[Ocfs-users] OCFS write/read across 2 nodes

Wim Coekaerts wim.coekaerts at oracle.com
Tue Jun 8 11:53:17 CDT 2004


eh this has ntohing to do with shared mmap

we should have actually killed off the pages after close of the cat, so
this seems to still be a problem,

this is a bug imo . we re supposed to invalidate buffers on close

On Tue, Jun 08, 2004 at 10:04:33AM -0700, Sunil Mushran wrote:
> Yes. ocfs does not supported shared mmap. So what's happening is that
> when you do the cat on node B in t1, the kernel does not read from the
> disk, just from the buffer cache. odirect reads will read the correct
> data.
> 
> On Tue, 2004-06-08 at 05:20, Dermot Reynolds wrote:
> > Hi,
> >  
> > If I have ocfs on 2 nodes should I be able to do the following:
> >  
> > t=0    
> >     Node A
> >         echo "1" > /ocfsmount/a
> >     
> >     Node B
> >         cat /ocfsmount/a
> >  
> > >>>>Question 1: At this point should I see "1"?
> >  
> > t=1    
> >     Node A
> >         echo "2" > /ocfsmount/a
> >     
> >     Node B
> >         cat /ocfsmount/a
> >  
> > >>>>Question 2: At this point should I see "2"?
> >  
> > The problem is that at t=1 I am still seeing "1", however when I run
> > ocfstool on node B I can see "2".
> >  
> > Assuming that this is correct behaviour this must imply that Oracle
> > writes and reads to/from the filesystem in a different manner.  
> > Is this where o_direct comes in?
> >  
> > Thanks
> > Dermot
> >  
> >  
> >  
> > 
> > ______________________________________________________________________
> > _______________________________________________
> > Ocfs-users mailing list
> > Ocfs-users at oss.oracle.com
> > http://oss.oracle.com/mailman/listinfo/ocfs-users
> 
> _______________________________________________
> Ocfs-users mailing list
> Ocfs-users at oss.oracle.com
> http://oss.oracle.com/mailman/listinfo/ocfs-users



More information about the Ocfs-users mailing list