Looks reasonable to me. All the patch does is make sure that the result
set is valid. Probably a check I should have done from the beginning,
or pg _autovacuum should be locking tables to make sure they aren't
dropped, but that sounds too intrusive, this is probably better.
Matt
daveg wrote:
> Apologies if this is old news, but pg_autovacuum in 8.0.x has the bad habit
> of SEGVing and exiting when a table gets dropped out from under it. This
> creates problems if you rely on pg_autovacuum for the bulk of your vacuuming
> as it forgets it's statistics when it is restarted and so will skip some
> desireable vacuums.
>
> I looked at the new autovacuum in 8.1 and it appears from casual inspection
> not to have the same problem.
>
> Below is a patch for this that should apply against any 8.0.x. The change
> verifies that the catalog query returned some rows before accessing the row
> data.
>
> -dg
>
> diff -Naur source/postgresql-8.0.2/contrib/pg_autovacuum/pg_autovacuum.c
build/postgresql-8.0.2/contrib/pg_autovacuum/pg_autovacuum.c
> --- source/postgresql-8.0.2/contrib/pg_autovacuum/pg_autovacuum.c 2005-04-02 16:02:03.000000000 -0800
> +++ build/postgresql-8.0.2/contrib/pg_autovacuum/pg_autovacuum.c 2005-09-28 22:15:25.428710172 -0700
> @@ -1013,6 +1013,7 @@
> static void
> perform_maintenance_command(db_info * dbi, tbl_info * tbl, int operation)
> {
> + PGresult *res;
> char buf[256];
>
> /*
> @@ -1069,10 +1070,16 @@
> fflush(LOGOUTPUT);
> }
>
> - send_query(buf, dbi);
> -
> - update_table_thresholds(dbi, tbl, operation);
> -
> + res = send_query(buf, dbi);
> + if (PQntuples(res)) {
> + update_table_thresholds(dbi, tbl, operation);
> + } else {
> + if (args->debug >= 1) {
> + sprintf(logbuffer, "Cannot refind table %s", tbl->table_name);
> + log_entry(logbuffer, LVL_DEBUG);
> + fflush(LOGOUTPUT);
> + }
> + }
> if (args->debug >= 2)
> print_table_info(tbl);
> }
>