minke.yu/sglang
1
0
Fork 0
Code Issues Pull Requests Actions 13 Packages Projects Releases Wiki Activity

Convert IPC dataclasses to msgspec.Struct with opt-in msgpack transport (#28688)

Browse Source 
Co-authored-by: Lianmin Zheng <lianminzheng@gmail.com>
This commit is contained in:
Rain Jiang
2026-06-26 12:04:03 -07:00
committed by GitHub
co-authored by Lianmin Zheng
parent 714011a40f
commit be1930133a
25 changed files with 694 additions and 354 deletions
Download Patch File Download Diff File Expand all files Collapse all files
python/sglang/srt/entrypoints/http_server.py
+16 -13
View File
@@ -173,6 +173,7 @@ from sglang.srt.utils.json_response import (
dumps_json,
orjson_response,
)
from sglang.srt.utils.msgspec_utils import msgspec_to_builtins
from sglang.srt.utils.watchdog import SubprocessWatchdog
from sglang.utils import get_exception_traceback
from sglang.version import __version__
@@ -698,16 +699,18 @@ async def server_info():
server_args = _global_state.tokenizer_manager.server_args
# server_args.model_config is not serializable but should be excluded by asdict.
return {
**dataclasses.asdict(server_args),
**_global_state.scheduler_info,
"internal_states": internal_states,
"version": __version__,
# Structured KV-event publisher descriptor for KV-aware routers.
# `None` when publishing is disabled or misconfigured; see
# `ServerArgs.describe_kv_events_publisher` for the precise contract.
"kv_events": server_args.describe_kv_events_publisher(),
}
return msgspec_to_builtins(
{
**dataclasses.asdict(server_args),
**_global_state.scheduler_info,
"internal_states": internal_states,
"version": __version__,
# Structured KV-event publisher descriptor for KV-aware routers.
# `None` when publishing is disabled or misconfigured; see
# `ServerArgs.describe_kv_events_publisher` for the precise contract.
"kv_events": server_args.describe_kv_events_publisher(),
}
)
@app.get("/get_load")
@@ -1417,7 +1420,7 @@ async def load_lora_adapter(
"""Load a new LoRA adapter without re-launching the server."""
result = await _global_state.tokenizer_manager.load_lora_adapter(obj, request)
status_code = HTTPStatus.OK if result.success else HTTPStatus.BAD_REQUEST
return ORJSONResponse(result, status_code=status_code)
return ORJSONResponse(msgspec_to_builtins(result), status_code=status_code)
@app.api_route("/load_lora_adapter_from_tensors", methods=["POST"])
@@ -1429,7 +1432,7 @@ async def load_lora_adapter_from_tensors(
obj, request
)
status_code = HTTPStatus.OK if result.success else HTTPStatus.BAD_REQUEST
return ORJSONResponse(result, status_code=status_code)
return ORJSONResponse(msgspec_to_builtins(result), status_code=status_code)
@app.api_route("/unload_lora_adapter", methods=["POST"])
@@ -1440,7 +1443,7 @@ async def unload_lora_adapter(
"""Load a new LoRA adapter without re-launching the server."""
result = await _global_state.tokenizer_manager.unload_lora_adapter(obj, request)
status_code = HTTPStatus.OK if result.success else HTTPStatus.BAD_REQUEST
return ORJSONResponse(result, status_code=status_code)
return ORJSONResponse(msgspec_to_builtins(result), status_code=status_code)
@app.api_route("/open_session", methods=["GET", "POST"])