Module documentation for 0.7.0
Haskell lists express pure computations using composable stream operations like
fold. Streamly is exactly like
lists except that it can express sequences of pure as well as monadic
computations aka streams. More importantly, it can express monadic sequences
with concurrent execution semantics without introducing any additional APIs.
Streamly expresses concurrency using standard, well known abstractions. Concurrency semantics are defined for list operations, semigroup, applicative and monadic compositions. Programmer does not need to know any low level notions of concurrency like threads, locking or synchronization. Concurrent and non-concurrent programs are fundamentally the same. A chosen segment of the program can be made concurrent by annotating it with an appropriate combinator. We can choose a combinator for lookahead style or asynchronous concurrency. Concurrency is automatically scaled up or down based on the demand from the consumer application, we can finally say goodbye to managing thread pools and associated sizing issues. The result is truly fearless and declarative monadic concurrency.
Where to use streamly?
Streamly is a general purpose programming framework. It can be used equally
efficiently from a simple
Hello World! program to a massively concurrent
application. The answer to the question, “where to use streamly?” - would be
similar to the answer to - “Where to use Haskell lists or the IO monad?”.
Streamly simplifies streaming and makes it as intuitive as plain lists. Unlike
other streaming libraries, no fancy types are required. Streamly is simply a
generalization of Haskell lists to monadic streaming optionally with concurrent
composition. The basic stream type in streamly
SerialT m a can be considered
as a list type
[a] parameterized by the monad
m. For example,
SerialT IO a is a moral equivalent of
[a] in the IO monad.
SerialT Identity a, is
equivalent to pure lists. Streams are constructed very much like lists, except
that they use
cons instead of
:. Unlike lists, streams
can be constructed from monadic effects, not just pure elements. Streams are
processed just like lists, with list like combinators, except that they are
monadic and work in a streaming fashion. In other words streamly just completes
what lists lack, you do not need to learn anything new. Please see streamly vs
lists for a detailed comparison.
Not surprisingly, the monad instance of streamly is a list transformer, with concurrency capability.
Why data flow programming?
If you need some convincing for using streaming or data flow programming paradigm itself then try to answer this question - why do we use lists in Haskell? It boils down to why we use functional programming in the first place. Haskell is successful in enforcing the functional data flow paradigm for pure computations using lists, but not for monadic computations. In the absence of a standard and easy to use data flow programming paradigm for monadic computations, and the IO monad providing an escape hatch to an imperative model, we just love to fall into the imperative trap, and start asking the same fundamental question again - why do we have to use the streaming data model?
High performance and simplicity are the two primary goals of streamly.
Streamly employs two different stream representations (CPS and direct style)
and interconverts between the two to get the best of both worlds on different
operations. It uses both foldr/build (for CPS style) and stream fusion (for
direct style) techniques to fuse operations. In terms of performance,
Streamly’s goal is to compete with equivalent C programs. Streamly redefines
“blazing fast” for streaming libraries, it competes with lists and
Other streaming libraries like “streaming”, “pipes” and “conduit” are orders of
magnitude slower on most microbenchmarks. See streaming
benchmarks for detailed
The following chart shows a comparison of those streamly and list operations where performance of the two differs by more than 10%. Positive y-axis displays how many times worse is a list operation compared to the same streamly operation, negative y-axis shows where streamly is worse compared to lists.
Streamly uses lock-free synchronization for concurrent operations. It employs
auto-scaling of the degree of concurrency based on demand. For CPU bound tasks
it tries to keep the threads close to the number of CPUs available whereas for
IO bound tasks more threads can be utilized. Parallelism can be utilized with
little overhead even if the task size is very small. See concurrency
benchmarks for detailed
performance results and a comparison with the
Installing and using
Please see INSTALL.md for instructions on how to use streamly with your Haskell build tool or package manager. You may want to go through it before jumping to run the examples below.
Streamly provides just the core stream types, type casting and
concurrency control combinators. Stream construction, transformation, folding,
merging, zipping combinators are found in
The following snippet provides a simple stream composition example that reads numbers from stdin, prints the squares of even numbers and exits if an even number more than 9 is entered.
import Streamly import qualified Streamly.Prelude as S import Data.Function ((&)) main = S.drain $ S.repeatM getLine & fmap read & S.filter even & S.takeWhile (<= 9) & fmap (\x -> x * x) & S.mapM print
conduit and like
composes stream data instead of stream processors (functions). A stream is
just like a list and is explicitly passed around to functions that process the
stream. Therefore, no special operator is needed to join stages in a streaming
pipeline, just the standard function application (
$) or reverse function
&) operator is enough.
Concurrent Stream Generation
consM or its operator form
|: can be used to construct a stream from
monadic actions. A stream constructed with
consM can run the monadic actions
in the stream concurrently when used with appropriate stream type combinator
The following code finishes in 3 seconds (6 seconds when serial), note the order of elements in the resulting output, the outputs are consumed as soon as each action is finished (asyncly):
> let p n = threadDelay (n * 1000000) >> return n > S.toList $ asyncly $ p 3 |: p 2 |: p 1 |: S.nil [1,2,3]
aheadly if you want speculative concurrency i.e. execute the actions in
the stream concurrently but consume the results in the specified order:
> S.toList $ aheadly $ p 3 |: p 2 |: p 1 |: S.nil [3,2,1]
Monadic stream generation functions e.g.
fromFoldableM etc. can work concurrently.
The following finishes in 10 seconds (100 seconds when serial):
S.drain $ asyncly $ S.replicateM 10 $ p 10
Concurrency Auto Scaling
Concurrency is auto-scaled i.e. more actions are executed concurrently if the
consumer is consuming the stream at a higher speed. How many tasks are executed
concurrently can be controlled by
maxThreads and how many results are
buffered ahead of consumption can be controlled by
maxBuffer. See the
documentation in the
Concurrent Streaming Pipelines
|$ to apply stream processing functions concurrently. The
following example prints a “hello” every second; if you use
& instead of
|& you will see that the delay doubles to 2 seconds instead because of serial
main = S.drain $ S.repeatM (threadDelay 1000000 >> return "hello") |& S.mapM (\x -> threadDelay 1000000 >> putStrLn x)
We can use
sequence functions concurrently on a stream.
> let p n = threadDelay (n * 1000000) >> return n > S.drain $ aheadly $ S.mapM (\x -> p 1 >> print x) (serially $ repeatM (p 1))
Serial and Concurrent Merging
Semigroup and Monoid instances can be used to fold streams serially or concurrently. In the following example we compose ten actions in the stream, each with a delay of 1 to 10 seconds, respectively. Since all the actions are concurrent we see one output printed every second:
import Streamly import qualified Streamly.Prelude as S import Control.Concurrent (threadDelay) main = S.toList $ parallely $ foldMap delay [1..10] where delay n = S.yieldM $ threadDelay (n * 1000000) >> print n
Streams can be combined together in many ways. We provide some examples
below, see the tutorial for more ways. We use the following
function in the examples to demonstrate the concurrency aspects:
import Streamly import qualified Streamly.Prelude as S import Control.Concurrent delay n = S.yieldM $ do threadDelay (n * 1000000) tid <- myThreadId putStrLn (show tid ++ ": Delay " ++ show n)
main = S.drain $ delay 3 <> delay 2 <> delay 1
ThreadId 36: Delay 3 ThreadId 36: Delay 2 ThreadId 36: Delay 1
main = S.drain . parallely $ delay 3 <> delay 2 <> delay 1
ThreadId 42: Delay 1 ThreadId 41: Delay 2 ThreadId 40: Delay 3
Nested Loops (aka List Transformer)
The monad instance composes like a list monad.
import Streamly import qualified Streamly.Prelude as S loops = do x <- S.fromFoldable [1,2] y <- S.fromFoldable [3,4] S.yieldM $ putStrLn $ show (x, y) main = S.drain loops
(1,3) (1,4) (2,3) (2,4)
Concurrent Nested Loops
To run the above code with speculative concurrency i.e. each iteration in the loop can run concurrently but the results are presented to the consumer of the output in the same order as serial execution:
main = S.drain $ aheadly $ loops
Different stream types execute the loop iterations in different ways. For
wSerially interleaves the loop iterations. There are several
concurrent stream styles to execute the loop iterations concurrently in
different ways, see the
Streamly.Tutorial module for a detailed treatment.
Streams can perform semigroup (<>) and monadic bind (>>=) operations
concurrently using combinators like
parallelly. For example,
to concurrently generate squares of a stream of numbers and then concurrently
sum the square roots of all combinations of two streams:
import Streamly import qualified Streamly.Prelude as S main = do s <- S.sum $ asyncly $ do -- Each square is performed concurrently, (<>) is concurrent x2 <- foldMap (\x -> return $ x * x) [1..100] y2 <- foldMap (\y -> return $ y * y) [1..100] -- Each addition is performed concurrently, monadic bind is concurrent return $ sqrt (x2 + y2) print s
Example: Listing Directories Recursively/Concurrently
The following code snippet lists a directory tree recursively, reading multiple directories concurrently:
import Control.Monad.IO.Class (liftIO) import Path.IO (listDir, getCurrentDir) -- from path-io package import Streamly (AsyncT, adapt) import qualified Streamly.Prelude as S listDirRecursive :: AsyncT IO () listDirRecursive = getCurrentDir >>= readdir >>= liftIO . mapM_ putStrLn where readdir dir = do (dirs, files) <- listDir dir S.yield (map show dirs ++ map show files) <> foldMap readdir dirs main :: IO () main = S.drain $ adapt $ listDirRecursive
AsyncT is a stream monad transformer. If you are familiar with a list
transformer, it is nothing but
ListT with concurrency semantics. For example,
the semigroup operation
<> is concurrent. This makes
too. You can replace
SerialT and the above code will become
serial, exactly equivalent to a
For bounded concurrent streams, stream yield rate can be specified. For example, to print hello once every second you can simply write this:
import Streamly import Streamly.Prelude as S main = S.drain $ asyncly $ avgRate 1 $ S.repeatM $ putStrLn "hello"
For some practical uses of rate control, see AcidRain.hs and CirclingSquare.hs . Concurrency of the stream is automatically controlled to match the specified rate. Rate control works precisely even at throughputs as high as millions of yields per second. For more sophisticated rate control see the haddock documentation.
Streamly.Memory.Array module provides immutable arrays. Arrays are the
computing duals of streams. Streams are good at sequential access and immutable
transformations of in-transit data whereas arrays are good at random access and
in-place transformations of buffered data. Unlike streams which are potentially
infinite, arrays are necessarily finite. Arrays can be used as an efficient
interface between streams and external storage systems like memory, files and
network. Streams and arrays complete each other to provide a general purpose
computing system. The design of streamly as a general purpose computing
framework is centered around these two fundamental aspects of computing and
Streamly.Memory.Array uses pinned memory outside GC and therefore avoid any
GC overhead for the storage in arrays. Streamly allows efficient
transformations over arrays using streams. It uses arrays to transfer data to
and from the operating system and to store data in memory.
Folds are consumers of streams.
Streamly.Data.Fold module provides a
type that represents a
foldl'. Such folds can be efficiently composed
allowing the compiler to perform stream fusion and therefore implement high
performance combinators for consuming streams. A stream can be distributed to
multiple folds, or it can be partitioned across multiple folds, or
demultiplexed over multiple folds, or unzipped to two folds. We can also use
folds to fold segments of stream generating a stream of the folded results.
If you are familiar with the
foldl library, these are the same composable
left folds but simpler and better integrated with streamly, and with many more
powerful ways of composing and applying them.
Unfolds are duals of folds. Folds help us compose consumers of streams
efficiently and unfolds help us compose producers of streams efficiently.
Streamly.Data.Unfold provides an
Unfold type that represents an
or a stream generator. Such generators can be combined together efficiently
allowing the compiler to perform stream fusion and implement high performance
stream merging combinators.
The following code snippets implement some common Unix command line utilities
using streamly. You can compile these with
ghc -O2 -fspec-constr-recursive=16 -fmax-worker-args=16 and compare the performance with regular GNU coreutils
available on your system. Though many of these are not most optimal solutions
to keep them short and elegant. Source file
in the examples directory includes these examples.
module Main where import qualified Streamly.Prelude as S import qualified Streamly.Data.Fold as FL import qualified Streamly.Memory.Array as A import qualified Streamly.FileSystem.Handle as FH import qualified System.IO as FH import Data.Char (ord) import System.Environment (getArgs) import System.IO (openFile, IOMode(..), stdout) withArg f = do (name : _) <- getArgs src <- openFile name ReadMode f src withArg2 f = do (sname : dname : _) <- getArgs src <- openFile sname ReadMode dst <- openFile dname WriteMode f src dst
cat = S.fold (FH.writeChunks stdout) . S.unfold FH.readChunks main = withArg cat
cp src dst = S.fold (FH.writeChunks dst) $ S.unfold FH.readChunks src main = withArg2 cp
wcl = S.length . S.splitOn (== 10) FL.drain . S.unfold FH.read main = withArg wcl >>= print
Average Line Length
avgll = S.fold avg . S.splitOn (== 10) FL.length . S.unfold FH.read where avg = (/) <$> toDouble FL.sum <*> toDouble FL.length toDouble = fmap (fromIntegral :: Int -> Double) main = withArg avgll >>= print
Line Length Histogram
classify is not released yet, and is available in
llhisto = S.fold (FL.classify FL.length) . S.map bucket . S.splitOn (== 10) FL.length . S.unfold FH.read where bucket n = let i = n `mod` 10 in if i > 9 then (9,n) else (i,n) main = withArg llhisto >>= print
Its easy to build concurrent client and server programs using streamly.
Streamly.Network.* modules provide easy combinators to build network servers
and client programs using streamly. See
in the examples directory.
Exceptions can be thrown at any point using the
MonadThrow instance. Standard
exception handling combinators like
onException are provided in
In presence of concurrency, synchronous exceptions work just the way they are supposed to work in non-concurrent code. When concurrent streams are combined together, exceptions from the constituent streams are propagated to the consumer stream. When an exception occurs in any of the constituent streams other concurrent streams are promptly terminated.
There is no notion of explicit threads in streamly, therefore, no
asynchronous exceptions to deal with. You can just ignore the zillions of
blogs, talks, caveats about async exceptions. Async exceptions just don’t
exist. Please don’t use things like
throwTo just for fun!
Reactive Programming (FRP)
Streamly is a foundation for first class reactive programming as well by virtue of integrating concurrency and streaming. See AcidRain.hs for a console based FRP game example and CirclingSquare.hs for an SDL based animation example.
Streamly, short for streaming concurrently, provides monadic streams, with a simple API, almost identical to standard lists, and an in-built support for concurrency. By using stream-style combinators on stream composition, streams can be generated, merged, chained, mapped, zipped, and consumed concurrently – providing a generalized high level programming framework unifying streaming and concurrency. Controlled concurrency allows even infinite streams to be evaluated concurrently. Concurrency is auto scaled based on feedback from the stream consumer. The programmer does not have to be aware of threads, locking or synchronization to write scalable concurrent programs.
Streamly is a programmer first library, designed to be useful and friendly to programmers for solving practical problems in a simple and concise manner. Some key points in favor of streamly are:
- Simplicity: Simple list like streaming API, if you know how to use lists then you know how to use streamly. This library is built with simplicity and ease of use as a design goal.
- Concurrency: Simple, powerful, and scalable concurrency. Concurrency is built-in, and not intrusive, concurrent programs are written exactly the same way as non-concurrent ones.
- Generality: Unifies functionality provided by several disparate packages (streaming, concurrency, list transformer, logic programming, reactive programming) in a concise API.
- Performance: Streamly is designed for high performance. It employs stream
fusion optimizations for best possible performance. Serial peformance is
equivalent to the venerable
vectorlibrary in most cases and even better in some cases. Concurrent performance is unbeatable. See streaming-benchmarks for a comparison of popular streaming libraries on micro-benchmarks.
The basic streaming functionality of streamly is equivalent to that provided by
streaming libraries like
In addition to providing streaming functionality, streamly subsumes
the functionality of list transformer libraries like
list-t, and also the logic
programming library logict. On
the concurrency side, it subsumes the functionality of the
async package, and provides even
higher level concurrent composition. Because it supports
streaming with concurrency we can write FRP applications similar in concept to
Comparison with existing packages section at the end of the
For more information, see:
- Detailed tutorial
- Reference documentation
- Streaming benchmarks
- Concurrency benchmarks
For additional unreleased/experimental APIs, build the haddock docs using:
$ cabal haddock --haddock-option="--show-all" $ stack haddock --haddock-arguments "--show-all" --no-haddock-deps
Please feel free to ask questions on the streamly gitter channel. If you require professional support, consulting, training or timely enhancements to the library please contact firstname.lastname@example.org.
The following authors/libraries have influenced or inspired this library in a significant way:
- Roman Leshchinskiy (vector)
- Gabriel Gonzalez (foldl)
- Alberto G. Corona (transient)
credits directory for full list of contributors, credits and licenses.
The code is available under BSD-3 license on github. Join the gitter chat channel for discussions. Please ask any questions on the gitter channel or contact the maintainer directly. All contributions are welcome!
- Change the signature of
foldrMto ensure that it is lazy
- Change the signature of
iterateMto ensure that it is lazy.
scanxwould now require an additional
ParallelTwas unaffected by
maxBuffercan limit the buffer of a
ParallelTstream as well. When the buffer becomes full, the producer threads block.
ParallelTstreams no longer have an unlimited buffer by default. Now the buffer for parallel streams is limited to 1500 by default, the same as other concurrent stream types.
runStreamhas been replaced by
runNhas been replaced by
runWhilehas been replaced by
fromHandlehas been deprecated. Please use
Streamly.Data.Fold.toListto split the stream to a stream of
Stringseparated by a newline.
toHandlehas been deprecated. Please use
concatUnfoldto add newlines to a stream,
Streamly.Data.Unicode.Stream.encodeUtf8for encoding and
Streamly.FileSystem.Handle.writefor writing to a file handle.
- Remove deprecated APIs
- Replace deprecated API
scanwith a new signature, to scan using Fold.
runStreamhas been deprecated, please use
Remove deprecated module
Streamly.Time(moved to Streamly.Internal.Data.Time)
Streamly.Internal(functionality moved to the Internal hierarchy)
- Fix a bug that caused
uniqfunction to yield the same element twice.
- Fix a bug that caused “thread blocked indefinitely in an MVar operation” exception in a parallel stream.
This release contains a lot of new features and major enhancements. For more details on the new features described below please see the haddock docs of the modules on hackage.
Streamly.Prelude for new exception handling combinators like
Streamly.Data.Fold module provides composable folds (stream consumers). Folds
allow splitting, grouping, partitioning, unzipping and nesting a stream onto
multiple folds without breaking the stream. Combinators are provided for
temporal and spatial window based fold operations, for example, to support
folding and aggregating data for timeout or inactivity based sessions.
Streamly.Data.Unfold module provides composable stream generators. Unfolds allow
high performance merging/flattening/combining of stream generators.
Streaming File IO
Streamly.FileSystem.Handle provides handle based streaming file IO
Streaming Network IO
Streamly.Network.Socketprovides socket based streaming network IO operations.
Streamly.Network.Inet.TCPprovides combinators to build Inet/TCP clients and servers.
Streamly.Prelude combinator performs a
concatMap using a supplied merge/concat strategy. This is a very
powerful combinator as you can, for example, concat streams
concurrently using this.
Add the following new features/modules:
- Unicode Strings:
Streamly.Data.Unicode.Streammodule provides encoding/decoding of character streams and other character stream operations.
Streamly.Memory.Arraymodule provides arrays for efficient in-memory buffering and efficient interfacing with IO.
- Unicode Strings:
Add the following to
concatUnfoldto concat a stream after unfolding each element
intersperseintersperse an element in between consecutive elements in stream
tracecombinator maps a monadic function on a stream just for side effects
tapredirects a copy of the stream to a
- Fix a bug that caused
maxThreadsdirective to be ignored when rate control was not used.
- Add GHCJS support
- Remove dependency on “clock” package
Monadconstraint may be needed on some of the existing APIs (
- Add the following functions to Streamly.Prelude:
- Following instances were added for
Identity: IsList, Eq, Ord, Show, Read, IsString, NFData, NFData1, Traversable
- Performance improvements
- Add benchmarks to measure composed and iterated operations
- Cleanup any pending threads when an exception occurs.
- Fixed a livelock in ahead style streams. The problem manifests sometimes when multiple streams are merged together in ahead style and one of them is a nil stream.
- As per expected concurrency semantics each forked concurrent task must run
with the monadic state captured at the fork point. This release fixes a bug,
which, in some cases caused an incorrect monadic state to be used for a
concurrent action, leading to unexpected behavior when concurrent streams are
used in a stateful monad e.g.
StateT. Particularly, this bug cannot affect
- Performance improvements, especially space consumption, for concurrent streams
- Leftover threads are now cleaned up as soon as the consumer is garbage collected.
- Fix a bug in concurrent function application that in certain cases would unnecessarily share the concurrency state resulting in incorrect output stream.
- Fix passing of state across
wSerialcombinators. Without this fix combinators that rely on state passing e.g.
maxBufferwon’t work across these combinators.
- Added rate limiting combinators
constRateto control the yield rate of a stream.
Streamly.Timemodule is now deprecated, its functionality is subsumed by the new rate limiting combinators.
- foldxM was not fully strict, fixed.
- Signatures of
- Some functions in prelude now require an additional
Monadconstraint on the underlying type of the stream.
oncehas been deprecated and renamed to
- Add concurrency control primitives
- Concurrency of a stream with bounded concurrency when used with
takeis now limited by the number of elements demanded by
- Significant performance improvements utilizing stream fusion optimizations.
yieldto construct a singleton stream from a pure value
repeatto generate an infinite stream by repeating a pure value
fromListMto generate streams from lists, faster than
mapas a synonym of fmap
scanlM', the monadic version of scanl’
- Some prelude functions, to whom concurrency capability has been added, will
now require a
- Fixed a race due to which, in a rare case, we might block indefinitely on an MVar due to a lost wakeup.
- Fixed an issue in adaptive concurrency. The issue caused us to stop creating more worker threads in some cases due to a race. This bug would not cause any functional issue but may reduce concurrency in some cases.
- Added a concurrent lookahead stream type
fromFoldableMAPI that creates a stream from a container of monadic actions
- Monadic stream generation functions
fromFoldableMcan now generate streams concurrently when used with concurrent stream types.
- Monad transformation functions
sequencecan now map actions concurrently when used at appropriate stream types.
- Added concurrent function application operators to run stages of a stream processing function application pipeline concurrently.
- Fixed a bug that caused some transformation ops to return incorrect results
when used with concurrent streams. The affected ops are
Changed the semantics of the Semigroup instance for
ParallelT. The new semantics are as follows:
<>operation interleaves two streams
<>now concurrently merges two streams in a left biased manner using demand based concurrency.
<>operation now concurrently meges the two streams in a fairly parallel manner.
To adapt to the new changes, replace
serialwherever it is used for stream types other than
Alternativeinstance. To adapt to this change replace any usage of
Stream type now defaults to the
SerialTtype unless explicitly specified using a type combinator or a monomorphic type. This change reduces puzzling type errors for beginners. It includes the following two changes:
- Change the type of all stream elimination functions to use
SerialTinstead of a polymorphic type. This makes sure that the stream type is always fixed at all exits.
- Change the type combinators (e.g.
parallely) to only fix the argument stream type and the output stream type remains polymorphic.
Stream types may have to be changed or type combinators may have to be added or removed to adapt to this change.
- Change the type of all stream elimination functions to use
Change the type of
foldrMto make it consistent with
asyncis renamed to
asyncis now a new API with a different meaning.
ZipAsyncis renamed to
ZipAsyncis now ZipAsyncM specialized to the IO Monad.
MonadErrorinstance as it was not working correctly for parallel compositions. Use
MonadThrowinstead for error propagation.
Remove Num/Fractional/Floating instances as they are not very useful. Use
- Deprecate and rename the following symbols:
- Deprecate the following symbols for future removal:
- Add the following functions:
|:operator to construct streams from monadic actions
onceto create a singleton stream from a monadic action
repeatMto construct a stream by repeating a monadic action
scanl'strict left scan
foldl'strict left fold
foldlM'strict left fold with a monadic fold function
serialrun two streams serially one after the other
asyncrun two streams asynchronously
parallelrun two streams in parallel (replaces
WAsyncTstream type for BFS version of
- Add simpler stream types that are specialized to the IO monad
- Put a bound (1500) on the output buffer used for asynchronous tasks
- Put a limit (1500) on the number of threads used for Async and WAsync types
- Fixed a bug that casued unexpected behavior when
purewas used to inject values in Applicative composition of
consright associative and provide an operator form
- Improve performance of some stream operations (
- Fix the
productoperation. Earlier, it always returned 0 due to a bug
- Fix the
lastoperation, which returned
Nothingfor singleton streams
- Initial release