On Tue, Jun 14, 2022 at 8:48 PM Álvaro Herrera <alvherre@alvh.no-ip.org> wrote:
> Oh, that definitely sounds like a bug in amcheck then. Please post the
> output of
> select * from heap_page_items(get_raw_page('the table', the_page));
> for some of these pages, preferrably just around the time the errors for
> these pages are shown; maybe someone can figure out which tuples cause
> these errors and how to fix amcheck. It might be difficult to catch
> tuples 'in the act' though.
I've switched to another table with less update traffic and was able
to catch the following:
(verify_heapam before)
blkno | offnum | attnum | msg
-------+--------+--------+-----------------------------------------------------------------
20365 | 8 | | xmax 1965224948 precedes relation freeze
threshold 1:1914858032
20365 | 12 | | xmin 1965224948 precedes relation freeze
threshold 1:1914858032
(2 rows)
(heap_page_items dump) - https://dpaste.org/fWzEj/raw
(verify_heapam after)
blkno | offnum | attnum | msg
-------+--------+--------+-----------------------------------------------------------------
20365 | 12 | | xmax 1965224955 precedes relation freeze
threshold 1:1914858032
20365 | 15 | | xmin 1965224955 precedes relation freeze
threshold 1:1914858032
(2 rows)
If this is not enough, I can try to catch even more matching before/after pairs.
Best regards,
Sergey Aleynikov