Is there a recommended pattern for long-running processes

[bruce kissinger]

I have a situation where I want to run a Java program as a CloudI service but the time that the process takes to run can vary significantly because of circumstances outside of my control.  I can increase the timeout values but that can cause other side effects.  Is there a preferred pattern for long running processes?

My current approach is to have a Linux shell script that runs some pseudo-code like this:

do {
  TransactionID = send_async(Process name);
  
   Response = recv_async(TransactionID);

   if (Response is not null) then {
     # Process the response
     # Set variable to exit loop

  } else {
     sleep for a while
  }

} while exit variable set


Regards,

Bruce

[Michael Truog]

The timeouts used only define what is an error condition, so you can use a maximum timeout value such that it should never be exceeded.  Timeouts are enforced for initialization, termination, and service request handling, so all service execution is covered (when considering all the common programming language functionality, i.e., Erlang usage of cloudi_service_handle_info/3 in the cloudi_service behaviour is the only thing I can think of without a timeout, and that is due to Erlang integration requirements).

If the service request handling timeout can not have a decent maximum (exceeding the 32bit timeout value, or too much variation causing fault-tolerance problems) you can choose to not use the return of the service request for the result and instead use a new service request to provide the result (when a service request provides no result, the service request destination is making it a completely asynchronous service request, also, when the service request provides no result, it is providing an empty binary, i.e., <<>> with Erlang syntax, for both ResponseInfo and Response).  That is the approach used by cloudi_service_queue when it is in 'both' mode, since its service requests must be able to survive a CloudI node restart.

So, with your example below, you need to either make sure your send_async timeout is the absolute maximum of any latency that is acceptable, or you should use send_async without recv_async and instead have a subscription that allows the response to arrive as a separate service request.

--
You received this message because you are subscribed to the Google Groups "CloudI Questions" group.
To unsubscribe from this group and stop receiving emails from it, send an email to cloudi-questio...@googlegroups.com.
For more options, visit https://groups.google.com/d/optout.