closedts: stop workloop on close when remote is unavailable

Previously if remote node containing replica was removed,
sidetransport connection will keep trying to connect to non-existing
node until server is stopped regardless of connection status.
This is not good as it will keep trying every second wasting
resources and also spamming log with message every 10 seconds.
This patch moves liveness check to the top of runloop to terminate
it early as soon as connection is closed.

Release note: None

pkg/kv/kvserver/closedts/sidetransport/BUILD.bazel

@@ -44,6 +44,7 @@ go_test(
         "//pkg/roachpb:with-mocks",
         "//pkg/rpc",
         "//pkg/settings/cluster",
+        "//pkg/testutils",
         "//pkg/util/hlc",
         "//pkg/util/leaktest",
         "//pkg/util/log",

pkg/kv/kvserver/closedts/sidetransport/sender.go

@@ -107,7 +107,8 @@ type streamState struct {
 }
 
 type connTestingKnobs struct {
-	beforeSend func(destNodeID roachpb.NodeID, msg *ctpb.Update)
+	beforeSend         func(destNodeID roachpb.NodeID, msg *ctpb.Update)
+	sleepOnErrOverride time.Duration
 }
 
 // trackedRange contains the information that the side-transport last published
@@ -666,6 +667,10 @@ type nodeDialer interface {
 	Dial(ctx context.Context, nodeID roachpb.NodeID, class rpc.ConnectionClass) (_ *grpc.ClientConn, err error)
 }
 
+// On sending errors, we sleep a bit as to not spin on a tripped
+// circuit-breaker in the Dialer.
+const sleepOnErr = time.Second
+
 // rpcConn is an implementation of conn that is implemented using a gRPC stream.
 //
 // The connection will read messages from producer.buf. If the buffer overflows
@@ -773,18 +778,23 @@ func (r *rpcConn) run(ctx context.Context, stopper *stop.Stopper) {
 			defer r.cleanupStream(nil /* err */)
 			everyN := log.Every(10 * time.Second)
 
-			// On sending errors, we sleep a bit as to not spin on a tripped
-			// circuit-breaker in the Dialer.
-			const sleepOnErr = time.Second
+			errSleepTime := sleepOnErr
+			if r.testingKnobs.sleepOnErrOverride > 0 {
+				errSleepTime = r.testingKnobs.sleepOnErrOverride
+			}
+
 			for {
 				if ctx.Err() != nil {
 					return
 				}
+				if atomic.LoadInt32(&r.closed) > 0 {
+					return
+				}
 				if err := r.maybeConnect(ctx, stopper); err != nil {
 					if everyN.ShouldLog() {
 						log.Infof(ctx, "side-transport failed to connect to n%d: %s", r.nodeID, err)
 					}
-					time.Sleep(sleepOnErr)
+					time.Sleep(errSleepTime)
 					continue
 				}
 
@@ -797,10 +807,6 @@ func (r *rpcConn) run(ctx context.Context, stopper *stop.Stopper) {
 				if !ok {
 					return
 				}
-				closed := atomic.LoadInt32(&r.closed) > 0
-				if closed {
-					return
-				}
 
 				if msg == nil {
 					// The sequence number we've requested is no longer in the buffer. We
@@ -827,7 +833,7 @@ func (r *rpcConn) run(ctx context.Context, stopper *stop.Stopper) {
 					// should have a blocking version of Dial() that we just leave hanging
 					// and get a notification when it succeeds.
 					r.cleanupStream(err)
-					time.Sleep(sleepOnErr)
+					time.Sleep(errSleepTime)
 				}
 			}
 		})

pkg/kv/kvserver/closedts/sidetransport/sender_test.go

@@ -14,6 +14,7 @@ import (
 	"context"
 	"fmt"
 	"net"
+	"sync/atomic"
 	"testing"
 	"time"
 
@@ -23,6 +24,7 @@ import (
 	"github.com/cockroachdb/cockroach/pkg/roachpb"
 	"github.com/cockroachdb/cockroach/pkg/rpc"
 	"github.com/cockroachdb/cockroach/pkg/settings/cluster"
+	"github.com/cockroachdb/cockroach/pkg/testutils"
 	"github.com/cockroachdb/cockroach/pkg/util/hlc"
 	"github.com/cockroachdb/cockroach/pkg/util/leaktest"
 	"github.com/cockroachdb/cockroach/pkg/util/log"
@@ -547,3 +549,55 @@ func TestSenderReceiverIntegration(t *testing.T) {
 	// Check that the other Receiver is still receiving updates.
 	<-receivers[2].testingKnobs[1].onMsg
 }
+
+type failingDialer struct {
+	dialCount int32
+}
+
+var _ nodeDialer = &failingDialer{}
+
+func (f *failingDialer) Dial(
+	ctx context.Context, nodeID roachpb.NodeID, class rpc.ConnectionClass,
+) (_ *grpc.ClientConn, err error) {
+	atomic.AddInt32(&f.dialCount, 1)
+	return nil, errors.New("failingDialer")
+}
+
+func (f *failingDialer) callCount() int32 {
+	return atomic.LoadInt32(&f.dialCount)
+}
+
+// TestRPCConnStopOnClose verifies that connections that are closed would stop
+// their work loops eagerly even when nodes they are talking to are unreachable.
+func TestRPCConnStopOnClose(t *testing.T) {
+	defer leaktest.AfterTest(t)()
+	defer log.Scope(t).Close(t)
+
+	ctx := context.Background()
+	stopper := stop.NewStopper()
+	defer stopper.Stop(ctx)
+
+	sleepTime := time.Millisecond
+
+	dialer := &failingDialer{}
+	factory := newRPCConnFactory(dialer, connTestingKnobs{sleepOnErrOverride: sleepTime})
+	connection := factory.new(nil, /* sender is not needed as dialer always fails Dial attempts */
+		roachpb.NodeID(1))
+	connection.run(ctx, stopper)
+
+	// Wait for first dial attempt for sanity reasons.
+	testutils.SucceedsSoon(t, func() error {
+		if dialer.callCount() == 0 {
+			return errors.New("connection didn't dial yet")
+		}
+		return nil
+	})
+	connection.close()
+	// Ensure that dialing stops once connection is stopped.
+	testutils.SucceedsSoon(t, func() error {
+		if stopper.NumTasks() > 0 {
+			return errors.New("connection worker didn't stop yet")
+		}
+		return nil
+	})
+}