oops should record the short traceback that caused each query?

This could probably be done in connection_raw_execute or the like.

I'd want to do some performance tests before pursuing this.

On 20 July 2010 02:35, Gary Poster <email address hidden> wrote:
> I'd want to do some performance tests before pursuing this.

Yes, there's definitely some risk there. Specifically if we're doing
an unreasonably large number of queries this will make us more likely
to oops; if we're doing a reasonably small number it will be fine.

Robert suggests we could turn this only only on non-lpnet servers.

timeit shows that traceback.format_stack takes about 0.24ms per call
(on a fairly short stack, so it may get longer.) So maybe it's
another 1ms per query we do, whether the page ultimately ends up
oopsing or not. jam says the fastest query is about 1ms but the
slowest ones are obviously

--
Martin

Once we get our peak action counts down to (say) 200 or so I'd be interested in pursuing this; right now we have many things up in the thousands, and this would add seconds to those pages.