[Gluster-devel] 1.4.0qa19 crash while untar (unify + afr-ns)

David Mayr - Hetzner Online AG david.mayr at hetzner.de
Wed Jun 18 13:11:57 UTC 2008


Hi Raghavendra,

> Its been a known issue and a fix is on its way to glusterfs--mainline--3.0.
> Meanwhile, you can use an aggregate-size of 64KB which will fix the issue
> in your case. Also window-size you've configured is too big. Please try
> using 512KB.


Thanks, that works now - I changed to the suggested values on all machines.
Now it seems to be much slower (even ls on fresh mounted glusterfs).

I'll wait for the next qa-release and try it again then.


> david.mayr at hetzner.de> wrote:
> > Forgot to add some information about the environment:
> >
> > all machines have debian etch 4.0 64bit installed, are running the
> > default kernel 2.6.18-6-amd64, use fuse-2.7.3glfs10 and are connected
> > through Gbit ethernet.
> >
> > Am Mittwoch 18 Juni 2008 12:38:47 schrieb David Mayr - Hetzner Online AG:
> > > Hi all,
> > >
> > > I'm just testing 1.4.0qa19 and hit a crash while untarring a file to a
> > > glusterfs mounted directory:
> > >
> > > $ glusterfsd -l glfs.log -L DEBUG -f
> > > /etc/glusterfs/glusterfs-client.vol /mnt/ $ mkdir /mnt/test
> > > $ tar xf Debian-40-etch-32-minimal.tar -C /mnt/test/
> > >  tar: bin/bash: Cannot close: Software caused connection abort
> > >  tar: bin/bash: Cannot utime: Transport endpoint is not connected
> > >  tar: bin/bash: Cannot change ownership to uid 0, gid 0: Transport
> >
> > endpoint
> >
> > > is not connected
> > >  tar: bin/znew: Cannot open: Transport endpoint is not connected
> > >  tar: Skipping to next header
> > > ...
> > >
> > > The tar file contains a complete root filesystem of a debian
> >
> > installation.
> >
> > > Find the client debug log here (inlcuding client config):
> > >   http://davey.de/glusterfsd.log
> > >
> > > Server config is the same on all servers and namespaces:
> > >   http://davey.de/glusterfs-server.vol
> > >
> > > All Servers log this at the time of error:
> > >   2008-06-18 12:16:43 E [tcp.c:89:__tcp_rwv] server: EOF from peer
> > >   2008-06-18 12:16:43 E [tcp.c:624:tcp_proto_state_machine] server:
> >
> > socket
> >
> > > read failed (Transport endpoint is not connected) in state 1
> > >   2008-06-18 12:16:43 E [tcp.c:1328:fini] server: transport 0x5924e0
> > > destroyed
> > >
> > > The servers filesystem is reiserfs with options "noatime,notail"
> > >
> > > Please tell me if I can give you more information to help you ...
> >
> > --
> >  David Mayr

-- 
  David Mayr





More information about the Gluster-devel mailing list