Re: [SSI-devel] Re: SSI-Lustre-1.2.4 initnode ssi_deadlock_check: cfs_kswapd hung?
Brought to you by:
brucewalker,
rogertsang
From: Roger T. <rog...@gm...> - 2005-04-05 01:48:35
|
Well I see that you already included the patch. To answer your question, yes it has been working for me since I posted it. I think the deadlock could be because I started MDS/OST and client on the same node which somehow caused deadlock on all nodes. It's been a while, so I don't remember the details. I won't be running that config again. I am using it as an MDS/OST with this version and testing it by running the client on another SSI node. So far it seems fine if you specify --fstype ext3 rather than leaving it at default. I've tested running MDS/OST on loopback device, DRBD, and raw disk partition. The performance on DRBD is close to Lustre using raw disk - which seems to be using journal_data. The fastest write performance seems to be loopback on top of non-chard ext3 fs. Also I had to setup at least two OST's per LOV to get df to show properly, but this could be I didn't reformat the MDS? I was using stripe_cnt 0 and stripe_pattern 0 which seems to be only option and must be specified during setup. -Roger On Apr 4, 2005 2:17 PM, Jaideep Dharap <Jai...@hp...> wrote: > Roger, > Did the lines below work? Let me know. I will merge them into the > patch and put the > new patch up on the website. > Btw, to answer a previous question, we have always tested Lustre-SSI > only as a client > so far. This goes for previous versions of lustre too. I think I might > have tried briefly > using it as an OST/MDS on previous versions. But definitely not on this > one. If you are > intending to use it as an OST or MDS, let me know if you run into > problems and how it goes. > Jai. > > Roger Tsang wrote: > > >I only found a few lines... > > > >--- lustre-1.2.4-ssi.patch 2005-03-29 20:55:37.000000000 -0500 > >+++ lustre-ssi.patch 2005-04-01 17:35:29.000000000 -0500 > >@@ -4673,6 +4673,11 @@ > > + list_del(&goal->d_child); > > + list_del(&dentry->d_child); > > + > >++ /* XXX I think the lustre patch need needs this, not just SSI */ > >++ if (goal->d_parent == goal) > >++ /* We're going to make de a child of goal */ > >++ dget(goal); > >++ > > + /* Switch the parents and the names.. */ > > + switch_names(goal, dentry); > > + do_switch(goal->d_parent, dentry->d_parent); > > > > > >On Apr 1, 2005 4:49 PM, Roger Tsang <rog...@gm...> wrote: > > > > > >>The bug is really annoying, so I'm looking for a merge soon. Which > >>earlier version? I'm currently trying to resolve conflicts between > >>the merge of the difference between 1.1.2.29 and 1.1.2.30 with the > >>lustre-1.2.4-ssi.patch. Might as well merge in this other "earlier" > >>version too. I wonder are you doing the merge as we speak? I guess > >>we don't need to be both doing the same thing, and I'll let you > >>finish. > >> > >>-Roger > >> > >>On Apr 1, 2005 2:31 PM, John Byrne <joh...@hp...> wrote: > >> > >> > >>>Roger Tsang wrote: > >>> > >>> > >>>>This looks like the same old cfs/nfs bug that John Byrne fixed earlier > >>>>back in SSI-1.1.1-FC. It's resurfaced again. All my nfsd processes > >>>>are in DW state. > >>>> > >>>> > >>>Jai missed merging the fix I checked into version 1.1.2.30 of the old > >>>lustre-ssi patch. He also missed an earlier change, but that might not > >>>be signifying. (It depends on whether asserts are enabled.) I'll talk to > >>>him about fixing it Monday or Tuesday. > >>> > >>>John > >>> > >>><...snipped...> > >>> > >>> > >>> > > > > > >------------------------------------------------------- > >SF email is sponsored by - The IT Product Guide > >Read honest & candid reviews on hundreds of IT Products from real users. > >Discover which products truly live up to the hype. Start reading now. > >http://ads.osdn.com/?ad_id=6595&alloc_id=14396&op=click > >_______________________________________________ > >ssic-linux-devel mailing list > >ssi...@li... > >https://lists.sourceforge.net/lists/listinfo/ssic-linux-devel > > > > > > > > |