On Sun, Aug 23, 2009 at 2:18 PM, Jeff Ross<jross@wykids.org> wrote:
> I had to modify your query slightly to make it run--hope I got what you are
> after!
>
> select (h).* from (select
> heap_page_items(get_raw_page('pg_namespace',0)) as h) as x;
>
> http://www.openvistas.net/pageinspect.html
Incidentally, may as well ask the usual questions:
Did you upgrade this via pg_dump/pg_restore or via pg_migrator?
Are you running with fsync=off?
Are you running with synchronous_commit=off?
Have you had any system crashes?
Have you had any Postgres crashes?
Of the above only fsync=off combined with a system crash ought to be
unsafe, but it could be good to know which other events have occurred
to know which code paths have been exercised.
Do you still have pg_clog/0000 (that's kept longer than the xlog so I
would expect you would), how large is it? could you send that file? It
has a record of which transactions have committed or aborted, no
private data.
--
greg
http://mit.edu/~gsstark/resume.pdf