priority: release references to child policies which are removed

internal/balancergroup/balancergroup.go

@@ -97,7 +97,7 @@ func (sbc *subBalancerWrapper) startBalancer() {
 	if sbc.balancer == nil {
 		sbc.balancer = gracefulswitch.NewBalancer(sbc, sbc.buildOpts)
 	}
-	sbc.group.logger.Infof("Creating child policy of type %v", sbc.builder.Name())
+	sbc.group.logger.Infof("Creating child policy of type %q for locality %q", sbc.builder.Name(), sbc.id)
 	sbc.balancer.SwitchTo(sbc.builder)
 	if sbc.ccState != nil {
 		sbc.balancer.UpdateClientConnState(*sbc.ccState)
@@ -298,17 +298,29 @@ func (bg *BalancerGroup) Start() {
 	bg.outgoingMu.Unlock()
 }
 
-// Add adds a balancer built by builder to the group, with given id.
-func (bg *BalancerGroup) Add(id string, builder balancer.Builder) {
+// AddWithClientConn adds a balancer with the given id to the group. The
+// balancer is built with a balancer builder registered with balancerName. The
+// given ClientConn is passed to the newly built balancer instead of the
+// onepassed to balancergroup.New().
+//
+// TODO: Get rid of the existing Add() API and replace it with this.
+func (bg *BalancerGroup) AddWithClientConn(id, balancerName string, cc balancer.ClientConn) error {
+	bg.logger.Infof("Adding child policy of type %q for locality %q", balancerName, id)
+	builder := balancer.Get(balancerName)
+	if builder == nil {
+		return fmt.Errorf("unregistered balancer name %q", balancerName)
+	}
+
 	// Store data in static map, and then check to see if bg is started.
 	bg.outgoingMu.Lock()
+	defer bg.outgoingMu.Unlock()
 	var sbc *subBalancerWrapper
 	// If outgoingStarted is true, search in the cache. Otherwise, cache is
 	// guaranteed to be empty, searching is unnecessary.
 	if bg.outgoingStarted {
 		if old, ok := bg.balancerCache.Remove(id); ok {
 			sbc, _ = old.(*subBalancerWrapper)
-			if sbc != nil && sbc.builder != builder {
+			if sbc != nil && sbc.builder.Name() != builder.Name() {
 				// If the sub-balancer in cache was built with a different
 				// balancer builder, don't use it, cleanup this old-balancer,
 				// and behave as sub-balancer is not found in cache.
@@ -326,7 +338,7 @@ func (bg *BalancerGroup) Add(id string, builder balancer.Builder) {
 	}
 	if sbc == nil {
 		sbc = &subBalancerWrapper{
-			ClientConn: bg.cc,
+			ClientConn: cc,
 			id:         id,
 			group:      bg,
 			builder:    builder,
@@ -343,11 +355,18 @@ func (bg *BalancerGroup) Add(id string, builder balancer.Builder) {
 		sbc.updateBalancerStateWithCachedPicker()
 	}
 	bg.idToBalancerConfig[id] = sbc
-	bg.outgoingMu.Unlock()
+	return nil
+}
+
+// Add adds a balancer built by builder to the group, with given id.
+func (bg *BalancerGroup) Add(id string, builder balancer.Builder) {
+	bg.AddWithClientConn(id, builder.Name(), bg.cc)
 }
 
 // UpdateBuilder updates the builder for a current child, starting the Graceful
 // Switch process for that child.
+//
+// TODO: update this API to take the name of the new builder instead.
 func (bg *BalancerGroup) UpdateBuilder(id string, builder balancer.Builder) {
 	bg.outgoingMu.Lock()
 	// This does not deal with the balancer cache because this call should come
@@ -369,6 +388,7 @@ func (bg *BalancerGroup) UpdateBuilder(id string, builder balancer.Builder) {
 // closed after timeout. Cleanup work (closing sub-balancer and removing
 // subconns) will be done after timeout.
 func (bg *BalancerGroup) Remove(id string) {
+	bg.logger.Infof("Removing child policy for locality %q", id)
 	bg.outgoingMu.Lock()
 	if sbToRemove, ok := bg.idToBalancerConfig[id]; ok {
 		if bg.outgoingStarted {

internal/balancergroup/balancergroup_test.go

@@ -374,11 +374,19 @@ func (s) TestBalancerGroup_locality_caching_not_readd_within_timeout(t *testing.
 	}
 }
 
-// Wrap the rr builder, so it behaves the same, but has a different pointer.
+// Wrap the rr builder, so it behaves the same, but has a different name.
 type noopBalancerBuilderWrapper struct {
 	balancer.Builder
 }
 
+func init() {
+	balancer.Register(&noopBalancerBuilderWrapper{Builder: rrBuilder})
+}
+
+func (*noopBalancerBuilderWrapper) Name() string {
+	return "noopBalancerBuilderWrapper"
+}
+
 // After removing a sub-balancer, re-add with same ID, but different balancer
 // builder. Old subconns should be removed, and new subconns should be created.
 func (s) TestBalancerGroup_locality_caching_readd_with_different_builder(t *testing.T) {

xds/internal/balancer/clusterimpl/clusterimpl.go

@@ -331,6 +331,7 @@ func (b *clusterImplBalancer) Close() {
 	if b.childLB != nil {
 		b.childLB.Close()
 		b.childLB = nil
+		b.childState = balancer.State{}
 	}
 	<-b.done.Done()
 	b.logger.Infof("Shutdown")

xds/internal/balancer/clusterresolver/clusterresolver.go

@@ -194,7 +194,6 @@ func (b *clusterResolverBalancer) handleWatchUpdate(update *resourceUpdate) {
 		return
 	}
 
-	b.logger.Infof("resource update: %+v", pretty.ToJSON(update.priorities))
 	b.watchUpdateReceived = true
 	b.priorities = update.priorities
 

xds/internal/balancer/priority/balancer.go

@@ -127,7 +127,7 @@ func (b *priorityBalancer) UpdateClientConnState(s balancer.ClientConnState) err
 			// This is a new child, add it to the children list. But note that
 			// the balancer isn't built, because this child can be a low
 			// priority. If necessary, it will be built when syncing priorities.
-			cb := newChildBalancer(name, b, bb)
+			cb := newChildBalancer(name, b, bb.Name(), b.cc)
 			cb.updateConfig(newSubConfig, resolver.State{
 				Addresses:     addressesSplit[name],
 				ServiceConfig: s.ResolverState.ServiceConfig,
@@ -141,9 +141,9 @@ func (b *priorityBalancer) UpdateClientConnState(s balancer.ClientConnState) err
 
 		// The balancing policy name is changed, close the old child. But don't
 		// rebuild, rebuild will happen when syncing priorities.
-		if currentChild.bb.Name() != bb.Name() {
+		if currentChild.balancerName != bb.Name() {
 			currentChild.stop()
-			currentChild.updateBuilder(bb)
+			currentChild.updateBalancerName(bb.Name())
 		}
 
 		// Update config and address, but note that this doesn't send the
@@ -155,10 +155,11 @@ func (b *priorityBalancer) UpdateClientConnState(s balancer.ClientConnState) err
 			Attributes:    s.ResolverState.Attributes,
 		})
 	}
-	// Remove child from children if it's not in new config.
+	// Cleanup resources used by children removed from the config.
 	for name, oldChild := range b.children {
 		if _, ok := newConfig.Children[name]; !ok {
 			oldChild.stop()
+			delete(b.children, name)
 		}
 	}
 

xds/internal/balancer/priority/balancer_child.go

@@ -29,9 +29,11 @@ import (
 )
 
 type childBalancer struct {
-	name   string
-	parent *priorityBalancer
-	bb     *ignoreResolveNowBalancerBuilder
+	name         string
+	parent       *priorityBalancer
+	parentCC     balancer.ClientConn
+	balancerName string
+	cc           *ignoreResolveNowClientConn
 
 	ignoreReresolutionRequests bool
 	config                     serviceconfig.LoadBalancingConfig
@@ -53,12 +55,14 @@ type childBalancer struct {
 
 // newChildBalancer creates a child balancer place holder, but doesn't
 // build/start the child balancer.
-func newChildBalancer(name string, parent *priorityBalancer, bb balancer.Builder) *childBalancer {
+func newChildBalancer(name string, parent *priorityBalancer, balancerName string, cc balancer.ClientConn) *childBalancer {
 	return &childBalancer{
-		name:    name,
-		parent:  parent,
-		bb:      newIgnoreResolveNowBalancerBuilder(bb, false),
-		started: false,
+		name:         name,
+		parent:       parent,
+		parentCC:     cc,
+		balancerName: balancerName,
+		cc:           newIgnoreResolveNowClientConn(cc, false),
+		started:      false,
 		// Start with the connecting state and picker with re-pick error, so
 		// that when a priority switch causes this child picked before it's
 		// balancing policy is created, a re-pick will happen.
@@ -69,9 +73,13 @@ func newChildBalancer(name string, parent *priorityBalancer, bb balancer.Builder
 	}
 }
 
-// updateBuilder updates builder for the child, but doesn't build.
-func (cb *childBalancer) updateBuilder(bb balancer.Builder) {
-	cb.bb = newIgnoreResolveNowBalancerBuilder(bb, cb.ignoreReresolutionRequests)
+// updateBalancerName updates balancer name for the child, but doesn't build a
+// new one. The parent priority LB always closes the child policy before
+// updating the balancer name, and the new balancer is built when it gets added
+// to the balancergroup as part of start().
+func (cb *childBalancer) updateBalancerName(balancerName string) {
+	cb.balancerName = balancerName
+	cb.cc = newIgnoreResolveNowClientConn(cb.parentCC, cb.ignoreReresolutionRequests)
 }
 
 // updateConfig sets childBalancer's config and state, but doesn't send update to
@@ -93,14 +101,14 @@ func (cb *childBalancer) start() {
 		return
 	}
 	cb.started = true
-	cb.parent.bg.Add(cb.name, cb.bb)
+	cb.parent.bg.AddWithClientConn(cb.name, cb.balancerName, cb.cc)
 	cb.startInitTimer()
 	cb.sendUpdate()
 }
 
 // sendUpdate sends the addresses and config to the child balancer.
 func (cb *childBalancer) sendUpdate() {
-	cb.bb.updateIgnoreResolveNow(cb.ignoreReresolutionRequests)
+	cb.cc.updateIgnoreResolveNow(cb.ignoreReresolutionRequests)
 	// TODO: return and aggregate the returned error in the parent.
 	err := cb.parent.bg.UpdateClientConnState(cb.name, balancer.ClientConnState{
 		ResolverState:  cb.rState,

xds/internal/balancer/priority/balancer_priority.go

@@ -162,6 +162,9 @@ func (b *priorityBalancer) handleChildStateUpdate(childName string, s balancer.S
 		b.logger.Warningf("priority: child balancer not found for child %v", childName)
 		return
 	}
+	if !child.started {
+		return
+	}
 	child.state = s
 
 	// We start/stop the init timer of this child based on the new connectivity