I've got several (let's say 4 for the sake of this post) NFS mounts which contain various stages and save points of the same files.

*please note I'm over simplifying my example for the sake of this post*

For my build I want to create a blending of the 4 NFS mounts using the latest files from each into a staging area. However, I don't want to copy the files.

So the end result.

code.c -> /net/server1/code.c
code.h -> /net/server4/code.h
index.html -> /net/server2/index.html
index.php -> /net/server2/index.html
somefolder/index.html -> /net/server3/somefolder/index.html

The reason I don't want to copy the files is the 4 mount points are +1T in size... where the stage area is only a few 100M. If I symlink I can build the stage area the way I need it without using up any space.

I don't think you'll manage that without a 'find' expression to locate files after a certain date piped to ln -sf to link them. I wouldn't dream of attempting the arguments without the exact situation. This presumes that any newest file can appear on any server. If you can narrow that down, you will be able to save a lot of disk thrashing which is gloriously inefficient over nfs.

The exact scenario. One NFS is the master in one datacentre. Then there are at least 4 other data centres, with up to 10 labels each (snapshots of the code at points in time), sync'd from a DB.

When building the DB copy it looks at a stage area. but the master copy is in one city and the closest NFS mount is on one of the other 4. So when building the stage area makes sense to copy as much from the local datacentre as possible, then copy the missing from the master copy.

Problem is we don't know at the time of the script which data centre is the local one, and which of the 10 labels contains the most up to-date file. Hence the hunt. The resent problem is were running out of space on the stage area so the rsync currently in place is falling over.

So the idea is to symlink them to the NFS mount with the latest file... which will be the local datacentre, latest label, and any missing end up pulling from the master.

I'm going to do this by script at the moment but wanted to see if there was any way to get rsync (or other product) to sync as symbolic links to the original location rather than copy.

Sounds like a crazy way to do software development. Has git been considered? I thought that was what it was designed for.

It's not for designing the software its for distributing the compile objects (or in my case more like the third-party libraries needed to compile the objects and other software).