Phil Endecott <spam_from_postgresql_general@chezphil.org> writes:
> Tom Lane wrote:
>> In principle, therefore, the kernel could hold temp table data in its
>> own disk buffers and never write it out to disk until the file is
>> deleted. In practice, of course, the kernel doesn't know the data is
>> transient and will probably push it out whenever it has nothing else to
>> do.
> That makes sense. I suspect that I am seeing writes every 5 seconds,
> which looks like bdflush / update.
> But my connections normally only last for a second at most. In this
> case, surely the table would normally have been deleted before the
> kernel decided to write anything.
That does seem a bit odd, then. Can you strace a typical backend
session and see if it's doing anything to force a disk write?
(I'm too lazy to go check right now whether 7.4 handled temp tables
exactly the same as CVS tip does. I think it's the same but I might
be wrong.)
regards, tom lane