HACMP/ES 4.4.1 NFS problem

HACMP/ES 4.4.1 NFS problem

Post by Wouter » Thu, 18 Oct 2001 20:59:41



Hello unix friends !

Who can help us with an NFS problem in a HACMP environment ?

We have 2 p660 in a HACMP mutual takeover situation.

Machine A does an export of a filesystem. (service addr=  .70 at the end,
boot addr=.71)

Machine B has to mount the filesystem from machine A. (service addr=  .80 at
the end, boot addr=.81)

If we do a takeover from machine B to machine A, everything is OK.
When we restart the HACMP software on machine B, it seems to hang for 10
minutes
on the NFS mount (manually mounting the NFS FS works immediately). Every
time we do a df-k
we can see the send-queue increases (netstat) on the connection from .81 to
.70.  It seems like then NFS
subsystem is still working on the boot-address although HACMP already
changed the ip to the service address.

Does this sound familiar to anybody ?  Is there any place where I can look
for problems ?  (we can't find anything about
the NFS mount in hacmp.out)

Thank you in advance !

Wouter V.
Unix System Engineer.

 
 
 

HACMP/ES 4.4.1 NFS problem

Post by Clive Sparke » Thu, 18 Oct 2001 22:42:18



> Hello unix friends !

> Who can help us with an NFS problem in a HACMP environment ?

> We have 2 p660 in a HACMP mutual takeover situation.

> Machine A does an export of a filesystem. (service addr=  .70 at the
> end,
> boot addr=.71)

> Machine B has to mount the filesystem from machine A. (service addr=
> .80 at
> the end, boot addr=.81)

> If we do a takeover from machine B to machine A, everything is OK.
> When we restart the HACMP software on machine B, it seems to hang for 10
> minutes
> on the NFS mount (manually mounting the NFS FS works immediately). Every
> time we do a df-k
> we can see the send-queue increases (netstat) on the connection from .81
> to
> .70.  It seems like then NFS
> subsystem is still working on the boot-address although HACMP already
> changed the ip to the service address.

> Does this sound familiar to anybody ?  Is there any place where I can
> look
> for problems ?  (we can't find anything about
> the NFS mount in hacmp.out)

> Thank you in advance !

> Wouter V.
> Unix System Engineer.

Guess 1, the ping_client_list isn't correct/complete and the 10 minutes
is caused by waiting for the mtu route discovery to kick in and update
the arp cache.

--
Regards,
Clive

 
 
 

HACMP/ES 4.4.1 NFS problem

Post by Simon Marches » Fri, 19 Oct 2001 05:25:27




> > We have 2 p660 in a HACMP mutual takeover situation.

> > Machine A does an export of a filesystem. (service addr=  .70 at the
> > end,
> > boot addr=.71)

> > Machine B has to mount the filesystem from machine A. (service addr=
> > .80 at
> > the end, boot addr=.81)

> > If we do a takeover from machine B to machine A, everything is OK.
> > When we restart the HACMP software on machine B, it seems to hang for 10
> > minutes
> > on the NFS mount (manually mounting the NFS FS works immediately). Every
> > time we do a df-k
> > we can see the send-queue increases (netstat) on the connection from .81
> > to
> > .70.  It seems like then NFS
> > subsystem is still working on the boot-address although HACMP already
> > changed the ip to the service address.

> > Does this sound familiar to anybody ?  Is there any place where I can
> > look
> > for problems ?  (we can't find anything about
> > the NFS mount in hacmp.out)

> > Thank you in advance !

> > Wouter V.
> > Unix System Engineer.

> Guess 1, the ping_client_list isn't correct/complete and the 10 minutes
> is caused by waiting for the mtu route discovery to kick in and update
> the arp cache.

Beware - mounting an NFS-exported filesystem from another cluster node can be
a major pain because if any process on B is using any files/directories in
your filesystem when machine A fails, B cannot unmount the filesystem unless
those processes are killed. Unless you use the "new method" of exports, which
requires a filesystem which is NFS mounted within the cluster to be NFS
mounted on *all* nodes, even the one local to the mounted JFS filesystem.
 
 
 

1. Weird behaviour after upgrading from HACMP/ES 4.4.0 to HACMP/ES 4.4.1

All,

I recently upgraded a 2 nodes cluster (cascading) from HACMP/ES 4.4.0
to HACMP/ES 4.4.1, and the cluster behaves weirdly! After upgrading
the primary node, and after a reboot (as per documentation), I restart
cluster manager. The node is joigning the cluster, then the resource
group travel gently from the backup node to the fresh upgraded primary
node. Everything is fine, filesystems are mounted, application server
is started, then when you think you are done, a swap adapter event
occurred on the node, and the standby adapter becomes the service
adapter. After all the software and hardware checking, I couldn't find
any trouble on the adapter...

Did you notice such behaviour?

Thierry

2. ** Wabi **

3. Difference between HACMP and HACMP/ES

4. Printer

5. HACMP/es + nfs + ops

6. DIP error

7. HACMP/ES 4.4 clinfoES problem

8. unresolved symbols

9. Where to get or buy ES-255 and ES-220 or exchanging with ES-310 ?

10. Etherchannel and HACMP ES

11. HACMP/ES support for Gigabit adapters

12. HACMP/ES or HAGEO

13. Oracle Parellel Server on HACMP/ES (Clarion, Fastt500)