X-Git-Url: https://git.arvados.org/arvados.git/blobdiff_plain/07dbd859d0b05d79680570c00c43a722102d7ea3..68f6929a58bf9aab7df691a0db1567e27bc942b8:/lib/service/cmd.go diff --git a/lib/service/cmd.go b/lib/service/cmd.go index 7f2f78ee9a..92c554ce53 100644 --- a/lib/service/cmd.go +++ b/lib/service/cmd.go @@ -2,7 +2,7 @@ // // SPDX-License-Identifier: Apache-2.0 -// package service provides a cmd.Handler that brings up a system service. +// Package service provides a cmd.Handler that brings up a system service. package service import ( @@ -12,6 +12,7 @@ import ( "io" "net" "net/http" + _ "net/http/pprof" "net/url" "os" "strings" @@ -20,8 +21,10 @@ import ( "git.arvados.org/arvados.git/lib/config" "git.arvados.org/arvados.git/sdk/go/arvados" "git.arvados.org/arvados.git/sdk/go/ctxlog" + "git.arvados.org/arvados.git/sdk/go/health" "git.arvados.org/arvados.git/sdk/go/httpserver" "github.com/coreos/go-systemd/daemon" + "github.com/julienschmidt/httprouter" "github.com/prometheus/client_golang/prometheus" "github.com/sirupsen/logrus" ) @@ -29,6 +32,9 @@ import ( type Handler interface { http.Handler CheckHealth() error + // Done returns a channel that closes when the handler shuts + // itself down, or nil if this never happens. + Done() <-chan struct{} } type NewHandlerFunc func(_ context.Context, _ *arvados.Cluster, token string, registry *prometheus.Registry) Handler @@ -68,17 +74,27 @@ func (c *command) RunCommand(prog string, args []string, stdin io.Reader, stdout loader := config.NewLoader(stdin, log) loader.SetupFlags(flags) + + // prog is [keepstore, keep-web, git-httpd, ...] but the + // legacy config flags are [-legacy-keepstore-config, + // -legacy-keepweb-config, -legacy-git-httpd-config, ...] + legacyFlag := "-legacy-" + strings.Replace(prog, "keep-", "keep", 1) + "-config" + args = loader.MungeLegacyConfigArgs(log, args, legacyFlag) + versionFlag := flags.Bool("version", false, "Write version information to stdout and exit 0") - err = flags.Parse(args) - if err == flag.ErrHelp { - err = nil - return 0 - } else if err != nil { - return 2 + pprofAddr := flags.String("pprof", "", "Serve Go profile data at `[addr]:port`") + if ok, code := cmd.ParseFlags(flags, prog, args, "", stderr); !ok { + return code } else if *versionFlag { return cmd.Version.RunCommand(prog, args, stdin, stdout, stderr) } + if *pprofAddr != "" { + go func() { + log.Println(http.ListenAndServe(*pprofAddr, nil)) + }() + } + if strings.HasSuffix(prog, "controller") { // Some config-loader checks try to make API calls via // controller. Those can't be expected to work if this @@ -100,34 +116,48 @@ func (c *command) RunCommand(prog string, args []string, stdin io.Reader, stdout // logger with a new one according to the logging config. log = ctxlog.New(stderr, cluster.SystemLogs.Format, cluster.SystemLogs.LogLevel) logger := log.WithFields(logrus.Fields{ - "PID": os.Getpid(), + "PID": os.Getpid(), + "ClusterID": cluster.ClusterID, }) ctx := ctxlog.Context(c.ctx, logger) - listenURL, err := getListenAddr(cluster.Services, c.svcName, log) + listenURL, internalURL, err := getListenAddr(cluster.Services, c.svcName, log) if err != nil { return 1 } - ctx = context.WithValue(ctx, contextKeyURL{}, listenURL) + ctx = context.WithValue(ctx, contextKeyURL{}, internalURL) reg := prometheus.NewRegistry() + loader.RegisterMetrics(reg) + + // arvados_version_running{version="1.2.3~4"} 1.0 + mVersion := prometheus.NewGaugeVec(prometheus.GaugeOpts{ + Namespace: "arvados", + Name: "version_running", + Help: "Indicated version is running.", + }, []string{"version"}) + mVersion.WithLabelValues(cmd.Version.String()).Set(1) + reg.MustRegister(mVersion) + handler := c.newHandler(ctx, cluster, cluster.SystemRootToken, reg) if err = handler.CheckHealth(); err != nil { return 1 } instrumented := httpserver.Instrument(reg, log, - httpserver.HandlerWithContext(ctx, + httpserver.HandlerWithDeadline(cluster.API.RequestTimeout.Duration(), httpserver.AddRequestIDs( httpserver.LogRequests( - httpserver.NewRequestLimiter(cluster.API.MaxConcurrentRequests, handler, reg))))) + interceptHealthReqs(cluster.ManagementToken, handler.CheckHealth, + httpserver.NewRequestLimiter(cluster.API.MaxConcurrentRequests, handler, reg)))))) srv := &httpserver.Server{ Server: http.Server{ - Handler: instrumented.ServeAPI(cluster.ManagementToken, instrumented), + Handler: ifCollectionInHost(instrumented, instrumented.ServeAPI(cluster.ManagementToken, instrumented)), + BaseContext: func(net.Listener) context.Context { return ctx }, }, Addr: listenURL.Host, } - if listenURL.Scheme == "https" { + if listenURL.Scheme == "https" || listenURL.Scheme == "wss" { tlsconfig, err := tlsConfigWithCertUpdater(cluster, logger) if err != nil { logger.WithError(err).Errorf("cannot start %s service on %s", c.svcName, listenURL.String()) @@ -143,14 +173,21 @@ func (c *command) RunCommand(prog string, args []string, stdin io.Reader, stdout "URL": listenURL, "Listen": srv.Addr, "Service": c.svcName, + "Version": cmd.Version.String(), }).Info("listening") if _, err := daemon.SdNotify(false, "READY=1"); err != nil { logger.WithError(err).Errorf("error notifying init daemon") } go func() { + // Shut down server if caller cancels context <-ctx.Done() srv.Close() }() + go func() { + // Shut down server if handler dies + <-handler.Done() + srv.Close() + }() err = srv.Wait() if err != nil { return 1 @@ -158,27 +195,92 @@ func (c *command) RunCommand(prog string, args []string, stdin io.Reader, stdout return 0 } -const rfc3339NanoFixed = "2006-01-02T15:04:05.000000000Z07:00" +// If an incoming request's target vhost has an embedded collection +// UUID or PDH, handle it with hTrue, otherwise handle it with +// hFalse. +// +// Facilitates routing "http://collections.example/metrics" to metrics +// and "http://{uuid}.collections.example/metrics" to a file in a +// collection. +func ifCollectionInHost(hTrue, hFalse http.Handler) http.Handler { + return http.HandlerFunc(func(w http.ResponseWriter, r *http.Request) { + if arvados.CollectionIDFromDNSName(r.Host) != "" { + hTrue.ServeHTTP(w, r) + } else { + hFalse.ServeHTTP(w, r) + } + }) +} + +func interceptHealthReqs(mgtToken string, checkHealth func() error, next http.Handler) http.Handler { + mux := httprouter.New() + mux.Handler("GET", "/_health/ping", &health.Handler{ + Token: mgtToken, + Prefix: "/_health/", + Routes: health.Routes{"ping": checkHealth}, + }) + mux.NotFound = next + return ifCollectionInHost(next, mux) +} -func getListenAddr(svcs arvados.Services, prog arvados.ServiceName, log logrus.FieldLogger) (arvados.URL, error) { +func getListenAddr(svcs arvados.Services, prog arvados.ServiceName, log logrus.FieldLogger) (arvados.URL, arvados.URL, error) { svc, ok := svcs.Map()[prog] if !ok { - return arvados.URL{}, fmt.Errorf("unknown service name %q", prog) + return arvados.URL{}, arvados.URL{}, fmt.Errorf("unknown service name %q", prog) } if want := os.Getenv("ARVADOS_SERVICE_INTERNAL_URL"); want == "" { } else if url, err := url.Parse(want); err != nil { - return arvados.URL{}, fmt.Errorf("$ARVADOS_SERVICE_INTERNAL_URL (%q): %s", want, err) + return arvados.URL{}, arvados.URL{}, fmt.Errorf("$ARVADOS_SERVICE_INTERNAL_URL (%q): %s", want, err) } else { - return arvados.URL(*url), nil + if url.Path == "" { + url.Path = "/" + } + internalURL := arvados.URL(*url) + listenURL := arvados.URL(*url) + if svc.ListenAddress != "" { + listenURL.Host = svc.ListenAddress + } + return listenURL, internalURL, nil + } + + if svc.ListenAddress != "" { + scheme := "" + for internalURL := range svc.InternalURLs { + if internalURL.Host == svc.ListenAddress { + if len(svc.InternalURLs) > 1 { + log.Warnf("possible configuration error: multiple InternalURLs entries exist for %s but only %q will ever be used because it matches ListenAddress", prog, internalURL.String()) + } + return internalURL, internalURL, nil + } + switch scheme { + case "": + scheme = internalURL.Scheme + case internalURL.Scheme: + default: + scheme = "-" // different InternalURLs have different schemes + } + } + if scheme == "-" { + return arvados.URL{}, arvados.URL{}, fmt.Errorf("cannot use ListenAddress %q: InternalURLs use multiple schemes and none have host %q", svc.ListenAddress, svc.ListenAddress) + } + if scheme == "" { + // No entries at all in InternalURLs + scheme = "http" + } + listenURL := arvados.URL{} + listenURL.Host = svc.ListenAddress + listenURL.Scheme = scheme + listenURL.Path = "/" + return listenURL, listenURL, nil } errors := []string{} - for url := range svc.InternalURLs { - listener, err := net.Listen("tcp", url.Host) + for internalURL := range svc.InternalURLs { + listener, err := net.Listen("tcp", internalURL.Host) if err == nil { listener.Close() - return url, nil + return internalURL, internalURL, nil } else if strings.Contains(err.Error(), "cannot assign requested address") { // If 'Host' specifies a different server than // the current one, it'll resolve the hostname @@ -186,13 +288,13 @@ func getListenAddr(svcs arvados.Services, prog arvados.ServiceName, log logrus.F // can't bind an IP address it doesn't own. continue } else { - errors = append(errors, fmt.Sprintf("tried %v, got %v", url, err)) + errors = append(errors, fmt.Sprintf("tried %v, got %v", internalURL, err)) } } if len(errors) > 0 { - return arvados.URL{}, fmt.Errorf("could not enable the %q service on this host: %s", prog, strings.Join(errors, "; ")) + return arvados.URL{}, arvados.URL{}, fmt.Errorf("could not enable the %q service on this host: %s", prog, strings.Join(errors, "; ")) } - return arvados.URL{}, fmt.Errorf("configuration does not enable the %q service on this host", prog) + return arvados.URL{}, arvados.URL{}, fmt.Errorf("configuration does not enable the %q service on this host", prog) } type contextKeyURL struct{}