@@ -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)
@@ -23,51 +28,73 @@ 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_)
43−import Data.Maybe (isJust, isNothing, catMaybes)
60+import 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_
59−import qualified Control.Exception as Exception (catch)
78+import qualified Control.Exception as Exception
60−import Control.Concurrent (forkIO, mkWeakThreadId, ThreadId)
84+import Control.Concurrent (forkIO, killThread)
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
70import Control.Concurrent.Chan (newChan, writeChan, readChan)
96import Control.Concurrent.Chan (newChan, writeChan, readChan)
97+import qualified Control.Concurrent.MVar as MVar (newEmptyMVar, takeMVar)
71import Control.Concurrent.STM
98import Control.Concurrent.STM
72 ( atomically
99 ( atomically
@@ -76,6 +103,7 @@ import Control.Distributed.Process.Internal.CQueue
76 , enqueue
103 , enqueue
77 , newCQueue
104 , newCQueue
78 , mkWeakCQueue
105 , mkWeakCQueue
80import qualified Network.Transport as NT
108import qualified Network.Transport as NT
81 ( Transport
109 ( Transport
@@ -87,8 +115,8 @@ import qualified Network.Transport as NT
87 , TransportError(..)
115 , TransportError(..)
88 , address
116 , address
89 , closeEndPoint
117 , closeEndPoint
91 , Connection
118 , Connection
93 , EndPointAddress
121 , EndPointAddress
94 , Reliability(ReliableOrdered)
122 , Reliability(ReliableOrdered)
@@ -105,8 +133,11 @@ import Control.Distributed.Process.Internal.Types
105 , LocalProcessId(..)
133 , LocalProcessId(..)
106 , ProcessId(..)
134 , ProcessId(..)
107 , LocalNode(..)
135 , LocalNode(..)
108− , Tracer(InactiveTracer)
136+ , MxEventBus(..)
109 , LocalNodeState(..)
137 , LocalNodeState(..)
138+ , ValidLocalNodeState(..)
139+ , withValidLocalState
140+ , modifyValidLocalState
110 , LocalProcess(..)
141 , LocalProcess(..)
111 , LocalProcessState(..)
142 , LocalProcessState(..)
112 , Process(..)
143 , Process(..)
@@ -116,9 +147,11 @@ import Control.Distributed.Process.Internal.Types
116 , localPidCounter
147 , localPidCounter
117 , localPidUnique
148 , localPidUnique
118 , localProcessWithId
149 , localProcessWithId
119 , localConnections
151 , localConnections
120 , forever'
152 , forever'
121 , MonitorRef(..)
153 , MonitorRef(..)
154+ , NodeClosedException(..)
122 , ProcessMonitorNotification(..)
155 , ProcessMonitorNotification(..)
123 , NodeMonitorNotification(..)
156 , NodeMonitorNotification(..)
124 , PortMonitorNotification(..)
157 , PortMonitorNotification(..)
@@ -138,52 +171,81 @@ import Control.Distributed.Process.Internal.Types
139 , ProcessInfo(..)
172 , ProcessInfo(..)
140 , ProcessInfoNone(..)
173 , ProcessInfoNone(..)
141 , SendPortId(..)
175 , SendPortId(..)
142 , typedChannelWithId
176 , typedChannelWithId
143 , RegisterReply(..)
177 , RegisterReply(..)
144 , WhereIsReply(..)
178 , WhereIsReply(..)
145 , payloadToMessage
179 , payloadToMessage
146 , createUnencodedMessage
180 , createUnencodedMessage
181+ , unsafeCreateUnencodedMessage
147 , runLocalProcess
182 , runLocalProcess
148 , firstNonReservedProcessId
183 , firstNonReservedProcessId
149 , ImplicitReconnect(WithImplicitReconnect)
184 , ImplicitReconnect(WithImplicitReconnect)
151−import Control.Distributed.Process.Internal.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
195+import Control.Distributed.Process.Management.Internal.Trace.Tracer
198+import Control.Distributed.Process.Management.Internal.Trace.Types
152 ( TraceArg(..)
199 ( TraceArg(..)
153− , traceFormat
200+ , traceEvent
154− , startTracing
201+ , traceLogFmt
155− , stopTracer
202+ , enableTrace
157import Control.Distributed.Process.Serializable (Serializable)
204import Control.Distributed.Process.Serializable (Serializable)
158import Control.Distributed.Process.Internal.Messaging
205import Control.Distributed.Process.Internal.Messaging
159 ( sendBinary
206 ( sendBinary
161 , closeImplicitReconnections
207 , closeImplicitReconnections
162 , impliesDeathOf
208 , impliesDeathOf
164import Control.Distributed.Process.Internal.Primitives
210import Control.Distributed.Process.Internal.Primitives
165 ( register
211 ( register
167 , receiveWait
212 , receiveWait
169 , sendChan
214 , sendChan
171−import Control.Distributed.Process.Internal.Types (SendPort)
221+import Control.Distributed.Process.Internal.Types (SendPort, Tracer(..))
172import qualified Control.Distributed.Process.Internal.Closure.BuiltIn as BuiltIn (remoteTable)
222import qualified Control.Distributed.Process.Internal.Closure.BuiltIn as BuiltIn (remoteTable)
173import Control.Distributed.Process.Internal.WeakTQueue (TQueue, writeTQueue)
223import Control.Distributed.Process.Internal.WeakTQueue (TQueue, writeTQueue)
174import qualified Control.Distributed.Process.Internal.StrictContainerAccessors as DAC
224import qualified Control.Distributed.Process.Internal.StrictContainerAccessors as DAC
175 ( mapMaybe
225 ( mapMaybe
176 , mapDefault
226 , mapDefault
228+import Control.Monad.Catch (try)
229+import GHC.IO (IO(..), unsafeUnmask)
230+import GHC.Base ( maskAsyncExceptions# )
179import 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
181--------------------------------------------------------------------------------
243--------------------------------------------------------------------------------
182-- Initialization --
244-- Initialization --
183--------------------------------------------------------------------------------
245--------------------------------------------------------------------------------
185initRemoteTable :: RemoteTable
247initRemoteTable :: RemoteTable
186−initRemoteTable = BuiltIn.remoteTable Static.initRemoteTable
248+initRemoteTable = Trace.remoteTable $ BuiltIn.remoteTable Static.initRemoteTable
188-- | Initialize a new local node.
250-- | Initialize a new local node.
189newLocalNode :: NT.Transport -> RemoteTable -> IO LocalNode
251newLocalNode :: NT.Transport -> RemoteTable -> IO LocalNode
@@ -200,7 +262,7 @@ newLocalNode transport rtable = do
200createBareLocalNode :: NT.EndPoint -> RemoteTable -> IO LocalNode
262createBareLocalNode :: NT.EndPoint -> RemoteTable -> IO LocalNode
201createBareLocalNode endPoint rtable = do
263createBareLocalNode endPoint rtable = do
202 unq <- randomIO
264 unq <- randomIO
203− state <- newMVar LocalNodeState
265+ state <- newMVar $ LocalNodeValid $ ValidLocalNodeState
204 { _localProcesses = Map.empty
266 { _localProcesses = Map.empty
205 , _localPidCounter = firstNonReservedProcessId
267 , _localPidCounter = firstNonReservedProcessId
206 , _localPidUnique = unq
268 , _localPidUnique = unq
@@ -211,24 +273,19 @@ createBareLocalNode endPoint rtable = do
211 , localEndPoint = endPoint
273 , localEndPoint = endPoint
212 , localState = state
274 , localState = state
213 , localCtrlChan = ctrlChan
275 , localCtrlChan = ctrlChan
214− , localTracer = InactiveTracer
276+ , localEventBus = MxEventBusInitialising
215 , remoteTable = rtable
277 , remoteTable = rtable
217− tracedNode <- startTracing node
279+ tracedNode <- startMxAgent node
219− ncGo <- newEmptyMVar
281+ -- Once the NC terminates, the endpoint isn't much use,
220− ncTid <- forkIO $ do
221− tid <- takeMVar ncGo
222− (runNodeController tracedNode
282+ void $ forkIO $ Exception.finally (runNodeController tracedNode)
223− `Exception.catch` \(e :: SomeException) -> reThrow tid e)
283+ (NT.closeEndPoint (localEndPoint node))
225− weakNcTid <- mkWeakThreadId ncTid
285+ -- whilst a closed/failing endpoint will terminate the NC
226− evTid <- forkIO $ do
227− (handleIncomingMessages tracedNode >> (stopNC node))
286+ void $ forkIO $ Exception.finally (handleIncomingMessages tracedNode)
228− `Exception.catch` \(e :: SomeException) -> reThrow weakNcTid e
287+ (stopNC node)
230− weakEvTid <- mkWeakThreadId evTid
231− putMVar ncGo weakEvTid
232 return tracedNode
289 return tracedNode
234 stopNC node =
291 stopNC node =
@@ -237,55 +294,100 @@ createBareLocalNode endPoint rtable = do
237 , ctrlMsgSignal = SigShutdown
294 , ctrlMsgSignal = SigShutdown
240− reThrow :: (Exception e) => Weak ThreadId -> e -> IO ()
297+startMxAgent :: LocalNode -> IO LocalNode
241− reThrow wTid ex = do
298+startMxAgent node = do
242− tid <- deRefWeak wTid
299+ -- see note [tracer/forkProcess races]
300+ let fork = forkProcess node
301+ mv <- MVar.newEmptyMVar
302+ pid <- fork $ mxAgentController fork mv
303+ (tracer', wqRef, mxNew') <- MVar.takeMVar mv
304+ return node { localEventBus = (MxEventBus pid tracer' wqRef mxNew') }
306+startDefaultTracer :: LocalNode -> IO ()
307+startDefaultTracer node' = do
308+ let t = localEventBus node'
243− case tid of
309+ case t of
244− Nothing -> return ()
310+ MxEventBus _ (Tracer pid _) _ _ -> do
245− Just t -> throwTo t ex
311+ runProcess node' $ register "trace.controller" pid
312+ pid' <- forkProcess node' defaultTracer
313+ enableTrace (localEventBus node') pid'
314+ runProcess node' $ register "tracer.initial" pid'
317+-- TODO: we need a better mechanism for defining and registering services
247-- | Start and register the service processes on a node
319-- | Start and register the service processes on a node
248−-- (for now, this is only the logger)
249startServiceProcesses :: LocalNode -> IO ()
320startServiceProcesses :: LocalNode -> IO ()
250startServiceProcesses node = do
321startServiceProcesses node = do
322+ -- tracing /spawns/ relies on the tracer being enabled, but we start
323+ -- the default tracer first, even though it might @nsend@ to the logger
324+ -- before /that/ process has started - this is a totally harmless race
325+ -- however, so we deliberably ignore it
326+ startDefaultTracer node
251 logger <- forkProcess node loop
327 logger <- forkProcess node loop
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.
252− runProcess node $ register "logger" logger
333+ register "trace.logger" logger
254− loop = do
335+ loop = do
255− receiveWait
336+ receiveWait
256− [ match $ \((time, pid, string) ::(String, ProcessId, String)) -> do
337+ [ match $ \(SayMessage time pid string) -> do
338+ let time' = formatTime defaultTimeLocale "%c" time
257− liftIO . hPutStrLn stderr $ time ++ " " ++ show pid ++ ": " ++ string
339+ liftIO . hPutStrLn stderr $ time' ++ " " ++ show pid ++ ": " ++ string
259− , match $ \((time, string) :: (String, String)) -> do
341+ , match $ \((time, string) :: (String, String)) -> do
260− -- this is a 'trace' message from the local node tracer
342+ -- this is a 'trace' message from the local node tracer
261− liftIO . hPutStrLn stderr $ time ++ " [trace] " ++ string
343+ liftIO . hPutStrLn stderr $ time ++ " [trace] " ++ string
263− , match $ \(ch :: SendPort ()) -> -- a shutdown request
345+ , match $ \(ch :: SendPort ()) -> -- a shutdown request
264− sendChan ch ()
346+ sendChan ch ()
267−-- | Force-close a local node
349+-- | Force-close a local node, killing all processes on that node.
269−-- TODO: for now we just close the associated endpoint
270closeLocalNode :: LocalNode -> IO ()
350closeLocalNode :: LocalNode -> IO ()
271−closeLocalNode node =
351+closeLocalNode node = do
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').
272 NT.closeEndPoint (localEndPoint node)
368 NT.closeEndPoint (localEndPoint node)
274-- | 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
275runProcess :: LocalNode -> Process () -> IO ()
371runProcess :: LocalNode -> Process () -> IO ()
276runProcess node proc = do
372runProcess node proc = do
277 done <- newEmptyMVar
373 done <- newEmptyMVar
374+ -- TODO; When forkProcess inherits the masking state, protect the forked
375+ -- thread against async exceptions that could occur before 'try' is evaluated.
278− void $ forkProcess node (proc `finally` liftIO (putMVar done ()))
376+ void $ forkProcess node $ try proc >>= liftIO . putMVar done
279− takeMVar done
377+ takeMVar done >>= either (throwIO :: SomeException -> IO a) return
281-- | Spawn a new process on a local node
379-- | Spawn a new process on a local node
282forkProcess :: LocalNode -> Process () -> IO ProcessId
380forkProcess :: LocalNode -> Process () -> IO ProcessId
381+forkProcess node proc = do
382+ ms <- getMaskingState
283−forkProcess node proc = modifyMVar (localState node) startProcess
383+ modifyMVarMasked (localState node) (startProcess ms)
385+ startProcess :: MaskingState
285− startProcess :: LocalNodeState -> IO (LocalNodeState, ProcessId)
387+ -> IO (LocalNodeState, ProcessId)
286− startProcess st = do
388+ startProcess ms (LocalNodeValid vst) = do
287− let lpid = LocalProcessId { lpidCounter = st ^. localPidCounter
389+ let lpid = LocalProcessId { lpidCounter = vst ^. localPidCounter
288− , lpidUnique = st ^. localPidUnique
390+ , lpidUnique = vst ^. localPidUnique
290 let pid = ProcessId { processNodeId = localNodeId node
392 let pid = ProcessId { processNodeId = localNodeId node
291 , processLocalId = lpid
393 , processLocalId = lpid
@@ -305,43 +407,81 @@ forkProcess node proc = modifyMVar (localState node) startProcess
305 , processThread = tid
407 , processThread = tid
306 , processNode = node
408 , processNode = node
410+ -- Rewrite this code when this is fixed:
411+ -- https://ghc.haskell.org/trac/ghc/ticket/10149
412+ let unmask = case ms of
414+ MaskedInterruptible -> block
415+ MaskedUninterruptible -> id
308− tid' <- forkIO $ do
416+ tid' <- uninterruptibleMask_ $ forkIO $ do
309− reason <- Exception.catch
417+ reason <- Exception.catches
310− (runLocalProcess lproc proc >> return DiedNormal)
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
311− (return . DiedException . (show :: SomeException -> String))
425+ (return . DiedException . (show :: SomeException -> String)))]
313 -- [Unified: Table 4, rules termination and exiting]
427 -- [Unified: Table 4, rules termination and exiting]
314− 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
315 writeChan (localCtrlChan node) NCMsg
434 writeChan (localCtrlChan node) NCMsg
316 { ctrlMsgSender = ProcessIdentifier pid
435 { ctrlMsgSender = ProcessIdentifier pid
317 , ctrlMsgSignal = Died (ProcessIdentifier pid) reason
436 , ctrlMsgSignal = Died (ProcessIdentifier pid) reason
319 return (tid', lproc)
438 return (tid', lproc)
440+ -- see note [tracer/forkProcess races]
441+ trace node (MxSpawned pid)
321 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!?
323 newUnique <- randomIO
447 newUnique <- randomIO
448+ return ( LocalNodeValid
324− return ( (localProcessWithId lpid ^= Just lproc)
449+ $ (localProcessWithId lpid ^= Just lproc)
325 . (localPidCounter ^= firstNonReservedProcessId)
450 . (localPidCounter ^= firstNonReservedProcessId)
326 . (localPidUnique ^= newUnique)
451 . (localPidUnique ^= newUnique)
456+ return ( LocalNodeValid
331− return ( (localProcessWithId lpid ^= Just lproc)
457+ $ (localProcessWithId lpid ^= Just lproc)
332 . (localPidCounter ^: (+ 1))
458 . (localPidCounter ^: (+ 1))
462+ startProcess _ LocalNodeClosed =
463+ throwIO $ NodeClosedException $ localNodeId node
337− cleanupProcess :: ProcessId -> LocalNodeState -> IO LocalNodeState
465+ cleanupProcess :: ProcessId
466+ -> ValidLocalNodeState
467+ -> IO (ValidLocalNodeState, [NT.Connection])
338− cleanupProcess pid st = do
468+ cleanupProcess pid vst = do
339 let pid' = ProcessIdentifier pid
469 let pid' = ProcessIdentifier pid
340− 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)
341− mapM_ (NT.close . fst) (Map.elems affected)
342− return $ (localProcessWithId (processLocalId pid) ^= Nothing)
471+ return ( (localProcessWithId (processLocalId pid) ^= Nothing)
343− . (localConnections ^= unaffected)
472+ . (localConnections ^= unaffected)
474+ , map fst $ Map.elems affected
477+-- note [tracer/forkProcess races]
479+-- Our startTracing function uses forkProcess to start the trace controller
480+-- process, and of course forkProcess attempts to call traceEvent once the
481+-- process has started. This is harmless, as the localEventBus is not updated
482+-- until /after/ the initial forkProcess completes, so the first call to
483+-- traceEvent behaves as if tracing were disabled (i.e., it is ignored).
346--------------------------------------------------------------------------------
486--------------------------------------------------------------------------------
347-- Handle incoming messages --
487-- Handle incoming messages --
@@ -351,8 +491,8 @@ type IncomingConnection = (NT.EndPointAddress, IncomingTarget)
352data IncomingTarget =
492data IncomingTarget =
354− | ToProc (Weak (CQueue Message))
494+ | ToProc ProcessId (Weak (CQueue Message))
355− | ToChan TypedChannel
495+ | ToChan SendPortId TypedChannel
358data ConnectionState = ConnectionState {
498data ConnectionState = ConnectionState {
@@ -379,6 +519,7 @@ incomingFrom addr = aux >>> DAC.mapDefault Set.empty addr
380handleIncomingMessages :: LocalNode -> IO ()
520handleIncomingMessages :: LocalNode -> IO ()
381handleIncomingMessages node = go initConnectionState
521handleIncomingMessages node = go initConnectionState
522+ `Exception.catch` \(NodeClosedException _) -> return ()
383 go :: ConnectionState -> IO ()
524 go :: ConnectionState -> IO ()
384 go !st = do
525 go !st = do
@@ -386,28 +527,37 @@ handleIncomingMessages node = go initConnectionState
386 case event of
527 case event of
387 NT.ConnectionOpened cid rel theirAddr ->
528 NT.ConnectionOpened cid rel theirAddr ->
388 if rel == NT.ReliableOrdered
529 if rel == NT.ReliableOrdered
531+ trace node (MxConnected cid theirAddr)
389− then go ( (incomingAt cid ^= Just (theirAddr, Uninit))
533+ (incomingAt cid ^= Just (theirAddr, Uninit))
390 . (incomingFrom theirAddr ^: Set.insert cid)
534 . (incomingFrom theirAddr ^: Set.insert cid)
393− else invalidRequest cid st
537+ else invalidRequest cid st $
538+ "attempt to connect with unsupported reliability " ++ show rel
394 NT.Received cid payload ->
539 NT.Received cid payload ->
395 case st ^. incomingAt cid of
540 case st ^. incomingAt cid of
396− Just (_, ToProc weakQueue) -> do
541+ Just (_, ToProc pid weakQueue) -> do
397 mQueue <- deRefWeak weakQueue
542 mQueue <- deRefWeak weakQueue
398 forM_ mQueue $ \queue -> do
543 forM_ mQueue $ \queue -> do
399 -- TODO: if we find that the queue is Nothing, should we remove
544 -- TODO: if we find that the queue is Nothing, should we remove
400 -- it from the NC state? (and same for channels, below)
545 -- it from the NC state? (and same for channels, below)
401 let msg = payloadToMessage payload
546 let msg = payloadToMessage payload
402 enqueue queue msg -- 'enqueue' is strict
547 enqueue queue msg -- 'enqueue' is strict
548+ trace node (MxReceived pid msg)
404− Just (_, ToChan (TypedChannel chan')) -> do
550+ Just (_, ToChan chId (TypedChannel chan')) -> do
405 mChan <- deRefWeak chan'
551 mChan <- deRefWeak chan'
406 -- 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
407 -- the channel and we simply ignore the incoming message
553 -- the channel and we simply ignore the incoming message
408− forM_ mChan $ \chan -> atomically $
554+ forM_ mChan $ \chan -> do
555+ msg' <- atomically $ do
556+ msg <- return $! decode (BSL.fromChunks payload)
409− -- We make sure the message is fully decoded when it is enqueued
557+ -- We make sure the message is fully decoded when it is enqueued
410− writeTQueue chan $! decode (BSL.fromChunks payload)
558+ writeTQueue chan msg
560+ trace node $ MxReceivedPort chId $ unsafeCreateUnencodedMessage msg'
412 Just (_, ToNode) -> do
562 Just (_, ToNode) -> do
413 let ctrlMsg = decode . BSL.fromChunks $ payload
563 let ctrlMsg = decode . BSL.fromChunks $ payload
@@ -417,37 +567,49 @@ handleIncomingMessages node = go initConnectionState
417 case decode (BSL.fromChunks payload) of
567 case decode (BSL.fromChunks payload) of
418 ProcessIdentifier pid -> do
568 ProcessIdentifier pid -> do
419 let lpid = processLocalId pid
569 let lpid = processLocalId pid
420− mProc <- withMVar state $ return . (^. localProcessWithId lpid)
570+ mProc <- withValidLocalState node $ return . (^. localProcessWithId lpid)
421 case mProc of
571 case mProc of
422 Just proc ->
572 Just proc ->
423− go (incomingAt cid ^= Just (src, ToProc (processWeakQ proc)) $ st)
573+ go (incomingAt cid ^= Just (src, ToProc pid (processWeakQ proc)) $ st)
424 Nothing ->
574 Nothing ->
425− invalidRequest cid st
575+ -- incoming attempt to connect to unknown process - might
577+ go (incomingAt cid ^= Nothing $ st)
426 SendPortIdentifier chId -> do
578 SendPortIdentifier chId -> do
427 let lcid = sendPortLocalId chId
579 let lcid = sendPortLocalId chId
428 lpid = processLocalId (sendPortProcessId chId)
580 lpid = processLocalId (sendPortProcessId chId)
429− mProc <- withMVar state $ return . (^. localProcessWithId lpid)
581+ mProc <- withValidLocalState node $ return . (^. localProcessWithId lpid)
430 case mProc of
582 case mProc of
431 Just proc -> do
583 Just proc -> do
432 mChannel <- withMVar (processState proc) $ return . (^. typedChannelWithId lcid)
584 mChannel <- withMVar (processState proc) $ return . (^. typedChannelWithId lcid)
433 case mChannel of
585 case mChannel of
434 Just channel ->
586 Just channel ->
435− go (incomingAt cid ^= Just (src, ToChan channel) $ st)
587+ go (incomingAt cid ^= Just (src, ToChan chId channel) $ st)
436 Nothing ->
588 Nothing ->
437− invalidRequest cid st
589+ invalidRequest cid st $
590+ "incoming attempt to connect to unknown channel of"
591+ ++ " process " ++ show (sendPortProcessId chId)
438 Nothing ->
592 Nothing ->
439− invalidRequest cid st
593+ -- incoming attempt to connect to channel of unknown
594+ -- process - might be dead already
595+ go (incomingAt cid ^= Nothing $ st)
440 NodeIdentifier nid ->
596 NodeIdentifier nid ->
441 if nid == localNodeId node
597 if nid == localNodeId node
442 then go (incomingAt cid ^= Just (src, ToNode) $ st)
598 then go (incomingAt cid ^= Just (src, ToNode) $ st)
443− 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 "
444 Nothing ->
604 Nothing ->
445 invalidRequest cid st
605 invalidRequest cid st
606+ "message received from an unknown connection"
446 NT.ConnectionClosed cid ->
607 NT.ConnectionClosed cid ->
447 case st ^. incomingAt cid of
608 case st ^. incomingAt cid of
448 Nothing ->
609 Nothing ->
449− invalidRequest cid st
610+ invalidRequest cid st "closed unknown connection"
450− Just (src, _) ->
611+ Just (src, _) -> do
612+ trace node (MxDisconnected cid src)
451 go ( (incomingAt cid ^= Nothing)
613 go ( (incomingAt cid ^= Nothing)
452 . (incomingFrom src ^: Set.delete cid)
614 . (incomingFrom src ^: Set.delete cid)
@@ -466,32 +628,30 @@ handleIncomingMessages node = go initConnectionState
468 NT.ErrorEvent (NT.TransportError NT.EventEndPointFailed str) ->
630 NT.ErrorEvent (NT.TransportError NT.EventEndPointFailed str) ->
469− fatal $ "Cloud Haskell fatal error: end point failed: " ++ str
631+ fail $ "Cloud Haskell fatal error: end point failed: " ++ str
470 NT.ErrorEvent (NT.TransportError NT.EventTransportFailed str) ->
632 NT.ErrorEvent (NT.TransportError NT.EventTransportFailed str) ->
471− fatal $ "Cloud Haskell fatal error: transport failed: " ++ str
633+ fail $ "Cloud Haskell fatal error: transport failed: " ++ str
472 NT.EndPointClosed ->
634 NT.EndPointClosed ->
473− stopTracer (localTracer node) >> return ()
635+ return ()
474 NT.ReceivedMulticast _ _ ->
636 NT.ReceivedMulticast _ _ ->
475 -- If we received a multicast message, something went horribly wrong
637 -- If we received a multicast message, something went horribly wrong
476 -- and we just give up
638 -- and we just give up
477− fatal "Cloud Haskell fatal error: received unexpected multicast"
639+ fail "Cloud Haskell fatal error: received unexpected multicast"
479− fatal :: String -> IO ()
480− fatal msg = stopTracer (localTracer node) >> fail msg
482− invalidRequest :: NT.ConnectionId -> ConnectionState -> IO ()
641+ invalidRequest :: NT.ConnectionId -> ConnectionState -> String -> IO ()
483− invalidRequest cid st = do
642+ invalidRequest cid st msg = do
484 -- 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
485 -- 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
486 -- should close incoming connections (this requires a Transport layer
645 -- should close incoming connections (this requires a Transport layer
487 -- extension).
646 -- extension).
488− traceEventFmtIO node "" [(TraceStr "[network] invalid request: "),
647+ traceEventFmtIO node "" [ TraceStr $ " [network] invalid request"
648+ ++ " (" ++ msg ++ "): "
489− (Trace cid)]
649+ , (Trace cid)
490 go ( incomingAt cid ^= Nothing
651 go ( incomingAt cid ^= Nothing
494− state = localState node
495 endpoint = localEndPoint node
655 endpoint = localEndPoint node
496 ctrlChan = localCtrlChan node
656 ctrlChan = localCtrlChan node
@@ -500,8 +660,9 @@ handleIncomingMessages node = go initConnectionState
500--------------------------------------------------------------------------------
660--------------------------------------------------------------------------------
502runNodeController :: LocalNode -> IO ()
662runNodeController :: LocalNode -> IO ()
503−runNodeController =
663+runNodeController node =
504− runReaderT (evalStateT (unNC nodeController) initNCState)
664+ runReaderT (evalStateT (unNC nodeController) initNCState) node
665+ `Exception.catch` \(NodeClosedException _) -> return ()
506--------------------------------------------------------------------------------
667--------------------------------------------------------------------------------
507-- Internal data types --
668-- Internal data types --
@@ -509,20 +670,26 @@ runNodeController =
510data NCState = NCState
671data NCState = NCState
511 { -- Mapping from remote processes to linked local processes
672 { -- Mapping from remote processes to linked local processes
512− _links :: !(Map Identifier (Set ProcessId))
673+ _links :: !(BiMultiMap Identifier ProcessId ())
513 -- Mapping from remote processes to monitoring local processes
674 -- Mapping from remote processes to monitoring local processes
514− , _monitors :: !(Map Identifier (Set (ProcessId, MonitorRef)))
675+ , _monitors :: !(BiMultiMap Identifier ProcessId MonitorRef)
515 -- Process registry: names and where they live, mapped to the PIDs
676 -- Process registry: names and where they live, mapped to the PIDs
516 , _registeredHere :: !(Map String ProcessId)
677 , _registeredHere :: !(Map String ProcessId)
517 , _registeredOnNodes :: !(Map ProcessId [(NodeId,Int)])
678 , _registeredOnNodes :: !(Map ProcessId [(NodeId,Int)])
520newtype NC a = NC { unNC :: StateT NCState (ReaderT LocalNode IO) a }
681newtype NC a = NC { unNC :: StateT NCState (ReaderT LocalNode IO) a }
682+ deriving ( Applicative
686+ , MonadState NCState
521− deriving (Functor, Monad, MonadIO, MonadState NCState, MonadReader LocalNode)
687+ , MonadReader LocalNode
523initNCState :: NCState
690initNCState :: NCState
524−initNCState = NCState { _links = Map.empty
691+initNCState = NCState { _links = BiMultiMap.empty
525− , _monitors = Map.empty
692+ , _monitors = BiMultiMap.empty
526 , _registeredHere = Map.empty
693 , _registeredHere = Map.empty
527 , _registeredOnNodes = Map.empty
694 , _registeredOnNodes = Map.empty
@@ -538,34 +705,60 @@ instance Show ProcessKillException where
538 show (ProcessKillException pid reason) =
705 show (ProcessKillException pid reason) =
539 "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
541--------------------------------------------------------------------------------
735--------------------------------------------------------------------------------
542-- Tracing/Debugging --
736-- Tracing/Debugging --
543--------------------------------------------------------------------------------
737--------------------------------------------------------------------------------
545−-- [Issue #104]
739+-- [Issue #104 / DP-13]
547traceNotifyDied :: LocalNode -> Identifier -> DiedReason -> NC ()
741traceNotifyDied :: LocalNode -> Identifier -> DiedReason -> NC ()
548traceNotifyDied node ident reason =
742traceNotifyDied node ident reason =
743+ -- TODO: sendPortDied notifications
744+ liftIO $ withLocalTracer node $ \t ->
549− case reason of
745+ case ident of
746+ (NodeIdentifier nid) -> traceEvent t (MxNodeDied nid reason)
747+ (ProcessIdentifier pid) -> traceEvent t (MxProcessDied pid reason)
550− DiedNormal -> return ()
748+ _ -> return ()
551− _ -> traceNcEventFmt node " "
552− [(TraceStr "[node-controller]"),
556−traceNcEventFmt :: LocalNode -> String -> [TraceArg] -> NC ()
557−traceNcEventFmt node fmt args =
558− liftIO $ traceEventFmtIO node fmt args
560traceEventFmtIO :: LocalNode
750traceEventFmtIO :: LocalNode
561 -> String
751 -> String
562 -> [TraceArg]
752 -> [TraceArg]
564traceEventFmtIO node fmt args =
754traceEventFmtIO node fmt args =
565− withLocalTracer node $ \t -> traceFormat t fmt args
755+ withLocalTracer node $ \t -> traceLogFmt t fmt args
757+trace :: LocalNode -> MxEvent -> IO ()
758+trace node ev = withLocalTracer node $ \t -> traceEvent t ev
567−withLocalTracer :: LocalNode -> (Tracer -> IO ()) -> IO ()
760+withLocalTracer :: LocalNode -> (MxEventBus -> IO ()) -> IO ()
568−withLocalTracer node act = act (localTracer node)
761+withLocalTracer node act = act (localEventBus node)
570--------------------------------------------------------------------------------
763--------------------------------------------------------------------------------
571-- Core functionality --
764-- Core functionality --
@@ -581,11 +774,7 @@ nodeController = do
581 -- [Unified: Table 7, rule nc_forward]
774 -- [Unified: Table 7, rule nc_forward]
582 case destNid (ctrlMsgSignal msg) of
775 case destNid (ctrlMsgSignal msg) of
583 Just nid' | nid' /= localNodeId node ->
776 Just nid' | nid' /= localNodeId node ->
584− liftIO $ sendBinary node
777+ ncSendToNode nid' msg
586− (NodeIdentifier nid')
587− WithImplicitReconnect
590 return ()
779 return ()
@@ -608,8 +797,10 @@ nodeController = do
608 ncEffectWhereIs from label
797 ncEffectWhereIs from label
609 NCMsg _ (NamedSend label msg') ->
798 NCMsg _ (NamedSend label msg') ->
610 ncEffectNamedSend label msg'
799 ncEffectNamedSend label msg'
800+ NCMsg _ (UnreliableSend lpid msg') ->
801+ ncEffectLocalSend node (ProcessId (localNodeId node) lpid) msg'
611 NCMsg _ (LocalSend to msg') ->
802 NCMsg _ (LocalSend to msg') ->
612− ncEffectLocalSend to msg'
803+ ncEffectLocalSend node to msg'
613 NCMsg _ (LocalPortSend to msg') ->
804 NCMsg _ (LocalPortSend to msg') ->
614 ncEffectLocalPortSend to msg'
805 ncEffectLocalPortSend to msg'
615 NCMsg (ProcessIdentifier from) (Kill to reason) ->
806 NCMsg (ProcessIdentifier from) (Kill to reason) ->
@@ -619,7 +810,11 @@ nodeController = do
619 NCMsg (ProcessIdentifier from) (GetInfo pid) ->
810 NCMsg (ProcessIdentifier from) (GetInfo pid) ->
620 ncEffectGetInfo from pid
811 ncEffectGetInfo from pid
621 NCMsg _ SigShutdown ->
812 NCMsg _ SigShutdown ->
622− liftIO $ throwIO ThreadKilled -- seems to make more sense than fail/error
813+ liftIO $ do
814+ NT.closeEndPoint (localEndPoint node)
815+ `Exception.finally` throwIO (NodeClosedException $ localNodeId node)
816+ NCMsg (ProcessIdentifier from) (GetNodeStats nid) ->
817+ ncEffectGetNodeStats from nid
623 unexpected ->
818 unexpected ->
624 error $ "nodeController: unexpected message " ++ show unexpected
819 error $ "nodeController: unexpected message " ++ show unexpected
@@ -637,22 +832,18 @@ ncEffectMonitor from them mRef = do
637 case (shouldLink, isLocal node (ProcessIdentifier from)) of
832 case (shouldLink, isLocal node (ProcessIdentifier from)) of
638 (True, _) -> -- [Unified: first rule]
833 (True, _) -> -- [Unified: first rule]
639 case mRef of
834 case mRef of
640− Just ref -> modify' $ monitorsFor them ^: Set.insert (from, ref)
835+ Just ref -> modify' $ monitors ^: BiMultiMap.insert them from ref
641− Nothing -> modify' $ linksFor them ^: Set.insert from
836+ Nothing -> modify' $ links ^: BiMultiMap.insert them from ()
642 (False, True) -> -- [Unified: second rule]
837 (False, True) -> -- [Unified: second rule]
643 notifyDied from them DiedUnknownId mRef
838 notifyDied from them DiedUnknownId mRef
644 (False, False) -> -- [Unified: third rule]
839 (False, False) -> -- [Unified: third rule]
645 -- TODO: this is the right sender according to the Unified semantics,
840 -- TODO: this is the right sender according to the Unified semantics,
646 -- but perhaps having 'them' as the sender would make more sense
841 -- but perhaps having 'them' as the sender would make more sense
647 -- (see also: notifyDied)
842 -- (see also: notifyDied)
648− liftIO $ sendBinary node
649− (NodeIdentifier $ localNodeId node)
650− (NodeIdentifier $ processNodeId from)
843+ ncSendToNode (processNodeId from) $ NCMsg
651− WithImplicitReconnect
653− { ctrlMsgSender = NodeIdentifier (localNodeId node)
844+ { ctrlMsgSender = NodeIdentifier (localNodeId node)
654− , ctrlMsgSignal = Died them DiedUnknownId
845+ , ctrlMsgSignal = Died them DiedUnknownId
657-- [Unified: Table 11]
848-- [Unified: Table 11]
658ncEffectUnlink :: ProcessId -> Identifier -> NC ()
849ncEffectUnlink :: ProcessId -> Identifier -> NC ()
@@ -666,7 +857,7 @@ ncEffectUnlink from them = do
666 postAsMessage from $ DidUnlinkNode nid
857 postAsMessage from $ DidUnlinkNode nid
667 SendPortIdentifier cid ->
858 SendPortIdentifier cid ->
668 postAsMessage from $ DidUnlinkPort cid
859 postAsMessage from $ DidUnlinkPort cid
669− modify' $ linksFor them ^: Set.delete from
860+ modify' $ links ^: BiMultiMap.delete them from ()
671-- [Unified: Table 11]
862-- [Unified: Table 11]
672ncEffectUnmonitor :: ProcessId -> MonitorRef -> NC ()
863ncEffectUnmonitor :: ProcessId -> MonitorRef -> NC ()
@@ -674,7 +865,7 @@ ncEffectUnmonitor from ref = do
674 node <- ask
865 node <- ask
675 when (isLocal node (ProcessIdentifier from)) $
866 when (isLocal node (ProcessIdentifier from)) $
676 postAsMessage from $ DidUnmonitor ref
867 postAsMessage from $ DidUnmonitor ref
677− modify' $ monitorsFor (monitorRefIdent ref) ^: Set.delete (from, ref)
868+ modify' $ monitors ^: BiMultiMap.delete (monitorRefIdent ref) from ref
679-- [Unified: Table 12]
870-- [Unified: Table 12]
680ncEffectDied :: Identifier -> DiedReason -> NC ()
871ncEffectDied :: Identifier -> DiedReason -> NC ()
@@ -689,7 +880,7 @@ ncEffectDied ident reason = do
689 let localOnly = case ident of NodeIdentifier _ -> True ; _ -> False
880 let localOnly = case ident of NodeIdentifier _ -> True ; _ -> False
691 forM_ (Map.toList affectedLinks) $ \(them, uss) ->
882 forM_ (Map.toList affectedLinks) $ \(them, uss) ->
692− forM_ uss $ \us ->
883+ forM_ uss $ \(us, _) ->
693 when (localOnly <= isLocal node (ProcessIdentifier us)) $
884 when (localOnly <= isLocal node (ProcessIdentifier us)) $
694 notifyDied us them reason Nothing
885 notifyDied us them reason Nothing
@@ -698,9 +889,35 @@ ncEffectDied ident reason = do
698 when (localOnly <= isLocal node (ProcessIdentifier us)) $
889 when (localOnly <= isLocal node (ProcessIdentifier us)) $
699 notifyDied us them reason (Just ref)
890 notifyDied us them reason (Just ref)
892+ -- Notify remote nodes that the process died so it can be removed from monitor
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
701− modify' $ (links ^= unaffectedLinks) . (monitors ^= unaffectedMons)
914+ modify' $ (links ^= unaffectedLinks') . (monitors ^= unaffectedMons')
916+ -- we now consider all labels for this identifier unregistered
703− modify' $ registeredHere ^: Map.filter (\pid -> not $ ident `impliesDeathOf` ProcessIdentifier pid)
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
705 remaining <- fmap Map.toList (gets (^. registeredOnNodes)) >>=
922 remaining <- fmap Map.toList (gets (^. registeredOnNodes)) >>=
706 mapM (\(pid,nidlist) ->
923 mapM (\(pid,nidlist) ->
@@ -708,20 +925,15 @@ ncEffectDied ident reason = do
709 do forM_ nidlist $ \(nid,_) ->
926 do forM_ nidlist $ \(nid,_) ->
710 when (not $ isLocal node (NodeIdentifier nid))
927 when (not $ isLocal node (NodeIdentifier nid))
711− (forwardNameDeath node nid)
928+ (forwardDeath node nid)
712 return Nothing
929 return Nothing
713 False -> return $ Just (pid,nidlist) )
930 False -> return $ Just (pid,nidlist) )
714 modify' $ registeredOnNodes ^= (Map.fromList (catMaybes remaining))
931 modify' $ registeredOnNodes ^= (Map.fromList (catMaybes remaining))
716− forwardNameDeath node nid =
933+ forwardDeath node nid = ncSendToNode nid
717− liftIO $ sendBinary node
718− (NodeIdentifier $ localNodeId node)
719− (NodeIdentifier $ nid)
720− WithImplicitReconnect
722− { ctrlMsgSender = NodeIdentifier (localNodeId node)
934+ NCMsg { ctrlMsgSender = NodeIdentifier (localNodeId node)
723− , ctrlMsgSignal = Died ident reason
935+ , ctrlMsgSignal = Died ident reason
726-- [Unified: Table 13]
938-- [Unified: Table 13]
727ncEffectSpawn :: ProcessId -> Closure (Process ()) -> SpawnRef -> NC ()
939ncEffectSpawn :: ProcessId -> Closure (Process ()) -> SpawnRef -> NC ()
@@ -729,16 +941,13 @@ ncEffectSpawn pid cProc ref = do
729 mProc <- unClosure cProc
941 mProc <- unClosure cProc
730 -- 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
731 -- This allows the remote node to find out what's happening
943 -- This allows the remote node to find out what's happening
732 let proc = case mProc of
945 let proc = case mProc of
733 Left err -> fail $ "Error: Could not resolve closure: " ++ err
946 Left err -> fail $ "Error: Could not resolve closure: " ++ err
734 Right p -> p
947 Right p -> p
735 node <- ask
948 node <- ask
736 pid' <- liftIO $ forkProcess node proc
949 pid' <- liftIO $ forkProcess node proc
737− liftIO $ sendMessage node
950+ ncSendToProcess pid $ unsafeCreateUnencodedMessage $ DidSpawn ref pid'
738− (NodeIdentifier (localNodeId node))
739− (ProcessIdentifier pid)
740− WithImplicitReconnect
743-- Unified semantics does not explicitly describe how to implement 'register',
952-- Unified semantics does not explicitly describe how to implement 'register',
744-- but mentions it's "very similar to nsend" (Table 14)
953-- but mentions it's "very similar to nsend" (Table 14)
@@ -756,14 +965,19 @@ ncEffectRegister from label atnode mPid reregistration = do
756 return $ (isNothing currentVal /= reregistration) &&
965 return $ (isNothing currentVal /= reregistration) &&
757 (not (isLocal node (ProcessIdentifier thepid) ) || isvalidlocal )
966 (not (isLocal node (ProcessIdentifier thepid) ) || isvalidlocal )
758 if isLocal node (NodeIdentifier atnode)
967 if isLocal node (NodeIdentifier atnode)
759− then do when (isOk) $
968+ then do when isOk $
760 do modify' $ registeredHereFor label ^= mPid
969 do modify' $ registeredHereFor label ^= mPid
761 updateRemote node currentVal mPid
970 updateRemote node currentVal mPid
762− liftIO $ sendMessage node
971+ case mPid of
763− (NodeIdentifier (localNodeId node))
972+ (Just p) -> do
764− (ProcessIdentifier from)
973+ if reregistration
765− WithImplicitReconnect
974+ then liftIO $ trace node (MxUnRegistered (fromJust currentVal) label)
976+ liftIO $ trace node (MxRegistered p label)
977+ Nothing -> liftIO $ trace node (MxUnRegistered (fromJust currentVal) label)
978+ newVal <- gets (^. registeredHereFor label)
979+ ncSendToProcess from $ unsafeCreateUnencodedMessage $
766− (RegisterReply label isOk)
980+ RegisterReply label isOk newVal
767 else let operation =
981 else let operation =
768 case reregistration of
982 case reregistration of
769 True -> flip decList
983 True -> flip decList
@@ -781,6 +995,7 @@ ncEffectRegister from label atnode mPid reregistration = do
781 updateRemote _ _ _ = return ()
995 updateRemote _ _ _ = return ()
782 maybeify f Nothing = unmaybeify $ f []
996 maybeify f Nothing = unmaybeify $ f []
783 maybeify f (Just x) = unmaybeify $ f x
997 maybeify f (Just x) = unmaybeify $ f x
784 unmaybeify [] = Nothing
999 unmaybeify [] = Nothing
785 unmaybeify x = Just x
1000 unmaybeify x = Just x
786 incList [] tag = [(tag,1)]
1001 incList [] tag = [(tag,1)]
@@ -792,45 +1007,44 @@ ncEffectRegister from label atnode mPid reregistration = do
792 decList (x:xs) tag = x:decList xs tag
1007 decList (x:xs) tag = x:decList xs tag
793 forward node to reg =
1008 forward node to reg =
794 when (not $ isLocal node (NodeIdentifier to)) $
1009 when (not $ isLocal node (NodeIdentifier to)) $
795− liftIO $ sendBinary node
796− (ProcessIdentifier from)
798− WithImplicitReconnect
800− { ctrlMsgSender = ProcessIdentifier from
1010+ ncSendToNode to $ NCMsg { ctrlMsgSender = ProcessIdentifier from
801− , ctrlMsgSignal = reg
1011+ , ctrlMsgSignal = reg
805-- Unified semantics does not explicitly describe 'whereis'
1015-- Unified semantics does not explicitly describe 'whereis'
806ncEffectWhereIs :: ProcessId -> String -> NC ()
1016ncEffectWhereIs :: ProcessId -> String -> NC ()
807ncEffectWhereIs from label = do
1017ncEffectWhereIs from label = do
809 mPid <- gets (^. registeredHereFor label)
1018 mPid <- gets (^. registeredHereFor label)
810− liftIO $ sendMessage node
811− (NodeIdentifier (localNodeId node))
812− (ProcessIdentifier from)
813− WithImplicitReconnect
814− (WhereIsReply label mPid)
1019+ ncSendToProcess from $ unsafeCreateUnencodedMessage $ WhereIsReply label mPid
816-- [Unified: Table 14]
1021-- [Unified: Table 14]
817ncEffectNamedSend :: String -> Message -> NC ()
1022ncEffectNamedSend :: String -> Message -> NC ()
818ncEffectNamedSend label msg = do
1023ncEffectNamedSend label msg = do
819 mPid <- gets (^. registeredHereFor label)
1024 mPid <- gets (^. registeredHereFor label)
820 -- 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)
821− 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
823-- [Issue #DP-20]
1031-- [Issue #DP-20]
824−ncEffectLocalSend :: ProcessId -> Message -> NC ()
1032+ncEffectLocalSend :: LocalNode -> ProcessId -> Message -> NC ()
825−ncEffectLocalSend = postMessage
1033+ncEffectLocalSend = ncEffectLocalSendAndTrace True
1035+ncEffectLocalSendAndTrace :: Bool -> LocalNode -> ProcessId -> Message -> NC ()
1036+ncEffectLocalSendAndTrace shouldTrace node to msg =
1037+ liftIO $ withLocalProc node to $ \p -> do
1038+ enqueue (processQueue p) msg
1039+ when shouldTrace $ trace node (MxReceived to msg)
827-- [Issue #DP-20]
1041-- [Issue #DP-20]
828ncEffectLocalPortSend :: SendPortId -> Message -> NC ()
1042ncEffectLocalPortSend :: SendPortId -> Message -> NC ()
829−ncEffectLocalPortSend from msg =
1043+ncEffectLocalPortSend from msg = do
830 let pid = sendPortProcessId from
1045 let pid = sendPortProcessId from
831 cid = sendPortLocalId from
1046 cid = sendPortLocalId from
833− withLocalProc pid $ \proc -> do
1047+ liftIO $ withLocalProc node pid $ \proc -> do
834 mChan <- withMVar (processState proc) $ return . (^. typedChannelWithId cid)
1048 mChan <- withMVar (processState proc) $ return . (^. typedChannelWithId cid)
835 case mChan of
1049 case mChan of
836 -- in the unlikely event we know nothing about this channel id,
1050 -- in the unlikely event we know nothing about this channel id,
@@ -840,11 +1054,12 @@ ncEffectLocalPortSend from msg =
840 -- 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
841 -- the channel and we simply ignore the incoming message - this
1055 -- the channel and we simply ignore the incoming message - this
842 ch <- deRefWeak chan'
1056 ch <- deRefWeak chan'
843− forM_ ch $ \chan -> deliverChan msg chan
1057+ forM_ ch $ \chan -> deliverChan node from msg chan
844− where deliverChan :: forall a . Message -> TQueue a -> IO ()
1058+ where deliverChan :: forall a . LocalNode -> SendPortId -> Message -> TQueue a -> IO ()
845− deliverChan (UnencodedMessage _ raw) chan' =
1059+ deliverChan n p (UnencodedMessage _ raw) chan' = do
846 atomically $ writeTQueue chan' ((unsafeCoerce raw) :: a)
1060 atomically $ writeTQueue chan' ((unsafeCoerce raw) :: a)
1061+ trace n (MxReceivedPort p $ unsafeCreateUnencodedMessage raw)
847− deliverChan (EncodedMessage _ _) _ =
1062+ deliverChan _ _ (EncodedMessage _ _) _ =
848 -- this will not happen unless someone screws with Primitives.hs
1063 -- this will not happen unless someone screws with Primitives.hs
849 error "invalid local channel delivery"
1064 error "invalid local channel delivery"
@@ -869,46 +1084,56 @@ ncEffectGetInfo from pid =
869 them = (ProcessIdentifier pid)
1084 them = (ProcessIdentifier pid)
871 node <- ask
1086 node <- ask
872− mProc <- liftIO $
1087+ mProc <- liftIO $ withValidLocalState node
873− withMVar (localState node) $ return . (^. localProcessWithId lpid)
1088+ $ return . (^. localProcessWithId lpid)
874 case mProc of
1089 case mProc of
875 Nothing -> dispatch (isLocal node (ProcessIdentifier from))
1090 Nothing -> dispatch (isLocal node (ProcessIdentifier from))
876− from node (ProcessInfoNone DiedUnknownId)
1091+ from (ProcessInfoNone DiedUnknownId)
877− Just _ -> do
1092+ Just proc -> do
878− itsLinks <- gets (^. linksFor them)
1093+ itsLinks <- Set.map fst . BiMultiMap.lookupBy1st them <$>
879− itsMons <- gets (^. monitorsFor them)
1094+ gets (^. links)
1095+ itsMons <- BiMultiMap.lookupBy1st them <$> gets (^. monitors)
880 registered <- gets (^. registeredHere)
1096 registered <- gets (^. registeredHere)
1097+ size <- liftIO $ queueSize $ processQueue $ proc
882 let reg = registeredNames registered
1099 let reg = registeredNames registered
883 dispatch (isLocal node (ProcessIdentifier from))
1100 dispatch (isLocal node (ProcessIdentifier from))
886 ProcessInfo {
1102 ProcessInfo {
887 infoNode = (processNodeId pid)
1103 infoNode = (processNodeId pid)
888 , infoRegisteredNames = reg
1104 , infoRegisteredNames = reg
889− -- we cannot populate this field
890− , infoMessageQueueLength = Nothing
1105+ , infoMessageQueueLength = size
891− , infoMonitors = Set.toList itsMons
1106+ , infoMonitors = Set.toList itsMons
892− , infoLinks = Set.toList itsLinks
1107+ , infoLinks = Set.toList itsLinks
894− where dispatch :: (Serializable a, Show a)
1109+ where dispatch :: (Serializable a)
896 -> ProcessId
1111 -> ProcessId
899 -> NC ()
1113 -> NC ()
900− dispatch True dest _ pInfo = postAsMessage dest $ pInfo
1114+ dispatch True dest pInfo = postAsMessage dest $ pInfo
901− dispatch False dest node pInfo = do
1115+ dispatch False dest pInfo =
902− liftIO $ sendMessage node
1116+ ncSendToProcess dest $ unsafeCreateUnencodedMessage pInfo
903− (NodeIdentifier (localNodeId node))
904− (ProcessIdentifier dest)
905− WithImplicitReconnect
908 registeredNames = Map.foldlWithKey (\ks k v -> if v == pid
1118 registeredNames = Map.foldlWithKey (\ks k v -> if v == pid
909 then (k:ks)
1119 then (k:ks)
910 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
912--------------------------------------------------------------------------------
1137--------------------------------------------------------------------------------
913-- Auxiliary --
1138-- Auxiliary --
914--------------------------------------------------------------------------------
1139--------------------------------------------------------------------------------
@@ -935,14 +1160,10 @@ notifyDied dest src reason mRef = do
935 throwException dest $ PortLinkException pid reason
1160 throwException dest $ PortLinkException pid reason
936 (False, _, _) ->
1161 (False, _, _) ->
937 -- The change in sender comes from [Unified: Table 10]
1162 -- The change in sender comes from [Unified: Table 10]
938− liftIO $ sendBinary node
939− (NodeIdentifier $ localNodeId node)
940− (NodeIdentifier $ processNodeId dest)
1163+ ncSendToNode (processNodeId dest) $ NCMsg
941− WithImplicitReconnect
943− { ctrlMsgSender = NodeIdentifier (localNodeId node)
1164+ { ctrlMsgSender = NodeIdentifier (localNodeId node)
944− , ctrlMsgSignal = Died src reason
1165+ , ctrlMsgSignal = Died src reason
947-- | [Unified: Table 8]
1168-- | [Unified: Table 8]
948destNid :: ProcessSignal -> Maybe NodeId
1169destNid :: ProcessSignal -> Maybe NodeId
@@ -954,6 +1175,7 @@ destNid (Spawn _ _) = Nothing
954destNid (Register _ _ _ _) = Nothing
1175destNid (Register _ _ _ _) = Nothing
955destNid (WhereIs _) = Nothing
1176destNid (WhereIs _) = Nothing
956destNid (NamedSend _ _) = Nothing
1177destNid (NamedSend _ _) = Nothing
1178+destNid (UnreliableSend _ _) = Nothing
957-- 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,
958-- 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
959-- care of notifying remote nodes
1181-- care of notifying remote nodes
@@ -961,6 +1183,7 @@ destNid (Died _ _) = Nothing
961destNid (Kill pid _) = Just $ processNodeId pid
1183destNid (Kill pid _) = Just $ processNodeId pid
962destNid (Exit pid _) = Just $ processNodeId pid
1184destNid (Exit pid _) = Just $ processNodeId pid
963destNid (GetInfo pid) = Just $ processNodeId pid
1185destNid (GetInfo pid) = Just $ processNodeId pid
1186+destNid (GetNodeStats nid) = Just nid
964destNid (LocalSend pid _) = Just $ processNodeId pid
1187destNid (LocalSend pid _) = Just $ processNodeId pid
965destNid (LocalPortSend cid _) = Just $ processNodeId (sendPortProcessId cid)
1188destNid (LocalPortSend cid _) = Just $ processNodeId (sendPortProcessId cid)
966destNid (SigShutdown) = Nothing
1189destNid (SigShutdown) = Nothing
@@ -979,7 +1202,7 @@ unClosure closure = do
979isValidLocalIdentifier :: Identifier -> NC Bool
1202isValidLocalIdentifier :: Identifier -> NC Bool
980isValidLocalIdentifier ident = do
1203isValidLocalIdentifier ident = do
981 node <- ask
1204 node <- ask
982− liftIO . withMVar (localState node) $ \nSt ->
1205+ liftIO . withValidLocalState node $ \nSt ->
983 case ident of
1206 case ident of
984 NodeIdentifier nid ->
1207 NodeIdentifier nid ->
985 return $ nid == localNodeId node
1208 return $ nid == localNodeId node
@@ -1004,30 +1227,33 @@ postAsMessage pid = postMessage pid . createUnencodedMessage
1005postMessage :: ProcessId -> Message -> NC ()
1228postMessage :: ProcessId -> Message -> NC ()
1006postMessage pid msg = do
1229postMessage pid msg = do
1007− withLocalProc pid $ \p -> enqueue (processQueue p) msg
1231+ liftIO $ withLocalProc node pid $ \p -> enqueue (processQueue p) msg
1009throwException :: Exception e => ProcessId -> e -> NC ()
1233throwException :: Exception e => ProcessId -> e -> NC ()
1010−throwException pid e = withLocalProc pid $ \p ->
1234+throwException pid e = do
1011− throwTo (processThread p) e
1013−withLocalProc :: ProcessId -> (LocalProcess -> IO ()) -> NC ()
1014−withLocalProc pid p = do
1015 node <- ask
1235 node <- ask
1016− liftIO $ do
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.
1239+ liftIO $ withLocalProc node pid $ \p -> throwTo (processThread p) e
1241+withLocalProc :: LocalNode -> ProcessId -> (LocalProcess -> IO ()) -> IO ()
1242+withLocalProc node pid p =
1017− -- By [Unified: table 6, rule missing_process] messages to dead processes
1243+ -- By [Unified: table 6, rule missing_process] messages to dead processes
1018− -- can silently be dropped
1244+ -- can silently be dropped
1019− let lpid = processLocalId pid
1245+ let lpid = processLocalId pid in do
1246+ join $ withValidLocalState node $ \vst ->
1020− mProc <- withMVar (localState node) $ return . (^. localProcessWithId lpid)
1247+ return $ forM_ (vst ^. localProcessWithId lpid) p
1023--------------------------------------------------------------------------------
1249--------------------------------------------------------------------------------
1024-- Accessors --
1250-- Accessors --
1025--------------------------------------------------------------------------------
1251--------------------------------------------------------------------------------
1027−links :: Accessor NCState (Map Identifier (Set ProcessId))
1253+links :: Accessor NCState (BiMultiMap Identifier ProcessId ())
1028links = accessor _links (\ls st -> st { _links = ls })
1254links = accessor _links (\ls st -> st { _links = ls })
1030−monitors :: Accessor NCState (Map Identifier (Set (ProcessId, MonitorRef)))
1256+monitors :: Accessor NCState (BiMultiMap Identifier ProcessId MonitorRef)
1031monitors = accessor _monitors (\ms st -> st { _monitors = ms })
1257monitors = accessor _monitors (\ms st -> st { _monitors = ms })
1033registeredHere :: Accessor NCState (Map String ProcessId)
1259registeredHere :: Accessor NCState (Map String ProcessId)
@@ -1036,12 +1262,6 @@ registeredHere = accessor _registeredHere (\ry st -> st { _registeredHere = ry }
1036registeredOnNodes :: Accessor NCState (Map ProcessId [(NodeId, Int)])
1262registeredOnNodes :: Accessor NCState (Map ProcessId [(NodeId, Int)])
1037registeredOnNodes = accessor _registeredOnNodes (\ry st -> st { _registeredOnNodes = ry })
1263registeredOnNodes = accessor _registeredOnNodes (\ry st -> st { _registeredOnNodes = ry })
1039−linksFor :: Identifier -> Accessor NCState (Set ProcessId)
1040−linksFor ident = links >>> DAC.mapDefault Set.empty ident
1042−monitorsFor :: Identifier -> Accessor NCState (Set (ProcessId, MonitorRef))
1043−monitorsFor ident = monitors >>> DAC.mapDefault Set.empty ident
1045registeredHereFor :: String -> Accessor NCState (Maybe ProcessId)
1265registeredHereFor :: String -> Accessor NCState (Maybe ProcessId)
1046registeredHereFor ident = registeredHere >>> DAC.mapMaybe ident
1266registeredHereFor ident = registeredHere >>> DAC.mapMaybe ident
@@ -1068,10 +1288,12 @@ registeredOnNodesFor ident = registeredOnNodes >>> DAC.mapMaybe ident
1068-- * the notifications for typed channels to that process.
1288-- * the notifications for typed channels to that process.
1070-- 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.
1291+splitNotif :: (Ord a, Ord v)
1071−splitNotif :: Identifier
1292+ => Identifier
1072− -> Map Identifier a
1293+ -> BiMultiMap Identifier a v
1073− -> (Map Identifier a, Map Identifier a)
1294+ -> (Map Identifier (Set (a,v)), BiMultiMap Identifier a v)
1074−splitNotif ident = Map.partitionWithKey (\k !_v -> ident `impliesDeathOf` k)
1295+splitNotif ident =
1296+ BiMultiMap.partitionWithKeyBy1st (\k !_v -> ident `impliesDeathOf` k)
1076--------------------------------------------------------------------------------
1298--------------------------------------------------------------------------------
1077-- Auxiliary --
1299-- Auxiliary --