[OpenAFS-devel] Re: Multiple databases in ubik

[Derrick Brashear]

On Sun, Dec 18, 2011 at 9:54 AM, Jeffrey Altman
<jaltman@your-file-system.com> wrote:
> On 12/18/2011 9:31 AM, Derrick Brashear wrote:
>> On Sun, Dec 18, 2011 at 9:30 AM, Derrick Brashear <shadow@gmail.com> wro=
te:
>>> Moved to openafs-devel
>>>
>>> On Sun, Dec 18, 2011 at 9:27 AM, Hartmut Reuter <reuter@rzg.mpg.de> wro=
te:
>>>> Derrick Brashear wrote:
>>>>>
>>>>> well, i can tell you the first problem: it reuses RPCs but changes
>>>>> their signatures.
>>>>
>>>>
>>>> This is exactly what it should do in order to make sure that any mixed
>>>> configuration doesn't work. They will get back RXGEN_SS_UNMARSHAL erro=
rs
>>>> and give up.
>>>
>>> Disagree. If you want to cause them to get an error, don't implement
>>> the (different numbered) RPCs, and then RXGEN_OPCODE saves you.
>>
>> (e.g. leave them out of the .xg... or, optionally leave them in and
>> have dummy stubs which return RXGEN_OPCODE)
>
> RXGEN_OPCODE is an indication that the RPC is intentionally not
> supported and can be used to indicate failover to an alternate RPC.
>
> RXGEN_SS_UNMARSHAL is an indication that the server and client mismatch
> and the client should stop communicating with the server to prevent data
> corruption.

I'd still rather see dummy stubs that return that, then, explicitly,
rather than RPC number reuse.

> There is consensus that updating ubik RPCs does not require afs3
> standardization. =A0ubik is an internal protocol of OpenAFS. =A0We just n=
eed
> consensus regarding the design on this list.

No argument.

--=20
Derrick