liftIO but need smth like liftEff[F[_]: Concurrent](eff: F[T]): Fetch[F, T] as not everyone using IO, as making new DataSources for every case is quite annoying and instantiating new class on every call with generic isn't a good performance practice it seems, lmk if I should open an issue
Hi, I have an API that provides a method get(key: String): Future[Thing]. In my implementation, I would like to transparently batch requests within a certain window. As I understand, fetch is not made to do this.
Is my understanding correct? If so, is there an extension point to make it do this? Or do you know an alternative library that would help with such a scenario?
Hi, I have an API that provides a method
get(key: String): Future[Thing]. In my implementation, I would like to transparently batch requests within a certain window. As I understand, fetch is not made to do this.Is my understanding correct? If so, is there an extension point to make it do this? Or do you know an alternative library that would help with such a scenario?
you mean time window @rethab ?
@purrgrammer ty for liftF!
I have another case which might be quite interesting related to tracing within Fetch
Context
My underlying Concurrent type is TraceIO[A] (type TraceIO[A] = TraceT[IO, A]) which I'm using for tracing, so in FetchTIO (type FetchTIO = Fetch[TraceIO, A])
I wonder how to properly generate new spans (TraceT has method def newSpan(...) which returns TraceT with updated fiber context) when composing, combining, mapping over fetches
Current attempts
I tried using helper
def withTrace[A](ftio: FetchTIO[A], spanInfo: SpanInfo) : FetchTIO[A] = {
Fetch.liftF(Fetch.run(methodTIO(...)).newSpan(spanInfo))
}but it breaks Fetch optimizations even though does create all spans as I want
Questions/Suggestions
1) Is there anything that can be done to allow this kind of behavior? Maybe smth like
withUpdateF[F[_]: Concurrent, A](f: Fetch[F, A])(upd: F[A] => F[A]): Fetch[F, A]method? or even more general transformer
withUpdateF[F[_]: Concurrent, G[_]: Concurrent, A](f: Fetch[F, A])(upd: F[A] => G[A]): Fetch[G, A]?
2) If above isn't possible and/or there's a better alternative wonder what it could be? Can maybe FetchLog be used somehow to collect execution path info?
FetchLog stores a log of a Fetch execution, and is a public data structure that users of the library have access to. You can run a fetch recording its execution using `Fetch#runLog, see http://47deg.github.io/fetch/docs.html#debugging-7 for more details.
@purrgrammer
1) yes, I do that as well, but it only gets my leaves of execution tree, I'd like to record branches as well as it is one of the reasons of having local tracing in the first place.
Like if my Fetch.run(returnUser()) 10 different method calls away from fetch db calls it's still baffling how we got to these db calls.
And these 10 methods along the way also return Fetch like
def branchMethod(...): Fetch[..., ...] = for {
(resA, resB) <- (leafMethodA, leafMethodB).tupled
resC <- leafMethodC
} yield {
doSmthWithABC(resA, resB, resC)
}^ I'd be very nice to get an aggregated span for this guy (even if we managed to get spans for leaves)
does this make sense?
2) yeah, I'm looking right now of how to use/manipulate Rounds to get data I can use once Fetch part is done to build spans from the resulted Log
Sangria to use Fetch for a couple weeks and keep getting the implementation working. Wondering if anyone else has tried at this? Sangria already supports a concept of Deferreds which are very similar to the Fetch but without all the magic that cats grants and I'm trying to make the Sangria resolvers accept a Fetch natively.
I had a question about these lines in the documentation:
Fetch provides its own instance of Applicative[Fetch]. Whenever we use applicative operations on more than one Fetch, we know that the fetches are independent meaning we can perform optimizations such as batching and concurrent requests.
If we were to use the default Applicative[Fetch] operations, which are implemented in terms of flatMap, we wouldn’t have information about the independency of multiple fetches.
I had a conversion yesterday where someone pointed me to parTraverse and https://typelevel.org/cats/typeclasses/parallel.html?.
In light of all of that, I'd expect Fetch to have, Monadand Applicative instances with a Parallel instance mapping between them, and one would need parTraverse in order to get batching from the monad.
But it seems like there's really only a Monad instance. So at the very least is seems like either the documentation is misleading or I can't find the instance of Applicative[Fetch] that it mentions.
So I was wondering on whether Fetch had the capability to batch across different requests, turns out it currently doesn't and I've gone and implemented it for my uses.
As an example of why this could be useful, say you have a CRUD HTTP endpoint where the user can only obtain one data element at a time. The implementation could then just focus on the individual request while behind the scenes Fetch does batching of the requests coming in for that DataSource across all the Fetches taking place simultaneously. I have so far an implementation done in Cats Effect 3 that does this by wrapping the original DataSource and letting the developer choose the allowed delay for the request. The API I have right now is the following signature:
def batchAcrossFetches[F[_], I, A](current: DataSource[F, I, A], delayPerBatch: FiniteDuration)(
implicit F: Async[F]
): Resource[F, DataSource[F, I, A]]It's a Resource due to having to setup a background Fiber that groups requests together in a given window and using a CE 3 Queue for processing incoming fetch requests
I'm posting this here to gauge interest, as that could be a potential feature PR for the library, I just don't know if there's interest in it or where to put it, so far I've stored this in the companion object of DataSource
runAll doesn't have all the data in it when mixing DataSources. I've created a gist with the issue, just re-using the Users and Posts DataSources from the examples. When running a Fetch for 2 users they both end up in the cache. When running e.g. a Fetch for one user and one Post only the User ends up in the cache. https://gist.github.com/byarr/1491b0715dac8cc1f222f6e6fc7f852b