18947: Refactor keep-web as arvados-server command.
[arvados.git] / lib / service / cmd.go
index d5619b87b307c53e781ea77b78ba943b20f78735..43357998d8b6c79cbb065f82e4ce92f71c080d54 100644 (file)
@@ -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 (
@@ -12,16 +12,19 @@ import (
        "io"
        "net"
        "net/http"
+       _ "net/http/pprof"
        "net/url"
        "os"
        "strings"
 
-       "git.curoverse.com/arvados.git/lib/cmd"
-       "git.curoverse.com/arvados.git/lib/config"
-       "git.curoverse.com/arvados.git/sdk/go/arvados"
-       "git.curoverse.com/arvados.git/sdk/go/ctxlog"
-       "git.curoverse.com/arvados.git/sdk/go/httpserver"
+       "git.arvados.org/arvados.git/lib/cmd"
+       "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
@@ -59,7 +65,7 @@ func (c *command) RunCommand(prog string, args []string, stdin io.Reader, stdout
        var err error
        defer func() {
                if err != nil {
-                       log.WithError(err).Info("exiting")
+                       log.WithError(err).Error("exiting")
                }
        }()
 
@@ -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
@@ -100,33 +116,17 @@ 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)
+       listenURL, err := getListenAddr(cluster.Services, c.svcName, log)
        if err != nil {
                return 1
        }
        ctx = context.WithValue(ctx, contextKeyURL{}, listenURL)
 
-       if cluster.SystemRootToken == "" {
-               logger.Warn("SystemRootToken missing from cluster config, falling back to ARVADOS_API_TOKEN environment variable")
-               cluster.SystemRootToken = os.Getenv("ARVADOS_API_TOKEN")
-       }
-       if cluster.Services.Controller.ExternalURL.Host == "" {
-               logger.Warn("Services.Controller.ExternalURL missing from cluster config, falling back to ARVADOS_API_HOST(_INSECURE) environment variables")
-               u, err := url.Parse("https://" + os.Getenv("ARVADOS_API_HOST"))
-               if err != nil {
-                       err = fmt.Errorf("ARVADOS_API_HOST: %s", err)
-                       return 1
-               }
-               cluster.Services.Controller.ExternalURL = arvados.URL(*u)
-               if i := os.Getenv("ARVADOS_API_HOST_INSECURE"); i != "" && i != "0" {
-                       cluster.TLS.Insecure = true
-               }
-       }
-
        reg := prometheus.NewRegistry()
        handler := c.newHandler(ctx, cluster, cluster.SystemRootToken, reg)
        if err = handler.CheckHealth(); err != nil {
@@ -134,13 +134,15 @@ func (c *command) RunCommand(prog string, args []string, stdin io.Reader, stdout
        }
 
        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,
        }
@@ -160,14 +162,21 @@ 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
@@ -175,24 +184,70 @@ 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) (arvados.URL, error) {
+func getListenAddr(svcs arvados.Services, prog arvados.ServiceName, log logrus.FieldLogger) (arvados.URL, error) {
        svc, ok := svcs.Map()[prog]
        if !ok {
                return arvados.URL{}, fmt.Errorf("unknown service name %q", prog)
        }
-       for url := range svc.InternalURLs {
-               if strings.HasPrefix(url.Host, "localhost:") {
-                       return url, nil
+
+       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)
+       } else {
+               if url.Path == "" {
+                       url.Path = "/"
                }
+               return arvados.URL(*url), nil
+       }
+
+       errors := []string{}
+       for url := range svc.InternalURLs {
                listener, err := net.Listen("tcp", url.Host)
                if err == nil {
                        listener.Close()
                        return url, 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
+                       // to IP address, and then fail because it
+                       // can't bind an IP address it doesn't own.
+                       continue
+               } else {
+                       errors = append(errors, fmt.Sprintf("tried %v, got %v", url, err))
                }
        }
-       return arvados.URL{}, fmt.Errorf("configuration does not enable the %s service on this host", prog)
+       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{}, fmt.Errorf("configuration does not enable the %q service on this host", prog)
 }
 
 type contextKeyURL struct{}