Data corruption problem...

Hello - Ran into an interesting problem last week with data unexpectedly
being deleted. Has anyone else experienced this and have any suggestions
on how to prevent this in the future?

Here is the environment:

Server: 8.0.3 on Solaris 9
JDBC Driver: 8.0.311

This is the series of statements that cause the problem, but only under
a certain condition listed below.

delete from X where id in (select bad_id from Z where name='qwerty');
delete from Y where id in (select bad_id from Z where name='qwerty');
delete from Z where name='qwerty');

The first 2 statements have an error, column bad_id does not exist in
table Z. If I run this normally from PGAdmin the first 2 silently fail,
and the last succeeds. In this case tables X and Y are untouched. An
error for this would be nice but the data is OK, so far so good.

Now we have J2EE (Tomcat hosted) app that access this database via
connection pool / JDBC driver. The problem scenario is as follows:

1) Start the J2EE app and do some work to access the database, although
I don't think this matters.
2) Shut down the app server, and as result the connection pool starts to
shut down.
3) Once the connection pool has started to shut down execute the
statements listed above from PGAdmin.
4) PGAmin will hang for a second like it's waiting for locks.
5) The statement will return normally, reporting the rows affected from
the last statement.
6) Tables X and Y are now empty. This is not good.

Now if I fix the first 2 statements the problem doesn't happen, I can
execute it during shutdown and the data is OK. Obviously I would really
prefer not to have times where executing a bad SQL statement deletes the
whole table. Has anyone seen this before, any suggestions how to debug
this or where to start looking?

Thanks!!!

-Nate