Using NFS in a TruCluster environment - limitations

[Vijayasekar Rajsekar]

I am trying to set up an automount of a partition in TruCluster and get the following error message "connection refused". The TruCluster manual says that when a cluster acts as an NFS server, clients must use the default cluster alias to specify the host when mounting file systems served by the cluster. If a client attempts to mount a file system from the cluster the above error message is returned.

In my case it is something like this:
A. /xxx/cluster --- can be mounted
B. /xxx/cluster1 --- cannot be mounted
where cluster is the "alias" name whereas "cluster1" is the specific partition name.

Has anyone got a workaround with this limitation ? My software needs to access files on the cluster partition rather than the alias. Any comments will be appreciated.

[Johan Brusche · Accepted solution]

You do not specify the TruCluster version, but in the earlier versions you could only NFS-export filesystems towards the same subnet as the DEFAULT cluster alias. If a client tried to NFS-import from within another subnet it got the "connection refused" message.

With the current versions (since V5.1A) you can also NFS-export filesystems via secondary cluster aliases. To enable this you must enter the secondary aliases in the file /etc/exports.aliases. If you haven't done that yet, you also get the "refused" message.

See also manpage "export.aliases" and the comments in that file.

Johan.

[Mobeen_1]

Vijayasekar,
If i understand you correctly, you are saying that you have a cluster setup as follows

Cluster Alias - Cluster
Cluster Node - Cluster1
Cluster Node - Cluster2

From your post, it looks like you are able to access NFS using the alias Cluster? But seem to have issues using Cluster1?

In that case, have you tried configuring NFS on Cluster1 just like you would do, if it were a non-clustered system?

regards
Mobeen

[Johan Brusche]

NFS exports via a "membername" are NOT supported in a TruCluster, clients that connect to an NFS filesystem via a membername also get the "connection refused" message.
To work around this restriction you can define a secondary cluster alias which is joined by only one member. For further setup details refer to my previous response.

Johan.

[Vijayasekar Rajsekar]

Thanks Johan and I am running Cluster version 5.1B. You are correct and following are the steps done:

1. Get two new IP addresses on the same network as cluster interface ips.
2. Decide the hostnames and make entries in /etc/hosts file on cluster and DNS. Each hostname for these two new IPs is cluster alias which we will be using from NFS client to import the file system.
3. Make sure /etc/clu_aliases.config is CDSL link.
4. Make /etc/exports a CDSL link (by default it is shared between two nodes) to /cluster/members/{memb}/etc/exports.
5. Make an entry in /etc/exports file for exporting the required directory on each node. Here we are restricting the NFS access for only one filesystem.
6. Add hostnames (for new IPs) entries in /etc/exports.aliases file.
7. Login on first node (root) and issue following command:
#cluamgr -a alias=hostname1,join
8. Edit /etc/clu_alias.config (CDSL link) on each node and add an entry for assigned new hostname for new cluster alias. This entry should look like this.
/usr/sbin/cluamgr -a selw=1,selp=1,join,alias=
On second member it should be:
/usr/sbin/cluamgr -a selw=1,selp=1,join,alias=. Make sure no other entry is removed.
9. Issue the following commands on any node:
#cluamgr -r stop
#cluamgr -r start
10. Restart /sbin/rc3.d/S19nfs and S20nfsmount
11. You should be able to access two different filesystem from each node on this workstation.
12. On any cluster node issue following command to ensure cluster aliases are running.
#cluamgr -s all