15954: Add comments.
[arvados.git] / lib / boot / cmd.go
index 6762446b1bcb3b3be9aa76230043663a622b5b64..992c7c6e510f669fc5d884c6fd3a1af962dae9a8 100644 (file)
@@ -26,6 +26,7 @@ import (
        "git.arvados.org/arvados.git/lib/cmd"
        "git.arvados.org/arvados.git/lib/config"
        "git.arvados.org/arvados.git/lib/controller"
+       "git.arvados.org/arvados.git/lib/dispatchcloud"
        "git.arvados.org/arvados.git/sdk/go/arvados"
        "git.arvados.org/arvados.git/sdk/go/ctxlog"
        "github.com/sirupsen/logrus"
@@ -38,8 +39,11 @@ type bootCommand struct {
        libPath     string // e.g., /var/lib/arvados
        clusterType string // e.g., production
 
-       stdout io.Writer
-       stderr io.Writer
+       cluster *arvados.Cluster
+       stdout  io.Writer
+       stderr  io.Writer
+
+       tempdir string
 
        setupRubyOnce sync.Once
        setupRubyErr  error
@@ -97,16 +101,19 @@ func (boot *bootCommand) RunCommand(prog string, args []string, stdin io.Reader,
                return 1
        }
 
-       tempdir, err := ioutil.TempDir("", "arvados-server-boot-")
+       boot.tempdir, err = ioutil.TempDir("", "arvados-server-boot-")
        if err != nil {
                return 1
        }
-       defer os.RemoveAll(tempdir)
+       defer os.RemoveAll(boot.tempdir)
 
        // Fill in any missing config keys, and write the resulting
        // config in the temp dir for child services to use.
-       autofillConfig(cfg, log)
-       conffile, err := os.OpenFile(filepath.Join(tempdir, "config.yml"), os.O_CREATE|os.O_WRONLY, 0777)
+       err = boot.autofillConfig(cfg, log)
+       if err != nil {
+               return 1
+       }
+       conffile, err := os.OpenFile(filepath.Join(boot.tempdir, "config.yml"), os.O_CREATE|os.O_WRONLY, 0777)
        if err != nil {
                return 1
        }
@@ -120,16 +127,16 @@ func (boot *bootCommand) RunCommand(prog string, args []string, stdin io.Reader,
                return 1
        }
        os.Setenv("ARVADOS_CONFIG", conffile.Name())
-
+       arvados.DefaultConfigFile = conffile.Name()
        os.Setenv("RAILS_ENV", boot.clusterType)
 
        // Now that we have the config, replace the bootstrap logger
        // with a new one according to the logging config.
-       cluster, err := cfg.GetCluster("")
+       boot.cluster, err = cfg.GetCluster("")
        if err != nil {
                return 1
        }
-       log = ctxlog.New(stderr, cluster.SystemLogs.Format, cluster.SystemLogs.LogLevel)
+       log = ctxlog.New(stderr, boot.cluster.SystemLogs.Format, boot.cluster.SystemLogs.LogLevel)
        logger := log.WithFields(logrus.Fields{
                "PID": os.Getpid(),
        })
@@ -165,9 +172,17 @@ func (boot *bootCommand) RunCommand(prog string, args []string, stdin io.Reader,
 
        var wg sync.WaitGroup
        for _, cmpt := range []component{
-               {name: "controller", svc: cluster.Services.Controller, cmdArgs: []string{"-config", conffile.Name()}, cmdHandler: controller.Command},
-               // {name: "dispatchcloud", cmdArgs: []string{"-config", conffile.Name()}, cmdHandler: dispatchcloud.Command},
-               {name: "railsAPI", svc: cluster.Services.RailsAPI, src: "services/api"},
+               {name: "nginx", runFunc: runNginx},
+               {name: "controller", cmdHandler: controller.Command},
+               {name: "dispatchcloud", cmdHandler: dispatchcloud.Command, notIfTest: true},
+               {name: "git-httpd", goProg: "services/arv-git-httpd"},
+               {name: "health", goProg: "services/health"},
+               {name: "keep-balance", goProg: "services/keep-balance", notIfTest: true},
+               {name: "keepproxy", goProg: "services/keepproxy"},
+               {name: "keepstore", goProg: "services/keepstore", svc: boot.cluster.Services.Keepstore},
+               {name: "keep-web", goProg: "services/keep-web"},
+               {name: "railsAPI", svc: boot.cluster.Services.RailsAPI, railsApp: "services/api"},
+               {name: "ws", goProg: "services/ws"},
        } {
                cmpt := cmpt
                wg.Add(1)
@@ -189,8 +204,7 @@ func (boot *bootCommand) RunCommand(prog string, args []string, stdin io.Reader,
 func (boot *bootCommand) installGoProgram(ctx context.Context, srcpath string) error {
        boot.goMutex.Lock()
        defer boot.goMutex.Unlock()
-       env := append([]string{"GOPATH=" + boot.libPath}, os.Environ()...)
-       return boot.RunProgram(ctx, filepath.Join(boot.sourcePath, srcpath), nil, env, "go", "install")
+       return boot.RunProgram(ctx, filepath.Join(boot.sourcePath, srcpath), nil, []string{"GOPATH=" + boot.libPath}, "go", "install")
 }
 
 func (boot *bootCommand) setupRubyEnv() error {
@@ -207,6 +221,14 @@ func (boot *bootCommand) setupRubyEnv() error {
        return boot.setupRubyErr
 }
 
+// Run prog with args, using dir as working directory. If ctx is
+// cancelled while the child is running, RunProgram terminates the
+// child, waits for it to exit, then returns.
+//
+// Child's environment will have our env vars, plus any given in env.
+//
+// Child's stdout will be written to output if non-nil, otherwise the
+// boot command's stderr.
 func (boot *bootCommand) RunProgram(ctx context.Context, dir string, output io.Writer, env []string, prog string, args ...string) error {
        cmdline := fmt.Sprintf("%s", append([]string{prog}, args...))
        fmt.Fprintf(boot.stderr, "%s executing in %s\n", cmdline, dir)
@@ -223,17 +245,22 @@ func (boot *bootCommand) RunProgram(ctx context.Context, dir string, output io.W
                cmd.Dir = filepath.Join(boot.sourcePath, dir)
        }
        if env != nil {
-               cmd.Env = env
+               cmd.Env = append(env, os.Environ()...)
        }
        go func() {
                <-ctx.Done()
-               cmd.Process.Signal(syscall.SIGINT)
-               for range time.Tick(5 * time.Second) {
-                       if cmd.ProcessState != nil {
-                               break
+               log := ctxlog.FromContext(ctx).WithFields(logrus.Fields{"dir": dir, "cmdline": cmdline})
+               for cmd.ProcessState == nil {
+                       // Child hasn't exited yet
+                       if cmd.Process == nil {
+                               log.Infof("waiting for child process to start")
+                               time.Sleep(time.Second)
+                       } else {
+                               log.WithField("PID", cmd.Process.Pid).Info("sending SIGTERM")
+                               cmd.Process.Signal(syscall.SIGTERM)
+                               log.WithField("PID", cmd.Process.Pid).Info("waiting for child process to exit after SIGTERM")
+                               time.Sleep(5 * time.Second)
                        }
-                       ctxlog.FromContext(ctx).WithField("process", cmd.Process).Infof("waiting for child process to exit after SIGINT")
-                       cmd.Process.Signal(syscall.SIGINT)
                }
        }()
        err := cmd.Run()
@@ -247,17 +274,24 @@ type component struct {
        name       string
        svc        arvados.Service
        cmdHandler cmd.Handler
-       cmdArgs    []string
-       src        string // source dir in arvados tree, e.g., "services/keepstore"
+       runFunc    func(ctx context.Context, boot *bootCommand, stdout, stderr io.Writer) error
+       railsApp   string // source dir in arvados tree, e.g., "services/api"
+       goProg     string // source dir in arvados tree, e.g., "services/keepstore"
+       notIfTest  bool   // don't run this component on a test cluster
 }
 
 func (cmpt *component) Run(ctx context.Context, boot *bootCommand, stdout, stderr io.Writer) error {
+       if cmpt.notIfTest && boot.clusterType == "test" {
+               fmt.Fprintf(stderr, "skipping component %q in %s mode\n", cmpt.name, boot.clusterType)
+               <-ctx.Done()
+               return nil
+       }
        fmt.Fprintf(stderr, "starting component %q\n", cmpt.name)
        if cmpt.cmdHandler != nil {
                errs := make(chan error, 1)
                go func() {
                        defer close(errs)
-                       exitcode := cmpt.cmdHandler.RunCommand(cmpt.name, cmpt.cmdArgs, bytes.NewBuffer(nil), stdout, stderr)
+                       exitcode := cmpt.cmdHandler.RunCommand(cmpt.name, nil, bytes.NewBuffer(nil), stdout, stderr)
                        if exitcode != 0 {
                                errs <- fmt.Errorf("exit code %d", exitcode)
                        }
@@ -272,59 +306,73 @@ func (cmpt *component) Run(ctx context.Context, boot *bootCommand, stdout, stder
                        return nil
                }
        }
-       if cmpt.src != "" {
-               port := "-"
-               for u := range cmpt.svc.InternalURLs {
-                       if _, p, err := net.SplitHostPort(u.Host); err != nil {
-                               return err
-                       } else if p != "" {
-                               port = p
-                       } else if u.Scheme == "https" {
-                               port = "443"
-                       } else {
-                               port = "80"
+       if cmpt.goProg != "" {
+               boot.RunProgram(ctx, cmpt.goProg, nil, nil, "go", "install")
+               if ctx.Err() != nil {
+                       return nil
+               }
+               _, basename := filepath.Split(cmpt.goProg)
+               if len(cmpt.svc.InternalURLs) > 0 {
+                       // Run one for each URL
+                       var wg sync.WaitGroup
+                       for u := range cmpt.svc.InternalURLs {
+                               u := u
+                               wg.Add(1)
+                               go func() {
+                                       defer wg.Done()
+                                       boot.RunProgram(ctx, boot.tempdir, nil, []string{"ARVADOS_SERVICE_INTERNAL_URL=" + u.String()}, basename)
+                               }()
                        }
-                       break
+                       wg.Wait()
+                       return nil
+               } else {
+                       // Just run one
+                       boot.RunProgram(ctx, boot.tempdir, nil, nil, basename)
                }
-               if port == "-" {
+       }
+       if cmpt.runFunc != nil {
+               return cmpt.runFunc(ctx, boot, stdout, stderr)
+       }
+       if cmpt.railsApp != "" {
+               port, err := internalPort(cmpt.svc)
+               if err != nil {
                        return fmt.Errorf("bug: no InternalURLs for component %q: %v", cmpt.name, cmpt.svc.InternalURLs)
                }
-
-               err := boot.setupRubyEnv()
+               err = boot.setupRubyEnv()
                if err != nil {
                        return err
                }
                var buf bytes.Buffer
-               err = boot.RunProgram(ctx, cmpt.src, &buf, nil, "gem", "list", "--details", "bundler")
+               err = boot.RunProgram(ctx, cmpt.railsApp, &buf, nil, "gem", "list", "--details", "bundler")
                if err != nil {
                        return err
                }
                for _, version := range []string{"1.11.0", "1.17.3", "2.0.2"} {
                        if !strings.Contains(buf.String(), "("+version+")") {
-                               err = boot.RunProgram(ctx, cmpt.src, nil, nil, "gem", "install", "--user", "bundler:1.11", "bundler:1.17.3", "bundler:2.0.2")
+                               err = boot.RunProgram(ctx, cmpt.railsApp, nil, nil, "gem", "install", "--user", "bundler:1.11", "bundler:1.17.3", "bundler:2.0.2")
                                if err != nil {
                                        return err
                                }
                                break
                        }
                }
-               err = boot.RunProgram(ctx, cmpt.src, nil, nil, "bundle", "install", "--jobs", "4", "--path", filepath.Join(os.Getenv("HOME"), ".gem"))
+               err = boot.RunProgram(ctx, cmpt.railsApp, nil, nil, "bundle", "install", "--jobs", "4", "--path", filepath.Join(os.Getenv("HOME"), ".gem"))
                if err != nil {
                        return err
                }
-               err = boot.RunProgram(ctx, cmpt.src, nil, nil, "bundle", "exec", "passenger-config", "build-native-support")
+               err = boot.RunProgram(ctx, cmpt.railsApp, nil, nil, "bundle", "exec", "passenger-config", "build-native-support")
                if err != nil {
                        return err
                }
-               err = boot.RunProgram(ctx, cmpt.src, nil, nil, "bundle", "exec", "passenger-config", "install-standalone-runtime")
+               err = boot.RunProgram(ctx, cmpt.railsApp, nil, nil, "bundle", "exec", "passenger-config", "install-standalone-runtime")
                if err != nil {
                        return err
                }
-               err = boot.RunProgram(ctx, cmpt.src, nil, nil, "bundle", "exec", "passenger-config", "validate-install")
+               err = boot.RunProgram(ctx, cmpt.railsApp, nil, nil, "bundle", "exec", "passenger-config", "validate-install")
                if err != nil {
                        return err
                }
-               err = boot.RunProgram(ctx, cmpt.src, nil, nil, "bundle", "exec", "passenger", "start", "-p", port)
+               err = boot.RunProgram(ctx, cmpt.railsApp, nil, nil, "bundle", "exec", "passenger", "start", "-p", port)
                if err != nil {
                        return err
                }
@@ -332,39 +380,93 @@ func (cmpt *component) Run(ctx context.Context, boot *bootCommand, stdout, stder
        return fmt.Errorf("bug: component %q has nothing to run", cmpt.name)
 }
 
-func autofillConfig(cfg *arvados.Config, log logrus.FieldLogger) {
+func (boot *bootCommand) autofillConfig(cfg *arvados.Config, log logrus.FieldLogger) error {
        cluster, err := cfg.GetCluster("")
        if err != nil {
-               panic(err)
+               return err
        }
        port := 9000
        for _, svc := range []*arvados.Service{
                &cluster.Services.Controller,
                &cluster.Services.DispatchCloud,
+               &cluster.Services.GitHTTP,
+               &cluster.Services.Health,
+               &cluster.Services.Keepproxy,
+               &cluster.Services.Keepstore,
                &cluster.Services.RailsAPI,
+               &cluster.Services.WebDAV,
+               &cluster.Services.WebDAVDownload,
+               &cluster.Services.Websocket,
        } {
+               if svc == &cluster.Services.DispatchCloud && boot.clusterType == "test" {
+                       continue
+               }
                if len(svc.InternalURLs) == 0 {
                        port++
                        svc.InternalURLs = map[arvados.URL]arvados.ServiceInstance{
                                arvados.URL{Scheme: "http", Host: fmt.Sprintf("localhost:%d", port)}: arvados.ServiceInstance{},
                        }
                }
-       }
-       if cluster.Services.Controller.ExternalURL.Host == "" {
-               for k := range cluster.Services.Controller.InternalURLs {
-                       cluster.Services.Controller.ExternalURL = k
+               if svc.ExternalURL.Host == "" && (svc == &cluster.Services.Controller ||
+                       svc == &cluster.Services.GitHTTP ||
+                       svc == &cluster.Services.Keepproxy ||
+                       svc == &cluster.Services.WebDAV ||
+                       svc == &cluster.Services.WebDAVDownload ||
+                       svc == &cluster.Services.Websocket) {
+                       port++
+                       svc.ExternalURL = arvados.URL{Scheme: "https", Host: fmt.Sprintf("localhost:%d", port)}
                }
        }
        if cluster.SystemRootToken == "" {
                cluster.SystemRootToken = randomHexString(64)
        }
+       if cluster.ManagementToken == "" {
+               cluster.ManagementToken = randomHexString(64)
+       }
        if cluster.API.RailsSessionSecretToken == "" {
                cluster.API.RailsSessionSecretToken = randomHexString(64)
        }
        if cluster.Collections.BlobSigningKey == "" {
                cluster.Collections.BlobSigningKey = randomHexString(64)
        }
+       if boot.clusterType != "production" && cluster.Containers.DispatchPrivateKey == "" {
+               buf, err := ioutil.ReadFile(filepath.Join(boot.sourcePath, "lib", "dispatchcloud", "test", "sshkey_dispatch"))
+               if err != nil {
+                       return err
+               }
+               cluster.Containers.DispatchPrivateKey = string(buf)
+       }
+       if boot.clusterType != "production" {
+               cluster.TLS.Insecure = true
+       }
+       if boot.clusterType == "test" {
+               // Add a second keepstore process.
+               port++
+               cluster.Services.Keepstore.InternalURLs[arvados.URL{Scheme: "http", Host: fmt.Sprintf("localhost:%d", port)}] = arvados.ServiceInstance{}
+
+               // Create a directory-backed volume for each keepstore
+               // process.
+               cluster.Volumes = map[string]arvados.Volume{}
+               for url := range cluster.Services.Keepstore.InternalURLs {
+                       volnum := len(cluster.Volumes)
+                       datadir := fmt.Sprintf("%s/keep%d.data", boot.tempdir, volnum)
+                       if _, err = os.Stat(datadir + "/."); err == nil {
+                       } else if !os.IsNotExist(err) {
+                               return err
+                       } else if err = os.Mkdir(datadir, 0777); err != nil {
+                               return err
+                       }
+                       cluster.Volumes[fmt.Sprintf("zzzzz-nyw5e-%015d", volnum)] = arvados.Volume{
+                               Driver:           "Directory",
+                               DriverParameters: json.RawMessage(fmt.Sprintf(`{"Root":%q}`, datadir)),
+                               AccessViaHosts: map[arvados.URL]arvados.VolumeAccess{
+                                       url: {},
+                               },
+                       }
+               }
+       }
        cfg.Clusters[cluster.ClusterID] = *cluster
+       return nil
 }
 
 func randomHexString(chars int) string {
@@ -375,3 +477,30 @@ func randomHexString(chars int) string {
        }
        return fmt.Sprintf("%x", b)
 }
+
+func internalPort(svc arvados.Service) (string, error) {
+       for u := range svc.InternalURLs {
+               if _, p, err := net.SplitHostPort(u.Host); err != nil {
+                       return "", err
+               } else if p != "" {
+                       return p, nil
+               } else if u.Scheme == "https" {
+                       return "443", nil
+               } else {
+                       return "80", nil
+               }
+       }
+       return "", fmt.Errorf("service has no InternalURLs")
+}
+
+func externalPort(svc arvados.Service) (string, error) {
+       if _, p, err := net.SplitHostPort(svc.ExternalURL.Host); err != nil {
+               return "", err
+       } else if p != "" {
+               return p, nil
+       } else if svc.ExternalURL.Scheme == "https" {
+               return "443", nil
+       } else {
+               return "80", nil
+       }
+}