Hi,
The http response on Rasa Open Source for sequence tagging and intent extraction (/model/parse endpoint) is very slow (more than 1500ms on average) with so many timeouts. The rasa 1.10.1-full docker is used on an OpenShift based cloud with 2Gi of ram and 2 CPU cores and horizontal scaling does not help so much. What is the problem here that the rasa server is so slow?
Tests are done with wrk tool with 20 threads and 300 connections.
Hi Kamyar,
Have you run htop or anything to look at the resource usage. 2Gb doesn’t sound like enough memory for that kind of load. The recommended minimum resource requirements start with 4GB memory.
Greg
I have tested it with up to 8GB of memory. The problem is rasa would be very slow with large lookup table files in run time.