Hi,

I remember that this has been tought in every cluster-training I
attended - AIX-HACMP, Sun-Cluster, Veritas-Cluster. I also remember an
elderly white paper from sun where nfs was explained in full detail.

The basic - from what I remember - is the NFS-protocol and the resulting
client-behaviour.

I think the problem is the IP/RPC-protocol:
Once you have a NFS-connection the client retries its connection with
rpc-retries.

If the server comes up with IP first there is no rpc-service that could
answer the rpc-request, so the reconnect fails on IP-level, thus
possibly terminating the client-nfs-session.

I did a search but could not find a complete state transition diagram
for nfs.

But anyway - if the server-ip comes up last, the rpc-retry will be
served and the nfs-reconnect will be successful (if the
nfs-client-handle: IP, device major number and inode-number of the file
has not changed during failover).

Kind regards, Nils


> -----Original Message-----
> From: [EMAIL PROTECTED] 
> [mailto:[EMAIL PROTECTED] On Behalf Of Ryan Thomson
> Sent: Tuesday, May 20, 2008 6:30 PM
> To: General Linux-HA mailing list
> Subject: Re: [Linux-HA] NFS Configuration
> 
> Hi,
> 
> What is the technical explanation for starting the IP 
> resource LAST? I've experienced both successful and 
> unsuccessful NFS failovers and I haven't yet had a chance to 
> look much further than the well documented tweaks and 
> settings (if this is well documented, I must have missed it). 
> My IP resource does start FIRST and perhaps this may start to 
> explain the bad failovers I've seen.
> 
> Any additional information regarding this would be very helpful.
> 
> Thanks,
> 
> --
> Ryan Thomson
> 
> Hildebrand, Nils, 232 wrote:
> > Hi,
> > 
> > just a short remark:
> > With NFS the IP-Resource should be the LAST resource to 
> start - else 
> > your NFS-Clients will run into trouble in case of failover.
> > 
> > Kind regards,
> > 
> > Nils
> >  
> > 
> >> -----Original Message-----
> >> From: [EMAIL PROTECTED]
> >> [mailto:[EMAIL PROTECTED] On Behalf Of Beren 
> >> Gamble
> >> Sent: Wednesday, May 07, 2008 4:53 PM
> >> To: [EMAIL PROTECTED]; [email protected]
> >> Subject: Re: [Linux-HA] NFS Configuration
> >>
> >>  
> >>> I guess then that it depends on the service. AFAIK, most
> >> services read
> >>> configuration files only at startup. On the other hand, most also 
> >>> cause things like flushing buffers at exit, so that may be
> >> the reason
> >>> for failed stops. Of course, the heartbeat configuration
> >> itself has to
> >>> be on a local node. Which services do you run?
> >> Thanks for the reply...
> >>
> >> This is our haresources file
> >>
> >> web1n1 10.1.66.44
> >> Filesystem::hcfiler2:/vol/vol_Web::/Web::nfs::soft mysql-web 
> >> apache::/etc/apache2-web/apache2-web.conf mon tomcat5-web
> >> web1n2 10.1.66.45
> >> Filesystem::hcfiler2:/vol/vol_Intranet::/Intranet::nfs::soft
> >> mysql-intranet
> >> apache::/etc/apache2-intranet/apache2-intranet.conf mon 
> >> tomcat5-intranet
> >>
> >>
> >> Cheers!
> >>
> >> Beren
> >>
> >> [...]
> > _______________________________________________
> > Linux-HA mailing list
> > [email protected]
> > http://lists.linux-ha.org/mailman/listinfo/linux-ha
> > See also: http://linux-ha.org/ReportingProblems
> _______________________________________________
> Linux-HA mailing list
> [email protected]
> http://lists.linux-ha.org/mailman/listinfo/linux-ha
> See also: http://linux-ha.org/ReportingProblems
> 
> 
_______________________________________________
Linux-HA mailing list
[email protected]
http://lists.linux-ha.org/mailman/listinfo/linux-ha
See also: http://linux-ha.org/ReportingProblems

Reply via email to