grpc 1.41.1 → 1.44.0.pre2

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

@@ -33,12 +33,10 @@
 #include "src/core/lib/channel/channel_args.h"
 #include "src/core/lib/debug/trace.h"
 #include "src/core/lib/gpr/string.h"
-#include "src/core/lib/gpr/useful.h"
 #include "src/core/lib/gprpp/ref_counted_ptr.h"
 #include "src/core/lib/gprpp/sync.h"
 #include "src/core/lib/transport/connectivity_state.h"
 #include "src/core/lib/transport/error_utils.h"
-#include "src/core/lib/transport/static_metadata.h"
 
 namespace grpc_core {
 
@@ -107,6 +105,7 @@ class RingHashLbConfig : public LoadBalancingPolicy::Config {
 //
 // ring_hash LB policy
 //
+
 class RingHash : public LoadBalancingPolicy {
  public:
   explicit RingHash(Args args);
@@ -119,8 +118,9 @@ class RingHash : public LoadBalancingPolicy {
  private:
   ~RingHash() override;
 
-  // Forward declaration.
+  // Forward declarations.
   class RingHashSubchannelList;
+  class Ring;
 
   // Data for a particular subchannel in a subchannel list.
   // This subclass adds the following functionality:
@@ -137,9 +137,10 @@ class RingHash : public LoadBalancingPolicy {
         : SubchannelData(subchannel_list, address, std::move(subchannel)),
           address_(address) {}
 
-    grpc_connectivity_state connectivity_state() const {
-      return last_connectivity_state_;
+    grpc_connectivity_state GetConnectivityState() const {
+      return connectivity_state_for_picker_.load(std::memory_order_relaxed);
     }
+
     const ServerAddress& address() const { return address_; }
 
     bool seen_failure_since_ready() const { return seen_failure_since_ready_; }
@@ -157,6 +158,8 @@ class RingHash : public LoadBalancingPolicy {
 
     ServerAddress address_;
     grpc_connectivity_state last_connectivity_state_ = GRPC_CHANNEL_SHUTDOWN;
+    std::atomic<grpc_connectivity_state> connectivity_state_for_picker_{
+        GRPC_CHANNEL_IDLE};
     bool seen_failure_since_ready_ = false;
   };
 
@@ -194,6 +197,9 @@ class RingHash : public LoadBalancingPolicy {
     // Transient Failure.
     bool UpdateRingHashConnectivityStateLocked();
 
+    // Create a new ring from this subchannel list.
+    RefCountedPtr<Ring> MakeRing();
+
    private:
     size_t num_idle_ = 0;
     size_t num_ready_ = 0;
@@ -201,20 +207,31 @@ class RingHash : public LoadBalancingPolicy {
     size_t num_transient_failure_ = 0;
   };
 
+  class Ring : public RefCounted<Ring> {
+   public:
+    struct Entry {
+      uint64_t hash;
+      RingHashSubchannelData* subchannel;
+    };
+
+    Ring(RingHash* parent,
+         RefCountedPtr<RingHashSubchannelList> subchannel_list);
+
+    const std::vector<Entry>& ring() const { return ring_; }
+
+   private:
+    RefCountedPtr<RingHashSubchannelList> subchannel_list_;
+    std::vector<Entry> ring_;
+  };
+
   class Picker : public SubchannelPicker {
    public:
-    Picker(RefCountedPtr<RingHash> parent,
-           RingHashSubchannelList* subchannel_list);
+    Picker(RefCountedPtr<RingHash> parent, RefCountedPtr<Ring> ring)
+        : parent_(std::move(parent)), ring_(std::move(ring)) {}
 
     PickResult Pick(PickArgs args) override;
 
    private:
-    struct RingEntry {
-      uint64_t hash;
-      RefCountedPtr<SubchannelInterface> subchannel;
-      grpc_connectivity_state connectivity_state;
-    };
-
     // A fire-and-forget class that schedules subchannel connection attempts
     // on the control plane WorkSerializer.
     class SubchannelConnectionAttempter : public Orphanable {
@@ -236,7 +253,7 @@ class RingHash : public LoadBalancingPolicy {
       }
 
      private:
-      static void RunInExecCtx(void* arg, grpc_error* /*error*/) {
+      static void RunInExecCtx(void* arg, grpc_error_handle /*error*/) {
         auto* self = static_cast<SubchannelConnectionAttempter*>(arg);
         self->ring_hash_lb_->work_serializer()->Run(
             [self]() {
@@ -256,9 +273,7 @@ class RingHash : public LoadBalancingPolicy {
     };
 
     RefCountedPtr<RingHash> parent_;
-
-    // A ring of subchannels.
-    std::vector<RingEntry> ring_;
+    RefCountedPtr<Ring> ring_;
   };
 
   void ShutdownLocked() override;
@@ -270,16 +285,19 @@ class RingHash : public LoadBalancingPolicy {
   OrphanablePtr<RingHashSubchannelList> subchannel_list_;
   // indicating if we are shutting down.
   bool shutdown_ = false;
+
+  // Current ring.
+  RefCountedPtr<Ring> ring_;
 };
 
 //
-// RingHash::Picker
+// RingHash::Ring
 //
 
-RingHash::Picker::Picker(RefCountedPtr<RingHash> parent,
-                         RingHashSubchannelList* subchannel_list)
-    : parent_(std::move(parent)) {
-  size_t num_subchannels = subchannel_list->num_subchannels();
+RingHash::Ring::Ring(RingHash* parent,
+                     RefCountedPtr<RingHashSubchannelList> subchannel_list)
+    : subchannel_list_(std::move(subchannel_list)) {
+  size_t num_subchannels = subchannel_list_->num_subchannels();
   // Store the weights while finding the sum.
   struct AddressWeight {
     std::string address;
@@ -292,7 +310,7 @@ RingHash::Picker::Picker(RefCountedPtr<RingHash> parent,
   size_t sum = 0;
   address_weights.reserve(num_subchannels);
   for (size_t i = 0; i < num_subchannels; ++i) {
-    RingHashSubchannelData* sd = subchannel_list->subchannel(i);
+    RingHashSubchannelData* sd = subchannel_list_->subchannel(i);
     const ServerAddressWeightAttribute* weight_attribute = static_cast<
         const ServerAddressWeightAttribute*>(sd->address().GetAttribute(
         ServerAddressWeightAttribute::kServerAddressWeightAttributeKey));
@@ -323,8 +341,8 @@ RingHash::Picker::Picker(RefCountedPtr<RingHash> parent,
   // weights aren't provided, all hosts should get an equal number of hashes. In
   // the case where this number exceeds the max_ring_size, it's scaled back down
   // to fit.
-  const size_t min_ring_size = parent_->config_->min_ring_size();
-  const size_t max_ring_size = parent_->config_->max_ring_size();
+  const size_t min_ring_size = parent->config_->min_ring_size();
+  const size_t max_ring_size = parent->config_->max_ring_size();
   const double scale = std::min(
       std::ceil(min_normalized_weight * min_ring_size) / min_normalized_weight,
       static_cast<double>(max_ring_size));
@@ -348,17 +366,13 @@ RingHash::Picker::Picker(RefCountedPtr<RingHash> parent,
     auto offset_start = hash_key_buffer.end();
     target_hashes += scale * address_weights[i].normalized_weight;
     size_t count = 0;
-    auto current_state =
-        subchannel_list->subchannel(i)->subchannel()->CheckConnectivityState();
     while (current_hashes < target_hashes) {
       const std::string count_str = absl::StrCat(count);
       hash_key_buffer.insert(offset_start, count_str.begin(), count_str.end());
       absl::string_view hash_key(hash_key_buffer.data(),
                                  hash_key_buffer.size());
       const uint64_t hash = XXH64(hash_key.data(), hash_key.size(), 0);
-      ring_.push_back({hash,
-                       subchannel_list->subchannel(i)->subchannel()->Ref(),
-                       current_state});
+      ring_.push_back({hash, subchannel_list_->subchannel(i)});
       ++count;
       ++current_hashes;
       hash_key_buffer.erase(offset_start, hash_key_buffer.end());
@@ -369,17 +383,21 @@ RingHash::Picker::Picker(RefCountedPtr<RingHash> parent,
         std::max(static_cast<uint64_t>(i), max_hashes_per_host);
   }
   std::sort(ring_.begin(), ring_.end(),
-            [](const RingEntry& lhs, const RingEntry& rhs) -> bool {
+            [](const Entry& lhs, const Entry& rhs) -> bool {
               return lhs.hash < rhs.hash;
             });
   if (GRPC_TRACE_FLAG_ENABLED(grpc_lb_ring_hash_trace)) {
     gpr_log(GPR_INFO,
-            "[RH %p picker %p] created picker from subchannel_list=%p "
+            "[RH %p picker %p] created ring from subchannel_list=%p "
             "with %" PRIuPTR " ring entries",
-            parent_.get(), this, subchannel_list, ring_.size());
+            parent, this, subchannel_list_.get(), ring_.size());
   }
 }
 
+//
+// RingHash::Picker
+//
+
 RingHash::PickResult RingHash::Picker::Pick(PickArgs args) {
   auto hash =
       args.call_state->ExperimentalGetCallAttribute(kRequestRingHashAttribute);
@@ -388,20 +406,21 @@ RingHash::PickResult RingHash::Picker::Pick(PickArgs args) {
     return PickResult::Fail(
         absl::InternalError("xds ring hash value is not a number"));
   }
+  const std::vector<Ring::Entry>& ring = ring_->ring();
   // Ported from https://github.com/RJ/ketama/blob/master/libketama/ketama.c
   // (ketama_get_server) NOTE: The algorithm depends on using signed integers
   // for lowp, highp, and first_index. Do not change them!
   int64_t lowp = 0;
-  int64_t highp = ring_.size();
+  int64_t highp = ring.size();
   int64_t first_index = 0;
   while (true) {
     first_index = (lowp + highp) / 2;
-    if (first_index == static_cast<int64_t>(ring_.size())) {
+    if (first_index == static_cast<int64_t>(ring.size())) {
       first_index = 0;
       break;
     }
-    uint64_t midval = ring_[first_index].hash;
-    uint64_t midval1 = first_index == 0 ? 0 : ring_[first_index - 1].hash;
+    uint64_t midval = ring[first_index].hash;
+    uint64_t midval1 = first_index == 0 ? 0 : ring[first_index - 1].hash;
     if (h <= midval && h > midval1) {
       break;
     }
@@ -424,35 +443,41 @@ RingHash::PickResult RingHash::Picker::Pick(PickArgs args) {
         }
         subchannel_connection_attempter->AddSubchannel(std::move(subchannel));
       };
-  switch (ring_[first_index].connectivity_state) {
+  switch (ring[first_index].subchannel->GetConnectivityState()) {
     case GRPC_CHANNEL_READY:
-      return PickResult::Complete(ring_[first_index].subchannel);
+      return PickResult::Complete(
+          ring[first_index].subchannel->subchannel()->Ref());
     case GRPC_CHANNEL_IDLE:
-      ScheduleSubchannelConnectionAttempt(ring_[first_index].subchannel);
+      ScheduleSubchannelConnectionAttempt(
+          ring[first_index].subchannel->subchannel()->Ref());
       ABSL_FALLTHROUGH_INTENDED;
     case GRPC_CHANNEL_CONNECTING:
       return PickResult::Queue();
     default:  // GRPC_CHANNEL_TRANSIENT_FAILURE
       break;
   }
-  ScheduleSubchannelConnectionAttempt(ring_[first_index].subchannel);
+  ScheduleSubchannelConnectionAttempt(
+      ring[first_index].subchannel->subchannel()->Ref());
   // Loop through remaining subchannels to find one in READY.
   // On the way, we make sure the right set of connection attempts
   // will happen.
   bool found_second_subchannel = false;
   bool found_first_non_failed = false;
-  for (size_t i = 1; i < ring_.size(); ++i) {
-    const RingEntry& entry = ring_[(first_index + i) % ring_.size()];
-    if (entry.subchannel == ring_[first_index].subchannel) {
+  for (size_t i = 1; i < ring.size(); ++i) {
+    const Ring::Entry& entry = ring[(first_index + i) % ring.size()];
+    if (entry.subchannel == ring[first_index].subchannel) {
       continue;
     }
-    if (entry.connectivity_state == GRPC_CHANNEL_READY) {
-      return PickResult::Complete(entry.subchannel);
+    grpc_connectivity_state connectivity_state =
+        entry.subchannel->GetConnectivityState();
+    if (connectivity_state == GRPC_CHANNEL_READY) {
+      return PickResult::Complete(entry.subchannel->subchannel()->Ref());
     }
     if (!found_second_subchannel) {
-      switch (entry.connectivity_state) {
+      switch (connectivity_state) {
         case GRPC_CHANNEL_IDLE:
-          ScheduleSubchannelConnectionAttempt(entry.subchannel);
+          ScheduleSubchannelConnectionAttempt(
+              entry.subchannel->subchannel()->Ref());
           ABSL_FALLTHROUGH_INTENDED;
         case GRPC_CHANNEL_CONNECTING:
           return PickResult::Queue();
@@ -462,11 +487,13 @@ RingHash::PickResult RingHash::Picker::Pick(PickArgs args) {
       found_second_subchannel = true;
     }
     if (!found_first_non_failed) {
-      if (entry.connectivity_state == GRPC_CHANNEL_TRANSIENT_FAILURE) {
-        ScheduleSubchannelConnectionAttempt(entry.subchannel);
+      if (connectivity_state == GRPC_CHANNEL_TRANSIENT_FAILURE) {
+        ScheduleSubchannelConnectionAttempt(
+            entry.subchannel->subchannel()->Ref());
       } else {
-        if (entry.connectivity_state == GRPC_CHANNEL_IDLE) {
-          ScheduleSubchannelConnectionAttempt(entry.subchannel);
+        if (connectivity_state == GRPC_CHANNEL_IDLE) {
+          ScheduleSubchannelConnectionAttempt(
+              entry.subchannel->subchannel()->Ref());
         }
         found_first_non_failed = true;
       }
@@ -499,7 +526,7 @@ void RingHash::RingHashSubchannelList::StartWatchingLocked() {
   p->channel_control_helper()->UpdateState(
       GRPC_CHANNEL_READY, absl::Status(),
       absl::make_unique<Picker>(p->Ref(DEBUG_LOCATION, "RingHashPicker"),
-                                this));
+                                p->ring_));
 }
 
 void RingHash::RingHashSubchannelList::UpdateStateCountersLocked(
@@ -548,7 +575,7 @@ bool RingHash::RingHashSubchannelList::UpdateRingHashConnectivityStateLocked() {
     p->channel_control_helper()->UpdateState(
         GRPC_CHANNEL_READY, absl::Status(),
         absl::make_unique<Picker>(p->Ref(DEBUG_LOCATION, "RingHashPicker"),
-                                  this));
+                                  p->ring_));
     return false;
   }
   if (num_connecting_ > 0 && num_transient_failure_ < 2) {
@@ -561,7 +588,7 @@ bool RingHash::RingHashSubchannelList::UpdateRingHashConnectivityStateLocked() {
     p->channel_control_helper()->UpdateState(
         GRPC_CHANNEL_IDLE, absl::Status(),
         absl::make_unique<Picker>(p->Ref(DEBUG_LOCATION, "RingHashPicker"),
-                                  this));
+                                  p->ring_));
     return false;
   }
   absl::Status status =
@@ -572,6 +599,11 @@ bool RingHash::RingHashSubchannelList::UpdateRingHashConnectivityStateLocked() {
   return true;
 }
 
+RefCountedPtr<RingHash::Ring> RingHash::RingHashSubchannelList::MakeRing() {
+  RingHash* p = static_cast<RingHash*>(policy());
+  return MakeRefCounted<Ring>(p, Ref(DEBUG_LOCATION, "Ring"));
+}
+
 //
 // RingHash::RingHashSubchannelData
 //
@@ -582,7 +614,7 @@ void RingHash::RingHashSubchannelData::UpdateConnectivityStateLocked(
   if (GRPC_TRACE_FLAG_ENABLED(grpc_lb_ring_hash_trace)) {
     gpr_log(
         GPR_INFO,
-        "[RR %p] connectivity changed for subchannel %p, subchannel_list %p "
+        "[RH %p] connectivity changed for subchannel %p, subchannel_list %p "
         "(index %" PRIuPTR " of %" PRIuPTR "): prev_state=%s new_state=%s",
         p, subchannel(), subchannel_list(), Index(),
         subchannel_list()->num_subchannels(),
@@ -615,6 +647,9 @@ void RingHash::RingHashSubchannelData::ProcessConnectivityChangeLocked(
     grpc_connectivity_state connectivity_state) {
   RingHash* p = static_cast<RingHash*>(subchannel_list()->policy());
   GPR_ASSERT(subchannel() != nullptr);
+  // Update connectivity state used by picker.
+  connectivity_state_for_picker_.store(connectivity_state,
+                                       std::memory_order_relaxed);
   // If the new state is TRANSIENT_FAILURE, re-resolve.
   // Only do this if we've started watching, not at startup time.
   // Otherwise, if the subchannel was already in state TRANSIENT_FAILURE
@@ -624,7 +659,7 @@ void RingHash::RingHashSubchannelData::ProcessConnectivityChangeLocked(
   if (connectivity_state == GRPC_CHANNEL_TRANSIENT_FAILURE) {
     if (GRPC_TRACE_FLAG_ENABLED(grpc_lb_ring_hash_trace)) {
       gpr_log(GPR_INFO,
-              "[RR %p] Subchannel %p has gone into TRANSIENT_FAILURE. "
+              "[RH %p] Subchannel %p has gone into TRANSIENT_FAILURE. "
               "Requesting re-resolution",
               p, subchannel());
     }
@@ -681,36 +716,52 @@ void RingHash::ShutdownLocked() {
   }
   shutdown_ = true;
   subchannel_list_.reset();
+  ring_.reset(DEBUG_LOCATION, "RingHash");
 }
 
 void RingHash::ResetBackoffLocked() { subchannel_list_->ResetBackoffLocked(); }
 
 void RingHash::UpdateLocked(UpdateArgs args) {
-  if (GRPC_TRACE_FLAG_ENABLED(grpc_lb_ring_hash_trace)) {
-    gpr_log(GPR_INFO, "[RR %p] received update with %" PRIuPTR " addresses",
-            this, args.addresses.size());
-  }
   config_ = std::move(args.config);
-  // Filter out any address with weight 0.
   ServerAddressList addresses;
-  addresses.reserve(args.addresses.size());
-  for (ServerAddress& address : args.addresses) {
-    const ServerAddressWeightAttribute* weight_attribute =
-        static_cast<const ServerAddressWeightAttribute*>(address.GetAttribute(
-            ServerAddressWeightAttribute::kServerAddressWeightAttributeKey));
-    if (weight_attribute == nullptr || weight_attribute->weight() > 0) {
-      addresses.push_back(std::move(address));
+  if (args.addresses.ok()) {
+    if (GRPC_TRACE_FLAG_ENABLED(grpc_lb_ring_hash_trace)) {
+      gpr_log(GPR_INFO, "[RH %p] received update with %" PRIuPTR " addresses",
+              this, args.addresses->size());
+    }
+    // Filter out any address with weight 0.
+    addresses.reserve(args.addresses->size());
+    for (ServerAddress& address : *args.addresses) {
+      const ServerAddressWeightAttribute* weight_attribute =
+          static_cast<const ServerAddressWeightAttribute*>(address.GetAttribute(
+              ServerAddressWeightAttribute::kServerAddressWeightAttributeKey));
+      if (weight_attribute == nullptr || weight_attribute->weight() > 0) {
+        addresses.emplace_back(std::move(address));
+      }
+    }
+  } else {
+    if (GRPC_TRACE_FLAG_ENABLED(grpc_lb_ring_hash_trace)) {
+      gpr_log(GPR_INFO, "[RH %p] received update with addresses error: %s",
+              this, args.addresses.status().ToString().c_str());
     }
+    // If we already have a subchannel list, then ignore the resolver
+    // failure and keep using the existing list.
+    if (subchannel_list_ != nullptr) return;
   }
   subchannel_list_ = MakeOrphanable<RingHashSubchannelList>(
       this, &grpc_lb_ring_hash_trace, std::move(addresses), *args.args);
   if (subchannel_list_->num_subchannels() == 0) {
     // If the new list is empty, immediately transition to TRANSIENT_FAILURE.
-    absl::Status status = absl::UnavailableError("Empty update");
+    absl::Status status =
+        args.addresses.ok() ? absl::UnavailableError(absl::StrCat(
+                                  "empty address list: ", args.resolution_note))
+                            : args.addresses.status();
     channel_control_helper()->UpdateState(
         GRPC_CHANNEL_TRANSIENT_FAILURE, status,
         absl::make_unique<TransientFailurePicker>(status));
   } else {
+    // Build the ring.
+    ring_ = subchannel_list_->MakeRing();
     // Start watching the new list.
     subchannel_list_->StartWatchingLocked();
   }
@@ -730,7 +781,7 @@ class RingHashFactory : public LoadBalancingPolicyFactory {
   const char* name() const override { return kRingHash; }
 
   RefCountedPtr<LoadBalancingPolicy::Config> ParseLoadBalancingConfig(
-      const Json& json, grpc_error** error) const override {
+      const Json& json, grpc_error_handle* error) const override {
     size_t min_ring_size;
     size_t max_ring_size;
     std::vector<grpc_error_handle> error_list;
@@ -748,9 +799,8 @@ class RingHashFactory : public LoadBalancingPolicyFactory {
 }  // namespace
 
 void GrpcLbPolicyRingHashInit() {
-  grpc_core::LoadBalancingPolicyRegistry::Builder::
-      RegisterLoadBalancingPolicyFactory(
-          absl::make_unique<grpc_core::RingHashFactory>());
+  LoadBalancingPolicyRegistry::Builder::RegisterLoadBalancingPolicyFactory(
+      absl::make_unique<RingHashFactory>());
 }
 
 void GrpcLbPolicyRingHashShutdown() {}