[OpenAFS-devel] on sockets and devices in AFS

Tom Keiser tkeiser@sinenomine.net
Thu, 16 Sep 2010 11:14:25 +0200


On Thu, Sep 16, 2010 at 9:58 AM, Derrick Brashear <shadow@dementia.org> wrote:
> Long has it been the case that people have wanted the ability to create special devices and sockets in AFS, yet due to the non-portability of these we haven't been able to do it.
>
> Can I suggest we consider
> 1) allowing these to be created but not sent to the server
> 2) if created locally they would cover a later file of the same name if one is created
> 3) when disconnected AFS persists data across reboots, it be used to persist this data on the local client.
> 4) until then, and even then for memcache, a config file which is updated as these are created and removed, and reread at AFS start to recreate state.
>

This is an interesting concept.  I do have two small concerns:

1) to some degree, the local filename overlay semantics violate the
principle of least surprise; and
2) it should be done in such a manner that any envisioned support for
distributed sockets/named pipes not be precluded.

Regards,

-Tom