WIP

Author: markdroth

src/core/xds/grpc/xds_server_grpc.cc

@@ -36,6 +36,9 @@ namespace {
 constexpr absl::string_view kServerFeatureIgnoreResourceDeletion =
     "ignore_resource_deletion";
 
+constexpr absl::string_view kServerFeatureResourceTimerIsTransientFailure =
+    "resource_timer_is_transient_error";
+
 constexpr absl::string_view kServerFeatureTrustedXdsServer =
     "trusted_xds_server";
 
@@ -46,6 +49,12 @@ bool GrpcXdsServer::IgnoreResourceDeletion() const {
              kServerFeatureIgnoreResourceDeletion)) != server_features_.end();
 }
 
+bool GrpcXdsServer::ResourceTimerIsTransientFailure() const {
+  return server_features_.find(std::string(
+             kServerFeatureResourceTimerIsTransientFailure)) !=
+             server_features_.end();
+}
+
 bool GrpcXdsServer::TrustedXdsServer() const {
   return server_features_.find(std::string(kServerFeatureTrustedXdsServer)) !=
          server_features_.end();
@@ -126,6 +135,8 @@ void GrpcXdsServer::JsonPostLoad(const Json& json, const JsonArgs& args,
         for (const Json& feature_json : array) {
           if (feature_json.type() == Json::Type::kString &&
               (feature_json.string() == kServerFeatureIgnoreResourceDeletion ||
+               feature_json.string() ==
+                   kServerFeatureResourceTimerIsTransientFailure ||
                feature_json.string() == kServerFeatureTrustedXdsServer)) {
             server_features_.insert(feature_json.string());
           }

src/core/xds/grpc/xds_server_grpc.h

@@ -35,6 +35,7 @@ class GrpcXdsServer final : public XdsBootstrap::XdsServer {
   const std::string& server_uri() const override { return server_uri_; }
 
   bool IgnoreResourceDeletion() const override;
+  bool ResourceTimerIsTransientFailure() const override;
 
   bool TrustedXdsServer() const;
 

src/core/xds/xds_client/xds_bootstrap.h

@@ -46,7 +46,9 @@ class XdsBootstrap {
     virtual ~XdsServer() = default;
 
     virtual const std::string& server_uri() const = 0;
+
     virtual bool IgnoreResourceDeletion() const = 0;
+    virtual bool ResourceTimerIsTransientFailure() const = 0;
 
     virtual bool Equals(const XdsServer& other) const = 0;
 

src/core/xds/xds_client/xds_client.cc

@@ -208,9 +208,15 @@ class XdsClient::XdsChannel::AdsCall final
       if (state.HasResource()) return;
       // Start timer.
       ads_call_ = std::move(ads_call);
+      Duration timeout = ads_call_->xds_client()->request_timeout_;
+      if (timeout == Duration::Zero()) {
+        timeout =
+            ads_call_->xds_channel()->server_.ResourceTimerIsTransientFailure()
+                ? Duration::Seconds(30)
+                : Duration::Seconds(15);
+      }
       timer_handle_ = ads_call_->xds_client()->engine()->RunAfter(
-          ads_call_->xds_client()->request_timeout_,
-          [self = Ref(DEBUG_LOCATION, "timer")]() {
+          timeout, [self = Ref(DEBUG_LOCATION, "timer")]() {
             ApplicationCallbackExecCtx callback_exec_ctx;
             ExecCtx exec_ctx;
             self->OnTimer();
@@ -237,9 +243,15 @@ class XdsClient::XdsChannel::AdsCall final
               << "} from xds server";
           resource_seen_ = true;
           state.SetDoesNotExist();
+          absl::Status status =
+              ads_call_->xds_channel()->server_
+                  .ResourceTimerIsTransientFailure()
+                  ? absl::UnavailableError(absl::StrCat(
+                        "xDS server ", ads_call_->xds_channel()->server_uri(),
+                        " not responding"))
+                  : absl::NotFoundError("does not exist");
           ads_call_->xds_client()->NotifyWatchersOnResourceChanged(
-              absl::NotFoundError("does not exist"), state.watchers(),
-              ReadDelayHandle::NoWait());
+              std::move(status), state.watchers(), ReadDelayHandle::NoWait());
         }
       }
       ads_call_->xds_client()->work_serializer_.DrainQueue();

src/core/xds/xds_client/xds_client.h

@@ -89,7 +89,9 @@ class XdsClient : public DualRefCounted<XdsClient> {
       std::shared_ptr<grpc_event_engine::experimental::EventEngine> engine,
       std::unique_ptr<XdsMetricsReporter> metrics_reporter,
       std::string user_agent_name, std::string user_agent_version,
-      Duration resource_request_timeout = Duration::Seconds(15));
+      // This parameter overrides the timer duration for testing
+      // purposes only -- do not use in production.
+      Duration resource_request_timeout = Duration::Zero());
   ~XdsClient() override;
 
   // Start and cancel watch for a resource.

test/core/xds/xds_client_test.cc

@@ -137,6 +137,9 @@ class XdsClientTest : public ::testing::Test {
       bool IgnoreResourceDeletion() const override {
         return ignore_resource_deletion_;
       }
+      bool ResourceTimerIsTransientFailure() const override {
+        return false;  // FIXME
+      }
       bool Equals(const XdsServer& other) const override {
         const auto& o = static_cast<const FakeXdsServer&>(other);
         return server_uri_ == o.server_uri_ &&