> The fix is to acquire the next page's write lock before we drop the
> current page's. This is deadlock-free since no writer ever tries
> to chain write-locks to the left (in fact the same thing is done in
> the page split logic).
>
> I am not convinced that this explains any of the field reports of
> duplicate rows that we've seen, but it's clearly a bug. Will commit
> the fix shortly.
Sounds good to me. Good find.
--
Bruce Momjian | http://candle.pha.pa.us
pgman(at)candle(dot)pha(dot)pa(dot)us | (610) 853-3000
+ If your life is a hard drive, | 830 Blythe Avenue
+ Christ can be your backup. | Drexel Hill, Pennsylvania 19026