On Mon, 10 Jan 2011 10:27:25 +0100 Nicolas Bourbaki <[email protected]> wrote:
> Hi all, > > Would it be possible for someone to explain why using Unix domain > socket with AFS seems impossible. As an example, I tried to use Chrome > in conjunction with a homedir stored in AFS. The browser creates its > profile directory in the AFS space and tries to "bind" to a socket in > this directory. From that point, the browser fails to start. I guess > this is a current limitation of AFS and I'd appreciate some technical > info about that. Is this a bad idea to use Unix domain sockets and is > there a workaround to be able to use applications which make use of > them ? It's not clear to everyone exactly how a socket created in AFS should behave, and some people want different behaviors. e.g. Should stuff send to the socket stay local to the client, or should it be distributed like AFS file data in some way? Should the client tell the fileserver about the socket at all? That is not a permanent unsurmountable problem, but it probably made it harder for anyone to implement sockets in AFS. A few months ago Derrick raised the idea for a solution that at least seems to meet the needs of some socket users, without precluding a more general solution later. Some technical points about this were discussed in this thread: <https://lists.openafs.org/pipermail/openafs-devel/2010-September/017891.html> As far as I know, nobody has worked on the code to implement that yet, however. -- Andrew Deason [email protected] _______________________________________________ OpenAFS-info mailing list [email protected] https://lists.openafs.org/mailman/listinfo/openafs-info
