Remote Procedure Call

We use the term Remote Procedure Call here to mean any type of procedure call between one process and another. All of the remote procedure calls we're talking about here (up to now) are all really between processes on the same machine, not across machines.

This greatly simplifies the security aspects of this design, since ClientAuthentication[1] is much simpler when clients are local on the same machine.

Normal (synchronous) remote procedure calls are performed by:

Client packaging the arguments up in a big blob
Client sends them to the server
Server receives the request and unpackages the arguments
Server processes the request
Server packages up return results into a blob
Server sends return results to client

Asynchronous remote procedure calls are performed similarly, but the return result is sent later.

Client packaging the arguments up in a big blob
Client sends them to the server
Server receives the request and unpackages the arguments
Server validates the request, assigns it an AsyncRequestId[2], and queues it for execution
Server packages the AsyncRequestId[2] (or error return)
Server sends the immediate return result to the client
Server performs the request specified by the AsyncRequestId[2]
Request completes on the server
Server packages up return results into a blob
Server sends return results to client as an AsyncEventMessage[3]
Client reads AsyncEventMessage[3] to learn the real return result from the request.

The process of "packaging up arguments into a big blob" and doing the unpackaging goes by various names. Common names include marshalling/demarshalling, pickling, or serializing.

References

[1]	http://www.linux-ha.org/ClientAuthentication
[2]	http://www.linux-ha.org/AsyncRequestId
[3]	http://www.linux-ha.org/AsyncEventMessage