Right now package tests will produce a manifest of all files and packages as an xml file.
This file can get quite big and long to review.
So it would be useful to have a tool to encode usual business rules and verify that the packages described in the manifest conform to these rules.
Example of rules are:
- any jar starting with "hadoop-" should be a symlink to /usr/lib/hadoop/ except for hadoop packages. Same for other packages such as zookeeper
- files outside of a bin/ should usually have the rights 644
- Checking that there is no duplicate jar with different versions
This would not ensure all error would be caught, but this would help catching most of them. It would also become more effective as we add more rules to it.