Enter transient failure on resolver creation failure

Author: arjan-bal

clientconn.go

@@ -357,9 +357,7 @@ func (cc *ClientConn) exitIdleMode() (err error) {
 	// This needs to be called without cc.mu because this builds a new resolver
 	// which might update state or report error inline, which would then need to
 	// acquire cc.mu.
-	if err := cc.resolverWrapper.start(); err != nil {
-		return err
-	}
+	cc.resolverWrapper.start()
 
 	cc.addTraceEvent("exiting idle mode")
 	return nil

clientconn_parsed_target_test.go

@@ -27,6 +27,7 @@ import (
 	"time"
 
 	"github.com/google/go-cmp/cmp"
+	"google.golang.org/grpc/connectivity"
 	"google.golang.org/grpc/credentials/insecure"
 	"google.golang.org/grpc/internal"
 	"google.golang.org/grpc/internal/testutils"
@@ -185,6 +186,8 @@ func (s) TestParsedTarget_Success_WithoutCustomDialer(t *testing.T) {
 }
 
 func (s) TestParsedTarget_Failure_WithoutCustomDialer(t *testing.T) {
+	ctx, cancel := context.WithTimeout(context.Background(), defaultTestTimeout)
+	defer cancel()
 	targets := []string{
 		"",
 		"unix://a/b/c",
@@ -195,10 +198,13 @@ func (s) TestParsedTarget_Failure_WithoutCustomDialer(t *testing.T) {
 
 	for _, target := range targets {
 		t.Run(target, func(t *testing.T) {
-			if cc, err := Dial(target, WithTransportCredentials(insecure.NewCredentials())); err == nil {
-				defer cc.Close()
-				t.Fatalf("Dial(%q) succeeded cc.parsedTarget = %+v, expected to fail", target, cc.parsedTarget)
+			cc, err := NewClient(target, WithTransportCredentials(insecure.NewCredentials()))
+			if err != nil {
+				t.Fatalf("NewClient(%q) failed: %v", target, err)
 			}
+			defer cc.Close()
+			cc.Connect()
+			testutils.AwaitState(ctx, t, cc, connectivity.TransientFailure)
 		})
 	}
 }

resolver_wrapper.go

@@ -20,6 +20,7 @@ package grpc
 
 import (
 	"context"
+	"fmt"
 	"strings"
 	"sync"
 
@@ -62,11 +63,15 @@ func newCCResolverWrapper(cc *ClientConn) *ccResolverWrapper {
 	}
 }
 
-// start builds the name resolver using the resolver.Builder in cc and returns
-// any error encountered.  It must always be the first operation performed on
-// any newly created ccResolverWrapper, except that close may be called instead.
-func (ccr *ccResolverWrapper) start() error {
-	errCh := make(chan error)
+// start builds the name resolver using the resolver.Builder in cc.
+// If an error is encountered, it will report the error to the load balancing
+// policy via cc.ReportError(). This action allows the policy to set the channel
+// state to TransientFailure and ensures the error is propagated to new RPCs,
+// causing them to fail.
+// It must always be the first operation performed on any newly created
+// ccResolverWrapper, except that close may be called instead.
+func (ccr *ccResolverWrapper) start() {
+	doneCh := make(chan struct{})
 	ccr.serializer.TrySchedule(func(ctx context.Context) {
 		if ctx.Err() != nil {
 			return
@@ -90,11 +95,24 @@ func (ccr *ccResolverWrapper) start() error {
 		} else {
 			ccr.resolver, err = delegatingresolver.New(ccr.cc.parsedTarget, ccr, opts, ccr.cc.resolverBuilder, ccr.cc.dopts.enableLocalDNSResolution)
 		}
-		errCh <- err
+
+		if err != nil {
+			ccr.resolver = &nopResolver{}
+			ccr.ReportError(fmt.Errorf("resolver creation failed: %v", err))
+		}
+
+		doneCh <- struct{}{}
 	})
-	return <-errCh
+	<-doneCh
 }
 
+type nopResolver struct {
+}
+
+func (*nopResolver) ResolveNow(resolver.ResolveNowOptions) {}
+
+func (*nopResolver) Close() {}
+
 func (ccr *ccResolverWrapper) resolveNow(o resolver.ResolveNowOptions) {
 	ccr.serializer.TrySchedule(func(ctx context.Context) {
 		if ctx.Err() != nil || ccr.resolver == nil {