Today we have one single huge rpm that installs everything under /opt.
But we have following deployment scenarios:
- memcached only nodes.
- memcached+ep-engine+vbucketmigrator nodes.
- pecl-memcache+mcmux on web nodes.
Later, we will have cluster related packages also deployed on these nodes.
Ideally, we would want these components to be packaged in their own separate rpms that installs them into system standard paths. And if all of them are required to be installed, they can be installed by using meta-packages that express dependency on these component packages.