Bots in production

Rasa Open Source
65

As long as you give each user a unique sender_id in the requests you make to your input channel, this shouldn’t be an issue – each person will have their own unique conversation with the bot.

66

Can you share with me how can we do this, any configuration changes required or we have to code the same?

67

@shubham3582 I too want to know how to handle chat with multiple users at the same time, you are talking about making request with unique sender_id. you are also talking about application session/bearer token. I don’t know how to achieve this. If you can help me out it will be a great support from your side. Is it something to do with tracker.sender_id inside actions.py?

68

I might help you better, if you elaborate the exact scenario.

69

@shubham3582 suppose I have a rasa bot running on my local and connected with slack group channel. In group there are multiple users who are chatting with rasa bot at the same time. There is problem that rasa bot can’t handle multiple users at same time. Is it due to my rasa bot running local that’s why it can’t handle it. If I deploy me bot to cloud using aws or gcp or Azure services will bot handle multiple users at same time.

Is there a way to handle each user in that situation?

71

I want to run multiple instances of Rasa on different servers so that we can handle 100s conversations at the same time. How can I do that?

73

@TQuy I think… Its related to how you deploy rasa… I think u deploy through docker with kubernetes on server cloud platform will help you to scale the rasa… I haven’t tried that yet… what u think in this

74

Hi, Is there anyone in Asia who has deployed a bot for any bank? I love to you hear your story about building a banking chatbot using RASA.

75

Do you guys know about some companies using it in Brazil?

1 Like
76

Hi @bayesianwannabe rs. I’m brazilian and I’m working on a chatbot (look for my name at introduction thread on this forum). The best case I’ve found was Lappis, built by a group from UnB (GitHub - lappis-unb/rasa-ptbr-boilerplate: Um template para criar um FAQ chatbot usando Rasa, Rocket.chat, elastic search). There is a telegram group with interested people either, take a look at the repo.

77

Hey Luan,

Thanks for your kind reply. Do you have a chatbot using Rasa in production? Can you tell more about the number of intents and your greatest challenge?

Lappis is the biggest case I know and the boilerplate project seems neat (I still need to test, but I am now more dedicated in a design/model prototyping part of the project), but I wonder if there are big companies on Brazil using this technology like banks or big industries.

78

I’m waiting for correct answer regarding bots in production. I built a FAQ bot and my organisation is not happy with it as it cannot be deployed in website and I said I will do that but I’m not getting exact resource in the rasa docs so please help me regarding this. If I want to deploy in a website what is the exact process for it please share it @Juste. Thank you

79

Hello @skjainmiah. I just finished deploying RASA in a WordPress website so I’ll show you what I did and maybe this could help:

  1. I’m running my RASA chatbot using ngrok, so I have my ngrok URL opening a specific port to the internet (in this case, port 5005).

  2. I added the following code to credentials.yml, to be able to use socketio:

    socketio: user_message_evt: user_uttered bot_message_evt: bot_uttered session_persistence: true

  3. I used webchat and deployed in my frontend using the tag you can find here: GitHub - botfront/rasa-webchat: A feature-rich chat widget for Rasa and Botfront

  4. I changed “socketUrl” with my ngrok URL.

And that’s it. Bot is working in frontend.

Hope this can help. :slight_smile:

1 Like
80

Hi @tacopanaya. It was good to hear you deployed on wordpress. How many people can access your chatbot at a time? does ngrok running continuously in the backend?

81

Hi @skjainmiah. We are still in dev alpha stage. Up until now we were using Telegram for testing purposes but we faced some issues with button generation that weren’t present when we tested in webchat. As we are probably going to move to a mobile app in the future for full control over what we are creating, we decided moving to a webchat was a good next step. So we open ngrok before testing and whenever we need to show it to someone. We are probably going to go for an Amazon EC3 server in the future whenever we decide to open for public testing. :slight_smile:

However, as far as I’ve seen in Rasa documentation here at the bottom, ngrok limit is set to 40 connections/minute.

1 Like
82

I want to know if a rasa core model can handle 2k-3k concurrent request on a 15gb ram server?

2 Likes
83

Hi, We are using Rasa Open Source for NLU. The http response for sequence tagging and intent extraction (/model/parse endpoint) is very slow (about than 2000ms on average and about 22 rps that is very low) 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. We have tseted it in a local machine with 64GB of ram and 12 i7 cores, the results are the same. What is the problem here that the rasa server is so slow for model parsing? Tests are done with wrk tool with 20 threads and 30 connections. With higher connection numbers than 20 or 30 it will even make all the requests to time out. Lower than 20 or 30 connections it is as I have described.

84

Hi @bayesianwannabe, take a look here.

85

Nice! Thank you, Geovana. It’s always good to check new local cases using the Rasa framework.

1 Like
86

Hi can you guide me on the cost of having this? I am trying to pitch it internally in my company and require a basic cost. I would have similar numbers as you defined earlier.

Also if you could let me know if you are using AWS or some other platform