git.iko.soy/distributed-process
on master help sign in
10 files changed+476−223
since splitsnapshots
unifiedsplit
distributed-process-tests/distributed-process-tests.cabal+1−1 distributed-process-tests/src/Control/Distributed/Process/Tests/CH.hs+110−32 distributed-process-tests/src/Control/Distributed/Process/Tests/Internal/Utils.hs+5−2 distributed-process-tests/src/Control/Distributed/Process/Tests/Mx.hs+204−81 distributed-process.cabal+1−1 src/Control/Distributed/Process/Internal/Primitives.hs+54−39 src/Control/Distributed/Process/Management.hs+34−2 src/Control/Distributed/Process/Management/Internal/Types.hs+5 src/Control/Distributed/Process/Node.hs+15−10 src/Control/Distributed/Process/UnsafePrimitives.hs+47−55
distributed-process-tests/distributed-process-tests.cabal+1−1
@@ -6,7 +6,7 @@ description: Tests and test suite for Cloud Haskell libraries.
66license: BSD3
77license-file: LICENSE
88Author: Duncan Coutts, Nicolas Wu, Edsko de Vries
9−Maintainer: Facundo Domínguez <facundo.dominguez@tweag.io>
9+Maintainer: Tim Watson <watson.timothy@gmail.com>
1010copyright: Well-Typed LLP
1111category: Control, Cloud Haskell
1212build-type: Simple
distributed-process-tests/src/Control/Distributed/Process/Tests/CH.hs+110−32
@@ -44,13 +44,13 @@ import Control.Distributed.Process.Internal.Types
4444import Control.Distributed.Process.Node
4545import Control.Distributed.Process.Debug
4646import Control.Distributed.Process.Management.Internal.Types
47−import Control.Distributed.Process.Tests.Internal.Utils (shouldBe)
47+import Control.Distributed.Process.Tests.Internal.Utils (shouldBe, pause)
4848import Control.Distributed.Process.Serializable (Serializable)
49−
49+import Data.Maybe (isNothing, isJust)
5050import Test.HUnit (Assertion, assertBool, assertFailure)
5151import Test.Framework (Test, testGroup)
5252import Test.Framework.Providers.HUnit (testCase)
53−import Control.Rematch hiding (match)
53+import Control.Rematch hiding (match, isNothing, isJust)
5454import Control.Rematch.Run (Match(..))
5555
5656newtype Ping = Ping ProcessId
@@ -87,10 +87,9 @@ verifyClient :: String -> MVar Bool -> IO ()
8787verifyClient s b = takeMVar b >>= assertBool s
8888
8989expectPing :: MVar Bool -> Process ()
90−expectPing mv = expect >>= liftIO . putMVar mv . checkPing
90+expectPing mv = expect >>= liftIO . putMVar mv . checkPing
9191 where
9292 checkPing (Ping _) = True
93− checkPing _ = False
9493
9594-- | Quick and dirty synchronous version of whereisRemoteAsync
9695whereisRemote :: NodeId -> String -> Process (Maybe ProcessId)
@@ -240,10 +239,10 @@ testPing TestTransport{..} = do
240239 pid <- getSelfPid
241240 replicateM_ numPings $ do
242241 send pingServer (Pong pid)
243− p <- expect
242+ p <- expectTimeout 3000000
244243 case p of
245− Ping _ -> return ()
246− _ -> die "Unexpected message"
244+ Just (Ping _) -> return ()
245+ Nothing -> die "Failed to receive Ping"
247246
248247 putMVar clientDone ()
249248
@@ -560,7 +559,7 @@ testMergeChannels TestTransport{..} = do
560559 ss <- liftIO $ mapM readMVar vs
561560 case ss of
562561 [sa, sb, sc] ->
563− mapM_ ((>> liftIO (threadDelay 10000)) . uncurry sendChan)
562+ mapM_ ((>> pause 10000) . uncurry sendChan)
564563 [ -- a, b, c
565564 (sa, 'a')
566565 , (sb, 'b')
@@ -736,7 +735,6 @@ testRemoteRegistry :: TestTransport -> Assertion
736735testRemoteRegistry TestTransport{..} = do
737736 node1 <- newLocalNode testTransport initRemoteTable
738737 node2 <- newLocalNode testTransport initRemoteTable
739− done <- newEmptyMVar
740738
741739 pingServer <- forkProcess node1 ping
742740 deadProcess <- forkProcess node1 (return ())
@@ -751,6 +749,7 @@ testRemoteRegistry TestTransport{..} = do
751749
752750 pid <- verifyWhereIsRemote nid1 "ping"
753751 liftIO $ assertBool "Expected pindServer to match pid" $ pingServer == pid
752+
754753 us <- getSelfPid
755754 nsendRemote nid1 "ping" (Pong us)
756755 receiveWait [
@@ -760,22 +759,22 @@ testRemoteRegistry TestTransport{..} = do
760759 -- test that if process was not registered Nothing is returned
761760 -- in owner field.
762761 registerRemoteAsync nid1 "dead" deadProcess
763− receiveWait [
764− matchIf (\(RegisterReply label' False Nothing) -> "dead" == label')
765− (\(RegisterReply _ _ _) -> return ()) ]
762+ receiveWait [ matchIf (\(RegisterReply label' f mPid) -> "dead" == label')
763+ (\(RegisterReply _ f mPid) -> return (not f && isNothing mPid))
764+ ] >>= liftIO . assertBool "Expected False Nothing in RegisterReply"
765+
766766 registerRemoteAsync nid1 "ping" deadProcess
767767 receiveWait [
768− matchIf (\(RegisterReply label' False (Just pid'')) ->
769− "ping" == label' && pid'' == pingServer)
770− (\(RegisterReply _ _ _) -> return ()) ]
768+ matchIf (\(RegisterReply label' False mPid) ->
769+ "ping" == label' && isJust mPid)
770+ (\(RegisterReply _ f (Just pid'')) -> return (not f && pid'' == pingServer))
771+ ] >>= liftIO . assertBool "Expected False and (Just alreadyRegisteredPid) in RegisterReply"
772+
771773 unregisterRemoteAsync nid1 "dead"
772774 receiveWait [
773− matchIf (\(RegisterReply label' False Nothing) ->
774− "dead" == label' && pid == pingServer)
775− (\(RegisterReply _ _ _) -> return ()) ]
776− liftIO $ putMVar done ()
777−
778− takeMVar done
775+ matchIf (\(RegisterReply label' _ _) -> "dead" == label')
776+ (\(RegisterReply _ f mPid) -> return (not f && isNothing mPid))
777+ ] >>= liftIO . assertBool "Expected False and Nothing in RegisterReply"
779778
780779testRemoteRegistryRemoteProcess :: TestTransport -> Assertion
781780testRemoteRegistryRemoteProcess TestTransport{..} = do
@@ -1273,6 +1272,76 @@ testReceiveChanFeatures TestTransport{..} = do
12731272
12741273 takeMVar done
12751274
1275+testChanLifecycle :: TestTransport -> Assertion
1276+testChanLifecycle TestTransport{..} = let delay = 3000000 in do
1277+ result <- newEmptyMVar
1278+ tchMV <- newEmptyMVar
1279+ localNode <- newLocalNode testTransport initRemoteTable
1280+ runProcess localNode $ do
1281+
1282+ pid <- spawnLocal $ do tCh <- newChan :: Process (SendPort (), ReceivePort ())
1283+ liftIO $ putMVar tchMV tCh
1284+ expect :: Process ()
1285+ let (sp, _) = tCh
1286+ sendChan sp ()
1287+ expect :: Process ()
1288+
1289+ mRefPid <- monitor pid
1290+
1291+ cPid <- spawnLocal $ do
1292+ (sp', rp) <- liftIO $ takeMVar tchMV
1293+ -- say "picked up our test channel"
1294+ send pid ()
1295+ -- say "told pid to continue"
1296+ res <- receiveChanTimeout delay rp
1297+ case res of
1298+ Nothing -> say "initial chan () missing!" >> (liftIO $ putMVar result False)
1299+ Just () -> do mr <- monitor pid
1300+ pause 10000
1301+ -- say "sending pid a second () will cause it to exit"
1302+ send pid ()
1303+
1304+ -- say "make sure we see a DOWN notification for pid having stopped"
1305+ receiveWait [ match (\(_ :: ProcessMonitorNotification) -> return ()) ]
1306+
1307+ -- now that pid has died, the send port should be useless...
1308+ liftIO $ putMVar tchMV (sp', rp)
1309+
1310+ -- let's verify that we do not see the message from our
1311+ -- parent process on the channel, once pid has died...
1312+ recv <- receiveChanTimeout delay rp
1313+ -- say $ "finished waiting for second (), writing result" ++ (show recv)
1314+ liftIO $ putMVar result $ isNothing recv
1315+
1316+ mRefCPid <- monitor cPid
1317+
1318+ receiveWait
1319+ [ matchIf (\(ProcessMonitorNotification r _ _) -> r == mRefPid)
1320+ (\_ -> return ())
1321+ ]
1322+
1323+ -- say "seen first pid die..."
1324+
1325+ (sendPort, _) <- liftIO $ takeMVar tchMV
1326+ sendChan sendPort ()
1327+ -- say "sent () after owning pid died"
1328+
1329+ -- let cPid know we've written to the channel...
1330+ send cPid ()
1331+
1332+ receiveWait
1333+ [ matchIf (\(ProcessMonitorNotification r _ _) -> r == mRefCPid)
1334+ (\_ -> return ())
1335+ ]
1336+
1337+ -- say "seen both pids die now..."
1338+
1339+ -- and wait on the result back in IO land...
1340+ testRes <- takeMVar result
1341+ -- runProcess localNode $ say "got result..."
1342+ assertBool "Expected sending on the channel to fail, but received data!" testRes
1343+
1344+
12761345testKillLocal :: TestTransport -> Assertion
12771346testKillLocal TestTransport{..} = do
12781347 localNode <- newLocalNode testTransport initRemoteTable
@@ -1487,20 +1556,24 @@ testRegistryMonitoring :: TestTransport -> Assertion
14871556testRegistryMonitoring TestTransport{..} = do
14881557 node1 <- newLocalNode testTransport initRemoteTable
14891558 node2 <- newLocalNode testTransport initRemoteTable
1490− waitH <- newEmptyMVar
14911559
14921560 let nid = localNodeId node2
14931561 pid <- forkProcess node1 $ do
1494− getSelfPid >>= runUntilRegistered nid
1495− liftIO $ takeMVar waitH
1562+ self <- getSelfPid
1563+ runUntilRegistered nid self
1564+ say $ (show self) ++ " registered as " ++ regName
1565+ expect :: Process ()
1566+ say $ (show self) ++ " exiting normally"
14961567
14971568 runProcess node2 $ do
14981569 register regName pid
1570+ say $ regName ++ " registered to " ++ show pid
14991571 res <- whereis regName
1572+ send pid ()
1573+ say $ " sent finish signal to " ++ show pid
15001574 us <- getSelfPid
1501− liftIO $ do
1502− putMVar waitH ()
1503− assertBool "expected (Just pid)" $ res == (Just pid)
1575+ liftIO $ assertBool "expected (Just pid)" $ res == (Just pid)
1576+
15041577
15051578 -- This delay isn't essential!
15061579 -- The test case passes perfectly fine without it (feel free to comment out
@@ -1525,7 +1598,9 @@ testRegistryMonitoring TestTransport{..} = do
15251598 regHere <- newEmptyMVar
15261599 runProcess node2 $ whereis regName >>= liftIO . putMVar regHere
15271600 res <- takeMVar regHere
1528− assertBool "expected Nothing, but process still registered" (res == Nothing)
1601+ case res of
1602+ Nothing -> return ()
1603+ Just pid -> assertBool ("expected Nothing, but got " ++ show pid) False
15291604
15301605 where
15311606 runUntilRegistered nid us = do
@@ -1537,11 +1612,13 @@ testRegistryMonitoring TestTransport{..} = do
15371612
15381613 delayUntilMaybeUnregistered nid p = do
15391614 whereisRemoteAsync nid regName
1540− receiveTimeout 20000000 {- 20 sec delay -} [
1541− matchIf (\(WhereIsReply n p) -> n == regName && p == Nothing)
1615+ res <- receiveTimeout 20000000 {- 20 sec delay -} [
1616+ matchIf (\(WhereIsReply n p) -> n == regName && isNothing p)
15421617 (const $ return ())
15431618 ]
1544− return ()
1619+ case res of
1620+ Just () -> return ()
1621+ Nothing -> delayUntilMaybeUnregistered nid p
15451622
15461623 regName = "testRegisterRemote"
15471624
@@ -1737,6 +1814,7 @@ tests testtrans = return [
17371814 , testCase "MatchMessageUnwrap" (testMatchMessageWithUnwrap testtrans)
17381815 , testCase "ReceiveChanTimeout" (testReceiveChanTimeout testtrans)
17391816 , testCase "ReceiveChanFeatures" (testReceiveChanFeatures testtrans)
1817+ , testCase "ChanLifecycle" (testChanLifecycle testtrans)
17401818 , testCase "KillLocal" (testKillLocal testtrans)
17411819 , testCase "KillRemote" (testKillRemote testtrans)
17421820 , testCase "Die" (testDie testtrans)
distributed-process-tests/src/Control/Distributed/Process/Tests/Internal/Utils.hs+5−2
@@ -74,12 +74,13 @@ import Control.Concurrent
7474import Control.Concurrent.MVar
7575 ( putMVar
7676 )
77−import Control.Distributed.Process
77+import Control.Distributed.Process hiding (finally, catch)
7878import Control.Distributed.Process.Node
7979import Control.Distributed.Process.Serializable()
8080
8181import Control.Exception (AsyncException(ThreadKilled), SomeException)
8282import Control.Monad (forever, void)
83+import Control.Monad.Catch (finally, catch)
8384import Control.Monad.STM (atomically)
8485import Control.Rematch hiding (match)
8586import Control.Rematch.Run
@@ -90,6 +91,7 @@ import Test.HUnit (Assertion, assertFailure)
9091import Test.HUnit.Base (assertBool)
9192
9293import GHC.Generics
94+import System.Timeout (timeout)
9395
9496-- | A mutable cell containing a test result.
9597type TestResult a = MVar a
@@ -232,7 +234,8 @@ testProcessReport pid = do
232234tryRunProcess :: LocalNode -> Process () -> IO ()
233235tryRunProcess node p = do
234236 tid <- liftIO myThreadId
235− runProcess node $ catch p (\e -> liftIO $ throwTo tid (e::SomeException))
237+ void $ timeout (1000000 * 60 * 5 :: Int) $
238+ runProcess node $ catch p (\e -> liftIO $ throwTo tid (e::SomeException))
236239
237240tryForkProcess :: LocalNode -> Process () -> IO ProcessId
238241tryForkProcess node p = do
distributed-process-tests/src/Control/Distributed/Process/Tests/Mx.hs+204−81
@@ -1,16 +1,22 @@
1−{-# LANGUAGE DeriveGeneric #-}
1+{-# LANGUAGE DeriveGeneric #-}
2+{-# LANGUAGE ParallelListComp #-}
3+
24module Control.Distributed.Process.Tests.Mx (tests) where
35
46import Control.Distributed.Process.Tests.Internal.Utils
57import Network.Transport.Test (TestTransport(..))
6−
7−import Control.Distributed.Process
8+import Control.Distributed.Process hiding (bracket, finally, try)
9+import Control.Distributed.Process.Internal.Types
10+ ( ProcessExitException(..)
11+ , unsafeCreateUnencodedMessage
12+ )
813import Control.Distributed.Process.Node
914import qualified Control.Distributed.Process.UnsafePrimitives as Unsafe
1015 ( send
1116 , nsend
1217 , nsendRemote
1318 , usend
19+ , sendChan
1420 )
1521import Control.Distributed.Process.Management
1622 ( MxEvent(..)
@@ -28,12 +34,13 @@ import Control.Distributed.Process.Management
2834 , mxBroadcast
2935 )
3036import Control.Monad (void, unless)
37+import Control.Monad.Catch(finally, bracket, try)
3138import Control.Rematch (equalTo)
3239import Data.Binary
33−import Data.List (find, sort)
34−import Data.Maybe (isJust, isNothing)
40+import Data.List (find, sort, intercalate)
41+import Data.Maybe (isJust, fromJust, isNothing, fromMaybe, catMaybes)
3542import Data.Typeable
36−import GHC.Generics
43+import GHC.Generics hiding (from)
3744#if ! MIN_VERSION_base(4,6,0)
3845import Prelude hiding (catch, log)
3946#endif
@@ -57,9 +64,9 @@ awaitExit pid =
5764 (\_ -> return ())
5865 ]
5966 where
60− withMonitorRef pid = bracket (monitor pid) unmonitor
67+ withMonitorRef p = bracket (monitor p) unmonitor
6168
62−testAgentBroadcast :: TestResult () -> Process ()
69+testAgentBroadcast :: TestResult (Maybe ()) -> Process ()
6370testAgentBroadcast result = do
6471 (resultSP, resultRP) <- newChan :: Process (SendPort (), ReceivePort ())
6572
@@ -74,10 +81,10 @@ testAgentBroadcast result = do
7481 mxNotify ()
7582 -- Once the publisher has seen our message, it will broadcast the Publish
7683 -- and the consumer will see that and send the result to our typed channel.
77− stash result =<< receiveChan resultRP
84+ stash result =<< receiveChanTimeout 10000000 resultRP
7885
79− kill publisher "finished" >> awaitExit publisher
80− kill consumer "finished" >> awaitExit consumer
86+ kill publisher "finished"
87+ kill consumer "finished"
8188
8289testAgentDualInput :: TestResult (Maybe Int) -> Process ()
8390testAgentDualInput result = do
@@ -92,6 +99,8 @@ testAgentDualInput result = do
9299 else mxReady)
93100 ]
94101
102+ mRef <- monitor s
103+
95104 mxNotify (1 :: Int)
96105 nsend "sum-agent" (3 :: Int)
97106 mxNotify (2 :: Int)
@@ -99,7 +108,10 @@ testAgentDualInput result = do
99108 mxNotify (5 :: Int)
100109
101110 stash result =<< receiveChanTimeout 10000000 rp
102− awaitExit s
111+ died <- receiveTimeout 10000000 [
112+ matchIf (\(ProcessMonitorNotification r _ _) -> r == mRef) (const $ return True)
113+ ]
114+ died `shouldBe` equalTo (Just True)
103115
104116testAgentPrioritisation :: TestResult [String] -> Process ()
105117testAgentPrioritisation result = do
@@ -113,10 +125,10 @@ testAgentPrioritisation result = do
113125 (sp, rp) <- newChan
114126 s <- mxAgent (MxAgentId name) ["first"] [
115127 mxSink (\(s :: String) -> do
116− mxUpdateLocal ((s:))
128+ mxUpdateLocal (s:)
117129 st <- mxGetLocal
118130 case length st of
119− n | n == 5 -> do liftMX $ sendChan sp st
131+ n | n == 5 -> do liftMX $ sendChan sp (sort st)
120132 mxDeactivate "finished"
121133 _ -> mxReceive -- go to the mailbox
122134 )
@@ -127,7 +139,7 @@ testAgentPrioritisation result = do
127139 mxNotify "fourth"
128140 nsend name "fifth"
129141
130− stash result . sort =<< receiveChan rp
142+ stash result . fromMaybe [] =<< receiveChanTimeout 10000000 rp
131143 awaitExit s
132144
133145testAgentMailboxHandling :: TestResult (Maybe ()) -> Process ()
@@ -140,12 +152,19 @@ testAgentMailboxHandling result = do
140152 nsend "mailbox-agent" ()
141153
142154 stash result =<< receiveChanTimeout 1000000 rp
143− kill agent "finished" >> awaitExit agent
155+ kill agent "finished"
144156
145157testAgentEventHandling :: TestResult Bool -> Process ()
146158testAgentEventHandling result = do
147− let initState = [] :: [MxEvent]
159+ us <- getSelfPid
160+ -- because this test is a bit racy, let's ensure it can't run indefinitely
161+ timer <- spawnLocal $ do
162+ pause (1000000 * 60 * 5)
163+ -- okay we've waited 5 mins, let's kill the test off if it's stuck...
164+ stash result False
165+ kill us "Test Timed Out"
148166
167+ let initState = [] :: [MxEvent]
149168 (rc, rs) <- newChan
150169 agentPid <- mxAgent (MxAgentId "lifecycle-listener-agent") initState [
151170 (mxSink $ \ev -> do
@@ -178,8 +197,7 @@ testAgentEventHandling result = do
178197 -- TODO: yes, this is racy, but we're at the mercy of the scheduler here...
179198 faff 2000000 rs
180199
181− _ <- monitor agentPid
182− (sp, rp) <- newChan
200+ (sp, rp) <- newChan :: Process (SendPort (), ReceivePort ())
183201
184202 pid <- spawnLocal $ expect >>= sendChan sp
185203
@@ -188,9 +206,11 @@ testAgentEventHandling result = do
188206 monitor pid
189207
190208 send pid ()
191− () <- receiveChan rp
209+ rct <- receiveChanTimeout 10000000 rp
210+ unless (isJust rct) $ die "No response on channel"
192211
193− receiveWait [ match (\(ProcessMonitorNotification _ _ _) -> return ()) ]
212+ pmn <- receiveTimeout 2000000 [ match (\ProcessMonitorNotification{} -> return ()) ]
213+ unless (isJust pmn) $ die "No monitor notification arrived"
194214
195215 -- TODO: yes, this is racy, but we're at the mercy of the scheduler here...
196216 faff 2000000 rs
@@ -199,17 +219,16 @@ testAgentEventHandling result = do
199219 mxNotify (MxSpawned pid, replyTo)
200220 mxNotify (MxProcessDied pid DiedNormal, replyTo)
201221
202− seenAlive <- receiveChan reply
203− seenDead <- receiveChan reply
222+ seenAlive <- receiveChanTimeout 10000000 reply
223+ seenDead <- receiveChanTimeout 10000000 reply
204224
205− stash result $ seenAlive && seenDead
206− kill agentPid "test-complete" >> awaitExit agentPid
225+ stash result $ foldr (&&) True $ catMaybes [seenAlive, seenDead]
226+ kill timer "test-complete"
227+ kill agentPid "test-complete"
207228 where
208229 faff delay port = do
209230 res <- receiveChanTimeout delay port
210− unless (isNothing res) $ do
211− pause delay
212− faff delay port
231+ unless (isNothing res) $ pause delay
213232
214233testMxRegEvents :: TestResult () -> Process ()
215234testMxRegEvents result = do
@@ -256,7 +275,6 @@ testMxRegEvents result = do
256275 reg3 `shouldBe` equalTo (Just (label, p1))
257276
258277 mapM_ (flip kill $ "test-complete") [agent, p1, p2]
259− awaitExit agent
260278
261279testMxRegMon :: LocalNode -> TestResult () -> Process ()
262280testMxRegMon remoteNode result = do
@@ -314,106 +332,211 @@ testMxRegMon remoteNode result = do
314332 evts `shouldContain` (Just (label1, p1))
315333 evts `shouldContain` (Just (label2, p1))
316334
317− kill agent "test-complete" >> awaitExit agent
335+ kill agent "test-complete"
318336
319337ensure :: Process () -> Process () -> Process ()
320338ensure = flip finally
321339
340+testNSend :: (String -> () -> Process ())
341+ -> Maybe LocalNode
342+ -> Process ()
343+testNSend op n = do
344+ us <- getSelfPid
345+ let delay = 5000000
346+ let label = "testMxSend" ++ (show us)
347+ let isValid = isValidLabel n label
348+
349+ testMxSend n label $ \p1 sink -> do
350+ register label p1
351+ reg1 <- receiveChanTimeout delay sink
352+ case reg1 of
353+ Just (MxRegistered pd lb)
354+ | pd == p1 && isValid lb -> return ()
355+ _ -> die $ "Reg-Failed: " ++ show reg1
322356
323−testMxUnsafeSendEvents :: LocalNode -> Process ()
324−testMxUnsafeSendEvents remoteNode = do
357+ op label ()
325358
326− -- ensure that when a registered process dies, we get a notification that
327− -- it has been unregistered as well as seeing the name get removed
359+ sent <- receiveChanTimeout delay sink
360+ case sent of
361+ Just (MxSentToName lb by _)
362+ | by == us && isValid lb -> return True
363+ _ -> die $ "Send-Failed: " ++ show sent
328364
329− let label1 = "aaaaa"
330− let label2 = "bbbbb"
331− let isValid l = l ==label1 || l == label2
332− let agentLabel = "listener-agent"
333− let delay = 1000000
334− (regChan, regSink) <- newChan
335− (unRegChan, unRegSink) <- newChan
336− agent <- mxAgent (MxAgentId agentLabel) () [
365+ where
366+ isValidLabel nd l1 l2
367+ | l2 == l2 = True
368+ | isJust nd = l2 == l1 ++ "@" ++ show (localNodeId $ fromJust nd)
369+ | otherwise = False
370+
371+testSend :: (ProcessId -> () -> Process ())
372+ -> Maybe LocalNode
373+ -> Process ()
374+testSend op n = do
375+ us <- getSelfPid
376+ let label = "testMxSend" ++ (show us)
377+ testMxSend n label $ \p1 sink -> do
378+ -- initiate a send
379+ op p1 ()
380+
381+ -- verify the management event
382+ sent <- receiveChanTimeout 5000000 sink
383+ case sent of
384+ Just (MxSent pidTo pidFrom _)
385+ | pidTo == p1 && pidFrom == us -> return True
386+ _ -> return False
387+
388+testChan :: (SendPort () -> () -> Process ())
389+ -> Maybe LocalNode
390+ -> Process ()
391+testChan op n = testMxSend n "" $ \p1 sink -> do
392+
393+ us <- getSelfPid
394+ send p1 us
395+
396+ cleared <- receiveChanTimeout 2000000 sink
397+ case cleared of
398+ Just (MxSent pidTo pidFrom _)
399+ | pidTo == p1 && pidFrom == us -> return ()
400+ _ -> die "Received uncleared Mx Event"
401+
402+ chan <- receiveTimeout 5000000 [ match return ]
403+ let ch' = fromMaybe (error "No reply chan received") chan
404+
405+ -- initiate a send
406+ op ch' ()
407+
408+ -- verify the management event
409+ sent <- receiveChanTimeout 5000000 sink
410+ case sent of
411+ Just (MxSentToPort sId spId _)
412+ | sId == us && spId == sendPortId ch' -> return True
413+ _ -> return False
414+
415+type SendTest = ProcessId -> ReceivePort MxEvent -> Process Bool
416+
417+testMxSend :: Maybe LocalNode -> String -> SendTest -> Process ()
418+testMxSend mNode label test = do
419+ us <- getSelfPid
420+ (sp, rp) <- newChan
421+ (chan, sink) <- newChan
422+ agent <- mxAgent (MxAgentId $ agentLabel us) () [
337423 mxSink $ \ev -> do
338424 case ev of
339− MxRegistered pid label
340− | isValid label -> liftMX $ sendChan regChan (label, pid)
341− MxUnRegistered pid label
342− | isValid label -> liftMX $ sendChan unRegChan (label, pid)
425+ m@(MxSentToPort _ cid _)
426+ | cid /= sendPortId chan
427+ && cid /= sendPortId sp -> liftMX $ sendChan chan m
428+ m@(MxSent _ fromPid _)
429+ | fromPid == us -> liftMX $ sendChan chan m
430+ m@(MxSentToName _ fromPid _)
431+ | fromPid == us -> liftMX $ sendChan chan m
432+ m@(MxRegistered _ name)
433+ | name == label -> liftMX $ sendChan chan m
343434 _ -> return ()
344435 mxReady
345436 ]
346437
347− (sp, rp) <- newChan
348− liftIO $ forkProcess remoteNode $ do
349− getSelfPid >>= sendChan sp
350− expect :: Process ()
351−
352− p1 <- receiveChan rp
438+ case mNode of
439+ Nothing -> void $ spawnLocal (proc sp)
440+ Just remoteNode -> void $ liftIO $ forkProcess remoteNode $ proc sp
353441
354− register label1 p1
355− reg1 <- receiveChanTimeout delay regSink
356− reg1 `shouldBe` equalTo (Just (label1, p1))
442+ p1 <- receiveChanTimeout 2000000 rp
443+ unless (isJust p1) $ die "Timed out waiting for ProcessId"
444+ res <- try $ test (fromJust p1) sink
357445
358− register label2 p1
359− reg2 <- receiveChanTimeout delay regSink
360− reg2 `shouldBe` equalTo (Just (label2, p1))
446+ kill agent "bye"
447+ kill (fromJust p1) "bye"
361448
362− n1 <- whereis label1
363− n1 `shouldBe` equalTo (Just p1)
449+ case res of
450+ Left (ProcessExitException _ m) -> (liftIO $ putStrLn $ "SomeException-" ++ show m) >> die m
451+ Right tr -> tr `shouldBe` equalTo True
364452
365− n2 <- whereis label2
366− n2 `shouldBe` equalTo (Just p1)
367453
368− kill p1 "goodbye"
369−
370− unreg1 <- receiveChanTimeout delay unRegSink
371− unreg2 <- receiveChanTimeout delay unRegSink
454+ where
455+ agentLabel s = "mx-unsafe-check-agent-" ++ show s
456+ proc sp' = getSelfPid >>= sendChan sp' >> go Nothing
372457
373− sort [unreg1, unreg2]
374− `shouldBe` equalTo [Just (label1, p1), Just (label2, p1)]
458+ go :: Maybe (ReceivePort ()) -> Process ()
459+ go Nothing = receiveTimeout 5000000 [ match replyChannel ] >>= go
460+ go c@(Just c') = receiveWait [ matchChan c' return ] >> go c
375461
376− kill agent "test-complete"
462+ replyChannel p' = do
463+ (s, r) <- newChan
464+ send p' s
465+ return r
377466
378467tests :: TestTransport -> IO [Test]
379468tests TestTransport{..} = do
380469 node1 <- newLocalNode testTransport initRemoteTable
381470 node2 <- newLocalNode testTransport initRemoteTable
382471 return [
383− testGroup "Mx Agents" [
384− testCase "Event Handling"
472+ testGroup "MxAgents" [
473+ testCase "EventHandling"
385474 (delayedAssertion
386475 "expected True, but events where not as expected"
387476 node1 True testAgentEventHandling)
388− , testCase "Inter-Agent Broadcast"
477+ , testCase "InterAgentBroadcast"
389478 (delayedAssertion
390479 "expected (), but no broadcast was received"
391− node1 () testAgentBroadcast)
392− , testCase "Agent Mailbox Handling"
480+ node1 (Just ()) testAgentBroadcast)
481+ , testCase "AgentMailboxHandling"
393482 (delayedAssertion
394483 "expected (Just ()), but no regular (mailbox) input was handled"
395484 node1 (Just ()) testAgentMailboxHandling)
396− , testCase "Agent Dual Input Handling"
485+ , testCase "AgentDualInputHandling"
397486 (delayedAssertion
398487 "expected sum = 15, but the result was Nothing"
399488 node1 (Just 15 :: Maybe Int) testAgentDualInput)
400− , testCase "Agent Input Prioritisation"
489+ , testCase "AgentInputPrioritisation"
401490 (delayedAssertion
402491 "expected [first, second, third, fourth, fifth], but result diverged"
403492 node1 (sort ["first", "second",
404493 "third", "fourth",
405494 "fifth"]) testAgentPrioritisation)
406495 ]
407− , testGroup "Mx Events" [
408− testCase "Name Registration Events"
496+ , testGroup "MxEvents" [
497+ testCase "NameRegistrationEvents"
409498 (delayedAssertion
410499 "expected registration events to map to the correct ProcessId"
411500 node1 () testMxRegEvents)
412− , testCase "Post Death Name UnRegistration Events"
501+ , testCase "PostDeathNameUnRegistrationEvents"
413502 (delayedAssertion
414503 "expected process deaths to result in unregistration events"
415504 node1 () (testMxRegMon node2))
416− , testCase "Monitor Events"
417− (runProcess node1 (testMxUnsafeSendEvents node2))
505+ , testGroup "SendEvents" $ buildTestCases node1 node2
418506 ]
419507 ]
508+ where
509+ buildTestCases n1 n2 = let nid = localNodeId n2 in build n1 n2 [
510+ ("NSend", [
511+ ("nsend", testNSend nsend)
512+ , ("Unsafe.nsend", testNSend Unsafe.nsend)
513+ , ("nsendRemote", testNSend (nsendRemote nid))
514+ , ("Unsafe.nsendRemote", testNSend (Unsafe.nsendRemote nid))
515+ ])
516+ , ("Send", [
517+ ("send", testSend send)
518+ , ("Unsafe.send", testSend Unsafe.send)
519+ , ("usend", testSend usend)
520+ , ("Unsafe.usend", testSend Unsafe.usend)
521+ , ("sendChan", testChan sendChan)
522+ , ("Unsafe.sendChan", testChan Unsafe.sendChan)
523+ ])
524+ , ("Forward", [
525+ ("forward", testSend (\p m -> forward (unsafeCreateUnencodedMessage m) p))
526+ , ("uforward", testSend (\p m -> uforward (unsafeCreateUnencodedMessage m) p))
527+ ])
528+ ]
529+
530+ build :: LocalNode
531+ -> LocalNode
532+ -> [(String, [(String, (Maybe LocalNode -> Process ()))])]
533+ -> [Test]
534+ build n ln specs =
535+ [ testGroup (intercalate "-" [groupName, caseSuffix]) [
536+ testCase (intercalate "-" [caseName, caseSuffix])
537+ (runProcess n (caseImpl caseNode))
538+ | (caseName, caseImpl) <- groupCases
539+ ]
540+ | (groupName, groupCases) <- specs
541+ , (caseSuffix, caseNode) <- [("RemotePid", Just ln), ("LocalPid", Nothing)]
542+ ]
distributed-process.cabal+1−1
@@ -6,7 +6,7 @@ License: BSD3
66License-File: LICENSE
77Copyright: Well-Typed LLP, Tweag I/O Limited
88Author: Duncan Coutts, Nicolas Wu, Edsko de Vries
9−Maintainer: Facundo Domínguez <facundo.dominguez@tweag.io>
9+Maintainer: Tim Watson <watson.timothy@gmail.com>
1010Stability: experimental
1111Homepage: http://haskell-distributed.github.com/
1212Bug-Reports: https://github.com/haskell-distributed/distributed-process/issues
src/Control/Distributed/Process/Internal/Primitives.hs+54−39
@@ -254,17 +254,15 @@ send them msg = do
254254 node = processNode proc
255255 nodeId = localNodeId node
256256 destNode = (processNodeId them) in do
257− case destNode == nodeId of
258− True -> sendLocal them msg
259− False -> liftIO $ sendMessage (processNode proc)
260− (ProcessIdentifier (processId proc))
261− (ProcessIdentifier them)
262− NoImplicitReconnect
263− msg
264− -- We do not fire the trace event until after the sending is complete;
265− -- In the remote case, 'sendMessage' can block in the networking stack.
266257 liftIO $ traceEvent (localEventBus node)
267258 (MxSent them us (createUnencodedMessage msg))
259+ if destNode == nodeId
260+ then sendLocal them msg
261+ else liftIO $ sendMessage (processNode proc)
262+ (ProcessIdentifier (processId proc))
263+ (ProcessIdentifier them)
264+ NoImplicitReconnect
265+ msg
268266
269267-- | /Unsafe/ variant of 'send'. This function makes /no/ attempt to serialize
270268-- and (in the case when the destination process resides on the same local
@@ -284,9 +282,12 @@ unsafeSend = Unsafe.send
284282--
285283usend :: Serializable a => ProcessId -> a -> Process ()
286284usend them msg = do
287− here <- getSelfNode
285+ proc <- ask
288286 let there = processNodeId them
289− if here == there
287+ let (us, node) = (processId proc, processNode proc)
288+ let msg' = wrapMessage msg
289+ liftIO $ traceEvent (localEventBus node) (MxSent them us msg')
290+ if localNodeId (processNode proc) == there
290291 then sendLocal them msg
291292 else sendCtrlMsg (Just there) $ UnreliableSend (processLocalId them)
292293 (createMessage msg)
@@ -306,7 +307,13 @@ expect = receiveWait [match return]
306307-- Channels --
307308--------------------------------------------------------------------------------
308309
309−-- | Create a new typed channel
310+-- | Create a new typed channel, bound to the calling Process
311+--
312+-- Note that the channel is bound to the lifecycle of the process that evaluates
313+-- this function, such that when it dies/exits, the channel will no longer
314+-- function, but will remain accessible. Thus reading from the ReceivePort will
315+-- fail silently thereafter, blocking indefinitely (unless a timeout is used).
316+--
310317newChan :: Serializable a => Process (SendPort a, ReceivePort a)
311318newChan = do
312319 proc <- ask
@@ -334,13 +341,16 @@ newChan = do
334341sendChan :: Serializable a => SendPort a -> a -> Process ()
335342sendChan (SendPort cid) msg = do
336343 proc <- ask
337− let node = localNodeId (processNode proc)
338− destNode = processNodeId (sendPortProcessId cid) in do
339− case destNode == node of
344+ let node = processNode proc
345+ pid = processId proc
346+ us = localNodeId node
347+ them = processNodeId (sendPortProcessId cid) in do
348+ liftIO $ traceEvent (localEventBus node) (MxSentToPort pid cid $ wrapMessage msg)
349+ case them == us of
340350 True -> sendChanLocal cid msg
341351 False -> do
342− liftIO $ sendBinary (processNode proc)
343− (ProcessIdentifier (processId proc))
352+ liftIO $ sendBinary node
353+ (ProcessIdentifier pid)
344354 (SendPortIdentifier cid)
345355 NoImplicitReconnect
346356 msg
@@ -484,15 +494,14 @@ forward msg them = do
484494 us = processId proc
485495 nid = localNodeId node
486496 destNode = (processNodeId them) in do
487− case destNode == nid of
488− True -> sendCtrlMsg Nothing (LocalSend them msg)
489− False ->
490− liftIO $ do
491− sendPayload (processNode proc) (ProcessIdentifier (processId proc)) (ProcessIdentifier them) NoImplicitReconnect (messageToPayload msg)
492− -- We do not fire the trace event until after the sending is complete;
493− -- In this remote case, 'sendMessage' can block in the networking stack.
494− liftIO $ traceEvent (localEventBus node)
495− (MxSent them us msg)
497+ liftIO $ traceEvent (localEventBus node) (MxSent them us msg)
498+ if destNode == nid
499+ then sendCtrlMsg Nothing (LocalSend them msg)
500+ else liftIO $ sendPayload (processNode proc)
501+ (ProcessIdentifier (processId proc))
502+ (ProcessIdentifier them)
503+ NoImplicitReconnect
504+ (messageToPayload msg)
496505
497506-- | Forward a raw 'Message' to the given 'ProcessId'.
498507--
@@ -506,14 +515,10 @@ uforward msg them = do
506515 us = processId proc
507516 nid = localNodeId node
508517 destNode = (processNodeId them) in do
509− case destNode == nid of
510− True -> sendCtrlMsg Nothing (LocalSend them msg)
511− False -> do
512− sendCtrlMsg (Just destNode) $ UnreliableSend (processLocalId them) msg
513− -- We do not fire the trace event until after the sending is complete;
514− -- In the remote case, 'sendCtrlMsg' can block in the networking stack.
515− liftIO $ traceEvent (localEventBus node)
516− (MxSent them us msg)
518+ liftIO $ traceEvent (localEventBus node) (MxSent them us msg)
519+ if destNode == nid
520+ then sendCtrlMsg Nothing (LocalSend them msg)
521+ else sendCtrlMsg (Just destNode) $ UnreliableSend (processLocalId them) msg
517522
518523-- | Wrap a 'Serializable' value in a 'Message'. Note that 'Message's are
519524-- 'Serializable' - like the datum they contain - but also note, deserialising
@@ -1222,8 +1227,12 @@ whereisRemoteAsync nid label = do
12221227
12231228-- | Named send to a process in the local registry (asynchronous)
12241229nsend :: Serializable a => String -> a -> Process ()
1225−nsend label msg =
1226− sendCtrlMsg Nothing (NamedSend label (createUnencodedMessage msg))
1230+nsend label msg = do
1231+ proc <- ask
1232+ let msg' = createUnencodedMessage msg
1233+ liftIO $ traceEvent (localEventBus (processNode proc))
1234+ (MxSentToName label (processId proc) msg')
1235+ sendCtrlMsg Nothing (NamedSend label msg')
12271236
12281237-- | Named send to a process in the local registry (asynchronous).
12291238-- This function makes /no/ attempt to serialize and (in the case when the
@@ -1235,9 +1244,15 @@ unsafeNSend = Unsafe.nsend
12351244-- | Named send to a process in a remote registry (asynchronous)
12361245nsendRemote :: Serializable a => NodeId -> String -> a -> Process ()
12371246nsendRemote nid label msg = do
1238− here <- getSelfNode
1239− if here == nid then nsend label msg
1240− else sendCtrlMsg (Just nid) (NamedSend label (createMessage msg))
1247+ proc <- ask
1248+ let us = processId proc
1249+ let node = processNode proc
1250+ if localNodeId node == nid
1251+ then nsend label msg
1252+ else let lbl = label ++ "@" ++ show nid in do
1253+ liftIO $ traceEvent (localEventBus node)
1254+ (MxSentToName lbl us (wrapMessage msg))
1255+ sendCtrlMsg (Just nid) (NamedSend label (createMessage msg))
12411256
12421257-- | Named send to a process in a remote registry (asynchronous)
12431258-- This function makes /no/ attempt to serialize and (in the case when the
src/Control/Distributed/Process/Management.hs+34−2
@@ -76,6 +76,40 @@
7676--
7777-- * Whether messages will be taken from the mailbox first, or the event bus.
7878--
79+-- Since the event bus uses STM broadcast channels to communicate with agents,
80+-- no message written to the bus successfully can be lost.
81+--
82+-- Agents can also receive messages via their mailboxes - these are subject to
83+-- the same guarantees as all inter-process message sending.
84+--
85+-- Messages dispatched on an STM broadcast channel (i.e., management event bus)
86+-- are guaranteed to be delivered with the same FIFO ordering guarantees that
87+-- exist between two communicating processes, such that communication from the
88+-- node controller's threads (i.e., MxEvent's) will never be re-ordered, but
89+-- messages dispatched to the event bus by other processes (including, but not
90+-- limited to agents) are only guaranteed to be ordered between one sender and
91+-- one receiver.
92+--
93+-- No guarantee exists for the ordering in which messages sent to an agent's
94+-- mailbox will be delivered, vs messages dispatched via the event bus.
95+--
96+-- Because of the above, there are no ordering guarantees for messages sent
97+-- between agents, or for processes to agents, except for those that apply to
98+-- messages sent between regular processes, since agents are
99+-- implemented as such.
100+--
101+-- The event bus is serial and single threaded. Anything that is published by
102+-- the node controller will be seen in FIFO order. There are no ordering
103+-- guarantees pertaining to entries published to the event bus by other
104+-- processes or agents.
105+--
106+-- It should not be possible to see, for example, an @MxReceived@ before the
107+-- corresponding @MxSent@ event, since the places where we issue the @MxSent@
108+-- write directly to the event bus (using STM) in the calling (green) thread,
109+-- before dispatching instructions to the node controller to perform the
110+-- necessary routing to deliver the message to a process (or registered name,
111+-- or typed channel) locally or remotely.
112+--
79113-- [Management Data API]
80114--
81115-- Both management agents and clients of the API have access to a variety of
@@ -144,8 +178,6 @@
144178-- > monitorNames = getSelfPid >>= nsend "name-monitor"
145179-- > monitorNames2 = getSelfPid >>= mxNotify
146180--
147−-- For some real-world examples, see the distributed-process-platform package.
148−--
149181-- [Performance, Stablity and Scalability]
150182--
151183-- /Management Agents/ offer numerous advantages over regular processes:
src/Control/Distributed/Process/Management/Internal/Types.hs+5
@@ -21,6 +21,7 @@ import Control.Concurrent.STM
2121import Control.Distributed.Process.Internal.Types
2222 ( Process
2323 , ProcessId
24+ , SendPortId
2425 , Message
2526 , DiedReason
2627 , NodeId
@@ -59,8 +60,12 @@ data MxEvent =
5960 -- ^ fired whenever a message is sent from a local process
6061 | MxSentToName String ProcessId Message
6162 -- ^ fired whenever a named send occurs
63+ | MxSentToPort ProcessId SendPortId Message
64+ -- ^ fired whenever a sendChan occurs
6265 | MxReceived ProcessId Message
6366 -- ^ fired whenever a message is received by a local process
67+ | MxReceivedPort SendPortId Message
68+ -- ^ fired whenever a message is received via a typed channel
6469 | MxConnected ConnectionId EndPointAddress
6570 -- ^ fired when a network-transport connection is first established
6671 | MxDisconnected ConnectionId EndPointAddress
src/Control/Distributed/Process/Node.hs+15−10
@@ -492,7 +492,7 @@ type IncomingConnection = (NT.EndPointAddress, IncomingTarget)
492492data IncomingTarget =
493493 Uninit
494494 | ToProc ProcessId (Weak (CQueue Message))
495− | ToChan TypedChannel
495+ | ToChan SendPortId TypedChannel
496496 | ToNode
497497
498498data ConnectionState = ConnectionState {
@@ -547,13 +547,17 @@ handleIncomingMessages node = go initConnectionState
547547 enqueue queue msg -- 'enqueue' is strict
548548 trace node (MxReceived pid msg)
549549 go st
550− Just (_, ToChan (TypedChannel chan')) -> do
550+ Just (_, ToChan chId (TypedChannel chan')) -> do
551551 mChan <- deRefWeak chan'
552552 -- If mChan is Nothing, the process has given up the read end of
553553 -- the channel and we simply ignore the incoming message
554− forM_ mChan $ \chan -> atomically $
555− -- We make sure the message is fully decoded when it is enqueued
556− writeTQueue chan $! decode (BSL.fromChunks payload)
554+ forM_ mChan $ \chan -> do
555+ msg' <- atomically $ do
556+ msg <- return $! decode (BSL.fromChunks payload)
557+ -- We make sure the message is fully decoded when it is enqueued
558+ writeTQueue chan msg
559+ return msg
560+ trace node $ MxReceivedPort chId $ unsafeCreateUnencodedMessage msg'
557561 go st
558562 Just (_, ToNode) -> do
559563 let ctrlMsg = decode . BSL.fromChunks $ payload
@@ -580,7 +584,7 @@ handleIncomingMessages node = go initConnectionState
580584 mChannel <- withMVar (processState proc) $ return . (^. typedChannelWithId lcid)
581585 case mChannel of
582586 Just channel ->
583− go (incomingAt cid ^= Just (src, ToChan channel) $ st)
587+ go (incomingAt cid ^= Just (src, ToChan chId channel) $ st)
584588 Nothing ->
585589 invalidRequest cid st $
586590 "incoming attempt to connect to unknown channel of"
@@ -1050,11 +1054,12 @@ ncEffectLocalPortSend from msg = do
10501054 -- If ch is Nothing, the process has given up the read end of
10511055 -- the channel and we simply ignore the incoming message - this
10521056 ch <- deRefWeak chan'
1053− forM_ ch $ \chan -> deliverChan msg chan
1054− where deliverChan :: forall a . Message -> TQueue a -> IO ()
1055− deliverChan (UnencodedMessage _ raw) chan' =
1057+ forM_ ch $ \chan -> deliverChan node from msg chan
1058+ where deliverChan :: forall a . LocalNode -> SendPortId -> Message -> TQueue a -> IO ()
1059+ deliverChan n p (UnencodedMessage _ raw) chan' = do
10561060 atomically $ writeTQueue chan' ((unsafeCoerce raw) :: a)
1057− deliverChan (EncodedMessage _ _) _ =
1061+ trace n (MxReceivedPort p $ unsafeCreateUnencodedMessage raw)
1062+ deliverChan _ _ (EncodedMessage _ _) _ =
10581063 -- this will not happen unless someone screws with Primitives.hs
10591064 error "invalid local channel delivery"
10601065
src/Control/Distributed/Process/UnsafePrimitives.hs+47−55
@@ -104,39 +104,26 @@ nsend :: Serializable a => String -> a -> Process ()
104104nsend label msg = do
105105 proc <- ask
106106 let us = processId proc
107− let node = localNodeId (processNode proc)
108107 let msg' = wrapMessage msg
109108 -- see [note: tracing]
110109 liftIO $ traceEvent (localEventBus (processNode proc))
111110 (MxSentToName label us msg')
112111 sendCtrlMsg Nothing (NamedSend label msg')
113112
114−-- [note: tracing]
115−-- In the remote case, we do not fire a trace event until after sending is
116−-- complete, since 'sendMessage' can block in the networking stack.
117−--
118−-- In addition, MxSent trace messages are dispatched by the node controller's
119−-- thread for local sends, which also covers local nsend, and local usend.
120−--
121−-- Also note that tracing writes to the local node's control channel, and this
122−-- module explicitly specifies to its clients that it does unsafe message
123−-- encoding. The same is true for the messages it puts onto the Management
124−-- event bus, however we do *not* want unevaluated thunks hitting the event
125−-- bus control thread. Hence the word /Unsafe/ in this module's name!
126−--
127−
128113-- | Named send to a process in a remote registry (asynchronous)
129114nsendRemote :: Serializable a => NodeId -> String -> a -> Process ()
130115nsendRemote nid label msg = do
131116 proc <- ask
132117 let us = processId proc
133118 let node = processNode proc
134− if localNodeId (processNode proc) == nid
119+ if localNodeId node == nid
135120 then nsend label msg
136− else do -- see [note: tracing] NB: this is a remote call to another NC...
137− sendCtrlMsg (Just nid) (NamedSend label (createMessage msg))
138− liftIO $ traceEvent (localEventBus node)
139− (MxSentToName label us (wrapMessage msg))
121+ else
122+ let lbl = label ++ "@" ++ show nid in do
123+ -- see [note: tracing] NB: this is a remote call to another NC...
124+ liftIO $ traceEvent (localEventBus node)
125+ (MxSentToName lbl us (wrapMessage msg))
126+ sendCtrlMsg (Just nid) (NamedSend label (createMessage msg))
140127
141128-- | Send a message
142129send :: Serializable a => ProcessId -> a -> Process ()
@@ -144,17 +131,18 @@ send them msg = do
144131 proc <- ask
145132 let node = localNodeId (processNode proc)
146133 destNode = (processNodeId them)
147− us = (processId proc) in do
148− case destNode == node of
149− True -> unsafeSendLocal them msg
150− False -> liftIO $ do sendMessage (processNode proc)
151− (ProcessIdentifier (processId proc))
152− (ProcessIdentifier them)
153− NoImplicitReconnect
154− msg
155− -- see [note: tracing]
156− liftIO $ traceEvent (localEventBus (processNode proc))
157− (MxSent them us (wrapMessage msg))
134+ us = (processId proc)
135+ msg' = wrapMessage msg in do
136+ -- see [note: tracing]
137+ liftIO $ traceEvent (localEventBus (processNode proc))
138+ (MxSent them us msg')
139+ if destNode == node
140+ then sendCtrlMsg Nothing $ LocalSend them msg'
141+ else liftIO $ sendMessage (processNode proc)
142+ (ProcessIdentifier (processId proc))
143+ (ProcessIdentifier them)
144+ NoImplicitReconnect
145+ msg
158146
159147-- | Send a message unreliably.
160148--
@@ -170,38 +158,42 @@ usend them msg = do
170158 proc <- ask
171159 let there = processNodeId them
172160 let (us, node) = (processId proc, processNode proc)
161+ let msg' = wrapMessage msg
162+ -- see [note: tracing]
163+ liftIO $ traceEvent (localEventBus node) (MxSent them us msg')
173164 if localNodeId (processNode proc) == there
174− then unsafeSendLocal them msg
175− else do sendCtrlMsg (Just there) $ UnreliableSend (processLocalId them)
176− (createMessage msg)
177− -- I would assert that is it *still* cheaper to not encode here...
178− liftIO $ traceEvent (localEventBus node)
179− (MxSent them us (wrapMessage msg))
165+ then sendCtrlMsg Nothing $ LocalSend them msg'
166+ else sendCtrlMsg (Just there) $ UnreliableSend (processLocalId them)
167+ (createMessage msg)
180168
181−unsafeSendLocal :: (Serializable a) => ProcessId -> a -> Process ()
182−unsafeSendLocal pid msg =
183− let msg' = wrapMessage msg in do
184− sendCtrlMsg Nothing $ LocalSend pid msg'
185− -- see [note: tracing]
169+-- [note: tracing]
170+-- Note that tracing writes to the local node's control channel, and this
171+-- module explicitly specifies to its clients that it does unsafe message
172+-- encoding. The same is true for the messages it puts onto the Management
173+-- event bus, however we do *not* want unevaluated thunks hitting the event
174+-- bus control thread. Hence the word /Unsafe/ in this module's name!
175+--
186176
187177-- | Send a message on a typed channel
188178sendChan :: Serializable a => SendPort a -> a -> Process ()
189179sendChan (SendPort cid) msg = do
190180 proc <- ask
191− let node = localNodeId (processNode proc)
192− destNode = processNodeId (sendPortProcessId cid) in do
193− case destNode == node of
194− True -> unsafeSendChanLocal cid msg
195− False -> do
196− liftIO $ sendBinary (processNode proc)
197− (ProcessIdentifier (processId proc))
198− (SendPortIdentifier cid)
199− NoImplicitReconnect
200− msg
181+ let node = processNode proc
182+ pid = processId proc
183+ us = localNodeId node
184+ them = processNodeId (sendPortProcessId cid)
185+ msg' = wrapMessage msg in do
186+ liftIO $ traceEvent (localEventBus node) (MxSentToPort pid cid msg')
187+ if them == us
188+ then unsafeSendChanLocal cid msg' -- NB: we wrap to P.Message !!!
189+ else liftIO $ sendBinary node
190+ (ProcessIdentifier pid)
191+ (SendPortIdentifier cid)
192+ NoImplicitReconnect
193+ msg
201194 where
202− unsafeSendChanLocal :: (Serializable a) => SendPortId -> a -> Process ()
203− unsafeSendChanLocal spId msg' =
204− sendCtrlMsg Nothing $ LocalPortSend spId (wrapMessage msg')
195+ unsafeSendChanLocal :: SendPortId -> Message -> Process ()
196+ unsafeSendChanLocal p m = sendCtrlMsg Nothing $ LocalPortSend p m
205197
206198-- | Create an unencoded @Message@ for any @Serializable@ type.
207199wrapMessage :: Serializable a => a -> Message