I think dfx has some artificial delay in processing requests and maybe that changed in dfx 0.15. @Severin might give some insight about that.
The response you’re seeing is a response saying that the system does not yet know about the particular request you are polling for. This happens for a brief period of time before the replica that received the ingress message has distributed it to the whole subnet. I.e. it is an expected response an you should just keep polling. Eventually the request id will become known and the system will provide its status.
The only thing I see in the release notes for 0.15.0 that seem applicable might be " fix!: http_request_streaming_callback and get_chunk now require the sha256 parameter to be set", but seems like it shouldn’t apply to this situation.
I can’t replicate it by hand Do you have an estimate how often this happens? Also, does it happen with any other functions too? I don’t think there’s any special treatment for http_request in the replica.
This changelog entry refers to something in the asset canister. To me it looks like the error comes from the replica, and not the asset canister because the error also happens in the motoko/http_request example which doesn’t use an asset canister.
Do you get any other errors, even if not all the time? And could you maybe try to set the replica log level to something more verbose? For that you have to set <network name>.replica.log_level to info in networks.json, like here
It happens very often, if I keep executing the curl command it’s easy to get the error. Seems maybe 20-30% of the time, if you curl 20-30 times I would imagine you’ll get the error.