feat: rate limiting

[Jaszkowic]

let's discuss what values to use for rate limiting. we want to avoid that the Parla API can be used to send too many requests to OpenAI.

[render]

Your Render PR Server URL is https://ki-anfragen-api-pr-59.onrender.com.

Follow its progress at https://dashboard.render.com/web/srv-cmju8amv3ddc739hh5rg.

[Jaszkowic]

This plugin will add an onRequest hook to check if a client (based on their IP address) has made too many requests in the given timeWindow.

https://github.com/fastify/fastify-rate-limit

[Jaszkowic]

Why 30 requests / min / IP ?

• 2 request per page load (document count)
• 2 requests per question (document_search and generate_answer)
• total 4 requests per question
• answer takes ~20s to load

in total: (2+2)*3 = 16 requests / min

adding some free bonus requests to account for unusual behavior:

total 30 requests / min should be enough

[ff6347]

30/min is already under attack. Maybe 10?

[ff6347]

30 Sounds good. 👌🏾