I have recently run into problems with the unfsd server reporting the following error in the file server's daemon.log: fd cache inconsistency! At around the same time on the clients in the cluster, files end up "dropping out of sight" and return an "Input/Output Error" message -- I can't even get the files' inode number by doing an 'ls -di <filename>'. In the client's syslog, I end up with the following messages: kernel: nfs_refresh_inode: inode 55279643 mode changed, 0120777 to 0100644 This issue seems to only crop up when we update files directly on the file server instead of going through an NFS mount. I've dug and dug around on the web for an answer to this particular issue, but haven't actually found an answer to it. Both the clients and the server are running a vanilla kernel version 2.4.18 straight out of the Debian apt archives. Any ideas or comments are welcome. =op P.S. I just had a thought when writing this email that it might be related to the client-side caching of file stats. If one of the clients is reading from a file, and then the file is updated outside of NFS at the same time, would this cause a similar problem? -- June Tate Software Engineer Centralized Technologies, Inc.
Attachment:
signature.asc
Description: This is a digitally signed message part