Ann,

> -----Original Message-----
> From: Ann W. Harrison [mailto:aharrison@...]
> Sent: Thursday, December 14, 2006 5:52 PM
>
> Vlad has done some interesting examination of the
> index handling code and offered a second possible
> explanation of these errors.
>
> My theory requires that the server or at least one
> of server processes crash. Without crashes, my
> theoretical problem can't occur.
>
> Vlad's theory requires that entries be removed and
> inserted in the same value range to the extend that
> an index page first splits and is then recombined.
>
>
> So, for those of you who have highly concurrent
> applications running 1.5.3 Classic and are seeing
> occasional corrupt indexes with a wrong page type
> in the index (expected 7 encountered 5) - do your
> environments resemble either of these?

I'm not sure about the crashes. Our systems are built to be tolerant
of database disconnects, and unfortunately we do not log this type of
activity.

However, within a single transaction we will delete between 10k - 70k
rows, and insert nearly the same amount. Most likely, 80-90% of the
newly inserted rows are duplicates of the just-deleted rows.

Bob M..