> I ended up with few only indexes on the operations table, because the
> processes that fill it up do minimal lookups to see if data is already in the
> table, if not do inserts. Then at regular intervals, the table is cleaned up -
> that is, a process to remove the duplicate is run. This unfortunately costs
> OIDs, but I found no other reasonable way to do the fast inserts. Perhaps the
> best way is to create the table without OIDs (but wouldn't this still waste
> OIDs?) use COPY and then clean afterwards?
No, WITHOUT OIDS is implemented specifically to not waste OIDs.
Chris