On Mon, Mar 19, 2018 at 4:12 PM, Peter Geoghegan <pg@bowt.ie> wrote:
On Mon, Mar 19, 2018 at 1:55 PM, Jeremy Finzel <finzelj@gmail.com> wrote: > @Peter : > > staging=# SELECT * FROM page_header(get_raw_page('pg_authid', 7)); > lsn | checksum | flags | lower | upper | special | pagesize | > version | prune_xid > ----------------+----------+-------+-------+-------+---------+----------+---------+----------- > 262B4/10FDC478 | 0 | 1 | 304 | 2224 | 8192 | 8192 | > 4 | 0 > (1 row)
Thanks.
That looks normal. I wonder if the contents of that page looks consistent with the rest of the table following manual inspection, though. I recently saw system catalog corruption on a 9.5 instance where an entirely different relation's page ended up in pg_attribute and pg_depend. They were actually pristine index pages from an application index. I still have no idea why this happened.
This is very much a guess, but it can't hurt to check if the contents of the tuples themselves are actually sane by inspecting them with "SELECT * FROM pg_authid". heap_page_items() doesn't actually care about the shape of the tuples in the page, so this might have been missed. -- Peter Geoghegan
The data all looks fine. I even forced the index scan on both indexes which also looks fine.
