Re: RPM Issue

We have included a workaround for this Owl upgrade issue into our latest
build of RPM (4.2-owl22).  When a pre-NPTL Owl system is upgraded, the
/var/lib/rpm/__db.00? files will get removed automatically, thereby
preventing the problem.  This automatic removal only happens once, when
going from a pre-NPTL to a with-NPTL build of RPM (e.g., when upgrading
Owl 2.0-stable to Owl-current - or to our upcoming release).

Simon - thank you for testing Owl-current and for reporting the issue!

On Sun, Mar 28, 2010 at 01:01:41AM +0300, Solar Designer wrote:
> ... I researched the issue some further.  It appears to be quite normal
> that the expected file format of /var/lib/rpm/__db.001 changes when RPM
> built against NPTL is installed on a system that was previously using
> RPM built against LinuxThreads.  That file is memory-mapped and it
> contains mutexes.  Unfortunately, RPM does not distinguish this file
> format change from it having accessed the file in an inconsistent state,
> so it treats this as a temporary error and keeps retrying with the
> confusing message that you saw.
> 
> The workaround is to ensure that there's no rpm process running and to
> "rm /var/lib/rpm/__db.00?".  There's no need to run "rpm --rebuilddb"
> after that, even though this is commonly suggested (cure for a different
> disease).  Once the old __db.00? files are removed, RPM works again (it
> silently re-creates the files) and the issue does not re-occur.
> 
> I still have no idea why the workaround didn't appear to work for Simon.
> It could have something to do with the order in which packages were
> being rebuilt/upgraded on Simon's system.
> 
> This same error message may appear in many other cases.  The above
> description pertains to the specific known issue with upgrading an Owl
> system to our latest packages (built on March 23).
> 
> Alexander