NFS server on cluster package.

[VKV]

Hi HP-UX gurus,

I have a MCSG 11.20 cluster with eight nodes, which runs a number of application/database packages.

There is a NFS package which exports NFS shares to almost all the packages and those NFS shares are critical for the applications to work.

The problem I face is,  if the NFS package failover to its secondary node, the mount points on which the NFS shares are mounted (on the NFS clients) become inaccessible. Although they won’t go stale, the mount points won’t respond to ls or bdf commands (They remain accessible on the NFS server). Once I bring the package back to the primary node the shares are accessible again.

NFS server was configured in an HA package believing that the shares will be immediately available to the clients again once the failover is complete. That is not happening.

Is it because NFS is aware about the MAC ID the NICs?

Could you please share your thoughts?

Thanks,

Vimal

--------------------------------------------------------------------------------------------------------------------------

PS.This thread has been moved from HP-UX> System Administration to Networking - Forum Moderator

[Emil Velez_2]

lokmat how you are mounting the filesystems. hard mounts, soft mounts or try the automounter. The automounter mounts the fs when u need it.

Might need to tweek some mount options

[Dave Olker]

Some questions:

• What OS is running on the NFS clients?
• What NFS mount options are they using (nfsstat -m)?
• Do all clients behave the same or does a subset of clients work and others fail?
• I assume you are configuring a floating IP address for the NFS package and that the NFS clients are mounting using the floating IP address?

Regards,

Dave

[VKV]

Hi Emil,

They are all hard mounts. The application always access these file systems that’s why automount was not considered while building the environment.

Hi Dave,

• What OS is running on the NFS clients?

           HP-UX B.11.31 (March 2012 patch) and ONCplus B.11.31.16

• What NFS mount options are they using (nfsstat -m)?

        Flags:         vers=4,proto=tcp,sec=sys,hard,intr,link,symlink,acl,devs,rsize=32768,wsize=32768,retrans=5,timeo=600

        Attr cache:    acregmin=3,acregmax=60,acdirmin=30,acdirmax=60

• Do all clients behave the same or does a subset of clients work and others fail?

                             All clients are in same subnet and all of them fail.

• I assume you are configuring a floating IP address for the NFS package and that the NFS clients are mounting using the floating IP address?

                           Yes. The clients are using the floating IP address (There is a DNS host entry  for that IP address and the clients use that).

Thanks

Vimal

[Dave Olker]

I've configured Serviceguard NFS packages many times and never had a problem with clients seeing a package migration between nodes. However, I've never used NFS v4 in this configuration. You could try mounting with NFS v3 (-o vers=3) and see if the same problem occurs.

Dave

[VKV]

Hi Dave,

Thanks for that.

However this is a production environment and it will take a huge effort (change request/ approvals) to experiment using version 3 mounts. Also I don’t have  similar sandbox environment where I can test version 3 mounts.

I will certainly try this if get an opportunity.

Thanks,

Vimal

[UshaLatha]

Hi Vimal,

Once the failover happens HANFS will unshare all the directories shared by hanfs on primary node
and reshare them on adoptive node , can you check if all the nfs shares shared via HANFS package on the
primary node are unshared properly and the volumes are unmounted properly by executing share command.

Can you please post if there are any errors in the log messages in /var/adm/cmcluster/log/<package-name>.log

And also once failover happens check using cmviewcl and see if hanfs package is started and running on adoptive node and the float ip is moved to adaoptive node and is pingable from clients.

Regards,

Usha Latha