perf: split up routers to separate lambdas

[juliusmarminge]

What does this PR do?

Splits up the all tRPC routers into their own API routes. (Edited by @keithwillcode)

New bundle sizes

The slots router is massive compared to all the rest endpoints together...

Old bundle size

Seems like the analyzer doesn't really compare apples to apples here, as it states the previous bundle size is tiny. Not sure what's going on there, does it not take into account the size of the appRouter since it's just joining in others?

But, given that the entire appRouter is imported which comes in at 334kB, that should give an idea of the actual size:

Fixes # (issue)

Environment: Staging(main branch) / Production

Type of change

• Bug fix (non-breaking change which fixes an issue)
• Chore (refactoring code, technical debt, workflow improvements)

How should this be tested?

• Test the entire app basically to ensure no tRPC routes have been broken.

[vercel]

The latest updates on your projects. Learn more about Vercel for Git ↗︎

Name	Status	Preview	Comments	Updated (UTC)
cal	✅ Ready (Inspect)	Visit Preview	💬 Add feedback	May 3, 2023 3:49pm
ui	✅ Ready (Inspect)	Visit Preview	💬 Add feedback	May 3, 2023 3:49pm
web	✅ Ready (Inspect)	Visit Preview	💬 Add feedback	May 3, 2023 3:49pm

[vercel]

@juliusmarminge is attempting to deploy a commit to the cal Team on Vercel.

A member of the Team first needs to authorize it.

[github-actions]

📦 Next.js Bundle Analysis for @calcom/web

This analysis was generated by the Next.js Bundle Analysis action. 🤖

This PR introduced no changes to the JavaScript bundle! 🙌

E2Es are fixed now. There were route assertions which changed.

[keithwillcode]

Documenting here for clarity. There was a separate test done in #8551 that verified when accessing the route /api/trpc/viewer.public.i18n under a different API route of /api/trpc/public/i18n, the time difference in cold start load was down to 2 seconds from 15+. When I realized the full path forward was to separate all routers into their own API routes so we avoid a compounding effect of loading lots of routers into 1 core router, I brought those changes back over here to build upon the work @juliusmarminge had started.

The effect this will have is if you look at the network traffic for cold starts on pages in the app, you will now see multiple tRPC calls take upwards of 3 seconds to load but since they run concurrently, the effect is not compounded, which was what resulted in such long load times.

[Udit-takkar]

LGTM. tested some core features locally

[roae]

Great job!! 🚀

Left a comment about how we can make sure this runs on separate lambdas. we can do it in other PR or add it here before we merge if you agree, your choice @keithwillcode

[roae]

to make sure this is deployed on different lambdas maybe we should have a slightly different configuration on each path as discussed in slack.

Here an example, this changes must be added to the vercel.json

cc @emrysal

b803c61

[keithwillcode]

The way it’s written now, since the api routes are separated into different folders, Vercel already creates separate functions for each one. We don’t have to include this config to ensure those are created distinctly. We can if we want to change config of course though.