grpc 1.49.0.pre1 → 1.50.0.pre1

data/src/core/ext/filters/client_channel/lb_policy/grpclb/grpclb.cc

@@ -114,6 +114,7 @@
 #include "src/core/lib/gprpp/ref_counted.h"
 #include "src/core/lib/gprpp/ref_counted_ptr.h"
 #include "src/core/lib/gprpp/time.h"
+#include "src/core/lib/gprpp/validation_errors.h"
 #include "src/core/lib/gprpp/work_serializer.h"
 #include "src/core/lib/iomgr/closure.h"
 #include "src/core/lib/iomgr/error.h"
@@ -124,6 +125,8 @@
 #include "src/core/lib/iomgr/socket_utils.h"
 #include "src/core/lib/iomgr/timer.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"
@@ -164,10 +167,47 @@ constexpr absl::string_view kGrpclb = "grpclb";
 
 class GrpcLbConfig : public LoadBalancingPolicy::Config {
  public:
-  GrpcLbConfig(RefCountedPtr<LoadBalancingPolicy::Config> child_policy,
-               std::string service_name)
-      : child_policy_(std::move(child_policy)),
-        service_name_(std::move(service_name)) {}
+  GrpcLbConfig() = default;
+
+  GrpcLbConfig(const GrpcLbConfig&) = delete;
+  GrpcLbConfig& operator=(const GrpcLbConfig&) = delete;
+
+  GrpcLbConfig(GrpcLbConfig&& other) = delete;
+  GrpcLbConfig& operator=(GrpcLbConfig&& other) = delete;
+
+  static const JsonLoaderInterface* JsonLoader(const JsonArgs&) {
+    static const auto* loader =
+        JsonObjectLoader<GrpcLbConfig>()
+            // Note: "childPolicy" field requires custom parsing, so
+            // it's handled in JsonPostLoad() instead.
+            .OptionalField("serviceName", &GrpcLbConfig::service_name_)
+            .Finish();
+    return loader;
+  }
+
+  void JsonPostLoad(const Json& json, const JsonArgs&,
+                    ValidationErrors* errors) {
+    ValidationErrors::ScopedField field(errors, ".childPolicy");
+    Json child_policy_config_json_tmp;
+    const Json* child_policy_config_json;
+    auto it = json.object_value().find("childPolicy");
+    if (it == json.object_value().end()) {
+      child_policy_config_json_tmp = Json::Array{Json::Object{
+          {"round_robin", Json::Object()},
+      }};
+      child_policy_config_json = &child_policy_config_json_tmp;
+    } else {
+      child_policy_config_json = &it->second;
+    }
+    auto child_policy_config =
+        CoreConfiguration::Get().lb_policy_registry().ParseLoadBalancingConfig(
+            *child_policy_config_json);
+    if (!child_policy_config.ok()) {
+      errors->AddError(child_policy_config.status().message());
+      return;
+    }
+    child_policy_ = std::move(*child_policy_config);
+  }
 
   absl::string_view name() const override { return kGrpclb; }
 
@@ -188,7 +228,7 @@ class GrpcLb : public LoadBalancingPolicy {
 
   absl::string_view name() const override { return kGrpclb; }
 
-  void UpdateLocked(UpdateArgs args) override;
+  absl::Status UpdateLocked(UpdateArgs args) override;
   void ResetBackoffLocked() override;
 
  private:
@@ -219,7 +259,6 @@ class GrpcLb : public LoadBalancingPolicy {
     void SendClientLoadReportLocked();
 
     // EventEngine callbacks
-    void MaybeSendClientLoadReport();
     void MaybeSendClientLoadReportLocked();
 
     static void ClientLoadReportDone(void* arg, grpc_error_handle error);
@@ -473,7 +512,7 @@ class GrpcLb : public LoadBalancingPolicy {
   void ShutdownLocked() override;
 
   // Helper functions used in UpdateLocked().
-  void UpdateBalancerChannelLocked(const ChannelArgs& args);
+  absl::Status UpdateBalancerChannelLocked(const ChannelArgs& args);
 
   void CancelBalancerChannelConnectivityWatchLocked();
 
@@ -868,7 +907,7 @@ GrpcLb::BalancerCallState::BalancerCallState(
   const Timestamp deadline =
       grpclb_policy()->lb_call_timeout_ == Duration::Zero()
           ? Timestamp::InfFuture()
-          : ExecCtx::Get()->Now() + grpclb_policy()->lb_call_timeout_;
+          : Timestamp::Now() + grpclb_policy()->lb_call_timeout_;
   lb_call_ = grpc_channel_create_pollset_set_call(
       grpclb_policy()->lb_channel_, nullptr, GRPC_PROPAGATE_DEFAULTS,
       grpclb_policy_->interested_parties(),
@@ -996,15 +1035,11 @@ void GrpcLb::BalancerCallState::ScheduleNextClientLoadReportLocked() {
       GetDefaultEventEngine()->RunAfter(client_stats_report_interval_, [this] {
         ApplicationCallbackExecCtx callback_exec_ctx;
         ExecCtx exec_ctx;
-        MaybeSendClientLoadReport();
+        grpclb_policy()->work_serializer()->Run(
+            [this] { MaybeSendClientLoadReportLocked(); }, DEBUG_LOCATION);
       });
 }
 
-void GrpcLb::BalancerCallState::MaybeSendClientLoadReport() {
-  grpclb_policy()->work_serializer()->Run(
-      [this] { MaybeSendClientLoadReportLocked(); }, DEBUG_LOCATION);
-}
-
 void GrpcLb::BalancerCallState::MaybeSendClientLoadReportLocked() {
   client_load_report_handle_.reset();
   if (this != grpclb_policy()->lb_calld_.get()) {
@@ -1358,6 +1393,47 @@ ServerAddressList ExtractBalancerAddresses(const ChannelArgs& args) {
 ChannelArgs BuildBalancerChannelArgs(
     FakeResolverResponseGenerator* response_generator,
     const ChannelArgs& args) {
+  ChannelArgs grpclb_channel_args;
+  const grpc_channel_args* lb_channel_specific_args =
+      args.GetPointer<grpc_channel_args>(
+          GRPC_ARG_EXPERIMENTAL_GRPCLB_CHANNEL_ARGS);
+  if (lb_channel_specific_args != nullptr) {
+    grpclb_channel_args = ChannelArgs::FromC(lb_channel_specific_args);
+  } else {
+    // Set grpclb_channel_args based on the parent channel's channel args.
+    grpclb_channel_args =
+        args
+            // LB policy name, since we want to use the default (pick_first) in
+            // the LB channel.
+            .Remove(GRPC_ARG_LB_POLICY_NAME)
+            // Strip out the service config, since we don't want the LB policy
+            // config specified for the parent channel to affect the LB channel.
+            .Remove(GRPC_ARG_SERVICE_CONFIG)
+            // The channel arg for the server URI, since that will be different
+            // for the LB channel than for the parent channel.  The client
+            // channel factory will re-add this arg with the right value.
+            .Remove(GRPC_ARG_SERVER_URI)
+            // The fake resolver response generator, because we are replacing it
+            // with the one from the grpclb policy, used to propagate updates to
+            // the LB channel.
+            .Remove(GRPC_ARG_FAKE_RESOLVER_RESPONSE_GENERATOR)
+            // The LB channel should use the authority indicated by the target
+            // authority table (see \a ModifyGrpclbBalancerChannelArgs),
+            // as opposed to the authority from the parent channel.
+            .Remove(GRPC_ARG_DEFAULT_AUTHORITY)
+            // Just as for \a GRPC_ARG_DEFAULT_AUTHORITY, the LB channel should
+            // be treated as a stand-alone channel and not inherit this argument
+            // from the args of the parent channel.
+            .Remove(GRPC_SSL_TARGET_NAME_OVERRIDE_ARG)
+            // Don't want to pass down channelz node from parent; the balancer
+            // channel will get its own.
+            .Remove(GRPC_ARG_CHANNELZ_CHANNEL_NODE)
+            // Remove the channel args for channel credentials and replace it
+            // with a version that does not contain call credentials. The
+            // loadbalancer is not necessarily trusted to handle bearer token
+            // credentials.
+            .Remove(GRPC_ARG_CHANNEL_CREDENTIALS);
+  }
   // Create channel args for channel credentials that does not contain bearer
   // token credentials.
   auto* channel_credentials = args.GetObject<grpc_channel_credentials>();
@@ -1365,36 +1441,7 @@ ChannelArgs BuildBalancerChannelArgs(
   RefCountedPtr<grpc_channel_credentials> creds_sans_call_creds =
       channel_credentials->duplicate_without_call_credentials();
   GPR_ASSERT(creds_sans_call_creds != nullptr);
-  return args
-      // LB policy name, since we want to use the default (pick_first) in
-      // the LB channel.
-      .Remove(GRPC_ARG_LB_POLICY_NAME)
-      // Strip out the service config, since we don't want the LB policy
-      // config specified for the parent channel to affect the LB channel.
-      .Remove(GRPC_ARG_SERVICE_CONFIG)
-      // The channel arg for the server URI, since that will be different for
-      // the LB channel than for the parent channel.  The client channel
-      // factory will re-add this arg with the right value.
-      .Remove(GRPC_ARG_SERVER_URI)
-      // The fake resolver response generator, because we are replacing it
-      // with the one from the grpclb policy, used to propagate updates to
-      // the LB channel.
-      .Remove(GRPC_ARG_FAKE_RESOLVER_RESPONSE_GENERATOR)
-      // The LB channel should use the authority indicated by the target
-      // authority table (see \a ModifyGrpclbBalancerChannelArgs),
-      // as opposed to the authority from the parent channel.
-      .Remove(GRPC_ARG_DEFAULT_AUTHORITY)
-      // Just as for \a GRPC_ARG_DEFAULT_AUTHORITY, the LB channel should be
-      // treated as a stand-alone channel and not inherit this argument from the
-      // args of the parent channel.
-      .Remove(GRPC_SSL_TARGET_NAME_OVERRIDE_ARG)
-      // Don't want to pass down channelz node from parent; the balancer
-      // channel will get its own.
-      .Remove(GRPC_ARG_CHANNELZ_CHANNEL_NODE)
-      // Remove the channel args for channel credentials and replace it
-      // with a version that does not contain call credentials. The loadbalancer
-      // is not necessarily trusted to handle bearer token credentials.
-      .Remove(GRPC_ARG_CHANNEL_CREDENTIALS)
+  return grpclb_channel_args
       // A channel arg indicating the target is a grpclb load balancer.
       .Set(GRPC_ARG_ADDRESS_IS_GRPCLB_LOAD_BALANCER, 1)
       // Tells channelz that this is an internal channel.
@@ -1512,7 +1559,7 @@ void GrpcLb::ResetBackoffLocked() {
   }
 }
 
-void GrpcLb::UpdateLocked(UpdateArgs args) {
+absl::Status GrpcLb::UpdateLocked(UpdateArgs args) {
   const bool is_initial_update = lb_channel_ == nullptr;
   config_ = args.config;
   GPR_ASSERT(config_ != nullptr);
@@ -1528,7 +1575,7 @@ void GrpcLb::UpdateLocked(UpdateArgs args) {
   }
   resolution_note_ = std::move(args.resolution_note);
   // Update balancer channel.
-  UpdateBalancerChannelLocked(args.args);
+  absl::Status status = UpdateBalancerChannelLocked(args.args);
   // Update the existing child policy, if any.
   if (child_policy_ != nullptr) CreateOrUpdateChildPolicyLocked();
   // If this is the initial update, start the fallback-at-startup checks
@@ -1536,7 +1583,7 @@ void GrpcLb::UpdateLocked(UpdateArgs args) {
   if (is_initial_update) {
     fallback_at_startup_checks_pending_ = true;
     // Start timer.
-    Timestamp deadline = ExecCtx::Get()->Now() + fallback_at_startup_timeout_;
+    Timestamp deadline = Timestamp::Now() + fallback_at_startup_timeout_;
     Ref(DEBUG_LOCATION, "on_fallback_timer").release();  // Ref for callback
     grpc_timer_init(&lb_fallback_timer_, deadline, &lb_on_fallback_);
     // Start watching the channel's connectivity state.  If the channel
@@ -1553,18 +1600,24 @@ void GrpcLb::UpdateLocked(UpdateArgs args) {
     // Start balancer call.
     StartBalancerCallLocked();
   }
+  return status;
 }
 
 //
 // helpers for UpdateLocked()
 //
 
-void GrpcLb::UpdateBalancerChannelLocked(const ChannelArgs& args) {
+absl::Status GrpcLb::UpdateBalancerChannelLocked(const ChannelArgs& args) {
   // Make sure that GRPC_ARG_LB_POLICY_NAME is set in channel args,
   // since we use this to trigger the client_load_reporting filter.
   args_ = args.Set(GRPC_ARG_LB_POLICY_NAME, "grpclb");
-  // Construct args for balancer channel.
+  // Get balancer addresses.
   ServerAddressList balancer_addresses = ExtractBalancerAddresses(args);
+  absl::Status status;
+  if (balancer_addresses.empty()) {
+    status = absl::UnavailableError("balancer address list must be non-empty");
+  }
+  // Construct args for balancer channel.
   ChannelArgs lb_channel_args =
       BuildBalancerChannelArgs(response_generator_.get(), args);
   // Create balancer channel if needed.
@@ -1592,6 +1645,8 @@ void GrpcLb::UpdateBalancerChannelLocked(const ChannelArgs& args) {
   result.addresses = std::move(balancer_addresses);
   result.args = lb_channel_args;
   response_generator_->SetResponse(std::move(result));
+  // Return status.
+  return status;
 }
 
 void GrpcLb::CancelBalancerChannelConnectivityWatchLocked() {
@@ -1623,7 +1678,7 @@ void GrpcLb::StartBalancerCallRetryTimerLocked() {
   Timestamp next_try = lb_call_backoff_.NextAttemptTime();
   if (GRPC_TRACE_FLAG_ENABLED(grpc_lb_glb_trace)) {
     gpr_log(GPR_INFO, "[grpclb %p] Connection to LB server lost...", this);
-    Duration timeout = next_try - ExecCtx::Get()->Now();
+    Duration timeout = next_try - Timestamp::Now();
     if (timeout > Duration::Zero()) {
       gpr_log(GPR_INFO, "[grpclb %p] ... retry_timer_active in %" PRId64 "ms.",
               this, timeout.millis());
@@ -1782,7 +1837,9 @@ void GrpcLb::CreateOrUpdateChildPolicyLocked() {
     gpr_log(GPR_INFO, "[grpclb %p] Updating child policy handler %p", this,
             child_policy_.get());
   }
-  child_policy_->UpdateLocked(std::move(update_args));
+  // TODO(roth): If we're in fallback mode and the child policy rejects the
+  // update, we should propagate that failure back to the resolver somehow.
+  (void)child_policy_->UpdateLocked(std::move(update_args));
 }
 
 //
@@ -1791,7 +1848,7 @@ void GrpcLb::CreateOrUpdateChildPolicyLocked() {
 
 void GrpcLb::CacheDeletedSubchannelLocked(
     RefCountedPtr<SubchannelInterface> subchannel) {
-  Timestamp deletion_time = ExecCtx::Get()->Now() + subchannel_cache_interval_;
+  Timestamp deletion_time = Timestamp::Now() + subchannel_cache_interval_;
   cached_subchannels_[deletion_time].push_back(std::move(subchannel));
   if (!subchannel_cache_timer_pending_) {
     Ref(DEBUG_LOCATION, "OnSubchannelCacheTimer").release();
@@ -1850,48 +1907,8 @@ class GrpcLbFactory : public LoadBalancingPolicyFactory {
 
   absl::StatusOr<RefCountedPtr<LoadBalancingPolicy::Config>>
   ParseLoadBalancingConfig(const Json& json) const override {
-    if (json.type() == Json::Type::JSON_NULL) {
-      return MakeRefCounted<GrpcLbConfig>(nullptr, "");
-    }
-    std::vector<std::string> error_list;
-    std::string service_name;
-    auto it = json.object_value().find("serviceName");
-    if (it != json.object_value().end()) {
-      const Json& service_name_json = it->second;
-      if (service_name_json.type() != Json::Type::STRING) {
-        error_list.emplace_back(
-            "field:serviceName error:type should be string");
-      } else {
-        service_name = service_name_json.string_value();
-      }
-    }
-    Json child_policy_config_json_tmp;
-    const Json* child_policy_config_json;
-    it = json.object_value().find("childPolicy");
-    if (it == json.object_value().end()) {
-      child_policy_config_json_tmp = Json::Array{Json::Object{
-          {"round_robin", Json::Object()},
-      }};
-      child_policy_config_json = &child_policy_config_json_tmp;
-    } else {
-      child_policy_config_json = &it->second;
-    }
-    auto child_policy_config =
-        LoadBalancingPolicyRegistry::ParseLoadBalancingConfig(
-            *child_policy_config_json);
-    if (!child_policy_config.ok()) {
-      error_list.emplace_back(
-          absl::StrCat("error parsing childPolicy field: ",
-                       child_policy_config.status().message()));
-    }
-    if (error_list.empty()) {
-      return MakeRefCounted<GrpcLbConfig>(std::move(*child_policy_config),
-                                          std::move(service_name));
-    } else {
-      return absl::InvalidArgumentError(
-          absl::StrCat("errors parsing grpclb LB policy config: [",
-                       absl::StrJoin(error_list, "; "), "]"));
-    }
+    return LoadRefCountedFromJson<GrpcLbConfig>(
+        json, JsonArgs(), "errors validating grpclb LB policy config");
   }
 };
 
@@ -1903,16 +1920,10 @@ class GrpcLbFactory : public LoadBalancingPolicyFactory {
 // Plugin registration
 //
 
-void grpc_lb_policy_grpclb_init() {
-  grpc_core::LoadBalancingPolicyRegistry::Builder::
-      RegisterLoadBalancingPolicyFactory(
-          absl::make_unique<grpc_core::GrpcLbFactory>());
-}
-
-void grpc_lb_policy_grpclb_shutdown() {}
-
 namespace grpc_core {
-void RegisterGrpcLbLoadReportingFilter(CoreConfiguration::Builder* builder) {
+void RegisterGrpcLbPolicy(CoreConfiguration::Builder* builder) {
+  builder->lb_policy_registry()->RegisterLoadBalancingPolicyFactory(
+      absl::make_unique<GrpcLbFactory>());
   builder->channel_init()->RegisterStage(
       GRPC_CLIENT_SUBCHANNEL, GRPC_CHANNEL_INIT_BUILTIN_PRIORITY,
       [](ChannelStackBuilder* builder) {

data/src/core/ext/filters/client_channel/lb_policy/oob_backend_metric.cc

@@ -76,7 +76,7 @@ class OrcaWatcher;
 // registered watchers.
 class OrcaProducer : public Subchannel::DataProducerInterface {
  public:
-  explicit OrcaProducer(RefCountedPtr<Subchannel> subchannel);
+  void Start(RefCountedPtr<Subchannel> subchannel);
 
   void Orphan() override;
 
@@ -288,9 +288,8 @@ class OrcaProducer::OrcaStreamEventHandler
 // OrcaProducer
 //
 
-OrcaProducer::OrcaProducer(RefCountedPtr<Subchannel> subchannel)
-    : subchannel_(std::move(subchannel)) {
-  subchannel_->AddDataProducer(this);
+void OrcaProducer::Start(RefCountedPtr<Subchannel> subchannel) {
+  subchannel_ = std::move(subchannel);
   connected_subchannel_ = subchannel_->connected_subchannel();
   auto connectivity_watcher = MakeRefCounted<ConnectivityWatcher>(WeakRef());
   connectivity_watcher_ = connectivity_watcher.get();
@@ -304,6 +303,7 @@ void OrcaProducer::Orphan() {
     MutexLock lock(&mu_);
     stream_client_.reset();
   }
+  GPR_ASSERT(subchannel_ != nullptr);  // Should not be called before Start().
   subchannel_->CancelConnectivityStateWatch(
       /*health_check_service_name=*/absl::nullopt, connectivity_watcher_);
   subchannel_->RemoveDataProducer(this);
@@ -384,14 +384,23 @@ OrcaWatcher::~OrcaWatcher() {
 }
 
 void OrcaWatcher::SetSubchannel(Subchannel* subchannel) {
+  bool created = false;
   // Check if our producer is already registered with the subchannel.
-  // If not, create a new one, which will register itself with the subchannel.
-  auto* p = static_cast<OrcaProducer*>(
-      subchannel->GetDataProducer(OrcaProducer::Type()));
-  if (p != nullptr) producer_ = p->RefIfNonZero();
-  if (producer_ == nullptr) {
-    producer_ = MakeRefCounted<OrcaProducer>(subchannel->Ref());
-  }
+  // If not, create a new one.
+  subchannel->GetOrAddDataProducer(
+      OrcaProducer::Type(), [&](Subchannel::DataProducerInterface** producer) {
+        if (*producer != nullptr) producer_ = (*producer)->RefIfNonZero();
+        if (producer_ == nullptr) {
+          producer_ = MakeRefCounted<OrcaProducer>();
+          *producer = producer_.get();
+          created = true;
+        }
+      });
+  // If we just created the producer, start it.
+  // This needs to be done outside of the lambda passed to
+  // GetOrAddDataProducer() to avoid deadlocking by re-acquiring the
+  // subchannel lock while already holding it.
+  if (created) producer_->Start(subchannel->Ref());
   // Register ourself with the producer.
   producer_->AddWatcher(this);
 }