Hi
I was using HACMP 4.5 with cascading resource groups for two years with
this configuration:

* two node cluster (dbnode,clientnode)
* two resource groups (APP,CLIENT)
DB resource group is bound to dbnode_serv IP address.
CLIENT resource group is bound to clientnode_serv IP address.
IPAT is done by IP replacement.

dbnode has all filesystem mounted (from sharedvg) and 3 of them are NFS
exported:
- /data
- /data/homes
- /LOGS

clientnode has these filesystems NFS mounted.
as:
dbnode:/data
dbnode:/data/homes
dbnode:/LOGS

So in HACMP attributes of DB resource group option in smit I have:

Filesystems/Directories to Export
/data /data/homes /LOGS

Filesystems/Directories to NFS Mount
/data /data/homes /LOGS


HACMP on dbnode is started. Filesystem mounted locally and exported.
dbnode_serv is set. The same on clientnode - filesystems NFS mounted,
clientnode_serv set. Great. Now we have a situation where dbnode dies.
clientnode forces unmounting NFS mounts, acquires sharedvg, mounts
filesystems locally, exports them, acuires dbnode_serv and
clientnode_serv Ip addresses. Works. That was a behaviour in 4.5

In HACMP 5.2 looks like I can not use the same config.
In 5.2 I couldn't set option for the resource group like before:
Filesystems/Directories to Export
/data /data/homes /LOGS

Filesystems/Directories to NFS Mount
/data /data/homes /LOGS

So I changed it to:

Filesystems/Directories to Export
/data /data/homes /LOGS

Filesystems/Directories to NFS Mount
/data;/data /data/homes;/data/homes /LOGS:/LOGS

When dbnode dies, clientnode doesn't unmount NFS mounts. It just aquires
sharedvg and mounts filesystems on previously mounted NFS mounts. So df
shows:
dbnode:/data
dbnode:/data/homes
dbnode:/LOGS
/dev/datalv /data
/dev/homeslv /data/homes
/dev/logslv /LOGS

Now I have a mess and additionally when dbnode is up again there is a
problem of unmounting resources properly.

Is there a way to have the same HACMP(NFS) behaviour in 5.2 as it was in
4.5?


Sincerely,

Jan Pomni <pocztex_noszpam@cogito.univ.gda.pl> wrote in
news:hdjqg2-8hd.ln1@airbus.lhsystems.pl:
>
> In HACMP 5.2 looks like I can not use the same config.
> In 5.2 I couldn't set option for the resource group like before:
> Filesystems/Directories to Export
> /data /data/homes /LOGS
>
> Filesystems/Directories to NFS Mount
> /data /data/homes /LOGS
>
> So I changed it to:
>
> Filesystems/Directories to Export
> /data /data/homes /LOGS
>
> Filesystems/Directories to NFS Mount
> /data;/data /data/homes;/data/homes /LOGS:/LOGS
>
> When dbnode dies, clientnode doesn't unmount NFS mounts. It just
> aquires sharedvg and mounts filesystems on previously mounted NFS
> mounts. So df shows:
> dbnode:/data
> dbnode:/data/homes
> dbnode:/LOGS
> /dev/datalv /data
> /dev/homeslv /data/homes
> /dev/logslv /LOGS
>
> Now I have a mess and additionally when dbnode is up again there is a
> problem of unmounting resources properly.

NFS filesystems mounted directly on the root dir (/) tend to be
unmountable during a takeover, since the server is no longer available.
This also causes a LOT of commands and scripts to hang.

A better way is to do the following:
- Unmount all the filesystems
- change the mount point of the filesystems to be under /export :
# chfs -m /exports/data /data
# chfs -m /exports/homes /data/homes
# chfs -m /exports/LOGS /LOGS
Note: either do this with c-spoc OR don't forget to export/import the VG
on the other node (don't forget to keep the same major number or NFS
will fail). - Change the Filesystems/Directories to Export to the new
mountpoints
>> /exports/data /exports/homes /exports/LOGS
Change the Filesystems/Directories to NFS Mount to the new mountpoints
and keep them mounted off a subdir of /:
>> /exports/data;/mounts/data /exports/homes;/mounts/homes
>> /exports/LOGS:/mounts/LOGS
- Mount the filesystems on /exports/*
- Inside /exports/data symlink homes
# rmdir /exports/data/homes
# ln -s /exports/homes /exports/data/homes

- On both nodes: Create the NFS mounts locations, remove the LOGS and
data from / on both machines and replace with a symlink to the NFS
mounts locations: # mkdir /mounts /mounts/data /mounts/LOGS
# rmdir /data /LOGS
# ln -s /mounts/data /mounts/LOGS /

- Syncronize the HACMP change and you'll probably see a warning about a
detail I probably forgot. :-) - Ready

Advantages:
1- Since there are no longer NFS mounts on the root dir, things tend to
go smoothly and faster. 2- The filesystem and NFS mountpoint are no
longer on the same place, so the problem you got won't happen again. 3-
All nodes use the NFS mountpoint instead of the filesystem directly.
This means that applications don't have to be killed to release the
usage mountpoint. 4- Almost forgot: the resource group must mount the
filesystems before aquiring the IP address or there will be a small
window there clinte applications may fail.


--
Doing AIX support was the most monty-pythonesque
activity available at the time.
Eagerly awaiting my thin chocolat mint.