[Rust Frontend] Add per-request timing metrics to chat and completions (#59314)

Signed-off-by: wxin <wxin@nvidia.com>
Signed-off-by: Bugen Zhao <i@bugenzhao.com>
Co-authored-by: Bugen Zhao <i@bugenzhao.com>
Co-authored-by: Claude Opus 5.5 <noreply@anthropic.com>

rust/src/cmd/src/cli.rs

@@ -348,6 +348,11 @@ pub struct SharedRuntimeArgs {
     #[serde(default)]
     pub enable_prompt_tokens_details: bool,
 
+    /// Include per-request timing metrics in API responses.
+    #[arg(long, default_missing_value = "true", num_args = 0..=1)]
+    #[serde(default)]
+    pub enable_per_request_metrics: bool,
+
     /// If specified, API server will add X-Request-Id header to responses.
     #[arg(
         long,
@@ -635,6 +640,7 @@ impl SharedRuntimeArgs {
         ApiServerOptions {
             enable_log_requests: self.enable_log_requests,
             enable_prompt_tokens_details: self.enable_prompt_tokens_details,
+            enable_per_request_metrics: self.enable_per_request_metrics,
             enable_request_id_headers: self.enable_request_id_headers,
             enable_scale_out: self.enable_scale_out,
             sse_keep_alive_interval: (self.sse_keep_alive_interval > 0)

rust/src/cmd/src/cli/tests.rs

@@ -199,6 +199,7 @@ fn serve_args_forward_python_flags_with_separator() {
                         chat_template_content_format: Auto,
                         enable_log_requests: false,
                         enable_prompt_tokens_details: false,
+                        enable_per_request_metrics: false,
                         enable_request_id_headers: false,
                         enable_scale_out: false,
                         sse_keep_alive_interval: 0,
@@ -970,6 +971,40 @@ fn serve_args_reject_unsupported_flag_arg() {
     "#]].assert_eq(&error.to_string());
 }
 
+#[test]
+fn serve_args_enable_per_request_metrics_requires_engine_stats() {
+    let cli = Cli::try_parse_from([
+        "vllm-rs",
+        "serve",
+        "Qwen/Qwen3-0.6B",
+        "--enable-per-request-metrics",
+    ])
+    .expect("per-request metrics flag is supported");
+    let Command::Serve(args) = cli.command else {
+        panic!("expected serve args");
+    };
+    let config = args.to_frontend_config("tcp://127.0.0.1:0".into());
+    assert!(config.api_server_options.enable_per_request_metrics);
+    assert!(config.validate().is_ok());
+
+    let cli = Cli::try_parse_from([
+        "vllm-rs",
+        "serve",
+        "Qwen/Qwen3-0.6B",
+        "--enable-per-request-metrics",
+        "--disable-log-stats",
+    ])
+    .expect("flags parse independently");
+    let Command::Serve(args) = cli.command else {
+        panic!("expected serve args");
+    };
+    let error = args.to_frontend_config("tcp://127.0.0.1:0".into()).validate().unwrap_err();
+    assert!(
+        format!("{error:#}")
+            .contains("--enable-per-request-metrics requires engine statistics logging")
+    );
+}
+
 #[test]
 fn serve_args_reject_custom_generation_config_source() {
     let error = Cli::try_parse_from([
@@ -1067,6 +1102,7 @@ fn frontend_args_accept_json() {
                         chat_template_content_format: Auto,
                         enable_log_requests: false,
                         enable_prompt_tokens_details: false,
+                        enable_per_request_metrics: false,
                         enable_request_id_headers: false,
                         enable_scale_out: false,
                         sse_keep_alive_interval: 0,
@@ -1738,6 +1774,7 @@ fn serve_args_accept_handshake_aliases() {
                         chat_template_content_format: Auto,
                         enable_log_requests: false,
                         enable_prompt_tokens_details: false,
+                        enable_per_request_metrics: false,
                         enable_request_id_headers: false,
                         enable_scale_out: false,
                         sse_keep_alive_interval: 0,
@@ -1895,6 +1932,7 @@ fn serve_frontend_config_uses_dp_address_as_advertised_host() {
             api_server_options: ApiServerOptions {
                 enable_log_requests: false,
                 enable_prompt_tokens_details: false,
+                enable_per_request_metrics: false,
                 enable_request_id_headers: false,
                 enable_scale_out: false,
                 sse_keep_alive_interval: None,
@@ -1989,6 +2027,7 @@ fn serve_frontend_config_keeps_tcp_transport_for_non_local_only_topology() {
             api_server_options: ApiServerOptions {
                 enable_log_requests: false,
                 enable_prompt_tokens_details: false,
+                enable_per_request_metrics: false,
                 enable_request_id_headers: false,
                 enable_scale_out: false,
                 sse_keep_alive_interval: None,
@@ -2127,6 +2166,7 @@ fn frontend_config_uses_external_coordinator_when_coordinator_address_is_present
             api_server_options: ApiServerOptions {
                 enable_log_requests: false,
                 enable_prompt_tokens_details: false,
+                enable_per_request_metrics: false,
                 enable_request_id_headers: false,
                 enable_scale_out: false,
                 sse_keep_alive_interval: None,

rust/src/cmd/src/cli/unsupported.rs

@@ -476,15 +476,6 @@ pub struct ServerUnsupportedArgs {
     #[arg(long)]
     pub max_log_len: Option<Unsupported>,
 
-    /// If set to True, include per-request timing metrics in API responses.
-    #[arg(
-        long,
-        visible_alias = "no-enable-per-request-metrics",
-        default_missing_value = "true",
-        num_args = 0..=1
-    )]
-    pub enable_per_request_metrics: Option<Unsupported>,
-
     /// If set to True, enable tracking server_load_metrics in the app state.
     #[arg(
         long,