[Rust Frontend][gRPC] Add server and model discovery (#49491)

Signed-off-by: Connor Carpenter <connorc@nvidia.com>
Co-authored-by: Nick Hill <nickhill123@gmail.com>

rust/src/engine-core-client/src/client.rs

@@ -394,6 +394,17 @@ impl EngineCoreClient {
         self.engines.iter().map(|engine| &engine.ready_response).collect()
     }
 
+    /// Return the first engine's ready response.
+    ///
+    /// Per-engine fields such as `data_parallel_rank` should be read through
+    /// [`ready_responses`](Self::ready_responses).
+    pub fn ready_response(&self) -> &EngineCoreReadyResponse {
+        &self
+            .engines
+            .first()
+            .expect("engine core client requires at least one engine")
+            .ready_response
+    }
     /// Return the engine-reported effective model dtype.
     pub fn model_dtype(&self) -> ModelDtype {
         self.engines

rust/src/engine-core-client/src/mock_engine.rs

@@ -57,6 +57,13 @@ pub fn default_ready_response() -> EngineCoreReadyResponse {
         vllm_version: "test-vllm-version".to_string(),
         world_size: 1,
         data_parallel_size: 1,
+        tensor_parallel_size: 1,
+        pipeline_parallel_size: 1,
+        decode_context_parallel_size: 1,
+        data_parallel_rank: 0,
+        max_num_seqs: 256,
+        max_num_batched_tokens: 8192,
+        instance_id: "test-instance".to_string(),
         kv_cache_size_tokens: None,
         kv_cache_max_concurrency: None,
     }

rust/src/engine-core-client/src/protocol/handshake.rs

@@ -52,6 +52,21 @@ pub struct EngineCoreReadyResponse {
     pub world_size: u64,
     /// Data parallelism size from the parallel config.
     pub data_parallel_size: u64,
+    // Required discovery metadata; EngineCore and client versions must match.
+    /// Tensor-parallel size of this engine.
+    pub tensor_parallel_size: u32,
+    /// Pipeline-parallel size of this engine.
+    pub pipeline_parallel_size: u32,
+    /// Decode-context-parallel size of this engine.
+    pub decode_context_parallel_size: u32,
+    /// This engine's data-parallel rank.
+    pub data_parallel_rank: u32,
+    /// Scheduler cap on concurrently running sequences.
+    pub max_num_seqs: u64,
+    /// Scheduler cap on batched tokens per step.
+    pub max_num_batched_tokens: u64,
+    /// Unique identifier for this server instance.
+    pub instance_id: String,
     /// Total KV cache capacity in tokens, if reported.
     pub kv_cache_size_tokens: Option<u64>,
     /// Maximum achievable request concurrency given the KV cache, if reported.

rust/src/engine-core-client/src/tests/python_compat.py

@@ -363,6 +363,13 @@ class EngineCoreReadyResponse:
     vllm_version: str
     world_size: int
     data_parallel_size: int
+    tensor_parallel_size: int
+    pipeline_parallel_size: int
+    decode_context_parallel_size: int
+    data_parallel_rank: int
+    max_num_seqs: int
+    max_num_batched_tokens: int
+    instance_id: str
     kv_cache_size_tokens: int | None = None
     kv_cache_max_concurrency: float | None = None
 
@@ -376,6 +383,13 @@ ready_response = EngineCoreReadyResponse(
     vllm_version="0.0.0",
     data_parallel_size=1,
     world_size=1,
+    tensor_parallel_size=1,
+    pipeline_parallel_size=1,
+    decode_context_parallel_size=1,
+    data_parallel_rank=0,
+    max_num_seqs=256,
+    max_num_batched_tokens=8192,
+    instance_id="test-instance",
 )
 
 print(msgspec.msgpack.encode(request).hex())