[AFS3-std] fileserver to fileserver RPCs

[Jeffrey Hutzelman]

--On Tuesday, May 11, 2010 02:20:13 PM -0400 Derrick Brashear 
<shadow@gmail.com> wrote:

> On Tue, May 11, 2010 at 1:56 PM, Jeffrey Hutzelman <jhutz@cmu.edu> wrote:
>> --On Tuesday, May 11, 2010 09:45:14 AM -0400 Derrick Brashear
>> <shadow@gmail.com> wrote:
>>
>>> As part of the work done by Vishal Powar, my Google Summer of Code
>>> student two years ago, we have the start of readwrite replication in
>>> RT (http://rt.central.org/rt/Ticket/Display.html?id=114116)
>>>
>>> It necessarily adds a server to server RPC protocol, which is below.
>>> However, it is effectively the needed subset of RPCs to create data,
>>> and the client ViceId of the caller.
>>>
>>> I intend to offer a standards proposal for this, instead using the
>>> signatures of the proposed "RPC refresh" RPCs, again with the client
>>> ViceIds added, with the idea that an additional Rx "service" is run on
>>> each fileserver to accept these.
>>>
>>> Would this be a reasonable proposal?
>>
>> Sounds reasonable.
>> For a new service, let's start RPC numbers at 1; there's no reason to
>> start with large values in a service where we don't have to worry about
>> IBM allocating numbers out-of-process.
>
> I meant to change those to XXX; That was from his reference patch,
> which used the existing Rx service and just added RPCs, while he was
> testing.

If you're creating a new service, you should go ahead and pick RPC numbers. 
New registries should come with initial contents, not an initial backlog of 
requests. :-)