Timeout with long lookup tables

Hi, I faced an issue after extending my lookup tables. The problem I’m facing is that I sometimes (90%) got a timeout error from async on the very first (text) message. However, if the initital messages gets processed correctly (but still very slow) the following messages get processed pretty much instantly.

I looked into the rasa code and found that the timeout is set to 10 secs, which coincidentally is the time it takes for the first message to process. So I guess it takes 10+ seconds in 90% of the cases and in 10% i get lucky and it finishes before.

When I raise the timelimit (to 100 seconds) it works every time (but ofc still around 10 secs on very first message).

So I guess the NLU component is setting up all the regex stuff on first run, which might take a lot of time. Of course an obvious fix would be to shorten the lookup tables (one is 270k and the other 50k) or extend the timeout limit, but Id like to know if theres another way or should this be considered a bug?

1 Like

Reported this as a bug @Github

Thanks for raising the bug, we’ll get back to you on github