Backport of Fix nil-pointer panics from proxycfg package. into release/1.15.x

agent/proxycfg/api_gateway.go

@@ -73,6 +73,7 @@ func (h *handlerAPIGateway) initialize(ctx context.Context) (ConfigSnapshot, err
 	snap.APIGateway.WatchedDiscoveryChains = make(map[UpstreamID]context.CancelFunc)
 	snap.APIGateway.WatchedGateways = make(map[UpstreamID]map[string]context.CancelFunc)
 	snap.APIGateway.WatchedGatewayEndpoints = make(map[UpstreamID]map[string]structs.CheckServiceNodes)
+	snap.APIGateway.WatchedLocalGWEndpoints = watch.NewMap[string, structs.CheckServiceNodes]()
 	snap.APIGateway.WatchedUpstreams = make(map[UpstreamID]map[string]context.CancelFunc)
 	snap.APIGateway.WatchedUpstreamEndpoints = make(map[UpstreamID]map[string]structs.CheckServiceNodes)
 

agent/proxycfg/ingress_gateway.go

@@ -67,6 +67,7 @@ func (s *handlerIngressGateway) initialize(ctx context.Context) (ConfigSnapshot,
 	snap.IngressGateway.WatchedUpstreamEndpoints = make(map[UpstreamID]map[string]structs.CheckServiceNodes)
 	snap.IngressGateway.WatchedGateways = make(map[UpstreamID]map[string]context.CancelFunc)
 	snap.IngressGateway.WatchedGatewayEndpoints = make(map[UpstreamID]map[string]structs.CheckServiceNodes)
+	snap.IngressGateway.WatchedLocalGWEndpoints = watch.NewMap[string, structs.CheckServiceNodes]()
 	snap.IngressGateway.Listeners = make(map[IngressListenerKey]structs.IngressListener)
 	snap.IngressGateway.UpstreamPeerTrustBundles = watch.NewMap[string, *pbpeering.PeeringTrustBundle]()
 	snap.IngressGateway.PeerUpstreamEndpoints = watch.NewMap[UpstreamID, structs.CheckServiceNodes]()

agent/proxycfg/manager.go

@@ -2,6 +2,7 @@ package proxycfg
 
 import (
 	"errors"
+	"runtime/debug"
 	"sync"
 
 	"github.com/hashicorp/go-hclog"
@@ -142,6 +143,20 @@ func (m *Manager) Register(id ProxyID, ns *structs.NodeService, source ProxySour
 	m.mu.Lock()
 	defer m.mu.Unlock()
 
+	defer func() {
+		if r := recover(); r != nil {
+			m.Logger.Error("unexpected panic during service manager registration",
+				"node", id.NodeName,
+				"service", id.ServiceID,
+				"message", r,
+				"stacktrace", string(debug.Stack()),
+			)
+		}
+	}()
+	return m.register(id, ns, source, token, overwrite)
+}
+
+func (m *Manager) register(id ProxyID, ns *structs.NodeService, source ProxySource, token string, overwrite bool) error {
 	state, ok := m.proxies[id]
 	if ok {
 		if state.source != source && !overwrite {

agent/proxycfg/state.go

@@ -6,6 +6,7 @@ import (
 	"fmt"
 	"net"
 	"reflect"
+	"runtime/debug"
 	"sync/atomic"
 	"time"
 
@@ -298,6 +299,21 @@ func newConfigSnapshotFromServiceInstance(s serviceInstance, config stateConfig)
 }
 
 func (s *state) run(ctx context.Context, snap *ConfigSnapshot) {
+	// Add a recover here so than any panics do not make their way up
+	// into the server / agent.
+	defer func() {
+		if r := recover(); r != nil {
+			s.logger.Error("unexpected panic while running proxycfg",
+				"node", s.serviceInstance.proxyID.NodeName,
+				"service", s.serviceInstance.proxyID.ServiceID,
+				"message", r,
+				"stacktrace", string(debug.Stack()))
+		}
+	}()
+	s.unsafeRun(ctx, snap)
+}
+
+func (s *state) unsafeRun(ctx context.Context, snap *ConfigSnapshot) {
 	// Close the channel we return from Watch when we stop so consumers can stop
 	// watching and clean up their goroutines. It's important we do this here and
 	// not in Close since this routine sends on this chan and so might panic if it