From Earlham CS Department
Revision as of 13:04, 26 May 2021 by Craigje (talk | contribs)
Jump to navigation Jump to search

We use NFS to share files (e.g. software, user home directories) across computers in the CS and cluster domains. This page will be largely informed by CS-side information, but much of it will carry over into the cluster world as well.

On the CS side, the main NFS server is bowie. On the cluster side, the main NFS server is hopperprime. Each cluster head node also has NFS exports. (This is not a comprehensive list.)


For the common pattern of /etc/exports and /etc/fstab, see how we mount software directories on Lovelace. There's nothing particularly special about the case of the Lovelace machines, so the pattern should largely hold up across our servers: list the directories you wish to export from the local system in /etc/exports and the directories/disks you wish to mount in /etc/fstab.

User directory troubleshooting

When a user logs in to one of our computers, they should land in a home directory. These are stored one machine in each subdomain and exported to the other machines by NFS. In the CS case, the home directories are stored on bowie.

One possible problem with user logins on machines other than smiley (i.e. any machine most people will try to access) is that the directories haven't been mounted. This should be done automatically but was not, for some time. We believe we've solved it but have placed this guide here for reference. Consult it, for example, if...

  • you receive bizarre login messages
  • you don't see your user files on login
  • you expect to authenticate via ssh keys but you end up having to use your Unix password instead

(This guide does assume you can log in somehow, at least as sysadmin or at the console.)

To check if there's a problem with how we're handling NFS mounting of user directories, and if so to fix it, do the following:

  • run df -h to see what filesystems are mounted; as of August 2018, for example, you should see a line resembling 4.0T 1.2T 2.7T 31% /eccs. If so, your problem probably isn't here. If not, proceed.
  • run sudo mount -a and then df -h. If the problem persists, proceed.
  • look for the right line in /etc/fstab - something like /eccs nfs auto,nofail,noatime,nolock,intr,tcp,actimeo=1800 0 2.

Allowing NFS in a firewall

Few of our servers run a firewall, but the cluster head nodes do (as of this writing). You want to make sure to allow calls to mount NFS, thusly:

firewall-cmd --permanent --add-service=nfs
firewall-cmd --permanent --add-service=mountd
firewall-cmd --permanent --add-service=rpc-bind
firewall-cmd --reload