Re: Statement-level rollback

On 2 November 2017 at 13:59, Vladimir Sitnikov
<sitnikov(dot)vladimir(at)gmail(dot)com> wrote:

> The performance overhead for "SELECT" statement (no columns, just select)
> statement over localhost is 36±4 us vs 38±3 us (savepoint is pipelined along
> with user-provided query). That is network overhead is close to negligible.

Yep. Not for psqlODBC or other libpq-based drives that can't pipeline
queries though.

> In other words, "savepoint; insert;savepoint; insert;savepoint;
> insert;savepoint; insert;savepoint; insert;" would allocate xids and might
> blow up backend's memory.

RELEASE SAVEPOINT, like psqlODBC does.

> Adding protocol messages would blow pgbouncer, etc things, so it makes sense
> to refrain from new messages unless it is absolutely required.

Yeah, it'd affect proxies, true. But it'd let us get rid of a lot of
very ugly log spam too. And unlike some of the prior protocol tweaks
I've been interested in, it'd be client-initiated so it should be
pretty safe.

--
Craig Ringer http://www.2ndQuadrant.com/
PostgreSQL Development, 24x7 Support, Training & Services