This shows the layout of the opt directory on our master
node. Note that it isn't that machine's /opt directory,
although it could be. On another network, we have exactly that
arrangement--the /opt directory is the master
opt directory for the whole network. On the machine
shown here (which is the "srvr01" from the overview),
the actual /opt
directory is set up to point to this directory, just as it is on
workstations. The only difference is that its /opt/mstr
link has the simpler value "/_c/opt.10".
The salient points of this structure are
pkg and site
subdirectories, and
/opt/appname directories
aren't directories at all, but are symbolic links.
The pkg subdirectory serves to contain the actual
application directories for net-supported packages. If you can
retrieve it from a public archive site, it goes into pkg.
The site directory contains things that we build
and maintain ourselves. In our particular case, this means locally
developed tools, not actual product code. As noted in the Open Questions
section, locally-developed code (even tools) at our site is kept
under configuration control (ClearCase), so we copy released files
into the appropriate /opt/site/appname-n.nn
directory as part of the release process.
The /opt/appname links serve several purposes:
/opt/appname, and trust that the system
administrators are providing the "right" version. This rule can be
broken only under the most unusual of circumstances.