Michael Alan Dorman > Memcached-Client-2.01 > NOTES.pod

Download:
Memcached-Client-2.01.tar.gz

Annotate this POD

CPAN RT

New  2
Open  2
View/Report Bugs
Source  

How it all fits together ^

Memcached::Client is the central hub

During the loading of the Memcached::Client::Request module, we insert a bunch of methods into the Memcached::Client namespace that are really just curried functions that create instances of the various Request classes.

When you make a request, things go like this (using get as an example):

$client->get calls the anonymous subroutine that was returned when M::C::Request::Get->generate (which, due to inheritance, is M::C::Request->generate) was called.

The anonymous sub expects the Memcached::Client object as its first parameter, and uses the remainder as arguments for its command. The command that was curried in when the subroutine was created is used to construct a hashref that will become the request object.

The last argument to the subroutine is examined. If it was something we can use as a callback (either a CODE reference or an AnyEvent::CondVar, it is stored, and we assume that somewhere else in the calling program there is a statement waiting on an AnyEvent::CondVar---so we will not be obliged to do so ourselves in order for the request to get processed.

Otherwise we create the aforementioned AnyEvent::CondVar and mark ourselves as needing to wait.

We then hand the rest of our arguments to the process routine, which is supposed to verify the object's arguments and returns one or more Request objects to be submitted to the appropriate server(s). If it does not return any objects, it is assumed that there was a problem with the submission, and an empty result is returned.

The __submit routine in the Memcached::Client module is handed the request objects to process.

__submit iterates through the objects, looking at whether they include a key member. If they do, the key is processed to determine what server it should be directed at, checked for validity, and if it's OK, it is submitted to the appropriate server's queue using the connection's enqueue method.

If the object doesn't have a key, it is assumed to be a broadcast message destined for all servers, and __submit uses the object's server factory to create subrequests for each server.

The connection object works its way through its queue (if there's nothing else in the queue when you add a new request, it will be handled immediately, but that's just an optimization) and comes to your request object.

It gets the name of the protocol function from your request object, and invokes that method on the protocol object, passing it a reference to itself and your request.

The protocol method constructs the command from the data members of the request object and submits it to the connection object.

When the response to the request has been recieved the protocol object decodes it and sends the result back to the request object using its result method, and notifies the connection that it's done with the request by calling complete.

syntax highlighting: