fix: Graceful shutdown for the API server (argoproj#18642)

Closes argoproj#18642

Implements a graceful shutdown the the API server. Without this, ArgoCD API server will eventually return 502 during rolling update.

Signed-off-by: Andrii Korotkov <andrii.korotkov@verkada.com>
Co-authored-by: Leonardo Luz Almeida <leonardo_almeida@intuit.com>
Co-authored-by: Michael Crenshaw <350466+crenshaw-dev@users.noreply.github.com>

cmd/argocd-server/commands/argocd_server.go

@@ -273,6 +273,9 @@ func NewCommand() *cobra.Command {
 				if closer != nil {
 					closer()
 				}
+				if argocd.ReceivedSignal() {
+					break
+				}
 			}
 		},
 		Example: templates.Examples(`

server/server.go

@@ -13,13 +13,15 @@ import (
 	"net/url"
 	"os"
 	"os/exec"
+	"os/signal"
 	"path"
 	"path/filepath"
 	"reflect"
 	"regexp"
 	go_runtime "runtime"
 	"strings"
 	gosync "sync"
+	"syscall"
 	"time"
 
 	// nolint:staticcheck
@@ -187,7 +189,7 @@ type ArgoCDServer struct {
 	db             db.ArgoDB
 
 	// stopCh is the channel which when closed, will shutdown the Argo CD server
-	stopCh            chan struct{}
+	stopCh            chan os.Signal
 	userStateStorage  util_session.UserStateStorage
 	indexDataInit     gosync.Once
 	indexData         []byte
@@ -198,6 +200,8 @@ type ArgoCDServer struct {
 	configMapInformer cache.SharedIndexInformer
 	serviceSet        *ArgoCDServiceSet
 	extensionManager  *extension.Manager
+	shutdown          func()
+	receivedSignal    bool
 }
 
 type ArgoCDServerOpts struct {
@@ -329,6 +333,9 @@ func NewServer(ctx context.Context, opts ArgoCDServerOpts, appsetOpts Applicatio
 	pg := extension.NewDefaultProjectGetter(projLister, dbInstance)
 	ug := extension.NewDefaultUserGetter(policyEnf)
 	em := extension.NewManager(logger, opts.Namespace, sg, ag, pg, enf, ug)
+	noopShutdown := func() {
+		log.Error("API Server Shutdown function called but server is not started yet.")
+	}
 
 	a := &ArgoCDServer{
 		ArgoCDServerOpts:   opts,
@@ -352,6 +359,8 @@ func NewServer(ctx context.Context, opts ArgoCDServerOpts, appsetOpts Applicatio
 		secretInformer:     secretInformer,
 		configMapInformer:  configMapInformer,
 		extensionManager:   em,
+		shutdown:           noopShutdown,
+		receivedSignal:     false,
 	}
 
 	err = a.logInClusterWarnings()
@@ -601,35 +610,108 @@ func (a *ArgoCDServer) Run(ctx context.Context, listeners *Listeners) {
 		log.Fatal("Timed out waiting for project cache to sync")
 	}
 
-	a.stopCh = make(chan struct{})
-	<-a.stopCh
+	shutdownFunc := func() {
+		log.Info("API Server shutdown initiated. Shutting down servers...")
+		sCtx, cancel := context.WithTimeout(context.Background(), 20*time.Second)
+		defer cancel()
+		var wg gosync.WaitGroup
+
+		// Shutdown http server
+		wg.Add(1)
+		go func() {
+			defer wg.Done()
+			err := httpS.Shutdown(sCtx)
+			if err != nil {
+				log.Errorf("Error shutting down http server: %s", err)
+			}
+		}()
+
+		if httpsS != nil {
+			// Shutdown https server
+			wg.Add(1)
+			go func() {
+				defer wg.Done()
+				err := httpsS.Shutdown(sCtx)
+				if err != nil {
+					log.Errorf("Error shutting down https server: %s", err)
+				}
+			}()
+		}
+
+		// Shutdown gRPC server
+		wg.Add(1)
+		go func() {
+			defer wg.Done()
+			grpcS.GracefulStop()
+		}()
+
+		// Shutdown metrics server
+		wg.Add(1)
+		go func() {
+			defer wg.Done()
+			err := metricsServ.Shutdown(sCtx)
+			if err != nil {
+				log.Errorf("Error shutting down metrics server: %s", err)
+			}
+		}()
+
+		if tlsm != nil {
+			// Shutdown tls server
+			wg.Add(1)
+			go func() {
+				defer wg.Done()
+				tlsm.Close()
+			}()
+		}
+
+		// Shutdown tcp server
+		wg.Add(1)
+		go func() {
+			defer wg.Done()
+			tcpm.Close()
+		}()
+
+		c := make(chan struct{})
+		// This goroutine will wait for all servers to conclude the shutdown
+		// process
+		go func() {
+			defer close(c)
+			wg.Wait()
+		}()
+
+		select {
+		case <-c:
+			log.Info("All servers were gracefully shutdown. Exiting...")
+		case <-sCtx.Done():
+			log.Warn("Graceful shutdown timeout. Exiting...")
+		}
+	}
+	a.shutdown = shutdownFunc
+
+	a.stopCh = make(chan os.Signal, 1)
+	signal.Notify(a.stopCh, os.Interrupt, syscall.SIGINT, syscall.SIGTERM)
+	signal := <-a.stopCh
+	log.Infof("API Server received signal: %s", signal.String())
+	a.receivedSignal = true
+	a.shutdown()
 }
 
 func (a *ArgoCDServer) Initialized() bool {
 	return a.projInformer.HasSynced() && a.appInformer.HasSynced()
 }
 
+// ReceivedSignal returns whether a shutdown was initiated by a signal as opposed to
+// a watch.
+func (a *ArgoCDServer) ReceivedSignal() bool {
+	return a.receivedSignal
+}
+
 // checkServeErr checks the error from a .Serve() call to decide if it was a graceful shutdown
 func (a *ArgoCDServer) checkServeErr(name string, err error) {
-	if err != nil {
-		if a.stopCh == nil {
-			// a nil stopCh indicates a graceful shutdown
-			log.Infof("graceful shutdown %s: %v", name, err)
-		} else {
-			log.Fatalf("%s: %v", name, err)
-		}
+	if err != nil && !errors.Is(err, http.ErrServerClosed) {
+		log.Errorf("Error received from server %s: %v", name, err)
 	} else {
-		log.Infof("graceful shutdown %s", name)
-	}
-}
-
-// Shutdown stops the Argo CD server
-func (a *ArgoCDServer) Shutdown() {
-	log.Info("Shut down requested")
-	stopCh := a.stopCh
-	a.stopCh = nil
-	if stopCh != nil {
-		close(stopCh)
+		log.Infof("Graceful shutdown of %s initiated", name)
 	}
 }
 
@@ -734,7 +816,7 @@ func (a *ArgoCDServer) watchSettings() {
 		}
 	}
 	log.Info("shutting down settings watch")
-	a.Shutdown()
+	a.shutdown()
 	a.settingsMgr.Unsubscribe(updateCh)
 	close(updateCh)
 }