I wrote:
> Alvaro Herrera <alvherre@2ndquadrant.com> writes:
>> Now that I actually tried this, it turned out that this problem is not
>> so simple. vacuum.c already has logic to use conditional acquire of the
>> table-level lock, and if not available it skips the table:
>> LOG: skipping vacuum of "pg_shdepend" --- lock not available
>> so an autovacuum worker is never "stuck" behind another worker trying to
>> vacuum the table.
> Hmm ... but then, how do we have the observed symptom of several workers
> concurrently trying to process the same shared catalog? Seems like all
> but one should fall out at this point.
Oh, see table_recheck_autovac:
tab->at_vacoptions = VACOPT_SKIPTOAST |
(dovacuum ? VACOPT_VACUUM : 0) |
(doanalyze ? VACOPT_ANALYZE : 0) |
(!wraparound ? VACOPT_NOWAIT : 0);
^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
I'll bet they're all trying to do anti-wraparound vacuums. This would
explain why the problem hasn't been observed often enough to have been
fixed long since.
regards, tom lane
