Compare commits
10 Commits
| Author | SHA1 | Date | |
|---|---|---|---|
| 7d3be849d9 | |||
| 0397af719d | |||
| 4fdc314fd9 | |||
| 639d82f5b4 | |||
| 4e2f2311d0 | |||
| 38782d89bc | |||
| 0185216ccb | |||
| b20d9e714c | |||
| b1eb65d75d | |||
| 1d09d7fe96 |
+2
-2
@@ -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"
|
||||
|
||||
@@ -33,33 +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]:
|
||||
|
||||
Reference in New Issue
Block a user