distributed-process-0.5.0: Cloud Haskell: Erlang-style concurrency in Haskell

Safe HaskellNone
LanguageHaskell98

Control.Distributed.Process

Contents

Description

Cloud Haskell

This is an implementation of Cloud Haskell, as described in Towards Haskell in the Cloud by Jeff Epstein, Andrew Black, and Simon Peyton Jones (see http://research.microsoft.com/en-us/um/people/simonpj/papers/parallel/), although some of the details are different. The precise message passing semantics are based on A unified semantics for future Erlang by Hans Svensson, Lars-Åke Fredlund and Clara Benac Earle.

For a detailed description of the package and other reference materials, please see the distributed-process wiki page on github: https://github.com/haskell-distributed/distributed-process/wiki.

Synopsis

Basic types

data SendPortId Source

A send port is identified by a SendPortId.

You cannot send directly to a SendPortId; instead, use newChan to create a SendPort.

processNodeId :: ProcessId -> NodeId Source

The ID of the node the process is running on

sendPortProcessId :: SendPortId -> ProcessId Source

The ID of the process that will receive messages sent on this port

liftIO :: MonadIO m => forall a. IO a -> m a

Lift a computation from the IO monad.

Basic messaging

send :: Serializable a => ProcessId -> a -> Process () Source

Send a message

expect :: forall a. Serializable a => Process a Source

Wait for a message of a specific type

expectTimeout :: forall a. Serializable a => Int -> Process (Maybe a) Source

Like expect but with a timeout

Channels

data ReceivePort a Source

The receive end of a typed channel (not serializable)

Note that ReceivePort implements Functor, Applicative, Alternative and Monad. This is especially useful when merging receive ports.

data SendPort a Source

The send send of a typed channel (serializable)

Instances

sendPortId :: SendPort a -> SendPortId Source

The (unique) ID of this send port

newChan :: Serializable a => Process (SendPort a, ReceivePort a) Source

Create a new typed channel

sendChan :: Serializable a => SendPort a -> a -> Process () Source

Send a message on a typed channel

receiveChan :: Serializable a => ReceivePort a -> Process a Source

Wait for a message on a typed channel

receiveChanTimeout :: Serializable a => Int -> ReceivePort a -> Process (Maybe a) Source

Like receiveChan but with a timeout. If the timeout is 0, do a non-blocking check for a message.

mergePortsBiased :: Serializable a => [ReceivePort a] -> Process (ReceivePort a) Source

Merge a list of typed channels.

The result port is left-biased: if there are messages available on more than one port, the first available message is returned.

mergePortsRR :: Serializable a => [ReceivePort a] -> Process (ReceivePort a) Source

Like mergePortsBiased, but with a round-robin scheduler (rather than left-biased)

Unsafe messaging variants

unsafeSend :: Serializable a => ProcessId -> a -> Process () Source

Unsafe variant of send. This function makes no attempt to serialize and (in the case when the destination process resides on the same local node) therefore ensure that the payload is fully evaluated before it is delivered.

unsafeSendChan :: Serializable a => SendPort a -> a -> Process () Source

Send a message on a typed channel. This function makes no attempt to serialize and (in the case when the ReceivePort resides on the same local node) therefore ensure that the payload is fully evaluated before it is delivered.

unsafeNSend :: Serializable a => String -> a -> Process () Source

Named send to a process in the local registry (asynchronous). This function makes no attempt to serialize and (in the case when the destination process resides on the same local node) therefore ensure that the payload is fully evaluated before it is delivered.

unsafeWrapMessage :: Serializable a => a -> Message Source

This is the unsafe variant of wrapMessage. See Control.Distributed.Process.UnsafePrimitives for details.

Advanced messaging

data Match b Source

Opaque type used in receiveWait and receiveTimeout

receiveWait :: [Match b] -> Process b Source

Test the matches in order against each message in the queue

receiveTimeout :: Int -> [Match b] -> Process (Maybe b) Source

Like receiveWait but with a timeout.

If the timeout is zero do a non-blocking check for matching messages. A non-zero timeout is applied only when waiting for incoming messages (that is, after we have checked the messages that are already in the mailbox).

match :: forall a b. Serializable a => (a -> Process b) -> Match b Source

Match against any message of the right type

matchIf :: forall a b. Serializable a => (a -> Bool) -> (a -> Process b) -> Match b Source

Match against any message of the right type that satisfies a predicate

matchUnknown :: Process b -> Match b Source

Remove any message from the queue

matchAny :: forall b. (Message -> Process b) -> Match b Source

Match against an arbitrary message. matchAny removes the first available message from the process mailbox. To handle arbitrary raw messages once removed from the mailbox, see handleMessage and unwrapMessage.

matchAnyIf :: forall a b. Serializable a => (a -> Bool) -> (Message -> Process b) -> Match b Source

Match against an arbitrary message. Intended for use with handleMessage and unwrapMessage, this function only removes a message from the process mailbox, if the supplied condition matches. The success (or failure) of runtime type checks deferred to handleMessage and friends is irrelevant here, i.e., if the condition evaluates to True then the message will be removed from the process mailbox and decoded, but that does not guarantee that an expression passed to handleMessage will pass the runtime type checks and therefore be evaluated.

matchChan :: ReceivePort a -> (a -> Process b) -> Match b Source

Match on a typed channel

matchSTM :: STM a -> (a -> Process b) -> Match b Source

Match on an arbitrary STM action.

This rather unusaul match primitive allows us to compose arbitrary STM actions with checks against our process' mailbox and/or any typed channel ReceivePorts we may hold.

This allows us to process multiple input streams along with our mailbox, in just the same way that matchChan supports checking both the mailbox and an arbitrary set of typed channels in one atomic transaction.

Note there are no ordering guarnatees with respect to these disparate input sources.

data Message Source

Messages consist of their typeRep fingerprint and their encoding

matchMessage :: (Message -> Process Message) -> Match Message Source

Match against any message, regardless of the underlying (contained) type

matchMessageIf :: (Message -> Bool) -> (Message -> Process Message) -> Match Message Source

Match against any message (regardless of underlying type) that satisfies a predicate

isEncoded :: Message -> Bool Source

internal use only.

wrapMessage :: Serializable a => a -> Message Source

Wrap a Serializable value in a Message. Note that Messages are Serializable - like the datum they contain - but also note, deserialising such a Message will yield a Message, not the type within it! To obtain the wrapped datum, use unwrapMessage or handleMessage with a specific type.

do
   self <- getSelfPid
   send self (wrapMessage "blah")
   Nothing  <- expectTimeout 1000000 :: Process (Maybe String)
   (Just m) <- expectTimeout 1000000 :: Process (Maybe Message)
   (Just "blah") <- unwrapMessage m :: Process (Maybe String)

unwrapMessage :: forall m a. (Monad m, Serializable a) => Message -> m (Maybe a) Source

Attempt to unwrap a raw Message. If the type of the decoded message payload matches the expected type, the value will be returned with Just, otherwise Nothing indicates the types do not match.

This expression, for example, will evaluate to Nothing > unwrapMessage (wrapMessage "foobar") :: Process (Maybe Int)

Whereas this expression, will yield Just "foo" > unwrapMessage (wrapMessage "foo") :: Process (Maybe String)

handleMessage :: forall m a b. (Monad m, Serializable a) => Message -> (a -> m b) -> m (Maybe b) Source

Attempt to handle a raw Message. If the type of the message matches the type of the first argument to the supplied expression, then the message will be decoded and the expression evaluated against its value. If this runtime type checking fails however, Nothing will be returned to indicate the fact. If the check succeeds and evaluation proceeds, the resulting value with be wrapped with Just.

Intended for use in catchesExit and matchAny primitives.

handleMessageIf :: forall m a b. (Monad m, Serializable a) => Message -> (a -> Bool) -> (a -> m b) -> m (Maybe b) Source

Conditionally handle a raw Message. If the predicate (a -> Bool) evaluates to True, invokes the supplied handler, other returns Nothing to indicate failure. See handleMessage for further information about runtime type checking.

handleMessage_ :: forall m a. (Monad m, Serializable a) => Message -> (a -> m ()) -> m () Source

As handleMessage but ignores result, which is useful if you don't care whether or not the handler succeeded.

handleMessageIf_ :: forall m a. (Monad m, Serializable a) => Message -> (a -> Bool) -> (a -> m ()) -> m () Source

Conditional version of handleMessage_.

forward :: Message -> ProcessId -> Process () Source

Forward a raw Message to the given ProcessId.

delegate :: ProcessId -> (Message -> Bool) -> Process () Source

Receives messages and forwards them to pid if p msg == True.

relay :: ProcessId -> Process () Source

A straight relay that forwards all messages to the supplied pid.

proxy :: Serializable a => ProcessId -> (a -> Process Bool) -> Process () Source

Proxies pid and forwards messages whenever proc evaluates to True. Unlike delegate the predicate proc runs in the Process monad, allowing for richer proxy behaviour. If proc returns False or the runtime type check fails, no action is taken and the proxy process will continue running.

Process management

spawn :: NodeId -> Closure (Process ()) -> Process ProcessId Source

Spawn a process

For more information about Closure, see Control.Distributed.Process.Closure.

See also call.

call :: Serializable a => Static (SerializableDict a) -> NodeId -> Closure (Process a) -> Process a Source

Run a process remotely and wait for it to reply

We monitor the remote process: if it dies before it can send a reply, we die too.

For more information about Static, SerializableDict, and Closure, see Control.Distributed.Process.Closure.

See also spawn.

terminate :: Process a Source

Terminate immediately (throws a ProcessTerminationException)

die :: Serializable a => a -> Process b Source

Die immediately - throws a ProcessExitException with the given reason.

kill :: ProcessId -> String -> Process () Source

Forceful request to kill a process. Where exit provides an exception that can be caught and handled, kill throws an unexposed exception type which cannot be handled explicitly (by type).

exit :: Serializable a => ProcessId -> a -> Process () Source

Graceful request to exit a process. Throws ProcessExitException with the supplied reason encoded as a message. Any exit signal raised in this manner can be handled using the catchExit family of functions.

catchExit :: forall a b. (Show a, Serializable a) => Process b -> (ProcessId -> a -> Process b) -> Process b Source

Catches ProcessExitException. The handler will not be applied unless its type matches the encoded data stored in the exception (see the reason argument given to the exit primitive). If the handler cannot be applied, the exception will be re-thrown.

To handle ProcessExitException without regard for reason, see catch. To handle multiple reasons of differing types, see catchesExit.

catchesExit :: Process b -> [ProcessId -> Message -> Process (Maybe b)] -> Process b Source

Lift catches (almost).

As ProcessExitException stores the exit reason as a typed, encoded message, a handler must accept inputs of the expected type. In order to handle a list of potentially different handlers (and therefore input types), a handler passed to catchesExit must accept Message and return Maybe (i.e., Just p if it handled the exit reason, otherwise Nothing).

See maybeHandleMessage and Message for more details.

data ProcessRegistrationException Source

Exception thrown when a process attempts to register a process under an already-registered name or to unregister a name that hasn't been registered

data SpawnRef Source

SpawnRef are used to return pids of spawned processes

getSelfPid :: Process ProcessId Source

Our own process ID

getSelfNode :: Process NodeId Source

Get the node ID of our local node

getProcessInfo :: ProcessId -> Process (Maybe ProcessInfo) Source

Get information about the specified process

getNodeStats :: NodeId -> Process (Either DiedReason NodeStats) Source

Get statistics about the specified node

getLocalNodeStats :: Process NodeStats Source

Get statistics about our local node

Monitoring and linking

link :: ProcessId -> Process () Source

Link to a remote process (asynchronous)

When process A links to process B (that is, process A calls link pidB) then an asynchronous exception will be thrown to process A when process B terminates (normally or abnormally), or when process A gets disconnected from process B. Although it is technically possible to catch these exceptions, chances are if you find yourself trying to do so you should probably be using monitor rather than link. In particular, code such as

link pidB   -- Link to process B
expect      -- Wait for a message from process B
unlink pidB -- Unlink again

doesn't quite do what one might expect: if process B sends a message to process A, and subsequently terminates, then process A might or might not be terminated too, depending on whether the exception is thrown before or after the unlink (i.e., this code has a race condition).

Linking is all-or-nothing: A is either linked to B, or it's not. A second call to link has no effect.

Note that link provides unidirectional linking (see spawnSupervised). Linking makes no distinction between normal and abnormal termination of the remote process.

linkNode :: NodeId -> Process () Source

Link to a node (asynchronous)

linkPort :: SendPort a -> Process () Source

Link to a channel (asynchronous)

unlink :: ProcessId -> Process () Source

Remove a link

This is synchronous in the sense that once it returns you are guaranteed that no exception will be raised if the remote process dies. However, it is asynchronous in the sense that we do not wait for a response from the remote node.

unlinkNode :: NodeId -> Process () Source

Remove a node link

This has the same synchronous/asynchronous nature as unlink.

unlinkPort :: SendPort a -> Process () Source

Remove a channel (send port) link

This has the same synchronous/asynchronous nature as unlink.

monitor :: ProcessId -> Process MonitorRef Source

Monitor another process (asynchronous)

When process A monitors process B (that is, process A calls monitor pidB) then process A will receive a ProcessMonitorNotification when process B terminates (normally or abnormally), or when process A gets disconnected from process B. You receive this message like any other (using expect); the notification includes a reason (DiedNormal, DiedException, DiedDisconnect, etc.).

Every call to monitor returns a new monitor reference MonitorRef; if multiple monitors are set up, multiple notifications will be delivered and monitors can be disabled individually using unmonitor.

monitorNode :: NodeId -> Process MonitorRef Source

Monitor a node (asynchronous)

monitorPort :: forall a. Serializable a => SendPort a -> Process MonitorRef Source

Monitor a typed channel (asynchronous)

unmonitor :: MonitorRef -> Process () Source

Remove a monitor

This has the same synchronous/asynchronous nature as unlink.

withMonitor :: ProcessId -> Process a -> Process a Source

Establishes temporary monitoring of another process.

withMonitor pid code sets up monitoring of pid for the duration of code. Note: although monitoring is no longer active when withMonitor returns, there might still be unreceived monitor messages in the queue.

data MonitorRef Source

MonitorRef is opaque for regular Cloud Haskell processes

data PortLinkException Source

Exception thrown when a linked channel (port) dies

data DiedReason Source

Why did a process die?

Constructors

DiedNormal

Normal termination

DiedException !String

The process exited with an exception (provided as String because Exception does not implement Binary)

DiedDisconnect

We got disconnected from the process node

DiedNodeDown

The process node died

DiedUnknownId

Invalid (processnodechannel) identifier

Closures

data Closure a :: * -> *

A closure is a static value and an encoded environment

Instances

Eq (Closure a) 
Ord (Closure a) 
Show (Closure a) 
Typeable * a => Binary (Closure a) 
Typeable (* -> *) Closure 

closure

Arguments

:: Static (ByteString -> a)

Decoder

-> ByteString

Encoded closure environment

-> Closure a 

data Static a :: * -> *

A static value. Static is opaque; see staticLabel and staticApply.

Instances

Eq (Static a) 
Ord (Static a) 
Show (Static a) 
Typeable * a => Binary (Static a) 
Typeable (* -> *) Static 

unStatic :: Typeable a => Static a -> Process a Source

Resolve a static value

unClosure :: Typeable a => Closure a -> Process a Source

Resolve a closure

data RemoteTable :: *

Runtime dictionary for unstatic lookups

Logging

say :: String -> Process () Source

Log a string

say message sends a message (time, pid of the current process, message) to the process registered as logger. By default, this process simply sends the string to stderr. Individual Cloud Haskell backends might replace this with a different logger process, however.

Registry

register :: String -> ProcessId -> Process () Source

Register a process with the local registry (asynchronous). This version will wait until a response is gotten from the management process. The name must not already be registered. The process need not be on this node. A bad registration will result in a ProcessRegistrationException

The process to be registered does not have to be local itself.

reregister :: String -> ProcessId -> Process () Source

Like register, but will replace an existing registration. The name must already be registered.

unregister :: String -> Process () Source

Remove a process from the local registry (asynchronous). This version will wait until a response is gotten from the management process. The name must already be registered.

whereis :: String -> Process (Maybe ProcessId) Source

Query the local process registry

nsend :: Serializable a => String -> a -> Process () Source

Named send to a process in the local registry (asynchronous)

registerRemoteAsync :: NodeId -> String -> ProcessId -> Process () Source

Register a process with a remote registry (asynchronous).

The process to be registered does not have to live on the same remote node. Reply wil come in the form of a RegisterReply message

See comments in whereisRemoteAsync

unregisterRemoteAsync :: NodeId -> String -> Process () Source

Remove a process from a remote registry (asynchronous).

Reply wil come in the form of a RegisterReply message

See comments in whereisRemoteAsync

whereisRemoteAsync :: NodeId -> String -> Process () Source

Query a remote process registry (asynchronous)

Reply will come in the form of a WhereIsReply message.

There is currently no synchronous version of whereisRemoteAsync: if you implement one yourself, be sure to take into account that the remote node might die or get disconnect before it can respond (i.e. you should use monitorNode and take appropriate action when you receive a NodeMonitorNotification).

nsendRemote :: Serializable a => NodeId -> String -> a -> Process () Source

Named send to a process in a remote registry (asynchronous)

data WhereIsReply Source

(Asynchronous) reply from whereis

data RegisterReply Source

(Asynchronous) reply from register and unregister

Constructors

RegisterReply String Bool 

Exception handling

catch :: Exception e => Process a -> (e -> Process a) -> Process a Source

Lift catch

data Handler a Source

You need this when using catches

Constructors

forall e . Exception e => Handler (e -> Process a) 

Instances

try :: Exception e => Process a -> Process (Either e a) Source

Lift try

mask :: ((forall a. Process a -> Process a) -> Process b) -> Process b Source

Lift mask

bracket :: Process a -> (a -> Process b) -> (a -> Process c) -> Process c Source

Lift bracket

Auxiliary API

spawnAsync :: NodeId -> Closure (Process ()) -> Process SpawnRef Source

Asynchronous version of spawn

(spawn is defined in terms of spawnAsync and expect)

spawnSupervised :: NodeId -> Closure (Process ()) -> Process (ProcessId, MonitorRef) Source

Spawn a child process, have the child link to the parent and the parent monitor the child

spawnLink :: NodeId -> Closure (Process ()) -> Process ProcessId Source

Spawn a process and link to it

Note that this is just the sequential composition of spawn and link. (The Unified semantics that underlies Cloud Haskell does not even support a synchronous link operation)

spawnMonitor :: NodeId -> Closure (Process ()) -> Process (ProcessId, MonitorRef) Source

Like spawnLink, but monitor the spawned process

spawnChannel :: forall a. Serializable a => Static (SerializableDict a) -> NodeId -> Closure (ReceivePort a -> Process ()) -> Process (SendPort a) Source

Spawn a new process, supplying it with a new ReceivePort and return the corresponding SendPort.

data DidSpawn Source

(Asynchronius) reply from spawn

Local versions of spawn

spawnLocal :: Process () -> Process ProcessId Source

Spawn a process on the local node

spawnChannelLocal :: Serializable a => (ReceivePort a -> Process ()) -> Process (SendPort a) Source

Create a new typed channel, spawn a process on the local node, passing it the receive port, and return the send port

Reconnecting

reconnect :: ProcessId -> Process () Source

Cloud Haskell provides the illusion of connection-less, reliable, ordered message passing. However, when network connections get disrupted this illusion cannot always be maintained. Once a network connection breaks (even temporarily) no further communication on that connection will be possible. For example, if process A sends a message to process B, and A is then notified (by monitor notification) that it got disconnected from B, A will not be able to send any further messages to B, unless A explicitly indicates that it is acceptable to attempt to reconnect to B using the Cloud Haskell reconnect primitive.

Importantly, when A calls reconnect it acknowledges that some messages to B might have been lost. For instance, if A sends messages m1 and m2 to B, then receives a monitor notification that its connection to B has been lost, calls reconnect and then sends m3, it is possible that B will receive m1 and m3 but not m2.

Note that reconnect does not mean reconnect now but rather /it is okay to attempt to reconnect on the next send/. In particular, if no further communication attempts are made to B then A can use reconnect to clean up its connection to B.

reconnectPort :: SendPort a -> Process () Source

Reconnect to a sendport. See reconnect for more information.