X-Git-Url: https://git.arvados.org/arvados.git/blobdiff_plain/c52c3788fbbf161db40118261c4cfff52ebf8ceb..7563e6276baade41eb3faa3ff167abab1cb0f890:/services/ws/router.go

diff --git a/services/ws/router.go b/services/ws/router.go
index 6aef64761b..878c282f8a 100644
--- a/services/ws/router.go
+++ b/services/ws/router.go
@@ -1,15 +1,21 @@
-package main
+// Copyright (C) The Arvados Authors. All rights reserved.
+//
+// SPDX-License-Identifier: AGPL-3.0
+
+package ws
 
 import (
-	"database/sql"
 	"io"
 	"net/http"
-	"strconv"
 	"sync"
+	"sync/atomic"
 	"time"
 
-	"git.curoverse.com/arvados.git/sdk/go/arvados"
-	"github.com/Sirupsen/logrus"
+	"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"
 )
 
@@ -21,74 +27,86 @@ type wsConn interface {
 }
 
 type router struct {
-	Config *Config
-
-	eventSource eventSource
-	mux         *http.ServeMux
-	setupOnce   sync.Once
+	client         *arvados.Client
+	cluster        *arvados.Cluster
+	eventSource    eventSource
+	newPermChecker func() permChecker
 
-	lastReqID  int64
-	lastReqMtx sync.Mutex
+	handler   *handler
+	mux       *http.ServeMux
+	setupOnce sync.Once
+	done      chan struct{}
+	reg       *prometheus.Registry
 }
 
-type sessionFactory func(wsConn, chan<- interface{}, arvados.Client, *sql.DB) (session, error)
-
 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("/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/",
+		Routes: health.Routes{
+			"db": rtr.eventSource.DBHealth,
+		},
+		Log: func(r *http.Request, err error) {
+			if err != nil {
+				ctxlog.FromContext(r.Context()).WithError(err).Error("error")
+			}
+		},
+	})
 }
 
-func (rtr *router) makeServer(newSession sessionFactory) *websocket.Server {
-	handler := &handler{
-		PingTimeout: rtr.Config.PingTimeout.Duration(),
-		QueueSize:   rtr.Config.ClientEventQueue,
-		NewSession: func(ws wsConn, sendq chan<- interface{}) (session, error) {
-			return newSession(ws, sendq, rtr.Config.Client, rtr.eventSource.DB())
-		},
-	}
+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
 		},
 		Handler: websocket.Handler(func(ws *websocket.Conn) {
 			t0 := time.Now()
-			sink := rtr.eventSource.NewSink()
-			log := logger(ws.Request().Context())
-			log.Info("connected")
+			logger := ctxlog.FromContext(ws.Request().Context())
+			atomic.AddInt64(&connected, 1)
+			gauge.Set(float64(atomic.LoadInt64(&connected)))
 
-			stats := handler.Handle(ws, sink.Channel())
+			stats := rtr.handler.Handle(ws, logger, rtr.eventSource,
+				func(ws wsConn, sendq chan<- interface{}) (session, error) {
+					return newSession(ws, sendq, rtr.eventSource.DB(), rtr.newPermChecker(), rtr.client)
+				})
 
-			log.WithFields(logrus.Fields{
-				"Elapsed": time.Now().Sub(t0).Seconds(),
-				"Stats":   stats,
-			}).Info("disconnect")
-
-			sink.Stop()
+			logger.WithFields(logrus.Fields{
+				"elapsed": time.Now().Sub(t0).Seconds(),
+				"stats":   stats,
+			}).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) ServeHTTP(resp http.ResponseWriter, req *http.Request) {
 	rtr.setupOnce.Do(rtr.setup)
-	logger := logger(req.Context()).
-		WithField("RequestID", rtr.newReqID())
-	ctx := contextWithLogger(req.Context(), logger)
-	req = req.WithContext(ctx)
-	logger.WithFields(logrus.Fields{
-		"RemoteAddr":      req.RemoteAddr,
-		"X-Forwarded-For": req.Header.Get("X-Forwarded-For"),
-	}).Info("accept request")
 	rtr.mux.ServeHTTP(resp, req)
 }
+
+func (rtr *router) CheckHealth() error {
+	rtr.setupOnce.Do(rtr.setup)
+	return rtr.eventSource.DBHealth()
+}
+
+func (rtr *router) Done() <-chan struct{} {
+	return rtr.done
+}