Re: Read Uncommitted

On Mon, 2008-05-26 at 13:25 -0400, Tom Lane wrote:
> Hannu Krosing <hannu(at)krosing(dot)net> writes:
> > On Mon, 2008-05-26 at 16:55 +0200, Peter Eisentraut wrote:
> >> If the data in a table never changes, why would VACUUM or HOT need to touch
> >> it? The use case isn't clear to me.
>
> > I guess the use-case is about a long read-write transaction doing
> > read-only access to an update-only table and thus blocking vacuum on
> > other tables.
>
> ... in which case the proposed kluge would result in unstable,
> unpredictable answers, so there is still no plausible use-case.

maybe it was meant as a super-power-user tool (and a big footgun) .

btw, when is a transaction id currently assigned to a transaction - when
INSERT/UPDATE/DELETE statement is first seen, or when data is actually
modified ?

that is when doing

INSERT INTO logtable
SELECT current_timestamp, count(*) FROM really_huge_table;

will there be a transaction id for just the tiny moment the returned row
is inserted or for the whole count(*) time ?

----------------
Hannu