[OpenAFS-devel] unix sockets (and maybe devices?)

[Tom Keiser]

On Thu, Sep 20, 2012 at 2:05 AM, Troy Benjegerdes <hozer@hozed.org> wrote:
>> Perhaps...once the following conditions are met:
>>
>> 1. we've either: a) resolved the local/distributed debate, or b) found
>> a way to do local-only now without precluding distributed from being a
>> future strategy (the complications of doing local-only--sans a
>> cohesive strategy for implementing distributed fifos in the
>> future--have been discussed at-length on this list in the past);
>
> Was there any sort of resolution, or did we just have a stalemate
> of opinion?
>

It seemed rather unresolved when I last paid attention...


> I see the point of having a distributed toolkit for heterogenous
> systems, and there could be some really interesting and powerful
> stuff you could do with distributed unix sockets protected with
> filesystem ACL's. (Say collaborative editing for a document, ala
> Etherpad, for instance)
>
> But in the meantime, the perfect is the enemy of the good, and
> lack of functioning sockets makes it a real pain to use the amazing
> distributed toolkit we *do* have for things using it as the rootfs
> for a 'cloud' based operating system.
>

Point taken.  My _hope_ would be that local-only could be architected
in a manner forwards-compatible with distributed sockets/fifos.  If
that goal cannot be met, then, alas.  I concede the good is likely
better than the perfect in this context.


> What's the roadmap for first, local-only unix sockts, and then
> defining the right rpcs so the code might someday be written?

Well, at least from my perspective, a big issue is: how do you handle
allocation?  Waiting for standardization of a next-generation vnode
create RPC would be ideal, however would likely require you to wait an
unacceptably long period of time.  I suppose you could take the
disconnected-like approach (of doing it as a purely-local vnode and
dirent allocation--that gets merged upon parent dir DV change),
although that necessarily involves some rather unfortunate collision
resolution semantics...

-Tom