NFS Locking Issues

NFS file locking has been known to cause problems on some platforms (notably Solaris and Linux). Stress testing your machine prior to deploying them is likely to show if you will be affected by this. Your NFS-mounted logfiles may show corruption or truncation you may additionally experience some performance degredation.

A work around would be to configure each machine in your cluster to log to a different filename on your file server and post-process the logs to merge them into a single file as a separate step. This will provide performance benefits as you will never be limited by the speed of NFS distributed locking.

You can configure individual servers in a cluster to deviate from the common configuration by following the instructions entitled 'Customizing Virtual Servers for Different Web Servers' in section 14.4 of the Zeus Web Server User Guide.

A simpler strategy would be to configure each web server child process to log to a different file by using the "%c" code in the log file name - "%c" will be replaced by the process id.

NFS Caching Issues

An interaction between the caching mechanism of Zeus Web Server and the file stat caching of NFS can result in unexpected NFS behavior when deleting files. This can occur if the web document root is stored on NFS, has quotas enabled, and is also directly accessible via FTP. In this case, if a user deletes then uploads a large amount of data, they can become over quota without it being obvious why.

Because Zeus Web Server has open file handles on cached files, any attempt to delete a cached file results in the NFS server renaming to files to ".nfs" followed by a random series of digits - instead of actually deleting them. These files will disappear within, by default, 2 minutes. However, they still occupy disk space, and so count towards any specified quotas.

This problem can be mitigated by altering the tuning!cache_flush_internval setting, and also specifiying the actimeo (or acdirmax and acregmax) option when the NFS filesystem is mounted. The setting for the NFS mount should be half or less than the tuning!cache_flush_internval value.

For further details, please see the NFS manual ('man 5 nfs').