Execute a distributed computation inside a TransIO computation. All the computations in the TransIO monad that enclose the cloud computation must be logged

Means that this computation will be executed in the current node. the result will be logged so the closure will be recovered if the computation is translated to other node by means of primitives like beamTo, forkTo, runAt, teleport, clustered, mclustered etc

Run a distributed computation inside the IO monad. Enables asynchronous console input (see keep).

Run a distributed computation inside the IO monad with no console input.

alternative to local It means that if the computation is translated to other node this will be executed again if this has not been executed inside a local computation.

onAll foo
local foo'
local $ do
      bar
      runCloud $ do
              onAll baz
              runAt node ....
runAt node' .....

foo bar and baz will e executed locally. But foo will be executed remotely also in node' while foo' bar and baz don't.

only executes if the result is demanded. It is useful when the conputation result is only used in the remote node, but it is not serializable. All the state changes executed in the argument with setData setState etc. are lost

executes a non-serilizable action in the remote node, whose result can be used by subsequent remote invocations

subsequent remote invocatioms will send logs to this closure. Therefore logs will be shorter.

Also, non serializable statements before it will not be re-executed

the Cloud monad has no MonadIO instance. `lliftIO= local . liftIO`

`localIO = lliftIO`

continue the execution in a new node

execute in the remote node a process with the same execution state

open a wormhole to another node and executes an action on it. currently by default it keep open the connection to receive additional requests and responses (streaming)

A connectionless version of callTo for long running remote calls

Within a connection to a node opened by wormhole, it run the computation in the remote node and return the result back to the original node.

If atRemote is executed in the remote node, then the computation is executed in the original node

wormhole node2 $ do
    t <- atRemote $ do
          r <- foo              -- executed in node2
          s <- atRemote bar r   -- executed in the original node
          baz s                 -- in node2
    bat t                      -- in the original node

Execute a computation in the node that initiated the connection.

if the sequence of connections is n1 -> n2 -> n3 then `atCallingNode $ atCallingNode foo` in n3 would execute foo in n1, -- while `atRemote $ atRemote foo` would execute it in n3 atCallingNode :: Loggable a => Cloud a -> Cloud a atCallingNode proc= connectCaller $ atRemote proc

synonymous of callTo

run a single thread with that action for each connection created. When the same action is re-executed within that connection, all the threads generated by the previous execution are killed

  box <-  foo
  r <- runAt node . local . single $ getMailbox box
  localIO $ print r

if foo return different mainbox indentifiers, the above code would print the messages of the last one. Without single, it would print the messages of all of them since each call would install a new getMailBox for each one of them

run an unique continuation for each connection. The first thread that execute unique is executed for that connection. The rest are ignored.

A wormhole opens a connection with another node anywhere in a computation. teleport uses this connection to translate the computation back and forth between the two nodes connected. If the connection fails, it search the network for suitable relay nodes to reach the destination node.

wormhole without searching for relay nodes.

set remote invocations synchronous this is necessary when data is transfered very fast from node to node in a stream non-deterministically in order to keep the continuation of the calling node unchanged until the arrival of the response since all the calls share a single continuation in the calling node.

If there is no response from the remote node, the streaming is interrupted

main= keep $ initNode $  onBrowser $  do
 local $ setSynchronous True
 line  <- local $  threads 0 $ choose[1..10::Int]
 localIO $ print ("1",line)
 atRemote $ localIO $ print line
 localIO $ print ("2", line)

One problem of forwarding closures for streaming is that it could transport not only the data but extra information that reconstruct the closure in the destination node. In a single in-single out interaction It may not be a problem, but think, for example, when I have to synchronize N editors by forwarding small modifications, or worst of all, when transmitting packets of audio or video. But the size of the closure, that is, the amount of variables that I have to transport increases when the code is more complex. But transient build closures upon closures, so It has to send only what has changed since the last interaction.

In one-to-one interactions whithin a wormhole, this is automatic, but when there are different wormholes involved, it is necessary to tell explicitly what is the closure that will continue the execution. this is what localFix does. otherwise it will use the closure 0.

main= do
     filename <- local input
     source <- atServer $ local $ readFile filename
     local $ render source inEditor
    --  send upto here one single time please,  so I only stream the deltas
     localFix
     delta <- react  onEachChange
     forallNodes $ update delta

if forwardChanges send to all the nodes editing the document, the data necessary to reconstruct the closure would include even the source code of the file on EACH change. Fortunately it is possible to fix a closure that will not change in all the remote nodes so after that, I only have to send the only necessary variable, the delta. This is as efficient as an hand-made socket writeforkThreadreadSocket loop for each node.

forward exceptions back to the calling node

copy a session data variable from the local to the remote node. If there is none set in the local node, The parameter is the default value. In this case, the default value is also set in the local node.

execute a Transient action in each of the nodes connected.

The response of each node is received by the invoking node and processed by the rest of the procedure. By default, each response is processed in a new thread. To restrict the number of threads use the thread control primitives.

this snippet receive a message from each of the simulated nodes:

main = keep $ do
   let nodes= map createLocalNode [2000..2005]
   addNodes nodes
   (foldl (<|>) empty $ map listen nodes) <|> return ()

   r <- clustered $ do
              Connection (Just(PortNumber port, _, _, _)) _ <- getSData
              return $ "hi from " ++ show port++ "\n"
   liftIO $ putStrLn r
   where
   createLocalNode n= createNode "localhost" (PortNumber n)

Setup the node to start listening for incoming connections.

filter out HTTP requests

if there is a remote job identified by th string identifier, it stop that job, and set the current remote operation (if any) as the current remote job for this identifier. The purpose is to have a single remote job. to identify the remote job, it should be used after the wormhole and before the remote call:

r <- wormhole node $ do
       stopRemoteJob "streamlog"
       atRemote myRemotejob

So:

runAtUnique ident node job= wormhole node $ do stopRemoteJob ident; atRemote job

kill the remote job. Usually, before starting a new one.

forward all the result of the Transient computation to the opened connection

Returns True if we are running in the browser.

Create a node from a hostname (or IP address), port number and a list of services.

Return the local node i.e. the node where this computation is running.

empty if the node is not set

Return the list of nodes in the cluster.

get the nodes that have the same service definition that the calling node

Add a list of nodes to the list of existing nodes know locally. If the node is already present, It add his services to the already present node services which have the first element equal (usually the "name" field) will be substituted if the match

set the list of nodes

Shuffle the list of cluster nodes and return the shuffled list.

add this node to the list of know nodes in the remote node connected by a wormhole. This is useful when the node is called back by the remote node. In the case of web nodes with webSocket connections, this is the way to add it to the list of known nodes in the server.

Add a node (first parameter) to the cluster using a node that is already part of the cluster (second parameter). The added node starts listening for incoming connections and the rest of the computation is executed on this newly added node.

Reconcile the list of nodes in the cluster using a remote node already part of the cluster. Reconciliation end up in each node in the cluster having the same list of nodes.

crawl the nodes executing the same action in each node and accumulate the results using a binary operator

only execute if the the program is executing in the browser. The code inside can contain calls to the server. Otherwise return empty (so it stop the computation and may execute alternative computations).

only executes the computaion if it is in the server, but the computation can call the browser. Otherwise return empty

If the computation is running in the server, translates i to the browser and return back. If it is already in the browser, just execute it

If the computation is running in the browser, translates i to the server and return back. If it is already in the server, just execute it