Show HN: Rill – Composable concurrency toolkit for Go

by destel • 2 years ago

208 122 2 years ago

destel18• 2 years ago

Hi everyone. Posting on HN for the first time. I'd like to share Rill - a toolkit for composable channel-based concurrency, that makes it easy to build concurrent programs from simple, reusable parts

Example of what it looks like:

    // Convert a slice into a channel
    ids := rill.FromSlice([]int{1, 2, 3, 4, 5, 6, 7, 8, 9, 10}, nil)


    // Read users from API with concurrency=3
    users := rill.Map(ids, 3, func(id int) (*User, error) {
        return api.GetUser(ctx, id)
    })

    // Process users with concurrency=2
    err := rill.ForEach(users, 2, func(u *User) error {
        if !u.IsActive {
            u.IsActive = true
            return api.SaveUser(ctx, u)
        }
        return nil
    })

    // Handle errors
    fmt.Println("Error:", err)

Key features:

  - Makes concurrent code composable and clean
  - Works for both simple cases and complex pipelines
  - Built-in batching support
  - Order preservation when needed
  - Centralized error handling
  - Zero dependencies

The library grew from solving real concurrency challenges in production. Happy to discuss the design decisions or answer any questions.

fillskills• 2 years ago

Very intuitive API. Thanks!

limit499karma• 2 years ago

Is there an underlying assumption that the channels are containers and not streams?

destel18• 2 years ago

No, it's the opposite - the library treats channels as streams, processing items as they arrive without needing to know the total size in advance. This is why it can handle infinite streams and large datasets that don't fit in memory.

slantedview• 2 years ago

I'm curious - what other technologies/libraries/APIs, including in other languages, did you draw on for inspiration, or would you say are similar to Rill?

destel18• 2 years ago

The short answer would be: I kept writing a code that spawns goroutines, that read from a channel, do some processing and write results to another channel. Add some wait/err groups to this and we'll get a lot of boilerplate repeated all over the place. I viewed this as "channel transformations" and wanted to abstract it away. When generics came out it became technically possible.

Surprisingly, part of my inspiration came from scala (which I haven't touched since 2014). Back then Scala had transformable streams and the "Try" typethen.

akshayshah2• 2 years ago

The channel-focused approach to stream processing reminds me of Heka [0]. It was a contemporary of Samza and Heron, and it was fairly prominent in the early Go ecosystem (maybe 10 years ago). As I recall it, quite foggily and quite a long while later, one of the final nails in Heka's coffin was that channel throughput didn't scale well. Do you have benchmarks for Rill, or is it not intended for high-throughput use cases?

[0] https://github.com/mozilla-services/heka

minus7• 2 years ago

Hey,

I found your library a few weeks ago when I was annoyed by nothing like this being built into the standard library. It’s been a breeze to use so far.

A neat trick I found to gauge bottlenecks in pipelines is using Buffers between steps and running a goroutine that periodically prints `len(buffered)/cap(buffered)`

destel18• 2 years ago

Thank you very much for the feedback. I thought about something similar some time ago. Buffer of size one, then measure the average time each item spends in the buffer. But for debugging your approach is simpler and more practical.

hu3• 2 years ago

Hi! Looks great. I might use this to fan out/in my RSS reader HTTP calls.

How would I implement timeout? In case a HTTP call takes too long?

lsaferite• 2 years ago

Based on the examples and documentation, rill doesn't manage context for you. You'd simply set the client timeout or give each http call a timeout context.

mariusor• 2 years ago

You might be interested by something that has been designed specifically for this problem. I created a state machine library for Go on top of which I mapped retry[1] and some other patterns. And funnily enough one of the first applications I implemented is an RSS reader[2]

[1] https://pkg.go.dev/git.sr.ht/~mariusor/ssm#example-Retry

[2] https://git.sr.ht/~mariusor/frankenjack/tree/master/item/sou...

destel18• 2 years ago

For now, the library is context-agnostic by design. For HTTP timeouts, you'd use Go's standard approaches: either set the HTTP client timeout or pass a context with timeout to each request. Please let me know more about your use case - I'll let you know if Rill isn't a good fit.

lspears• 2 years ago

This is great. I am working on a robotics application and this seems like a better abstraction than alternatives such local messaging servers. How do you deal with something like back pressure or not keeping up with incoming data?

destel18• 2 years ago

The lib is based on channels and inherits the channel behavior in terms of backpressure. Simply put if no-one reads on one side of the pipeline, it wouldn't be possible to write anything on the other side. Still, it's possible to add buffering at arbitrary points in the pipeline using the rill.Buffer function.

lyxell• 2 years ago

The API looks really nice and intuitive! What motivated you to build this?

destel18• 2 years ago

Thank you! There are two pieces of motivation here. The first one is removing boilerplate of spawning goroutines that read from one channel and write to another. Such code also needs wait/err group to properly close the output channel. I wanted to abstract this away as "channel transformation" with user controlled concurrency.

Another part is to provide solutions for some real problems I had. Most notably batching, error handling (in multi stage pipelines) and order preservation. I thought that they are generic enough to be the part of general purpose library.

c4pt0r• 2 years ago

Very handy!

noctane2• 2 years ago

It looks great. What are other existing tools? And how do they compare to them?

Scaevolus• 2 years ago

Sourcegraph Conc is broadly similar in providing pool helpers, but doesn't provide the same fine grained batching options: https://github.com/sourcegraph/conc

Uber CFF does code generation, and has more of a focus on readability and complex dependency chains: https://github.com/uber-go/cff

noctane2• 2 years ago

Thanks for sharing!

qaq• 2 years ago

any plans to add context support ?

destel18• 2 years ago

I am thinking on it. To be honest, the current design works fine for my use cases: simply put, the function that defines a pipeline should have context.WithCancel() and defer cancel() calls.

I need a feedback on this. What kind of builtin context support would work for you? Do you need something like errgroup's ability to automatically cancel the context on first error?

destel18• 2 years ago

I've just pushed few small changes to the readme that better explain context usage

tschellenbach• 2 years ago

this is a real problem in go, very easy to have bugs when working with channels and the way it handles errors etc.

latchkey19• 2 years ago

If you write comprehensive unit tests, it is not easy to have bugs in golang. Especially as things change over time. A library like this isn't going to protect you from having bugs.

TIL: HN doesn't like writing tests. The downvotes on this are hilarious. "Job security" ¯\_(ツ)_/¯.

bogota• 2 years ago

Hard disagree on this. Large production apps that use channels have very subtle bugs that cause all kinds of annoying issues that only come up under load in prod. I have been using go for ten years and still pick it as my language of choice for most projects however I stay away from channels and especially any complex use of them unless it’s 100% required to scale the application. Even then you can most of the time come up with a better solution by re architecting that part of the application. For pet projects go crazy with them though.

tmoertel7• 2 years ago

I think you're getting downvoted for the unsupported assertion that "If you write comprehensive unit tests, it is not easy to have bugs in golang." Probably because you made that assertion in the context of a discussion of channels, widely believed to have underlying concurrency semantics that are subtle and easy to misunderstand, making "write comprehensive unit tests" seem like a strategy that's apt to let real-world problems slip through (because a programmer's belief that their tests are "comprehensive" is likely to be mistaken).

hnlmorg7• 2 years ago

You’re getting downvoted because you’re essentially arguing that a language abstraction which is a known source of bugs can be solved simply by writing better code.

which misses the point of the OP.

dboreham2• 2 years ago

This is exactly the kind of bug that unit tests are poor at exposing.

jbendotnet21• 2 years ago

Looks good, similar to https://github.com/sourcegraph/conc which we've been using for a while. Will give this a look.

alpb• 2 years ago

There are also libraries like https://github.com/Jeffail/tunny or https://pkg.go.dev/go.uber.org/goleak or https://github.com/fatih/semgroup to help deal with concurrency limits and goroutine lifecycle management.

As the author of https://github.com/ahmetb/go-linq, it's hard to find adoption for libraries offering "syntactic sugar" in Go, as the language culture discourages those kind of abstractions and keeping the code straightforward.

izolate• 2 years ago

The batching concept is a cool idea and could be useful in the right context. That said, this feels like a JavaScript engineer's take on Go. Abstractions like Map and ForEach don't align with Go's emphasis on simplicity and explicitness. The lack of context.Context handling also seems like an oversight, especially when considering concurrency.

Judging by the praise, I'm probably in the minority, but as a code reviewer, I’d much rather see straightforward loops, channels, and Go's native constructs over something like Rill.

ARandomerDude2• 2 years ago

Map and ForEach don't align with Go's emphasis on simplicity and explicitness

I've never paid my bills with Go, but `Map` and `ForEach` don't seem all that different than `for _, u := range Users` to me. Yes, the former is "functional" but only mildly.

prisenco• 2 years ago

In that case there's no particular reason to use them. As far as Go's philosophy goes.

zendist• 2 years ago

If you were to build a library like `rill` in the Go-way, what would your Batch API usage look like?

hnlmorg7• 2 years ago

I don’t agree with your comment about Map and ForEach, just by virtue of the fact that sync.Map exists in Go’s standard library.

But your point about the lack of contexts is definitely a deal breaker for me personally too.

akshayshah2• 2 years ago

The "map" under discussion here is very different from sync.Map. The discussion here is focused on the "map" primitive from functional programming - transforming a collection by applying a function to each element.

sync.Map is a concurrency-safe hash map. Same name, totally different thing.

purpleidea2• 2 years ago

Nice! I do a lot of concurrency work with DAG's in https://github.com/purpleidea/mgmt/ and I would love to swap out some of those concurrency runners with a lib if possible.

I was wondering if this could be it... Any thoughts in that direction, please let me know!

destel18• 2 years ago

Thank you! I took a quick look at mgmt, it's quite a large and complex project. I'd need to better understand your DAG-based concurrency patterns to say if Rill would be a good fit. Could you share some examples of the concurrent runners you're thinking about? This would help me understand if/how Rill might be useful for your case.

purpleidea2• 2 years ago

So we have two concurrency "engines" in mgmt:

(1) One for running the function graph, and (2) one for running the resource graph.

(1) https://github.com/purpleidea/mgmt/tree/master/lang/funcs/da... (2) https://github.com/purpleidea/mgmt/tree/master/engine/graph

TBQH, I think I'm decent at concurrency, but I never considered it my passion or expertise and while both of those _work_ I do know that I have some concurrency bugs hanging out, and I'd love real professional help here.

In the odd chance you'd be willing to hack on things with me, I'd be particularly elated. There might be a possible symbiosis!

I'm @purpleidea on Mastodon, Matrix (where we have an #mgmtconfig room) and a few other places in case you'd like to chat more!

fidotron• 2 years ago

I think it is time to face the fact CSP style channels are a bad idea if you don’t also have the occam semantics for channel scope. (I know there are other people here that understood that sentence).

The problem in golang is the channel cleanup, particularly. is a mess. In occam they come and go simply by existing as variables in seq or par blocks. Occam is very static though, so the equivalent to goroutines are all allocated at build. (Occam pi attempted to resolve this iirc).

https://en.m.wikipedia.org/wiki/Occam_(programming_language)

Some of the patterns in this library are reminiscent of constructs occam has as basic constructs, such as the for each block, although the occam one must have the number of blocks known at build time.

The fact so many people in golang reach for mutexes constantly is a sign things are not all well there.

0x696C69616• 2 years ago

Channels are just another synchronization primitive in your toolbox. They do make some things much simpler, but there's no reason to reach for one if a mutex does the job.

The usage of mutexes doesn't make channels "bad" for the same reason that usage of atomics doesn't make mutexes bad.

dboreham2• 2 years ago

+1 for understanding the sentence.

0x696C69616• 2 years ago

I think that using iterators in the public API would have been better than channels.

destel18• 2 years ago

Rill might look like it tries to be a replacement for iterators, but it's not the case. It's a concurrency library, that's why it's based on channels

0x696C69616• 2 years ago

I disagree that using channels is necessary for concurrency. Consider the following iterator based signature for your Map function:

    func Map[A, B any](in iter.Seq[Try[A]], n int, f func(A) (B, error)) iter.Seq[Try[B]]

dangoodmanUT• 2 years ago

Love the idea, some weirdness though:

Here's a practical example: finding the first occurrence of a specific string among 1000 large files hosted online. Downloading all files at once would consume too much memory, processing them sequentially would be too slow, and traditional concurrency patterns do not preserve the order of files, making it challenging to find the first match.

But this example will process ALL items, it won't break when a batch of 5 finds something?

destel18• 2 years ago

It will. Otherwise the example wouldn't make sense. There's one important detail I haven't clarified enough in that part of the readme.

For proper pipeline termination the context has to be cancelled. So it should have been be Like:

func main() { ctx, cancel := context.WithCancel(context.Background()) defer cancel()

    urls := rill.Generate(func(send func(string), sendErr func(error)) {
        for i := 0; i < 1000 && ctx.Err() == nil; i++ {
                send(fmt.Sprintf("https://example.com/file-%d.txt", i))
        }
    })
    ...

One of the reasons I've ommited context cancellation in this and some other examples is because everything's happening inside the main function. I'll probably add cancellations to avoid confusion.

destel18• 2 years ago

I've also just pushed few small changes to the readme that clarify this things.

fatih-erikli-cg3• 2 years ago

:= for assinging a variable sounds and looks weird to me

steve_adams_864• 2 years ago

It's the walrus operator. Pascal and Python use it as well. You get used to it pretty quickly.

Pascal: https://www.freepascal.org/docs-html/ref/refse104.html#x224-... Python: https://docs.python.org/3/whatsnew/3.8.html#assignment-expre...

fatih-erikli-cg3• 2 years ago

It was something introduced like a 1 april joke. Clean implementation of any programming language won't have that.

indulona• 2 years ago

it's just a shorthand for var foo int = 5 vs foo := 5 where the type is derived from the assigned value.

linhns2• 2 years ago

I think it's good. 2 characters for declaration + assign.

pajeetz• 2 years ago

what sort of environment do you need to be in to have to compose concurrency like this instead of relying on native go's scaling?

dougbarrett• 2 years ago

Batching is a pattern I’ve had to manually build in the past to push large amounts of analytic data to a database. I’d push individual events to be logged, map reduce those in batches and then perform insert on duplicate update queries on the database, otherwise the threshold of incoming events was enough to saturate the connection pool making the app inoperable.

Even optimizing to where if an app instance new it ran the inert on update for a specific unique index by storing that in a hash map and only running updates from there on out to increase the count of occurrences of that event was enough to find significant performance gains as well.

jerf2• 2 years ago

The same sort of environment in which one uses such abstractions like "functions" instead of relying on the language's native ability to run sequential instructions.

It's generally good for languages to provide relatively low-level functionality and let libraries be able to build on top of it, because as the programming language development world has now learned many times over, the hardest code to change is the code in the language and its standard library. It isn't the job of the language itself to provide every possible useful iteration on the base primitives it provides.

gregwebs2• 2 years ago

Thanks for sharing what is working for you in production. I made a somewhat similar library that also does batching (docs are sparse, but I am updating docs on my libraries this week) [1].

I would call this parallelism rather than concurrency.

The main issue I have with this library's implementation is how errors are handled. Errors are retrieved rather than assigned- but assignment is preferable because it gets verified by tools. In my library I used a channel for errors- that gives ultimate flexiblitiy- it can be converted to wait and collect them to a slice or to perform a cancellation on first error.

[1] https://github.com/gregwebs/go-parallel

destel18• 2 years ago

Thank you for the feedback. My design decision is of course a tradeoff. When multiple channels are exposed to the users (not encapsulated inside the lib), this forces them to use "select". And this is very error prone in my experience

gregwebs2• 2 years ago

I never needed to use select on an error channel for my use cases because at the point I operate on the error channel I want to block for completion. And I provide helpers for the desired behavior for the channel so I don't even directly receive from it. I see that some of Rill is designed to operate on continuous streams, and in that light the design decision makes sense. For my use cases though the stream always had an end.

Groxx• 2 years ago

Hmmm. Some stuff to like, but I do feel like this should have a big, noticeable cautionary note that it does not wait if you end early (e.g. via Err). Any pending actions continue running and draining in background goroutines, and are potentially VERY delayed if internal/core.Delay is ever exposed or your funcs sleep.

I've seen that kind of pattern lead to TONS of surprise race conditions in user-code, because everyone always thinks that "it returned" means "it's done". Which is reasonable, nothing else is measurable on the caller side - changing that won't be noticeable by callers and may violate their expectations and cause crashes.

destel18• 2 years ago

Thank you for the feedback. I agree with your point.

The current solution is to make pipeline stages context-aware (which is often happens automatically) and cancel the context before returning. This is the responsibility of the user and can lead to problems you described.

I haven't yet found a better solution to this. On the other hand, exact same thing happens if your function spawns a goroutine and returns. That goroutine would run until done, unless it's context aware.

Regarding the "Delay" and "infiniteBuffer" functions - these are part of the work on adding support for feedback loops to Rill. I haven't yet found a reliable and user friendly way to do it, so this work is on hold for now.

icar• 2 years ago

This reminds me of rxjs.

pezo1919• 2 years ago

How does it compare to RxGo?

destel18• 2 years ago

I have to be honest - I haven't ever heard about it. Just checked and found it's very mature and popular, though it seems to have had no commits in the last 2 years.

After quick scanning, I'd highlight these 2 differences: - Rill is type safe and uses generics - Rill does not try to hide channels, it just provides primitives to transform them

kermatt• 2 years ago

Name overlap with another Go based package: https://github.com/rilldata/rill

taffit• 2 years ago

This also came to my mind when I heard `rill`, coming from https://www.rilldata.com/ .

linhns2• 2 years ago

Looks super neat! Just a small question: Why did you chose the name Try instead of Result?

destel18• 2 years ago

Thank you. I "stealed" the name from scala. They have the similar value+error type. Maybe in context it rill the better name could have been "Item"

Show HN: Rill – Composable concurrency toolkit for Go

Related Stories