NFS cacheing issue - HP-UX mounted to Linux

[schaepa]

I have an HP-UX 11.23 server that is providing an NFS mount to multiple RedHat 4u4 clients. I have occasionally seen the following behavior:

a file in the NFS share appears different on one of the clients than it does on the other clients or on the master. For example, today one file had a tilde at the end of a line of text on one client, while on the master and the other clients, the character was a carat. The file had not been changed, so it wasn't a case of storing a previous version of the file. There was a true difference - it wasn't just the font of the display, etc. - the character was different.

When I touched the file on the server, it updated correctly on all the clients. However, after that touch I unmounted and remounted the share on the client and the data error reappeared. I then unmounted for 5 minutes and remounted and the error had cleared.

Questions:
1. How could the file become different / corrupt on the client without an update
2. Under what circumstances will the NFS client refresh its data, and what is the scope of the refresh
3. How can I clear the NFS cache or take some other steps to purge bad data across the whole mount on a live production system?

Notes:
1. I can't reproduce the condition at will for testing purposes :(
2. There is a possibility this data corruption is only happening on one client host.