agent: do not send worker crash errors when agent is not active (stopped) (#1256)

Author: epoberezkin

src/Simplex/Messaging/Agent.hs

@@ -2112,8 +2112,7 @@ cleanupManager c@AgentClient {subQ} = do
   liftIO $ threadDelay' delay
   int <- asks (cleanupInterval . config)
   ttl <- asks $ storedMsgDataTTL . config
-  forever $ do
-    liftIO $ waitUntilActive c
+  forever $ waitActive $ do
     run ERR deleteConns
     run ERR $ withStore' c (`deleteRcvMsgHashesExpired` ttl)
     run ERR $ withStore' c (`deleteSndMsgsExpired` ttl)
@@ -2133,6 +2132,7 @@ cleanupManager c@AgentClient {subQ} = do
       step <- asks $ cleanupStepInterval . config
       liftIO $ threadDelay step
     -- we are catching it to avoid CRITICAL errors in tests when this is the only remaining handle to active
+    waitActive :: ReaderT Env IO a -> AM' ()
     waitActive a = liftIO (E.tryAny $ waitUntilActive c) >>= either (\_ -> pure ()) (\_ -> void a)
     deleteConns =
       withLock (deleteLock c) "cleanupManager" $ do

src/Simplex/Messaging/Agent/Client.hs

@@ -372,21 +372,23 @@ getAgentWorker' toW fromW name hasWork c key ws work = do
           restart <- atomically $ getWorker >>= maybe (pure False) (shouldRestart e_ (toW w) t maxRestarts)
           when restart runWork
         shouldRestart e_ Worker {workerId = wId, doWork, action, restarts} t maxRestarts w'
-          | wId == workerId (toW w') =
-              checkRestarts . updateRestartCount t =<< readTVar restarts
+          | wId == workerId (toW w') = do
+              rc <- readTVar restarts
+              isActive <- readTVar $ active c
+              checkRestarts isActive $ updateRestartCount t rc
           | otherwise =
               pure False -- there is a new worker in the map, no action
           where
-            checkRestarts rc
-              | restartCount rc < maxRestarts = do
+            checkRestarts isActive rc
+              | isActive && restartCount rc < maxRestarts = do
                   writeTVar restarts rc
                   hasWorkToDo' doWork
                   void $ tryPutTMVar action Nothing
                   notifyErr INTERNAL
                   pure True
               | otherwise = do
                   TM.delete key ws
-                  notifyErr $ CRITICAL True
+                  when isActive $ notifyErr $ CRITICAL True
                   pure False
               where
                 notifyErr err = do