grpc 1.53.2 → 1.54.0

data/src/core/ext/filters/client_channel/retry_filter.cc

@@ -35,13 +35,14 @@
 #include "absl/strings/strip.h"
 #include "absl/types/optional.h"
 
+#include <grpc/event_engine/event_engine.h>
 #include <grpc/grpc.h>
 #include <grpc/slice.h>
 #include <grpc/status.h>
-#include <grpc/support/atm.h>
 #include <grpc/support/log.h>
 
 #include "src/core/ext/filters/client_channel/client_channel.h"
+#include "src/core/ext/filters/client_channel/client_channel_internal.h"
 #include "src/core/ext/filters/client_channel/config_selector.h"
 #include "src/core/ext/filters/client_channel/retry_service_config.h"
 #include "src/core/ext/filters/client_channel/retry_throttle.h"
@@ -64,7 +65,6 @@
 #include "src/core/lib/iomgr/error.h"
 #include "src/core/lib/iomgr/exec_ctx.h"
 #include "src/core/lib/iomgr/polling_entity.h"
-#include "src/core/lib/iomgr/timer.h"
 #include "src/core/lib/resource_quota/arena.h"
 #include "src/core/lib/service_config/service_config.h"
 #include "src/core/lib/service_config/service_config_call_data.h"
@@ -132,6 +132,7 @@ namespace grpc_core {
 
 namespace {
 
+using grpc_event_engine::experimental::EventEngine;
 using internal::RetryGlobalConfig;
 using internal::RetryMethodConfig;
 using internal::RetryServiceConfigParser;
@@ -176,6 +177,7 @@ class RetryFilter {
 
   RetryFilter(const ChannelArgs& args, grpc_error_handle* error)
       : client_channel_(args.GetObject<ClientChannel>()),
+        event_engine_(args.GetObject<EventEngine>()),
         per_rpc_retry_buffer_size_(GetMaxPerRpcRetryBufferSize(args)),
         service_config_parser_index_(
             internal::RetryServiceConfigParser::ParserIndex()) {
@@ -212,6 +214,7 @@ class RetryFilter {
       const grpc_call_context_element* context);
 
   ClientChannel* client_channel_;
+  EventEngine* const event_engine_;
   size_t per_rpc_retry_buffer_size_;
   RefCountedPtr<ServerRetryThrottleData> retry_throttle_data_;
   const size_t service_config_parser_index_;
@@ -269,7 +272,7 @@ class RetryFilter::CallData {
     // We allocate one struct on the arena for each attempt at starting a
     // batch on a given LB call.
     class BatchData
-        : public RefCounted<BatchData, PolymorphicRefCount, kUnrefCallDtor> {
+        : public RefCounted<BatchData, PolymorphicRefCount, UnrefCallDtor> {
      public:
       BatchData(RefCountedPtr<CallAttempt> call_attempt, int refcount,
                 bool set_on_complete);
@@ -442,18 +445,17 @@ class RetryFilter::CallData {
     // Abandons the call attempt.  Unrefs any deferred batches.
     void Abandon();
 
-    static void OnPerAttemptRecvTimer(void* arg, grpc_error_handle error);
+    void OnPerAttemptRecvTimer();
     static void OnPerAttemptRecvTimerLocked(void* arg, grpc_error_handle error);
     void MaybeCancelPerAttemptRecvTimer();
 
     CallData* calld_;
     AttemptDispatchController attempt_dispatch_controller_;
-    OrphanablePtr<ClientChannel::LoadBalancedCall> lb_call_;
+    OrphanablePtr<ClientChannel::FilterBasedLoadBalancedCall> lb_call_;
     bool lb_call_committed_ = false;
 
-    grpc_timer per_attempt_recv_timer_;
     grpc_closure on_per_attempt_recv_timer_;
-    bool per_attempt_recv_timer_pending_ = false;
+    absl::optional<EventEngine::TaskHandle> per_attempt_recv_timer_handle_;
 
     // BatchData.batch.payload points to this.
     grpc_transport_stream_op_batch_payload batch_payload_;
@@ -548,14 +550,15 @@ class RetryFilter::CallData {
   // If server_pushback is nullopt, retry_backoff_ is used.
   void StartRetryTimer(absl::optional<Duration> server_pushback);
 
-  static void OnRetryTimer(void* arg, grpc_error_handle error);
-  static void OnRetryTimerLocked(void* arg, grpc_error_handle error);
+  void OnRetryTimer();
+  static void OnRetryTimerLocked(void* arg, grpc_error_handle /*error*/);
 
   // Adds a closure to closures to start a transparent retry.
   void AddClosureToStartTransparentRetry(CallCombinerClosureList* closures);
   static void StartTransparentRetry(void* arg, grpc_error_handle error);
 
-  OrphanablePtr<ClientChannel::LoadBalancedCall> CreateLoadBalancedCall(
+  OrphanablePtr<ClientChannel::FilterBasedLoadBalancedCall>
+  CreateLoadBalancedCall(
       ConfigSelector::CallDispatchController* call_dispatch_controller,
       bool is_transparent_retry);
 
@@ -586,7 +589,7 @@ class RetryFilter::CallData {
   // LB call used when we've committed to a call attempt and the retry
   // state for that attempt is no longer needed.  This provides a fast
   // path for long-running streaming calls that minimizes overhead.
-  OrphanablePtr<ClientChannel::LoadBalancedCall> committed_call_;
+  OrphanablePtr<ClientChannel::FilterBasedLoadBalancedCall> committed_call_;
 
   // When are are not yet fully committed to a particular call (i.e.,
   // either we might still retry or we have committed to the call but
@@ -601,26 +604,16 @@ class RetryFilter::CallData {
 
   // Retry state.
   bool retry_committed_ : 1;
-  bool retry_timer_pending_ : 1;
   bool retry_codepath_started_ : 1;
   bool sent_transparent_retry_not_seen_by_server_ : 1;
   int num_attempts_completed_ = 0;
-  grpc_timer retry_timer_;
+  absl::optional<EventEngine::TaskHandle> retry_timer_handle_;
   grpc_closure retry_closure_;
 
   // Cached data for retrying send ops.
   // send_initial_metadata
   bool seen_send_initial_metadata_ = false;
   grpc_metadata_batch send_initial_metadata_{arena_};
-  // TODO(roth): As part of implementing hedging, we'll probably need to
-  // have the LB call set a value in CallAttempt and then propagate it
-  // from CallAttempt to the parent call when we commit.  Otherwise, we
-  // may leave this with a value for a peer other than the one we
-  // actually commit to.  Alternatively, maybe see if there's a way to
-  // change the surface API such that the peer isn't available until
-  // after initial metadata is received?  (Could even change the
-  // transport API to return this with the recv_initial_metadata op.)
-  gpr_atm* peer_string_;
   // send_message
   // When we get a send_message op, we replace the original byte stream
   // with a CachingByteStream that caches the slices to a local buffer for
@@ -656,7 +649,7 @@ class RetryFilter::CallData {
 // on_call_stack_destruction closure from the surface.
 class RetryFilter::CallData::CallStackDestructionBarrier
     : public RefCounted<CallStackDestructionBarrier, PolymorphicRefCount,
-                        kUnrefCallDtor> {
+                        UnrefCallDtor> {
  public:
   CallStackDestructionBarrier() {}
 
@@ -723,23 +716,23 @@ RetryFilter::CallData::CallAttempt::CallAttempt(CallData* calld,
   // If per_attempt_recv_timeout is set, start a timer.
   if (calld->retry_policy_ != nullptr &&
       calld->retry_policy_->per_attempt_recv_timeout().has_value()) {
-    Timestamp per_attempt_recv_deadline =
-        Timestamp::Now() + *calld->retry_policy_->per_attempt_recv_timeout();
+    const Duration per_attempt_recv_timeout =
+        *calld->retry_policy_->per_attempt_recv_timeout();
     if (GRPC_TRACE_FLAG_ENABLED(grpc_retry_trace)) {
       gpr_log(GPR_INFO,
               "chand=%p calld=%p attempt=%p: per-attempt timeout in %" PRId64
               " ms",
-              calld->chand_, calld, this,
-              calld->retry_policy_->per_attempt_recv_timeout()->millis());
+              calld->chand_, calld, this, per_attempt_recv_timeout.millis());
     }
     // Schedule retry after computed delay.
-    GRPC_CLOSURE_INIT(&on_per_attempt_recv_timer_, OnPerAttemptRecvTimer, this,
-                      nullptr);
     GRPC_CALL_STACK_REF(calld->owning_call_, "OnPerAttemptRecvTimer");
     Ref(DEBUG_LOCATION, "OnPerAttemptRecvTimer").release();
-    per_attempt_recv_timer_pending_ = true;
-    grpc_timer_init(&per_attempt_recv_timer_, per_attempt_recv_deadline,
-                    &on_per_attempt_recv_timer_);
+    per_attempt_recv_timer_handle_ = calld_->chand_->event_engine_->RunAfter(
+        per_attempt_recv_timeout, [this] {
+          ApplicationCallbackExecCtx callback_exec_ctx;
+          ExecCtx exec_ctx;
+          OnPerAttemptRecvTimer();
+        });
   }
 }
 
@@ -803,7 +796,7 @@ void RetryFilter::CallData::CallAttempt::MaybeSwitchToFastPath() {
   // If we've already switched to fast path, there's nothing to do here.
   if (calld_->committed_call_ != nullptr) return;
   // If the perAttemptRecvTimeout timer is pending, we can't switch yet.
-  if (per_attempt_recv_timer_pending_) return;
+  if (per_attempt_recv_timer_handle_.has_value()) return;
   // If there are still send ops to replay, we can't switch yet.
   if (HaveSendOpsToReplay()) return;
   // If we started an internal batch for recv_trailing_metadata but have not
@@ -881,7 +874,7 @@ namespace {
 void StartBatchInCallCombiner(void* arg, grpc_error_handle /*ignored*/) {
   grpc_transport_stream_op_batch* batch =
       static_cast<grpc_transport_stream_op_batch*>(arg);
-  auto* lb_call = static_cast<ClientChannel::LoadBalancedCall*>(
+  auto* lb_call = static_cast<ClientChannel::FilterBasedLoadBalancedCall*>(
       batch->handler_private.extra_arg);
   // Note: This will release the call combiner.
   lb_call->StartTransportStreamOpBatch(batch);
@@ -895,7 +888,7 @@ void RetryFilter::CallData::CallAttempt::AddClosureForBatch(
   if (GRPC_TRACE_FLAG_ENABLED(grpc_retry_trace)) {
     gpr_log(GPR_INFO, "chand=%p calld=%p attempt=%p: adding batch (%s): %s",
             calld_->chand_, calld_, this, reason,
-            grpc_transport_stream_op_batch_string(batch).c_str());
+            grpc_transport_stream_op_batch_string(batch, false).c_str());
   }
   batch->handler_private.extra_arg = lb_call_.get();
   GRPC_CLOSURE_INIT(&batch->handler_private.closure, StartBatchInCallCombiner,
@@ -1249,14 +1242,11 @@ void RetryFilter::CallData::CallAttempt::Abandon() {
   on_complete_deferred_batches_.clear();
 }
 
-void RetryFilter::CallData::CallAttempt::OnPerAttemptRecvTimer(
-    void* arg, grpc_error_handle error) {
-  auto* call_attempt = static_cast<CallAttempt*>(arg);
-  GRPC_CLOSURE_INIT(&call_attempt->on_per_attempt_recv_timer_,
-                    OnPerAttemptRecvTimerLocked, call_attempt, nullptr);
-  GRPC_CALL_COMBINER_START(call_attempt->calld_->call_combiner_,
-                           &call_attempt->on_per_attempt_recv_timer_, error,
-                           "per-attempt timer fired");
+void RetryFilter::CallData::CallAttempt::OnPerAttemptRecvTimer() {
+  GRPC_CLOSURE_INIT(&on_per_attempt_recv_timer_, OnPerAttemptRecvTimerLocked,
+                    this, nullptr);
+  GRPC_CALL_COMBINER_START(calld_->call_combiner_, &on_per_attempt_recv_timer_,
+                           absl::OkStatus(), "per-attempt timer fired");
 }
 
 void RetryFilter::CallData::CallAttempt::OnPerAttemptRecvTimerLocked(
@@ -1266,35 +1256,33 @@ void RetryFilter::CallData::CallAttempt::OnPerAttemptRecvTimerLocked(
   if (GRPC_TRACE_FLAG_ENABLED(grpc_retry_trace)) {
     gpr_log(GPR_INFO,
             "chand=%p calld=%p attempt=%p: perAttemptRecvTimeout timer fired: "
-            "error=%s, per_attempt_recv_timer_pending_=%d",
+            "error=%s, per_attempt_recv_timer_handle_.has_value()=%d",
             calld->chand_, calld, call_attempt, StatusToString(error).c_str(),
-            call_attempt->per_attempt_recv_timer_pending_);
+            call_attempt->per_attempt_recv_timer_handle_.has_value());
   }
   CallCombinerClosureList closures;
-  if (error.ok() && call_attempt->per_attempt_recv_timer_pending_) {
-    call_attempt->per_attempt_recv_timer_pending_ = false;
-    // Cancel this attempt.
-    // TODO(roth): When implementing hedging, we should not cancel the
-    // current attempt.
-    call_attempt->MaybeAddBatchForCancelOp(
-        grpc_error_set_int(
-            GRPC_ERROR_CREATE("retry perAttemptRecvTimeout exceeded"),
-            StatusIntProperty::kRpcStatus, GRPC_STATUS_CANCELLED),
-        &closures);
-    // Check whether we should retry.
-    if (call_attempt->ShouldRetry(/*status=*/absl::nullopt,
-                                  /*server_pushback_ms=*/absl::nullopt)) {
-      // Mark current attempt as abandoned.
-      call_attempt->Abandon();
-      // We are retrying.  Start backoff timer.
-      calld->StartRetryTimer(/*server_pushback=*/absl::nullopt);
-    } else {
-      // Not retrying, so commit the call.
-      calld->RetryCommit(call_attempt);
-      // If retry state is no longer needed, switch to fast path for
-      // subsequent batches.
-      call_attempt->MaybeSwitchToFastPath();
-    }
+  call_attempt->per_attempt_recv_timer_handle_.reset();
+  // Cancel this attempt.
+  // TODO(roth): When implementing hedging, we should not cancel the
+  // current attempt.
+  call_attempt->MaybeAddBatchForCancelOp(
+      grpc_error_set_int(
+          GRPC_ERROR_CREATE("retry perAttemptRecvTimeout exceeded"),
+          StatusIntProperty::kRpcStatus, GRPC_STATUS_CANCELLED),
+      &closures);
+  // Check whether we should retry.
+  if (call_attempt->ShouldRetry(/*status=*/absl::nullopt,
+                                /*server_pushback_ms=*/absl::nullopt)) {
+    // Mark current attempt as abandoned.
+    call_attempt->Abandon();
+    // We are retrying.  Start backoff timer.
+    calld->StartRetryTimer(/*server_pushback=*/absl::nullopt);
+  } else {
+    // Not retrying, so commit the call.
+    calld->RetryCommit(call_attempt);
+    // If retry state is no longer needed, switch to fast path for
+    // subsequent batches.
+    call_attempt->MaybeSwitchToFastPath();
   }
   closures.RunClosures(calld->call_combiner_);
   call_attempt->Unref(DEBUG_LOCATION, "OnPerAttemptRecvTimer");
@@ -1302,15 +1290,19 @@ void RetryFilter::CallData::CallAttempt::OnPerAttemptRecvTimerLocked(
 }
 
 void RetryFilter::CallData::CallAttempt::MaybeCancelPerAttemptRecvTimer() {
-  if (per_attempt_recv_timer_pending_) {
+  if (per_attempt_recv_timer_handle_.has_value()) {
     if (GRPC_TRACE_FLAG_ENABLED(grpc_retry_trace)) {
       gpr_log(GPR_INFO,
               "chand=%p calld=%p attempt=%p: cancelling "
               "perAttemptRecvTimeout timer",
               calld_->chand_, calld_, this);
     }
-    per_attempt_recv_timer_pending_ = false;
-    grpc_timer_cancel(&per_attempt_recv_timer_);
+    if (calld_->chand_->event_engine_->Cancel(
+            *per_attempt_recv_timer_handle_)) {
+      Unref(DEBUG_LOCATION, "OnPerAttemptRecvTimer");
+      GRPC_CALL_STACK_UNREF(calld_->owning_call_, "OnPerAttemptRecvTimer");
+    }
+    per_attempt_recv_timer_handle_.reset();
   }
 }
 
@@ -1351,8 +1343,9 @@ RetryFilter::CallData::CallAttempt::BatchData::~BatchData() {
             this);
   }
   CallAttempt* call_attempt = std::exchange(call_attempt_, nullptr);
-  GRPC_CALL_STACK_UNREF(call_attempt->calld_->owning_call_, "Retry BatchData");
+  grpc_call_stack* owning_call = call_attempt->calld_->owning_call_;
   call_attempt->Unref(DEBUG_LOCATION, "~BatchData");
+  GRPC_CALL_STACK_UNREF(owning_call, "Retry BatchData");
 }
 
 void RetryFilter::CallData::CallAttempt::BatchData::
@@ -1879,7 +1872,8 @@ void RetryFilter::CallData::CallAttempt::BatchData::OnComplete(
             "got on_complete, error=%s, batch=%s",
             calld->chand_, calld, call_attempt, batch_data.get(),
             StatusToString(error).c_str(),
-            grpc_transport_stream_op_batch_string(&batch_data->batch_).c_str());
+            grpc_transport_stream_op_batch_string(&batch_data->batch_, false)
+                .c_str());
   }
   // If this attempt has been abandoned, then we're not going to propagate
   // the completion of this batch, so do nothing.
@@ -1954,7 +1948,8 @@ void RetryFilter::CallData::CallAttempt::BatchData::OnCompleteForCancelOp(
             "got on_complete for cancel_stream batch, error=%s, batch=%s",
             calld->chand_, calld, call_attempt, batch_data.get(),
             StatusToString(error).c_str(),
-            grpc_transport_stream_op_batch_string(&batch_data->batch_).c_str());
+            grpc_transport_stream_op_batch_string(&batch_data->batch_, false)
+                .c_str());
   }
   GRPC_CALL_COMBINER_STOP(
       calld->call_combiner_,
@@ -1986,7 +1981,6 @@ void RetryFilter::CallData::CallAttempt::BatchData::
   batch_.send_initial_metadata = true;
   batch_.payload->send_initial_metadata.send_initial_metadata =
       &call_attempt_->send_initial_metadata_;
-  batch_.payload->send_initial_metadata.peer_string = calld->peer_string_;
 }
 
 void RetryFilter::CallData::CallAttempt::BatchData::
@@ -2160,7 +2154,6 @@ RetryFilter::CallData::CallData(RetryFilter* chand,
       pending_send_message_(false),
       pending_send_trailing_metadata_(false),
       retry_committed_(false),
-      retry_timer_pending_(false),
       retry_codepath_started_(false),
       sent_transparent_retry_not_seen_by_server_(false) {}
 
@@ -2178,7 +2171,8 @@ void RetryFilter::CallData::StartTransportStreamOpBatch(
   if (GRPC_TRACE_FLAG_ENABLED(grpc_retry_trace) &&
       !GRPC_TRACE_FLAG_ENABLED(grpc_trace_channel)) {
     gpr_log(GPR_INFO, "chand=%p calld=%p: batch started from surface: %s",
-            chand_, this, grpc_transport_stream_op_batch_string(batch).c_str());
+            chand_, this,
+            grpc_transport_stream_op_batch_string(batch, false).c_str());
   }
   // If we have an LB call, delegate to the LB call.
   if (committed_call_ != nullptr) {
@@ -2220,13 +2214,15 @@ void RetryFilter::CallData::StartTransportStreamOpBatch(
       return;
     }
     // Cancel retry timer if needed.
-    if (retry_timer_pending_) {
+    if (retry_timer_handle_.has_value()) {
       if (GRPC_TRACE_FLAG_ENABLED(grpc_retry_trace)) {
         gpr_log(GPR_INFO, "chand=%p calld=%p: cancelling retry timer", chand_,
                 this);
       }
-      retry_timer_pending_ = false;  // Lame timer callback.
-      grpc_timer_cancel(&retry_timer_);
+      if (chand_->event_engine_->Cancel(*retry_timer_handle_)) {
+        GRPC_CALL_STACK_UNREF(owning_call_, "OnRetryTimer");
+      }
+      retry_timer_handle_.reset();
       FreeAllCachedSendOpData();
     }
     // We have no call attempt, so there's nowhere to send the cancellation
@@ -2240,7 +2236,7 @@ void RetryFilter::CallData::StartTransportStreamOpBatch(
   PendingBatch* pending = PendingBatchesAdd(batch);
   // If the timer is pending, yield the call combiner and wait for it to
   // run, since we don't want to start another call attempt until it does.
-  if (retry_timer_pending_) {
+  if (retry_timer_handle_.has_value()) {
     GRPC_CALL_COMBINER_STOP(call_combiner_,
                             "added pending batch while retry timer pending");
     return;
@@ -2300,7 +2296,7 @@ void RetryFilter::CallData::StartTransportStreamOpBatch(
   call_attempt_->StartRetriableBatches();
 }
 
-OrphanablePtr<ClientChannel::LoadBalancedCall>
+OrphanablePtr<ClientChannel::FilterBasedLoadBalancedCall>
 RetryFilter::CallData::CreateLoadBalancedCall(
     ConfigSelector::CallDispatchController* call_dispatch_controller,
     bool is_transparent_retry) {
@@ -2334,7 +2330,6 @@ void RetryFilter::CallData::MaybeCacheSendOpsForBatch(PendingBatch* pending) {
     grpc_metadata_batch* send_initial_metadata =
         batch->payload->send_initial_metadata.send_initial_metadata;
     send_initial_metadata_ = send_initial_metadata->Copy();
-    peer_string_ = batch->payload->send_initial_metadata.peer_string;
   }
   // Set up cache for send_message ops.
   if (batch->send_message) {
@@ -2575,42 +2570,40 @@ void RetryFilter::CallData::StartRetryTimer(
   // Reset call attempt.
   call_attempt_.reset(DEBUG_LOCATION, "StartRetryTimer");
   // Compute backoff delay.
-  Timestamp next_attempt_time;
+  Duration next_attempt_timeout;
   if (server_pushback.has_value()) {
     GPR_ASSERT(*server_pushback >= Duration::Zero());
-    next_attempt_time = Timestamp::Now() + *server_pushback;
+    next_attempt_timeout = *server_pushback;
     retry_backoff_.Reset();
   } else {
-    next_attempt_time = retry_backoff_.NextAttemptTime();
+    next_attempt_timeout = retry_backoff_.NextAttemptTime() - Timestamp::Now();
   }
   if (GRPC_TRACE_FLAG_ENABLED(grpc_retry_trace)) {
     gpr_log(GPR_INFO,
             "chand=%p calld=%p: retrying failed call in %" PRId64 " ms", chand_,
-            this, (next_attempt_time - Timestamp::Now()).millis());
+            this, next_attempt_timeout.millis());
   }
   // Schedule retry after computed delay.
-  GRPC_CLOSURE_INIT(&retry_closure_, OnRetryTimer, this, nullptr);
   GRPC_CALL_STACK_REF(owning_call_, "OnRetryTimer");
-  retry_timer_pending_ = true;
-  grpc_timer_init(&retry_timer_, next_attempt_time, &retry_closure_);
+  retry_timer_handle_ =
+      chand_->event_engine_->RunAfter(next_attempt_timeout, [this] {
+        ApplicationCallbackExecCtx callback_exec_ctx;
+        ExecCtx exec_ctx;
+        OnRetryTimer();
+      });
 }
 
-void RetryFilter::CallData::OnRetryTimer(void* arg, grpc_error_handle error) {
-  auto* calld = static_cast<CallData*>(arg);
-  GRPC_CLOSURE_INIT(&calld->retry_closure_, OnRetryTimerLocked, calld, nullptr);
-  GRPC_CALL_COMBINER_START(calld->call_combiner_, &calld->retry_closure_, error,
+void RetryFilter::CallData::OnRetryTimer() {
+  GRPC_CLOSURE_INIT(&retry_closure_, OnRetryTimerLocked, this, nullptr);
+  GRPC_CALL_COMBINER_START(call_combiner_, &retry_closure_, absl::OkStatus(),
                            "retry timer fired");
 }
 
 void RetryFilter::CallData::OnRetryTimerLocked(void* arg,
-                                               grpc_error_handle error) {
+                                               grpc_error_handle /*error*/) {
   auto* calld = static_cast<CallData*>(arg);
-  if (error.ok() && calld->retry_timer_pending_) {
-    calld->retry_timer_pending_ = false;
-    calld->CreateCallAttempt(/*is_transparent_retry=*/false);
-  } else {
-    GRPC_CALL_COMBINER_STOP(calld->call_combiner_, "retry timer cancelled");
-  }
+  calld->retry_timer_handle_.reset();
+  calld->CreateCallAttempt(/*is_transparent_retry=*/false);
   GRPC_CALL_STACK_UNREF(calld->owning_call_, "OnRetryTimer");
 }
 

data/src/core/ext/filters/client_channel/subchannel.cc

@@ -920,7 +920,8 @@ bool Subchannel::PublishTransportLocked() {
   // Construct channel stack.
   ChannelStackBuilderImpl builder("subchannel", GRPC_CLIENT_SUBCHANNEL,
                                   connecting_result_.channel_args);
-  builder.SetTransport(connecting_result_.transport);
+  // Builder takes ownership of transport.
+  builder.SetTransport(std::exchange(connecting_result_.transport, nullptr));
   if (!CoreConfiguration::Get().channel_init().CreateStack(&builder)) {
     return false;
   }
@@ -933,9 +934,6 @@ bool Subchannel::PublishTransportLocked() {
             key_.ToString().c_str(), StatusToString(error).c_str());
     return false;
   }
-  // Release the ownership since it is now owned by the connected filter in the
-  // channel stack (published).
-  connecting_result_.transport = nullptr;
   RefCountedPtr<channelz::SocketNode> socket =
       std::move(connecting_result_.socket_node);
   connecting_result_.Reset();

data/src/core/ext/filters/client_channel/subchannel_stream_client.cc

@@ -33,6 +33,7 @@
 #include "src/core/lib/gprpp/status_helper.h"
 #include "src/core/lib/gprpp/sync.h"
 #include "src/core/lib/gprpp/time.h"
+#include "src/core/lib/iomgr/exec_ctx.h"
 #include "src/core/lib/resource_quota/resource_quota.h"
 #include "src/core/lib/transport/error_utils.h"
 
@@ -43,6 +44,8 @@
 
 namespace grpc_core {
 
+using ::grpc_event_engine::experimental::EventEngine;
+
 //
 // SubchannelStreamClient
 //
@@ -69,12 +72,11 @@ SubchannelStreamClient::SubchannelStreamClient(
               .set_multiplier(SUBCHANNEL_STREAM_RECONNECT_BACKOFF_MULTIPLIER)
               .set_jitter(SUBCHANNEL_STREAM_RECONNECT_JITTER)
               .set_max_backoff(Duration::Seconds(
-                  SUBCHANNEL_STREAM_RECONNECT_MAX_BACKOFF_SECONDS))) {
+                  SUBCHANNEL_STREAM_RECONNECT_MAX_BACKOFF_SECONDS))),
+      event_engine_(connected_subchannel_->args().GetObject<EventEngine>()) {
   if (GPR_UNLIKELY(tracer_ != nullptr)) {
     gpr_log(GPR_INFO, "%s %p: created SubchannelStreamClient", tracer_, this);
   }
-  GRPC_CLOSURE_INIT(&retry_timer_callback_, OnRetryTimer, this,
-                    grpc_schedule_on_exec_ctx);
   StartCall();
 }
 
@@ -94,8 +96,9 @@ void SubchannelStreamClient::Orphan() {
     MutexLock lock(&mu_);
     event_handler_.reset();
     call_state_.reset();
-    if (retry_timer_callback_pending_) {
-      grpc_timer_cancel(&retry_timer_);
+    if (retry_timer_handle_.has_value()) {
+      event_engine_->Cancel(*retry_timer_handle_);
+      retry_timer_handle_.reset();
     }
   }
   Unref(DEBUG_LOCATION, "orphan");
@@ -124,11 +127,10 @@ void SubchannelStreamClient::StartRetryTimerLocked() {
   if (event_handler_ != nullptr) {
     event_handler_->OnRetryTimerStartLocked(this);
   }
-  Timestamp next_try = retry_backoff_.NextAttemptTime();
+  const Duration timeout = retry_backoff_.NextAttemptTime() - Timestamp::Now();
   if (GPR_UNLIKELY(tracer_ != nullptr)) {
     gpr_log(GPR_INFO, "%s %p: SubchannelStreamClient health check call lost...",
             tracer_, this);
-    Duration timeout = next_try - Timestamp::Now();
     if (timeout > Duration::Zero()) {
       gpr_log(GPR_INFO, "%s %p: ... will retry in %" PRId64 "ms.", tracer_,
               this, timeout.millis());
@@ -136,28 +138,27 @@ void SubchannelStreamClient::StartRetryTimerLocked() {
       gpr_log(GPR_INFO, "%s %p: ... retrying immediately.", tracer_, this);
     }
   }
-  // Ref for callback, tracked manually.
-  Ref(DEBUG_LOCATION, "health_retry_timer").release();
-  retry_timer_callback_pending_ = true;
-  grpc_timer_init(&retry_timer_, next_try, &retry_timer_callback_);
+  retry_timer_handle_ = event_engine_->RunAfter(
+      timeout, [self = Ref(DEBUG_LOCATION, "health_retry_timer")]() mutable {
+        ApplicationCallbackExecCtx callback_exec_ctx;
+        ExecCtx exec_ctx;
+        self->OnRetryTimer();
+        self.reset(DEBUG_LOCATION, "health_retry_timer");
+      });
 }
 
-void SubchannelStreamClient::OnRetryTimer(void* arg, grpc_error_handle error) {
-  auto* self = static_cast<SubchannelStreamClient*>(arg);
-  {
-    MutexLock lock(&self->mu_);
-    self->retry_timer_callback_pending_ = false;
-    if (self->event_handler_ != nullptr && error.ok() &&
-        self->call_state_ == nullptr) {
-      if (GPR_UNLIKELY(self->tracer_ != nullptr)) {
-        gpr_log(GPR_INFO,
-                "%s %p: SubchannelStreamClient restarting health check call",
-                self->tracer_, self);
-      }
-      self->StartCallLocked();
+void SubchannelStreamClient::OnRetryTimer() {
+  MutexLock lock(&mu_);
+  if (event_handler_ != nullptr && retry_timer_handle_.has_value() &&
+      call_state_ == nullptr) {
+    if (GPR_UNLIKELY(tracer_ != nullptr)) {
+      gpr_log(GPR_INFO,
+              "%s %p: SubchannelStreamClient restarting health check call",
+              tracer_, this);
     }
+    StartCallLocked();
   }
-  self->Unref(DEBUG_LOCATION, "health_retry_timer");
+  retry_timer_handle_.reset();
 }
 
 //
@@ -242,7 +243,6 @@ void SubchannelStreamClient::CallState::StartCallLocked() {
   GPR_ASSERT(error.ok());
   payload_.send_initial_metadata.send_initial_metadata =
       &send_initial_metadata_;
-  payload_.send_initial_metadata.peer_string = nullptr;
   batch_.send_initial_metadata = true;
   // Add send_message op.
   send_message_.Append(Slice(
@@ -257,7 +257,6 @@ void SubchannelStreamClient::CallState::StartCallLocked() {
   payload_.recv_initial_metadata.recv_initial_metadata =
       &recv_initial_metadata_;
   payload_.recv_initial_metadata.trailing_metadata_available = nullptr;
-  payload_.recv_initial_metadata.peer_string = nullptr;
   // recv_initial_metadata_ready callback takes ref, handled manually.
   call_->Ref(DEBUG_LOCATION, "recv_initial_metadata_ready").release();
   payload_.recv_initial_metadata.recv_initial_metadata_ready =

data/src/core/ext/filters/client_channel/subchannel_stream_client.h

@@ -27,6 +27,7 @@
 #include "absl/strings/string_view.h"
 #include "absl/types/optional.h"
 
+#include <grpc/event_engine/event_engine.h>
 #include <grpc/event_engine/memory_allocator.h>
 #include <grpc/slice.h>
 #include <grpc/status.h>
@@ -42,7 +43,6 @@
 #include "src/core/lib/iomgr/error.h"
 #include "src/core/lib/iomgr/iomgr_fwd.h"
 #include "src/core/lib/iomgr/polling_entity.h"
-#include "src/core/lib/iomgr/timer.h"
 #include "src/core/lib/resource_quota/arena.h"
 #include "src/core/lib/resource_quota/memory_quota.h"
 #include "src/core/lib/slice/slice.h"
@@ -196,7 +196,7 @@ class SubchannelStreamClient
   void StartCallLocked() ABSL_EXCLUSIVE_LOCKS_REQUIRED(&mu_);
 
   void StartRetryTimerLocked() ABSL_EXCLUSIVE_LOCKS_REQUIRED(&mu_);
-  static void OnRetryTimer(void* arg, grpc_error_handle error);
+  void OnRetryTimer() ABSL_LOCKS_EXCLUDED(mu_);
 
   RefCountedPtr<ConnectedSubchannel> connected_subchannel_;
   grpc_pollset_set* interested_parties_;  // Do not own.
@@ -212,9 +212,12 @@ class SubchannelStreamClient
 
   // Call retry state.
   BackOff retry_backoff_ ABSL_GUARDED_BY(mu_);
-  grpc_timer retry_timer_ ABSL_GUARDED_BY(mu_);
-  grpc_closure retry_timer_callback_ ABSL_GUARDED_BY(mu_);
-  bool retry_timer_callback_pending_ ABSL_GUARDED_BY(mu_) = false;
+  absl::optional<grpc_event_engine::experimental::EventEngine::TaskHandle>
+      retry_timer_handle_ ABSL_GUARDED_BY(mu_);
+  // A raw pointer will suffice since connected_subchannel_ holds a copy of the
+  // ChannelArgs which holds an std::shared_ptr of the EventEngine.
+  grpc_event_engine::experimental::EventEngine* event_engine_
+      ABSL_GUARDED_BY(mu_);
 };
 
 }  // namespace grpc_core

data/src/core/ext/filters/http/client/http_client_filter.cc

@@ -133,13 +133,13 @@ ArenaPromise<ServerMetadataHandle> HttpClientFilter::MakeCallPromise(
         return std::move(md);
       });
 
-  return Race(Map(next_promise_factory(std::move(call_args)),
+  return Race(initial_metadata_err->Wait(),
+              Map(next_promise_factory(std::move(call_args)),
                   [](ServerMetadataHandle md) -> ServerMetadataHandle {
                     auto r = CheckServerMetadata(md.get());
                     if (!r.ok()) return ServerMetadataFromStatus(r);
                     return md;
-                  }),
-              initial_metadata_err->Wait());
+                  }));
 }
 
 HttpClientFilter::HttpClientFilter(HttpSchemeMetadata::ValueType scheme,