Reputation
Badges 1
32 × Eureka!ok great I ll check what other changes we have missed yesterday
was allow_archived
removed from Task.query_tasks?
alright, so actually we noticed that the problem disappears if we use only sync requests. Meaning if I create a sleep endpoint that is async we get the 502 but if it's sync we don't
we have tried both and got the same issue (gunicorn vs uvcorn).
No I meant creating a
@router.post(
"/sleep",
tags=["temp"],
response_description="Return HTTP Status Code 200 (OK)",
status_code=status.HTTP_200_OK,
response_model=TestResponse,
)
# def here instead of async def
def post_sleep(time_sleep: float) -> TestResponse:
""" """
time.sleep(time_sleep)
return TestResponse(status="OK")
that's a fair point. Actually we have switched from using siege because we believe it is causing the issues and are using Locust now instead. We have been running for days at the same rate and don't see any errors being reported...
no requests are being served as in there is no traffic indeed
so they ping the werb server?
what is actually setting the task status to Aborted
?
my understanding was that the deamon thread was deserializing the task of the control plane every 300 seconds by default
We put back the additional changes and so far it seems that this has solved our issue. Thanks a lot for the quick turnaround on this.
Hi @<1523701087100473344:profile|SuccessfulKoala55> ,
I'm running in almost the same error (see below) but I want to connect the the free clearml server version at None so I have set up the corresponding env variables in example.env:
CLEARML_WEB_HOST="
"
CLEARML_API_HOST="
"
CLEARML_FILES_HOST="
"
CLEARML_API_ACCESS_KEY="---"
CLEARML_API_SECRET_KEY="---"
CLEARML_SERVING_TASK_ID="---"
I have set up the right values from...
Hi Alex,
thanks for your answer. I'm curious about your third point using OutputModel. I could not figure out from the documentation how do you actually use it. I constructed the OutputModel object as such:
out = OutputModel(task, name="my_model", framework="xgboost")
However, I could not find any method in the doc that would allow me to pass the model object to that instance or said otherwise, I can't understand how to use that Output model to register my model which would be stored in a...
I'm not sure what to do with that info I must say since the serve_model is async for good reasons I guess
I have tested with an endpoint that basically add two numbers and never managed to trigger the 502. I'm starting to wonder if we are not running just too many workers. I had it wrong that 2 vcpus should mean 5 workers should be good but I think i should probably be closer to 2 but I m not sure why that would lead requests being dropped
ACtually the request are never registered to the gunicorn app, and the ALB log show that there is no response from the target "-".
Hi Martin,
- Actually we are using ALB with a 30 seconds timeout
- we do not have GPUs instances
- docker version 1.3.0
Geez, I have been looking for this for a while, thanks for saving my day...again.
I'm assuming that task.data.script.requirements is not the right way to do this...
I can't be sure of the version I can't check at the moment, I have 1.3.0 from the top of my head but could be way off
yeah I don't know I think we are probably just trying to fit to high a throughput for that box but it's weird that the packet just get dropped i would have assumed the response time should degrade and requests be queued.
Hey tahnks a lot Alex, that's exactly what I was looking for. cheers
we are actually building from our fork of the code into our own images and helm charts
hey Marin real quick actually, on your update to the requirements.txt file isn't that constraint on fastapi inconsistent?
how can you be >= 0.109.1 and lower than 0.96
Hey Martin, I will, but it's a bit more tricky because we have modifications in the code that I have to merge on our side
Hi Martin, thanks a lot for looking into this so quickly. Will you let me know the version number once it's pushed? Thanks!
ok so I haven't looked at the latest changes after the sync this morning but the ones we put in yesterday seems to have fixed the issue, the service is still running this morning at least.