Subject | RE: [ib-support] Interpreting the iblockpr output.. |
---|---|
Author | Ignacio J. Ortega |
Post date | 2002-10-28T15:17:16Z |
> From: Ann W. Harrison [mailto:aharrison@...]Yes, but in any case you need to commit from phase to phase, because
> Sent: Monday, October 28, 2002 3:40 PM
>
>
> Hmmm... gbak deadlock. There was a bug fix in restore.e -
> commented even - that described such a situation. It's in
> the routine restore. I have no idea why gbak is using two
> transactions ... that seems like a recipe for disaster.
most of the work is done in DFW.. so to be able to compile SP i.e., you
would need to see relatiopns a so on..
what i dont really understand is why it uses overlapping transacions..,
this is the real recipe for disaster.. ;)
> Have you traced the error to a particular statement?Yes, i know it very well ;), the gbak part, its in one of the latest
>
phases of the restore process, the global overlapped transaction is
committed, and next the sp are are updated to set the owner, the
deadlock is produced after the commit, in the first procedure touched..
i dont have the exact line at hand, but it must be in some past message
to firebird-devel..
>:D, i definitely agree..
> The code is below - I HATE routines that start with
> if (flag_xxx) - they're almost always bandages on suppurating
> wounds.
>
Saludos,
Ignacio J. Ortega