This message was deleted.
# ask-for-helps
Slackbot
01/13/2023, 8:58 AMThis message was deleted.
j
Jiri Suchomel
02/03/2023, 1:31 PMHi @Benjamin Tan, did you find a solution to your issue? We’re experiencing it as well…
b
Benjamin Tan
02/03/2023, 1:32 PMOhai! could be proxy body size of the ingress. I had to set an annotation there.
Benjamin Tan
02/03/2023, 1:32 PMWhat's your setup?
j
Jiri Suchomel
02/03/2023, 1:34 PMwe have a docker container in AWS App runner, no Kubernetes
b
Benjamin Tan
02/03/2023, 1:35 PMMight have an option to change the proxy body size in app runner?
j
Jiri Suchomel
02/03/2023, 1:36 PMI’ll take a look, thanks for the hint
b
Benjamin Tan
02/03/2023, 1:36 PMDoes it work with something smaller?
Benjamin Tan
02/03/2023, 1:37 PMAt least isolate that as a problem first
j
Jiri Suchomel
02/03/2023, 1:38 PMcalls with smaller data seem to work, yes
b
Benjamin Tan
02/03/2023, 1:47 PMThere might be some logs in the api runner that might be useful
j
Jiri Suchomel
02/03/2023, 1:49 PMnothing really, I only found this aiohttp.client_exceptions.ServerDisconnectedError: Server disconnected that’s why I came here
b
Benjamin Tan
02/03/2023, 3:13 PMoh another t hing
Benjamin Tan
02/03/2023, 3:13 PMis your runner downloading stuff during first init?
Benjamin Tan
02/03/2023, 3:13 PMyou might need to increase the timeout
Benjamin Tan
02/03/2023, 3:15 PManother thing i tried was to increase the amount of resources too
j
Jiri Suchomel
02/03/2023, 3:49 PMno, no initial download
Jiri Suchomel
02/03/2023, 3:50 PMbut the timeout part might be a thing, based on what they write about aiohttp
Jiri Suchomel
02/03/2023, 3:51 PMthanks for hints
s
Sudeep Ghimire
03/13/2023, 5:04 PM@Jiri Suchomel did you solve this issue? Facing the same issue on our end.
j
Jiri Suchomel
03/13/2023, 6:20 PMNot really. I think it must be some bug in bentoml or aiohttp.
Jiri Suchomel
03/13/2023, 6:20 PMWe actually switched to different framework with this model 😞
Jiri Suchomel
03/13/2023, 6:21 PMActually, raising memory helped (when I reproduced the problem locally). But our infra setup did not allow to increase memory
6 Views