados.Service) (host, port string, err error) {
 975         if len(svc.InternalURLs) > 1 {
 976                 return "", "", errors.New("internalPort() doesn't work with multiple InternalURLs")
 977         }
 978         for u := range svc.InternalURLs {
 979                 u := url.URL(u)
 980                 host, port = u.Hostname(), u.Port()
 981                 switch {
 982                 case port != "":
 983                 case u.Scheme == "https", u.Scheme == "ws":
 984                         port = "443"
 985                 default:
 986                         port = "80"
 987                 }
 988                 return
 989         }
 990         return "", "", fmt.Errorf("service has no InternalURLs")
 991 }
 992
 993 func externalPort(svc arvados.Service) (string, error) {
 994         u := url.URL(svc.ExternalURL)
 995         if p := u.Port(); p != "" {
 996                 return p, nil
 997         } else if u.Scheme == "https" || u.Scheme == "wss" {
 998                 return "443", nil
 999         } else {
1000                 return "80", nil
1001         }
1002 }
1003
1004 func availablePort(host string) (string, error) {
1005         ln, err := net.Listen("tcp", net.JoinHostPort(host, "0"))
1006         if err != nil {
1007                 return "", err
1008         }
1009         defer ln.Close()
1010         _, port, err := net.SplitHostPort(ln.Addr().String())
1011         if err != nil {
1012                 return "", err
1013         }
1014         return port, nil
1015 }
1016
1017 // Try to connect to addr until it works, then close ch. Give up if
1018 // ctx cancels.
1019 func waitForConnect(ctx context.Context, addr string) error {
1020         ctxlog.FromContext(ctx).WithField("addr", addr).Info("waitForConnect")
1021         dialer := net.Dialer{Timeout: time.Second}
1022         for ctx.Err() == nil {
1023                 conn, err := dialer.DialContext(ctx, "tcp", addr)
1024                 if err != nil {
1025                         time.Sleep(time.Second / 10)
1026                         continue
1027                 }
1028                 conn.Close()
1029                 return nil
1030         }
1031         return ctx.Err()
1032 }
1033
1034 func copyConfig(cfg *arvados.Config) *arvados.Config {
1035         pr, pw := io.Pipe()
1036         go func() {
1037                 err := json.NewEncoder(pw).Encode(cfg)
1038                 if err != nil {
1039                         panic(err)
1040                 }
1041                 pw.Close()
1042         }()
1043         cfg2 := new(arvados.Config)
1044         err := json.NewDecoder(pr).Decode(cfg2)
1045         if err != nil {
1046                 panic(err)
1047         }
1048         return cfg2
1049 }
1050
1051 func watchConfig(ctx context.Context, logger logrus.FieldLogger, cfgPath string, prevcfg *arvados.Config, fn func()) {
1052         watcher, err := fsnotify.NewWatcher()
1053         if err != nil {
1054                 logger.WithError(err).Error("fsnotify setup failed")
1055                 return
1056         }
1057         defer watcher.Close()
1058
1059         err = watcher.Add(cfgPath)
1060         if err != nil {
1061                 logger.WithError(err).Error("fsnotify watcher failed")
1062                 return
1063         }
1064
1065         for {
1066                 select {
1067                 case <-ctx.Done():
1068                         return
1069                 case err, ok := <-watcher.Errors:
1070                         if !ok {
1071                                 return
1072                         }
1073                         logger.WithError(err).Warn("fsnotify watcher reported error")
1074                 case _, ok := <-watcher.Events:
1075                         if !ok {
1076                                 return
1077                         }
1078                         for len(watcher.Events) > 0 {
1079                                 <-watcher.Events
1080                         }
1081                         loader := config.NewLoader(&bytes.Buffer{}, &logrus.Logger{Out: ioutil.Discard})
1082                         loader.Path = cfgPath
1083                         loader.SkipAPICalls = true
1084                         cfg, err := loader.Load()
1085                         if err != nil {
1086                                 logger.WithError(err).Warn("error reloading config file after change detected; ignoring new config for now")
1087                         } else if reflect.DeepEqual(cfg, prevcfg) {
1088                                 logger.Debug("config file changed but is still DeepEqual to the existing config")
1089                         } else {
1090                                 logger.Debug("config changed, notifying supervisor")
1091                                 fn()
1092                                 prevcfg = cfg
1093                         }
1094                 }
1095         }
1096 }