HA NFS

[Sentinel_2]

Hi All,

could someone explain what are the reasons to use NFS cross-mounting in a cluster?
What are the advantages?

The guide says "every server has an identical
view of the file systems" but I don´t see so big advantage as to configure this.

Thanks

[Hakki Aydin Ucar]

if servers has to use common resource like active data with each other, it is useful method. Suppose you have 3 servers one of them GUI server and all GUI event on it, and needs collect data from other servers to display frequently..

[Ismail Azad]

Hi,

You will find one momentous advantage in the same manual which states that:-

"If a server is configured to use NFS over TCP and the client is the same machine as the server,
which results in a loopback NFS mount, the client may hang for about 5 minutes if the
package is moved to another node. The solution is to use NFS over UDP between
NFS-HA-server cross mounts."

Well that's one scenario. But I would like to add that....

With the use of the CFS feature which is supported on the HPUX operating system which has the wonderful ability of concurrent reads/writes, HP recommends that you do not use the "cross mounting" feature between two servers and use CFS instead. So in terms of what is reccommended, I would vouch for your last statement. Don't forget to assign points.

Regards
Ismail Azad

[Joel Shank]

We use HA NFS for our /home directory. We have a 7 node SG cluster and users can be logging onto any one of them depending on where their package is running. HA NFS allows them to see the same /home directory regardless which server they are on.

[Kenan Erdey]

i.e. in a serviceguard Oracle RAC environment, it's user for getting oracle archive logs from all nodes when restoring from one node.

[Sentinel_2]

But, i don´t understand why the node has the fs mounted by nfs to itself.
It already has the fs mounted locally by the cluster package.

[Kenan Erdey]

there is no need to nfs mount to itself. can it be other mounts coming from other nodes ?

[Ismail Azad]

Hi,

Here we talk about the package type as "failover". Now "cross mount" being the topic as you said...

> But, i dont understand why the node has the fs mounted by nfs to itself.

Here we talk about HANFS.... Yes, in a normal NFS environment, this would sound really strange. However, let's say the package fails over to the node it was supposed to failover with the appropriate policy in apt terminology the "adoptive node". If your primary node was node1 and adoptive was node2, then node2 cannot access the NFS filesystem in a "package" through the "local mount point" {the mount point on node1}.... let's say when it fails over.

So the whole question pops up, can the adoptive node access a serviceguard filesystem from a the "local mount point" of a package from the node it has failed over?! There are answers to this question and one of them is having "NFS cross mounts accross two servers"

This is one momentous reason why only in an HANFS environment we make use of the script nfs_xmnt and that's why part of your question states...

> the node has the fs mounted by nfs to itself.

Regards
Ismail Azad

[Andre Cornelissen · Accepted solution]

I am not sure whether the last reply want to explain the same as I am saying ;-)

We use this on SAP systems where "/usr/saptrans" is NFS mounted on all systems (the primary running Prod, the adoptive node running Q/A and standalone systems running Development)
The Production node mounts /export/saptrans and uses the nfsmount on /usr/saptrans.
in that case it can mount /export/saptrans on the adoptive node when it moves there.
If it mounted it on /usr/saptrans directly and used it there, there would be an issue during failover because the mountpoint is already in use. (the NFS mount is outside the cluster)

(hard to explain without a whiteboard ;-)
Regards,
Andre

[Viktor Balogh]

let's think of several nfs clients, who use an nfs filesystem mounted from a server. If that nfs server server fails, all the nodes get stale nfs. To avoid this situation, you can employ a HA-nfs solution which means a HA-nfs _server_. In fact, HP-s HA-nfs solution can switch the file handles with the nfs server package, so the clients only notice a switch as a short disruption and ip package re-send but not an IO error.

[Ismail Azad]

Hi,

There are many setups you can use but implementing a cross mount with two NFS servers probably defeats the purpose of the HANFS design of "cross mount" in your environment. One advantage lies based on my previous post and coming to cross mounts in your environment{although what you see in quotes is an advantage}, you can say

> The guide says "every server has an identical
view of the file systems" but I don´t see so big advantage as to configure this.

Because your "local mount point" at failover is duplicated elsewhere.

Regards
Ismail Azad

[Sentinel_2]

Reading again the manual (Serviceguard NFS Toolkit Administrator's Guide) and with the help of your answers I finally got it.

Thanks for your help

[Sentinel_2]

See above