Graceful shutdowns (#1734)

* Initial graceful stop

* Fix dendritejs

* Use process context for outbound federation requests in destination queues

* Reduce logging

* Fix log level

setup/base.go

@@ -15,22 +15,28 @@
 package setup
 
 import (
+	"context"
 	"crypto/tls"
 	"fmt"
 	"io"
 	"net"
 	"net/http"
 	"net/url"
+	"os"
+	"os/signal"
+	"syscall"
 	"time"
 
 	"github.com/matrix-org/dendrite/internal/caching"
 	"github.com/matrix-org/dendrite/internal/httputil"
 	"github.com/matrix-org/gomatrixserverlib"
 	"github.com/prometheus/client_golang/prometheus/promhttp"
+	"go.uber.org/atomic"
 	"golang.org/x/net/http2"
 	"golang.org/x/net/http2/h2c"
 
 	"github.com/matrix-org/dendrite/internal"
+	"github.com/matrix-org/dendrite/setup/process"
 	"github.com/matrix-org/dendrite/userapi/storage/accounts"
 
 	"github.com/gorilla/mux"
@@ -61,6 +67,7 @@ import (
 // should only be used during start up.
 // Must be closed when shutting down.
 type BaseDendrite struct {
+	*process.ProcessContext
 	componentName          string
 	tracerCloser           io.Closer
 	PublicClientAPIMux     *mux.Router
@@ -161,7 +168,9 @@ func NewBaseDendrite(cfg *config.Dendrite, componentName string, useHTTPAPIs boo
 	// We need to be careful with media APIs if they read from a filesystem to make sure they
 	// are not inadvertently reading paths without cleaning, else this could introduce a
 	// directory traversal attack e.g /../../../etc/passwd
+
 	return &BaseDendrite{
+		ProcessContext:         process.NewProcessContext(),
 		componentName:          componentName,
 		UseHTTPAPIs:            useHTTPAPIs,
 		tracerCloser:           closer,
@@ -354,14 +363,26 @@ func (b *BaseDendrite) SetupAndServeHTTP(
 
 	if internalAddr != NoListener && internalAddr != externalAddr {
 		go func() {
+			var internalShutdown atomic.Bool // RegisterOnShutdown can be called more than once
 			logrus.Infof("Starting internal %s listener on %s", b.componentName, internalServ.Addr)
+			b.ProcessContext.ComponentStarted()
+			internalServ.RegisterOnShutdown(func() {
+				if internalShutdown.CAS(false, true) {
+					b.ProcessContext.ComponentFinished()
+					logrus.Infof("Stopped internal HTTP listener")
+				}
+			})
 			if certFile != nil && keyFile != nil {
 				if err := internalServ.ListenAndServeTLS(*certFile, *keyFile); err != nil {
-					logrus.WithError(err).Fatal("failed to serve HTTPS")
+					if err != http.ErrServerClosed {
+						logrus.WithError(err).Fatal("failed to serve HTTPS")
+					}
 				}
 			} else {
 				if err := internalServ.ListenAndServe(); err != nil {
-					logrus.WithError(err).Fatal("failed to serve HTTP")
+					if err != http.ErrServerClosed {
+						logrus.WithError(err).Fatal("failed to serve HTTP")
+					}
 				}
 			}
 			logrus.Infof("Stopped internal %s listener on %s", b.componentName, internalServ.Addr)
@@ -370,19 +391,52 @@ func (b *BaseDendrite) SetupAndServeHTTP(
 
 	if externalAddr != NoListener {
 		go func() {
+			var externalShutdown atomic.Bool // RegisterOnShutdown can be called more than once
 			logrus.Infof("Starting external %s listener on %s", b.componentName, externalServ.Addr)
+			b.ProcessContext.ComponentStarted()
+			externalServ.RegisterOnShutdown(func() {
+				if externalShutdown.CAS(false, true) {
+					b.ProcessContext.ComponentFinished()
+					logrus.Infof("Stopped external HTTP listener")
+				}
+			})
 			if certFile != nil && keyFile != nil {
 				if err := externalServ.ListenAndServeTLS(*certFile, *keyFile); err != nil {
-					logrus.WithError(err).Fatal("failed to serve HTTPS")
+					if err != http.ErrServerClosed {
+						logrus.WithError(err).Fatal("failed to serve HTTPS")
+					}
 				}
 			} else {
 				if err := externalServ.ListenAndServe(); err != nil {
-					logrus.WithError(err).Fatal("failed to serve HTTP")
+					if err != http.ErrServerClosed {
+						logrus.WithError(err).Fatal("failed to serve HTTP")
+					}
 				}
 			}
 			logrus.Infof("Stopped external %s listener on %s", b.componentName, externalServ.Addr)
 		}()
 	}
 
-	select {}
+	<-b.ProcessContext.WaitForShutdown()
+
+	ctx, cancel := context.WithCancel(context.Background())
+	cancel()
+
+	_ = internalServ.Shutdown(ctx)
+	_ = externalServ.Shutdown(ctx)
+	logrus.Infof("Stopped HTTP listeners")
+}
+
+func (b *BaseDendrite) WaitForShutdown() {
+	sigs := make(chan os.Signal, 1)
+	signal.Notify(sigs, syscall.SIGINT, syscall.SIGTERM)
+	<-sigs
+	signal.Reset(syscall.SIGINT, syscall.SIGTERM)
+
+	logrus.Warnf("Shutdown signal received")
+
+	b.ProcessContext.ShutdownDendrite()
+	b.ProcessContext.WaitForComponentsToFinish()
+
+	logrus.Warnf("Dendrite is exiting now")
 }

setup/monolith.go

@@ -27,6 +27,7 @@ import (
 	"github.com/matrix-org/dendrite/mediaapi"
 	roomserverAPI "github.com/matrix-org/dendrite/roomserver/api"
 	"github.com/matrix-org/dendrite/setup/config"
+	"github.com/matrix-org/dendrite/setup/process"
 	serverKeyAPI "github.com/matrix-org/dendrite/signingkeyserver/api"
 	"github.com/matrix-org/dendrite/syncapi"
 	userapi "github.com/matrix-org/dendrite/userapi/api"
@@ -56,7 +57,7 @@ type Monolith struct {
 }
 
 // AddAllPublicRoutes attaches all public paths to the given router
-func (m *Monolith) AddAllPublicRoutes(csMux, ssMux, keyMux, mediaMux *mux.Router) {
+func (m *Monolith) AddAllPublicRoutes(process *process.ProcessContext, csMux, ssMux, keyMux, mediaMux *mux.Router) {
 	clientapi.AddPublicRoutes(
 		csMux, &m.Config.ClientAPI, m.AccountDB,
 		m.FedClient, m.RoomserverAPI,
@@ -71,7 +72,7 @@ func (m *Monolith) AddAllPublicRoutes(csMux, ssMux, keyMux, mediaMux *mux.Router
 	)
 	mediaapi.AddPublicRoutes(mediaMux, &m.Config.MediaAPI, m.UserAPI, m.Client)
 	syncapi.AddPublicRoutes(
-		csMux, m.UserAPI, m.RoomserverAPI,
+		process, csMux, m.UserAPI, m.RoomserverAPI,
 		m.KeyAPI, m.FedClient, &m.Config.SyncAPI,
 	)
 }

setup/process/process.go

@@ -0,0 +1,45 @@
+package process
+
+import (
+	"context"
+	"sync"
+)
+
+type ProcessContext struct {
+	wg       *sync.WaitGroup    // used to wait for components to shutdown
+	ctx      context.Context    // cancelled when Stop is called
+	shutdown context.CancelFunc // shut down Dendrite
+}
+
+func NewProcessContext() *ProcessContext {
+	ctx, shutdown := context.WithCancel(context.Background())
+	return &ProcessContext{
+		ctx:      ctx,
+		shutdown: shutdown,
+		wg:       &sync.WaitGroup{},
+	}
+}
+
+func (b *ProcessContext) Context() context.Context {
+	return context.WithValue(b.ctx, "scope", "process") // nolint:staticcheck
+}
+
+func (b *ProcessContext) ComponentStarted() {
+	b.wg.Add(1)
+}
+
+func (b *ProcessContext) ComponentFinished() {
+	b.wg.Done()
+}
+
+func (b *ProcessContext) ShutdownDendrite() {
+	b.shutdown()
+}
+
+func (b *ProcessContext) WaitForShutdown() <-chan struct{} {
+	return b.ctx.Done()
+}
+
+func (b *ProcessContext) WaitForComponentsToFinish() {
+	b.wg.Wait()
+}