Dmitry Tkach <dmitry(at)openratings(dot)com> writes:
> 3) I restart the server manually, and try again
> analyze mytable;
> ... it *works*
> 4) I let it run for a while, then try again:
> analyze mytable;
> ... it crashes.
Proves nothing, since ANALYZE only touches a random sample of the rows.
If you get that behavior with VACUUM, or a full-table SELECT (say,
"SELECT count(*) FROM foo"), then it'd be interesting.
What range of file names do you actually have in pg_clog/, anyway?
regards, tom lane