I'm de-duping a 45-million-row table (about 4 gigs) by creating a new table
with the unique rows of the first table (should be about 3 gigs). I'm on a
64-bit SuSE Linux with 2 gigs of memory and another gig of swap space. I
tracked postmaster's use of memory and swap space after I run the query, and
I noticed that as soon as postmaster first uses up all available memory and
swap space, I get this:
2004-08-02 19:35:52 LOG: checkpoint record is at 7/4AA2F590
2004-08-02 19:35:52 LOG: redo record is at 7/4AA2F590; undo record is at
0/0; shutdown TRUE
2004-08-02 19:35:52 LOG: next transaction ID: 740; next OID: 116212470
2004-08-02 19:35:52 LOG: database system is ready
2004-08-02 19:38:04 LOG: server process (PID 25302) was terminated by
signal 9
2004-08-02 19:38:04 LOG: terminating any other active server processes
2004-08-02 19:38:04 LOG: all server processes terminated; reinitializing
2004-08-02 19:38:04 FATAL: the database system is starting up
2004-08-02 19:38:04 LOG: database system was interrupted at 2004-08-02
19:35:52 PDT
2004-08-02 19:38:05 LOG: checkpoint record is at 7/4AA2F590
2004-08-02 19:38:05 LOG: redo record is at 7/4AA2F590; undo record is at
0/0; shutdown TRUE
2004-08-02 19:38:05 LOG: next transaction ID: 740; next OID: 116212470
2004-08-02 19:38:05 LOG: database system was not properly shut down;
automatic recovery in progress
2004-08-02 19:38:05 LOG: redo starts at 7/4AA2F5D8
2004-08-02 19:38:05 LOG: record with zero length at 7/4AA4D5C0
2004-08-02 19:38:05 LOG: redo done at 7/4AA4B560
2004-08-02 19:38:07 LOG: database system is ready
And the query never completes. This happens reproducibly and consistently. I
like to think that two gigs should be enough memory. What is killing the
postmaster process and how can I stop it? And, more importantly, is there
any way I can run this query? Thanks for any help you can provide,
Kevin