Arturo Perez <aperez@hayesinc.com> writes:
> Saturday I changed a table to add a varchar(24) and a TEXT column.
You didn't actually say which of these tables you changed?
> I'm not very good at reading these but it looks like sort memory might
> be too low?
The runtime seems to be entirely in the index scan on user_tracking.
I'm surprised it doesn't do something to avoid a full-table indexscan
--- in this case, hashing with extended_user as the inner relation would
seem like the obvious thing. Is user_id a hashable datatype?
It's possible that adding the columns would have affected the plan by
making it look like a sort or hash would take too much memory, but if
that were it then your hand increase in work_mem should have fixed it.
Tis odd. I don't suppose you know what plan was used before?
regards, tom lane