Clean rasa-x install containers don't start

Rasa version: 1.8.0 Rasa X version: 0.26.0 Python version: 3.6.9 Operating system: Ubuntu 18.04

Issue: I am installing Rasa X on a clean Ubuntu install on a new VPS using the docker one line method. When I run docker-compose up -d I can access the Rasa X UI through <domain>/login. However, the rasa-worker and rasa-production containers don’t start, they are in an endless restart loop with exit code 132 (docker container ls), also there are no logs for these containers (docker logs <container name>).

Do I have any missing dependencies on a clean ubuntu install? Or does anyone have any suggestion on how to fix this issue?


Through sheer luck I found out that this has to do with tensorflow. The scripts in the container try to import the tensorflow module and that dumps the 132 exit code when the server CPU does not support AVX.

Is there any way to define the tensorflow version that should be used?

You can use another TensorFlow version is to install from source. TensorFlow allows you to do a build with different CPU options.

I’ve only encountered the error you’re reporting on an old laptop. Do you have a system with a newer CPU that you can use?

I am bound to the hardware that the VPS hoster provides, they do not expose AVX at the moment.

I’m building tensorflow from source, but how do I tell the docker container to use my local build?

you would have to rebuild our Docker container from scratch i believe