RE: UPDATE many records

You don’t tell us if other users will be concurrently changing any of the records involved.  If you could guarantee that the table won’t be changed, you might be better off doing a CREATE TABLE table_new as SELECT … FROM table_old, dropping table_old, and finally renaming table_new.   Given the way Postgres handles updates, I would think that might perform significantly better.  Even if you did the work in batches (create a new table, insert/select from the old table, drop, rename), that could well be better.  Especially if you re-create the indexes after all the data is moved.

Thanks to a change in historical data, I have a need to update a large number of records (around 50 million). The update itself is straight forward, as I can just issue an "UPDATE table_name SET changed_field=new_value();" (yes, new_value is the result of a stored procedure, if that makes a difference) command via psql, and it should work. However, due to the large number of records this command will obviously take a while, and if anything goes wrong during the update (one bad value in row 45 million, lost connection, etc), all the work that has been done already will be lost due to the transactional nature of such commands (unless I am missing something).