git.iko.soy/distributed-process
on master help sign in
10 files changed+364−182
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+3−1 distributed-process-tests/src/Control/Distributed/Process/Tests/Mx.hs+165−116 distributed-process.cabal+1−1 src/Control/Distributed/Process/Internal/Primitives.hs+15−6 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+15−13
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+3−1
@@ -91,6 +91,7 @@ import Test.HUnit (Assertion, assertFailure)
9191import Test.HUnit.Base (assertBool)
9292
9393import GHC.Generics
94+import System.Timeout (timeout)
9495
9596-- | A mutable cell containing a test result.
9697type TestResult a = MVar a
@@ -233,7 +234,8 @@ testProcessReport pid = do
233234tryRunProcess :: LocalNode -> Process () -> IO ()
234235tryRunProcess node p = do
235236 tid <- liftIO myThreadId
236− 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))
237239
238240tryForkProcess :: LocalNode -> Process () -> IO ProcessId
239241tryForkProcess node p = do
distributed-process-tests/src/Control/Distributed/Process/Tests/Mx.hs+165−116
@@ -1,17 +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−import Control.Exception (SomeException)
78import Control.Distributed.Process hiding (bracket, finally, try)
8−import Control.Distributed.Process.Internal.Types (ProcessExitException(..))
9+import Control.Distributed.Process.Internal.Types
10+ ( ProcessExitException(..)
11+ , unsafeCreateUnencodedMessage
12+ )
913import Control.Distributed.Process.Node
1014import qualified Control.Distributed.Process.UnsafePrimitives as Unsafe
1115 ( send
1216 , nsend
1317 , nsendRemote
1418 , usend
19+ , sendChan
1520 )
1621import Control.Distributed.Process.Management
1722 ( MxEvent(..)
@@ -32,8 +37,8 @@ import Control.Monad (void, unless)
3237import Control.Monad.Catch(finally, bracket, try)
3338import Control.Rematch (equalTo)
3439import Data.Binary
35−import Data.List (find, sort)
36−import Data.Maybe (isJust, fromJust, isNothing)
40+import Data.List (find, sort, intercalate)
41+import Data.Maybe (isJust, fromJust, isNothing, fromMaybe, catMaybes)
3742import Data.Typeable
3843import GHC.Generics hiding (from)
3944#if ! MIN_VERSION_base(4,6,0)
@@ -61,7 +66,7 @@ awaitExit pid =
6166 where
6267 withMonitorRef p = bracket (monitor p) unmonitor
6368
64−testAgentBroadcast :: TestResult () -> Process ()
69+testAgentBroadcast :: TestResult (Maybe ()) -> Process ()
6570testAgentBroadcast result = do
6671 (resultSP, resultRP) <- newChan :: Process (SendPort (), ReceivePort ())
6772
@@ -76,10 +81,10 @@ testAgentBroadcast result = do
7681 mxNotify ()
7782 -- Once the publisher has seen our message, it will broadcast the Publish
7883 -- and the consumer will see that and send the result to our typed channel.
79− stash result =<< receiveChan resultRP
84+ stash result =<< receiveChanTimeout 10000000 resultRP
8085
81− kill publisher "finished" >> awaitExit publisher
82− kill consumer "finished" >> awaitExit consumer
86+ kill publisher "finished"
87+ kill consumer "finished"
8388
8489testAgentDualInput :: TestResult (Maybe Int) -> Process ()
8590testAgentDualInput result = do
@@ -94,6 +99,8 @@ testAgentDualInput result = do
9499 else mxReady)
95100 ]
96101
102+ mRef <- monitor s
103+
97104 mxNotify (1 :: Int)
98105 nsend "sum-agent" (3 :: Int)
99106 mxNotify (2 :: Int)
@@ -101,7 +108,10 @@ testAgentDualInput result = do
101108 mxNotify (5 :: Int)
102109
103110 stash result =<< receiveChanTimeout 10000000 rp
104− awaitExit s
111+ died <- receiveTimeout 10000000 [
112+ matchIf (\(ProcessMonitorNotification r _ _) -> r == mRef) (const $ return True)
113+ ]
114+ died `shouldBe` equalTo (Just True)
105115
106116testAgentPrioritisation :: TestResult [String] -> Process ()
107117testAgentPrioritisation result = do
@@ -115,10 +125,10 @@ testAgentPrioritisation result = do
115125 (sp, rp) <- newChan
116126 s <- mxAgent (MxAgentId name) ["first"] [
117127 mxSink (\(s :: String) -> do
118− mxUpdateLocal ((s:))
128+ mxUpdateLocal (s:)
119129 st <- mxGetLocal
120130 case length st of
121− n | n == 5 -> do liftMX $ sendChan sp st
131+ n | n == 5 -> do liftMX $ sendChan sp (sort st)
122132 mxDeactivate "finished"
123133 _ -> mxReceive -- go to the mailbox
124134 )
@@ -129,7 +139,7 @@ testAgentPrioritisation result = do
129139 mxNotify "fourth"
130140 nsend name "fifth"
131141
132− stash result . sort =<< receiveChan rp
142+ stash result . fromMaybe [] =<< receiveChanTimeout 10000000 rp
133143 awaitExit s
134144
135145testAgentMailboxHandling :: TestResult (Maybe ()) -> Process ()
@@ -142,12 +152,19 @@ testAgentMailboxHandling result = do
142152 nsend "mailbox-agent" ()
143153
144154 stash result =<< receiveChanTimeout 1000000 rp
145− kill agent "finished" >> awaitExit agent
155+ kill agent "finished"
146156
147157testAgentEventHandling :: TestResult Bool -> Process ()
148158testAgentEventHandling result = do
149− 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"
150166
167+ let initState = [] :: [MxEvent]
151168 (rc, rs) <- newChan
152169 agentPid <- mxAgent (MxAgentId "lifecycle-listener-agent") initState [
153170 (mxSink $ \ev -> do
@@ -180,8 +197,7 @@ testAgentEventHandling result = do
180197 -- TODO: yes, this is racy, but we're at the mercy of the scheduler here...
181198 faff 2000000 rs
182199
183− _ <- monitor agentPid
184− (sp, rp) <- newChan
200+ (sp, rp) <- newChan :: Process (SendPort (), ReceivePort ())
185201
186202 pid <- spawnLocal $ expect >>= sendChan sp
187203
@@ -190,9 +206,11 @@ testAgentEventHandling result = do
190206 monitor pid
191207
192208 send pid ()
193− () <- receiveChan rp
209+ rct <- receiveChanTimeout 10000000 rp
210+ unless (isJust rct) $ die "No response on channel"
194211
195− receiveWait [ match (\(ProcessMonitorNotification _ _ _) -> return ()) ]
212+ pmn <- receiveTimeout 2000000 [ match (\ProcessMonitorNotification{} -> return ()) ]
213+ unless (isJust pmn) $ die "No monitor notification arrived"
196214
197215 -- TODO: yes, this is racy, but we're at the mercy of the scheduler here...
198216 faff 2000000 rs
@@ -201,17 +219,16 @@ testAgentEventHandling result = do
201219 mxNotify (MxSpawned pid, replyTo)
202220 mxNotify (MxProcessDied pid DiedNormal, replyTo)
203221
204− seenAlive <- receiveChan reply
205− seenDead <- receiveChan reply
222+ seenAlive <- receiveChanTimeout 10000000 reply
223+ seenDead <- receiveChanTimeout 10000000 reply
206224
207− stash result $ seenAlive && seenDead
208− kill agentPid "test-complete" >> awaitExit agentPid
225+ stash result $ foldr (&&) True $ catMaybes [seenAlive, seenDead]
226+ kill timer "test-complete"
227+ kill agentPid "test-complete"
209228 where
210229 faff delay port = do
211230 res <- receiveChanTimeout delay port
212− unless (isNothing res) $ do
213− pause delay
214− faff delay port
231+ unless (isNothing res) $ pause delay
215232
216233testMxRegEvents :: TestResult () -> Process ()
217234testMxRegEvents result = do
@@ -258,7 +275,6 @@ testMxRegEvents result = do
258275 reg3 `shouldBe` equalTo (Just (label, p1))
259276
260277 mapM_ (flip kill $ "test-complete") [agent, p1, p2]
261− awaitExit agent
262278
263279testMxRegMon :: LocalNode -> TestResult () -> Process ()
264280testMxRegMon remoteNode result = do
@@ -316,64 +332,99 @@ testMxRegMon remoteNode result = do
316332 evts `shouldContain` (Just (label1, p1))
317333 evts `shouldContain` (Just (label2, p1))
318334
319− kill agent "test-complete" >> awaitExit agent
335+ kill agent "test-complete"
320336
321337ensure :: Process () -> Process () -> Process ()
322338ensure = flip finally
323339
324−type SendTest = ProcessId -> ReceivePort MxEvent -> Process Bool
325−
326340testNSend :: (String -> () -> Process ())
327341 -> Maybe LocalNode
328− -> TestResult Bool
329342 -> Process ()
330−testNSend op n r = testMxSend n r $ \p1 sink -> do
343+testNSend op n = do
344+ us <- getSelfPid
331345 let delay = 5000000
332− let label = "testMxSend"
346+ let label = "testMxSend" ++ (show us)
333347 let isValid = isValidLabel n label
334348
335− register label p1
336− reg1 <- receiveChanTimeout delay sink
337− case reg1 of
338− Just (MxRegistered pd lb)
339− | pd == p1 && isValid lb -> return ()
340− _ -> die $ "Reg-Failed: " ++ show reg1
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
341356
342− op label ()
357+ op label ()
343358
344− us <- getSelfPid
345− sent <- receiveChanTimeout delay sink
346− case sent of
347− Just (MxSentToName lb by _)
348− | by == us && isValid lb -> return True
349− _ -> die $ "Send-Failed: " ++ show sent
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
350364
351365 where
352− isValidLabel n l1 l2
366+ isValidLabel nd l1 l2
353367 | l2 == l2 = True
354− | isJust n = l2 == l1 ++ "@" ++ show (localNodeId $ fromJust n)
368+ | isJust nd = l2 == l1 ++ "@" ++ show (localNodeId $ fromJust nd)
355369 | otherwise = False
356370
357−testSend :: (ProcessId -> () -> Process ()) -> Maybe LocalNode -> TestResult Bool -> Process ()
358−testSend op n r = testMxSend n r $ \p1 sink -> do
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+
359405 -- initiate a send
360− op p1 ()
406+ op ch' ()
361407
362408 -- verify the management event
363− us <- getSelfPid
364409 sent <- receiveChanTimeout 5000000 sink
365410 case sent of
366− Just (MxSent pidTo pidFrom _)
367− | pidTo == p1 && pidFrom == us -> return True
368− _ -> return False
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
369416
370−testMxSend :: Maybe LocalNode -> TestResult Bool -> SendTest -> Process ()
371−testMxSend mNode result test = do
417+testMxSend :: Maybe LocalNode -> String -> SendTest -> Process ()
418+testMxSend mNode label test = do
372419 us <- getSelfPid
420+ (sp, rp) <- newChan
373421 (chan, sink) <- newChan
374422 agent <- mxAgent (MxAgentId $ agentLabel us) () [
375423 mxSink $ \ev -> do
376424 case ev of
425+ m@(MxSentToPort _ cid _)
426+ | cid /= sendPortId chan
427+ && cid /= sendPortId sp -> liftMX $ sendChan chan m
377428 m@(MxSent _ fromPid _)
378429 | fromPid == us -> liftMX $ sendChan chan m
379430 m@(MxSentToName _ fromPid _)
@@ -384,29 +435,39 @@ testMxSend mNode result test = do
384435 mxReady
385436 ]
386437
387− (sp, rp) <- newChan
388438 case mNode of
389439 Nothing -> void $ spawnLocal (proc sp)
390440 Just remoteNode -> void $ liftIO $ forkProcess remoteNode $ proc sp
391441
392− p1 <- receiveChan rp
393− res <- try (test p1 sink)
394− case res of
395− Left (ProcessExitException _ m) -> (liftIO $ putStrLn $ "SomeException-" ++ show m) >> stash result False
396− Right tr -> stash result tr
442+ p1 <- receiveChanTimeout 2000000 rp
443+ unless (isJust p1) $ die "Timed out waiting for ProcessId"
444+ res <- try $ test (fromJust p1) sink
445+
397446 kill agent "bye"
398− kill p1 "bye"
447+ kill (fromJust p1) "bye"
448+
449+ case res of
450+ Left (ProcessExitException _ m) -> (liftIO $ putStrLn $ "SomeException-" ++ show m) >> die m
451+ Right tr -> tr `shouldBe` equalTo True
452+
399453
400454 where
401− label = "testMxSend"
402455 agentLabel s = "mx-unsafe-check-agent-" ++ show s
403− proc sp' = getSelfPid >>= sendChan sp' >> expect :: Process ()
456+ proc sp' = getSelfPid >>= sendChan sp' >> go Nothing
457+
458+ go :: Maybe (ReceivePort ()) -> Process ()
459+ go Nothing = receiveTimeout 5000000 [ match replyChannel ] >>= go
460+ go c@(Just c') = receiveWait [ matchChan c' return ] >> go c
461+
462+ replyChannel p' = do
463+ (s, r) <- newChan
464+ send p' s
465+ return r
404466
405467tests :: TestTransport -> IO [Test]
406468tests TestTransport{..} = do
407469 node1 <- newLocalNode testTransport initRemoteTable
408470 node2 <- newLocalNode testTransport initRemoteTable
409− let nid = localNodeId node2
410471 return [
411472 testGroup "MxAgents" [
412473 testCase "EventHandling"
@@ -416,7 +477,7 @@ tests TestTransport{..} = do
416477 , testCase "InterAgentBroadcast"
417478 (delayedAssertion
418479 "expected (), but no broadcast was received"
419− node1 () testAgentBroadcast)
480+ node1 (Just ()) testAgentBroadcast)
420481 , testCase "AgentMailboxHandling"
421482 (delayedAssertion
422483 "expected (Just ()), but no regular (mailbox) input was handled"
@@ -441,53 +502,41 @@ tests TestTransport{..} = do
441502 (delayedAssertion
442503 "expected process deaths to result in unregistration events"
443504 node1 () (testMxRegMon node2))
444− , testGroup "SentEvents" [
445− testGroup "RemoteTargets" [
446− testCase "Unsafe.nsend"
447− (delayedAssertion "expected mx events failed"
448− node1 True (testNSend Unsafe.nsend $ Just node2))
449− , testCase "Unsafe.nsendRemote"
450− (delayedAssertion "expected mx events failed"
451− node1 True (testNSend (Unsafe.nsendRemote nid) $ Just node2))
452− , testCase "Unsafe.send"
453− (delayedAssertion "expected mx events failed"
454− node1 True (testSend Unsafe.send $ Just node2))
455− , testCase "Unsafe.usend"
456− (delayedAssertion "expected mx events failed"
457− node1 True (testSend Unsafe.usend $ Just node2))
458− , testCase "nsend"
459− (delayedAssertion "expected mx events failed"
460− node1 True (testNSend nsend $ Just node2))
461− , testCase "nsendRemote"
462− (delayedAssertion "expected mx events failed"
463− node1 True (testNSend (nsendRemote nid) $ Just node2))
464− , testCase "send"
465− (delayedAssertion "expected mx events failed"
466− node1 True (testSend send $ Just node2))
467− , testCase "usend"
468− (delayedAssertion "expected mx events failed"
469− node1 True (testSend usend $ Just node2))
470− ]
471− , testGroup "LocalTargets" [
472− testCase "Unsafe.nsend"
473− (delayedAssertion "expected mx events failed"
474− node1 True (testNSend Unsafe.nsend Nothing))
475− , testCase "Unsafe.send"
476− (delayedAssertion "expected mx events failed"
477− node1 True (testSend Unsafe.send Nothing))
478− , testCase "Unsafe.usend"
479− (delayedAssertion "expected mx events failed"
480− node1 True (testSend Unsafe.usend Nothing))
481− , testCase "nsend"
482− (delayedAssertion "expected mx events failed"
483− node1 True (testNSend nsend Nothing))
484− , testCase "send"
485− (delayedAssertion "expected mx events failed"
486− node1 True (testSend send Nothing))
487− , testCase "usend"
488− (delayedAssertion "expected mx events failed"
489− node1 True (testSend usend Nothing))
505+ , testGroup "SendEvents" $ buildTestCases node1 node2
506+ ]
507+ ]
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+ ])
490528 ]
491− ]
492− ]
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)]
493542 ]
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+15−6
@@ -307,7 +307,13 @@ expect = receiveWait [match return]
307307-- Channels --
308308--------------------------------------------------------------------------------
309309
310−-- | 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+--
311317newChan :: Serializable a => Process (SendPort a, ReceivePort a)
312318newChan = do
313319 proc <- ask
@@ -335,13 +341,16 @@ newChan = do
335341sendChan :: Serializable a => SendPort a -> a -> Process ()
336342sendChan (SendPort cid) msg = do
337343 proc <- ask
338− let node = localNodeId (processNode proc)
339− destNode = processNodeId (sendPortProcessId cid) in do
340− 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
341350 True -> sendChanLocal cid msg
342351 False -> do
343− liftIO $ sendBinary (processNode proc)
344− (ProcessIdentifier (processId proc))
352+ liftIO $ sendBinary node
353+ (ProcessIdentifier pid)
345354 (SendPortIdentifier cid)
346355 NoImplicitReconnect
347356 msg
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+15−13
@@ -178,20 +178,22 @@ usend them msg = do
178178sendChan :: Serializable a => SendPort a -> a -> Process ()
179179sendChan (SendPort cid) msg = do
180180 proc <- ask
181− let node = localNodeId (processNode proc)
182− destNode = processNodeId (sendPortProcessId cid) in do
183− case destNode == node of
184− True -> unsafeSendChanLocal cid msg
185− False ->
186− liftIO $ sendBinary (processNode proc)
187− (ProcessIdentifier (processId proc))
188− (SendPortIdentifier cid)
189− NoImplicitReconnect
190− 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
191194 where
192− unsafeSendChanLocal :: (Serializable a) => SendPortId -> a -> Process ()
193− unsafeSendChanLocal spId msg' =
194− sendCtrlMsg Nothing $ LocalPortSend spId (wrapMessage msg')
195+ unsafeSendChanLocal :: SendPortId -> Message -> Process ()
196+ unsafeSendChanLocal p m = sendCtrlMsg Nothing $ LocalPortSend p m
195197
196198-- | Create an unencoded @Message@ for any @Serializable@ type.
197199wrapMessage :: Serializable a => a -> Message