Botkit: Maintaining bot state between deployments.

[srbartlett]

Not an issue per se but a question on how a bot can maintain state between deployment.

For example, if a bot (slack platform) is having a private conversation and during the conversation a deployment is pushed the conversation thread is lost.

When a bot is deployed and new RTM connection is establish and clearly any state the bot has is memory is lost.

Continuous deployment is an important practice so I am wondering how it can be achieved using the very awesome bitkit library.

[jamesmgg]

@srbartlett, I 100% agree. I am facing this problem as well and am wondering if others have already solved it... If not, I was thinking of creating a fork of BotKit where I can store a hash in a db for every user which is associated to a conversation node... so if the server is reset (cough, cough Heroku...) the bot can smartly know where it is in the conversation flow.

[dfischer]

Definitely an issue and would be wise to borrow from existing solutions in http cycles

[DigitalKrony]

I'm leveraging the filesystem built directly into NODE to resolve this. See docs here.

[s5]

I'm facing the same problem. Ideally a solution would leverage botkit's custom storage system. Has anyone worked on a solution?

[alexbudin]

Can anyone help with saving the state of a conversation flow? Has anybody found any solutions?

[srbartlett]

@alexbudin Not exactly a solution but using the the Slack Events API allows you to be stateless protocol. https://api.slack.com/events

[dannyfreed]

I've found a solution for this 👋

When a user has a conversation with my bot, I create (or find an existing) session. In a session, I store "context" variables about the conversation/state which are also saved to my mongo database. I load all these active sessions into memory when restarting the app so I can access them locally.

Not sure if it's the best or the right way, but it works for me 😄

[lavaxun]

any plan to support multi-nodes and maintaining state in database?
going to try @dannyfreed solution at the moment ;)

[anglol]

Hi ! Something new about this point ?
@dannyfreed can you share your solution ?

Thanks !

[GautierT]

Hi !
@dannyfreed can you share your solution ? Thanks !

[adantoscano]

+1

[peterswimm]

merged this towards this more active thread https://github.com/howdyai/botkit/issues/506

[rsweetland]

@dannyfreed I was just about implement something similar. Is there a snippet or two you wouldn't mind sharing for where / how you captured and reinitialized the conversations within botkit? Is it just a matter of dumping / loading controller.tasks on every request / response? What other context are you managing?

[dannyfreed]

@rsweetland I didn't implement this using botkit, ended up building my own solution.

The gist of it was I have a sessions table in my database. I make a call to the database onReceivedMessage and search for the user's "session" based on their user id.

Each session row has a user id, and then a "context" which is essentially just a mutable object (key/value).

I update the state onSend and onReceive depending on the business logic.

Hope that helps!

[rsweetland]

Totally helps, thank you! 🙌

But one question (if you don't mind): How did you load the stored conversation / context back into botkit's memory? Just curious if you found a way to effectively serialize / deserialize this.

For example (wishful thinking / pseudo code):

  const message; //available from earlier
  const tasks = controller.tasks;
  for (let task of tasks) {
    for (let convo of task.convos) {
      const userid = convo.source_message.channel;
      const serializedConvo = await lookupStoredConversation(userid);
      const botkitConvo = magicalDeserialization(serializedConvo); // <-- ?
      const botkitConvo.handle(message);     
    }
  }
}

[dannyfreed]

It's been a while since I looked at any Botkit code. But if I remember correctly, I think you can add middleware that will do this. Basically in the middleware, which you'll want to call for each message you receive and/or send, you can make a call to your database and find the user context as I previously described.