On 2014-11-04 13:51:23 -0500, Tom Lane wrote:
> Bernd Helmle <mailings@oopsware.de> writes:
> > --On 3. November 2014 18:15:04 +0100 Sven Wegener
> > <sven.wegener@stealer.net> wrote:
> >> I've check git master and 9.x and all show the same behaviour. I came up
> >> with the patch below, which is against curent git master. The patch
> >> modifies the COPY TO code to create a new snapshot, after acquiring the
> >> necessary locks on the source tables, so that it sees any modification
> >> commited by other backends.
>
> > Well, i have the feeling that there's nothing wrong with it. The ALTER
> > TABLE command has rewritten all tuples with its own XID, thus the current
> > snapshot does not "see" these tuples anymore. I suppose that in
> > SERIALIZABLE or REPEATABLE READ transaction isolation your proposal still
> > doesn't return the tuples you'd like to see.
>
> Not sure. The OP's point is that in a SELECT, you do get unsurprising
> results, because a SELECT will acquire its execution snapshot after it's
> gotten AccessShareLock on the table. Arguably COPY should behave likewise.
> Or to be even more concrete, COPY (SELECT * FROM tab) TO ... probably
> already acts like he wants, so why isn't plain COPY equivalent to that?
Even a plain SELECT essentially acts that way if I recall correctly if
you use REPEATABLE READ+ and force a snapshot to be acquired
beforehand. It's imo not very surprising.
All ALTER TABLE rewrites just disregard visibility of existing
tuples. Only CLUSTER/VACUUM FULL do the full hangups to keep all the
necessary tuples + ctid chains around.
Greetings,
Andres Freund
--
Andres Freund http://www.2ndQuadrant.com/
PostgreSQL Development, 24x7 Support, Training & Services