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, 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
 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, ".", 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, ".", 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                 runServiceCommand{name: "controller", svc: super.cluster.Services.Controller, depends: []supervisedTask{seedDatabase{}}},
 359                 runServiceCommand{name: "git-httpd", svc: super.cluster.Services.GitHTTP},
 360                 runServiceCommand{name: "health", svc: super.cluster.Services.Health},
 361                 runServiceCommand{name: "keepproxy", svc: super.cluster.Services.Keepproxy, depends: []supervisedTask{runPassenger{src: "services/api"}}},
 362                 runServiceCommand{name: "keepstore", svc: super.cluster.Services.Keepstore},
 363                 runServiceCommand{name: "keep-web", svc: super.cluster.Services.WebDAV},
 364                 runServiceCommand{name: "ws", svc: super.cluster.Services.Websocket, depends: []supervisedTask{seedDatabase{}}},
 365                 installPassenger{src: "services/api", varlibdir: "railsapi"},
 366                 runPassenger{src: "services/api", varlibdir: "railsapi", svc: super.cluster.Services.RailsAPI, depends: []supervisedTask{createCertificates{}, seedDatabase{}, installPassenger{src: "services/api", varlibdir: "railsapi"}}},
 367                 seedDatabase{},
 368         }
 369         if !super.NoWorkbench1 {
 370                 tasks = append(tasks,
 371                         installPassenger{src: "apps/workbench", varlibdir: "workbench1", depends: []supervisedTask{seedDatabase{}}}, // dependency ensures workbench doesn't delay api install/startup
 372                         runPassenger{src: "apps/workbench", varlibdir: "workbench1", svc: super.cluster.Services.Workbench1, depends: []supervisedTask{installPassenger{src: "apps/workbench", varlibdir: "workbench1"}}},
 373                 )
 374         }
 375         if !super.NoWorkbench2 {
 376                 tasks = append(tasks,
 377                         runWorkbench2{svc: super.cluster.Services.Workbench2},
 378                 )
 379         }
 380         if super.ClusterType != "test" {
 381                 tasks = append(tasks,
 382                         runServiceCommand{name: "keep-balance", svc: super.cluster.Services.Keepbalance},
 383                 )
 384         }
 385         if super.cluster.Containers.CloudVMs.Enable {
 386                 tasks = append(tasks,
 387                         runServiceCommand{name: "dispatch-cloud", svc: super.cluster.Services.DispatchCloud},
 388                 )
 389         }
 390         super.tasksReady = map[string]chan bool{}
 391         for _, task := range tasks {
 392                 super.tasksReady[task.String()] = make(chan bool)
 393         }
 394         for _, task := range tasks {
 395                 task := task
 396                 fail := func(err error) {
 397                         if super.ctx.Err() != nil {
 398                                 return
 399                         }
 400                         super.cancel()
 401                         super.logger.WithField("task", task.String()).WithError(err).Error("task failed")
 402                 }
 403                 go func() {
 404                         super.logger.WithField("task", task.String()).Info("starting")
 405                         err := task.Run(super.ctx, fail, super)
 406                         if err != nil {
 407                                 fail(err)
 408                                 return
 409                         }
 410                         close(super.tasksReady[task.String()])
 411                 }()
 412         }
 413         err = super.wait(super.ctx, tasks...)
 414         if err != nil {
 415                 return err
 416         }
 417         super.logger.Info("all startup tasks are complete; starting health checks")
 418         super.healthChecker = &health.Aggregator{Cluster: super.cluster}
 419         <-super.ctx.Done()
 420         super.logger.Info("shutting down")
 421         super.waitShutdown.Wait()
 422         return super.ctx.Err()
 423 }
 424
 425 func (super *Supervisor) wait(ctx context.Context, tasks ...supervisedTask) error {
 426         ticker := time.NewTicker(15 * time.Second)
 427         defer ticker.Stop()
 428         for _, task := range tasks {
 429                 ch, ok := super.tasksReady[task.String()]
 430                 if !ok {
 431                         return fmt.Errorf("no such task: %s", task)
 432                 }
 433                 super.logger.WithField("task", task.String()).Info("waiting")
 434                 for {
 435                         select {
 436                         case <-ch:
 437                                 super.logger.WithField("task", task.String()).Info("ready")
 438                         case <-ctx.Done():
 439                                 super.logger.WithField("task", task.String()).Info("task was never ready")
 440                                 return ctx.Err()
 441                         case <-ticker.C:
 442                                 super.logger.WithField("task", task.String()).Info("still waiting...")
 443                                 continue
 444                         }
 445                         break
 446                 }
 447         }
 448         return nil
 449 }
 450
 451 // Stop shuts down all child processes and goroutines, and returns
 452 // when all of them have exited.
 453 func (super *Supervisor) Stop() {
 454         super.cancel()
 455         <-super.done
 456 }
 457
 458 // WaitReady waits for the cluster(s) to be ready to handle requests,
 459 // then returns true. If startup fails, it returns false.
 460 func (super *Supervisor) WaitReady() bool {
 461         if super.children != nil {
 462                 for id, super2 := range super.children {
 463                         super.logger.Infof("waiting for %s to be ready", id)
 464                         if !super2.WaitReady() {
 465                                 super.logger.Infof("%s startup failed", id)
 466                                 super.Stop()
 467                                 return false
 468                         }
 469                         super.logger.Infof("%s is ready", id)
 470                 }
 471                 super.logger.Info("all clusters are ready")
 472                 return true
 473         }
 474         ticker := time.NewTicker(time.Second)
 475         defer ticker.Stop()
 476         for waiting := "all"; waiting != ""; {
 477                 select {
 478                 case <-ticker.C:
 479                 case <-super.ctx.Done():
 480                         super.Stop()
 481                         return false
 482                 }
 483                 if super.healthChecker == nil {
 484                         // not set up yet
 485                         continue
 486                 }
 487                 resp := super.healthChecker.ClusterHealth()
 488                 // The overall health check (resp.Health=="OK") might
 489                 // never pass due to missing components (like
 490                 // arvados-dispatch-cloud in a test cluster), so
 491                 // instead we wait for all configured components to
 492                 // pass.
 493                 waiting = ""
 494                 for target, check := range resp.Checks {
 495                         if check.Health != "OK" {
 496                                 waiting += " " + target
 497                         }
 498                 }
 499                 if waiting != "" {
 500                         super.logger.WithField("targets", waiting[1:]).Info("waiting")
 501                 }
 502         }
 503         return true
 504 }
 505
 506 func (super *Supervisor) prependEnv(key, prepend string) {
 507         for i, s := range super.environ {
 508                 if strings.HasPrefix(s, key+"=") {
 509                         super.environ[i] = key + "=" + prepend + s[len(key)+1:]
 510                         return
 511                 }
 512         }
 513         super.environ = append(super.environ, key+"="+prepend)
 514 }
 515
 516 func (super *Supervisor) cleanEnv(prefixes []string) {
 517         var cleaned []string
 518         for _, s := range super.environ {
 519                 drop := false
 520                 for _, p := range prefixes {
 521                         if strings.HasPrefix(s, p) {
 522                                 drop = true
 523                                 break
 524                         }
 525                 }
 526                 if !drop {
 527                         cleaned = append(cleaned, s)
 528                 }
 529         }
 530         super.environ = cleaned
 531 }
 532
 533 func (super *Supervisor) setEnv(key, val string) {
 534         for i, s := range super.environ {
 535                 if strings.HasPrefix(s, key+"=") {
 536                         super.environ[i] = key + "=" + val
 537                         return
 538                 }
 539         }
 540         super.environ = append(super.environ, key+"="+val)
 541 }
 542
 543 // Remove all but the first occurrence of each env var.
 544 func dedupEnv(in []string) []string {
 545         saw := map[string]bool{}
 546         var out []string
 547         for _, kv := range in {
 548                 if split := strings.Index(kv, "="); split < 1 {
 549                         panic("invalid environment var: " + kv)
 550                 } else if saw[kv[:split]] {
 551                         continue
 552                 } else {
 553                         saw[kv[:split]] = true
 554                         out = append(out, kv)
 555                 }
 556         }
 557         return out
 558 }
 559
 560 func (super *Supervisor) installGoProgram(ctx context.Context, srcpath string) (string, error) {
 561         _, basename := filepath.Split(srcpath)
 562         binfile := filepath.Join(super.bindir, basename)
 563         if super.ClusterType == "production" {
 564                 return binfile, nil
 565         }
 566         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)
 567         return binfile, err
 568 }
 569
 570 func (super *Supervisor) usingRVM() bool {
 571         return os.Getenv("rvm_path") != ""
 572 }
 573
 574 func (super *Supervisor) setupRubyEnv() error {
 575         if !super.usingRVM() {
 576                 // (If rvm is in use, assume the caller has everything
 577                 // set up as desired)
 578                 super.cleanEnv([]string{
 579                         "GEM_HOME=",
 580                         "GEM_PATH=",
 581                 })
 582                 gem := "gem"
 583                 if _, err := os.Stat("/var/lib/arvados/bin/gem"); err == nil || super.ClusterType == "production" {
 584                         gem = "/var/lib/arvados/bin/gem"
 585                 }
 586                 cmd := exec.Command(gem, "env", "gempath")
 587                 if super.ClusterType == "production" {
 588                         cmd.Args = append([]string{"sudo", "-u", "www-data", "-E", "HOME=/var/www"}, cmd.Args...)
 589                         path, err := exec.LookPath("sudo")
 590                         if err != nil {
 591                                 return fmt.Errorf("LookPath(\"sudo\"): %w", err)
 592                         }
 593                         cmd.Path = path
 594                 }
 595                 cmd.Stderr = super.Stderr
 596                 cmd.Env = super.environ
 597                 buf, err := cmd.Output() // /var/lib/arvados/.gem/ruby/2.5.0/bin:...
 598                 if err != nil || len(buf) == 0 {
 599                         return fmt.Errorf("gem env gempath: %w", err)
 600                 }
 601                 gempath := string(bytes.Split(buf, []byte{':'})[0])
 602                 super.prependEnv("PATH", gempath+"/bin:")
 603                 super.setEnv("GEM_HOME", gempath)
 604                 super.setEnv("GEM_PATH", gempath)
 605         }
 606         // Passenger install doesn't work unless $HOME is ~user
 607         u, err := user.Current()
 608         if err != nil {
 609                 return err
 610         }
 611         super.setEnv("HOME", u.HomeDir)
 612         return nil
 613 }
 614
 615 func (super *Supervisor) lookPath(prog string) string {
 616         for _, val := range super.environ {
 617                 if strings.HasPrefix(val, "PATH=") {
 618                         for _, dir := range filepath.SplitList(val[5:]) {
 619                                 path := filepath.Join(dir, prog)
 620                                 if fi, err := os.Stat(path); err == nil && fi.Mode()&0111 != 0 {
 621                                         return path
 622                                 }
 623                         }
 624                 }
 625         }
 626         return prog
 627 }
 628
 629 type runOptions struct {
 630         output io.Writer // attach stdout
 631         env    []string  // add/replace environment variables
 632         user   string    // run as specified user
 633         stdin  io.Reader
 634 }
 635
 636 // RunProgram runs prog with args, using dir as working directory. If ctx is
 637 // cancelled while the child is running, RunProgram terminates the child, waits
 638 // for it to exit, then returns.
 639 //
 640 // Child's environment will have our env vars, plus any given in env.
 641 //
 642 // Child's stdout will be written to output if non-nil, otherwise the
 643 // boot command's stderr.
 644 func (super *Supervisor) RunProgram(ctx context.Context, dir string, opts runOptions, prog string, args ...string) error {
 645         cmdline := fmt.Sprintf("%s", append([]string{prog}, args...))
 646         super.logger.WithField("command", cmdline).WithField("dir", dir).Info("executing")
 647
 648         logprefix := prog
 649         {
 650                 innerargs := args
 651                 if logprefix == "sudo" {
 652                         for i := 0; i < len(args); i++ {
 653                                 if args[i] == "-u" {
 654                                         i++
 655                                 } else if args[i] == "-E" || strings.Contains(args[i], "=") {
 656                                 } else {
 657                                         logprefix = args[i]
 658                                         innerargs = args[i+1:]
 659                                         break
 660                                 }
 661                         }
 662                 }
 663                 logprefix = strings.TrimPrefix(logprefix, "/var/lib/arvados/bin/")
 664                 logprefix = strings.TrimPrefix(logprefix, super.tempdir+"/bin/")
 665                 if logprefix == "bundle" && len(innerargs) > 2 && innerargs[0] == "exec" {
 666                         _, dirbase := filepath.Split(dir)
 667                         logprefix = innerargs[1] + "@" + dirbase
 668                 } else if logprefix == "arvados-server" && len(args) > 1 {
 669                         logprefix = args[0]
 670                 }
 671                 if !strings.HasPrefix(dir, "/") {
 672                         logprefix = dir + ": " + logprefix
 673                 }
 674         }
 675
 676         cmd := exec.Command(super.lookPath(prog), args...)
 677         cmd.Stdin = opts.stdin
 678         stdout, err := cmd.StdoutPipe()
 679         if err != nil {
 680                 return err
 681         }
 682         stderr, err := cmd.StderrPipe()
 683         if err != nil {
 684                 return err
 685         }
 686         logwriter := &service.LogPrefixer{Writer: super.Stderr, Prefix: []byte("[" + logprefix + "] ")}
 687         var copiers sync.WaitGroup
 688         copiers.Add(1)
 689         go func() {
 690                 io.Copy(logwriter, stderr)
 691                 copiers.Done()
 692         }()
 693         copiers.Add(1)
 694         go func() {
 695                 if opts.output == nil {
 696                         io.Copy(logwriter, stdout)
 697                 } else {
 698                         io.Copy(opts.output, stdout)
 699                 }
 700                 copiers.Done()
 701         }()
 702
 703         if strings.HasPrefix(dir, "/") {
 704                 cmd.Dir = dir
 705         } else {
 706                 cmd.Dir = filepath.Join(super.SourcePath, dir)
 707         }
 708         env := append([]string(nil), opts.env...)
 709         env = append(env, super.environ...)
 710         cmd.Env = dedupEnv(env)
 711
 712         if opts.user != "" {
 713                 // Note: We use this approach instead of "sudo"
 714                 // because in certain circumstances (we are pid 1 in a
 715                 // docker container, and our passenger child process
 716                 // changes to pgid 1) the intermediate sudo process
 717                 // notices we have the same pgid as our child and
 718                 // refuses to propagate signals from us to our child,
 719                 // so we can't signal/shutdown our passenger/rails
 720                 // apps. "chpst" or "setuidgid" would work, but these
 721                 // few lines avoid depending on runit/daemontools.
 722                 u, err := user.Lookup(opts.user)
 723                 if err != nil {
 724                         return fmt.Errorf("user.Lookup(%q): %w", opts.user, err)
 725                 }
 726                 uid, _ := strconv.Atoi(u.Uid)
 727                 gid, _ := strconv.Atoi(u.Gid)
 728                 cmd.SysProcAttr = &syscall.SysProcAttr{
 729                         Credential: &syscall.Credential{
 730                                 Uid: uint32(uid),
 731                                 Gid: uint32(gid),
 732                         },
 733                 }
 734         }
 735
 736         exited := false
 737         defer func() { exited = true }()
 738         go func() {
 739                 <-ctx.Done()
 740                 log := ctxlog.FromContext(ctx).WithFields(logrus.Fields{"dir": dir, "cmdline": cmdline})
 741                 for !exited {
 742                         if cmd.Process == nil {
 743                                 log.Debug("waiting for child process to start")
 744                                 time.Sleep(time.Second / 2)
 745                         } else {
 746                                 log.WithField("PID", cmd.Process.Pid).Debug("sending SIGTERM")
 747                                 cmd.Process.Signal(syscall.SIGTERM)
 748                                 time.Sleep(5 * time.Second)
 749                                 if !exited {
 750                                         stdout.Close()
 751                                         stderr.Close()
 752                                         log.WithField("PID", cmd.Process.Pid).Warn("still waiting for child process to exit 5s after SIGTERM")
 753                                 }
 754                         }
 755                 }
 756         }()
 757
 758         err = cmd.Start()
 759         if err != nil {
 760                 return err
 761         }
 762         copiers.Wait()
 763         err = cmd.Wait()
 764         if ctx.Err() != nil {
 765                 // Return "context canceled", instead of the "killed"
 766                 // error that was probably caused by the context being
 767                 // canceled.
 768                 return ctx.Err()
 769         } else if err != nil {
 770                 return fmt.Errorf("%s: error: %v", cmdline, err)
 771         }
 772         return nil
 773 }
 774
 775 func (super *Supervisor) autofillConfig() error {
 776         usedPort := map[string]bool{}
 777         nextPort := func(host string) (string, error) {
 778                 for {
 779                         port, err := availablePort(host)
 780                         if err != nil {
 781                                 port, err = availablePort(super.ListenHost)
 782                         }
 783                         if err != nil {
 784                                 return "", err
 785                         }
 786                         if usedPort[port] {
 787                                 continue
 788                         }
 789                         usedPort[port] = true
 790                         return port, nil
 791                 }
 792         }
 793         if super.cluster.Services.Controller.ExternalURL.Host == "" {
 794                 h, p, err := net.SplitHostPort(super.ControllerAddr)
 795                 if err != nil && super.ControllerAddr != "" {
 796                         return fmt.Errorf("SplitHostPort(ControllerAddr %q): %w", super.ControllerAddr, err)
 797                 }
 798                 if h == "" {
 799                         h = super.ListenHost
 800                 }
 801                 if p == "0" || p == "" {
 802                         p, err = nextPort(h)
 803                         if err != nil {
 804                                 return err
 805                         }
 806                 }
 807                 super.cluster.Services.Controller.ExternalURL = arvados.URL{Scheme: "https", Host: net.JoinHostPort(h, p), Path: "/"}
 808         }
 809         u := url.URL(super.cluster.Services.Controller.ExternalURL)
 810         defaultExtHost := u.Hostname()
 811         for _, svc := range []*arvados.Service{
 812                 &super.cluster.Services.Controller,
 813                 &super.cluster.Services.DispatchCloud,
 814                 &super.cluster.Services.GitHTTP,
 815                 &super.cluster.Services.Health,
 816                 &super.cluster.Services.Keepproxy,
 817                 &super.cluster.Services.Keepstore,
 818                 &super.cluster.Services.RailsAPI,
 819                 &super.cluster.Services.WebDAV,
 820                 &super.cluster.Services.WebDAVDownload,
 821                 &super.cluster.Services.Websocket,
 822                 &super.cluster.Services.Workbench1,
 823                 &super.cluster.Services.Workbench2,
 824         } {
 825                 if svc.ExternalURL.Host == "" {
 826                         port, err := nextPort(defaultExtHost)
 827                         if err != nil {
 828                                 return err
 829                         }
 830                         host := net.JoinHostPort(defaultExtHost, port)
 831                         if svc == &super.cluster.Services.Controller ||
 832                                 svc == &super.cluster.Services.GitHTTP ||
 833                                 svc == &super.cluster.Services.Health ||
 834                                 svc == &super.cluster.Services.Keepproxy ||
 835                                 svc == &super.cluster.Services.WebDAV ||
 836                                 svc == &super.cluster.Services.WebDAVDownload ||
 837                                 svc == &super.cluster.Services.Workbench1 ||
 838                                 svc == &super.cluster.Services.Workbench2 {
 839                                 svc.ExternalURL = arvados.URL{Scheme: "https", Host: host, Path: "/"}
 840                         } else if svc == &super.cluster.Services.Websocket {
 841                                 svc.ExternalURL = arvados.URL{Scheme: "wss", Host: host, Path: "/websocket"}
 842                         }
 843                 }
 844                 if super.NoWorkbench1 && svc == &super.cluster.Services.Workbench1 ||
 845                         super.NoWorkbench2 && svc == &super.cluster.Services.Workbench2 ||
 846                         !super.cluster.Containers.CloudVMs.Enable && svc == &super.cluster.Services.DispatchCloud {
 847                         // When workbench1 is disabled, it gets an
 848                         // ExternalURL (so we have a valid listening
 849                         // port to write in our Nginx config) but no
 850                         // InternalURLs (so health checker doesn't
 851                         // complain).
 852                         continue
 853                 }
 854                 if len(svc.InternalURLs) == 0 {
 855                         port, err := nextPort(super.ListenHost)
 856                         if err != nil {
 857                                 return err
 858                         }
 859                         host := net.JoinHostPort(super.ListenHost, port)
 860                         svc.InternalURLs = map[arvados.URL]arvados.ServiceInstance{
 861                                 {Scheme: "http", Host: host, Path: "/"}: {},
 862                         }
 863                 }
 864         }
 865         if super.ClusterType != "production" {
 866                 if super.cluster.SystemRootToken == "" {
 867                         super.cluster.SystemRootToken = randomHexString(64)
 868                 }
 869                 if super.cluster.ManagementToken == "" {
 870                         super.cluster.ManagementToken = randomHexString(64)
 871                 }
 872                 if super.cluster.Collections.BlobSigningKey == "" {
 873                         super.cluster.Collections.BlobSigningKey = randomHexString(64)
 874                 }
 875                 if super.cluster.Users.AnonymousUserToken == "" {
 876                         super.cluster.Users.AnonymousUserToken = randomHexString(64)
 877                 }
 878                 if super.cluster.Containers.DispatchPrivateKey == "" {
 879                         buf, err := ioutil.ReadFile(filepath.Join(super.SourcePath, "lib", "dispatchcloud", "test", "sshkey_dispatch"))
 880                         if err != nil {
 881                                 return err
 882                         }
 883                         super.cluster.Containers.DispatchPrivateKey = string(buf)
 884                 }
 885                 super.cluster.TLS.Insecure = true
 886         }
 887         if super.ClusterType == "test" {
 888                 // Add a second keepstore process.
 889                 port, err := nextPort(super.ListenHost)
 890                 if err != nil {
 891                         return err
 892                 }
 893                 host := net.JoinHostPort(super.ListenHost, port)
 894                 super.cluster.Services.Keepstore.InternalURLs[arvados.URL{Scheme: "http", Host: host, Path: "/"}] = arvados.ServiceInstance{}
 895
 896                 // Create a directory-backed volume for each keepstore
 897                 // process.
 898                 super.cluster.Volumes = map[string]arvados.Volume{}
 899                 for url := range super.cluster.Services.Keepstore.InternalURLs {
 900                         volnum := len(super.cluster.Volumes)
 901                         datadir := fmt.Sprintf("%s/keep%d.data", super.tempdir, volnum)
 902                         if _, err = os.Stat(datadir + "/."); err == nil {
 903                         } else if !os.IsNotExist(err) {
 904                                 return err
 905                         } else if err = os.Mkdir(datadir, 0755); err != nil {
 906                                 return err
 907                         }
 908                         super.cluster.Volumes[fmt.Sprintf(super.cluster.ClusterID+"-nyw5e-%015d", volnum)] = arvados.Volume{
 909                                 Driver:           "Directory",
 910                                 DriverParameters: json.RawMessage(fmt.Sprintf(`{"Root":%q}`, datadir)),
 911                                 AccessViaHosts: map[arvados.URL]arvados.VolumeAccess{
 912                                         url: {},
 913                                 },
 914                                 StorageClasses: map[string]bool{
 915                                         "default": true,
 916                                         "foo":     true,
 917                                         "bar":     true,
 918                                 },
 919                         }
 920                 }
 921                 super.cluster.StorageClasses = map[string]arvados.StorageClassConfig{
 922                         "default": {Default: true},
 923                         "foo":     {},
 924                         "bar":     {},
 925                 }
 926         }
 927         if super.OwnTemporaryDatabase {
 928                 port, err := nextPort("localhost")
 929                 if err != nil {
 930                         return err
 931                 }
 932                 super.cluster.PostgreSQL.Connection = arvados.PostgreSQLConnection{
 933                         "client_encoding": "utf8",
 934                         "host":            "localhost",
 935                         "port":            port,
 936                         "dbname":          "arvados_test",
 937                         "user":            "arvados",
 938                         "password":        "insecure_arvados_test",
 939                 }
 940         }
 941         return nil
 942 }
 943
 944 func addrIsLocal(addr string) (bool, error) {
 945         if h, _, err := net.SplitHostPort(addr); err != nil {
 946                 return false, err
 947         } else {
 948                 addr = net.JoinHostPort(h, "0")
 949         }
 950         listener, err := net.Listen("tcp", addr)
 951         if err == nil {
 952                 listener.Close()
 953                 return true, nil
 954         } else if strings.Contains(err.Error(), "cannot assign requested address") {
 955                 return false, nil
 956         } else {
 957                 return false, err
 958         }
 959 }
 960
 961 func randomHexString(chars int) string {
 962         b := make([]byte, chars/2)
 963         _, err := rand.Read(b)
 964         if err != nil {
 965                 panic(err)
 966         }
 967         return fmt.Sprintf("%x", b)
 968 }
 969
 970 func internalPort(svc arvados.Service) (host, port string, err error) {
 971         if len(svc.InternalURLs) > 1 {
 972                 return "", "", errors.New("internalPort() doesn't work with multiple InternalURLs")
 973         }
 974         for u := range svc.InternalURLs {
 975                 u := url.URL(u)
 976                 host, port = u.Hostname(), u.Port()
 977                 switch {
 978                 case port != "":
 979                 case u.Scheme == "https", u.Scheme == "ws":
 980                         port = "443"
 981                 default:
 982                         port = "80"
 983                 }
 984                 return
 985         }
 986         return "", "", fmt.Errorf("service has no InternalURLs")
 987 }
 988
 989 func externalPort(svc arvados.Service) (string, error) {
 990         u := url.URL(svc.ExternalURL)
 991         if p := u.Port(); p != "" {
 992                 return p, nil
 993         } else if u.Scheme == "https" || u.Scheme == "wss" {
 994                 return "443", nil
 995         } else {
 996                 return "80", nil
 997         }
 998 }
 999
1000 func availablePort(host string) (string, error) {
1001         ln, err := net.Listen("tcp", net.JoinHostPort(host, "0"))
1002         if err != nil {
1003                 return "", err
1004         }
1005         defer ln.Close()
1006         _, port, err := net.SplitHostPort(ln.Addr().String())
1007         if err != nil {
1008                 return "", err
1009         }
1010         return port, nil
1011 }
1012
1013 // Try to connect to addr until it works, then close ch. Give up if
1014 // ctx cancels.
1015 func waitForConnect(ctx context.Context, addr string) error {
1016         ctxlog.FromContext(ctx).WithField("addr", addr).Info("waitForConnect")
1017         dialer := net.Dialer{Timeout: time.Second}
1018         for ctx.Err() == nil {
1019                 conn, err := dialer.DialContext(ctx, "tcp", addr)
1020                 if err != nil {
1021                         time.Sleep(time.Second / 10)
1022                         continue
1023                 }
1024                 conn.Close()
1025                 return nil
1026         }
1027         return ctx.Err()
1028 }
1029
1030 func copyConfig(cfg *arvados.Config) *arvados.Config {
1031         pr, pw := io.Pipe()
1032         go func() {
1033                 err := json.NewEncoder(pw).Encode(cfg)
1034                 if err != nil {
1035                         panic(err)
1036                 }
1037                 pw.Close()
1038         }()
1039         cfg2 := new(arvados.Config)
1040         err := json.NewDecoder(pr).Decode(cfg2)
1041         if err != nil {
1042                 panic(err)
1043         }
1044         return cfg2
1045 }
1046
1047 func watchConfig(ctx context.Context, logger logrus.FieldLogger, cfgPath string, prevcfg *arvados.Config, fn func()) {
1048         watcher, err := fsnotify.NewWatcher()
1049         if err != nil {
1050                 logger.WithError(err).Error("fsnotify setup failed")
1051                 return
1052         }
1053         defer watcher.Close()
1054
1055         err = watcher.Add(cfgPath)
1056         if err != nil {
1057                 logger.WithError(err).Error("fsnotify watcher failed")
1058                 return
1059         }
1060
1061         for {
1062                 select {
1063                 case <-ctx.Done():
1064                         return
1065                 case err, ok := <-watcher.Errors:
1066                         if !ok {
1067                                 return
1068                         }
1069                         logger.WithError(err).Warn("fsnotify watcher reported error")
1070                 case _, ok := <-watcher.Events:
1071                         if !ok {
1072                                 return
1073                         }
1074                         for len(watcher.Events) > 0 {
1075                                 <-watcher.Events
1076                         }
1077                         loader := config.NewLoader(&bytes.Buffer{}, &logrus.Logger{Out: ioutil.Discard})
1078                         loader.Path = cfgPath
1079                         loader.SkipAPICalls = true
1080                         cfg, err := loader.Load()
1081                         if err != nil {
1082                                 logger.WithError(err).Warn("error reloading config file after change detected; ignoring new config for now")
1083                         } else if reflect.DeepEqual(cfg, prevcfg) {
1084                                 logger.Debug("config file changed but is still DeepEqual to the existing config")
1085                         } else {
1086                                 logger.Debug("config changed, notifying supervisor")
1087                                 fn()
1088                                 prevcfg = cfg
1089                         }
1090                 }
1091         }
1092 }