[distcc] re: distributed compiler cache

[joerg.beyer at email.de]

MartinPool <mbp at samba.org> schrieb am 12.09.02 14:04:03:
> On 12 Sep 2002, Joerg Beyer <j.beyer at web.de> wrote:
> 
> > is that enough? It at least leave a small window between open() and
> > the ioctl() for a race condition. I dont what to be academic but...
> 
> I think the lock_fd will suffice to serialize access to the file.
> Since they don't do anything between opening the file and locking it
> I'm not sure the window matters.  Can you step through the race you
> think could cause a problem?

ok, now I see: if one process has a stats file locked and another
process will lock the same stats file, than the second process
will silently fail, because the fcntl(fd, F_SETLKW, &fl) in
util.c:lock_fd() will fail. Is this a problem? I am not sure.
> 
> > Martin, in your private mail from 8/31/2002 you indicated that 
> > using ccache by multiple users on the same machine at the same
> > time with unhappy timing will lead to problems.  Do you think it is
> > okay to use ccache as describe by a group of developers?
> 
> I think what I said was that I didn't know if it was a supported
> configuration.  File permissions may or may not cause a problem.  I
> haven't really looked in detail.

if the developers share a unix group (which seems resonable if they
work on the same project), then you at least could build something:
the don disk cache could use this group and ensure group read/write 
permissions - I have not checked this.
ccache and/or distcc could have the group sticky bit set.
> 
> If ccache is called by distccd as I suggest, then everything will
> happen in the same uid and it should be no different to a single user

no - without chmod g+sx ccache and/or distcc each
distcc and ccache process wil run with the uid of the calling
user.

> running make -j, which works.


    Joerg