Dear all,
for those wishing to install and play with DPM, make sure that
`hostname` gives you your hostname _without_ the domain name.
Otherwise rfio transfers fail saying ``Connection refused''. Killed at
least 2 hours figuring that out and I had to go to DPM sources to
get srmcp work. The piece of code which checks whether a machine is
a localhost (rfio_parseln) needs some looking into.
Graeme came up with a really cool installation guide, it's worth
checking it out. I really recommend the RFIO check he is suggesting.
It helped me a lot to get DPM working and it really makes no sense
to continue without making sure the rfdir works for you. For me the
crucial thing was to have
rfdir dev03:/tmp
(_without_ the [gridpp.rl.ac.uk] domain name) listing the contents
of the machine where I've installed DPM (dev03.gridpp.rl.ac.uk).
Otherwise I'm quite pleased with the current setup. Installation
worked first time for me, I only needed to tweak local machine setup
(not DPM specific). Not only I'm pleased with the current setup,
but it also seems to be much faster than dCache when doing srmcp
into/from it. Time to move of dCache? Well, time will tell I guess.
More tests to follow, probably not me though.
Bad news
~~~~~~~~
The bad news that the dpm.ftpd zombie always re-appears after an
srmcp or globus-url-copy.
Good news
~~~~~~~~~
- this doesn't seem to affect the transfers in any way, they simply work
- zombies don't seem to be accumulating :) There's always only one
zombie regardless of the number of copies you do. The zombie disappears
when
# /etc/init.d/dpm-gsiftp restart
is run.
Hope this additional DPM experience helps a bit.
Regards.
--
Jiri
Words written by `Graeme Stewart' on 01 Aug 2005 at 11:33:31 +0100 prompted:
> On Monday 01 Aug 2005 11:23, Jiri Mencak wrote:
> > Hi Greame,
> >
> > where do you report problems with DPM? Quite impressed so far, but
> > getting a zombie dpm.ftpd process when doing globus-url-copy...
> >
> > Thanks.
>
> Try: [log in to unmask]
>
> This is supposed to be archived, but I don't know where. Or put a bug in
> savannah - the developers (i.e. J-P B) are (is?) very responsive.
>
> Check that the pool filesystem is owned by dpmmgr (though that usually causes
> ftpd to fail, not hang). Attaching an strace to the daemon (and children) can
> often be useful.
>
> cheers
>
> g
>
> --
> --------------------------------------------------------------------
> Dr Graeme Stewart http://www.astro.gla.ac.uk/users/graeme/
> GridPP DM Wiki http://www.physics.gla.ac.uk/gridpp/datamanagement/
|