[Rust] Gate health on startup warmup completion (#37994)

This commit is contained in:
Po-Han Huang (NVIDIA)
2026-09-09 15:55:13 -07:00
committed by GitHub
parent beaf3d9252
commit 51c8581a26
6 changed files with 131 additions and 6 deletions
@@ -2205,6 +2205,10 @@ def _execute_server_warmup(server_args: ServerArgs):
url = server_args.url()
if get_serving().api_key:
headers["Authorization"] = f"Bearer {get_serving().api_key}"
if envs.SGLANG_RUST_SERVER.get():
# The Rust listener binds before this request so /model_info is
# available, but health stays 503 until this marked request succeeds.
headers["x-sglang-startup-warmup"] = "1"
ssl_verify = ssl_verify_of(server_args)
+1
View File
@@ -62,6 +62,7 @@ def _build_server_args(scheduler: Scheduler) -> ServerArgs:
tokenizer_worker_num=get_serving().tokenizer_worker_num,
detokenizer_worker_num=get_serving().detokenizer_worker_num,
skip_tokenizer_init=get_serving().skip_tokenizer_init,
skip_server_warmup=get_serving().skip_server_warmup,
incremental_streaming_output=get_serving().incremental_streaming_output,
disaggregation_mode=disaggregation_mode,
model_config=ext.ModelConfig(