Rasa deployment on Heroku

Rasa Open Source
1

I’ve tried deploying my rasa chatbot using a blog but it gives this error: 2021-06-10T07:16:27.636769+00:00 heroku[router]: at=error code=H10 desc=“App crashed” method=GET path="/mul_bot/test.html" host=newmul.herokuapp.com request_id=4050b39f-2c23-4a6c-aa5e-844ec3476067 fwd=“203.130.19.74” dyno= connect= service= status=503 bytes= protocol=https

For deployment, I have used this tutorial:

Moreover, I also want to do “rasa action run” to run forms…

2

Hi, I was trying to do exactly the same as you and after wasting a lot of time I found out that it´s because there isn´t enough memory storage.

I don´t have a solution for you. I´m still trying to deploy my chatbot in any kind of platform haha, but I wish you luck and contact me if you discover a way to deploy the bot

1 Like
3

I would highly recommend Google Cloud Run and their 2GB free containers. a lot more memory for Rasa model inference. I have been using Cloud Run, beside a small cold problem, it uses KNative and scales quite well if in future you would like to move on to a more orchestration based solution

4

Hi! Have you got any tutorial of how to do it? Thanks

5

Just created one. I suppose this might be useful

Let me know if this helps. I tested it now and it does work for me. You will have a cold start on the first request but then it is quite fast. For prototypes and even for running POCs, this is quite useful. Upgrade if you have always on functionality. Similar to Heroku.

6

Hi, currently I’ve not found any solution to this problem but I think I need to try over GCP to deploy my bot

7

Hi, @souvikg10 I’m looking more into a free test version for deployment as I’m a student. So, I didn’t tried for GCP as I need to enter my card info for sign up. Can you guide me to deploy it into any similar platfrom where I can test it for free without entering any card info…

8

You can take a look at this list

Most of them provide a free k8s cluster without signup options.

Dockerize your rasa bot and create a helm chart. then you can install in any cluster. keep in mind, as your bot grows bigger, your memory requirements too. There is no forever free solution for rasa.

But you can get the most out of it using k8s , some provide containers with 2vCPU and 1Gi as forever free but i guess they will be rate limited. Still 1Gi is decent to run rasa bot but you need to be resourceful.

This looks the best according to me: Okteto