grpc 1.73.0 → 1.75.0

data/src/core/client_channel/retry_interceptor.cc

@@ -35,11 +35,10 @@ size_t GetMaxPerRpcRetryBufferSize(const ChannelArgs& args) {
 
 namespace retry_detail {
 
-RetryState::RetryState(
-    const internal::RetryMethodConfig* retry_policy,
-    RefCountedPtr<internal::ServerRetryThrottleData> retry_throttle_data)
+RetryState::RetryState(const internal::RetryMethodConfig* retry_policy,
+                       RefCountedPtr<internal::RetryThrottler> retry_throttler)
     : retry_policy_(retry_policy),
-      retry_throttle_data_(std::move(retry_throttle_data)),
+      retry_throttler_(std::move(retry_throttler)),
       retry_backoff_(
           BackOff::Options()
               .set_initial_backoff(retry_policy_ == nullptr
@@ -67,8 +66,8 @@ std::optional<Duration> RetryState::ShouldRetry(
   const auto status = md.get(GrpcStatusMetadata());
   if (status.has_value()) {
     if (GPR_LIKELY(*status == GRPC_STATUS_OK)) {
-      if (retry_throttle_data_ != nullptr) {
-        retry_throttle_data_->RecordSuccess();
+      if (retry_throttler_ != nullptr) {
+        retry_throttler_->RecordSuccess();
       }
       GRPC_TRACE_LOG(retry, INFO)
           << lazy_attempt_debug_string() << " call succeeded";
@@ -89,8 +88,7 @@ std::optional<Duration> RetryState::ShouldRetry(
   // things like failures due to malformed requests (INVALID_ARGUMENT).
   // Conversely, it's important for this to come before the remaining
   // checks, so that we don't fail to record failures due to other factors.
-  if (retry_throttle_data_ != nullptr &&
-      !retry_throttle_data_->RecordFailure()) {
+  if (retry_throttler_ != nullptr && !retry_throttler_->RecordFailure()) {
     GRPC_TRACE_LOG(retry, INFO)
         << lazy_attempt_debug_string() << " retries throttled";
     return std::nullopt;
@@ -131,55 +129,43 @@ std::optional<Duration> RetryState::ShouldRetry(
   return next_attempt_timeout;
 }
 
-absl::StatusOr<RefCountedPtr<internal::ServerRetryThrottleData>>
-ServerRetryThrottleDataFromChannelArgs(const ChannelArgs& args) {
-  // Get retry throttling parameters from service config.
-  auto* service_config = args.GetObject<ServiceConfig>();
-  if (service_config == nullptr) return nullptr;
-  const auto* config = static_cast<const internal::RetryGlobalConfig*>(
-      service_config->GetGlobalParsedConfig(
-          internal::RetryServiceConfigParser::ParserIndex()));
-  if (config == nullptr) return nullptr;
-  // Get server name from target URI.
-  auto server_uri = args.GetString(GRPC_ARG_SERVER_URI);
-  if (!server_uri.has_value()) {
-    return GRPC_ERROR_CREATE(
-        "server URI channel arg missing or wrong type in client channel "
-        "filter");
-  }
-  absl::StatusOr<URI> uri = URI::Parse(*server_uri);
-  if (!uri.ok() || uri->path().empty()) {
-    return GRPC_ERROR_CREATE("could not extract server name from target URI");
-  }
-  std::string server_name(absl::StripPrefix(uri->path(), "/"));
-  // Get throttling config for server_name.
-  return internal::ServerRetryThrottleMap::Get()->GetDataForServer(
-      server_name, config->max_milli_tokens(), config->milli_token_ratio());
-}
-
 }  // namespace retry_detail
 
 ////////////////////////////////////////////////////////////////////////////////
 // RetryInterceptor
 
 absl::StatusOr<RefCountedPtr<RetryInterceptor>> RetryInterceptor::Create(
-    const ChannelArgs& args, const FilterArgs&) {
-  auto retry_throttle_data =
-      retry_detail::ServerRetryThrottleDataFromChannelArgs(args);
-  if (!retry_throttle_data.ok()) {
-    return retry_throttle_data.status();
+    const ChannelArgs& args, const FilterArgs& filter_args) {
+  auto retry_throttler = filter_args.GetState<internal::RetryThrottler>("");
+  return MakeRefCounted<RetryInterceptor>(args, std::move(retry_throttler));
+}
+
+void RetryInterceptor::UpdateBlackboard(const ServiceConfig& service_config,
+                                        const Blackboard* old_blackboard,
+                                        Blackboard* new_blackboard) {
+  const auto* config = static_cast<const internal::RetryGlobalConfig*>(
+      service_config.GetGlobalParsedConfig(
+          internal::RetryServiceConfigParser::ParserIndex()));
+  if (config == nullptr) return;
+  // Get existing throttle state.
+  RefCountedPtr<internal::RetryThrottler> throttler;
+  if (old_blackboard != nullptr) {
+    throttler = old_blackboard->Get<internal::RetryThrottler>("");
   }
-  return MakeRefCounted<RetryInterceptor>(args,
-                                          std::move(*retry_throttle_data));
+  throttler = internal::RetryThrottler::Create(config->max_milli_tokens(),
+                                               config->milli_token_ratio(),
+                                               std::move(throttler));
+  CHECK_NE(new_blackboard, nullptr);
+  new_blackboard->Set("", std::move(throttler));
 }
 
 RetryInterceptor::RetryInterceptor(
     const ChannelArgs& args,
-    RefCountedPtr<internal::ServerRetryThrottleData> retry_throttle_data)
+    RefCountedPtr<internal::RetryThrottler> retry_throttler)
     : per_rpc_retry_buffer_size_(GetMaxPerRpcRetryBufferSize(args)),
       service_config_parser_index_(
           internal::RetryServiceConfigParser::ParserIndex()),
-      retry_throttle_data_(std::move(retry_throttle_data)) {}
+      retry_throttler_(std::move(retry_throttler)) {}
 
 void RetryInterceptor::InterceptCall(
     UnstartedCallHandler unstarted_call_handler) {
@@ -206,7 +192,7 @@ RetryInterceptor::Call::Call(RefCountedPtr<RetryInterceptor> interceptor,
     : call_handler_(std::move(call_handler)),
       interceptor_(std::move(interceptor)),
       retry_state_(interceptor_->GetRetryPolicy(),
-                   interceptor_->retry_throttle_data_) {
+                   interceptor_->retry_throttler_) {
   GRPC_TRACE_LOG(retry, INFO)
       << DebugTag() << " retry call created: " << retry_state_;
 }

data/src/core/client_channel/retry_interceptor.h

@@ -26,11 +26,11 @@
 namespace grpc_core {
 
 namespace retry_detail {
+
 class RetryState {
  public:
-  RetryState(
-      const internal::RetryMethodConfig* retry_policy,
-      RefCountedPtr<internal::ServerRetryThrottleData> retry_throttle_data);
+  RetryState(const internal::RetryMethodConfig* retry_policy,
+             RefCountedPtr<internal::RetryThrottler> retry_throttler);
 
   // if nullopt --> commit & don't retry
   // if duration --> retry after duration
@@ -41,36 +41,37 @@ class RetryState {
 
   template <typename Sink>
   friend void AbslStringify(Sink& sink, const RetryState& state) {
-    sink.Append(absl::StrCat(
-        "policy:{",
-        state.retry_policy_ != nullptr ? absl::StrCat(*state.retry_policy_)
-                                       : "none",
-        "} throttle:", state.retry_throttle_data_ != nullptr,
-        " attempts:", state.num_attempts_completed_));
+    sink.Append(absl::StrCat("policy:{",
+                             state.retry_policy_ != nullptr
+                                 ? absl::StrCat(*state.retry_policy_)
+                                 : "none",
+                             "} throttler:", state.retry_throttler_ != nullptr,
+                             " attempts:", state.num_attempts_completed_));
   }
 
  private:
   const internal::RetryMethodConfig* const retry_policy_;
-  RefCountedPtr<internal::ServerRetryThrottleData> retry_throttle_data_;
+  RefCountedPtr<internal::RetryThrottler> retry_throttler_;
   int num_attempts_completed_ = 0;
   BackOff retry_backoff_;
 };
 
-absl::StatusOr<RefCountedPtr<internal::ServerRetryThrottleData>>
-ServerRetryThrottleDataFromChannelArgs(const ChannelArgs& args);
 }  // namespace retry_detail
 
 class RetryInterceptor : public Interceptor {
  public:
-  RetryInterceptor(
-      const ChannelArgs& args,
-      RefCountedPtr<internal::ServerRetryThrottleData> retry_throttle_data);
+  RetryInterceptor(const ChannelArgs& args,
+                   RefCountedPtr<internal::RetryThrottler> retry_throttler);
 
   static absl::StatusOr<RefCountedPtr<RetryInterceptor>> Create(
-      const ChannelArgs& args, const FilterArgs&);
+      const ChannelArgs& args, const FilterArgs& filter_args);
 
   void Orphaned() override {}
 
+  static void UpdateBlackboard(const ServiceConfig& service_config,
+                               const Blackboard* old_blackboard,
+                               Blackboard* new_blackboard);
+
  protected:
   void InterceptCall(UnstartedCallHandler unstarted_call_handler) override;
 
@@ -149,7 +150,7 @@ class RetryInterceptor : public Interceptor {
 
   const size_t per_rpc_retry_buffer_size_;
   const size_t service_config_parser_index_;
-  const RefCountedPtr<internal::ServerRetryThrottleData> retry_throttle_data_;
+  const RefCountedPtr<internal::RetryThrottler> retry_throttler_;
 };
 
 }  // namespace grpc_core

data/src/core/client_channel/retry_throttle.cc

@@ -18,14 +18,9 @@
 
 #include "src/core/client_channel/retry_throttle.h"
 
-#include <grpc/support/port_platform.h>
-
 #include <atomic>
 #include <cstdint>
 #include <limits>
-#include <map>
-#include <string>
-#include <utility>
 
 #include "src/core/util/useful.h"
 
@@ -33,6 +28,7 @@ namespace grpc_core {
 namespace internal {
 
 namespace {
+
 template <typename T>
 T ClampedAdd(std::atomic<T>& value, T delta, T min, T max) {
   T prev_value = value.load(std::memory_order_relaxed);
@@ -43,45 +39,74 @@ T ClampedAdd(std::atomic<T>& value, T delta, T min, T max) {
                                         std::memory_order_relaxed));
   return new_value;
 }
+
 }  // namespace
 
 //
-// ServerRetryThrottleData
+// RetryThrottler
 //
 
-ServerRetryThrottleData::ServerRetryThrottleData(uintptr_t max_milli_tokens,
-                                                 uintptr_t milli_token_ratio,
-                                                 uintptr_t milli_tokens)
+RefCountedPtr<RetryThrottler> RetryThrottler::Create(
+    uintptr_t max_milli_tokens, uintptr_t milli_token_ratio,
+    RefCountedPtr<RetryThrottler> previous) {
+  if (previous != nullptr && previous->max_milli_tokens_ == max_milli_tokens &&
+      previous->milli_token_ratio_ == milli_token_ratio) {
+    return previous;
+  }
+  // previous is null or has different parameters.  Create a new one.
+  uintptr_t initial_milli_tokens = max_milli_tokens;
+  // If there was a pre-existing entry for this server name, initialize
+  // the token count by scaling proportionately to the old data.  This
+  // ensures that if we're already throttling retries on the old scale,
+  // we will start out doing the same thing on the new one.
+  if (previous != nullptr) {
+    double token_fraction = static_cast<double>(previous->milli_tokens_) /
+                            static_cast<double>(previous->max_milli_tokens_);
+    initial_milli_tokens =
+        static_cast<uintptr_t>(token_fraction * max_milli_tokens);
+  }
+  auto throttle_data = MakeRefCounted<RetryThrottler>(
+      max_milli_tokens, milli_token_ratio, initial_milli_tokens);
+  if (previous != nullptr) previous->SetReplacement(throttle_data);
+  return throttle_data;
+}
+
+UniqueTypeName RetryThrottler::Type() {
+  static UniqueTypeName::Factory factory("retry_throttle");
+  return factory.Create();
+}
+
+RetryThrottler::RetryThrottler(uintptr_t max_milli_tokens,
+                               uintptr_t milli_token_ratio,
+                               uintptr_t milli_tokens)
     : max_milli_tokens_(max_milli_tokens),
       milli_token_ratio_(milli_token_ratio),
       milli_tokens_(milli_tokens) {}
 
-ServerRetryThrottleData::~ServerRetryThrottleData() {
-  ServerRetryThrottleData* replacement =
-      replacement_.load(std::memory_order_acquire);
+RetryThrottler::~RetryThrottler() {
+  RetryThrottler* replacement = replacement_.load(std::memory_order_acquire);
   if (replacement != nullptr) {
     replacement->Unref();
   }
 }
 
-void ServerRetryThrottleData::SetReplacement(
-    RefCountedPtr<ServerRetryThrottleData> replacement) {
+void RetryThrottler::SetReplacement(RefCountedPtr<RetryThrottler> replacement) {
   replacement_.store(replacement.release(), std::memory_order_release);
 }
 
-void ServerRetryThrottleData::GetReplacementThrottleDataIfNeeded(
-    ServerRetryThrottleData** throttle_data) {
+void RetryThrottler::GetReplacementThrottleDataIfNeeded(
+    RetryThrottler** throttle_data) {
   while (true) {
-    ServerRetryThrottleData* new_throttle_data =
+    RetryThrottler* new_throttle_data =
         (*throttle_data)->replacement_.load(std::memory_order_acquire);
     if (new_throttle_data == nullptr) return;
     *throttle_data = new_throttle_data;
   }
 }
 
-bool ServerRetryThrottleData::RecordFailure() {
+bool RetryThrottler::RecordFailure() {
   // First, check if we are stale and need to be replaced.
-  ServerRetryThrottleData* throttle_data = this;
+  RetryThrottler* throttle_data = this;
   GetReplacementThrottleDataIfNeeded(&throttle_data);
   // We decrement milli_tokens by 1000 (1 token) for each failure.
   const uintptr_t new_value = ClampedAdd<intptr_t>(
@@ -93,9 +118,9 @@ bool ServerRetryThrottleData::RecordFailure() {
   return new_value > throttle_data->max_milli_tokens_ / 2;
 }
 
-void ServerRetryThrottleData::RecordSuccess() {
+void RetryThrottler::RecordSuccess() {
   // First, check if we are stale and need to be replaced.
-  ServerRetryThrottleData* throttle_data = this;
+  RetryThrottler* throttle_data = this;
   GetReplacementThrottleDataIfNeeded(&throttle_data);
   // We increment milli_tokens by milli_token_ratio for each success.
   ClampedAdd<intptr_t>(
@@ -105,45 +130,5 @@ void ServerRetryThrottleData::RecordSuccess() {
                                  std::numeric_limits<intptr_t>::max())));
 }
 
-//
-// ServerRetryThrottleMap
-//
-
-ServerRetryThrottleMap* ServerRetryThrottleMap::Get() {
-  static ServerRetryThrottleMap* m = new ServerRetryThrottleMap();
-  return m;
-}
-
-RefCountedPtr<ServerRetryThrottleData> ServerRetryThrottleMap::GetDataForServer(
-    const std::string& server_name, uintptr_t max_milli_tokens,
-    uintptr_t milli_token_ratio) {
-  MutexLock lock(&mu_);
-  auto& throttle_data = map_[server_name];
-  if (throttle_data == nullptr ||
-      throttle_data->max_milli_tokens() != max_milli_tokens ||
-      throttle_data->milli_token_ratio() != milli_token_ratio) {
-    // Entry not found, or found with old parameters.  Create a new one.
-    auto old_throttle_data = std::move(throttle_data);
-    uintptr_t initial_milli_tokens = max_milli_tokens;
-    // If there was a pre-existing entry for this server name, initialize
-    // the token count by scaling proportionately to the old data.  This
-    // ensures that if we're already throttling retries on the old scale,
-    // we will start out doing the same thing on the new one.
-    if (old_throttle_data != nullptr) {
-      double token_fraction =
-          static_cast<double>(old_throttle_data->milli_tokens()) /
-          static_cast<double>(old_throttle_data->max_milli_tokens());
-      initial_milli_tokens =
-          static_cast<uintptr_t>(token_fraction * max_milli_tokens);
-    }
-    throttle_data = MakeRefCounted<ServerRetryThrottleData>(
-        max_milli_tokens, milli_token_ratio, initial_milli_tokens);
-    if (old_throttle_data != nullptr) {
-      old_throttle_data->SetReplacement(throttle_data);
-    }
-  }
-  return throttle_data;
-}
-
 }  // namespace internal
 }  // namespace grpc_core

data/src/core/client_channel/retry_throttle.h

@@ -19,30 +19,29 @@
 #ifndef GRPC_SRC_CORE_CLIENT_CHANNEL_RETRY_THROTTLE_H
 #define GRPC_SRC_CORE_CLIENT_CHANNEL_RETRY_THROTTLE_H
 
-#include <grpc/support/port_platform.h>
 #include <stdint.h>
 
 #include <atomic>
-#include <map>
-#include <string>
 
-#include "absl/base/thread_annotations.h"
-#include "src/core/util/ref_counted.h"
+#include "src/core/filter/blackboard.h"
 #include "src/core/util/ref_counted_ptr.h"
-#include "src/core/util/sync.h"
 
 namespace grpc_core {
 namespace internal {
 
-class ServerRetryThrottleMap;
-
 /// Tracks retry throttling data for an individual server name.
-class ServerRetryThrottleData final
-    : public RefCounted<ServerRetryThrottleData> {
+class RetryThrottler final : public Blackboard::Entry {
  public:
-  ServerRetryThrottleData(uintptr_t max_milli_tokens,
-                          uintptr_t milli_token_ratio, uintptr_t milli_tokens);
-  ~ServerRetryThrottleData() override;
+  static RefCountedPtr<RetryThrottler> Create(
+      uintptr_t max_milli_tokens, uintptr_t milli_token_ratio,
+      RefCountedPtr<RetryThrottler> previous);
+
+  static UniqueTypeName Type();
+
+  // Do not instantiate directly -- use Create() instead.
+  RetryThrottler(uintptr_t max_milli_tokens, uintptr_t milli_token_ratio,
+                 uintptr_t milli_tokens);
+  ~RetryThrottler() override;
 
   /// Records a failure.  Returns true if it's okay to send a retry.
   bool RecordFailure();
@@ -50,6 +49,7 @@ class ServerRetryThrottleData final
   /// Records a success.
   void RecordSuccess();
 
+  // Exposed for testing purposes only.
   uintptr_t max_milli_tokens() const { return max_milli_tokens_; }
   uintptr_t milli_token_ratio() const { return milli_token_ratio_; }
   intptr_t milli_tokens() const {
@@ -57,39 +57,17 @@ class ServerRetryThrottleData final
   }
 
  private:
-  friend ServerRetryThrottleMap;
+  void SetReplacement(RefCountedPtr<RetryThrottler> replacement);
 
-  void SetReplacement(RefCountedPtr<ServerRetryThrottleData> replacement);
-
-  void GetReplacementThrottleDataIfNeeded(
-      ServerRetryThrottleData** throttle_data);
+  void GetReplacementThrottleDataIfNeeded(RetryThrottler** throttle_data);
 
   const uintptr_t max_milli_tokens_;
   const uintptr_t milli_token_ratio_;
   std::atomic<intptr_t> milli_tokens_;
-  // A pointer to the replacement for this ServerRetryThrottleData entry.
+  // A pointer to the replacement for this RetryThrottler entry.
   // If non-nullptr, then this entry is stale and must not be used.
   // We hold a reference to the replacement.
-  std::atomic<ServerRetryThrottleData*> replacement_{nullptr};
-};
-
-/// Global map of server name to retry throttle data.
-class ServerRetryThrottleMap final {
- public:
-  static ServerRetryThrottleMap* Get();
-
-  /// Returns the failure data for \a server_name, creating a new entry if
-  /// needed.
-  RefCountedPtr<ServerRetryThrottleData> GetDataForServer(
-      const std::string& server_name, uintptr_t max_milli_tokens,
-      uintptr_t milli_token_ratio);
-
- private:
-  using StringToDataMap =
-      std::map<std::string, RefCountedPtr<ServerRetryThrottleData>>;
-
-  Mutex mu_;
-  StringToDataMap map_ ABSL_GUARDED_BY(mu_);
+  std::atomic<RetryThrottler*> replacement_{nullptr};
 };
 
 }  // namespace internal

data/src/core/client_channel/subchannel.cc

@@ -117,7 +117,7 @@ class LegacyConnectedSubchannel : public ConnectedSubchannel {
     channel_stack_.reset(DEBUG_LOCATION, "ConnectedSubchannel");
   }
 
-  channelz::SubchannelNode* channelz_node() const {
+  channelz::SubchannelNode* channelz_node() const override {
     return channelz_node_.get();
   }
 
@@ -220,6 +220,8 @@ class NewConnectedSubchannel : public ConnectedSubchannel {
     Crash("legacy ping method called in call v3 impl");
   }
 
+  channelz::SubchannelNode* channelz_node() const override { return nullptr; }
+
  private:
   RefCountedPtr<UnstartedCallDestination> call_destination_;
   RefCountedPtr<TransportCallDestination> transport_;
@@ -410,23 +412,37 @@ class Subchannel::ConnectedSubchannelStateWatcher final
       // we will see TRANSIENT_FAILURE followed by SHUTDOWN, but if not, we
       // will see only SHUTDOWN.  Either way, we react to the first one we
       // see, ignoring anything that happens after that.
-      if (c->connected_subchannel_ == nullptr) return;
       if (new_state == GRPC_CHANNEL_TRANSIENT_FAILURE ||
           new_state == GRPC_CHANNEL_SHUTDOWN) {
+        RefCountedPtr<ConnectedSubchannel> connected_subchannel =
+            std::move(c->connected_subchannel_);
+        if (connected_subchannel == nullptr) return;
         GRPC_TRACE_LOG(subchannel, INFO)
             << "subchannel " << c << " " << c->key_.ToString()
-            << ": Connected subchannel " << c->connected_subchannel_.get()
+            << ": Connected subchannel " << connected_subchannel.get()
             << " reports " << ConnectivityStateName(new_state) << ": "
             << status;
-        c->connected_subchannel_.reset();
         if (c->channelz_node() != nullptr) {
-          c->channelz_node()->SetChildSocket(nullptr);
+          if (connected_subchannel->channelz_node() != nullptr) {
+            connected_subchannel->channelz_node()->RemoveParent(
+                c->channelz_node());
+          }
         }
-        // Even though we're reporting IDLE instead of TRANSIENT_FAILURE here,
-        // pass along the status from the transport, since it may have
-        // keepalive info attached to it that the channel needs.
-        // TODO(roth): Consider whether there's a cleaner way to do this.
-        c->SetConnectivityStateLocked(GRPC_CHANNEL_IDLE, status);
+        // If the subchannel was created from an endpoint, then we report
+        // TRANSIENT_FAILURE here instead of IDLE. The subchannel will never
+        // leave TRANSIENT_FAILURE state, because there is no way for us to
+        // establish a new connection.
+        //
+        // Otherwise, we report IDLE here. Note that even though we're not
+        // reporting TRANSIENT_FAILURE, we pass along the status from the
+        // transport, since it may have keepalive info attached to it that the
+        // channel needs.
+        // TODO(roth): Consider whether there's a cleaner way to propagate the
+        // keepalive info.
+        c->SetConnectivityStateLocked(c->created_from_endpoint_
+                                          ? GRPC_CHANNEL_TRANSIENT_FAILURE
+                                          : GRPC_CHANNEL_IDLE,
+                                      status);
         c->backoff_.Reset();
       }
     }
@@ -509,6 +525,7 @@ Subchannel::Subchannel(SubchannelKey key,
                                      ? "Subchannel"
                                      : nullptr),
       key_(std::move(key)),
+      created_from_endpoint_(args.Contains(GRPC_ARG_SUBCHANNEL_ENDPOINT)),
       args_(args),
       pollset_set_(grpc_pollset_set_create()),
       connector_(std::move(connector)),
@@ -546,9 +563,7 @@ Subchannel::Subchannel(SubchannelKey key,
         grpc_sockaddr_to_uri(&key_.address())
             .value_or("<unknown address type>"),
         channel_tracer_max_memory);
-    channelz_node_->AddTraceEvent(
-        channelz::ChannelTrace::Severity::Info,
-        grpc_slice_from_static_string("subchannel created"));
+    GRPC_CHANNELZ_LOG(channelz_node_) << "subchannel created";
     channelz_node_->SetChannelArgs(args_);
     args_ = args_.SetObject<channelz::BaseNode>(channelz_node_);
   }
@@ -556,9 +571,7 @@ Subchannel::Subchannel(SubchannelKey key,
 
 Subchannel::~Subchannel() {
   if (channelz_node_ != nullptr) {
-    channelz_node_->AddTraceEvent(
-        channelz::ChannelTrace::Severity::Info,
-        grpc_slice_from_static_string("Subchannel destroyed"));
+    GRPC_CHANNELZ_LOG(channelz_node_) << "Subchannel destroyed";
     channelz_node_->UpdateConnectivityState(GRPC_CHANNEL_SHUTDOWN);
   }
   connector_.reset();
@@ -578,6 +591,15 @@ RefCountedPtr<Subchannel> Subchannel::Create(
     return c;
   }
   c = MakeRefCounted<Subchannel>(std::move(key), std::move(connector), args);
+  if (c->created_from_endpoint_) {
+    // We don't interact with the subchannel pool in this case.
+    // Instead, we unconditionally return the newly created subchannel.
+    // Before returning, we explicitly trigger a connection attempt
+    // by calling RequestConnection(), which sets the subchannel’s
+    // connectivity state to CONNECTING.
+    c->RequestConnection();
+    return c;
+  }
   // Try to register the subchannel before setting the subchannel pool.
   // Otherwise, in case of a registration race, unreffing c in
   // RegisterSubchannel() will cause c to be tried to be unregistered, while
@@ -702,12 +724,15 @@ void Subchannel::SetConnectivityStateLocked(grpc_connectivity_state state,
   }
   if (channelz_node_ != nullptr) {
     channelz_node_->UpdateConnectivityState(state);
-    channelz_node_->AddTraceEvent(
-        channelz::ChannelTrace::Severity::Info,
-        grpc_slice_from_cpp_string(absl::StrCat(
-            "Subchannel connectivity state changed to ",
-            ConnectivityStateName(state),
-            status.ok() ? "" : absl::StrCat(": ", status_.ToString()))));
+    if (status.ok()) {
+      GRPC_CHANNELZ_LOG(channelz_node_)
+          << "Subchannel connectivity state changed to "
+          << ConnectivityStateName(state);
+    } else {
+      GRPC_CHANNELZ_LOG(channelz_node_)
+          << "Subchannel connectivity state changed to "
+          << ConnectivityStateName(state) << ": " << status;
+    }
   }
   // Notify watchers.
   watcher_list_.NotifyLocked(state, status_);
@@ -767,10 +792,15 @@ void Subchannel::OnConnectingFinishedLocked(grpc_error_handle error) {
         next_attempt_time_ - Timestamp::Now();
     GRPC_TRACE_LOG(subchannel, INFO)
         << "subchannel " << this << " " << key_.ToString()
-        << ": connect failed (" << StatusToString(error)
-        << "), backing off for " << time_until_next_attempt.millis() << " ms";
+        << ": connect failed (" << StatusToString(error) << ")"
+        << (created_from_endpoint_
+                ? ", no retry will be attempted (created from endpoint); "
+                  "remaining in TRANSIENT_FAILURE"
+                : ", backing off for " +
+                      std::to_string(time_until_next_attempt.millis()) + " ms");
     SetConnectivityStateLocked(GRPC_CHANNEL_TRANSIENT_FAILURE,
                                grpc_error_to_absl_status(error));
+    if (created_from_endpoint_) return;
     retry_timer_handle_ = event_engine_->RunAfter(
         time_until_next_attempt,
         [self = WeakRef(DEBUG_LOCATION, "RetryTimer")]() mutable {
@@ -856,7 +886,9 @@ bool Subchannel::PublishTransportLocked() {
       << "subchannel " << this << " " << key_.ToString()
       << ": new connected subchannel at " << connected_subchannel_.get();
   if (channelz_node_ != nullptr) {
-    channelz_node_->SetChildSocket(std::move(socket_node));
+    if (socket_node != nullptr) {
+      socket_node->AddParent(channelz_node_.get());
+    }
   }
   // Start watching connected subchannel.
   connected_subchannel_->StartWatch(

data/src/core/client_channel/subchannel.h

@@ -58,6 +58,11 @@
 #include "src/core/util/unique_type_name.h"
 #include "src/core/util/work_serializer.h"
 
+/** This arg is intended for internal use only, primarily
+ *  for passing endpoint information during subchannel creation or connection.
+ */
+#define GRPC_ARG_SUBCHANNEL_ENDPOINT "grpc.internal.subchannel_endpoint"
+
 namespace grpc_core {
 
 class SubchannelCall;
@@ -80,6 +85,8 @@ class ConnectedSubchannel : public RefCounted<ConnectedSubchannel> {
   virtual size_t GetInitialCallSizeEstimate() const = 0;
   virtual void Ping(grpc_closure* on_initiate, grpc_closure* on_ack) = 0;
 
+  virtual channelz::SubchannelNode* channelz_node() const = 0;
+
  protected:
   explicit ConnectedSubchannel(const ChannelArgs& args);
 
@@ -328,6 +335,9 @@ class Subchannel final : public DualRefCounted<Subchannel> {
   RefCountedPtr<SubchannelPoolInterface> subchannel_pool_;
   // Subchannel key that identifies this subchannel in the subchannel pool.
   const SubchannelKey key_;
+  // boolean value that identifies this subchannel is created from event engine
+  // endpoint.
+  const bool created_from_endpoint_;
   // Actual address to connect to.  May be different than the address in
   // key_ if overridden by proxy mapper.
   grpc_resolved_address address_for_connect_;

data/src/core/config/config_vars.cc

@@ -30,7 +30,9 @@
 #endif  // !GPR_DEFAULT_LOG_VERBOSITY_STRING
 
 #ifdef GRPC_ENABLE_FORK_SUPPORT
+#ifndef GRPC_ENABLE_FORK_SUPPORT_DEFAULT
 #define GRPC_ENABLE_FORK_SUPPORT_DEFAULT true
+#endif  // !defined(GRPC_ENABLE_FORK_SUPPORT_DEFAULT)
 #else
 #define GRPC_ENABLE_FORK_SUPPORT_DEFAULT false
 #endif  // GRPC_ENABLE_FORK_SUPPORT