Distributed Computing?

Creating a remote request?

We can connect to a remote host and perform remote computation via a remote isolate in much the same way as we can create a new isolate. The syntax is both elegant and concise. Let's look at an example:

rm = Remote('localhost')//establish a connection to a remote Concurnas server
rm.connect()//optionally called
result int: = {1+1}!(rm.onfail(e=>e.retry()))//result => 2
rm.close()//close wait for result of our isolate above to be set

Alternatively we can define the isolate to execute as follows:

rm = Remote('localhost')
rm.connect()//optionally called
def func() => 1+1
result int: = func()!(rm.onfail(e=>e.retry()))//result => 2
rm.close()//close wait for result of our isolate above to be set

In the above cases, first we establish a connection to a remote Concurnas server by creating a Remote connector with a hostname/ip address and optional port (which defaults to 42000 if unspecified): Remote('localhost'). Note that Remote is automatically imported.

We next call connect() to establish a network connection to our remote server. This is an optional step, it will be called upon first submission of an isolate if it a connection has not already been established.

Next we attempt to submit the isolate {1+1}! for execution remotely. This occurs as normal but except for us passing an executor on the right hand side of the bang ! operator. This executor is returned from the remote objects onfail method, which obliges the remote request initiator to handle the error case of a isolate task submission failing at inception by passing an object of type com.concurnas.dist.RemoteFailureHandler to handle this occurrence. RemoteFailureHandler is a SAM type thus we are able to use the compact lambda syntax to pass in a handler above. Our error handling strategy above is to simply re-attempt to submit the isolate task if it cannot be initially submit or a server side execution error occurs.

An object of type com.concurnas.dist.RemoteFailureContext is passed to the provided RemoteFailureHandler. This contains the following noteworthy utility methods:

• retry() - Attempt to re run the isolate.

• result() - Returns the ref to which the return value of the remotely spawned isolate would have its return value assigned. Returns null if the isolate does not return a value.

• reason() - Returns the Exception indicating the cause of failure.

The error handler can even be stateful, for example:

from com.concurnas.dist import RemoteFailureContext, RemoteFailureContext

rem = Remote('localhost')
result = {1+1}!(rem.onfail( new RemoteFailureHandler{  
  attempts = 1
	
  def handle(e RemoteFailureContext){
    if(attempts++ > 10){//retry 10 times then give up
      e.result():setException(e.reason())
    }else{
      e.retry()
    }
  }
} ))
//result == 2

If the error handler itself throws an exception, then this exception will be set on the ref returned from the isolate. If no ref is returned from the isolate, then this exception will be passed to the default isolate error handler (this is generally not a desirable situation).

Remember that an exception occurring within the bounds of the isolate itself, does not constitute a distributed computing communication (or otherwise) failure, and is recorded on the exception value of the ref as normal - the error handler discussed here does not catch this.

Being an isolate, of which the resultant type is a ref, allows us to otherwise carry on with execution asynchronously until the point where the value of the ref is required. This is an incredibly powerful mechanism, it's feasible to envisage hundreds or even thousands of these remote requests being spawned across thousands of machines in order to solve a complex problem.

When we are finished with the remote connection we can call the close() method in order to wait for the result of any pending isolates and finally disconnect from the remote server once the results (positive or an exception) are known. By implementing the close method in this way, we can use remote connections within try-with-resources blocks. For example:

try(rm = Remote('localhost')){//close rm on completion of this block
  rm.connect()//optionally called
  result int: = {1+1}!(rem.onfail(e=>e.retry()))//result => 2
}

A second variant of the close() method exists for remote connections. The close() method takes a boolean parameter, close(hard boolean). If this is set to true then the connection will be closed immediately, any outstanding remotely spawned isolates will have an exception set on their resultant ref:

rm = Remote('localhost')
rm.connect()//optionally called
result int: = {1+1}!(rem.onfail(e=>e.retry()))//result => 2
rm.close(hard = true)//do not wait for the result above to complete

In the above case of an explicit hard disconnection any outstanding isolates will have an ClientPrematureDisconnection exception set on their result refs.

Some care should be taken with hard disconnections, since we are not waiting for the result of our submit isolate before closing our connection. Closing a connection will prevent as yet unstarted remote isolates from starting.

It is essential to call the close method when one is finished using the remote connection because just like other forms of io (e.g. files), failing to do so will keep the connection open resulting in a resource leak.

Creating a remote server?

A remote server can easily be created in Concurnas by using the following code:

from com.concurnas.lang.dist import RemoteServer
rmServer = RemoteServer()
remServer.startServer()//start remote server
//...use remote server...
rmServer.close()

A port may optionally be specified, if not then the server will be spawned on port: 42000.

It is essential to call the close method when one is finished using the remote server because just like other forms of I/O (e.g. files), failing to do so will keep all its client connections open resulting in a resource leak.

from java.security import PermissionCollection, Permissions, AllPermission
perms = Permissions()
perms.add(new AllPermission())

remServer = RemoteServer(permissions = perms)