Does anyone know? It feels that way from fetching sub fields it adds to latency.
Can it be done in parallel?
#Are ResolveField fetched sequentially?
soft walrus
lunar moat
Node.js is single-threaded, so you really can't do anything in parallel. You could implement DataLoader and fetch multiple results at once.
soft walrus
but async allows to do concurrently
I have a resolver where if I query sub objects latency just adds up
and we use prisma it does what dataloader do
lunar moat
Well, concurrently still means they add up: https://tsh.io/blog/simple-guide-concurrency-node-js/
soft walrus
that's the point, it's sync instead of async. If I add a delay function in a ResolveField. It waits for that ResolveField to be done to call the other ResolveFields
soft walrus
if I add a delay to a resolveField of 5 seconds to a query that takes 1 second the graphql takes 6 seconds instead of around 5 seconds
lunar moat
Oh, I see. Have you tried with vanilla apollo/mercurius server? I'm don't have deep knowledge of how Nest integrates the drivers. It would be good to know whether it's Nest issue, or the problem comes from the driver (or even graphql.js).
If I remember correctly, the graphql specs don't guarantee that the field are resolved in the requested order, so it probably should work like you expect.
soft walrus
I'm doing this with Insomnia to send graphql queries
lunar moat
if I add a delay to a resolveField of 5 seconds to a query that takes 1 second t...
This sounds correct though. The top level query has to resolve first, before the fields on that object can begin resolving.
If you have:
- query that takes 1s
- field A resolver that takes 2s
- field B resolver that takes 2s
it should be around 3s, at least that's what I'd expect.
soft walrus
In this example I have a model where we query properties +- 200 ms and adding 5 sub models looks like additive as latency increase for each model added
maiden quiver
https://youtu.be/cCOL7MC4Pl0?si=QrqY0-5W7qjJ_X6o this is a good video on how the event loop works
oh wait that is for the browser
maybe this will cut it then
it is really mind bending sometimes on what is when processed and why
difference in microtask and promise queue and such