UNIX for Dummies Questions & Answers

I have a somewhat interesting problem, we've decided to load balance a java application and as such I'll be running our application on four physical machines as opposed to the single machine it's currently running on.

I've centralized the directory that the application requires (including the application itself), along with all of its subdirectories and necessary files/libraires/etc and shared/exported out via NFS to the four machines that will each run a copy of the application. I will be using Nginx to load balance through these 4 instances via round robin.

However, for logging purposes it's best NOT to share one log file amongst the four physical servers that will each be running an instance of the application (basically each one is running its own jvm), as it's most likely an invitation to log corruption at some point.

Unfortunately, the configuration files to tell each instance to write its own logfile (with a unique name) live in a subdirectory of the parent directory which I'm sharing via NFS. And the java application was written in a way such that the relative path of the configuration files is hard coded within the application itself, and thus I can't just move the configuration directory to a non-NFS local directory on each machine.

Is there anyway I can have a local only subdirectory (for each machine) inside of an NFS share? I'm assuming no, b/c of the fact that the NFS share itself does not exist on the local machines but rather is basically just a link to the central machine...But I thought I'd ask if there was a way this could be done.

symlinks perhaps this will satisfy your needs.

Remove the subdirectory in question on NFS server and create a link with same name inside that mountpoint on each NFS client pointing to local directory.

Hope that helps
Regards
Peasant.

Yes, a symbolic link on the nfs is the way to go.
A link tmp -> /tmp is quite common.
The same applies to a log directory; the destination directory with correct attributes must exist on each client.
Eventual log rotation/cleanup must run on each client - the best method is to let the application do it.
--
An alternative is to use automounter and a hierarchical mount. Probably more complicated and less preferrable.