This is a known bug, and if fixed in CVS. It will be released as part
of 7.4.3, or just download pg_autovacuum.c and pg_autovacuum.h from cvs
and compile a new executable.
Matthew O'Connor
Jim C. Nasby wrote:
> There's a bug in pg_autovacuum that makes it vacuum large tables way to
> frequently.
>
> ogr=# select reltuples from pg_class where relname='ogr_summary';
> reltuples
> -------------
> 2.64411e+07
> (1 row)
>
> [2004-04-27 02:03:08 PM] table name: ogr."public"."ogr_results"
> [2004-04-27 02:03:08 PM] relid: 615097; relisshared: 0
> [2004-04-27 02:03:08 PM] reltuples: 7; relpages: 519951
> [2004-04-27 02:03:08 PM] curr_analyze_count: 9904180; cur_delete_count: 168282
> [2004-04-27 02:03:08 PM] ins_at_last_analyze: 9904180; del_at_last_vacuum: 168282
> [2004-04-27 02:03:08 PM] insert_threshold: 507; delete_threshold 1014
> [2004-04-27 02:03:08 PM] added table: ogr."public"."ogr_results"
>
> I suspect the issue is on line 121:
> new_tbl->reltuples = atoi(PQgetvalue(res, row, PQfnumber(res, "reltuples")));
>
> I'm not a C coder, but doesn't atoi just take a text value and convert
> it to an int? Shouldn't all the stats counters be stored as reals?