Improving rpm resilience

[pmatilai]

@j-mracek wrote elsewhere:

I know that this is not exactly about RPM6 format but it is related to a new major version of RPM.

There are few things in our environment that are painful for our users but it is not easy to resolve them:

• broken systems

• There are multiple ways how to get to that state and sometimes we have a responsibility - lost power, interupted process, scriptlets, bug in DNF, rpm, ..

• There is no way how to easily revert the transaction or restore the systems. It requires very skilled administrator and a lot of time for support

• Any improvement even small will make a difference

  • DNF should improve history redo, undo operation
  • Is there anything what RPM could improve? (more robust transaction, recovery functionality, reverting transaction, ...)

• replacements of file, symlink, directories by each other during rpm upgrade

• I know that some of those operation are not supported and it makes problem with downgrades

• files, symlinks, directories are mentioned as file in descriptions of problems (file conflicts)

• it makes difficult debugging specially in case when symlink is going to be replaced by directory and the operation fails.

• we would like to see reports more in structures then in string - to prevent parsing of outputs - see in dnf.base.py - def _trans_error_summary(self, errstring):

I understand that what I mentioned above is not easy to resolve, but I thing that RPM6 is a good opportunity to look also in those directions.

[pmatilai]

Linking any and every possible rpm improvement idea to version 6 is exactly what we DON'T want.

Improvements need to be looked at on their own merits, v6 is primarily about the package format facelift. From functionality point of view it's just another release.

As for resilience against power outages and the like - sure, desireable features of course but I can't see it happening in the next couple of years.

[pmatilai]

FWIW, for example this would make a very reasonable ticket on it's own:

files, symlinks, directories are mentioned as file in descriptions of problems (file conflicts)

it makes difficult debugging specially in case when symlink is going to be replaced by directory and the operation fails.

Edit: filed as #2482