[postgis-devel] Introducing wagyu to validate MVT polygons

[Raúl Marín Rodríguez]

Hi,

> Looking at the graphs I can't grok meaningfulness of mean 99% CI on graph page 13 - it feels like mean is in the saddle of camel-shaped distribution.
Why are there two distinct peaks?

I'm not sure either. I suspect it's either some background job in
Postgres (I'm not using the database for anything else while I run the
benchmarks, but I haven't disabled any periodic tasks), CPU freq
scaling (which isn't disabled in the machine either) or node garbage
collector (since the queries are html request -> node -> postgres ->
node -> html response). Someday I hope to be able to measure times
directly with pgbench.

> Otherwise I like how you fixed up wagyu. Can the changes be upstreamed so that pulling new version is a matter of file copy later?.

The only thing that isn't already upstreamed (and merged) is the
interruption code. I haven't come up with the a code that fits
upstream so, at least for now, I've left a comment in the readme [1].
I'll see if there is a proper way to send that change too (maybe a
compile time flag in wagyu `WAGYU_ENABLE_INTERRUPTS`?).


1 - https://github.com/postgis/postgis/pull/356/files#diff-04ce11e2ad3a22c644aca202237ccdf9R28




--
Raúl Marín Rodríguez
carto.com