Re: Patch to show individual statement latencies in pgbench output

Florian Pflug wrote:
> Attached is an updated version (v4).
>

I've attached a v5. No real code changes from Florian's version, just
some wording/style fixes and rework on the documentation. The user side
is now consistent about calling these statement latencies for example,
even though the internals still call them command latencies most places.

Since using this new feature will introduce a whole stack of new calls
to get the system time, I also added a warning about that impacting results:

Note that collecting the additional timing information needed for
detailed latency computation does add some overhead. This will slow
average execution speed and lower the computed TPS. The exact amount
of slowdown varies significantly based on platform and hardware.
Comparing average TPS values with and without latency reporting enabled
is a good way to measure if the timing overhead is significant.

I wasn't able to see any significant slowdown on my modern Linux systems
doing such a test:

$ ./pgbench -T 10 -S -c 8 -j 4 pgbench
tps = 6716.039813 (including connections establishing)
tps = 6720.238878 (excluding connections establishing)
$ ./pgbench -T 10 -S -c 8 -j 4 -r pgbench
tps = 6708.544618 (including connections establishing)
tps = 6712.728526 (excluding connections establishing)

But I know gettimeofday is fast here. Worth including a warning for
though I think.

I'm out of things to check here, marking this one ready for a committer
review. The patch hasn't had a committer assigned yet, so whoever wants
to claim it should mark the CF app.

--
Greg Smith 2ndQuadrant US Baltimore, MD
PostgreSQL Training, Services and Support
greg(at)2ndQuadrant(dot)com www.2ndQuadrant.us