Streamly

Streaming Concurrently

Streamly is a monad transformer unifying non-determinism (list-t/logict), concurrency (async), streaming (conduit/pipes), and FRP (Yampa/reflex) functionality in a concise and intuitive API. High level concurrency makes concurrent applications almost indistinguishable from non-concurrent ones. By changing a single combinator you can control whether the code runs serially or concurrently. It naturally integrates concurrency with streaming rather than adding it as an afterthought. Moreover, it interworks with the popular streaming libraries.

See the haddock documentation for full reference. It is recommended that you read Streamly.Tutorial first. Also see Streamly.Examples for some working examples.

Streamly has best in class performance even though it generalizes streaming to concurrent composition that does not mean it sacrifices non-concurrent performance. See streaming-benchmarks for detailed performance comparison with regular streaming libraries and the explanation of the benchmarks. The following graphs show a summary, the first one measures how four pipeline stages in a series perform, the second one measures the performance of individual stream operations; in both cases the stream processes a million elements:

Non-determinism

The monad instance composes like a list monad.

import Streamly
import qualified Streamly.Prelude as S

loops = do
    x <- S.each [1,2]
    y <- S.each [3,4]
    liftIO $ putStrLn $ show (x, y)

main = runStreaming $ serially $ loops

(1,3)
(1,4)
(2,3)
(2,4)

Magical Concurrency

To run the above code with demand-driven concurrency i.e. each iteration in the loops can run concurrently depending on the consumer rate:

main = runStreaming $ asyncly $ loops

To run it with full parallelism irrespective of demand:

main = runStreaming $ parallely $ loops

To run it serially but interleaving the outer and inner loop iterations:

main = runStreaming $ interleaving $ loops

You can fold multiple streams or IO actions using parallel combinators like <|, <|>. For example, to concurrently generate the squares and then concurrently sum the square roots of all combinations:

import Streamly
import qualified Streamly.Prelude as S

main = do
    s <- S.sum $ asyncly $ do
        -- Squaring is concurrent (<|)
        x2 <- forEachWith (<|) [1..100] $ \x -> return $ x * x
        y2 <- forEachWith (<|) [1..100] $ \y -> return $ y * y
        -- sqrt is concurrent (asyncly)
        return $ sqrt (x2 + y2)
    print s

Of course, the actions running in parallel could be arbitrary IO actions. To concurrently list the contents of a directory tree recursively:

import Path.IO (listDir, getCurrentDir)
import Streamly

main = runStreaming $ serially $ getCurrentDir >>= readdir
   where readdir d = do
            (dirs, files) <- lift $ listDir d
            liftIO $ mapM_ putStrLn $ map show files
            -- read the subdirs concurrently
            foldMapWith (<|>) readdir dirs

In the above examples we do not think in terms of threads, locking or synchronization, rather we think in terms of what can run in parallel, the rest is taken care of automatically. With asyncly and <| the programmer does not have to worry about how many threads are to be created they are automatically adjusted based on the demand of the consumer.

The concurrency facilities provided by streamly can be compared with OpenMP and Cilk but with a more declarative expression. Concurrency support does not compromise performance in non-concurrent cases, the performance of the library is at par or better than most of the existing streaming libraries.

Streaming

Streaming is effortless, simple and straightforward. Streamly data type behaves just like a list and combinators are provided in Streamly.Prelude to transform or fold streamly streams. Unlike other libraries and like streaming library the combinators explicitly consume a stream and produce a stream, therefore, no special operator is needed to join stream stages, just a forward ($) or reverse (&) function application operator is enough.

import Streamly
import qualified Streamly.Prelude as S
import Data.Function ((&))

main = S.each [1..10]
     & fmap (+ 1)
     & S.drop 2
     & S.filter even
     & fmap (* 3)
     & S.takeWhile (< 25)
     & S.mapM (\x -> putStrLn ("saw " ++ show x) >> return x)
     & S.toList . serially
     >>= print

Fold style combinators can be used to fold purely or monadically. You can also use the beautiful foldl library for folding.

main = S.each [1..10]
     & serially
     & S.foldl (+) 0 id
     >>= print

Streams can be combined together in multiple ways:

main = do
    let p s = (toList . serially) s >>= print
    p $ return 1 <> return 2               -- serial, combine atoms
    p $ S.each [1..10] <> S.each [11..20]  -- serial
    p $ S.each [1..10] <| S.each [11..20]  -- demand driven parallel
    p $ S.each [1..10] <=> S.each [11..20] -- serial but interleaved
    p $ S.each [1..10] <|> S.each [11..20] -- fully parallel

As we have already seen streams can be combined using monadic composition in a non-deterministic manner. This allows arbitrary manipulation and combining of streams. See Streamly.Examples.MergeSortedStreams for a more complicated example.

Reactive Programming (FRP)

Streamly is a foundation for first class reactive programming as well by virtue of integrating concurrency and streaming. See Streamly.Examples.AcidRainGame and Streamly.Examples.CirclingSquare for an SDL based animation example.

Contributing

The code is available under BSD-3 license on github. Join the gitter chat channel for discussions. You can find some of the todo items on the github wiki. Please ask on the gitter channel or contact the maintainer directly for more details on each item. All contributions are welcome!

This library was originally inspired by the transient package authored by Alberto G. Corona.