Cargo.toml workspace

Cargo.toml

@@ -0,0 +1,3 @@
+[workspace]
+members = ["source/extensions/dynamic_modules/sdk/rust", "test/extensions/dynamic_modules/test_data/rust"]
+resolver = "2"

api/envoy/service/health/v3/hds.proto

@@ -10,6 +10,7 @@ import "envoy/config/endpoint/v3/endpoint_components.proto";
 
 import "google/api/annotations.proto";
 import "google/protobuf/duration.proto";
+import "google/protobuf/struct.proto";
 
 import "envoy/annotations/deprecation.proto";
 import "udpa/annotations/status.proto";
@@ -109,6 +110,19 @@ message EndpointHealth {
   config.endpoint.v3.Endpoint endpoint = 1;
 
   config.core.v3.HealthStatus health_status = 2;
+
+  // Optional metadata about the health check result, populated by the active
+  // health checker and forwarded to the management server for richer health
+  // state interpretation.
+  //
+  // Well-known keys:
+  //
+  // ``http_status_code`` (number)
+  //   Set by the HTTP health checker. Contains the HTTP response status code
+  //   returned by the upstream endpoint during the most recent health check,
+  //   e.g. ``200``, ``503``. Only present when the health check received a
+  //   complete HTTP response; absent on connection failures or timeouts.
+  google.protobuf.Struct health_metadata = 3;
 }
 
 // Group endpoint health by locality under each cluster.

envoy/upstream/upstream.h

@@ -323,6 +323,19 @@ class Host : virtual public HostDescription {
    * Set true to disable active health check for the host.
    */
   virtual void setDisableActiveHealthCheck(bool disable_active_health_check) PURE;
+
+  /**
+   * Store the HTTP status code from the last active health check response.
+   * Used by HDS to report richer health state to the control plane.
+   * 0 means no response has been recorded yet.
+   */
+  virtual void setLastHealthCheckHttpStatus(uint64_t) PURE;
+
+  /**
+   * @return the HTTP status code from the last active health check response, or
+   * 0 if no response has been recorded.
+   */
+  virtual absl::optional<uint64_t> lastHealthCheckHttpStatus() const PURE;
 };
 
 using HostConstSharedPtr = std::shared_ptr<const Host>;

source/common/upstream/health_discovery_service.cc

@@ -124,6 +124,14 @@ envoy::service::health::v3::HealthCheckRequestOrEndpointHealthResponse HdsDelega
             }
           }
 
+          // If a HTTP health check has run, attach the last response code to the
+          // HDS report so the control plane can interpret richer health states.
+          auto http_status = host->lastHealthCheckHttpStatus();
+          if (http_status.has_value()) {
+            (*endpoint->mutable_health_metadata()->mutable_fields())["http_status_code"]
+                .set_number_value(http_status.value());
+          }
+
           // TODO(drewsortega): remove this once we are on v4 and endpoint_health_response is
           // removed. Copy this endpoint's health info to the legacy flat-list.
           response.mutable_endpoint_health_response()->add_endpoints_health()->MergeFrom(*endpoint);

source/common/upstream/upstream_impl.h

@@ -383,6 +383,11 @@ class HostImplBase : public Host,
   uint32_t healthFlagsGetAll() const override { return health_flags_; }
   void healthFlagsSetAll(uint32_t bits) override { health_flags_ |= bits; }
 
+  void setLastHealthCheckHttpStatus(uint64_t status) override { last_hc_http_status_ = status; }
+  absl::optional<uint64_t> lastHealthCheckHttpStatus() const override {
+    return last_hc_http_status_;
+  }
+
   Host::HealthStatus healthStatus() const override {
     // Evaluate active health status first.
 
@@ -471,6 +476,7 @@ class HostImplBase : public Host,
   // flag access? May be we could refactor HealthFlag to contain all these statuses and flags in the
   // future.
   std::atomic<Host::HealthStatus> eds_health_status_{};
+  absl::optional<std::atomic<uint64_t>> last_hc_http_status_ = absl::nullopt;
 
   struct HostHandleImpl : HostHandle {
     HostHandleImpl(const std::shared_ptr<const HostImplBase>& parent) : parent_(parent) {