Documenting learning on how to reduce the db queries on the server

[tordans]

The goal…

Query the DB for only the IDs of an object (or any other subset of columns) to optimize the data "cost".

Here are a few things I learned – or at least my current state of know how.

1. Tanstack query has a select option on the tanstack useQuery hook. That would give type save return types.
   • BUT, it would only work in the Frontend and we want to reduce the select on our DB level, not just the frontend after all data was transfered
   • BUT, we actually don't use tanstack query (more below) direclty
2. Passing select to the getProject query has a zod and a types component. Not clear how to best specify the zod part. Not clear how to remove the redundancy.
3. We actually need to research this stuff on the Prisma level. The prisma call is the one that is responsible to receive the select option and thus to specify the return type.
   • TODO maybe more research needed; the documented takes are all hacky/no nice

I looked into how Tanstack query would handle the return types. There is good docs like https://tkdodo.eu/blog/react-query-and-type-script#the-four-generics. BUT, my understanding is now that the blitz useQuery uses Tanstack useQuery internally but exposes its own hook under the same name with a completely different signature (WHY…). https://blitzjs.com/docs/query-usage — I asked in Discord and might PR a docs update to clarify.

Update: This is the blitz > tanstack wrapper code https://github.com/blitz-js/blitz/blob/main/packages/blitz-rpc/src/data-client/react-query.tsx

We can optimize on different layer

1. the db query – this would be optimal; minimal data queried and transferred
2. what we return from the server function – this would be good, the db is not optimized but the client receives minimal data
3. what we return from useQuery in our frontend; this is what Tanstack query would do (if we could use it). However, that is not really a optimization since the data "cost" is already paid.

Conclusion

Right now, it looks like we either duplicate the getProject file and create specific files per query, like getProjectId.
Or, we just pay the data "cost" and optimize later…