18947: Refactor keep-balance as arvados-server subcommand.
[arvados.git] / services / ws / router.go
index b3403dabd00a3740d87f3136a21076ac4640fdc1..878c282f8a6c57f17192b777faba760485757b86 100644 (file)
@@ -5,18 +5,16 @@
 package ws
 
 import (
-       "encoding/json"
        "io"
        "net/http"
-       "strconv"
        "sync"
        "sync/atomic"
        "time"
 
-       "git.arvados.org/arvados.git/lib/cmd"
        "git.arvados.org/arvados.git/sdk/go/arvados"
        "git.arvados.org/arvados.git/sdk/go/ctxlog"
        "git.arvados.org/arvados.git/sdk/go/health"
+       "github.com/prometheus/client_golang/prometheus"
        "github.com/sirupsen/logrus"
        "golang.org/x/net/websocket"
 )
@@ -38,33 +36,25 @@ type router struct {
        mux       *http.ServeMux
        setupOnce sync.Once
        done      chan struct{}
-
-       lastReqID  int64
-       lastReqMtx sync.Mutex
-
-       status routerDebugStatus
-}
-
-type routerDebugStatus struct {
-       ReqsReceived int64
-       ReqsActive   int64
-}
-
-type debugStatuser interface {
-       DebugStatus() interface{}
+       reg       *prometheus.Registry
 }
 
 func (rtr *router) setup() {
+       mSockets := prometheus.NewGaugeVec(prometheus.GaugeOpts{
+               Namespace: "arvados",
+               Subsystem: "ws",
+               Name:      "sockets",
+               Help:      "Number of connected sockets",
+       }, []string{"version"})
+       rtr.reg.MustRegister(mSockets)
+
        rtr.handler = &handler{
                PingTimeout: time.Duration(rtr.cluster.API.SendTimeout),
                QueueSize:   rtr.cluster.API.WebsocketClientEventQueue,
        }
        rtr.mux = http.NewServeMux()
-       rtr.mux.Handle("/websocket", rtr.makeServer(newSessionV0))
-       rtr.mux.Handle("/arvados/v1/events.ws", rtr.makeServer(newSessionV1))
-       rtr.mux.Handle("/debug.json", rtr.jsonHandler(rtr.DebugStatus))
-       rtr.mux.Handle("/status.json", rtr.jsonHandler(rtr.Status))
-
+       rtr.mux.Handle("/websocket", rtr.makeServer(newSessionV0, mSockets.WithLabelValues("0")))
+       rtr.mux.Handle("/arvados/v1/events.ws", rtr.makeServer(newSessionV1, mSockets.WithLabelValues("1")))
        rtr.mux.Handle("/_health/", &health.Handler{
                Token:  rtr.cluster.ManagementToken,
                Prefix: "/_health/",
@@ -79,7 +69,8 @@ func (rtr *router) setup() {
        })
 }
 
-func (rtr *router) makeServer(newSession sessionFactory) *websocket.Server {
+func (rtr *router) makeServer(newSession sessionFactory, gauge prometheus.Gauge) *websocket.Server {
+       var connected int64
        return &websocket.Server{
                Handshake: func(c *websocket.Config, r *http.Request) error {
                        return nil
@@ -87,7 +78,8 @@ func (rtr *router) makeServer(newSession sessionFactory) *websocket.Server {
                Handler: websocket.Handler(func(ws *websocket.Conn) {
                        t0 := time.Now()
                        logger := ctxlog.FromContext(ws.Request().Context())
-                       logger.Info("connected")
+                       atomic.AddInt64(&connected, 1)
+                       gauge.Set(float64(atomic.LoadInt64(&connected)))
 
                        stats := rtr.handler.Handle(ws, logger, rtr.eventSource,
                                func(ws wsConn, sendq chan<- interface{}) (session, error) {
@@ -97,71 +89,19 @@ func (rtr *router) makeServer(newSession sessionFactory) *websocket.Server {
                        logger.WithFields(logrus.Fields{
                                "elapsed": time.Now().Sub(t0).Seconds(),
                                "stats":   stats,
-                       }).Info("disconnect")
+                       }).Info("client disconnected")
                        ws.Close()
+                       atomic.AddInt64(&connected, -1)
+                       gauge.Set(float64(atomic.LoadInt64(&connected)))
                }),
        }
 }
 
-func (rtr *router) newReqID() string {
-       rtr.lastReqMtx.Lock()
-       defer rtr.lastReqMtx.Unlock()
-       id := time.Now().UnixNano()
-       if id <= rtr.lastReqID {
-               id = rtr.lastReqID + 1
-       }
-       return strconv.FormatInt(id, 36)
-}
-
-func (rtr *router) DebugStatus() interface{} {
-       s := map[string]interface{}{
-               "HTTP":     rtr.status,
-               "Outgoing": rtr.handler.DebugStatus(),
-       }
-       if es, ok := rtr.eventSource.(debugStatuser); ok {
-               s["EventSource"] = es.DebugStatus()
-       }
-       return s
-}
-
-func (rtr *router) Status() interface{} {
-       return map[string]interface{}{
-               "Clients": atomic.LoadInt64(&rtr.status.ReqsActive),
-               "Version": cmd.Version.String(),
-       }
-}
-
 func (rtr *router) ServeHTTP(resp http.ResponseWriter, req *http.Request) {
        rtr.setupOnce.Do(rtr.setup)
-       atomic.AddInt64(&rtr.status.ReqsReceived, 1)
-       atomic.AddInt64(&rtr.status.ReqsActive, 1)
-       defer atomic.AddInt64(&rtr.status.ReqsActive, -1)
-
-       logger := ctxlog.FromContext(req.Context()).
-               WithField("RequestID", rtr.newReqID())
-       ctx := ctxlog.Context(req.Context(), logger)
-       req = req.WithContext(ctx)
-       logger.WithFields(logrus.Fields{
-               "remoteAddr":      req.RemoteAddr,
-               "reqForwardedFor": req.Header.Get("X-Forwarded-For"),
-       }).Info("accept request")
        rtr.mux.ServeHTTP(resp, req)
 }
 
-func (rtr *router) jsonHandler(fn func() interface{}) http.Handler {
-       return http.HandlerFunc(func(w http.ResponseWriter, r *http.Request) {
-               logger := ctxlog.FromContext(r.Context())
-               w.Header().Set("Content-Type", "application/json")
-               enc := json.NewEncoder(w)
-               err := enc.Encode(fn())
-               if err != nil {
-                       msg := "encode failed"
-                       logger.WithError(err).Error(msg)
-                       http.Error(w, msg, http.StatusInternalServerError)
-               }
-       })
-}
-
 func (rtr *router) CheckHealth() error {
        rtr.setupOnce.Do(rtr.setup)
        return rtr.eventSource.DBHealth()