Problem with VACUUM after very large delete?

Hello all,

(7.2.1 on Debian Woody 3.0 with various 7.2.3 patches backported)

Earlier today I deleted about 31 million records from one of our tables with a very complex query which required a full table scan. This query took quite a few hours to run (in fact, overnight).

Later, I did a VACUUM and then an ANALYZE.

However, it does not appear that either:

1) The records were actually deleted or

2) The VACUUM is working properly

Why? The VACUUM doesn't seem to be vacuuming the two indexes on the table. I'm not sure what's going on, but if you could make any suggestions, I'd appreciate it.

Details are below.

Thanks,

Doug

pexicast_lg=# ANALYZE VERBOSE audit_log;
NOTICE:  Analyzing audit_log
ANALYZE
pexicast_lg=# VACUUM VERBOSE audit_log;
NOTICE:  --Relation audit_log--
NOTICE:  Pages 2083605: Changed 2961, Empty 0; Tup 105773248: Vac 0, Keep 31106432, UnUsed 2.
        Total CPU 89.49s/13.67u sec elapsed 1101.17 sec.
VACUUM
pexicast_lg=# \d audit_log
                  Table "audit_log"
  Column  |           Type           |   Modifiers
----------+--------------------------+---------------
 at       | timestamp with time zone | default now()
 type     | integer                  |
 severity | integer                  |
 who      | integer                  |
 target   | integer                  |
 var1     | integer                  |
 var2     | integer                  |
 message  | character varying(255)   |
 ip       | character(15)            |
 login    | integer                  |
Indexes: audit_type_target_idx,
         audit_type_var1_idx