git.iko.soy/distributed-process
on b3fa53c help sign in
distributed-process/src/Control/Distributed/Process/Node.hs
1134 lines 45 kB history raw
1-- | Local nodes
2--
3module Control.Distributed.Process.Node
4 ( LocalNode
5 , newLocalNode
6 , closeLocalNode
7 , forkProcess
8 , runProcess
9 , initRemoteTable
10 , localNodeId
11 ) where
13-- TODO: Calls to 'sendBinary' and co (by the NC) may stall the node controller.
15#if ! MIN_VERSION_base(4,6,0)
16import Prelude hiding (catch)
17#endif
19import System.IO (fixIO, hPutStrLn, stderr)
20import System.Mem.Weak (Weak, deRefWeak)
21import qualified Data.ByteString.Lazy as BSL (fromChunks)
22import Data.Binary (decode)
23import Data.Map (Map)
24import qualified Data.Map as Map
25 ( empty
26 , toList
27 , fromList
28 , filter
29 , partitionWithKey
30 , elems
31 , filterWithKey
32 , foldlWithKey
33 )
34import Data.Set (Set)
35import qualified Data.Set as Set
36 ( empty
37 , insert
38 , delete
39 , member
40 , toList
41 )
42import Data.Foldable (forM_)
43import Data.Maybe (isJust, fromJust, isNothing, catMaybes)
44import Data.Typeable (Typeable)
45import Control.Category ((>>>))
46import Control.Applicative ((<$>))
47import Control.Monad (void, when)
48import Control.Monad.IO.Class (MonadIO, liftIO)
49import Control.Monad.State.Strict (MonadState, StateT, evalStateT, gets)
50import qualified Control.Monad.State.Strict as StateT (get, put)
51import Control.Monad.Reader (MonadReader, ReaderT, runReaderT, ask)
52import Control.Exception
53 ( throwIO
54 , AsyncException(ThreadKilled)
55 , SomeException
56 , Exception
57 , throwTo
58 )
59import qualified Control.Exception as Exception (catch, finally)
60import Control.Concurrent (forkIO)
61import Control.Distributed.Process.Internal.StrictMVar
62 ( newMVar
63 , withMVar
64 , modifyMVar
65 , modifyMVar_
66 , newEmptyMVar
67 , putMVar
68 , takeMVar
69 )
70import Control.Concurrent.Chan (newChan, writeChan, readChan)
71import qualified Control.Concurrent.MVar as MVar (newEmptyMVar, takeMVar)
72import Control.Concurrent.STM
73 ( atomically
74 )
75import Control.Distributed.Process.Internal.CQueue
76 ( CQueue
77 , enqueue
78 , newCQueue
79 , mkWeakCQueue
80 )
81import qualified Network.Transport as NT
82 ( Transport
83 , EndPoint
84 , newEndPoint
85 , receive
86 , Event(..)
87 , EventErrorCode(..)
88 , TransportError(..)
89 , address
90 , closeEndPoint
91 , ConnectionId
92 , Connection
93 , close
94 , EndPointAddress
95 , Reliability(ReliableOrdered)
96 )
97import Data.Accessor (Accessor, accessor, (^.), (^=), (^:))
98import System.Random (randomIO)
99import Control.Distributed.Static (RemoteTable, Closure)
100import qualified Control.Distributed.Static as Static
101 ( unclosure
102 , initRemoteTable
103 )
104import Control.Distributed.Process.Internal.Types
105 ( NodeId(..)
106 , LocalProcessId(..)
107 , ProcessId(..)
108 , LocalNode(..)
109 , Tracer(..)
110 , LocalNodeState(..)
111 , LocalProcess(..)
112 , LocalProcessState(..)
113 , Process(..)
114 , DiedReason(..)
115 , NCMsg(..)
116 , ProcessSignal(..)
117 , localPidCounter
118 , localPidUnique
119 , localProcessWithId
120 , localConnections
121 , forever'
122 , MonitorRef(..)
123 , ProcessMonitorNotification(..)
124 , NodeMonitorNotification(..)
125 , PortMonitorNotification(..)
126 , ProcessExitException(..)
127 , ProcessLinkException(..)
128 , NodeLinkException(..)
129 , PortLinkException(..)
130 , DidUnmonitor(..)
131 , DidUnlinkProcess(..)
132 , DidUnlinkNode(..)
133 , DidUnlinkPort(..)
134 , SpawnRef
135 , DidSpawn(..)
136 , Message(..)
137 , TypedChannel(..)
138 , Identifier(..)
139 , nodeOf
140 , ProcessInfo(..)
141 , ProcessInfoNone(..)
142 , SendPortId(..)
143 , typedChannelWithId
144 , RegisterReply(..)
145 , WhereIsReply(..)
146 , payloadToMessage
147 , createUnencodedMessage
148 , runLocalProcess
149 , firstNonReservedProcessId
150 , ImplicitReconnect(WithImplicitReconnect)
151 )
152import Control.Distributed.Process.Internal.Trace.Primitives
153 ( traceLogFmt
154 , traceEvent
155 )
156import Control.Distributed.Process.Internal.Trace.Types
157 ( TraceArg(..)
158 , TraceEvent(..)
159 )
160import Control.Distributed.Process.Internal.Trace.Tracer
161 ( traceController
162 , defaultTracer
163 )
164import Control.Distributed.Process.Serializable (Serializable)
165import Control.Distributed.Process.Internal.Messaging
166 ( sendBinary
167 , sendMessage
168 , closeImplicitReconnections
169 , impliesDeathOf
170 )
171import Control.Distributed.Process.Internal.Primitives
172 ( register
173 , finally
174 , receiveWait
175 , match
176 , sendChan
177 )
178import Control.Distributed.Process.Internal.Types (SendPort)
179import qualified Control.Distributed.Process.Internal.Closure.BuiltIn as BuiltIn (remoteTable)
180import Control.Distributed.Process.Internal.WeakTQueue (TQueue, writeTQueue)
181import qualified Control.Distributed.Process.Internal.StrictContainerAccessors as DAC
182 ( mapMaybe
183 , mapDefault
184 )
185import System.Environment (getEnv)
186import Unsafe.Coerce
188--------------------------------------------------------------------------------
189-- Initialization --
190--------------------------------------------------------------------------------
192initRemoteTable :: RemoteTable
193initRemoteTable = BuiltIn.remoteTable Static.initRemoteTable
195-- | Initialize a new local node.
196newLocalNode :: NT.Transport -> RemoteTable -> IO LocalNode
197newLocalNode transport rtable = do
198 mEndPoint <- NT.newEndPoint transport
199 case mEndPoint of
200 Left ex -> throwIO ex
201 Right endPoint -> do
202 localNode <- createBareLocalNode endPoint rtable
203 startServiceProcesses localNode
204 return localNode
206-- | Create a new local node (without any service processes running)
207createBareLocalNode :: NT.EndPoint -> RemoteTable -> IO LocalNode
208createBareLocalNode endPoint rtable = do
209 unq <- randomIO
210 state <- newMVar LocalNodeState
211 { _localProcesses = Map.empty
212 , _localPidCounter = firstNonReservedProcessId
213 , _localPidUnique = unq
214 , _localConnections = Map.empty
215 }
216 ctrlChan <- newChan
217 let node = LocalNode { localNodeId = NodeId $ NT.address endPoint
218 , localEndPoint = endPoint
219 , localState = state
220 , localCtrlChan = ctrlChan
221 , localTracer = InactiveTracer
222 , remoteTable = rtable
223 }
224 tracedNode <- startTracing node
226 -- Once the NC terminates, the endpoint isn't much use,
227 void $ forkIO $ Exception.finally (runNodeController tracedNode)
228 (NT.closeEndPoint (localEndPoint node))
230 -- whilst a closed/failing endpoint will terminate the NC
231 void $ forkIO $ Exception.finally (handleIncomingMessages tracedNode)
232 (stopNC node)
234 return tracedNode
235 where
236 stopNC node =
237 writeChan (localCtrlChan node) NCMsg
238 { ctrlMsgSender = NodeIdentifier (localNodeId node)
239 , ctrlMsgSignal = SigShutdown
240 }
242startTracing :: LocalNode -> IO LocalNode
243startTracing node =
244 Exception.catch
245 (getEnv "DISTRIBUTED_PROCESS_TRACE_ENABLED" >> procTracer node)
246 (\(_ :: IOError) -> return node)
247 where procTracer :: LocalNode -> IO LocalNode
248 procTracer n = do
249 -- see note [tracer/forkProcess races]
250 mv <- MVar.newEmptyMVar
251 pid <- forkProcess n (traceController mv)
252 wqRef <- MVar.takeMVar mv
253 return n { localTracer = (ActiveTracer pid wqRef) }
255startDefaultTracer :: LocalNode -> IO (Maybe ProcessId)
256startDefaultTracer node =
257 Exception.catch
258 (getEnv "DISTRIBUTED_PROCESS_TRACE_ENABLED" >> go node)
259 (\(_ :: IOError) -> return Nothing)
260 where go :: LocalNode -> IO (Maybe ProcessId)
261 go node' = do
262 pid <- forkProcess node' defaultTracer
263 return (Just pid)
265-- | Start and register the service processes on a node
266startServiceProcesses :: LocalNode -> IO ()
267startServiceProcesses node = do
268 -- tracing /spawns/ relies on the tracer being enabled, but we start
269 -- the default tracer first, even though it might @nsend@ to the logger
270 -- before /that/ process has started - this is a totally harmless race
271 mPid <- startDefaultTracer node
272 case mPid of
273 Nothing -> return ()
274 Just pid ->
275 runProcess node (register "tracer" pid >> register "tracer.initial" pid)
277 logger <- forkProcess node loop
278 runProcess node $ register "logger" logger
279 where
280 loop = do
281 receiveWait
282 [ match $ \((time, pid, string) ::(String, ProcessId, String)) -> do
283 liftIO . hPutStrLn stderr $ time ++ " " ++ show pid ++ ": " ++ string
284 loop
285 , match $ \((time, string) :: (String, String)) -> do
286 -- this is a 'trace' message from the local node tracer
287 liftIO . hPutStrLn stderr $ time ++ " [trace] " ++ string
288 loop
289 , match $ \(ch :: SendPort ()) -> -- a shutdown request
290 sendChan ch ()
291 ]
293-- | Force-close a local node
294--
295-- TODO: for now we just close the associated endpoint
296closeLocalNode :: LocalNode -> IO ()
297closeLocalNode node =
298 -- TODO: close all our processes, surely!?
299 NT.closeEndPoint (localEndPoint node)
301-- | Run a process on a local node and wait for it to finish
302runProcess :: LocalNode -> Process () -> IO ()
303runProcess node proc = do
304 done <- newEmptyMVar
305 void $ forkProcess node (proc `finally` liftIO (putMVar done ()))
306 takeMVar done
308-- | Spawn a new process on a local node
309forkProcess :: LocalNode -> Process () -> IO ProcessId
310forkProcess node proc = modifyMVar (localState node) startProcess
311 where
312 startProcess :: LocalNodeState -> IO (LocalNodeState, ProcessId)
313 startProcess st = do
314 let lpid = LocalProcessId { lpidCounter = st ^. localPidCounter
315 , lpidUnique = st ^. localPidUnique
316 }
317 let pid = ProcessId { processNodeId = localNodeId node
318 , processLocalId = lpid
319 }
320 pst <- newMVar LocalProcessState { _monitorCounter = 0
321 , _spawnCounter = 0
322 , _channelCounter = 0
323 , _typedChannels = Map.empty
324 }
325 queue <- newCQueue
326 weakQueue <- mkWeakCQueue queue (return ())
327 (_, lproc) <- fixIO $ \ ~(tid, _) -> do
328 let lproc = LocalProcess { processQueue = queue
329 , processWeakQ = weakQueue
330 , processId = pid
331 , processState = pst
332 , processThread = tid
333 , processNode = node
334 }
335 tid' <- forkIO $ do
336 reason <- Exception.catch
337 (runLocalProcess lproc proc >> return DiedNormal)
338 (return . DiedException . (show :: SomeException -> String))
340 -- [Unified: Table 4, rules termination and exiting]
341 modifyMVar_ (localState node) (cleanupProcess pid)
342 writeChan (localCtrlChan node) NCMsg
343 { ctrlMsgSender = ProcessIdentifier pid
344 , ctrlMsgSignal = Died (ProcessIdentifier pid) reason
345 }
346 return (tid', lproc)
349 -- see note [tracer/forkProcess races]
350 trace node (TraceEvSpawned pid)
352 if lpidCounter lpid == maxBound
353 then do
354 newUnique <- randomIO
355 return ( (localProcessWithId lpid ^= Just lproc)
356 . (localPidCounter ^= firstNonReservedProcessId)
357 . (localPidUnique ^= newUnique)
358 $ st
359 , pid
360 )
361 else
362 return ( (localProcessWithId lpid ^= Just lproc)
363 . (localPidCounter ^: (+ 1))
364 $ st
365 , pid
366 )
368 cleanupProcess :: ProcessId -> LocalNodeState -> IO LocalNodeState
369 cleanupProcess pid st = do
370 let pid' = ProcessIdentifier pid
371 let (affected, unaffected) = Map.partitionWithKey (\(fr, _to) !_v -> impliesDeathOf pid' fr) (st ^. localConnections)
372 mapM_ (NT.close . fst) (Map.elems affected)
373 return $ (localProcessWithId (processLocalId pid) ^= Nothing)
374 . (localConnections ^= unaffected)
375 $ st
377-- note [tracer/forkProcess races]
378--
379-- Our startTracing function uses forkProcess to start the trace controller
380-- process, and of course forkProcess attempts to call traceEvent once the
381-- process has started. This is harmless, as the localTracer is not updated
382-- until /after/ the initial forkProcess completes, so the first call to
383-- traceEvent behaves as if tracing were disabled (i.e., it is ignored).
384--
386--------------------------------------------------------------------------------
387-- Handle incoming messages --
388--------------------------------------------------------------------------------
390type IncomingConnection = (NT.EndPointAddress, IncomingTarget)
392data IncomingTarget =
393 Uninit
394 | ToProc ProcessId (Weak (CQueue Message))
395 | ToChan TypedChannel
396 | ToNode
398data ConnectionState = ConnectionState {
399 _incoming :: !(Map NT.ConnectionId IncomingConnection)
400 , _incomingFrom :: !(Map NT.EndPointAddress (Set NT.ConnectionId))
401 }
403initConnectionState :: ConnectionState
404initConnectionState = ConnectionState {
405 _incoming = Map.empty
406 , _incomingFrom = Map.empty
407 }
409incoming :: Accessor ConnectionState (Map NT.ConnectionId IncomingConnection)
410incoming = accessor _incoming (\conns st -> st { _incoming = conns })
412incomingAt :: NT.ConnectionId -> Accessor ConnectionState (Maybe IncomingConnection)
413incomingAt cid = incoming >>> DAC.mapMaybe cid
415incomingFrom :: NT.EndPointAddress -> Accessor ConnectionState (Set NT.ConnectionId)
416incomingFrom addr = aux >>> DAC.mapDefault Set.empty addr
417 where
418 aux = accessor _incomingFrom (\fr st -> st { _incomingFrom = fr })
420handleIncomingMessages :: LocalNode -> IO ()
421handleIncomingMessages node = go initConnectionState
422 where
423 go :: ConnectionState -> IO ()
424 go !st = do
425 event <- NT.receive endpoint
426 case event of
427 NT.ConnectionOpened cid rel theirAddr ->
428 if rel == NT.ReliableOrdered
429 then
430 trace node (TraceEvConnected cid)
431 >> go (
432 (incomingAt cid ^= Just (theirAddr, Uninit))
433 . (incomingFrom theirAddr ^: Set.insert cid)
434 $ st
435 )
436 else invalidRequest cid st
437 NT.Received cid payload ->
438 case st ^. incomingAt cid of
439 Just (_, ToProc pid weakQueue) -> do
440 mQueue <- deRefWeak weakQueue
441 forM_ mQueue $ \queue -> do
442 -- TODO: if we find that the queue is Nothing, should we remove
443 -- it from the NC state? (and same for channels, below)
444 let msg = payloadToMessage payload
445 enqueue queue msg -- 'enqueue' is strict
446 trace node (TraceEvReceived pid msg)
447 go st
448 Just (_, ToChan (TypedChannel chan')) -> do
449 mChan <- deRefWeak chan'
450 -- If mChan is Nothing, the process has given up the read end of
451 -- the channel and we simply ignore the incoming message
452 forM_ mChan $ \chan -> atomically $
453 -- We make sure the message is fully decoded when it is enqueued
454 writeTQueue chan $! decode (BSL.fromChunks payload)
455 go st
456 Just (_, ToNode) -> do
457 let ctrlMsg = decode . BSL.fromChunks $ payload
458 writeChan ctrlChan $! ctrlMsg
459 go st
460 Just (src, Uninit) ->
461 case decode (BSL.fromChunks payload) of
462 ProcessIdentifier pid -> do
463 let lpid = processLocalId pid
464 mProc <- withMVar state $ return . (^. localProcessWithId lpid)
465 case mProc of
466 Just proc ->
467 go (incomingAt cid ^= Just (src, ToProc pid (processWeakQ proc)) $ st)
468 Nothing ->
469 invalidRequest cid st
470 SendPortIdentifier chId -> do
471 let lcid = sendPortLocalId chId
472 lpid = processLocalId (sendPortProcessId chId)
473 mProc <- withMVar state $ return . (^. localProcessWithId lpid)
474 case mProc of
475 Just proc -> do
476 mChannel <- withMVar (processState proc) $ return . (^. typedChannelWithId lcid)
477 case mChannel of
478 Just channel ->
479 go (incomingAt cid ^= Just (src, ToChan channel) $ st)
480 Nothing ->
481 invalidRequest cid st
482 Nothing ->
483 invalidRequest cid st
484 NodeIdentifier nid ->
485 if nid == localNodeId node
486 then go (incomingAt cid ^= Just (src, ToNode) $ st)
487 else invalidRequest cid st
488 Nothing ->
489 invalidRequest cid st
490 NT.ConnectionClosed cid ->
491 case st ^. incomingAt cid of
492 Nothing ->
493 invalidRequest cid st
494 Just (src, _) -> do
495 trace node (TraceEvDisconnected cid)
496 go ( (incomingAt cid ^= Nothing)
497 . (incomingFrom src ^: Set.delete cid)
498 $ st
499 )
500 NT.ErrorEvent (NT.TransportError (NT.EventConnectionLost theirAddr) _) -> do
501 -- [Unified table 9, rule node_disconnect]
502 let nid = NodeIdentifier $ NodeId theirAddr
503 writeChan ctrlChan NCMsg
504 { ctrlMsgSender = nid
505 , ctrlMsgSignal = Died nid DiedDisconnect
506 }
507 let notLost k = not (k `Set.member` (st ^. incomingFrom theirAddr))
508 closeImplicitReconnections node nid
509 go ( (incomingFrom theirAddr ^= Set.empty)
510 . (incoming ^: Map.filterWithKey (const . notLost))
511 $ st
512 )
513 NT.ErrorEvent (NT.TransportError NT.EventEndPointFailed str) ->
514 fail $ "Cloud Haskell fatal error: end point failed: " ++ str
515 NT.ErrorEvent (NT.TransportError NT.EventTransportFailed str) ->
516 fail $ "Cloud Haskell fatal error: transport failed: " ++ str
517 NT.EndPointClosed ->
518 return ()
519 NT.ReceivedMulticast _ _ ->
520 -- If we received a multicast message, something went horribly wrong
521 -- and we just give up
522 fail "Cloud Haskell fatal error: received unexpected multicast"
524 invalidRequest :: NT.ConnectionId -> ConnectionState -> IO ()
525 invalidRequest cid st = do
526 -- TODO: We should treat this as a fatal error on the part of the remote
527 -- node. That is, we should report the remote node as having died, and we
528 -- should close incoming connections (this requires a Transport layer
529 -- extension).
530 traceEventFmtIO node "" [(TraceStr " [network] invalid request: "),
531 (Trace cid)]
532 go ( incomingAt cid ^= Nothing
533 $ st
534 )
536 state = localState node
537 endpoint = localEndPoint node
538 ctrlChan = localCtrlChan node
540--------------------------------------------------------------------------------
541-- Top-level access to the node controller --
542--------------------------------------------------------------------------------
544runNodeController :: LocalNode -> IO ()
545runNodeController =
546 runReaderT (evalStateT (unNC nodeController) initNCState)
548--------------------------------------------------------------------------------
549-- Internal data types --
550--------------------------------------------------------------------------------
552data NCState = NCState
553 { -- Mapping from remote processes to linked local processes
554 _links :: !(Map Identifier (Set ProcessId))
555 -- Mapping from remote processes to monitoring local processes
556 , _monitors :: !(Map Identifier (Set (ProcessId, MonitorRef)))
557 -- Process registry: names and where they live, mapped to the PIDs
558 , _registeredHere :: !(Map String ProcessId)
559 , _registeredOnNodes :: !(Map ProcessId [(NodeId,Int)])
560 }
562newtype NC a = NC { unNC :: StateT NCState (ReaderT LocalNode IO) a }
563 deriving (Functor, Monad, MonadIO, MonadState NCState, MonadReader LocalNode)
565initNCState :: NCState
566initNCState = NCState { _links = Map.empty
567 , _monitors = Map.empty
568 , _registeredHere = Map.empty
569 , _registeredOnNodes = Map.empty
570 }
572-- | Thrown in response to the user invoking 'kill' (see Primitives.hs). This
573-- type is deliberately not exported so it cannot be caught explicitly.
574data ProcessKillException =
575 ProcessKillException !ProcessId !String
576 deriving (Typeable)
578instance Exception ProcessKillException
579instance Show ProcessKillException where
580 show (ProcessKillException pid reason) =
581 "killed-by=" ++ show pid ++ ",reason=" ++ reason
583--------------------------------------------------------------------------------
584-- Tracing/Debugging --
585--------------------------------------------------------------------------------
587-- [Issue #104]
589traceNotifyDied :: LocalNode -> Identifier -> DiedReason -> NC ()
590traceNotifyDied node ident reason =
591 -- TODO: sendPortDied notifications
592 liftIO $ withLocalTracer node $ \t ->
593 case ident of
594 (NodeIdentifier nid) -> traceEvent t (TraceEvNodeDied nid reason)
595 (ProcessIdentifier pid) -> traceEvent t (TraceEvDied pid reason)
596 _ -> return ()
598traceEventFmtIO :: LocalNode
599 -> String
600 -> [TraceArg]
601 -> IO ()
602traceEventFmtIO node fmt args =
603 withLocalTracer node $ \t -> traceLogFmt t fmt args
605trace :: LocalNode -> TraceEvent -> IO ()
606trace node ev = withLocalTracer node $ \t -> traceEvent t ev
608withLocalTracer :: LocalNode -> (Tracer -> IO ()) -> IO ()
609withLocalTracer node act = act (localTracer node)
611--------------------------------------------------------------------------------
612-- Core functionality --
613--------------------------------------------------------------------------------
615-- [Unified: Table 7]
616nodeController :: NC ()
617nodeController = do
618 node <- ask
619 forever' $ do
620 msg <- liftIO $ readChan (localCtrlChan node)
622 -- [Unified: Table 7, rule nc_forward]
623 case destNid (ctrlMsgSignal msg) of
624 Just nid' | nid' /= localNodeId node ->
625 liftIO $ sendBinary node
626 (ctrlMsgSender msg)
627 (NodeIdentifier nid')
628 WithImplicitReconnect
629 msg
630 _ ->
631 return ()
633 case msg of
634 NCMsg (ProcessIdentifier from) (Link them) ->
635 ncEffectMonitor from them Nothing
636 NCMsg (ProcessIdentifier from) (Monitor ref) ->
637 ncEffectMonitor from (monitorRefIdent ref) (Just ref)
638 NCMsg (ProcessIdentifier from) (Unlink them) ->
639 ncEffectUnlink from them
640 NCMsg (ProcessIdentifier from) (Unmonitor ref) ->
641 ncEffectUnmonitor from ref
642 NCMsg _from (Died ident reason) ->
643 ncEffectDied ident reason
644 NCMsg (ProcessIdentifier from) (Spawn proc ref) ->
645 ncEffectSpawn from proc ref
646 NCMsg (ProcessIdentifier from) (Register label atnode pid force) ->
647 ncEffectRegister from label atnode pid force
648 NCMsg (ProcessIdentifier from) (WhereIs label) ->
649 ncEffectWhereIs from label
650 NCMsg _ (NamedSend label msg') ->
651 ncEffectNamedSend label msg'
652 NCMsg _ (LocalSend to msg') ->
653 ncEffectLocalSend node to msg'
654 NCMsg _ (LocalPortSend to msg') ->
655 ncEffectLocalPortSend to msg'
656 NCMsg (ProcessIdentifier from) (Kill to reason) ->
657 ncEffectKill from to reason
658 NCMsg (ProcessIdentifier from) (Exit to reason) ->
659 ncEffectExit from to reason
660 NCMsg (ProcessIdentifier from) (GetInfo pid) ->
661 ncEffectGetInfo from pid
662 NCMsg _ SigShutdown ->
663 liftIO $ do
664 NT.closeEndPoint (localEndPoint node)
665 `Exception.finally` throwIO ThreadKilled
666 -- ThreadKilled seems to make more sense than fail/error here
667 unexpected ->
668 error $ "nodeController: unexpected message " ++ show unexpected
670-- [Unified: Table 10]
671ncEffectMonitor :: ProcessId -- ^ Who's watching?
672 -> Identifier -- ^ Who's being watched?
673 -> Maybe MonitorRef -- ^ 'Nothing' to link
674 -> NC ()
675ncEffectMonitor from them mRef = do
676 node <- ask
677 shouldLink <-
678 if not (isLocal node them)
679 then return True
680 else isValidLocalIdentifier them
681 case (shouldLink, isLocal node (ProcessIdentifier from)) of
682 (True, _) -> -- [Unified: first rule]
683 case mRef of
684 Just ref -> modify' $ monitorsFor them ^: Set.insert (from, ref)
685 Nothing -> modify' $ linksFor them ^: Set.insert from
686 (False, True) -> -- [Unified: second rule]
687 notifyDied from them DiedUnknownId mRef
688 (False, False) -> -- [Unified: third rule]
689 -- TODO: this is the right sender according to the Unified semantics,
690 -- but perhaps having 'them' as the sender would make more sense
691 -- (see also: notifyDied)
692 liftIO $ sendBinary node
693 (NodeIdentifier $ localNodeId node)
694 (NodeIdentifier $ processNodeId from)
695 WithImplicitReconnect
696 NCMsg
697 { ctrlMsgSender = NodeIdentifier (localNodeId node)
698 , ctrlMsgSignal = Died them DiedUnknownId
699 }
701-- [Unified: Table 11]
702ncEffectUnlink :: ProcessId -> Identifier -> NC ()
703ncEffectUnlink from them = do
704 node <- ask
705 when (isLocal node (ProcessIdentifier from)) $
706 case them of
707 ProcessIdentifier pid ->
708 postAsMessage from $ DidUnlinkProcess pid
709 NodeIdentifier nid ->
710 postAsMessage from $ DidUnlinkNode nid
711 SendPortIdentifier cid ->
712 postAsMessage from $ DidUnlinkPort cid
713 modify' $ linksFor them ^: Set.delete from
715-- [Unified: Table 11]
716ncEffectUnmonitor :: ProcessId -> MonitorRef -> NC ()
717ncEffectUnmonitor from ref = do
718 node <- ask
719 when (isLocal node (ProcessIdentifier from)) $
720 postAsMessage from $ DidUnmonitor ref
721 modify' $ monitorsFor (monitorRefIdent ref) ^: Set.delete (from, ref)
723-- [Unified: Table 12]
724ncEffectDied :: Identifier -> DiedReason -> NC ()
725ncEffectDied ident reason = do
726 node <- ask
727 traceNotifyDied node ident reason
728 (affectedLinks, unaffectedLinks) <- gets (splitNotif ident . (^. links))
729 (affectedMons, unaffectedMons) <- gets (splitNotif ident . (^. monitors))
731-- _registry :: !(Map (String,NodeId) ProcessId)
733 let localOnly = case ident of NodeIdentifier _ -> True ; _ -> False
735 forM_ (Map.toList affectedLinks) $ \(them, uss) ->
736 forM_ uss $ \us ->
737 when (localOnly <= isLocal node (ProcessIdentifier us)) $
738 notifyDied us them reason Nothing
740 forM_ (Map.toList affectedMons) $ \(them, refs) ->
741 forM_ refs $ \(us, ref) ->
742 when (localOnly <= isLocal node (ProcessIdentifier us)) $
743 notifyDied us them reason (Just ref)
745 modify' $ (links ^= unaffectedLinks) . (monitors ^= unaffectedMons)
747 modify' $ registeredHere ^: Map.filter (\pid -> not $ ident `impliesDeathOf` ProcessIdentifier pid)
749 remaining <- fmap Map.toList (gets (^. registeredOnNodes)) >>=
750 mapM (\(pid,nidlist) ->
751 case ident `impliesDeathOf` ProcessIdentifier pid of
752 True ->
753 do forM_ nidlist $ \(nid,_) ->
754 when (not $ isLocal node (NodeIdentifier nid))
755 (forwardNameDeath node nid)
756 return Nothing
757 False -> return $ Just (pid,nidlist) )
758 modify' $ registeredOnNodes ^= (Map.fromList (catMaybes remaining))
759 where
760 forwardNameDeath node nid =
761 liftIO $ sendBinary node
762 (NodeIdentifier $ localNodeId node)
763 (NodeIdentifier $ nid)
764 WithImplicitReconnect
765 NCMsg
766 { ctrlMsgSender = NodeIdentifier (localNodeId node)
767 , ctrlMsgSignal = Died ident reason
768 }
770-- [Unified: Table 13]
771ncEffectSpawn :: ProcessId -> Closure (Process ()) -> SpawnRef -> NC ()
772ncEffectSpawn pid cProc ref = do
773 mProc <- unClosure cProc
774 -- If the closure does not exist, we spawn a process that throws an exception
775 -- This allows the remote node to find out what's happening
776 -- TODO:
777 let proc = case mProc of
778 Left err -> fail $ "Error: Could not resolve closure: " ++ err
779 Right p -> p
780 node <- ask
781 pid' <- liftIO $ forkProcess node proc
782 liftIO $ sendMessage node
783 (NodeIdentifier (localNodeId node))
784 (ProcessIdentifier pid)
785 WithImplicitReconnect
786 (DidSpawn ref pid')
788-- Unified semantics does not explicitly describe how to implement 'register',
789-- but mentions it's "very similar to nsend" (Table 14)
790-- We send a response indicated if the operation is invalid
791ncEffectRegister :: ProcessId -> String -> NodeId -> Maybe ProcessId -> Bool -> NC ()
792ncEffectRegister from label atnode mPid reregistration = do
793 node <- ask
794 currentVal <- gets (^. registeredHereFor label)
795 isOk <-
796 case mPid of
797 Nothing -> -- unregister request
798 return $ isJust currentVal
799 Just thepid -> -- register request
800 do isvalidlocal <- isValidLocalIdentifier (ProcessIdentifier thepid)
801 return $ (isNothing currentVal /= reregistration) &&
802 (not (isLocal node (ProcessIdentifier thepid) ) || isvalidlocal )
803 if isLocal node (NodeIdentifier atnode)
804 then do when (isOk) $
805 do modify' $ registeredHereFor label ^= mPid
806 updateRemote node currentVal mPid
807 case mPid of
808 (Just p) -> liftIO $ trace node (TraceEvRegistered p label)
809 Nothing -> liftIO $ trace node (TraceEvUnRegistered (fromJust currentVal) label)
810 liftIO $ sendMessage node
811 (NodeIdentifier (localNodeId node))
812 (ProcessIdentifier from)
813 WithImplicitReconnect
814 (RegisterReply label isOk)
815 else let operation =
816 case reregistration of
817 True -> flip decList
818 False -> flip incList
819 in case mPid of
820 Nothing -> return ()
821 Just pid -> modify' $ registeredOnNodesFor pid ^: (maybeify $ operation atnode)
822 where updateRemote node (Just oldval) (Just newval) | processNodeId oldval /= processNodeId newval =
823 do forward node (processNodeId oldval) (Register label atnode (Just oldval) True)
824 forward node (processNodeId newval) (Register label atnode (Just newval) False)
825 updateRemote node Nothing (Just newval) =
826 forward node (processNodeId newval) (Register label atnode (Just newval) False)
827 updateRemote node (Just oldval) Nothing =
828 forward node (processNodeId oldval) (Register label atnode (Just oldval) True)
829 updateRemote _ _ _ = return ()
830 maybeify f Nothing = unmaybeify $ f []
831 maybeify f (Just x) = unmaybeify $ f x
833 unmaybeify [] = Nothing
834 unmaybeify x = Just x
835 incList [] tag = [(tag,1)]
836 incList ((atag,acount):xs) tag | tag==atag = (atag,acount+1) : xs
837 incList (x:xs) tag = x : incList xs tag
838 decList [] _ = []
839 decList ((atag,1):xs) tag | atag == tag = xs
840 decList ((atag,n):xs) tag | atag == tag = (atag,n-1):xs
841 decList (x:xs) tag = x:decList xs tag
842 forward node to reg =
843 when (not $ isLocal node (NodeIdentifier to)) $
844 liftIO $ sendBinary node
845 (ProcessIdentifier from)
846 (NodeIdentifier to)
847 WithImplicitReconnect
848 NCMsg
849 { ctrlMsgSender = ProcessIdentifier from
850 , ctrlMsgSignal = reg
851 }
854-- Unified semantics does not explicitly describe 'whereis'
855ncEffectWhereIs :: ProcessId -> String -> NC ()
856ncEffectWhereIs from label = do
857 node <- ask
858 mPid <- gets (^. registeredHereFor label)
859 liftIO $ sendMessage node
860 (NodeIdentifier (localNodeId node))
861 (ProcessIdentifier from)
862 WithImplicitReconnect
863 (WhereIsReply label mPid)
865-- [Unified: Table 14]
866ncEffectNamedSend :: String -> Message -> NC ()
867ncEffectNamedSend label msg = do
868 mPid <- gets (^. registeredHereFor label)
869 -- If mPid is Nothing, we just ignore the named send (as per Table 14)
870 forM_ mPid $ \pid -> postMessage pid msg
872-- [Issue #DP-20]
873ncEffectLocalSend :: LocalNode -> ProcessId -> Message -> NC ()
874ncEffectLocalSend node to msg =
875 liftIO $ withLocalProc node to $ \p -> do
876 enqueue (processQueue p) msg
877 trace node (TraceEvReceived to msg)
879-- [Issue #DP-20]
880ncEffectLocalPortSend :: SendPortId -> Message -> NC ()
881ncEffectLocalPortSend from msg = do
882 node <- ask
883 let pid = sendPortProcessId from
884 cid = sendPortLocalId from
885 liftIO $ withLocalProc node pid $ \proc -> do
886 mChan <- withMVar (processState proc) $ return . (^. typedChannelWithId cid)
887 case mChan of
888 -- in the unlikely event we know nothing about this channel id,
889 -- there's little to be done - perhaps some logging/tracing though...
890 Nothing -> return ()
891 Just (TypedChannel chan') -> do
892 -- If ch is Nothing, the process has given up the read end of
893 -- the channel and we simply ignore the incoming message - this
894 ch <- deRefWeak chan'
895 forM_ ch $ \chan -> deliverChan msg chan
896 where deliverChan :: forall a . Message -> TQueue a -> IO ()
897 deliverChan (UnencodedMessage _ raw) chan' =
898 atomically $ writeTQueue chan' ((unsafeCoerce raw) :: a)
899 deliverChan (EncodedMessage _ _) _ =
900 -- this will not happen unless someone screws with Primitives.hs
901 error "invalid local channel delivery"
903-- [Issue #69]
904ncEffectKill :: ProcessId -> ProcessId -> String -> NC ()
905ncEffectKill from to reason = do
906 node <- ask
907 when (isLocal node (ProcessIdentifier to)) $
908 throwException to $ ProcessKillException from reason
910-- [Issue #69]
911ncEffectExit :: ProcessId -> ProcessId -> Message -> NC ()
912ncEffectExit from to reason = do
913 node <- ask
914 when (isLocal node (ProcessIdentifier to)) $
915 throwException to $ ProcessExitException from reason
917-- [Issue #89]
918ncEffectGetInfo :: ProcessId -> ProcessId -> NC ()
919ncEffectGetInfo from pid =
920 let lpid = processLocalId pid
921 them = (ProcessIdentifier pid)
922 in do
923 node <- ask
924 mProc <- liftIO $
925 withMVar (localState node) $ return . (^. localProcessWithId lpid)
926 case mProc of
927 Nothing -> dispatch (isLocal node (ProcessIdentifier from))
928 from node (ProcessInfoNone DiedUnknownId)
929 Just _ -> do
930 itsLinks <- gets (^. linksFor them)
931 itsMons <- gets (^. monitorsFor them)
932 registered <- gets (^. registeredHere)
934 let reg = registeredNames registered
935 dispatch (isLocal node (ProcessIdentifier from))
936 from
937 node
938 ProcessInfo {
939 infoNode = (processNodeId pid)
940 , infoRegisteredNames = reg
941 -- we cannot populate this field
942 , infoMessageQueueLength = Nothing
943 , infoMonitors = Set.toList itsMons
944 , infoLinks = Set.toList itsLinks
945 }
946 where dispatch :: (Serializable a, Show a)
947 => Bool
948 -> ProcessId
949 -> LocalNode
950 -> a
951 -> NC ()
952 dispatch True dest _ pInfo = postAsMessage dest $ pInfo
953 dispatch False dest node pInfo = do
954 liftIO $ sendMessage node
955 (NodeIdentifier (localNodeId node))
956 (ProcessIdentifier dest)
957 WithImplicitReconnect
958 pInfo
960 registeredNames = Map.foldlWithKey (\ks k v -> if v == pid
961 then (k:ks)
962 else ks) []
964--------------------------------------------------------------------------------
965-- Auxiliary --
966--------------------------------------------------------------------------------
968notifyDied :: ProcessId -- ^ Who to notify?
969 -> Identifier -- ^ Who died?
970 -> DiedReason -- ^ How did they die?
971 -> Maybe MonitorRef -- ^ 'Nothing' for linking
972 -> NC ()
973notifyDied dest src reason mRef = do
974 node <- ask
975 case (isLocal node (ProcessIdentifier dest), mRef, src) of
976 (True, Just ref, ProcessIdentifier pid) ->
977 postAsMessage dest $ ProcessMonitorNotification ref pid reason
978 (True, Just ref, NodeIdentifier nid) ->
979 postAsMessage dest $ NodeMonitorNotification ref nid reason
980 (True, Just ref, SendPortIdentifier cid) ->
981 postAsMessage dest $ PortMonitorNotification ref cid reason
982 (True, Nothing, ProcessIdentifier pid) ->
983 throwException dest $ ProcessLinkException pid reason
984 (True, Nothing, NodeIdentifier pid) ->
985 throwException dest $ NodeLinkException pid reason
986 (True, Nothing, SendPortIdentifier pid) ->
987 throwException dest $ PortLinkException pid reason
988 (False, _, _) ->
989 -- The change in sender comes from [Unified: Table 10]
990 liftIO $ sendBinary node
991 (NodeIdentifier $ localNodeId node)
992 (NodeIdentifier $ processNodeId dest)
993 WithImplicitReconnect
994 NCMsg
995 { ctrlMsgSender = NodeIdentifier (localNodeId node)
996 , ctrlMsgSignal = Died src reason
997 }
999-- | [Unified: Table 8]
1000destNid :: ProcessSignal -> Maybe NodeId
1001destNid (Link ident) = Just $ nodeOf ident
1002destNid (Unlink ident) = Just $ nodeOf ident
1003destNid (Monitor ref) = Just $ nodeOf (monitorRefIdent ref)
1004destNid (Unmonitor ref) = Just $ nodeOf (monitorRefIdent ref)
1005destNid (Spawn _ _) = Nothing
1006destNid (Register _ _ _ _) = Nothing
1007destNid (WhereIs _) = Nothing
1008destNid (NamedSend _ _) = Nothing
1009-- We don't need to forward 'Died' signals; if monitoring/linking is setup,
1010-- then when a local process dies the monitoring/linking machinery will take
1011-- care of notifying remote nodes
1012destNid (Died _ _) = Nothing
1013destNid (Kill pid _) = Just $ processNodeId pid
1014destNid (Exit pid _) = Just $ processNodeId pid
1015destNid (GetInfo pid) = Just $ processNodeId pid
1016destNid (LocalSend pid _) = Just $ processNodeId pid
1017destNid (LocalPortSend cid _) = Just $ processNodeId (sendPortProcessId cid)
1018destNid (SigShutdown) = Nothing
1020-- | Check if a process is local to our own node
1021isLocal :: LocalNode -> Identifier -> Bool
1022isLocal nid ident = nodeOf ident == localNodeId nid
1024-- | Lookup a local closure
1025unClosure :: Typeable a => Closure a -> NC (Either String a)
1026unClosure closure = do
1027 rtable <- remoteTable <$> ask
1028 return (Static.unclosure rtable closure)
1030-- | Check if an identifier refers to a valid local object
1031isValidLocalIdentifier :: Identifier -> NC Bool
1032isValidLocalIdentifier ident = do
1033 node <- ask
1034 liftIO . withMVar (localState node) $ \nSt ->
1035 case ident of
1036 NodeIdentifier nid ->
1037 return $ nid == localNodeId node
1038 ProcessIdentifier pid -> do
1039 let mProc = nSt ^. localProcessWithId (processLocalId pid)
1040 return $ isJust mProc
1041 SendPortIdentifier cid -> do
1042 let pid = sendPortProcessId cid
1043 mProc = nSt ^. localProcessWithId (processLocalId pid)
1044 case mProc of
1045 Nothing -> return False
1046 Just proc -> withMVar (processState proc) $ \pSt -> do
1047 let mCh = pSt ^. typedChannelWithId (sendPortLocalId cid)
1048 return $ isJust mCh
1050--------------------------------------------------------------------------------
1051-- Messages to local processes --
1052--------------------------------------------------------------------------------
1054postAsMessage :: Serializable a => ProcessId -> a -> NC ()
1055postAsMessage pid = postMessage pid . createUnencodedMessage
1057postMessage :: ProcessId -> Message -> NC ()
1058postMessage pid msg = do
1059 node <- ask
1060 liftIO $ withLocalProc node pid $ \p -> enqueue (processQueue p) msg
1062throwException :: Exception e => ProcessId -> e -> NC ()
1063throwException pid e = do
1064 node <- ask
1065 liftIO $ withLocalProc node pid $ \p -> throwTo (processThread p) e
1067withLocalProc :: LocalNode -> ProcessId -> (LocalProcess -> IO ()) -> IO ()
1068withLocalProc node pid p =
1069 -- By [Unified: table 6, rule missing_process] messages to dead processes
1070 -- can silently be dropped
1071 let lpid = processLocalId pid in do
1072 mProc <- withMVar (localState node) $ return . (^. localProcessWithId lpid)
1073 forM_ mProc p
1075--------------------------------------------------------------------------------
1076-- Accessors --
1077--------------------------------------------------------------------------------
1079links :: Accessor NCState (Map Identifier (Set ProcessId))
1080links = accessor _links (\ls st -> st { _links = ls })
1082monitors :: Accessor NCState (Map Identifier (Set (ProcessId, MonitorRef)))
1083monitors = accessor _monitors (\ms st -> st { _monitors = ms })
1085registeredHere :: Accessor NCState (Map String ProcessId)
1086registeredHere = accessor _registeredHere (\ry st -> st { _registeredHere = ry })
1088registeredOnNodes :: Accessor NCState (Map ProcessId [(NodeId, Int)])
1089registeredOnNodes = accessor _registeredOnNodes (\ry st -> st { _registeredOnNodes = ry })
1091linksFor :: Identifier -> Accessor NCState (Set ProcessId)
1092linksFor ident = links >>> DAC.mapDefault Set.empty ident
1094monitorsFor :: Identifier -> Accessor NCState (Set (ProcessId, MonitorRef))
1095monitorsFor ident = monitors >>> DAC.mapDefault Set.empty ident
1097registeredHereFor :: String -> Accessor NCState (Maybe ProcessId)
1098registeredHereFor ident = registeredHere >>> DAC.mapMaybe ident
1100registeredOnNodesFor :: ProcessId -> Accessor NCState (Maybe [(NodeId,Int)])
1101registeredOnNodesFor ident = registeredOnNodes >>> DAC.mapMaybe ident
1103-- | @splitNotif ident@ splits a notifications map into those
1104-- notifications that should trigger when 'ident' fails and those links that
1105-- should not.
1107-- There is a hierarchy between identifiers: failure of a node implies failure
1108-- of all processes on that node, and failure of a process implies failure of
1109-- all typed channels to that process. In other words, if 'ident' refers to a
1110-- node, then the /should trigger/ set will include
1112-- * the notifications for the node specifically
1113-- * the notifications for processes on that node, and
1114-- * the notifications for typed channels to processes on that node.
1116-- Similarly, if 'ident' refers to a process, the /should trigger/ set will
1117-- include
1119-- * the notifications for that process specifically and
1120-- * the notifications for typed channels to that process.
1122-- See https://github.com/haskell/containers/issues/14 for the bang on _v.
1123splitNotif :: Identifier
1124 -> Map Identifier a
1125 -> (Map Identifier a, Map Identifier a)
1126splitNotif ident = Map.partitionWithKey (\k !_v -> ident `impliesDeathOf` k)
1128--------------------------------------------------------------------------------
1129-- Auxiliary --
1130--------------------------------------------------------------------------------
1132-- | Modify and evaluate the state
1133modify' :: MonadState s m => (s -> s) -> m ()
1134modify' f = StateT.get >>= \s -> StateT.put $! f s