grpc 1.50.0.pre1 → 1.52.0.pre2

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

@@ -24,22 +24,22 @@
 #include <algorithm>
 #include <atomic>
 #include <cmath>
-#include <cstdint>
 #include <map>
 #include <memory>
 #include <set>
 #include <string>
+#include <type_traits>
 #include <utility>
 #include <vector>
 
-#include "absl/memory/memory.h"
 #include "absl/random/random.h"
 #include "absl/status/status.h"
 #include "absl/status/statusor.h"
 #include "absl/strings/string_view.h"
 #include "absl/types/variant.h"
 
-#include <grpc/impl/codegen/connectivity_state.h>
+#include <grpc/event_engine/event_engine.h>
+#include <grpc/impl/connectivity_state.h>
 #include <grpc/support/log.h>
 
 #include "src/core/ext/filters/client_channel/lb_policy/child_policy_handler.h"
@@ -47,19 +47,15 @@
 #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/env.h"
 #include "src/core/lib/gprpp/orphanable.h"
 #include "src/core/lib/gprpp/ref_counted.h"
 #include "src/core/lib/gprpp/ref_counted_ptr.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"
+#include "src/core/lib/iomgr/exec_ctx.h"
 #include "src/core/lib/iomgr/iomgr_fwd.h"
 #include "src/core/lib/iomgr/pollset_set.h"
-#include "src/core/lib/iomgr/timer.h"
 #include "src/core/lib/json/json.h"
 #include "src/core/lib/load_balancing/lb_policy.h"
 #include "src/core/lib/load_balancing/lb_policy_factory.h"
@@ -72,17 +68,10 @@ namespace grpc_core {
 
 TraceFlag grpc_outlier_detection_lb_trace(false, "outlier_detection_lb");
 
-// TODO(donnadionne): Remove once outlier detection is no longer experimental
-bool XdsOutlierDetectionEnabled() {
-  auto value = GetEnv("GRPC_EXPERIMENTAL_ENABLE_OUTLIER_DETECTION");
-  if (!value.has_value()) return false;
-  bool parsed_value;
-  bool parse_succeeded = gpr_parse_bool_value(value->c_str(), &parsed_value);
-  return parse_succeeded && parsed_value;
-}
-
 namespace {
 
+using ::grpc_event_engine::experimental::EventEngine;
+
 constexpr absl::string_view kOutlierDetection =
     "outlier_detection_experimental";
 
@@ -307,8 +296,8 @@ class OutlierDetectionLb : public LoadBalancingPolicy {
     }
 
    private:
-    std::unique_ptr<Bucket> current_bucket_ = absl::make_unique<Bucket>();
-    std::unique_ptr<Bucket> backup_bucket_ = absl::make_unique<Bucket>();
+    std::unique_ptr<Bucket> current_bucket_ = std::make_unique<Bucket>();
+    std::unique_ptr<Bucket> backup_bucket_ = std::make_unique<Bucket>();
     // The bucket used to update call counts.
     // Points to either current_bucket or active_bucket.
     std::atomic<Bucket*> active_bucket_{current_bucket_.get()};
@@ -317,28 +306,17 @@ class OutlierDetectionLb : public LoadBalancingPolicy {
     std::set<SubchannelWrapper*> subchannels_;
   };
 
-  // A simple wrapper for ref-counting a picker from the child policy.
-  class RefCountedPicker : public RefCounted<RefCountedPicker> {
-   public:
-    explicit RefCountedPicker(std::unique_ptr<SubchannelPicker> picker)
-        : picker_(std::move(picker)) {}
-    PickResult Pick(PickArgs args) { return picker_->Pick(args); }
-
-   private:
-    std::unique_ptr<SubchannelPicker> picker_;
-  };
-
   // A picker that wraps the picker from the child to perform outlier detection.
   class Picker : public SubchannelPicker {
    public:
     Picker(OutlierDetectionLb* outlier_detection_lb,
-           RefCountedPtr<RefCountedPicker> picker, bool counting_enabled);
+           RefCountedPtr<SubchannelPicker> picker, bool counting_enabled);
 
     PickResult Pick(PickArgs args) override;
 
    private:
     class SubchannelCallTracker;
-    RefCountedPtr<RefCountedPicker> picker_;
+    RefCountedPtr<SubchannelPicker> picker_;
     bool counting_enabled_;
   };
 
@@ -354,9 +332,10 @@ class OutlierDetectionLb : public LoadBalancingPolicy {
     RefCountedPtr<SubchannelInterface> CreateSubchannel(
         ServerAddress address, const ChannelArgs& args) override;
     void UpdateState(grpc_connectivity_state state, const absl::Status& status,
-                     std::unique_ptr<SubchannelPicker> picker) override;
+                     RefCountedPtr<SubchannelPicker> picker) override;
     void RequestReresolution() override;
     absl::string_view GetAuthority() override;
+    grpc_event_engine::experimental::EventEngine* GetEventEngine() override;
     void AddTraceEvent(TraceSeverity severity,
                        absl::string_view message) override;
 
@@ -374,13 +353,10 @@ class OutlierDetectionLb : public LoadBalancingPolicy {
     Timestamp StartTime() const { return start_time_; }
 
    private:
-    static void OnTimer(void* arg, grpc_error_handle error);
-    void OnTimerLocked(grpc_error_handle);
+    void OnTimerLocked();
 
     RefCountedPtr<OutlierDetectionLb> parent_;
-    grpc_timer timer_;
-    grpc_closure on_timer_;
-    bool timer_pending_ = true;
+    absl::optional<EventEngine::TaskHandle> timer_handle_;
     Timestamp start_time_;
     absl::BitGen bit_gen_;
   };
@@ -407,7 +383,7 @@ class OutlierDetectionLb : public LoadBalancingPolicy {
   // Latest state and picker reported by the child policy.
   grpc_connectivity_state state_ = GRPC_CHANNEL_IDLE;
   absl::Status status_;
-  RefCountedPtr<RefCountedPicker> picker_;
+  RefCountedPtr<SubchannelPicker> picker_;
   std::map<std::string, RefCountedPtr<SubchannelState>> subchannel_state_map_;
   OrphanablePtr<EjectionTimer> ejection_timer_;
 };
@@ -434,7 +410,7 @@ void OutlierDetectionLb::SubchannelWrapper::WatchConnectivityState(
     std::unique_ptr<ConnectivityStateWatcherInterface> watcher) {
   ConnectivityStateWatcherInterface* watcher_ptr = watcher.get();
   auto watcher_wrapper =
-      absl::make_unique<WatcherWrapper>(std::move(watcher), ejected_);
+      std::make_unique<WatcherWrapper>(std::move(watcher), ejected_);
   watchers_.emplace(watcher_ptr, watcher_wrapper.get());
   wrapped_subchannel()->WatchConnectivityState(std::move(watcher_wrapper));
 }
@@ -501,7 +477,7 @@ class OutlierDetectionLb::Picker::SubchannelCallTracker
 //
 
 OutlierDetectionLb::Picker::Picker(OutlierDetectionLb* outlier_detection_lb,
-                                   RefCountedPtr<RefCountedPicker> picker,
+                                   RefCountedPtr<SubchannelPicker> picker,
                                    bool counting_enabled)
     : picker_(std::move(picker)), counting_enabled_(counting_enabled) {
   if (GRPC_TRACE_FLAG_ENABLED(grpc_outlier_detection_lb_trace)) {
@@ -530,7 +506,7 @@ LoadBalancingPolicy::PickResult OutlierDetectionLb::Picker::Pick(
     // not both success_rate_ejection and failure_percentage_ejection are unset.
     if (counting_enabled_) {
       complete_pick->subchannel_call_tracker =
-          absl::make_unique<SubchannelCallTracker>(
+          std::make_unique<SubchannelCallTracker>(
               std::move(complete_pick->subchannel_call_tracker),
               subchannel_wrapper->subchannel_state());
     }
@@ -693,7 +669,7 @@ absl::Status OutlierDetectionLb::UpdateLocked(UpdateArgs args) {
 void OutlierDetectionLb::MaybeUpdatePickerLocked() {
   if (picker_ != nullptr) {
     auto outlier_detection_picker =
-        absl::make_unique<Picker>(this, picker_, config_->CountingEnabled());
+        MakeRefCounted<Picker>(this, picker_, config_->CountingEnabled());
     if (GRPC_TRACE_FLAG_ENABLED(grpc_outlier_detection_lb_trace)) {
       gpr_log(GPR_INFO,
               "[outlier_detection_lb %p] updating connectivity: state=%s "
@@ -712,7 +688,7 @@ OrphanablePtr<LoadBalancingPolicy> OutlierDetectionLb::CreateChildPolicyLocked(
   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"));
+      std::make_unique<Helper>(Ref(DEBUG_LOCATION, "Helper"));
   OrphanablePtr<LoadBalancingPolicy> lb_policy =
       MakeOrphanable<ChildPolicyHandler>(std::move(lb_policy_args),
                                          &grpc_outlier_detection_lb_trace);
@@ -754,7 +730,7 @@ RefCountedPtr<SubchannelInterface> OutlierDetectionLb::Helper::CreateSubchannel(
 
 void OutlierDetectionLb::Helper::UpdateState(
     grpc_connectivity_state state, const absl::Status& status,
-    std::unique_ptr<SubchannelPicker> picker) {
+    RefCountedPtr<SubchannelPicker> picker) {
   if (outlier_detection_policy_->shutting_down_) return;
   if (GRPC_TRACE_FLAG_ENABLED(grpc_outlier_detection_lb_trace)) {
     gpr_log(GPR_INFO,
@@ -766,8 +742,7 @@ void OutlierDetectionLb::Helper::UpdateState(
   // Save the state and picker.
   outlier_detection_policy_->state_ = state;
   outlier_detection_policy_->status_ = status;
-  outlier_detection_policy_->picker_ =
-      MakeRefCounted<RefCountedPicker>(std::move(picker));
+  outlier_detection_policy_->picker_ = std::move(picker);
   // Wrap the picker and return it to the channel.
   outlier_detection_policy_->MaybeUpdatePickerLocked();
 }
@@ -781,6 +756,11 @@ absl::string_view OutlierDetectionLb::Helper::GetAuthority() {
   return outlier_detection_policy_->channel_control_helper()->GetAuthority();
 }
 
+grpc_event_engine::experimental::EventEngine*
+OutlierDetectionLb::Helper::GetEventEngine() {
+  return outlier_detection_policy_->channel_control_helper()->GetEventEngine();
+}
+
 void OutlierDetectionLb::Helper::AddTraceEvent(TraceSeverity severity,
                                                absl::string_view message) {
   if (outlier_detection_policy_->shutting_down_) return;
@@ -800,212 +780,201 @@ OutlierDetectionLb::EjectionTimer::EjectionTimer(
     gpr_log(GPR_INFO, "[outlier_detection_lb %p] ejection timer will run in %s",
             parent_.get(), interval.ToString().c_str());
   }
-  GRPC_CLOSURE_INIT(&on_timer_, OnTimer, this, nullptr);
-  Ref().release();
-  grpc_timer_init(&timer_, start_time_ + interval, &on_timer_);
+  timer_handle_ = parent_->channel_control_helper()->GetEventEngine()->RunAfter(
+      interval, [self = Ref(DEBUG_LOCATION, "EjectionTimer")]() mutable {
+        ApplicationCallbackExecCtx callback_exec_ctx;
+        ExecCtx exec_ctx;
+        auto self_ptr = self.get();
+        self_ptr->parent_->work_serializer()->Run(
+            [self = std::move(self)]() { self->OnTimerLocked(); },
+            DEBUG_LOCATION);
+      });
 }
 
 void OutlierDetectionLb::EjectionTimer::Orphan() {
-  if (timer_pending_) {
-    timer_pending_ = false;
-    grpc_timer_cancel(&timer_);
+  if (timer_handle_.has_value()) {
+    parent_->channel_control_helper()->GetEventEngine()->Cancel(*timer_handle_);
+    timer_handle_.reset();
   }
   Unref();
 }
 
-void OutlierDetectionLb::EjectionTimer::OnTimer(void* arg,
-                                                grpc_error_handle error) {
-  auto* self = static_cast<EjectionTimer*>(arg);
-  (void)GRPC_ERROR_REF(error);  // ref owned by lambda
-  self->parent_->work_serializer()->Run(
-      [self, error]() { self->OnTimerLocked(error); }, DEBUG_LOCATION);
-}
-
-void OutlierDetectionLb::EjectionTimer::OnTimerLocked(grpc_error_handle error) {
-  if (GRPC_ERROR_IS_NONE(error) && timer_pending_) {
-    if (GRPC_TRACE_FLAG_ENABLED(grpc_outlier_detection_lb_trace)) {
-      gpr_log(GPR_INFO, "[outlier_detection_lb %p] ejection timer running",
-              parent_.get());
+void OutlierDetectionLb::EjectionTimer::OnTimerLocked() {
+  if (!timer_handle_.has_value()) return;
+  timer_handle_.reset();
+  if (GRPC_TRACE_FLAG_ENABLED(grpc_outlier_detection_lb_trace)) {
+    gpr_log(GPR_INFO, "[outlier_detection_lb %p] ejection timer running",
+            parent_.get());
+  }
+  std::map<SubchannelState*, double> success_rate_ejection_candidates;
+  std::map<SubchannelState*, double> failure_percentage_ejection_candidates;
+  size_t ejected_host_count = 0;
+  double success_rate_sum = 0;
+  auto time_now = Timestamp::Now();
+  auto& config = parent_->config_->outlier_detection_config();
+  for (auto& state : parent_->subchannel_state_map_) {
+    auto* subchannel_state = state.second.get();
+    // For each address, swap the call counter's buckets in that address's
+    // map entry.
+    subchannel_state->RotateBucket();
+    // Gather data to run success rate algorithm or failure percentage
+    // algorithm.
+    if (subchannel_state->ejection_time().has_value()) {
+      ++ejected_host_count;
     }
-    std::map<SubchannelState*, double> success_rate_ejection_candidates;
-    std::map<SubchannelState*, double> failure_percentage_ejection_candidates;
-    size_t ejected_host_count = 0;
-    double success_rate_sum = 0;
-    auto time_now = Timestamp::Now();
-    auto& config = parent_->config_->outlier_detection_config();
-    for (auto& state : parent_->subchannel_state_map_) {
-      auto* subchannel_state = state.second.get();
-      // For each address, swap the call counter's buckets in that address's
-      // map entry.
-      subchannel_state->RotateBucket();
-      // Gather data to run success rate algorithm or failure percentage
-      // algorithm.
-      if (subchannel_state->ejection_time().has_value()) {
-        ++ejected_host_count;
-      }
-      absl::optional<std::pair<double, uint64_t>> host_success_rate_and_volume =
-          subchannel_state->GetSuccessRateAndVolume();
-      if (!host_success_rate_and_volume.has_value()) {
-        continue;
-      }
-      double success_rate = host_success_rate_and_volume->first;
-      uint64_t request_volume = host_success_rate_and_volume->second;
-      if (config.success_rate_ejection.has_value()) {
-        if (request_volume >= config.success_rate_ejection->request_volume) {
-          success_rate_ejection_candidates[subchannel_state] = success_rate;
-          success_rate_sum += success_rate;
-        }
+    absl::optional<std::pair<double, uint64_t>> host_success_rate_and_volume =
+        subchannel_state->GetSuccessRateAndVolume();
+    if (!host_success_rate_and_volume.has_value()) {
+      continue;
+    }
+    double success_rate = host_success_rate_and_volume->first;
+    uint64_t request_volume = host_success_rate_and_volume->second;
+    if (config.success_rate_ejection.has_value()) {
+      if (request_volume >= config.success_rate_ejection->request_volume) {
+        success_rate_ejection_candidates[subchannel_state] = success_rate;
+        success_rate_sum += success_rate;
       }
-      if (config.failure_percentage_ejection.has_value()) {
-        if (request_volume >=
-            config.failure_percentage_ejection->request_volume) {
-          failure_percentage_ejection_candidates[subchannel_state] =
-              success_rate;
-        }
+    }
+    if (config.failure_percentage_ejection.has_value()) {
+      if (request_volume >=
+          config.failure_percentage_ejection->request_volume) {
+        failure_percentage_ejection_candidates[subchannel_state] = success_rate;
       }
     }
+  }
+  if (GRPC_TRACE_FLAG_ENABLED(grpc_outlier_detection_lb_trace)) {
+    gpr_log(GPR_INFO,
+            "[outlier_detection_lb %p] found %" PRIuPTR
+            " success rate candidates and %" PRIuPTR
+            " failure percentage candidates; ejected_host_count=%" PRIuPTR
+            "; success_rate_sum=%.3f",
+            parent_.get(), success_rate_ejection_candidates.size(),
+            failure_percentage_ejection_candidates.size(), ejected_host_count,
+            success_rate_sum);
+  }
+  // success rate algorithm
+  if (!success_rate_ejection_candidates.empty() &&
+      success_rate_ejection_candidates.size() >=
+          config.success_rate_ejection->minimum_hosts) {
+    if (GRPC_TRACE_FLAG_ENABLED(grpc_outlier_detection_lb_trace)) {
+      gpr_log(GPR_INFO,
+              "[outlier_detection_lb %p] running success rate algorithm",
+              parent_.get());
+    }
+    // calculate ejection threshold: (mean - stdev *
+    // (success_rate_ejection.stdev_factor / 1000))
+    double mean = success_rate_sum / success_rate_ejection_candidates.size();
+    double variance = 0;
+    for (const auto& p : success_rate_ejection_candidates) {
+      variance += std::pow(p.second - mean, 2);
+    }
+    variance /= success_rate_ejection_candidates.size();
+    double stdev = std::sqrt(variance);
+    const double success_rate_stdev_factor =
+        static_cast<double>(config.success_rate_ejection->stdev_factor) / 1000;
+    double ejection_threshold = mean - stdev * success_rate_stdev_factor;
     if (GRPC_TRACE_FLAG_ENABLED(grpc_outlier_detection_lb_trace)) {
       gpr_log(GPR_INFO,
-              "[outlier_detection_lb %p] found %" PRIuPTR
-              " success rate candidates and %" PRIuPTR
-              " failure percentage candidates; ejected_host_count=%" PRIuPTR
-              "; success_rate_sum=%.3f",
-              parent_.get(), success_rate_ejection_candidates.size(),
-              failure_percentage_ejection_candidates.size(), ejected_host_count,
-              success_rate_sum);
+              "[outlier_detection_lb %p] stdev=%.3f, ejection_threshold=%.3f",
+              parent_.get(), stdev, ejection_threshold);
     }
-    // success rate algorithm
-    if (!success_rate_ejection_candidates.empty() &&
-        success_rate_ejection_candidates.size() >=
-            config.success_rate_ejection->minimum_hosts) {
-      if (GRPC_TRACE_FLAG_ENABLED(grpc_outlier_detection_lb_trace)) {
-        gpr_log(GPR_INFO,
-                "[outlier_detection_lb %p] running success rate algorithm",
-                parent_.get());
-      }
-      // calculate ejection threshold: (mean - stdev *
-      // (success_rate_ejection.stdev_factor / 1000))
-      double mean = success_rate_sum / success_rate_ejection_candidates.size();
-      double variance = 0;
-      for (const auto& p : success_rate_ejection_candidates) {
-        variance += std::pow(p.second - mean, 2);
-      }
-      variance /= success_rate_ejection_candidates.size();
-      double stdev = std::sqrt(variance);
-      const double success_rate_stdev_factor =
-          static_cast<double>(config.success_rate_ejection->stdev_factor) /
-          1000;
-      double ejection_threshold = mean - stdev * success_rate_stdev_factor;
+    for (auto& candidate : success_rate_ejection_candidates) {
       if (GRPC_TRACE_FLAG_ENABLED(grpc_outlier_detection_lb_trace)) {
         gpr_log(GPR_INFO,
-                "[outlier_detection_lb %p] stdev=%.3f, ejection_threshold=%.3f",
-                parent_.get(), stdev, ejection_threshold);
+                "[outlier_detection_lb %p] checking candidate %p: "
+                "success_rate=%.3f",
+                parent_.get(), candidate.first, candidate.second);
       }
-      for (auto& candidate : success_rate_ejection_candidates) {
+      if (candidate.second < ejection_threshold) {
+        uint32_t random_key = absl::Uniform(bit_gen_, 1, 100);
+        double current_percent =
+            100.0 * ejected_host_count / parent_->subchannel_state_map_.size();
         if (GRPC_TRACE_FLAG_ENABLED(grpc_outlier_detection_lb_trace)) {
           gpr_log(GPR_INFO,
-                  "[outlier_detection_lb %p] checking candidate %p: "
-                  "success_rate=%.3f",
-                  parent_.get(), candidate.first, candidate.second);
+                  "[outlier_detection_lb %p] random_key=%d "
+                  "ejected_host_count=%" PRIuPTR " current_percent=%.3f",
+                  parent_.get(), random_key, ejected_host_count,
+                  current_percent);
         }
-        if (candidate.second < ejection_threshold) {
-          uint32_t random_key = absl::Uniform(bit_gen_, 1, 100);
-          double current_percent = 100.0 * ejected_host_count /
-                                   parent_->subchannel_state_map_.size();
+        if (random_key < config.success_rate_ejection->enforcement_percentage &&
+            (ejected_host_count == 0 ||
+             (current_percent < config.max_ejection_percent))) {
+          // Eject and record the timestamp for use when ejecting addresses in
+          // this iteration.
           if (GRPC_TRACE_FLAG_ENABLED(grpc_outlier_detection_lb_trace)) {
-            gpr_log(GPR_INFO,
-                    "[outlier_detection_lb %p] random_key=%d "
-                    "ejected_host_count=%" PRIuPTR " current_percent=%.3f",
-                    parent_.get(), random_key, ejected_host_count,
-                    current_percent);
-          }
-          if (random_key <
-                  config.success_rate_ejection->enforcement_percentage &&
-              (ejected_host_count == 0 ||
-               (current_percent < config.max_ejection_percent))) {
-            // Eject and record the timestamp for use when ejecting addresses in
-            // this iteration.
-            if (GRPC_TRACE_FLAG_ENABLED(grpc_outlier_detection_lb_trace)) {
-              gpr_log(GPR_INFO, "[outlier_detection_lb %p] ejecting candidate",
-                      parent_.get());
-            }
-            candidate.first->Eject(time_now);
-            ++ejected_host_count;
+            gpr_log(GPR_INFO, "[outlier_detection_lb %p] ejecting candidate",
+                    parent_.get());
           }
+          candidate.first->Eject(time_now);
+          ++ejected_host_count;
         }
       }
     }
-    // failure percentage algorithm
-    if (!failure_percentage_ejection_candidates.empty() &&
-        failure_percentage_ejection_candidates.size() >=
-            config.failure_percentage_ejection->minimum_hosts) {
+  }
+  // failure percentage algorithm
+  if (!failure_percentage_ejection_candidates.empty() &&
+      failure_percentage_ejection_candidates.size() >=
+          config.failure_percentage_ejection->minimum_hosts) {
+    if (GRPC_TRACE_FLAG_ENABLED(grpc_outlier_detection_lb_trace)) {
+      gpr_log(GPR_INFO,
+              "[outlier_detection_lb %p] running failure percentage algorithm",
+              parent_.get());
+    }
+    for (auto& candidate : failure_percentage_ejection_candidates) {
       if (GRPC_TRACE_FLAG_ENABLED(grpc_outlier_detection_lb_trace)) {
-        gpr_log(
-            GPR_INFO,
-            "[outlier_detection_lb %p] running failure percentage algorithm",
-            parent_.get());
+        gpr_log(GPR_INFO,
+                "[outlier_detection_lb %p] checking candidate %p: "
+                "success_rate=%.3f",
+                parent_.get(), candidate.first, candidate.second);
       }
-      for (auto& candidate : failure_percentage_ejection_candidates) {
+      // Extra check to make sure success rate algorithm didn't already
+      // eject this backend.
+      if (candidate.first->ejection_time().has_value()) continue;
+      if ((100.0 - candidate.second) >
+          config.failure_percentage_ejection->threshold) {
+        uint32_t random_key = absl::Uniform(bit_gen_, 1, 100);
+        double current_percent =
+            100.0 * ejected_host_count / parent_->subchannel_state_map_.size();
         if (GRPC_TRACE_FLAG_ENABLED(grpc_outlier_detection_lb_trace)) {
           gpr_log(GPR_INFO,
-                  "[outlier_detection_lb %p] checking candidate %p: "
-                  "success_rate=%.3f",
-                  parent_.get(), candidate.first, candidate.second);
+                  "[outlier_detection_lb %p] random_key=%d "
+                  "ejected_host_count=%" PRIuPTR " current_percent=%.3f",
+                  parent_.get(), random_key, ejected_host_count,
+                  current_percent);
         }
-        // Extra check to make sure success rate algorithm didn't already
-        // eject this backend.
-        if (candidate.first->ejection_time().has_value()) continue;
-        if ((100.0 - candidate.second) >
-            config.failure_percentage_ejection->threshold) {
-          uint32_t random_key = absl::Uniform(bit_gen_, 1, 100);
-          double current_percent = 100.0 * ejected_host_count /
-                                   parent_->subchannel_state_map_.size();
+        if (random_key <
+                config.failure_percentage_ejection->enforcement_percentage &&
+            (ejected_host_count == 0 ||
+             (current_percent < config.max_ejection_percent))) {
+          // Eject and record the timestamp for use when ejecting addresses in
+          // this iteration.
           if (GRPC_TRACE_FLAG_ENABLED(grpc_outlier_detection_lb_trace)) {
-            gpr_log(GPR_INFO,
-                    "[outlier_detection_lb %p] random_key=%d "
-                    "ejected_host_count=%" PRIuPTR " current_percent=%.3f",
-                    parent_.get(), random_key, ejected_host_count,
-                    current_percent);
-          }
-          if (random_key <
-                  config.failure_percentage_ejection->enforcement_percentage &&
-              (ejected_host_count == 0 ||
-               (current_percent < config.max_ejection_percent))) {
-            // Eject and record the timestamp for use when ejecting addresses in
-            // this iteration.
-            if (GRPC_TRACE_FLAG_ENABLED(grpc_outlier_detection_lb_trace)) {
-              gpr_log(GPR_INFO, "[outlier_detection_lb %p] ejecting candidate",
-                      parent_.get());
-            }
-            candidate.first->Eject(time_now);
-            ++ejected_host_count;
+            gpr_log(GPR_INFO, "[outlier_detection_lb %p] ejecting candidate",
+                    parent_.get());
           }
+          candidate.first->Eject(time_now);
+          ++ejected_host_count;
         }
       }
     }
-    // For each address in the map:
-    //   If the address is not ejected and the multiplier is greater than 0,
-    //   decrease the multiplier by 1. If the address is ejected, and the
-    //   current time is after ejection_timestamp + min(base_ejection_time *
-    //   multiplier, max(base_ejection_time, max_ejection_time)), un-eject the
-    //   address.
-    for (auto& state : parent_->subchannel_state_map_) {
-      auto* subchannel_state = state.second.get();
-      const bool unejected =
-          subchannel_state->MaybeUneject(config.base_ejection_time.millis(),
-                                         config.max_ejection_time.millis());
-      if (unejected &&
-          GRPC_TRACE_FLAG_ENABLED(grpc_outlier_detection_lb_trace)) {
-        gpr_log(GPR_INFO, "[outlier_detection_lb %p] unejected address %s (%p)",
-                parent_.get(), state.first.c_str(), subchannel_state);
-      }
+  }
+  // For each address in the map:
+  //   If the address is not ejected and the multiplier is greater than 0,
+  //   decrease the multiplier by 1. If the address is ejected, and the
+  //   current time is after ejection_timestamp + min(base_ejection_time *
+  //   multiplier, max(base_ejection_time, max_ejection_time)), un-eject the
+  //   address.
+  for (auto& state : parent_->subchannel_state_map_) {
+    auto* subchannel_state = state.second.get();
+    const bool unejected = subchannel_state->MaybeUneject(
+        config.base_ejection_time.millis(), config.max_ejection_time.millis());
+    if (unejected && GRPC_TRACE_FLAG_ENABLED(grpc_outlier_detection_lb_trace)) {
+      gpr_log(GPR_INFO, "[outlier_detection_lb %p] unejected address %s (%p)",
+              parent_.get(), state.first.c_str(), subchannel_state);
     }
-    timer_pending_ = false;
-    parent_->ejection_timer_ =
-        MakeOrphanable<EjectionTimer>(parent_, Timestamp::Now());
   }
-  Unref(DEBUG_LOCATION, "Timer");
-  GRPC_ERROR_UNREF(error);
+  parent_->ejection_timer_ =
+      MakeOrphanable<EjectionTimer>(parent_, Timestamp::Now());
 }
 
 //
@@ -1035,8 +1004,6 @@ class OutlierDetectionLbFactory : public LoadBalancingPolicyFactory {
     OutlierDetectionConfig outlier_detection_config;
     RefCountedPtr<LoadBalancingPolicy::Config> child_policy;
     {
-      ValidationErrors::ScopedField field(
-          &errors, "[\"outlier_detection_experimental\"]");
       outlier_detection_config =
           LoadFromJson<OutlierDetectionConfig>(json, JsonArgs(), &errors);
       // Parse childPolicy manually.
@@ -1085,6 +1052,14 @@ OutlierDetectionConfig::SuccessRateEjection::JsonLoader(const JsonArgs&) {
   return loader;
 }
 
+void OutlierDetectionConfig::SuccessRateEjection::JsonPostLoad(
+    const Json&, const JsonArgs&, ValidationErrors* errors) {
+  if (enforcement_percentage > 100) {
+    ValidationErrors::ScopedField field(errors, ".enforcement_percentage");
+    errors->AddError("value must be <= 100");
+  }
+}
+
 const JsonLoaderInterface*
 OutlierDetectionConfig::FailurePercentageEjection::JsonLoader(const JsonArgs&) {
   static const auto* loader =
@@ -1100,6 +1075,18 @@ OutlierDetectionConfig::FailurePercentageEjection::JsonLoader(const JsonArgs&) {
   return loader;
 }
 
+void OutlierDetectionConfig::FailurePercentageEjection::JsonPostLoad(
+    const Json&, const JsonArgs&, ValidationErrors* errors) {
+  if (enforcement_percentage > 100) {
+    ValidationErrors::ScopedField field(errors, ".enforcement_percentage");
+    errors->AddError("value must be <= 100");
+  }
+  if (threshold > 100) {
+    ValidationErrors::ScopedField field(errors, ".threshold");
+    errors->AddError("value must be <= 100");
+  }
+}
+
 const JsonLoaderInterface* OutlierDetectionConfig::JsonLoader(const JsonArgs&) {
   static const auto* loader =
       JsonObjectLoader<OutlierDetectionConfig>()
@@ -1119,11 +1106,15 @@ const JsonLoaderInterface* OutlierDetectionConfig::JsonLoader(const JsonArgs&) {
 }
 
 void OutlierDetectionConfig::JsonPostLoad(const Json& json, const JsonArgs&,
-                                          ValidationErrors* /*errors*/) {
+                                          ValidationErrors* errors) {
   if (json.object_value().find("maxEjectionTime") ==
       json.object_value().end()) {
     max_ejection_time = std::max(base_ejection_time, Duration::Seconds(300));
   }
+  if (max_ejection_percent > 100) {
+    ValidationErrors::ScopedField field(errors, ".max_ejection_percent");
+    errors->AddError("value must be <= 100");
+  }
 }
 
 //
@@ -1131,10 +1122,8 @@ void OutlierDetectionConfig::JsonPostLoad(const Json& json, const JsonArgs&,
 //
 
 void RegisterOutlierDetectionLbPolicy(CoreConfiguration::Builder* builder) {
-  if (XdsOutlierDetectionEnabled()) {
-    builder->lb_policy_registry()->RegisterLoadBalancingPolicyFactory(
-        absl::make_unique<OutlierDetectionLbFactory>());
-  }
+  builder->lb_policy_registry()->RegisterLoadBalancingPolicyFactory(
+      std::make_unique<OutlierDetectionLbFactory>());
 }
 
 }  // namespace grpc_core