Re: Re: Re: SUG: Automatic RPM database verification and repair

[seth vidal <skvidal linux duke edu>]

On Wed, 2006-11-29 at 18:10 -0500, Jeff Johnson wrote:
> On 11/29/06, seth vidal <skvidal linux duke edu> wrote:
> > but we're not doing updates to a database in that way.
> >
> > The only time yum is doing the quick open-read-closes of the rpmdb is
> > when it is reading in the info from the rpmdb or getting prco info from
> > a package. Yum switched from:
> >  open rpmdb ro, and use that for all ro interactions
> >  to
> >  open rpmdb ro, get the index number of the header for a package, save
> > the index number into a dict, close the rpmdb (repeat)
> >
> 
> The index number is not guaranteed to remain constant just because yum opened
> an rpmdb ro and then closed.
> 
> >
> > There's no updating going on.
> >
> 
> Yep. Index retrieved, saved in dict, rpmdb closed. No update there.
> 
> The next question is
>     What does yum do to guarantee that the index number from the dict
> is meaningful
>      when it actually *does* get around to doing an update?
> 
> And if the answer is that
>     Yum has its own locks.
> well, that ain't good enough in the real world, /bin/rpm certainly
> does not pay attention
> to yum locks, for one easy reproducer.
> 

I'm not saying yum does guarantee those. I'm asking why does the above
cause the rpmdb to have errors?

-sv