[Samba] Re: locking.tdb: expand_file ftruncate to 8192 failed

--===============0138838918==
Content-Type: multipart/signed; micalg=pgp-sha1;
protocol="application/pgp-signature"; boundary="VS++wcV0S1rZb1Fb"
Content-Disposition: inline


--VS++wcV0S1rZb1Fb
Content-Type: text/plain; charset=us-ascii
Content-Disposition: inline
Content-Transfer-Encoding: quoted-printable

On Thu, Sep 22, 2005 at 08:19:18AM -0700, Jeremy Allison wrote:
> On Thu, Sep 22, 2005 at 10:15:08AM +0200, Axel Thimm wrote:
> > On Wed, Sep 21, 2005 at 04:34:32PM -0700, Jeremy Allison wrote:
> > > Why are you putting the locking db on a GFS filesystem anyway. That's
> > > madness !
> >=20
> > The reason is to have a poor-man's-clustered-samba by placing lock and
> > private dir on a common share and have the relocated smbd/nmbd pairs
> > access them. E.g. relocating within the cluster is effectively like
> > restarting smbd/nmbd on a node.
>=20
> That's never going to work (at least with acceptable speed). Talk
> to Volker for details...

We found the speed is very acceptable, its is faster than NFS
relocation and counts a couple of seconds, no more than a simple
non-clustered samba restart.

But it also isn't as transparent as we would wish it to be, as a TCP
RST doesn't have the cifs client retry as NFS does, which fortunately
is sometimes masked away in higher application levels like Office
saving dialogs, but still breaks a simple copy operation.

Anyway it does serve its purpose quite well, if it were not for the
mentioned bug in GFS. We now have several instances of smdb/nmbd pairs
freely floating within the cluster.

> > > As I said, I bet GFS isn't POSIX complient. Don't put locking
> > > tdb's on anything but local filesystems.
> >=20
> > Well, GFS claims to be POSIX and local-like in any way. Maybe it is
> > just a bug in GFS? Does POSIX ensure that you can open an fd under
> > some user and not lose access right to the fd when dropping
> > priviledges?
>=20
> Yes. That's why we wrote it this way. It's a bug in GFS. Open it
> with RedHat.

Already done so, the bugzilla link was in my previous mail :)

Thanks for the test case, it helped cornering the GFS bug.
--=20
Axel.Thimm at ATrpms.net

--VS++wcV0S1rZb1Fb
Content-Type: application/pgp-signature
Content-Disposition: inline

-----BEGIN PGP SIGNATURE-----
Version: GnuPG v1.4.1 (GNU/Linux)

iD8DBQFDMs+SQBVS1GOamfERAo2RAJ9lQ/wagBaRvtau++aRqVSROnp+SACdGN7U
qetkGE34jSdbbNn5RaJuIDQ=
=vetu
-----END PGP SIGNATURE-----

--VS++wcV0S1rZb1Fb--

--===============0138838918==
Content-Type: text/plain; charset="us-ascii"
MIME-Version: 1.0
Content-Transfer-Encoding: 7bit
Content-Disposition: inline

--
To unsubscribe from this list go to the following URL and read the
instructions: https://lists.samba.org/mailman/listinfo/samba
--===============0138838918==--