A while ago there was a blog (https://blog.softwaremill.com/correlation-ids-in-scala-using-monix-3aa11783db81) about using
ThreadLocal for storing a correlationId. This is useful for scoping log entries (via MDC). This was however bound to Monix, is there such a thing for cats-effect IO ?
I know there is http4s-tracer as well, but I think it’s a bit heavy solution
because it's not actually using
ThreadLocal, but TaskLocal: there are (possibly many) more Task/IO running on one thread
it should be possible to do it via other means though (e.g. Kleisli)
Hmm but introducing
Kleisli would have a penalty in terms of performance right?
but so will introducing
TaskLocal (although not as big I will imagine)
two things:
- actually check that the performance penalty matters for your use case: a couple of network calls and it stops being that important
- you can try something akin to passing the correlation
Idimplicitly but putting it behind aCorrelation[F]algebra
Wonder if I could decorate
ContextShiftto propagate the right MDC context as seen in http://code.hootsuite.com/logging-contextual-info-in-an-asynchronous-scala-application/
@Fristi in Monix you can use a TracingScheduler, out of which you can build the proper ContextShift, for IO too, however it gets tricky — and the reason for why the logic for Monix's TaskLocal resides in the Task run-loop itself.
For instance all it takes to screw your locals is something like this and your special ContextSwitch won't help 😉
IO.async { cb =>
new Thread(() -> cb(Right(()))).start()
}
Monix's
Task protects against this of course, by restoring the original locals after such async tasks are done executing.
For simple use-cases, you might get away with it though.
Note that you can easily build a
ContextSwitch out of a Monix TrackingScheduler and you can use the low-level Local in monix-execution, so you could do something with IO, but you have to keep in mind the above gotcha.
But I like using
ApplicativeAsk for something like that (from cats-mtl)
Unrelated question, I've been wondering something for two days - is there anything wrong with this?
def defaultContextShift[F[_]: Async](implicit ec: ExecutionContext): ContextShift[F] = new ContextShift[F] {
def shift: F[Unit] = Async.shift(ec)
def evalOn[A](ec: ExecutionContext)(fa: F[A]): F[A] =
Async[F].bracket(Async.shift(ec))(_ => fa)(_ => shift)
}
So no, I wouldn't use
defaultContextShift anywhere. If you need ContextShift[F], then get it as a function parameter.
So I have to implement
ContextShift myself ofc and there is no autoshiftinh
Argh on phone sorry about the typos
(curiosity)
You can't observe cancellation
Is there some initiaive to steal https://github.com/scalaz/scalaz-zio/blob/master/core/shared/src/main/scala/scalaz/zio/RefM.scala ?
I would do that if you like
@Odomontois
RefM would probably be cool, but I see that it needs a Queue too and that may be out of scope, queues are complicated.
Also if you do anything, it will need proper attribution.