vacuum verbose analyze lazy problem.

Hello,

I just have installed Vadim's patch for speeding up vacuum.
And have quite strange problem.
I have quite small table:
[root(at)mx src]# ls -l /home/postgres/data/base/db/users*
-rw------- 1 postgres postgres 4120576 Dec 14 08:48
/home/postgres/data/base/db/users
-rw------- 1 postgres postgres 483328 Dec 14 08:46
/home/postgres/data/base/db/users_id_key
-rw------- 1 postgres postgres 8192 Dec 14 08:20
/home/postgres/data/base/db/users_id_seq

I did vacuum verbose analyze lazy;, and it locks up (or better say do
something for a long time.

Before started vacuuming users.
19379 pts/0 R 2:24 /home/postgres/bin/postgres localhost postgres
db VACUUM

Before I kill the backend.
19379 pts/0 R 4:41 /home/postgres/bin/postgres localhost postgres
db VACUUM

It spends at least 2 minutes trying vacuuming users. Usually this table is
vacuumed & analyzed in few seconds.

Here is the output of vacuum verbose analyze, I done after this problem
arises.

db=# vacuum verbose analyze users;
NOTICE: --Relation users --
NOTICE: Pages 978: Changed 1, reaped 832, Empty 0, New 0; Tup 14280: Vac
13541, Keep/VTL 0/0, Crash 0, UnUsed 265, MinLen 248, MaxLen 340; Re-using:
Free/Avail. Space 3891320/3854076; EndEmpty/Avail. Pages 0/666. CPU
0.09s/1.25u sec.
NOTICE: Index users_id_key: Pages 35; Tuples 14280: Deleted 82. CPU
0.00s/0.05u sec.
NOTICE: Index ix_users_account_name: Pages 56; Tuples 14280: Deleted 82. CPU
0.01s/0.05u sec.
NOTICE: Index ix_users_blocked: Pages 31; Tuples 14280: Deleted 82. CPU
0.00s/0.05u sec.
NOTICE: Rel users: Pages: 978 --> 503; Tuple(s) moved: 640. CPU 0.22s/0.22u
sec.
NOTICE: Index users_id_key: Pages 59; Tuples 14280: Deleted 640. CPU
0.00s/0.04u sec.
NOTICE: Index ix_users_account_name: Pages 93; Tuples 14280: Deleted 640.
CPU 0.00s/0.04u sec.
NOTICE: Index ix_users_blocked: Pages 32; Tuples 14280: Deleted 640. CPU
0.00s/0.04u sec.
VACUUM

I wouldn't consider it's a bug, but from my point of view it is quite strange.
Any comments?

BTW, I did a backup, and can supply anyone interested with original table.

--
Sincerely Yours,
Denis Perchine

----------------------------------
E-Mail: dyp(at)perchine(dot)com
HomePage: http://www.perchine.com/dyp/
FidoNet: 2:5000/120.5
----------------------------------