Unverified Commit fe43e889 authored by fzyzcjy's avatar fzyzcjy Committed by GitHub
Browse files

Fix mini lb timeout issue (#9369)

parent 5ae5ecaa
......@@ -118,7 +118,13 @@ def main():
lb_args = LBArgs.from_cli_args(args)
prefill_configs = [PrefillConfig(url, port) for url, port in lb_args.prefill_infos]
run(prefill_configs, lb_args.decode_infos, lb_args.host, lb_args.port)
run(
prefill_configs,
lb_args.decode_infos,
lb_args.host,
lb_args.port,
lb_args.timeout,
)
if __name__ == "__main__":
......
......@@ -50,10 +50,16 @@ class PrefillConfig:
class MiniLoadBalancer:
def __init__(self, prefill_configs: List[PrefillConfig], decode_servers: List[str]):
def __init__(
self,
prefill_configs: List[PrefillConfig],
decode_servers: List[str],
timeout: int,
):
self.prefill_configs = prefill_configs
self.prefill_servers = [p.url for p in prefill_configs]
self.decode_servers = decode_servers
self.timeout = timeout
def add_prefill_server(self, new_prefill_config: PrefillConfig):
self.prefill_configs.append(new_prefill_config)
......@@ -78,7 +84,7 @@ class MiniLoadBalancer:
async with aiohttp.ClientSession(
timeout=aiohttp.ClientTimeout(
total=3600
total=self.timeout
) # Add timeout for request reliability
) as session:
tasks = [
......@@ -117,7 +123,7 @@ class MiniLoadBalancer:
async def stream_results():
async with aiohttp.ClientSession(
timeout=aiohttp.ClientTimeout(
total=3600
total=self.timeout
) # Add timeout for request reliability
) as session:
# Create the tasks for both prefill and decode requests
......@@ -401,9 +407,9 @@ async def register(obj: PDRegistryRequest):
return Response(status_code=200)
def run(prefill_configs, decode_addrs, host, port):
def run(prefill_configs, decode_addrs, host, port, timeout):
global load_balancer
load_balancer = MiniLoadBalancer(prefill_configs, decode_addrs)
load_balancer = MiniLoadBalancer(prefill_configs, decode_addrs, timeout=timeout)
uvicorn.run(app, host=host, port=port)
......
Markdown is supported
0% or .
You are about to add 0 people to the discussion. Proceed with caution.
Finish editing this message first!
Please register or to comment