"Josh Berkus" <josh@agliodbs.com> writes:
> Given this, why bother with system-generated OIDs on user rows at all?
> Why not simply reserve the OIDs for the system tables?
An option to not generate OIDs unless requested (on a table-by-table
basis) has been discussed. It seems like a fine near-term solution
to me. 8-byte OIDs are a longer-term solution, because they'll break
a lot of things (including clients...)
>> This is certainly not ideal, but it's not nearly as big a problem as
>> transaction ID wraparound. You can live with it, whereas right now
>> xact ID wraparound is catastrophic. That we gotta work on, soon.
> Nothing like reassuring us commercial DB users, Tom. :-P
> Can you describe what you're talking about?
It's in the archives: after 4G transactions, your database curls up
and dies. When your pg_log starts to approach 1Gbyte (2 bits per
transaction) you'd better plan on dump/initdb/reload.
regards, tom lane
