Compare commits

...

16 Commits

Author SHA1 Message Date
Lucas Armand 7d3be849d9 Handle errors from model for comfyui-json 2025-10-08 12:00:45 -07:00
Colter-Downing 0397af719d Merge pull request #37 from robballantyne/bugfix/healthcheck-endpoint
Fix healthcheck endpoint URL

Tested and merged by Colter
2025-10-06 15:11:27 -07:00
Rob Ballantyne 4fdc314fd9 Fix healthcheck endpoint URL 2025-10-06 22:16:09 +01:00
Colter-Downing 639d82f5b4 Merge pull request #35 from vast-ai/AUTO-664--Healthcheck-error
Fix healthcheck with separate session
2025-10-02 12:51:19 -07:00
Colter Downing 25db78e39d Fix healthcheck with separate session 2025-10-01 18:04:31 -07:00
Scott-Laytart 4e2f2311d0 Merge pull request #33 from vast-ai/comfy-blind-fix-override
undo the fix for comfy yesterday.
2025-09-03 11:50:07 -07:00
abiola-vastai 38782d89bc undo the fix for comfy yesterday. 2025-09-03 17:12:35 +00:00
Scott-Laytart 0185216ccb Merge pull request #32 from vast-ai/blindhotfix_comfy_ui_default_port
Blind hotfix to see if comfy UI default is needed. if it does work we…
2025-09-02 18:26:25 -07:00
abiola-vastai b20d9e714c Blind hotfix to see if comfy UI default is needed. if it does work we would revert back. 2025-09-03 01:20:09 +00:00
Rob Ballantyne b1eb65d75d Merge pull request #31 from vast-ai/bugfix/startup-script-20250901
Update uv venv creation command
2025-09-01 18:19:17 +01:00
Rob Ballantyne 1d09d7fe96 Update uv venv creation command 2025-09-01 16:55:20 +01:00
Colter-Downing 1b37054dec Merge pull request #28 from vast-ai/bugfix/backend-timeout-infinite
Bugfix/backend timeout infinite
2025-08-28 11:22:33 -07:00
Colter-Downing 1a1e4174b8 Merge pull request #29 from vast-ai/bugfix/comfyui-json-cost-fix
Set cost to 100
2025-08-28 11:22:21 -07:00
Rob Ballantyne 1e4fa87437 Prevent timeout and allow long running connections 2025-08-28 15:48:57 +01:00
Rob Ballantyne 4c5fa03c7b adds import for ClientTimeout 2025-08-27 20:54:27 +01:00
Rob Ballantyne a8fe74f771 Remove default 300s timeout 2025-08-27 18:34:45 +01:00
3 changed files with 56 additions and 33 deletions
+21 -4
View File
@@ -11,7 +11,7 @@ from functools import cached_property
from distutils.util import strtobool
from anyio import open_file
from aiohttp import web, ClientResponse, ClientSession, ClientConnectorError
from aiohttp import web, ClientResponse, ClientSession, ClientConnectorError, ClientTimeout, TCPConnector
import requests
from Crypto.Signature import pkcs1_15
@@ -75,7 +75,13 @@ class Backend:
@cached_property
def session(self):
log.debug(f"starting session with {self.model_server_url}")
return ClientSession(self.model_server_url)
connector = TCPConnector(
force_close=True, # Required for long running jobs
enable_cleanup_closed=True,
)
timeout = ClientTimeout(total=None)
return ClientSession(self.model_server_url, timeout=timeout, connector=connector)
def create_handler(
self,
@@ -184,18 +190,30 @@ class Backend:
log.debug(f"Exception in main handler loop {e}")
return web.Response(status=500)
@cached_property
def healthcheck_session(self):
"""Dedicated session for healthchecks to avoid conflicts with API session"""
log.debug("creating dedicated healthcheck session")
connector = TCPConnector(
force_close=True, # Keep this for isolation
enable_cleanup_closed=True,
)
timeout = ClientTimeout(total=10) # Reasonable timeout for healthchecks
return ClientSession(timeout=timeout, connector=connector)
async def __healthcheck(self):
health_check_url = self.benchmark_handler.healthcheck_endpoint
if health_check_url is None:
log.debug("No healthcheck endpoint defined, skipping healthcheck")
return
while True:
await sleep(10)
if self.__start_healthcheck is False:
continue
try:
log.debug(f"Performing healthcheck on {health_check_url}")
async with self.session.get(health_check_url) as response:
async with self.healthcheck_session.get(health_check_url) as response:
if response.status == 200:
log.debug("Healthcheck successful")
elif response.status == 503:
@@ -204,7 +222,6 @@ class Backend:
f"Healthcheck failed with status: {response.status}"
)
else:
# endpoint not ready yet so bail
log.debug(f"Healthcheck Endpoint not ready: {response.status}")
except Exception as e:
log.debug(f"Healthcheck failed with exception: {e}")
+2 -2
View File
@@ -59,12 +59,12 @@ then
fi
# Fork testing
git clone "${PYWORKER_REPO:-https://github.com/vast-ai/pyworker}" "$SERVER_DIR"
[[ ! -d $SERVER_DIR ]] && git clone "${PYWORKER_REPO:-https://github.com/vast-ai/pyworker}" "$SERVER_DIR"
if [[ -n ${PYWORKER_REF:-} ]]; then
(cd "$SERVER_DIR" && git checkout "$PYWORKER_REF")
fi
uv venv --managed-python "$ENV_PATH" -p 3.10
uv venv --python-preference only-managed "$ENV_PATH" -p 3.10
source "$ENV_PATH/bin/activate"
uv pip install -r "${SERVER_DIR}/requirements.txt"
+33 -27
View File
@@ -33,33 +33,39 @@ log = logging.getLogger(__file__)
async def generate_client_response(
client_request: web.Request, model_response: ClientResponse
) -> Union[web.Response, web.StreamResponse]:
# Check if the response is actually streaming based on response headers/content-type
is_streaming_response = (
model_response.content_type == "text/event-stream"
or model_response.content_type == "application/x-ndjson"
or model_response.headers.get("Transfer-Encoding") == "chunked"
or "stream" in model_response.content_type.lower()
)
match model_response.status:
case 200:
log.debug("SUCCESS")
# Check if the response is actually streaming based on response headers/content-type
is_streaming_response = (
model_response.content_type == "text/event-stream"
or model_response.content_type == "application/x-ndjson"
or model_response.headers.get("Transfer-Encoding") == "chunked"
or "stream" in model_response.content_type.lower()
)
if is_streaming_response:
log.debug("Detected streaming response...")
res = web.StreamResponse()
res.content_type = model_response.content_type
await res.prepare(client_request)
async for chunk in model_response.content:
await res.write(chunk)
await res.write_eof()
log.debug("Done streaming response")
return res
else:
log.debug("Detected non-streaming response...")
content = await model_response.read()
return web.Response(
body=content,
status=model_response.status,
content_type=model_response.content_type
)
if is_streaming_response:
log.debug("Detected streaming response...")
res = web.StreamResponse()
res.content_type = model_response.content_type
await res.prepare(client_request)
async for chunk in model_response.content:
await res.write(chunk)
await res.write_eof()
log.debug("Done streaming response")
return res
else:
log.debug("Detected non-streaming response...")
content = await model_response.read()
return web.Response(
body=content,
status=model_response.status,
content_type=model_response.content_type
)
case code:
log.debug(f"Model responded with error {code}")
return web.Response(status=code)
@dataclasses.dataclass
class ComfyWorkflowHandler(EndpointHandler[ComfyWorkflowData]):
@@ -70,7 +76,7 @@ class ComfyWorkflowHandler(EndpointHandler[ComfyWorkflowData]):
@property
def healthcheck_endpoint(self) -> Optional[str]:
return "/health"
return f"{MODEL_SERVER_URL}/health"
@classmethod
def payload_cls(cls) -> Type[ComfyWorkflowData]: