grpc 1.48.0 → 1.50.0

data/src/core/ext/filters/client_channel/lb_policy/xds/xds_cluster_resolver.cc

@@ -27,11 +27,11 @@
 #include <utility>
 #include <vector>
 
-#include "absl/container/inlined_vector.h"
 #include "absl/memory/memory.h"
 #include "absl/status/status.h"
 #include "absl/status/statusor.h"
 #include "absl/strings/str_cat.h"
+#include "absl/strings/str_join.h"
 #include "absl/strings/string_view.h"
 #include "absl/types/optional.h"
 
@@ -39,34 +39,36 @@
 #include <grpc/impl/codegen/grpc_types.h>
 #include <grpc/support/log.h>
 
-#include "src/core/ext/filters/client_channel/lb_policy.h"
 #include "src/core/ext/filters/client_channel/lb_policy/address_filtering.h"
 #include "src/core/ext/filters/client_channel/lb_policy/child_policy_handler.h"
 #include "src/core/ext/filters/client_channel/lb_policy/outlier_detection/outlier_detection.h"
 #include "src/core/ext/filters/client_channel/lb_policy/ring_hash/ring_hash.h"
 #include "src/core/ext/filters/client_channel/lb_policy/xds/xds.h"
 #include "src/core/ext/filters/client_channel/lb_policy/xds/xds_channel_args.h"
-#include "src/core/ext/filters/client_channel/lb_policy_factory.h"
-#include "src/core/ext/filters/client_channel/lb_policy_registry.h"
 #include "src/core/ext/filters/client_channel/resolver/fake/fake_resolver.h"
-#include "src/core/ext/filters/client_channel/subchannel_interface.h"
 #include "src/core/ext/xds/xds_bootstrap.h"
+#include "src/core/ext/xds/xds_bootstrap_grpc.h"
 #include "src/core/ext/xds/xds_client.h"
+#include "src/core/ext/xds/xds_client_grpc.h"
 #include "src/core/ext/xds/xds_client_stats.h"
 #include "src/core/ext/xds/xds_endpoint.h"
 #include "src/core/ext/xds/xds_resource_type_impl.h"
 #include "src/core/lib/channel/channel_args.h"
 #include "src/core/lib/config/core_configuration.h"
 #include "src/core/lib/debug/trace.h"
-#include "src/core/lib/gpr/string.h"
 #include "src/core/lib/gprpp/debug_location.h"
 #include "src/core/lib/gprpp/orphanable.h"
 #include "src/core/lib/gprpp/ref_counted_ptr.h"
-#include "src/core/lib/gprpp/time.h"
-#include "src/core/lib/iomgr/error.h"
+#include "src/core/lib/gprpp/validation_errors.h"
+#include "src/core/lib/gprpp/work_serializer.h"
 #include "src/core/lib/iomgr/pollset_set.h"
-#include "src/core/lib/iomgr/work_serializer.h"
 #include "src/core/lib/json/json.h"
+#include "src/core/lib/json/json_args.h"
+#include "src/core/lib/json/json_object_loader.h"
+#include "src/core/lib/load_balancing/lb_policy.h"
+#include "src/core/lib/load_balancing/lb_policy_factory.h"
+#include "src/core/lib/load_balancing/lb_policy_registry.h"
+#include "src/core/lib/load_balancing/subchannel_interface.h"
 #include "src/core/lib/resolver/resolver.h"
 #include "src/core/lib/resolver/resolver_registry.h"
 #include "src/core/lib/resolver/server_address.h"
@@ -82,14 +84,15 @@ const char* kXdsLocalityNameAttributeKey = "xds_locality_name";
 
 namespace {
 
-constexpr char kXdsClusterResolver[] = "xds_cluster_resolver_experimental";
+constexpr absl::string_view kXdsClusterResolver =
+    "xds_cluster_resolver_experimental";
 
 // Config for EDS LB policy.
 class XdsClusterResolverLbConfig : public LoadBalancingPolicy::Config {
  public:
   struct DiscoveryMechanism {
     std::string cluster_name;
-    absl::optional<XdsBootstrap::XdsServer> lrs_load_reporting_server;
+    absl::optional<GrpcXdsBootstrap::GrpcXdsServer> lrs_load_reporting_server;
     uint32_t max_concurrent_requests;
     enum DiscoveryMechanismType {
       EDS,
@@ -98,6 +101,13 @@ class XdsClusterResolverLbConfig : public LoadBalancingPolicy::Config {
     DiscoveryMechanismType type;
     std::string eds_service_name;
     std::string dns_hostname;
+
+    // This is type Json::Object instead of OutlierDetectionConfig, because we
+    // don't actually need to validate the contents of the outlier detection
+    // config here.  In this case, the JSON is generated by the CDS policy
+    // instead of coming from service config, so it's not actually any better
+    // to catch the problem here than it is to catch it in the
+    // outlier_detection policy itself, so here we just act as a pass-through.
     absl::optional<Json::Object> outlier_detection_lb_config;
 
     bool operator==(const DiscoveryMechanism& other) const {
@@ -109,23 +119,37 @@ class XdsClusterResolverLbConfig : public LoadBalancingPolicy::Config {
               dns_hostname == other.dns_hostname &&
               outlier_detection_lb_config == other.outlier_detection_lb_config);
     }
+
+    static const JsonLoaderInterface* JsonLoader(const JsonArgs&);
+    void JsonPostLoad(const Json& json, const JsonArgs& args,
+                      ValidationErrors* errors);
   };
 
-  XdsClusterResolverLbConfig(
-      std::vector<DiscoveryMechanism> discovery_mechanisms, Json xds_lb_policy)
-      : discovery_mechanisms_(std::move(discovery_mechanisms)),
-        xds_lb_policy_(std::move(xds_lb_policy)) {}
+  XdsClusterResolverLbConfig() = default;
+
+  XdsClusterResolverLbConfig(const XdsClusterResolverLbConfig&) = delete;
+  XdsClusterResolverLbConfig& operator=(const XdsClusterResolverLbConfig&) =
+      delete;
+
+  XdsClusterResolverLbConfig(XdsClusterResolverLbConfig&& other) = delete;
+  XdsClusterResolverLbConfig& operator=(XdsClusterResolverLbConfig&& other) =
+      delete;
+
+  absl::string_view name() const override { return kXdsClusterResolver; }
 
-  const char* name() const override { return kXdsClusterResolver; }
   const std::vector<DiscoveryMechanism>& discovery_mechanisms() const {
     return discovery_mechanisms_;
   }
 
   const Json& xds_lb_policy() const { return xds_lb_policy_; }
 
+  static const JsonLoaderInterface* JsonLoader(const JsonArgs&);
+  void JsonPostLoad(const Json& json, const JsonArgs& args,
+                    ValidationErrors* errors);
+
  private:
   std::vector<DiscoveryMechanism> discovery_mechanisms_;
-  Json xds_lb_policy_;
+  Json xds_lb_policy_ = Json::Object{{"ROUND_ROBIN", Json::Object()}};
 };
 
 // Xds Cluster Resolver LB policy.
@@ -133,9 +157,9 @@ class XdsClusterResolverLb : public LoadBalancingPolicy {
  public:
   XdsClusterResolverLb(RefCountedPtr<XdsClient> xds_client, Args args);
 
-  const char* name() const override { return kXdsClusterResolver; }
+  absl::string_view name() const override { return kXdsClusterResolver; }
 
-  void UpdateLocked(UpdateArgs args) override;
+  absl::Status UpdateLocked(UpdateArgs args) override;
   void ResetBackoffLocked() override;
   void ExitIdleLocked() override;
 
@@ -224,16 +248,44 @@ class XdsClusterResolverLb : public LoadBalancingPolicy {
       // in methods of this class rather than in lambdas to work around an MSVC
       // bug.
       void OnResourceChangedHelper(XdsEndpointResource update) {
+        std::string resolution_note;
+        if (update.priorities.empty()) {
+          resolution_note = absl::StrCat(
+              "EDS resource ", discovery_mechanism_->GetEdsResourceName(),
+              " contains no localities");
+        } else {
+          std::set<std::string> empty_localities;
+          for (const auto& priority : update.priorities) {
+            for (const auto& p : priority.localities) {
+              if (p.second.endpoints.empty()) {
+                empty_localities.insert(p.first->AsHumanReadableString());
+              }
+            }
+          }
+          if (!empty_localities.empty()) {
+            resolution_note = absl::StrCat(
+                "EDS resource ", discovery_mechanism_->GetEdsResourceName(),
+                " contains empty localities: [",
+                absl::StrJoin(empty_localities, "; "), "]");
+          }
+        }
         discovery_mechanism_->parent()->OnEndpointChanged(
-            discovery_mechanism_->index(), std::move(update));
+            discovery_mechanism_->index(), std::move(update),
+            std::move(resolution_note));
       }
       void OnErrorHelper(absl::Status status) {
-        discovery_mechanism_->parent()->OnError(discovery_mechanism_->index(),
-                                                status);
+        discovery_mechanism_->parent()->OnError(
+            discovery_mechanism_->index(),
+            absl::StrCat("EDS watcher error for resource ",
+                         discovery_mechanism_->GetEdsResourceName(), " (",
+                         status.ToString(), ")"));
       }
       void OnResourceDoesNotExistHelper() {
         discovery_mechanism_->parent()->OnResourceDoesNotExist(
-            discovery_mechanism_->index());
+            discovery_mechanism_->index(),
+            absl::StrCat("EDS resource ",
+                         discovery_mechanism_->GetEdsResourceName(),
+                         " does not exist"));
       }
       RefCountedPtr<EdsDiscoveryMechanism> discovery_mechanism_;
     };
@@ -288,6 +340,11 @@ class XdsClusterResolverLb : public LoadBalancingPolicy {
     // access protected member in base class.
     friend class ResolverResultHandler;
 
+    absl::string_view GetDnsHostname() const {
+      auto& config = parent()->config_->discovery_mechanisms()[index()];
+      return config.dns_hostname;
+    }
+
     OrphanablePtr<Resolver> resolver_;
   };
 
@@ -295,6 +352,8 @@ class XdsClusterResolverLb : public LoadBalancingPolicy {
     OrphanablePtr<DiscoveryMechanism> discovery_mechanism;
     // Most recent update reported by the discovery mechanism.
     absl::optional<XdsEndpointResource> latest_update;
+    // Last resolution note reported by the discovery mechanism, if any.
+    std::string resolution_note;
     // State used to retain child policy names for priority policy.
     std::vector<size_t /*child_number*/> priority_child_numbers;
     size_t next_available_child_number = 0;
@@ -317,7 +376,7 @@ class XdsClusterResolverLb : public LoadBalancingPolicy {
     }
 
     RefCountedPtr<SubchannelInterface> CreateSubchannel(
-        ServerAddress address, const grpc_channel_args& args) override;
+        ServerAddress address, const ChannelArgs& args) override;
     void UpdateState(grpc_connectivity_state state, const absl::Status& status,
                      std::unique_ptr<SubchannelPicker> picker) override;
     // This is a no-op, because we get the addresses from the xds
@@ -335,25 +394,26 @@ class XdsClusterResolverLb : public LoadBalancingPolicy {
 
   void ShutdownLocked() override;
 
-  void OnEndpointChanged(size_t index, XdsEndpointResource update);
-  void OnError(size_t index, absl::Status status);
-  void OnResourceDoesNotExist(size_t index);
+  void OnEndpointChanged(size_t index, XdsEndpointResource update,
+                         std::string resolution_note);
+  void OnError(size_t index, std::string resolution_note);
+  void OnResourceDoesNotExist(size_t index, std::string resolution_note);
 
   void MaybeDestroyChildPolicyLocked();
 
-  void UpdateChildPolicyLocked();
+  absl::Status UpdateChildPolicyLocked();
   OrphanablePtr<LoadBalancingPolicy> CreateChildPolicyLocked(
-      const grpc_channel_args* args);
+      const ChannelArgs& args);
   ServerAddressList CreateChildPolicyAddressesLocked();
+  std::string CreateChildPolicyResolutionNoteLocked();
   RefCountedPtr<Config> CreateChildPolicyConfigLocked();
-  grpc_channel_args* CreateChildPolicyArgsLocked(
-      const grpc_channel_args* args_in);
+  ChannelArgs CreateChildPolicyArgsLocked(const ChannelArgs& args_in);
 
   // The xds client and endpoint watcher.
   RefCountedPtr<XdsClient> xds_client_;
 
   // Current channel args and config from the resolver.
-  const grpc_channel_args* args_ = nullptr;
+  ChannelArgs args_;
   RefCountedPtr<XdsClusterResolverLbConfig> config_;
 
   // Internal state.
@@ -371,7 +431,7 @@ class XdsClusterResolverLb : public LoadBalancingPolicy {
 
 RefCountedPtr<SubchannelInterface>
 XdsClusterResolverLb::Helper::CreateSubchannel(ServerAddress address,
-                                               const grpc_channel_args& args) {
+                                               const ChannelArgs& args) {
   if (xds_cluster_resolver_policy_->shutting_down_) return nullptr;
   return xds_cluster_resolver_policy_->channel_control_helper()
       ->CreateSubchannel(std::move(address), args);
@@ -441,30 +501,26 @@ void XdsClusterResolverLb::EdsDiscoveryMechanism::Orphan() {
 //
 
 void XdsClusterResolverLb::LogicalDNSDiscoveryMechanism::Start() {
-  std::string target =
-      parent()->config_->discovery_mechanisms()[index()].dns_hostname;
-  grpc_channel_args* args = nullptr;
-  FakeResolverResponseGenerator* fake_resolver_response_generator =
-      grpc_channel_args_find_pointer<FakeResolverResponseGenerator>(
-          parent()->args_,
+  std::string target;
+  ChannelArgs args = parent()->args_;
+  auto* fake_resolver_response_generator =
+      args.GetPointer<FakeResolverResponseGenerator>(
           GRPC_ARG_XDS_LOGICAL_DNS_CLUSTER_FAKE_RESOLVER_RESPONSE_GENERATOR);
   if (fake_resolver_response_generator != nullptr) {
-    target = absl::StrCat("fake:", target);
-    grpc_arg new_arg = FakeResolverResponseGenerator::MakeChannelArg(
-        fake_resolver_response_generator);
-    args = grpc_channel_args_copy_and_add(parent()->args_, &new_arg, 1);
+    target = absl::StrCat("fake:", GetDnsHostname());
+    args = args.SetObject(fake_resolver_response_generator->Ref());
   } else {
-    target = absl::StrCat("dns:", target);
-    args = grpc_channel_args_copy(parent()->args_);
+    target = absl::StrCat("dns:", GetDnsHostname());
   }
   resolver_ = CoreConfiguration::Get().resolver_registry().CreateResolver(
       target.c_str(), args, parent()->interested_parties(),
       parent()->work_serializer(),
       absl::make_unique<ResolverResultHandler>(
           Ref(DEBUG_LOCATION, "LogicalDNSDiscoveryMechanism")));
-  grpc_channel_args_destroy(args);
   if (resolver_ == nullptr) {
-    parent()->OnResourceDoesNotExist(index());
+    parent()->OnResourceDoesNotExist(
+        index(),
+        absl::StrCat("error creating DNS resolver for ", GetDnsHostname()));
     return;
   }
   resolver_->StartLocked();
@@ -494,14 +550,18 @@ void XdsClusterResolverLb::LogicalDNSDiscoveryMechanism::Orphan() {
 
 void XdsClusterResolverLb::LogicalDNSDiscoveryMechanism::ResolverResultHandler::
     ReportResult(Resolver::Result result) {
+  XdsClusterResolverLb* lb_policy = discovery_mechanism_->parent();
+  size_t index = discovery_mechanism_->index();
   if (!result.addresses.ok()) {
-    discovery_mechanism_->parent()->OnError(discovery_mechanism_->index(),
-                                            result.addresses.status());
+    if (result.resolution_note.empty()) {
+      result.resolution_note = absl::StrCat(
+          "DNS resolution failed for ", discovery_mechanism_->GetDnsHostname(),
+          " (", result.addresses.status().ToString(), ")");
+    }
+    lb_policy->OnError(index, result.resolution_note);
     return;
   }
   // Convert resolver result to EDS update.
-  // TODO(roth): Figure out a way to pass resolution_note through to the
-  // child policy.
   XdsEndpointResource update;
   XdsEndpointResource::Priority::Locality locality;
   locality.name = MakeRefCounted<XdsLocalityName>("", "", "");
@@ -510,8 +570,8 @@ void XdsClusterResolverLb::LogicalDNSDiscoveryMechanism::ResolverResultHandler::
   XdsEndpointResource::Priority priority;
   priority.localities.emplace(locality.name.get(), std::move(locality));
   update.priorities.emplace_back(std::move(priority));
-  discovery_mechanism_->parent()->OnEndpointChanged(
-      discovery_mechanism_->index(), std::move(update));
+  lb_policy->OnEndpointChanged(index, std::move(update),
+                               std::move(result.resolution_note));
 }
 
 //
@@ -560,9 +620,7 @@ void XdsClusterResolverLb::ShutdownLocked() {
   MaybeDestroyChildPolicyLocked();
   discovery_mechanisms_.clear();
   xds_client_.reset(DEBUG_LOCATION, "XdsClusterResolverLb");
-  // Destroy channel args.
-  grpc_channel_args_destroy(args_);
-  args_ = nullptr;
+  args_ = ChannelArgs();
 }
 
 void XdsClusterResolverLb::MaybeDestroyChildPolicyLocked() {
@@ -573,20 +631,19 @@ void XdsClusterResolverLb::MaybeDestroyChildPolicyLocked() {
   }
 }
 
-void XdsClusterResolverLb::UpdateLocked(UpdateArgs args) {
+absl::Status XdsClusterResolverLb::UpdateLocked(UpdateArgs args) {
   if (GRPC_TRACE_FLAG_ENABLED(grpc_lb_xds_cluster_resolver_trace)) {
     gpr_log(GPR_INFO, "[xds_cluster_resolver_lb %p] Received update", this);
   }
-  const bool is_initial_update = args_ == nullptr;
+  const bool is_initial_update = args_ == ChannelArgs();
   // Update config.
   auto old_config = std::move(config_);
   config_ = std::move(args.config);
   // Update args.
-  grpc_channel_args_destroy(args_);
-  args_ = args.args;
-  args.args = nullptr;
+  args_ = std::move(args.args);
   // Update child policy if needed.
-  if (child_policy_ != nullptr) UpdateChildPolicyLocked();
+  absl::Status status;
+  if (child_policy_ != nullptr) status = UpdateChildPolicyLocked();
   // Create endpoint watcher if needed.
   if (is_initial_update) {
     for (const auto& config : config_->discovery_mechanisms()) {
@@ -612,6 +669,7 @@ void XdsClusterResolverLb::UpdateLocked(UpdateArgs args) {
       discovery_mechanism.discovery_mechanism->Start();
     }
   }
+  return status;
 }
 
 void XdsClusterResolverLb::ResetBackoffLocked() {
@@ -625,13 +683,14 @@ void XdsClusterResolverLb::ExitIdleLocked() {
 }
 
 void XdsClusterResolverLb::OnEndpointChanged(size_t index,
-                                             XdsEndpointResource update) {
+                                             XdsEndpointResource update,
+                                             std::string resolution_note) {
   if (shutting_down_) return;
   if (GRPC_TRACE_FLAG_ENABLED(grpc_lb_xds_cluster_resolver_trace)) {
     gpr_log(GPR_INFO,
             "[xds_cluster_resolver_lb %p] Received update from xds client"
-            " for discovery mechanism %" PRIuPTR "",
-            this, index);
+            " for discovery mechanism %" PRIuPTR " (resolution_note=\"%s\")",
+            this, index, resolution_note.c_str());
   }
   DiscoveryMechanismEntry& discovery_entry = discovery_mechanisms_[index];
   // We need at least one priority for each discovery mechanism, just so that we
@@ -704,6 +763,7 @@ void XdsClusterResolverLb::OnEndpointChanged(size_t index,
   }
   // Save update.
   discovery_entry.latest_update = std::move(update);
+  discovery_entry.resolution_note = std::move(resolution_note);
   discovery_entry.priority_child_numbers = std::move(priority_child_numbers);
   // If any discovery mechanism has not received its first update,
   // wait until that happens before creating the child policy.
@@ -718,30 +778,33 @@ void XdsClusterResolverLb::OnEndpointChanged(size_t index,
     if (!mechanism.latest_update.has_value()) return;
   }
   // Update child policy.
-  UpdateChildPolicyLocked();
+  // TODO(roth): If the child policy reports an error with the update,
+  // we need to propagate that error back to the resolver somehow.
+  (void)UpdateChildPolicyLocked();
 }
 
-void XdsClusterResolverLb::OnError(size_t index, absl::Status status) {
+void XdsClusterResolverLb::OnError(size_t index, std::string resolution_note) {
   gpr_log(GPR_ERROR,
           "[xds_cluster_resolver_lb %p] discovery mechanism %" PRIuPTR
-          " xds watcher reported error: %s",
-          this, index, status.ToString().c_str());
+          " reported error: %s",
+          this, index, resolution_note.c_str());
   if (shutting_down_) return;
   if (!discovery_mechanisms_[index].latest_update.has_value()) {
-    // Call OnEndpointChanged with an empty update just like
-    // OnResourceDoesNotExist.
-    OnEndpointChanged(index, XdsEndpointResource());
+    // Call OnEndpointChanged() with an empty update just like
+    // OnResourceDoesNotExist().
+    OnEndpointChanged(index, XdsEndpointResource(), std::move(resolution_note));
   }
 }
 
-void XdsClusterResolverLb::OnResourceDoesNotExist(size_t index) {
+void XdsClusterResolverLb::OnResourceDoesNotExist(size_t index,
+                                                  std::string resolution_note) {
   gpr_log(GPR_ERROR,
           "[xds_cluster_resolver_lb %p] discovery mechanism %" PRIuPTR
-          " resource does not exist",
-          this, index);
+          " resource does not exist: %s",
+          this, index, resolution_note.c_str());
   if (shutting_down_) return;
-  // Call OnEndpointChanged with an empty update.
-  OnEndpointChanged(index, XdsEndpointResource());
+  // Call OnEndpointChanged() with an empty update.
+  OnEndpointChanged(index, XdsEndpointResource(), std::move(resolution_note));
 }
 
 //
@@ -790,6 +853,16 @@ ServerAddressList XdsClusterResolverLb::CreateChildPolicyAddressesLocked() {
   return addresses;
 }
 
+std::string XdsClusterResolverLb::CreateChildPolicyResolutionNoteLocked() {
+  std::vector<absl::string_view> resolution_notes;
+  for (const auto& discovery_entry : discovery_mechanisms_) {
+    if (!discovery_entry.resolution_note.empty()) {
+      resolution_notes.push_back(discovery_entry.resolution_note);
+    }
+  }
+  return absl::StrJoin(resolution_notes, "; ");
+}
+
 RefCountedPtr<LoadBalancingPolicy::Config>
 XdsClusterResolverLb::CreateChildPolicyConfigLocked() {
   Json::Object priority_children;
@@ -885,10 +958,6 @@ XdsClusterResolverLb::CreateChildPolicyConfigLocked() {
         if (discovery_entry.config().outlier_detection_lb_config.has_value()) {
           outlier_detection_config =
               discovery_entry.config().outlier_detection_lb_config.value();
-        } else {
-          // outlier detection will be a no-op
-          outlier_detection_config["interval"] =
-              Duration::Infinity().ToJsonString();
         }
         outlier_detection_config["childPolicy"] = Json::Array{Json::Object{
             {"xds_cluster_impl_experimental",
@@ -930,17 +999,17 @@ XdsClusterResolverLb::CreateChildPolicyConfigLocked() {
         "[xds_cluster_resolver_lb %p] generated config for child policy: %s",
         this, json_str.c_str());
   }
-  grpc_error_handle error = GRPC_ERROR_NONE;
-  RefCountedPtr<LoadBalancingPolicy::Config> config =
-      LoadBalancingPolicyRegistry::ParseLoadBalancingConfig(json, &error);
-  if (!GRPC_ERROR_IS_NONE(error)) {
+  auto config =
+      CoreConfiguration::Get().lb_policy_registry().ParseLoadBalancingConfig(
+          json);
+  if (!config.ok()) {
     // This should never happen, but if it does, we basically have no
     // way to fix it, so we put the channel in TRANSIENT_FAILURE.
     gpr_log(GPR_ERROR,
             "[xds_cluster_resolver_lb %p] error parsing generated child policy "
             "config -- "
             "will put channel in TRANSIENT_FAILURE: %s",
-            this, grpc_error_std_string(error).c_str());
+            this, config.status().ToString().c_str());
     absl::Status status = absl::InternalError(
         "xds_cluster_resolver LB policy: error parsing generated child policy "
         "config");
@@ -949,15 +1018,16 @@ XdsClusterResolverLb::CreateChildPolicyConfigLocked() {
         absl::make_unique<TransientFailurePicker>(status));
     return nullptr;
   }
-  return config;
+  return std::move(*config);
 }
 
-void XdsClusterResolverLb::UpdateChildPolicyLocked() {
-  if (shutting_down_) return;
+absl::Status XdsClusterResolverLb::UpdateChildPolicyLocked() {
+  if (shutting_down_) return absl::OkStatus();
   UpdateArgs update_args;
   update_args.config = CreateChildPolicyConfigLocked();
-  if (update_args.config == nullptr) return;
+  if (update_args.config == nullptr) return absl::OkStatus();
   update_args.addresses = CreateChildPolicyAddressesLocked();
+  update_args.resolution_note = CreateChildPolicyResolutionNoteLocked();
   update_args.args = CreateChildPolicyArgsLocked(args_);
   if (child_policy_ == nullptr) {
     child_policy_ = CreateChildPolicyLocked(update_args.args);
@@ -966,29 +1036,25 @@ void XdsClusterResolverLb::UpdateChildPolicyLocked() {
     gpr_log(GPR_INFO, "[xds_cluster_resolver_lb %p] Updating child policy %p",
             this, child_policy_.get());
   }
-  child_policy_->UpdateLocked(std::move(update_args));
+  return child_policy_->UpdateLocked(std::move(update_args));
 }
 
-grpc_channel_args* XdsClusterResolverLb::CreateChildPolicyArgsLocked(
-    const grpc_channel_args* args) {
-  absl::InlinedVector<grpc_arg, 2> new_args = {
-      // Inhibit client-side health checking, since the balancer does this
-      // for us.
-      grpc_channel_arg_integer_create(
-          const_cast<char*>(GRPC_ARG_INHIBIT_HEALTH_CHECKING), 1),
-  };
-  return grpc_channel_args_copy_and_add(args, new_args.data(), new_args.size());
+ChannelArgs XdsClusterResolverLb::CreateChildPolicyArgsLocked(
+    const ChannelArgs& args) {
+  // Inhibit client-side health checking, since the balancer does this
+  // for us.
+  return args.Set(GRPC_ARG_INHIBIT_HEALTH_CHECKING, 1);
 }
 
 OrphanablePtr<LoadBalancingPolicy>
-XdsClusterResolverLb::CreateChildPolicyLocked(const grpc_channel_args* args) {
+XdsClusterResolverLb::CreateChildPolicyLocked(const ChannelArgs& args) {
   LoadBalancingPolicy::Args lb_policy_args;
   lb_policy_args.work_serializer = work_serializer();
   lb_policy_args.args = args;
   lb_policy_args.channel_control_helper =
       absl::make_unique<Helper>(Ref(DEBUG_LOCATION, "Helper"));
   OrphanablePtr<LoadBalancingPolicy> lb_policy =
-      LoadBalancingPolicyRegistry::CreateLoadBalancingPolicy(
+      CoreConfiguration::Get().lb_policy_registry().CreateLoadBalancingPolicy(
           "priority_experimental", std::move(lb_policy_args));
   if (GPR_UNLIKELY(lb_policy == nullptr)) {
     gpr_log(GPR_ERROR,
@@ -1012,225 +1078,158 @@ XdsClusterResolverLb::CreateChildPolicyLocked(const grpc_channel_args* args) {
 // factory
 //
 
-class XdsClusterResolverLbFactory : public LoadBalancingPolicyFactory {
- public:
-  OrphanablePtr<LoadBalancingPolicy> CreateLoadBalancingPolicy(
-      LoadBalancingPolicy::Args args) const override {
-    RefCountedPtr<XdsClient> xds_client =
-        XdsClient::GetFromChannelArgs(*args.args);
-    if (xds_client == nullptr) {
-      gpr_log(GPR_ERROR,
-              "XdsClient not present in channel args -- cannot instantiate "
-              "xds_cluster_resolver LB policy");
-      return nullptr;
+const JsonLoaderInterface*
+XdsClusterResolverLbConfig::DiscoveryMechanism::JsonLoader(const JsonArgs&) {
+  static const auto* loader =
+      JsonObjectLoader<DiscoveryMechanism>()
+          // Note: Several fields requires custom processing,
+          // so they are handled in JsonPostLoad() instead.
+          .Field("clusterName", &DiscoveryMechanism::cluster_name)
+          .OptionalField("lrsLoadReportingServer",
+                         &DiscoveryMechanism::lrs_load_reporting_server)
+          .OptionalField("max_concurrent_requests",
+                         &DiscoveryMechanism::max_concurrent_requests)
+          .OptionalField("outlierDetection",
+                         &DiscoveryMechanism::outlier_detection_lb_config,
+                         "outlier_detection")
+          .Finish();
+  return loader;
+}
+
+void XdsClusterResolverLbConfig::DiscoveryMechanism::JsonPostLoad(
+    const Json& json, const JsonArgs& args, ValidationErrors* errors) {
+  // Parse "type".
+  {
+    auto type_field = LoadJsonObjectField<std::string>(json.object_value(),
+                                                       args, "type", errors);
+    if (type_field.has_value()) {
+      if (*type_field == "EDS") {
+        type = DiscoveryMechanismType::EDS;
+      } else if (*type_field == "LOGICAL_DNS") {
+        type = DiscoveryMechanismType::LOGICAL_DNS;
+      } else {
+        ValidationErrors::ScopedField field(errors, ".type");
+        errors->AddError(absl::StrCat("unknown type \"", *type_field, "\""));
+      }
     }
-    return MakeOrphanable<XdsClusterResolverChildHandler>(std::move(xds_client),
-                                                          std::move(args));
   }
+  // Parse "edsServiceName" if type is EDS.
+  if (type == DiscoveryMechanismType::EDS) {
+    auto value = LoadJsonObjectField<std::string>(json.object_value(), args,
+                                                  "edsServiceName", errors,
+                                                  /*required=*/false);
+    if (value.has_value()) eds_service_name = std::move(*value);
+  }
+  // Parse "dnsHostname" if type is LOGICAL_DNS.
+  if (type == DiscoveryMechanismType::LOGICAL_DNS) {
+    auto value = LoadJsonObjectField<std::string>(json.object_value(), args,
+                                                  "dnsHostname", errors);
+    if (value.has_value()) dns_hostname = std::move(*value);
+  }
+}
 
-  const char* name() const override { return kXdsClusterResolver; }
+const JsonLoaderInterface* XdsClusterResolverLbConfig::JsonLoader(
+    const JsonArgs&) {
+  static const auto* loader =
+      JsonObjectLoader<XdsClusterResolverLbConfig>()
+          // Note: The "xdsLbPolicy" field requires custom processing,
+          // so it's handled in JsonPostLoad() instead.
+          .Field("discoveryMechanisms",
+                 &XdsClusterResolverLbConfig::discovery_mechanisms_)
+          .Finish();
+  return loader;
+}
 
-  RefCountedPtr<LoadBalancingPolicy::Config> ParseLoadBalancingConfig(
-      const Json& json, grpc_error_handle* error) const override {
-    GPR_DEBUG_ASSERT(error != nullptr && GRPC_ERROR_IS_NONE(*error));
-    if (json.type() == Json::Type::JSON_NULL) {
-      // xds_cluster_resolver was mentioned as a policy in the deprecated
-      // loadBalancingPolicy field or in the client API.
-      *error = GRPC_ERROR_CREATE_FROM_STATIC_STRING(
-          "field:loadBalancingPolicy error:xds_cluster_resolver policy "
-          "requires configuration. "
-          "Please use loadBalancingConfig field of service config instead.");
-      return nullptr;
-    }
-    std::vector<grpc_error_handle> error_list;
-    std::vector<XdsClusterResolverLbConfig::DiscoveryMechanism>
-        discovery_mechanisms;
-    auto it = json.object_value().find("discoveryMechanisms");
-    if (it == json.object_value().end()) {
-      error_list.push_back(GRPC_ERROR_CREATE_FROM_STATIC_STRING(
-          "field:discoveryMechanisms error:required field missing"));
-    } else if (it->second.type() != Json::Type::ARRAY) {
-      error_list.push_back(GRPC_ERROR_CREATE_FROM_STATIC_STRING(
-          "field:discoveryMechanisms error:type should be array"));
-    } else {
-      const Json::Array& array = it->second.array_value();
-      for (size_t i = 0; i < array.size(); ++i) {
-        XdsClusterResolverLbConfig::DiscoveryMechanism discovery_mechanism;
-        std::vector<grpc_error_handle> discovery_mechanism_errors =
-            ParseDiscoveryMechanism(array[i], &discovery_mechanism);
-        if (!discovery_mechanism_errors.empty()) {
-          grpc_error_handle error = GRPC_ERROR_CREATE_FROM_CPP_STRING(
-              absl::StrCat("field:discovery_mechanism element: ", i, " error"));
-          for (const grpc_error_handle& discovery_mechanism_error :
-               discovery_mechanism_errors) {
-            error = grpc_error_add_child(error, discovery_mechanism_error);
-          }
-          error_list.push_back(error);
-        }
-        discovery_mechanisms.emplace_back(std::move(discovery_mechanism));
-      }
-    }
-    if (discovery_mechanisms.empty()) {
-      error_list.push_back(GRPC_ERROR_CREATE_FROM_STATIC_STRING(
-          "field:discovery_mechanism error:list is missing or empty"));
+void XdsClusterResolverLbConfig::JsonPostLoad(const Json& json,
+                                              const JsonArgs& args,
+                                              ValidationErrors* errors) {
+  // Validate discoveryMechanisms.
+  {
+    ValidationErrors::ScopedField field(errors, ".discoveryMechanisms");
+    if (!errors->FieldHasErrors() && discovery_mechanisms_.empty()) {
+      errors->AddError("must be non-empty");
     }
-    Json xds_lb_policy = Json::Object{
-        {"ROUND_ROBIN", Json::Object()},
-    };
-    it = json.object_value().find("xdsLbPolicy");
+  }
+  // Parse "xdsLbPolicy".
+  {
+    ValidationErrors::ScopedField field(errors, ".xdsLbPolicy");
+    auto it = json.object_value().find("xdsLbPolicy");
     if (it != json.object_value().end()) {
       if (it->second.type() != Json::Type::ARRAY) {
-        error_list.push_back(GRPC_ERROR_CREATE_FROM_STATIC_STRING(
-            "field:xdsLbPolicy error:type should be array"));
+        errors->AddError("is not an array");
       } else {
         const Json::Array& array = it->second.array_value();
         for (size_t i = 0; i < array.size(); ++i) {
+          ValidationErrors::ScopedField field(errors,
+                                              absl::StrCat("[", i, "]"));
           if (array[i].type() != Json::Type::OBJECT) {
-            error_list.push_back(GRPC_ERROR_CREATE_FROM_STATIC_STRING(
-                "field:xdsLbPolicy error:element should be of type object"));
+            errors->AddError("is not an object");
             continue;
           }
           const Json::Object& policy = array[i].object_value();
           auto policy_it = policy.find("ROUND_ROBIN");
           if (policy_it != policy.end()) {
+            ValidationErrors::ScopedField field(errors, "[\"ROUND_ROBIN\"]");
             if (policy_it->second.type() != Json::Type::OBJECT) {
-              error_list.push_back(GRPC_ERROR_CREATE_FROM_STATIC_STRING(
-                  "field:ROUND_ROBIN error:type should be object"));
+              errors->AddError("is not an object");
             }
             break;
           }
-          policy_it = policy.find("RING_HASH");
-          if (policy_it != policy.end()) {
-            xds_lb_policy = array[i];
-            size_t min_ring_size;
-            size_t max_ring_size;
-            ParseRingHashLbConfig(policy_it->second, &min_ring_size,
-                                  &max_ring_size, &error_list);
+          {
+            ValidationErrors::ScopedField field(errors, "[\"RING_HASH\"]");
+            policy_it = policy.find("RING_HASH");
+            if (policy_it != policy.end()) {
+              LoadFromJson<RingHashConfig>(policy_it->second, args, errors);
+              xds_lb_policy_ = array[i];
+            }
           }
         }
       }
     }
-    // Construct config.
-    if (error_list.empty()) {
-      return MakeRefCounted<XdsClusterResolverLbConfig>(
-          std::move(discovery_mechanisms), std::move(xds_lb_policy));
-    } else {
-      *error = GRPC_ERROR_CREATE_FROM_VECTOR(
-          "xds_cluster_resolver_experimental LB policy config", &error_list);
+  }
+}
+
+class XdsClusterResolverLbFactory : public LoadBalancingPolicyFactory {
+ public:
+  OrphanablePtr<LoadBalancingPolicy> CreateLoadBalancingPolicy(
+      LoadBalancingPolicy::Args args) const override {
+    auto xds_client = args.args.GetObjectRef<GrpcXdsClient>(
+        DEBUG_LOCATION, "XdsClusterResolverLbFactory");
+    if (xds_client == nullptr) {
+      gpr_log(GPR_ERROR,
+              "XdsClient not present in channel args -- cannot instantiate "
+              "xds_cluster_resolver LB policy");
       return nullptr;
     }
+    return MakeOrphanable<XdsClusterResolverChildHandler>(std::move(xds_client),
+                                                          std::move(args));
   }
 
- private:
-  static std::vector<grpc_error_handle> ParseDiscoveryMechanism(
-      const Json& json,
-      XdsClusterResolverLbConfig::DiscoveryMechanism* discovery_mechanism) {
-    std::vector<grpc_error_handle> error_list;
-    if (json.type() != Json::Type::OBJECT) {
-      error_list.push_back(GRPC_ERROR_CREATE_FROM_STATIC_STRING(
-          "value should be of type object"));
-      return error_list;
-    }
-    // Cluster name.
-    auto it = json.object_value().find("clusterName");
-    if (it == json.object_value().end()) {
-      error_list.push_back(GRPC_ERROR_CREATE_FROM_STATIC_STRING(
-          "field:clusterName error:required field missing"));
-    } else if (it->second.type() != Json::Type::STRING) {
-      error_list.push_back(GRPC_ERROR_CREATE_FROM_STATIC_STRING(
-          "field:clusterName error:type should be string"));
-    } else {
-      discovery_mechanism->cluster_name = it->second.string_value();
-    }
-    // LRS load reporting server name.
-    it = json.object_value().find("lrsLoadReportingServer");
-    if (it != json.object_value().end()) {
-      if (it->second.type() != Json::Type::OBJECT) {
-        error_list.push_back(GRPC_ERROR_CREATE_FROM_STATIC_STRING(
-            "field:lrsLoadReportingServer error:type should be object"));
-      } else {
-        grpc_error_handle parse_error;
-        discovery_mechanism->lrs_load_reporting_server.emplace(
-            XdsBootstrap::XdsServer::Parse(it->second, &parse_error));
-        if (!GRPC_ERROR_IS_NONE(parse_error)) {
-          error_list.push_back(GRPC_ERROR_CREATE_FROM_CPP_STRING(
-              absl::StrCat("errors parsing lrs_load_reporting_server")));
-          error_list.push_back(parse_error);
-        }
-      }
-    }
-    // Max concurrent requests.
-    discovery_mechanism->max_concurrent_requests = 1024;
-    it = json.object_value().find("max_concurrent_requests");
-    if (it != json.object_value().end()) {
-      if (it->second.type() != Json::Type::NUMBER) {
-        error_list.push_back(GRPC_ERROR_CREATE_FROM_STATIC_STRING(
-            "field:max_concurrent_requests error:must be of type number"));
-      } else {
-        discovery_mechanism->max_concurrent_requests =
-            gpr_parse_nonnegative_int(it->second.string_value().c_str());
-      }
-    }
-    if (XdsOutlierDetectionEnabled()) {
-      it = json.object_value().find("outlierDetection");
-      if (it != json.object_value().end()) {
-        if (it->second.type() != Json::Type::OBJECT) {
-          error_list.push_back(GRPC_ERROR_CREATE_FROM_STATIC_STRING(
-              "field:outlierDetection error:type should be object"));
-        } else {
-          // No need to validate the contents of the outlier detection config,
-          // because in this particular case, the JSON is generated by the CDS
-          // policy instead of coming from service config, so it's not actually
-          // any better to catch the problem here than it is to catch it in the
-          // outlier_detection policy itself, so here we just act as a
-          // pass-through.
-          discovery_mechanism->outlier_detection_lb_config =
-              it->second.object_value();
-        }
-      }
+  absl::string_view name() const override { return kXdsClusterResolver; }
+
+  absl::StatusOr<RefCountedPtr<LoadBalancingPolicy::Config>>
+  ParseLoadBalancingConfig(const Json& json) const override {
+    if (json.type() == Json::Type::JSON_NULL) {
+      // xds_cluster_resolver was mentioned as a policy in the deprecated
+      // loadBalancingPolicy field or in the client API.
+      return absl::InvalidArgumentError(
+          "field:loadBalancingPolicy error:xds_cluster_resolver policy "
+          "requires configuration. "
+          "Please use loadBalancingConfig field of service config instead.");
     }
-    // Discovery Mechanism type
-    it = json.object_value().find("type");
-    if (it == json.object_value().end()) {
-      error_list.push_back(GRPC_ERROR_CREATE_FROM_STATIC_STRING(
-          "field:type error:required field missing"));
-    } else if (it->second.type() != Json::Type::STRING) {
-      error_list.push_back(GRPC_ERROR_CREATE_FROM_STATIC_STRING(
-          "field:type error:type should be string"));
-    } else {
-      if (it->second.string_value() == "EDS") {
-        discovery_mechanism->type = XdsClusterResolverLbConfig::
-            DiscoveryMechanism::DiscoveryMechanismType::EDS;
-        it = json.object_value().find("edsServiceName");
-        if (it != json.object_value().end()) {
-          if (it->second.type() != Json::Type::STRING) {
-            error_list.push_back(GRPC_ERROR_CREATE_FROM_STATIC_STRING(
-                "field:edsServiceName error:type should be string"));
-          } else {
-            discovery_mechanism->eds_service_name = it->second.string_value();
-          }
-        }
-      } else if (it->second.string_value() == "LOGICAL_DNS") {
-        discovery_mechanism->type = XdsClusterResolverLbConfig::
-            DiscoveryMechanism::DiscoveryMechanismType::LOGICAL_DNS;
-        it = json.object_value().find("dnsHostname");
-        if (it == json.object_value().end()) {
-          error_list.push_back(GRPC_ERROR_CREATE_FROM_STATIC_STRING(
-              "field:dnsHostname error:required field missing"));
-        } else if (it->second.type() != Json::Type::STRING) {
-          error_list.push_back(GRPC_ERROR_CREATE_FROM_STATIC_STRING(
-              "field:dnsHostname error:type should be string"));
-        } else {
-          discovery_mechanism->dns_hostname = it->second.string_value();
-        }
-      } else {
-        error_list.push_back(GRPC_ERROR_CREATE_FROM_STATIC_STRING(
-            "field:type error:invalid type"));
+    class XdsJsonArgs : public JsonArgs {
+     public:
+      bool IsEnabled(absl::string_view key) const override {
+        if (key == "outlier_detection") return XdsOutlierDetectionEnabled();
+        return true;
       }
-    }
-    return error_list;
+    };
+    return LoadRefCountedFromJson<XdsClusterResolverLbConfig>(
+        json, XdsJsonArgs(),
+        "errors validating xds_cluster_resolver LB policy config");
   }
 
+ private:
   class XdsClusterResolverChildHandler : public ChildPolicyHandler {
    public:
     XdsClusterResolverChildHandler(RefCountedPtr<XdsClient> xds_client,
@@ -1239,6 +1238,10 @@ class XdsClusterResolverLbFactory : public LoadBalancingPolicyFactory {
                              &grpc_lb_xds_cluster_resolver_trace),
           xds_client_(std::move(xds_client)) {}
 
+    ~XdsClusterResolverChildHandler() override {
+      xds_client_.reset(DEBUG_LOCATION, "XdsClusterResolverChildHandler");
+    }
+
     bool ConfigChangeRequiresNewPolicyInstance(
         LoadBalancingPolicy::Config* old_config,
         LoadBalancingPolicy::Config* new_config) const override {
@@ -1248,13 +1251,38 @@ class XdsClusterResolverLbFactory : public LoadBalancingPolicyFactory {
           static_cast<XdsClusterResolverLbConfig*>(old_config);
       XdsClusterResolverLbConfig* new_xds_cluster_resolver_config =
           static_cast<XdsClusterResolverLbConfig*>(new_config);
-      return old_xds_cluster_resolver_config->discovery_mechanisms() !=
-             new_xds_cluster_resolver_config->discovery_mechanisms();
+      if (old_xds_cluster_resolver_config->discovery_mechanisms().size() !=
+          new_xds_cluster_resolver_config->discovery_mechanisms().size()) {
+        return true;
+      }
+      for (size_t i = 0;
+           i < old_xds_cluster_resolver_config->discovery_mechanisms().size();
+           ++i) {
+        auto& old_discovery_mechanism =
+            old_xds_cluster_resolver_config->discovery_mechanisms()[i];
+        auto& new_discovery_mechanism =
+            new_xds_cluster_resolver_config->discovery_mechanisms()[i];
+        if (old_discovery_mechanism.type != new_discovery_mechanism.type ||
+            old_discovery_mechanism.cluster_name !=
+                new_discovery_mechanism.cluster_name ||
+            old_discovery_mechanism.eds_service_name !=
+                new_discovery_mechanism.eds_service_name ||
+            old_discovery_mechanism.dns_hostname !=
+                new_discovery_mechanism.dns_hostname ||
+            !(old_discovery_mechanism.lrs_load_reporting_server ==
+              new_discovery_mechanism.lrs_load_reporting_server)) {
+          return true;
+        }
+      }
+      return false;
     }
 
     OrphanablePtr<LoadBalancingPolicy> CreateLoadBalancingPolicy(
-        const char* /*name*/, LoadBalancingPolicy::Args args) const override {
-      return MakeOrphanable<XdsClusterResolverLb>(xds_client_, std::move(args));
+        absl::string_view /*name*/,
+        LoadBalancingPolicy::Args args) const override {
+      return MakeOrphanable<XdsClusterResolverLb>(
+          xds_client_->Ref(DEBUG_LOCATION, "XdsClusterResolverLb"),
+          std::move(args));
     }
 
    private:
@@ -1264,16 +1292,9 @@ class XdsClusterResolverLbFactory : public LoadBalancingPolicyFactory {
 
 }  // namespace
 
-}  // namespace grpc_core
-
-//
-// Plugin registration
-//
-
-void grpc_lb_policy_xds_cluster_resolver_init() {
-  grpc_core::LoadBalancingPolicyRegistry::Builder::
-      RegisterLoadBalancingPolicyFactory(
-          absl::make_unique<grpc_core::XdsClusterResolverLbFactory>());
+void RegisterXdsClusterResolverLbPolicy(CoreConfiguration::Builder* builder) {
+  builder->lb_policy_registry()->RegisterLoadBalancingPolicyFactory(
+      absl::make_unique<XdsClusterResolverLbFactory>());
 }
 
-void grpc_lb_policy_xds_cluster_resolver_shutdown() {}
+}  // namespace grpc_core