Compare commits
3 Commits
| Author | SHA1 | Date | |
|---|---|---|---|
| fdd50a2aaa | |||
| 70f8a8f534 | |||
| 7be8aa6397 |
+1
-1
@@ -66,7 +66,7 @@ class AuthData:
|
|||||||
"""data used to authenticate requester"""
|
"""data used to authenticate requester"""
|
||||||
|
|
||||||
cost: str
|
cost: str
|
||||||
endpoint_id: int
|
endpoint: str
|
||||||
reqnum: int
|
reqnum: int
|
||||||
request_idx: int
|
request_idx: int
|
||||||
signature: str
|
signature: str
|
||||||
|
|||||||
+3
-7
@@ -75,7 +75,6 @@ def print_truncate_res(res: str):
|
|||||||
@dataclass
|
@dataclass
|
||||||
class ClientState:
|
class ClientState:
|
||||||
endpoint_group_name: str
|
endpoint_group_name: str
|
||||||
endpoint_id: int
|
|
||||||
api_key: str
|
api_key: str
|
||||||
server_url: str
|
server_url: str
|
||||||
worker_endpoint: str
|
worker_endpoint: str
|
||||||
@@ -96,7 +95,7 @@ class ClientState:
|
|||||||
self.status = ClientStatus.Error
|
self.status = ClientStatus.Error
|
||||||
return
|
return
|
||||||
route_payload = {
|
route_payload = {
|
||||||
"endpoint_id": self.endpoint_id,
|
"endpoint": self.endpoint_group_name,
|
||||||
"api_key": self.api_key,
|
"api_key": self.api_key,
|
||||||
"cost": self.payload.count_workload(),
|
"cost": self.payload.count_workload(),
|
||||||
}
|
}
|
||||||
@@ -245,19 +244,16 @@ def run_test(
|
|||||||
print_thread = threading.Thread(target=print_state, args=(clients, num_requests))
|
print_thread = threading.Thread(target=print_state, args=(clients, num_requests))
|
||||||
print_thread.daemon = True # makes threads get killed on program exit
|
print_thread.daemon = True # makes threads get killed on program exit
|
||||||
print_thread.start()
|
print_thread.start()
|
||||||
endpoint_info = Endpoint.get_endpoint_info(
|
endpoint_api_key = Endpoint.get_endpoint_api_key(
|
||||||
endpoint_name=endpoint_group_name, account_api_key=api_key, instance=instance
|
endpoint_name=endpoint_group_name, account_api_key=api_key, instance=instance
|
||||||
)
|
)
|
||||||
if not endpoint_info:
|
if not endpoint_api_key:
|
||||||
log.debug(f"Endpoint {endpoint_group_name} not found for API key")
|
log.debug(f"Endpoint {endpoint_group_name} not found for API key")
|
||||||
return
|
return
|
||||||
endpoint_id = endpoint_info["id"]
|
|
||||||
endpoint_api_key = endpoint_info["api_key"]
|
|
||||||
try:
|
try:
|
||||||
for _ in range(num_requests):
|
for _ in range(num_requests):
|
||||||
client = ClientState(
|
client = ClientState(
|
||||||
endpoint_group_name=endpoint_group_name,
|
endpoint_group_name=endpoint_group_name,
|
||||||
endpoint_id=endpoint_id,
|
|
||||||
api_key=endpoint_api_key,
|
api_key=endpoint_api_key,
|
||||||
server_url=server_url,
|
server_url=server_url,
|
||||||
worker_endpoint=worker_endpoint,
|
worker_endpoint=worker_endpoint,
|
||||||
|
|||||||
+4
-2
@@ -1,4 +1,6 @@
|
|||||||
aiohttp[speedups]==3.10.1
|
aiohttp==3.10.1
|
||||||
|
aiodns~=3.6.0
|
||||||
|
pycares~=4.11.0
|
||||||
anyio~=4.4
|
anyio~=4.4
|
||||||
lib~=4.0
|
lib~=4.0
|
||||||
nltk~=3.9
|
nltk~=3.9
|
||||||
@@ -8,4 +10,4 @@ Requests~=2.32
|
|||||||
transformers~=4.52
|
transformers~=4.52
|
||||||
utils==1.0.*
|
utils==1.0.*
|
||||||
hf_transfer>=0.1.9
|
hf_transfer>=0.1.9
|
||||||
vastai-sdk>=0.2.0
|
vastai-sdk>=0.2.0
|
||||||
|
|||||||
+11
-14
@@ -13,11 +13,11 @@ from vastai import Serverless
|
|||||||
ENDPOINT_NAME = "my-comfyui-endpoint"
|
ENDPOINT_NAME = "my-comfyui-endpoint"
|
||||||
COST = 100 # Use a constant cost for image generation
|
COST = 100 # Use a constant cost for image generation
|
||||||
|
|
||||||
def call_default_workflow(endpoint_id: int, api_key: str, server_url: str) -> None:
|
def call_default_workflow(client: Serverless) -> None:
|
||||||
WORKER_ENDPOINT = "/prompt"
|
WORKER_ENDPOINT = "/prompt"
|
||||||
COST = 100
|
COST = 100
|
||||||
route_payload = {
|
route_payload = {
|
||||||
"endpoint_id": endpoint_id,
|
"endpoint": endpoint_group_name,
|
||||||
"api_key": api_key,
|
"api_key": api_key,
|
||||||
"cost": COST,
|
"cost": COST,
|
||||||
}
|
}
|
||||||
@@ -32,7 +32,7 @@ def call_default_workflow(endpoint_id: int, api_key: str, server_url: str) -> No
|
|||||||
auth_data = dict(
|
auth_data = dict(
|
||||||
signature=message["signature"],
|
signature=message["signature"],
|
||||||
cost=message["cost"],
|
cost=message["cost"],
|
||||||
endpoint_id=message["endpoint_id"],
|
endpoint=message["endpoint"],
|
||||||
reqnum=message["reqnum"],
|
reqnum=message["reqnum"],
|
||||||
url=message["url"],
|
url=message["url"],
|
||||||
)
|
)
|
||||||
@@ -52,12 +52,12 @@ def call_default_workflow(endpoint_id: int, api_key: str, server_url: str) -> No
|
|||||||
|
|
||||||
|
|
||||||
def call_custom_workflow_for_sd3(
|
def call_custom_workflow_for_sd3(
|
||||||
endpoint_id: int, api_key: str, server_url: str
|
endpoint_group_name: str, api_key: str, server_url: str
|
||||||
) -> None:
|
) -> None:
|
||||||
WORKER_ENDPOINT = "/custom-workflow"
|
WORKER_ENDPOINT = "/custom-workflow"
|
||||||
COST = 100
|
COST = 100
|
||||||
route_payload = {
|
route_payload = {
|
||||||
"endpoint_id": endpoint_id,
|
"endpoint": endpoint_group_name,
|
||||||
"api_key": api_key,
|
"api_key": api_key,
|
||||||
"cost": COST,
|
"cost": COST,
|
||||||
}
|
}
|
||||||
@@ -72,7 +72,7 @@ def call_custom_workflow_for_sd3(
|
|||||||
auth_data = dict(
|
auth_data = dict(
|
||||||
signature=message["signature"],
|
signature=message["signature"],
|
||||||
cost=message["cost"],
|
cost=message["cost"],
|
||||||
endpoint_id=message["endpoint_id"],
|
endpoint=message["endpoint"],
|
||||||
reqnum=message["reqnum"],
|
reqnum=message["reqnum"],
|
||||||
url=message["url"],
|
url=message["url"],
|
||||||
request_idx=message["request_idx"],
|
request_idx=message["request_idx"],
|
||||||
@@ -146,28 +146,25 @@ def call_custom_workflow_for_sd3(
|
|||||||
if __name__ == "__main__":
|
if __name__ == "__main__":
|
||||||
from lib.test_utils import test_args
|
from lib.test_utils import test_args
|
||||||
|
|
||||||
log = logging.getLogger(__name__)
|
|
||||||
args = test_args.parse_args()
|
args = test_args.parse_args()
|
||||||
endpoint_info = Endpoint.get_endpoint_info(
|
endpoint_api_key = Endpoint.get_endpoint_api_key(
|
||||||
endpoint_name=args.endpoint_group_name,
|
endpoint_name=args.endpoint_group_name,
|
||||||
account_api_key=args.api_key,
|
account_api_key=args.api_key,
|
||||||
instance=args.instance,
|
instance=args.instance,
|
||||||
)
|
)
|
||||||
if endpoint_info:
|
if endpoint_api_key:
|
||||||
endpoint_id = endpoint_info["id"]
|
|
||||||
endpoint_api_key = endpoint_info["api_key"]
|
|
||||||
try:
|
try:
|
||||||
call_default_workflow(
|
call_default_workflow(
|
||||||
endpoint_id=endpoint_id,
|
|
||||||
api_key=endpoint_api_key,
|
api_key=endpoint_api_key,
|
||||||
|
endpoint_group_name=args.endpoint_group_name,
|
||||||
server_url=args.server_url,
|
server_url=args.server_url,
|
||||||
)
|
)
|
||||||
call_custom_workflow_for_sd3(
|
call_custom_workflow_for_sd3(
|
||||||
endpoint_id=endpoint_id,
|
|
||||||
api_key=endpoint_api_key,
|
api_key=endpoint_api_key,
|
||||||
|
endpoint_group_name=args.endpoint_group_name,
|
||||||
server_url=args.server_url,
|
server_url=args.server_url,
|
||||||
)
|
)
|
||||||
except Exception as e:
|
except Exception as e:
|
||||||
log.error(f"Error during API call: {e}")
|
log.error(f"Error during API call: {e}")
|
||||||
else:
|
else:
|
||||||
log.error(f"Failed to get endpoint info for {args.endpoint_group_name}")
|
log.error(f"Failed to get API key for endpoint {args.endpoint_group_name} ")
|
||||||
|
|||||||
@@ -60,7 +60,7 @@ def do_one(endpoint_name: str,
|
|||||||
worker_session):
|
worker_session):
|
||||||
try:
|
try:
|
||||||
workload = payload.count_workload()
|
workload = payload.count_workload()
|
||||||
route_payload = {"endpoint_id": endpoint_id, "api_key": endpoint_api_key, "cost": workload}
|
route_payload = {"endpoint": endpoint_name, "api_key": endpoint_api_key, "cost": workload}
|
||||||
headers = {"Authorization": f"Bearer {endpoint_api_key}"}
|
headers = {"Authorization": f"Bearer {endpoint_api_key}"}
|
||||||
start = time.time()
|
start = time.time()
|
||||||
r0 = route_session.post(urljoin(server_url, "/route/"), json=route_payload, headers=headers, timeout=4)
|
r0 = route_session.post(urljoin(server_url, "/route/"), json=route_payload, headers=headers, timeout=4)
|
||||||
|
|||||||
Reference in New Issue
Block a user