diff --git a/lib/test_utils.py b/lib/test_utils.py index 791b7dd..ba97611 100644 --- a/lib/test_utils.py +++ b/lib/test_utils.py @@ -53,6 +53,13 @@ test_args.add_argument( default="https://run.vast.ai", help="Call local autoscaler instead of prod, for dev use only", ) +test_args.add_argument( + "-i", + dest="instance", + type=str, + default="prod", + help="Autoscaler shard to run the command against, default: prod", +) GetPayloadAndWorkload = Callable[[], Tuple[Dict[str, Any], float]] @@ -70,6 +77,7 @@ class ClientState: api_key: str server_url: str worker_endpoint: str + instance: str payload: ApiPayload url: str = "" status: ClientStatus = ClientStatus.FetchEndpoint @@ -79,11 +87,7 @@ class ClientState: def make_call(self): self.status = ClientStatus.FetchEndpoint - endpoint_api_key = Endpoint.get_endpoint_api_key( - endpoint_name=self.endpoint_group_name, - account_api_key=self.api_key, - ) - if not endpoint_api_key: + if not self.api_key: self.as_error.append( f"Endpoint {self.endpoint_group_name} not found for API key", ) @@ -91,12 +95,14 @@ class ClientState: return route_payload = { "endpoint": self.endpoint_group_name, - "api_key": endpoint_api_key, + "api_key": self.api_key, "cost": self.payload.count_workload(), } + headers = {"Authorization": f"Bearer {self.api_key}"} response = requests.post( urljoin(self.server_url, "/route/"), json=route_payload, + headers=headers, timeout=4, ) if response.status_code != 200: @@ -135,6 +141,7 @@ class ClientState: try: self.make_call() except Exception as e: + print(e) self.status = ClientStatus.Error _ = e self.conn_errors[self.url] += 1 @@ -226,6 +233,7 @@ def run_test( server_url: str, worker_endpoint: str, payload_cls: Type[ApiPayload], + instance: str, ): threads = [] @@ -234,8 +242,7 @@ def run_test( print_thread.daemon = True # makes threads get killed on program exit print_thread.start() endpoint_api_key = Endpoint.get_endpoint_api_key( - endpoint_name=endpoint_group_name, - account_api_key=api_key, + endpoint_name=endpoint_group_name, account_api_key=api_key, instance=instance ) if not endpoint_api_key: log.debug(f"Endpoint {endpoint_group_name} not found for API key") @@ -248,6 +255,7 @@ def run_test( server_url=server_url, worker_endpoint=worker_endpoint, payload=payload_cls.for_test(), + instance=instance, ) clients.append(client) thread = threading.Thread(target=client.simulate_user, args=()) @@ -281,12 +289,19 @@ def test_load_cmd( args = arg_parser.parse_args() if hasattr(args, "comfy_model"): os.environ["COMFY_MODEL"] = args.comfy_model + server_url = dict( + prod="https://run.vast.ai", + alpha="https://run-alpha.vast.ai", + candidate="https://run-candidate.vast.ai", + local="http://localhost:8080", + )[args.instance] run_test( num_requests=args.num_requests, requests_per_second=args.requests_per_second, api_key=args.api_key, - server_url=args.server_url, + server_url=server_url, endpoint_group_name=args.endpoint_group_name, worker_endpoint=endpoint, payload_cls=payload_cls, + instance=args.instance, ) diff --git a/start_server.sh b/start_server.sh index 57a96e8..7f54a5b 100755 --- a/start_server.sh +++ b/start_server.sh @@ -87,17 +87,17 @@ if [ "$USE_SSL" = true ]; then IP.1 = 0.0.0.0 EOF -openssl req -newkey rsa:2048 -subj "/C=US/ST=CA/CN=pyworker.vast.ai/" \ - -nodes \ - -sha256 \ - -keyout /etc/instance.key \ - -out /etc/instance.csr \ - -config /etc/openssl-san.cnf + openssl req -newkey rsa:2048 -subj "/C=US/ST=CA/CN=pyworker.vast.ai/" \ + -nodes \ + -sha256 \ + -keyout /etc/instance.key \ + -out /etc/instance.csr \ + -config /etc/openssl-san.cnf -curl --header 'Content-Type: application/octet-stream' \ - --data-binary @//etc/instance.csr \ - -X \ - POST "https://console.vast.ai/api/v0/sign_cert/?instance_id=$CONTAINER_ID" > /etc/instance.crt; + curl --header 'Content-Type: application/octet-stream' \ + --data-binary @//etc/instance.csr \ + -X \ + POST "https://console.vast.ai/api/v0/sign_cert/?instance_id=$CONTAINER_ID" > /etc/instance.crt; fi diff --git a/utils/endpoint_util.py b/utils/endpoint_util.py index 42fea1c..48c4cdb 100644 --- a/utils/endpoint_util.py +++ b/utils/endpoint_util.py @@ -17,7 +17,9 @@ class Endpoint: """ @staticmethod - def get_endpoint_api_key(endpoint_name: str, account_api_key: str) -> Optional[str]: + def get_endpoint_api_key( + endpoint_name: str, account_api_key: str, instance: str + ) -> Optional[str]: """ Fetch endpoint API key from VastAI console following the healthcheck pattern. @@ -33,7 +35,9 @@ class Endpoint: try: log.debug(f"Fetching endpoint API key for endpoint: {endpoint_name}") - response = requests.get(vast_console_url, headers=headers) + response = requests.get( + f"{vast_console_url}?autoscaler_instance={instance}", headers=headers + ) if response.status_code != 200: error_msg = f"Failed to fetch endpoint API key: {response.status_code} - {response.text}" diff --git a/workers/comfyui/client.py b/workers/comfyui/client.py index 23c8ca0..6563e00 100644 --- a/workers/comfyui/client.py +++ b/workers/comfyui/client.py @@ -153,6 +153,7 @@ if __name__ == "__main__": endpoint_api_key = Endpoint.get_endpoint_api_key( endpoint_name=args.endpoint_group_name, account_api_key=args.api_key, + instance=args.instance, ) if endpoint_api_key: try: diff --git a/workers/tgi/client.py b/workers/tgi/client.py index cfa91f8..7e4f1bb 100644 --- a/workers/tgi/client.py +++ b/workers/tgi/client.py @@ -100,6 +100,7 @@ if __name__ == "__main__": endpoint_api_key = Endpoint.get_endpoint_api_key( endpoint_name=args.endpoint_group_name, account_api_key=args.api_key, + instance=args.instance, ) if endpoint_api_key: try: