@@ -1,3 +1,12 @@
2+{-# LANGUAGE DeriveDataTypeable #-}
3+{-# LANGUAGE ScopedTypeVariables #-}
4+{-# LANGUAGE RankNTypes #-}
5+{-# LANGUAGE BangPatterns #-}
6+{-# LANGUAGE GeneralizedNewtypeDeriving #-}
7+{-# LANGUAGE RecordWildCards #-}
8+{-# LANGUAGE MagicHash #-}
1-- | Local nodes
10-- | Local nodes
3module Control.Distributed.Process.Node
12module Control.Distributed.Process.Node
@@ -12,10 +21,6 @@ module Control.Distributed.Process.Node
13-- TODO: Calls to 'sendBinary' and co (by the NC) may stall the node controller.
22-- TODO: Calls to 'sendBinary' and co (by the NC) may stall the node controller.
15−#if ! MIN_VERSION_base(4,6,0)
16−import Prelude hiding (catch)
19import System.IO (fixIO, hPutStrLn, stderr)
24import System.IO (fixIO, hPutStrLn, stderr)
20import System.Mem.Weak (Weak, deRefWeak)
25import System.Mem.Weak (Weak, deRefWeak)
21import qualified Data.ByteString.Lazy as BSL (fromChunks)
26import qualified Data.ByteString.Lazy as BSL (fromChunks)
@@ -23,46 +28,67 @@ import Data.Binary (decode)
23import Data.Map (Map)
28import Data.Map (Map)
24import qualified Data.Map as Map
29import qualified Data.Map as Map
27 , fromList
33 , fromList
29 , partitionWithKey
37 , partitionWithKey
31 , filterWithKey
40 , filterWithKey
32 , foldlWithKey
41 , foldlWithKey
43+import Data.Time.Format (formatTime)
44+#if MIN_VERSION_time(1,5,0)
45+import Data.Time.Format (defaultTimeLocale)
47+import System.Locale (defaultTimeLocale)
34import Data.Set (Set)
49import Data.Set (Set)
35import qualified Data.Set as Set
50import qualified Data.Set as Set
42import Data.Foldable (forM_)
59import Data.Foldable (forM_)
43import Data.Maybe (isJust, fromJust, isNothing, catMaybes)
60import Data.Maybe (isJust, fromJust, isNothing, catMaybes)
44import Data.Typeable (Typeable)
61import Data.Typeable (Typeable)
45import Control.Category ((>>>))
62import Control.Category ((>>>))
46−import Control.Applicative ((<$>))
63+import Control.Applicative
47−import Control.Monad (void, when)
64+import Control.Monad (void, when, join)
48import Control.Monad.IO.Class (MonadIO, liftIO)
65import Control.Monad.IO.Class (MonadIO, liftIO)
49import Control.Monad.State.Strict (MonadState, StateT, evalStateT, gets)
66import Control.Monad.State.Strict (MonadState, StateT, evalStateT, gets)
50import qualified Control.Monad.State.Strict as StateT (get, put)
67import qualified Control.Monad.State.Strict as StateT (get, put)
51import Control.Monad.Reader (MonadReader, ReaderT, runReaderT, ask)
68import Control.Monad.Reader (MonadReader, ReaderT, runReaderT, ask)
52import Control.Exception
69import Control.Exception
54− , AsyncException(ThreadKilled)
55 , SomeException
71 , SomeException
56 , Exception
72 , Exception
74+ , uninterruptibleMask_
78+import qualified Control.Exception as Exception
59−import qualified Control.Exception as Exception (catch, finally)
84+import Control.Concurrent (forkIO, killThread)
60−import Control.Concurrent (forkIO)
85+import Control.Distributed.Process.Internal.BiMultiMap (BiMultiMap)
86+import qualified Control.Distributed.Process.Internal.BiMultiMap as BiMultiMap
61import Control.Distributed.Process.Internal.StrictMVar
87import Control.Distributed.Process.Internal.StrictMVar
63 , withMVar
89 , withMVar
64 , modifyMVar
91 , modifyMVar
66 , newEmptyMVar
92 , newEmptyMVar
68 , takeMVar
94 , takeMVar
@@ -77,6 +103,7 @@ import Control.Distributed.Process.Internal.CQueue
77 , enqueue
103 , enqueue
78 , newCQueue
104 , newCQueue
79 , mkWeakCQueue
105 , mkWeakCQueue
81import qualified Network.Transport as NT
108import qualified Network.Transport as NT
82 ( Transport
109 ( Transport
@@ -88,8 +115,8 @@ import qualified Network.Transport as NT
88 , TransportError(..)
115 , TransportError(..)
89 , address
116 , address
90 , closeEndPoint
117 , closeEndPoint
92 , Connection
118 , Connection
94 , EndPointAddress
121 , EndPointAddress
95 , Reliability(ReliableOrdered)
122 , Reliability(ReliableOrdered)
@@ -106,8 +133,11 @@ import Control.Distributed.Process.Internal.Types
106 , LocalProcessId(..)
133 , LocalProcessId(..)
107 , ProcessId(..)
134 , ProcessId(..)
108 , LocalNode(..)
135 , LocalNode(..)
109− , Tracer(..)
136+ , MxEventBus(..)
110 , LocalNodeState(..)
137 , LocalNodeState(..)
138+ , ValidLocalNodeState(..)
139+ , withValidLocalState
140+ , modifyValidLocalState
111 , LocalProcess(..)
141 , LocalProcess(..)
112 , LocalProcessState(..)
142 , LocalProcessState(..)
113 , Process(..)
143 , Process(..)
@@ -117,9 +147,11 @@ import Control.Distributed.Process.Internal.Types
117 , localPidCounter
147 , localPidCounter
118 , localPidUnique
148 , localPidUnique
119 , localProcessWithId
149 , localProcessWithId
120 , localConnections
151 , localConnections
121 , forever'
152 , forever'
122 , MonitorRef(..)
153 , MonitorRef(..)
154+ , NodeClosedException(..)
123 , ProcessMonitorNotification(..)
155 , ProcessMonitorNotification(..)
124 , NodeMonitorNotification(..)
156 , NodeMonitorNotification(..)
125 , PortMonitorNotification(..)
157 , PortMonitorNotification(..)
@@ -139,52 +171,74 @@ import Control.Distributed.Process.Internal.Types
140 , ProcessInfo(..)
172 , ProcessInfo(..)
141 , ProcessInfoNone(..)
173 , ProcessInfoNone(..)
142 , SendPortId(..)
175 , SendPortId(..)
143 , typedChannelWithId
176 , typedChannelWithId
144 , RegisterReply(..)
177 , RegisterReply(..)
145 , WhereIsReply(..)
178 , WhereIsReply(..)
146 , payloadToMessage
179 , payloadToMessage
147 , createUnencodedMessage
180 , createUnencodedMessage
181+ , unsafeCreateUnencodedMessage
148 , runLocalProcess
182 , runLocalProcess
149 , firstNonReservedProcessId
183 , firstNonReservedProcessId
150 , ImplicitReconnect(WithImplicitReconnect)
184 , ImplicitReconnect(WithImplicitReconnect)
152−import qualified Control.Distributed.Process.Internal.Trace.Remote as Trace
186+import Control.Distributed.Process.Management.Internal.Agent
189+import Control.Distributed.Process.Management.Internal.Types
192+import qualified Control.Distributed.Process.Management.Internal.Trace.Remote as Trace
153 ( remoteTable
193 ( remoteTable
155−import Control.Distributed.Process.Internal.Trace.Types
195+import Control.Distributed.Process.Management.Internal.Trace.Tracer
198+import Control.Distributed.Process.Management.Internal.Trace.Types
156 ( TraceArg(..)
199 ( TraceArg(..)
158 , traceEvent
200 , traceEvent
159 , traceLogFmt
201 , traceLogFmt
161−import Control.Distributed.Process.Internal.Trace.Tracer
165import Control.Distributed.Process.Serializable (Serializable)
204import Control.Distributed.Process.Serializable (Serializable)
166import Control.Distributed.Process.Internal.Messaging
205import Control.Distributed.Process.Internal.Messaging
167 ( sendBinary
206 ( sendBinary
169 , closeImplicitReconnections
207 , closeImplicitReconnections
170 , impliesDeathOf
208 , impliesDeathOf
172import Control.Distributed.Process.Internal.Primitives
210import Control.Distributed.Process.Internal.Primitives
173 ( register
211 ( register
175 , receiveWait
212 , receiveWait
177 , sendChan
214 , sendChan
179−import Control.Distributed.Process.Internal.Types (SendPort)
221+import Control.Distributed.Process.Internal.Types (SendPort, Tracer(..))
180import qualified Control.Distributed.Process.Internal.Closure.BuiltIn as BuiltIn (remoteTable)
222import qualified Control.Distributed.Process.Internal.Closure.BuiltIn as BuiltIn (remoteTable)
181import Control.Distributed.Process.Internal.WeakTQueue (TQueue, writeTQueue)
223import Control.Distributed.Process.Internal.WeakTQueue (TQueue, writeTQueue)
182import qualified Control.Distributed.Process.Internal.StrictContainerAccessors as DAC
224import qualified Control.Distributed.Process.Internal.StrictContainerAccessors as DAC
183 ( mapMaybe
225 ( mapMaybe
184 , mapDefault
226 , mapDefault
186−import System.Environment (getEnv)
228+import Control.Monad.Catch (try)
229+import GHC.IO (IO(..), unsafeUnmask)
230+import GHC.Base ( maskAsyncExceptions# )
187import Unsafe.Coerce
232import Unsafe.Coerce
235+-- Remove these definitions when the fix for
236+-- https://ghc.haskell.org/trac/ghc/ticket/10149
237+-- is included in all supported compilers:
238+block :: IO a -> IO a
239+block (IO io) = IO $ maskAsyncExceptions# io
240+unblock :: IO a -> IO a
241+unblock = unsafeUnmask
189--------------------------------------------------------------------------------
243--------------------------------------------------------------------------------
190-- Initialization --
244-- Initialization --
@@ -208,7 +262,7 @@ newLocalNode transport rtable = do
208createBareLocalNode :: NT.EndPoint -> RemoteTable -> IO LocalNode
262createBareLocalNode :: NT.EndPoint -> RemoteTable -> IO LocalNode
209createBareLocalNode endPoint rtable = do
263createBareLocalNode endPoint rtable = do
210 unq <- randomIO
264 unq <- randomIO
211− state <- newMVar LocalNodeState
265+ state <- newMVar $ LocalNodeValid $ ValidLocalNodeState
212 { _localProcesses = Map.empty
266 { _localProcesses = Map.empty
213 , _localPidCounter = firstNonReservedProcessId
267 , _localPidCounter = firstNonReservedProcessId
214 , _localPidUnique = unq
268 , _localPidUnique = unq
@@ -219,10 +273,10 @@ createBareLocalNode endPoint rtable = do
219 , localEndPoint = endPoint
273 , localEndPoint = endPoint
220 , localState = state
274 , localState = state
221 , localCtrlChan = ctrlChan
275 , localCtrlChan = ctrlChan
222− , localTracer = InactiveTracer
276+ , localEventBus = MxEventBusInitialising
223 , remoteTable = rtable
277 , remoteTable = rtable
225− tracedNode <- startTracing node
279+ tracedNode <- startMxAgent node
227 -- Once the NC terminates, the endpoint isn't much use,
281 -- Once the NC terminates, the endpoint isn't much use,
228 void $ forkIO $ Exception.finally (runNodeController tracedNode)
282 void $ forkIO $ Exception.finally (runNodeController tracedNode)
@@ -240,37 +294,27 @@ createBareLocalNode endPoint rtable = do
240 , ctrlMsgSignal = SigShutdown
294 , ctrlMsgSignal = SigShutdown
243−startTracing :: LocalNode -> IO LocalNode
297+startMxAgent :: LocalNode -> IO LocalNode
244−startTracing node =
298+startMxAgent node = do
245− -- TODO: use tryGetEnv once we drop support for GHC 7.2.x
299+ -- see note [tracer/forkProcess races]
246− Exception.catch
300+ let fork = forkProcess node
247− (getEnv "DISTRIBUTED_PROCESS_TRACE_ENABLED" >> procTracer node)
301+ mv <- MVar.newEmptyMVar
248− (\(_ :: IOError) -> return node)
302+ pid <- fork $ mxAgentController fork mv
249− where procTracer :: LocalNode -> IO LocalNode
303+ (tracer', wqRef, mxNew') <- MVar.takeMVar mv
250− procTracer n = do
304+ return node { localEventBus = (MxEventBus pid tracer' wqRef mxNew') }
251− -- see note [tracer/forkProcess races]
305+
252− mv <- MVar.newEmptyMVar
306+startDefaultTracer :: LocalNode -> IO ()
253− pid <- forkProcess n (traceController mv)
307+startDefaultTracer node' = do
254− wqRef <- MVar.takeMVar mv
308+ let t = localEventBus node'
255− return n { localTracer = (ActiveTracer pid wqRef) }
309+ case t of
256−
310+ MxEventBus _ (Tracer pid _) _ _ -> do
257−startDefaultTracer :: LocalNode -> IO (Maybe ProcessId)
311+ runProcess node' $ register "trace.controller" pid
258−startDefaultTracer node =
312+ pid' <- forkProcess node' defaultTracer
259− Exception.catch
313+ enableTrace (localEventBus node') pid'
260− (getEnv "DISTRIBUTED_PROCESS_TRACE_ENABLED" >> go node)
314+ runProcess node' $ register "tracer.initial" pid'
261− (\(_ :: IOError) -> return Nothing)
315+ _ -> return ()
262− where go :: LocalNode -> IO (Maybe ProcessId)
316+
263− go node' = do
317+-- TODO: we need a better mechanism for defining and registering services
264− registerTraceController node'
265− pid <- forkProcess node' defaultTracer
268− registerTraceController :: LocalNode -> IO ()
269− registerTraceController n =
270− let t = localTracer n in
272− InactiveTracer -> return ()
273− (ActiveTracer pid _) -> runProcess n $ register "trace.controller" pid
275-- | Start and register the service processes on a node
319-- | Start and register the service processes on a node
276startServiceProcesses :: LocalNode -> IO ()
320startServiceProcesses :: LocalNode -> IO ()
@@ -279,50 +323,71 @@ startServiceProcesses node = do
279 -- the default tracer first, even though it might @nsend@ to the logger
323 -- the default tracer first, even though it might @nsend@ to the logger
280 -- before /that/ process has started - this is a totally harmless race
324 -- before /that/ process has started - this is a totally harmless race
281 -- however, so we deliberably ignore it
325 -- however, so we deliberably ignore it
282− mPid <- startDefaultTracer node
326+ startDefaultTracer node
284− Nothing -> return ()
285− Just pid -> runProcess node $ register "tracer.initial" pid
287 logger <- forkProcess node loop
327 logger <- forkProcess node loop
288− runProcess node $ register "logger" logger
328+ runProcess node $ do
329+ register "logger" logger
330+ -- The trace.logger is used for tracing to the console to avoid feedback
331+ -- loops during tracing if the user reregisters the "logger" with a custom
332+ -- process which uses 'send' or other primitives which are traced.
333+ register "trace.logger" logger
290− loop = do
335+ loop = do
291− receiveWait
336+ receiveWait
292− [ match $ \((time, pid, string) ::(String, ProcessId, String)) -> do
337+ [ match $ \(SayMessage time pid string) -> do
293− liftIO . hPutStrLn stderr $ time ++ " " ++ show pid ++ ": " ++ string
338+ let time' = formatTime defaultTimeLocale "%c" time
294− loop
339+ liftIO . hPutStrLn stderr $ time' ++ " " ++ show pid ++ ": " ++ string
295− , match $ \((time, string) :: (String, String)) -> do
340+ loop
296− -- this is a 'trace' message from the local node tracer
341+ , match $ \((time, string) :: (String, String)) -> do
297− liftIO . hPutStrLn stderr $ time ++ " [trace] " ++ string
342+ -- this is a 'trace' message from the local node tracer
298− loop
343+ liftIO . hPutStrLn stderr $ time ++ " [trace] " ++ string
299− , match $ \(ch :: SendPort ()) -> -- a shutdown request
344+ loop
300− sendChan ch ()
345+ , match $ \(ch :: SendPort ()) -> -- a shutdown request
301− ]
346+ sendChan ch ()
303−-- | Force-close a local node
348+
304−--
349+-- | Force-close a local node, killing all processes on that node.
305−-- TODO: for now we just close the associated endpoint
306closeLocalNode :: LocalNode -> IO ()
350closeLocalNode :: LocalNode -> IO ()
307−closeLocalNode node =
351+closeLocalNode node = do
308− -- TODO: close all our processes, surely!?
352+ -- Kill processes after refilling the mvar. Otherwise, there is potential for
353+ -- deadlock as a dying process tries to get the mvar while masking exceptions
355+ join $ modifyMVar (localState node) $ \st -> case st of
356+ LocalNodeValid vst -> do
357+ return ( LocalNodeClosed
358+ , forM_ (vst ^. localProcesses) $ \lproc ->
359+ -- Semantics of 'throwTo' guarantee that target thread will get
360+ -- delivered an exception. Therefore, target thread will be
361+ -- killed eventually and that's as good as we can do. No need
362+ -- to wait for thread to actually finish dying.
363+ killThread (processThread lproc)
365+ LocalNodeClosed -> return (LocalNodeClosed, return ())
366+ -- This call will have the effect of shutting down the NC as well (see
367+ -- 'createBareLocalNode').
309 NT.closeEndPoint (localEndPoint node)
368 NT.closeEndPoint (localEndPoint node)
311-- | Run a process on a local node and wait for it to finish
370-- | Run a process on a local node and wait for it to finish
312runProcess :: LocalNode -> Process () -> IO ()
371runProcess :: LocalNode -> Process () -> IO ()
313runProcess node proc = do
372runProcess node proc = do
314 done <- newEmptyMVar
373 done <- newEmptyMVar
315− void $ forkProcess node (proc `finally` liftIO (putMVar done ()))
374+ -- TODO; When forkProcess inherits the masking state, protect the forked
316− takeMVar done
375+ -- thread against async exceptions that could occur before 'try' is evaluated.
376+ void $ forkProcess node $ try proc >>= liftIO . putMVar done
377+ takeMVar done >>= either (throwIO :: SomeException -> IO a) return
318-- | Spawn a new process on a local node
379-- | Spawn a new process on a local node
319forkProcess :: LocalNode -> Process () -> IO ProcessId
380forkProcess :: LocalNode -> Process () -> IO ProcessId
320−forkProcess node proc = modifyMVar (localState node) startProcess
381+forkProcess node proc = do
382+ ms <- getMaskingState
383+ modifyMVarMasked (localState node) (startProcess ms)
322− startProcess :: LocalNodeState -> IO (LocalNodeState, ProcessId)
385+ startProcess :: MaskingState
323− startProcess st = do
386+ -> LocalNodeState
324− let lpid = LocalProcessId { lpidCounter = st ^. localPidCounter
387+ -> IO (LocalNodeState, ProcessId)
325− , lpidUnique = st ^. localPidUnique
388+ startProcess ms (LocalNodeValid vst) = do
389+ let lpid = LocalProcessId { lpidCounter = vst ^. localPidCounter
390+ , lpidUnique = vst ^. localPidUnique
327 let pid = ProcessId { processNodeId = localNodeId node
392 let pid = ProcessId { processNodeId = localNodeId node
328 , processLocalId = lpid
393 , processLocalId = lpid
@@ -342,53 +407,78 @@ forkProcess node proc = modifyMVar (localState node) startProcess
342 , processThread = tid
407 , processThread = tid
343 , processNode = node
408 , processNode = node
345− tid' <- forkIO $ do
410+ -- Rewrite this code when this is fixed:
346− reason <- Exception.catch
411+ -- https://ghc.haskell.org/trac/ghc/ticket/10149
347− (runLocalProcess lproc proc >> return DiedNormal)
412+ let unmask = case ms of
348− (return . DiedException . (show :: SomeException -> String))
413+ Unmasked -> unblock
414+ MaskedInterruptible -> block
415+ MaskedUninterruptible -> id
416+ tid' <- uninterruptibleMask_ $ forkIO $ do
417+ reason <- Exception.catches
418+ (unmask $ runLocalProcess lproc proc >> return DiedNormal)
419+ [ (Exception.Handler (\ex@(ProcessExitException from msg) -> do
420+ mMsg <- unwrapMessage msg :: IO (Maybe String)
422+ Nothing -> return $ DiedException $ show ex
423+ Just m -> return $ DiedException ("exit-from=" ++ (show from) ++ ",reason=" ++ m)))
424+ , (Exception.Handler
425+ (return . DiedException . (show :: SomeException -> String)))]
350 -- [Unified: Table 4, rules termination and exiting]
427 -- [Unified: Table 4, rules termination and exiting]
351− modifyMVar_ (localState node) (cleanupProcess pid)
428+ mconns <- modifyValidLocalState node (cleanupProcess pid)
429+ -- XXX: Revisit after agreeing on the bigger picture for the semantics
430+ -- of transport operations.
431+ -- https://github.com/haskell-distributed/distributed-process/issues/204
432+ forM_ mconns $ forkIO . mapM_ NT.close
352 writeChan (localCtrlChan node) NCMsg
434 writeChan (localCtrlChan node) NCMsg
353 { ctrlMsgSender = ProcessIdentifier pid
435 { ctrlMsgSender = ProcessIdentifier pid
354 , ctrlMsgSignal = Died (ProcessIdentifier pid) reason
436 , ctrlMsgSignal = Died (ProcessIdentifier pid) reason
356 return (tid', lproc)
438 return (tid', lproc)
359 -- see note [tracer/forkProcess races]
440 -- see note [tracer/forkProcess races]
360− trace node (TraceEvSpawned pid)
441+ trace node (MxSpawned pid)
362 if lpidCounter lpid == maxBound
443 if lpidCounter lpid == maxBound
445+ -- TODO: this doesn't look right at all - how do we know
446+ -- that newUnique represents a process id that is available!?
364 newUnique <- randomIO
447 newUnique <- randomIO
365− return ( (localProcessWithId lpid ^= Just lproc)
448+ return ( LocalNodeValid
449+ $ (localProcessWithId lpid ^= Just lproc)
366 . (localPidCounter ^= firstNonReservedProcessId)
450 . (localPidCounter ^= firstNonReservedProcessId)
367 . (localPidUnique ^= newUnique)
451 . (localPidUnique ^= newUnique)
372− return ( (localProcessWithId lpid ^= Just lproc)
456+ return ( LocalNodeValid
457+ $ (localProcessWithId lpid ^= Just lproc)
373 . (localPidCounter ^: (+ 1))
458 . (localPidCounter ^: (+ 1))
462+ startProcess _ LocalNodeClosed =
463+ throwIO $ NodeClosedException $ localNodeId node
378− cleanupProcess :: ProcessId -> LocalNodeState -> IO LocalNodeState
465+ cleanupProcess :: ProcessId
379− cleanupProcess pid st = do
466+ -> ValidLocalNodeState
467+ -> IO (ValidLocalNodeState, [NT.Connection])
468+ cleanupProcess pid vst = do
380 let pid' = ProcessIdentifier pid
469 let pid' = ProcessIdentifier pid
381− let (affected, unaffected) = Map.partitionWithKey (\(fr, _to) !_v -> impliesDeathOf pid' fr) (st ^. localConnections)
470+ let (affected, unaffected) = Map.partitionWithKey (\(fr, _to) !_v -> impliesDeathOf pid' fr) (vst ^. localConnections)
382− mapM_ (NT.close . fst) (Map.elems affected)
471+ return ( (localProcessWithId (processLocalId pid) ^= Nothing)
383− return $ (localProcessWithId (processLocalId pid) ^= Nothing)
472+ . (localConnections ^= unaffected)
384− . (localConnections ^= unaffected)
473+ $ vst
385− $ st
474+ , map fst $ Map.elems affected
387-- note [tracer/forkProcess races]
477-- note [tracer/forkProcess races]
389-- Our startTracing function uses forkProcess to start the trace controller
479-- Our startTracing function uses forkProcess to start the trace controller
390-- process, and of course forkProcess attempts to call traceEvent once the
480-- process, and of course forkProcess attempts to call traceEvent once the
391−-- process has started. This is harmless, as the localTracer is not updated
481+-- process has started. This is harmless, as the localEventBus is not updated
392-- until /after/ the initial forkProcess completes, so the first call to
482-- until /after/ the initial forkProcess completes, so the first call to
393-- traceEvent behaves as if tracing were disabled (i.e., it is ignored).
483-- traceEvent behaves as if tracing were disabled (i.e., it is ignored).
@@ -402,7 +492,7 @@ type IncomingConnection = (NT.EndPointAddress, IncomingTarget)
402data IncomingTarget =
492data IncomingTarget =
404 | ToProc ProcessId (Weak (CQueue Message))
494 | ToProc ProcessId (Weak (CQueue Message))
405− | ToChan TypedChannel
495+ | ToChan SendPortId TypedChannel
408data ConnectionState = ConnectionState {
498data ConnectionState = ConnectionState {
@@ -429,6 +519,7 @@ incomingFrom addr = aux >>> DAC.mapDefault Set.empty addr
430handleIncomingMessages :: LocalNode -> IO ()
520handleIncomingMessages :: LocalNode -> IO ()
431handleIncomingMessages node = go initConnectionState
521handleIncomingMessages node = go initConnectionState
522+ `Exception.catch` \(NodeClosedException _) -> return ()
433 go :: ConnectionState -> IO ()
524 go :: ConnectionState -> IO ()
434 go !st = do
525 go !st = do
@@ -437,13 +528,14 @@ handleIncomingMessages node = go initConnectionState
437 NT.ConnectionOpened cid rel theirAddr ->
528 NT.ConnectionOpened cid rel theirAddr ->
438 if rel == NT.ReliableOrdered
529 if rel == NT.ReliableOrdered
440− trace node (TraceEvConnected cid)
531+ trace node (MxConnected cid theirAddr)
442 (incomingAt cid ^= Just (theirAddr, Uninit))
533 (incomingAt cid ^= Just (theirAddr, Uninit))
443 . (incomingFrom theirAddr ^: Set.insert cid)
534 . (incomingFrom theirAddr ^: Set.insert cid)
446− else invalidRequest cid st
537+ else invalidRequest cid st $
538+ "attempt to connect with unsupported reliability " ++ show rel
447 NT.Received cid payload ->
539 NT.Received cid payload ->
448 case st ^. incomingAt cid of
540 case st ^. incomingAt cid of
449 Just (_, ToProc pid weakQueue) -> do
541 Just (_, ToProc pid weakQueue) -> do
@@ -453,15 +545,19 @@ handleIncomingMessages node = go initConnectionState
453 -- it from the NC state? (and same for channels, below)
545 -- it from the NC state? (and same for channels, below)
454 let msg = payloadToMessage payload
546 let msg = payloadToMessage payload
455 enqueue queue msg -- 'enqueue' is strict
547 enqueue queue msg -- 'enqueue' is strict
456− trace node (TraceEvReceived pid msg)
548+ trace node (MxReceived pid msg)
458− Just (_, ToChan (TypedChannel chan')) -> do
550+ Just (_, ToChan chId (TypedChannel chan')) -> do
459 mChan <- deRefWeak chan'
551 mChan <- deRefWeak chan'
460 -- If mChan is Nothing, the process has given up the read end of
552 -- If mChan is Nothing, the process has given up the read end of
461 -- the channel and we simply ignore the incoming message
553 -- the channel and we simply ignore the incoming message
462− forM_ mChan $ \chan -> atomically $
554+ forM_ mChan $ \chan -> do
463− -- We make sure the message is fully decoded when it is enqueued
555+ msg' <- atomically $ do
464− writeTQueue chan $! decode (BSL.fromChunks payload)
556+ msg <- return $! decode (BSL.fromChunks payload)
557+ -- We make sure the message is fully decoded when it is enqueued
558+ writeTQueue chan msg
560+ trace node $ MxReceivedPort chId $ unsafeCreateUnencodedMessage msg'
466 Just (_, ToNode) -> do
562 Just (_, ToNode) -> do
467 let ctrlMsg = decode . BSL.fromChunks $ payload
563 let ctrlMsg = decode . BSL.fromChunks $ payload
@@ -471,38 +567,49 @@ handleIncomingMessages node = go initConnectionState
471 case decode (BSL.fromChunks payload) of
567 case decode (BSL.fromChunks payload) of
472 ProcessIdentifier pid -> do
568 ProcessIdentifier pid -> do
473 let lpid = processLocalId pid
569 let lpid = processLocalId pid
474− mProc <- withMVar state $ return . (^. localProcessWithId lpid)
570+ mProc <- withValidLocalState node $ return . (^. localProcessWithId lpid)
475 case mProc of
571 case mProc of
476 Just proc ->
572 Just proc ->
477 go (incomingAt cid ^= Just (src, ToProc pid (processWeakQ proc)) $ st)
573 go (incomingAt cid ^= Just (src, ToProc pid (processWeakQ proc)) $ st)
478 Nothing ->
574 Nothing ->
479− invalidRequest cid st
575+ -- incoming attempt to connect to unknown process - might
577+ go (incomingAt cid ^= Nothing $ st)
480 SendPortIdentifier chId -> do
578 SendPortIdentifier chId -> do
481 let lcid = sendPortLocalId chId
579 let lcid = sendPortLocalId chId
482 lpid = processLocalId (sendPortProcessId chId)
580 lpid = processLocalId (sendPortProcessId chId)
483− mProc <- withMVar state $ return . (^. localProcessWithId lpid)
581+ mProc <- withValidLocalState node $ return . (^. localProcessWithId lpid)
484 case mProc of
582 case mProc of
485 Just proc -> do
583 Just proc -> do
486 mChannel <- withMVar (processState proc) $ return . (^. typedChannelWithId lcid)
584 mChannel <- withMVar (processState proc) $ return . (^. typedChannelWithId lcid)
487 case mChannel of
585 case mChannel of
488 Just channel ->
586 Just channel ->
489− go (incomingAt cid ^= Just (src, ToChan channel) $ st)
587+ go (incomingAt cid ^= Just (src, ToChan chId channel) $ st)
490 Nothing ->
588 Nothing ->
491− invalidRequest cid st
589+ invalidRequest cid st $
590+ "incoming attempt to connect to unknown channel of"
591+ ++ " process " ++ show (sendPortProcessId chId)
492 Nothing ->
592 Nothing ->
493− invalidRequest cid st
593+ -- incoming attempt to connect to channel of unknown
594+ -- process - might be dead already
595+ go (incomingAt cid ^= Nothing $ st)
494 NodeIdentifier nid ->
596 NodeIdentifier nid ->
495 if nid == localNodeId node
597 if nid == localNodeId node
496 then go (incomingAt cid ^= Just (src, ToNode) $ st)
598 then go (incomingAt cid ^= Just (src, ToNode) $ st)
497− else invalidRequest cid st
599+ else invalidRequest cid st $
600+ "incoming attempt to connect to a different node -"
601+ ++ " I'm " ++ show (localNodeId node)
602+ ++ " but the remote peer wants to connect to "
498 Nothing ->
604 Nothing ->
499 invalidRequest cid st
605 invalidRequest cid st
606+ "message received from an unknown connection"
500 NT.ConnectionClosed cid ->
607 NT.ConnectionClosed cid ->
501 case st ^. incomingAt cid of
608 case st ^. incomingAt cid of
502 Nothing ->
609 Nothing ->
503− invalidRequest cid st
610+ invalidRequest cid st "closed unknown connection"
504 Just (src, _) -> do
611 Just (src, _) -> do
505− trace node (TraceEvDisconnected cid)
612+ trace node (MxDisconnected cid src)
506 go ( (incomingAt cid ^= Nothing)
613 go ( (incomingAt cid ^= Nothing)
507 . (incomingFrom src ^: Set.delete cid)
614 . (incomingFrom src ^: Set.delete cid)
@@ -531,19 +638,20 @@ handleIncomingMessages node = go initConnectionState
531 -- and we just give up
638 -- and we just give up
532 fail "Cloud Haskell fatal error: received unexpected multicast"
639 fail "Cloud Haskell fatal error: received unexpected multicast"
534− invalidRequest :: NT.ConnectionId -> ConnectionState -> IO ()
641+ invalidRequest :: NT.ConnectionId -> ConnectionState -> String -> IO ()
535− invalidRequest cid st = do
642+ invalidRequest cid st msg = do
536 -- TODO: We should treat this as a fatal error on the part of the remote
643 -- TODO: We should treat this as a fatal error on the part of the remote
537 -- node. That is, we should report the remote node as having died, and we
644 -- node. That is, we should report the remote node as having died, and we
538 -- should close incoming connections (this requires a Transport layer
645 -- should close incoming connections (this requires a Transport layer
539 -- extension).
646 -- extension).
540− traceEventFmtIO node "" [(TraceStr " [network] invalid request: "),
647+ traceEventFmtIO node "" [ TraceStr $ " [network] invalid request"
541− (Trace cid)]
648+ ++ " (" ++ msg ++ "): "
542 go ( incomingAt cid ^= Nothing
651 go ( incomingAt cid ^= Nothing
546− state = localState node
547 endpoint = localEndPoint node
655 endpoint = localEndPoint node
548 ctrlChan = localCtrlChan node
656 ctrlChan = localCtrlChan node
@@ -552,8 +660,9 @@ handleIncomingMessages node = go initConnectionState
552--------------------------------------------------------------------------------
660--------------------------------------------------------------------------------
554runNodeController :: LocalNode -> IO ()
662runNodeController :: LocalNode -> IO ()
555−runNodeController =
663+runNodeController node =
556− runReaderT (evalStateT (unNC nodeController) initNCState)
664+ runReaderT (evalStateT (unNC nodeController) initNCState) node
665+ `Exception.catch` \(NodeClosedException _) -> return ()
558--------------------------------------------------------------------------------
667--------------------------------------------------------------------------------
559-- Internal data types --
668-- Internal data types --
@@ -561,20 +670,26 @@ runNodeController =
562data NCState = NCState
671data NCState = NCState
563 { -- Mapping from remote processes to linked local processes
672 { -- Mapping from remote processes to linked local processes
564− _links :: !(Map Identifier (Set ProcessId))
673+ _links :: !(BiMultiMap Identifier ProcessId ())
565 -- Mapping from remote processes to monitoring local processes
674 -- Mapping from remote processes to monitoring local processes
566− , _monitors :: !(Map Identifier (Set (ProcessId, MonitorRef)))
675+ , _monitors :: !(BiMultiMap Identifier ProcessId MonitorRef)
567 -- Process registry: names and where they live, mapped to the PIDs
676 -- Process registry: names and where they live, mapped to the PIDs
568 , _registeredHere :: !(Map String ProcessId)
677 , _registeredHere :: !(Map String ProcessId)
569 , _registeredOnNodes :: !(Map ProcessId [(NodeId,Int)])
678 , _registeredOnNodes :: !(Map ProcessId [(NodeId,Int)])
572newtype NC a = NC { unNC :: StateT NCState (ReaderT LocalNode IO) a }
681newtype NC a = NC { unNC :: StateT NCState (ReaderT LocalNode IO) a }
573− deriving (Functor, Monad, MonadIO, MonadState NCState, MonadReader LocalNode)
682+ deriving ( Applicative
686+ , MonadState NCState
687+ , MonadReader LocalNode
575initNCState :: NCState
690initNCState :: NCState
576−initNCState = NCState { _links = Map.empty
691+initNCState = NCState { _links = BiMultiMap.empty
577− , _monitors = Map.empty
692+ , _monitors = BiMultiMap.empty
578 , _registeredHere = Map.empty
693 , _registeredHere = Map.empty
579 , _registeredOnNodes = Map.empty
694 , _registeredOnNodes = Map.empty
@@ -590,19 +705,46 @@ instance Show ProcessKillException where
590 show (ProcessKillException pid reason) =
705 show (ProcessKillException pid reason) =
591 "killed-by=" ++ show pid ++ ",reason=" ++ reason
706 "killed-by=" ++ show pid ++ ",reason=" ++ reason
708+ncSendToProcess :: ProcessId -> Message -> NC ()
709+ncSendToProcess = ncSendToProcessAndTrace True
711+ncSendToProcessAndTrace :: Bool -> ProcessId -> Message -> NC ()
712+ncSendToProcessAndTrace shouldTrace pid msg = do
714+ if processNodeId pid == localNodeId node
715+ then ncEffectLocalSendAndTrace shouldTrace node pid msg
716+ else liftIO $ sendBinary node
717+ (NodeIdentifier $ localNodeId node)
718+ (NodeIdentifier $ processNodeId pid)
719+ WithImplicitReconnect
720+ NCMsg { ctrlMsgSender = NodeIdentifier $ localNodeId node
721+ , ctrlMsgSignal = UnreliableSend (processLocalId pid) msg
724+ncSendToNode :: NodeId -> NCMsg -> NC ()
725+ncSendToNode to msg = do
727+ liftIO $ if to == localNodeId node
728+ then writeChan (localCtrlChan node) $! msg
729+ else sendBinary node
730+ (NodeIdentifier $ localNodeId node)
732+ WithImplicitReconnect
593--------------------------------------------------------------------------------
735--------------------------------------------------------------------------------
594-- Tracing/Debugging --
736-- Tracing/Debugging --
595--------------------------------------------------------------------------------
737--------------------------------------------------------------------------------
597−-- [Issue #104]
739+-- [Issue #104 / DP-13]
599traceNotifyDied :: LocalNode -> Identifier -> DiedReason -> NC ()
741traceNotifyDied :: LocalNode -> Identifier -> DiedReason -> NC ()
600traceNotifyDied node ident reason =
742traceNotifyDied node ident reason =
601 -- TODO: sendPortDied notifications
743 -- TODO: sendPortDied notifications
602 liftIO $ withLocalTracer node $ \t ->
744 liftIO $ withLocalTracer node $ \t ->
603 case ident of
745 case ident of
604− (NodeIdentifier nid) -> traceEvent t (TraceEvNodeDied nid reason)
746+ (NodeIdentifier nid) -> traceEvent t (MxNodeDied nid reason)
605− (ProcessIdentifier pid) -> traceEvent t (TraceEvDied pid reason)
747+ (ProcessIdentifier pid) -> traceEvent t (MxProcessDied pid reason)
606 _ -> return ()
748 _ -> return ()
608traceEventFmtIO :: LocalNode
750traceEventFmtIO :: LocalNode
@@ -612,11 +754,11 @@ traceEventFmtIO :: LocalNode
612traceEventFmtIO node fmt args =
754traceEventFmtIO node fmt args =
613 withLocalTracer node $ \t -> traceLogFmt t fmt args
755 withLocalTracer node $ \t -> traceLogFmt t fmt args
615−trace :: LocalNode -> TraceEvent -> IO ()
757+trace :: LocalNode -> MxEvent -> IO ()
616trace node ev = withLocalTracer node $ \t -> traceEvent t ev
758trace node ev = withLocalTracer node $ \t -> traceEvent t ev
618−withLocalTracer :: LocalNode -> (Tracer -> IO ()) -> IO ()
760+withLocalTracer :: LocalNode -> (MxEventBus -> IO ()) -> IO ()
619−withLocalTracer node act = act (localTracer node)
761+withLocalTracer node act = act (localEventBus node)
621--------------------------------------------------------------------------------
763--------------------------------------------------------------------------------
622-- Core functionality --
764-- Core functionality --
@@ -632,11 +774,7 @@ nodeController = do
632 -- [Unified: Table 7, rule nc_forward]
774 -- [Unified: Table 7, rule nc_forward]
633 case destNid (ctrlMsgSignal msg) of
775 case destNid (ctrlMsgSignal msg) of
634 Just nid' | nid' /= localNodeId node ->
776 Just nid' | nid' /= localNodeId node ->
635− liftIO $ sendBinary node
777+ ncSendToNode nid' msg
637− (NodeIdentifier nid')
638− WithImplicitReconnect
641 return ()
779 return ()
@@ -659,6 +797,8 @@ nodeController = do
659 ncEffectWhereIs from label
797 ncEffectWhereIs from label
660 NCMsg _ (NamedSend label msg') ->
798 NCMsg _ (NamedSend label msg') ->
661 ncEffectNamedSend label msg'
799 ncEffectNamedSend label msg'
800+ NCMsg _ (UnreliableSend lpid msg') ->
801+ ncEffectLocalSend node (ProcessId (localNodeId node) lpid) msg'
662 NCMsg _ (LocalSend to msg') ->
802 NCMsg _ (LocalSend to msg') ->
663 ncEffectLocalSend node to msg'
803 ncEffectLocalSend node to msg'
664 NCMsg _ (LocalPortSend to msg') ->
804 NCMsg _ (LocalPortSend to msg') ->
@@ -672,8 +812,9 @@ nodeController = do
672 NCMsg _ SigShutdown ->
812 NCMsg _ SigShutdown ->
673 liftIO $ do
813 liftIO $ do
674 NT.closeEndPoint (localEndPoint node)
814 NT.closeEndPoint (localEndPoint node)
675− `Exception.finally` throwIO ThreadKilled
815+ `Exception.finally` throwIO (NodeClosedException $ localNodeId node)
676− -- ThreadKilled seems to make more sense than fail/error here
816+ NCMsg (ProcessIdentifier from) (GetNodeStats nid) ->
817+ ncEffectGetNodeStats from nid
677 unexpected ->
818 unexpected ->
678 error $ "nodeController: unexpected message " ++ show unexpected
819 error $ "nodeController: unexpected message " ++ show unexpected
@@ -691,22 +832,18 @@ ncEffectMonitor from them mRef = do
691 case (shouldLink, isLocal node (ProcessIdentifier from)) of
832 case (shouldLink, isLocal node (ProcessIdentifier from)) of
692 (True, _) -> -- [Unified: first rule]
833 (True, _) -> -- [Unified: first rule]
693 case mRef of
834 case mRef of
694− Just ref -> modify' $ monitorsFor them ^: Set.insert (from, ref)
835+ Just ref -> modify' $ monitors ^: BiMultiMap.insert them from ref
695− Nothing -> modify' $ linksFor them ^: Set.insert from
836+ Nothing -> modify' $ links ^: BiMultiMap.insert them from ()
696 (False, True) -> -- [Unified: second rule]
837 (False, True) -> -- [Unified: second rule]
697 notifyDied from them DiedUnknownId mRef
838 notifyDied from them DiedUnknownId mRef
698 (False, False) -> -- [Unified: third rule]
839 (False, False) -> -- [Unified: third rule]
699 -- TODO: this is the right sender according to the Unified semantics,
840 -- TODO: this is the right sender according to the Unified semantics,
700 -- but perhaps having 'them' as the sender would make more sense
841 -- but perhaps having 'them' as the sender would make more sense
701 -- (see also: notifyDied)
842 -- (see also: notifyDied)
702− liftIO $ sendBinary node
843+ ncSendToNode (processNodeId from) $ NCMsg
703− (NodeIdentifier $ localNodeId node)
844+ { ctrlMsgSender = NodeIdentifier (localNodeId node)
704− (NodeIdentifier $ processNodeId from)
845+ , ctrlMsgSignal = Died them DiedUnknownId
705− WithImplicitReconnect
846+ }
707− { ctrlMsgSender = NodeIdentifier (localNodeId node)
708− , ctrlMsgSignal = Died them DiedUnknownId
711-- [Unified: Table 11]
848-- [Unified: Table 11]
712ncEffectUnlink :: ProcessId -> Identifier -> NC ()
849ncEffectUnlink :: ProcessId -> Identifier -> NC ()
@@ -720,7 +857,7 @@ ncEffectUnlink from them = do
720 postAsMessage from $ DidUnlinkNode nid
857 postAsMessage from $ DidUnlinkNode nid
721 SendPortIdentifier cid ->
858 SendPortIdentifier cid ->
722 postAsMessage from $ DidUnlinkPort cid
859 postAsMessage from $ DidUnlinkPort cid
723− modify' $ linksFor them ^: Set.delete from
860+ modify' $ links ^: BiMultiMap.delete them from ()
725-- [Unified: Table 11]
862-- [Unified: Table 11]
726ncEffectUnmonitor :: ProcessId -> MonitorRef -> NC ()
863ncEffectUnmonitor :: ProcessId -> MonitorRef -> NC ()
@@ -728,7 +865,7 @@ ncEffectUnmonitor from ref = do
728 node <- ask
865 node <- ask
729 when (isLocal node (ProcessIdentifier from)) $
866 when (isLocal node (ProcessIdentifier from)) $
730 postAsMessage from $ DidUnmonitor ref
867 postAsMessage from $ DidUnmonitor ref
731− modify' $ monitorsFor (monitorRefIdent ref) ^: Set.delete (from, ref)
868+ modify' $ monitors ^: BiMultiMap.delete (monitorRefIdent ref) from ref
733-- [Unified: Table 12]
870-- [Unified: Table 12]
734ncEffectDied :: Identifier -> DiedReason -> NC ()
871ncEffectDied :: Identifier -> DiedReason -> NC ()
@@ -743,7 +880,7 @@ ncEffectDied ident reason = do
743 let localOnly = case ident of NodeIdentifier _ -> True ; _ -> False
880 let localOnly = case ident of NodeIdentifier _ -> True ; _ -> False
745 forM_ (Map.toList affectedLinks) $ \(them, uss) ->
882 forM_ (Map.toList affectedLinks) $ \(them, uss) ->
746− forM_ uss $ \us ->
883+ forM_ uss $ \(us, _) ->
747 when (localOnly <= isLocal node (ProcessIdentifier us)) $
884 when (localOnly <= isLocal node (ProcessIdentifier us)) $
748 notifyDied us them reason Nothing
885 notifyDied us them reason Nothing
@@ -752,9 +889,35 @@ ncEffectDied ident reason = do
752 when (localOnly <= isLocal node (ProcessIdentifier us)) $
889 when (localOnly <= isLocal node (ProcessIdentifier us)) $
753 notifyDied us them reason (Just ref)
890 notifyDied us them reason (Just ref)
755− modify' $ (links ^= unaffectedLinks) . (monitors ^= unaffectedMons)
892+ -- Notify remote nodes that the process died so it can be removed from monitor
757− modify' $ registeredHere ^: Map.filter (\pid -> not $ ident `impliesDeathOf` ProcessIdentifier pid)
894+ mapM_ (forwardDeath node) $
895+ [ nid | ProcessIdentifier pid <- [ident]
896+ , i <- Set.toList $ Set.union
897+ (Set.map fst $ BiMultiMap.lookupBy2nd pid unaffectedLinks)
898+ (Set.map fst $ BiMultiMap.lookupBy2nd pid unaffectedMons)
899+ , let nid = nodeOf i
900+ , nid /= localNodeId node
903+ -- Delete monitors in the local node.
904+ let deleteDeads :: (Ord a, Ord v)
905+ => BiMultiMap a ProcessId v -> BiMultiMap a ProcessId v
906+ deleteDeads = case ident of
907+ -- deleteAllBy2nd is faster than partitionWithKeyBy2nd
908+ ProcessIdentifier pid -> BiMultiMap.deleteAllBy2nd pid
909+ _ -> snd . BiMultiMap.partitionWithKeyBy2nd
910+ (\pid _ -> ident `impliesDeathOf` ProcessIdentifier pid)
911+ unaffectedLinks' = deleteDeads unaffectedLinks
912+ unaffectedMons' = deleteDeads unaffectedMons
914+ modify' $ (links ^= unaffectedLinks') . (monitors ^= unaffectedMons')
916+ -- we now consider all labels for this identifier unregistered
917+ let toDrop pid = not $ ident `impliesDeathOf` ProcessIdentifier pid
918+ (keepNames, dropNames) <- Map.partition toDrop <$> gets (^. registeredHere)
919+ mapM_ (\(p, l) -> liftIO $ trace node (MxUnRegistered l p)) (Map.toList dropNames)
920+ modify' $ registeredHere ^= keepNames
759 remaining <- fmap Map.toList (gets (^. registeredOnNodes)) >>=
922 remaining <- fmap Map.toList (gets (^. registeredOnNodes)) >>=
760 mapM (\(pid,nidlist) ->
923 mapM (\(pid,nidlist) ->
@@ -762,20 +925,15 @@ ncEffectDied ident reason = do
763 do forM_ nidlist $ \(nid,_) ->
926 do forM_ nidlist $ \(nid,_) ->
764 when (not $ isLocal node (NodeIdentifier nid))
927 when (not $ isLocal node (NodeIdentifier nid))
765− (forwardNameDeath node nid)
928+ (forwardDeath node nid)
766 return Nothing
929 return Nothing
767 False -> return $ Just (pid,nidlist) )
930 False -> return $ Just (pid,nidlist) )
768 modify' $ registeredOnNodes ^= (Map.fromList (catMaybes remaining))
931 modify' $ registeredOnNodes ^= (Map.fromList (catMaybes remaining))
770− forwardNameDeath node nid =
933+ forwardDeath node nid = ncSendToNode nid
771− liftIO $ sendBinary node
934+ NCMsg { ctrlMsgSender = NodeIdentifier (localNodeId node)
772− (NodeIdentifier $ localNodeId node)
935+ , ctrlMsgSignal = Died ident reason
773− (NodeIdentifier $ nid)
936+ }
774− WithImplicitReconnect
776− { ctrlMsgSender = NodeIdentifier (localNodeId node)
777− , ctrlMsgSignal = Died ident reason
780-- [Unified: Table 13]
938-- [Unified: Table 13]
781ncEffectSpawn :: ProcessId -> Closure (Process ()) -> SpawnRef -> NC ()
939ncEffectSpawn :: ProcessId -> Closure (Process ()) -> SpawnRef -> NC ()
@@ -783,17 +941,13 @@ ncEffectSpawn pid cProc ref = do
783 mProc <- unClosure cProc
941 mProc <- unClosure cProc
784 -- If the closure does not exist, we spawn a process that throws an exception
942 -- If the closure does not exist, we spawn a process that throws an exception
785 -- This allows the remote node to find out what's happening
943 -- This allows the remote node to find out what's happening
786− -- TODO:
944+ -- TODO:
787 let proc = case mProc of
945 let proc = case mProc of
788 Left err -> fail $ "Error: Could not resolve closure: " ++ err
946 Left err -> fail $ "Error: Could not resolve closure: " ++ err
789 Right p -> p
947 Right p -> p
790 node <- ask
948 node <- ask
791 pid' <- liftIO $ forkProcess node proc
949 pid' <- liftIO $ forkProcess node proc
792− liftIO $ sendMessage node
950+ ncSendToProcess pid $ unsafeCreateUnencodedMessage $ DidSpawn ref pid'
793− (NodeIdentifier (localNodeId node))
794− (ProcessIdentifier pid)
795− WithImplicitReconnect
798-- Unified semantics does not explicitly describe how to implement 'register',
952-- Unified semantics does not explicitly describe how to implement 'register',
799-- but mentions it's "very similar to nsend" (Table 14)
953-- but mentions it's "very similar to nsend" (Table 14)
@@ -811,17 +965,19 @@ ncEffectRegister from label atnode mPid reregistration = do
811 return $ (isNothing currentVal /= reregistration) &&
965 return $ (isNothing currentVal /= reregistration) &&
812 (not (isLocal node (ProcessIdentifier thepid) ) || isvalidlocal )
966 (not (isLocal node (ProcessIdentifier thepid) ) || isvalidlocal )
813 if isLocal node (NodeIdentifier atnode)
967 if isLocal node (NodeIdentifier atnode)
814− then do when (isOk) $
968+ then do when isOk $
815 do modify' $ registeredHereFor label ^= mPid
969 do modify' $ registeredHereFor label ^= mPid
816 updateRemote node currentVal mPid
970 updateRemote node currentVal mPid
817 case mPid of
971 case mPid of
818− (Just p) -> liftIO $ trace node (TraceEvRegistered p label)
972+ (Just p) -> do
819− Nothing -> liftIO $ trace node (TraceEvUnRegistered (fromJust currentVal) label)
973+ if reregistration
820− liftIO $ sendMessage node
974+ then liftIO $ trace node (MxUnRegistered (fromJust currentVal) label)
821− (NodeIdentifier (localNodeId node))
975+ else return ()
822− (ProcessIdentifier from)
976+ liftIO $ trace node (MxRegistered p label)
823− WithImplicitReconnect
977+ Nothing -> liftIO $ trace node (MxUnRegistered (fromJust currentVal) label)
824− (RegisterReply label isOk)
978+ newVal <- gets (^. registeredHereFor label)
979+ ncSendToProcess from $ unsafeCreateUnencodedMessage $
980+ RegisterReply label isOk newVal
825 else let operation =
981 else let operation =
826 case reregistration of
982 case reregistration of
827 True -> flip decList
983 True -> flip decList
@@ -851,40 +1007,36 @@ ncEffectRegister from label atnode mPid reregistration = do
851 decList (x:xs) tag = x:decList xs tag
1007 decList (x:xs) tag = x:decList xs tag
852 forward node to reg =
1008 forward node to reg =
853 when (not $ isLocal node (NodeIdentifier to)) $
1009 when (not $ isLocal node (NodeIdentifier to)) $
854− liftIO $ sendBinary node
1010+ ncSendToNode to $ NCMsg { ctrlMsgSender = ProcessIdentifier from
855− (ProcessIdentifier from)
1011+ , ctrlMsgSignal = reg
856− (NodeIdentifier to)
1012+ }
857− WithImplicitReconnect
859− { ctrlMsgSender = ProcessIdentifier from
860− , ctrlMsgSignal = reg
864-- Unified semantics does not explicitly describe 'whereis'
1015-- Unified semantics does not explicitly describe 'whereis'
865ncEffectWhereIs :: ProcessId -> String -> NC ()
1016ncEffectWhereIs :: ProcessId -> String -> NC ()
866ncEffectWhereIs from label = do
1017ncEffectWhereIs from label = do
868 mPid <- gets (^. registeredHereFor label)
1018 mPid <- gets (^. registeredHereFor label)
869− liftIO $ sendMessage node
1019+ ncSendToProcess from $ unsafeCreateUnencodedMessage $ WhereIsReply label mPid
870− (NodeIdentifier (localNodeId node))
871− (ProcessIdentifier from)
872− WithImplicitReconnect
873− (WhereIsReply label mPid)
875-- [Unified: Table 14]
1021-- [Unified: Table 14]
876ncEffectNamedSend :: String -> Message -> NC ()
1022ncEffectNamedSend :: String -> Message -> NC ()
877ncEffectNamedSend label msg = do
1023ncEffectNamedSend label msg = do
878 mPid <- gets (^. registeredHereFor label)
1024 mPid <- gets (^. registeredHereFor label)
879 -- If mPid is Nothing, we just ignore the named send (as per Table 14)
1025 -- If mPid is Nothing, we just ignore the named send (as per Table 14)
880− forM_ mPid $ \pid -> postMessage pid msg
1026+ forM_ mPid $ \to ->
1027+ -- If this is a trace message we don't trace it to avoid entering a loop
1028+ -- where trace messages produce more trace messages.
1029+ ncSendToProcessAndTrace (label /= "trace.logger") to msg
882-- [Issue #DP-20]
1031-- [Issue #DP-20]
883ncEffectLocalSend :: LocalNode -> ProcessId -> Message -> NC ()
1032ncEffectLocalSend :: LocalNode -> ProcessId -> Message -> NC ()
884−ncEffectLocalSend node to msg =
1033+ncEffectLocalSend = ncEffectLocalSendAndTrace True
1035+ncEffectLocalSendAndTrace :: Bool -> LocalNode -> ProcessId -> Message -> NC ()
1036+ncEffectLocalSendAndTrace shouldTrace node to msg =
885 liftIO $ withLocalProc node to $ \p -> do
1037 liftIO $ withLocalProc node to $ \p -> do
886 enqueue (processQueue p) msg
1038 enqueue (processQueue p) msg
887− trace node (TraceEvReceived to msg)
1039+ when shouldTrace $ trace node (MxReceived to msg)
889-- [Issue #DP-20]
1041-- [Issue #DP-20]
890ncEffectLocalPortSend :: SendPortId -> Message -> NC ()
1042ncEffectLocalPortSend :: SendPortId -> Message -> NC ()
@@ -902,11 +1054,12 @@ ncEffectLocalPortSend from msg = do
902 -- If ch is Nothing, the process has given up the read end of
1054 -- If ch is Nothing, the process has given up the read end of
903 -- the channel and we simply ignore the incoming message - this
1055 -- the channel and we simply ignore the incoming message - this
904 ch <- deRefWeak chan'
1056 ch <- deRefWeak chan'
905− forM_ ch $ \chan -> deliverChan msg chan
1057+ forM_ ch $ \chan -> deliverChan node from msg chan
906− where deliverChan :: forall a . Message -> TQueue a -> IO ()
1058+ where deliverChan :: forall a . LocalNode -> SendPortId -> Message -> TQueue a -> IO ()
907− deliverChan (UnencodedMessage _ raw) chan' =
1059+ deliverChan n p (UnencodedMessage _ raw) chan' = do
908 atomically $ writeTQueue chan' ((unsafeCoerce raw) :: a)
1060 atomically $ writeTQueue chan' ((unsafeCoerce raw) :: a)
909− deliverChan (EncodedMessage _ _) _ =
1061+ trace n (MxReceivedPort p $ unsafeCreateUnencodedMessage raw)
1062+ deliverChan _ _ (EncodedMessage _ _) _ =
910 -- this will not happen unless someone screws with Primitives.hs
1063 -- this will not happen unless someone screws with Primitives.hs
911 error "invalid local channel delivery"
1064 error "invalid local channel delivery"
@@ -931,46 +1084,56 @@ ncEffectGetInfo from pid =
931 them = (ProcessIdentifier pid)
1084 them = (ProcessIdentifier pid)
933 node <- ask
1086 node <- ask
934− mProc <- liftIO $
1087+ mProc <- liftIO $ withValidLocalState node
935− withMVar (localState node) $ return . (^. localProcessWithId lpid)
1088+ $ return . (^. localProcessWithId lpid)
936 case mProc of
1089 case mProc of
937 Nothing -> dispatch (isLocal node (ProcessIdentifier from))
1090 Nothing -> dispatch (isLocal node (ProcessIdentifier from))
938− from node (ProcessInfoNone DiedUnknownId)
1091+ from (ProcessInfoNone DiedUnknownId)
939− Just _ -> do
1092+ Just proc -> do
940− itsLinks <- gets (^. linksFor them)
1093+ itsLinks <- Set.map fst . BiMultiMap.lookupBy1st them <$>
941− itsMons <- gets (^. monitorsFor them)
1094+ gets (^. links)
1095+ itsMons <- BiMultiMap.lookupBy1st them <$> gets (^. monitors)
942 registered <- gets (^. registeredHere)
1096 registered <- gets (^. registeredHere)
1097+ size <- liftIO $ queueSize $ processQueue $ proc
944 let reg = registeredNames registered
1099 let reg = registeredNames registered
945 dispatch (isLocal node (ProcessIdentifier from))
1100 dispatch (isLocal node (ProcessIdentifier from))
948 ProcessInfo {
1102 ProcessInfo {
949 infoNode = (processNodeId pid)
1103 infoNode = (processNodeId pid)
950 , infoRegisteredNames = reg
1104 , infoRegisteredNames = reg
951− -- we cannot populate this field
1105+ , infoMessageQueueLength = size
952− , infoMessageQueueLength = Nothing
1106+ , infoMonitors = Set.toList itsMons
953− , infoMonitors = Set.toList itsMons
1107+ , infoLinks = Set.toList itsLinks
954− , infoLinks = Set.toList itsLinks
956− where dispatch :: (Serializable a, Show a)
1109+ where dispatch :: (Serializable a)
958 -> ProcessId
1111 -> ProcessId
961 -> NC ()
1113 -> NC ()
962− dispatch True dest _ pInfo = postAsMessage dest $ pInfo
1114+ dispatch True dest pInfo = postAsMessage dest $ pInfo
963− dispatch False dest node pInfo = do
1115+ dispatch False dest pInfo =
964− liftIO $ sendMessage node
1116+ ncSendToProcess dest $ unsafeCreateUnencodedMessage pInfo
965− (NodeIdentifier (localNodeId node))
966− (ProcessIdentifier dest)
967− WithImplicitReconnect
970 registeredNames = Map.foldlWithKey (\ks k v -> if v == pid
1118 registeredNames = Map.foldlWithKey (\ks k v -> if v == pid
971 then (k:ks)
1119 then (k:ks)
972 else ks) []
1120 else ks) []
1122+ncEffectGetNodeStats :: ProcessId -> NodeId -> NC ()
1123+ncEffectGetNodeStats from _nid = do
1125+ ncState <- StateT.get
1126+ nodeState <- liftIO $ withValidLocalState node return
1129+ nodeStatsNode = localNodeId node
1130+ , nodeStatsRegisteredNames = Map.size $ ncState ^. registeredHere
1131+ , nodeStatsMonitors = BiMultiMap.size $ ncState ^. monitors
1132+ , nodeStatsLinks = BiMultiMap.size $ ncState ^. links
1133+ , nodeStatsProcesses = Map.size (nodeState ^. localProcesses)
1135+ postAsMessage from stats
974--------------------------------------------------------------------------------
1137--------------------------------------------------------------------------------
975-- Auxiliary --
1138-- Auxiliary --
976--------------------------------------------------------------------------------
1139--------------------------------------------------------------------------------
@@ -997,14 +1160,10 @@ notifyDied dest src reason mRef = do
997 throwException dest $ PortLinkException pid reason
1160 throwException dest $ PortLinkException pid reason
998 (False, _, _) ->
1161 (False, _, _) ->
999 -- The change in sender comes from [Unified: Table 10]
1162 -- The change in sender comes from [Unified: Table 10]
1000− liftIO $ sendBinary node
1163+ ncSendToNode (processNodeId dest) $ NCMsg
1001− (NodeIdentifier $ localNodeId node)
1164+ { ctrlMsgSender = NodeIdentifier (localNodeId node)
1002− (NodeIdentifier $ processNodeId dest)
1165+ , ctrlMsgSignal = Died src reason
1003− WithImplicitReconnect
1166+ }
1005− { ctrlMsgSender = NodeIdentifier (localNodeId node)
1006− , ctrlMsgSignal = Died src reason
1009-- | [Unified: Table 8]
1168-- | [Unified: Table 8]
1010destNid :: ProcessSignal -> Maybe NodeId
1169destNid :: ProcessSignal -> Maybe NodeId
@@ -1016,6 +1175,7 @@ destNid (Spawn _ _) = Nothing
1016destNid (Register _ _ _ _) = Nothing
1175destNid (Register _ _ _ _) = Nothing
1017destNid (WhereIs _) = Nothing
1176destNid (WhereIs _) = Nothing
1018destNid (NamedSend _ _) = Nothing
1177destNid (NamedSend _ _) = Nothing
1178+destNid (UnreliableSend _ _) = Nothing
1019-- We don't need to forward 'Died' signals; if monitoring/linking is setup,
1179-- We don't need to forward 'Died' signals; if monitoring/linking is setup,
1020-- then when a local process dies the monitoring/linking machinery will take
1180-- then when a local process dies the monitoring/linking machinery will take
1021-- care of notifying remote nodes
1181-- care of notifying remote nodes
@@ -1023,6 +1183,7 @@ destNid (Died _ _) = Nothing
1023destNid (Kill pid _) = Just $ processNodeId pid
1183destNid (Kill pid _) = Just $ processNodeId pid
1024destNid (Exit pid _) = Just $ processNodeId pid
1184destNid (Exit pid _) = Just $ processNodeId pid
1025destNid (GetInfo pid) = Just $ processNodeId pid
1185destNid (GetInfo pid) = Just $ processNodeId pid
1186+destNid (GetNodeStats nid) = Just nid
1026destNid (LocalSend pid _) = Just $ processNodeId pid
1187destNid (LocalSend pid _) = Just $ processNodeId pid
1027destNid (LocalPortSend cid _) = Just $ processNodeId (sendPortProcessId cid)
1188destNid (LocalPortSend cid _) = Just $ processNodeId (sendPortProcessId cid)
1028destNid (SigShutdown) = Nothing
1189destNid (SigShutdown) = Nothing
@@ -1041,7 +1202,7 @@ unClosure closure = do
1041isValidLocalIdentifier :: Identifier -> NC Bool
1202isValidLocalIdentifier :: Identifier -> NC Bool
1042isValidLocalIdentifier ident = do
1203isValidLocalIdentifier ident = do
1043 node <- ask
1204 node <- ask
1044− liftIO . withMVar (localState node) $ \nSt ->
1205+ liftIO . withValidLocalState node $ \nSt ->
1045 case ident of
1206 case ident of
1046 NodeIdentifier nid ->
1207 NodeIdentifier nid ->
1047 return $ nid == localNodeId node
1208 return $ nid == localNodeId node
@@ -1072,6 +1233,9 @@ postMessage pid msg = do
1072throwException :: Exception e => ProcessId -> e -> NC ()
1233throwException :: Exception e => ProcessId -> e -> NC ()
1073throwException pid e = do
1234throwException pid e = do
1074 node <- ask
1235 node <- ask
1236+ -- throwTo blocks until the exception is received by the target thread.
1237+ -- We cannot easily make it happen asynchronpusly because then 'unlink'
1238+ -- semantics would break.
1075 liftIO $ withLocalProc node pid $ \p -> throwTo (processThread p) e
1239 liftIO $ withLocalProc node pid $ \p -> throwTo (processThread p) e
1077withLocalProc :: LocalNode -> ProcessId -> (LocalProcess -> IO ()) -> IO ()
1241withLocalProc :: LocalNode -> ProcessId -> (LocalProcess -> IO ()) -> IO ()
@@ -1079,17 +1243,17 @@ withLocalProc node pid p =
1079 -- By [Unified: table 6, rule missing_process] messages to dead processes
1243 -- By [Unified: table 6, rule missing_process] messages to dead processes
1080 -- can silently be dropped
1244 -- can silently be dropped
1081 let lpid = processLocalId pid in do
1245 let lpid = processLocalId pid in do
1082− mProc <- withMVar (localState node) $ return . (^. localProcessWithId lpid)
1246+ join $ withValidLocalState node $ \vst ->
1083− forM_ mProc p
1247+ return $ forM_ (vst ^. localProcessWithId lpid) p
1085--------------------------------------------------------------------------------
1249--------------------------------------------------------------------------------
1086-- Accessors --
1250-- Accessors --
1087--------------------------------------------------------------------------------
1251--------------------------------------------------------------------------------
1089−links :: Accessor NCState (Map Identifier (Set ProcessId))
1253+links :: Accessor NCState (BiMultiMap Identifier ProcessId ())
1090links = accessor _links (\ls st -> st { _links = ls })
1254links = accessor _links (\ls st -> st { _links = ls })
1092−monitors :: Accessor NCState (Map Identifier (Set (ProcessId, MonitorRef)))
1256+monitors :: Accessor NCState (BiMultiMap Identifier ProcessId MonitorRef)
1093monitors = accessor _monitors (\ms st -> st { _monitors = ms })
1257monitors = accessor _monitors (\ms st -> st { _monitors = ms })
1095registeredHere :: Accessor NCState (Map String ProcessId)
1259registeredHere :: Accessor NCState (Map String ProcessId)
@@ -1098,12 +1262,6 @@ registeredHere = accessor _registeredHere (\ry st -> st { _registeredHere = ry }
1098registeredOnNodes :: Accessor NCState (Map ProcessId [(NodeId, Int)])
1262registeredOnNodes :: Accessor NCState (Map ProcessId [(NodeId, Int)])
1099registeredOnNodes = accessor _registeredOnNodes (\ry st -> st { _registeredOnNodes = ry })
1263registeredOnNodes = accessor _registeredOnNodes (\ry st -> st { _registeredOnNodes = ry })
1101−linksFor :: Identifier -> Accessor NCState (Set ProcessId)
1102−linksFor ident = links >>> DAC.mapDefault Set.empty ident
1104−monitorsFor :: Identifier -> Accessor NCState (Set (ProcessId, MonitorRef))
1105−monitorsFor ident = monitors >>> DAC.mapDefault Set.empty ident
1107registeredHereFor :: String -> Accessor NCState (Maybe ProcessId)
1265registeredHereFor :: String -> Accessor NCState (Maybe ProcessId)
1108registeredHereFor ident = registeredHere >>> DAC.mapMaybe ident
1266registeredHereFor ident = registeredHere >>> DAC.mapMaybe ident
@@ -1130,10 +1288,12 @@ registeredOnNodesFor ident = registeredOnNodes >>> DAC.mapMaybe ident
1130-- * the notifications for typed channels to that process.
1288-- * the notifications for typed channels to that process.
1132-- See https://github.com/haskell/containers/issues/14 for the bang on _v.
1290-- See https://github.com/haskell/containers/issues/14 for the bang on _v.
1133−splitNotif :: Identifier
1291+splitNotif :: (Ord a, Ord v)
1134− -> Map Identifier a
1292+ => Identifier
1135− -> (Map Identifier a, Map Identifier a)
1293+ -> BiMultiMap Identifier a v
1136−splitNotif ident = Map.partitionWithKey (\k !_v -> ident `impliesDeathOf` k)
1294+ -> (Map Identifier (Set (a,v)), BiMultiMap Identifier a v)
1296+ BiMultiMap.partitionWithKeyBy1st (\k !_v -> ident `impliesDeathOf` k)
1138--------------------------------------------------------------------------------
1298--------------------------------------------------------------------------------
1139-- Auxiliary --
1299-- Auxiliary --