Subject: Re: pkgsrc fully over NFS
To: Michal Pasternak <email@example.com>
From: Juan RP <firstname.lastname@example.org>
Date: 01/16/2004 18:42:55
On Fri, 16 Jan 2004 17:32:14 +0100
Michal Pasternak <email@example.com> wrote:
> Let's look at situation like building mozilla with pkgsrc when everything
> (/usr/pkgsrc, /usr/pkgsrc/distfiles, WRKOBJDIR) is mounted via NFS:
> - checksum:
> - fetch whole mozilla source tarball via NFS
> - calculate checksum
> - extract:
> - fetch whole mozilla source tarball (again) via NFS
> - extract every piece of data back via NFS (size of tarball
> multipled by at least one point something)
> This can take some amounts of time. While doing one package, time doesn't
> matters as much, as when bulk building.
> Do you have any ideas how could I get around this?
> I thought about extracting packages on the NFS server first. Why not, this
> could possibly work (except I won't need package dependencies to be
> installed on NFS server, which is normal behaviour of "make extract") . The
> next problem is, anyway, that if I want to build single pkgsrc tree on some
> more hosts, than one, all of them will create "work" directory in pkgsrc
> tree, which will cause other builds to be confused by other (even with
> WRKOBJDIR symlinks are created). What solution -- except mount_union or
> obvious, cp -R pkgsrc pkgsrc2 do we have for such problem?
Why don't use WRKOBJDIR pointing to a local directory in the clients?
Juan RP <juan AT xtraeme DOT unixbsd DOT org>