Visualize and interact with micro benchmarks in performance report #65
Conversation
boshek
changed the title
|
No I can't, probably because he hasn't interacted with the repo so we have to tag @alistaire47 |
Co-authored-by: Edward Visel <1693477+alistaire47@users.noreply.github.com>
|
@alistaire47 thanks for such a thorough review. Hugely beneficial to improve this. I've address a number of things but I also have some follow-ups that I was hoping you could look at if possible. |
|
Great work and review everyone! It already looks super useful 🚀 |
| opt_table_font(font = google_font("Roboto Mono")) | ||
| ``` | ||
|
|
||
| Because of the large number of benchmarks, the top 20 benchmark results that deviate most from the baseline in both the positive and negative directions are presented below. |
There was a problem hiding this comment.
I like the suite grouping, that does help. But the explorer doesn't really help find the most extreme results because you can't deselect / select all on the suite and name filters. The data would get big if you did, but maybe it's worth it and just truncating head/tail with a warning? That would get to my ask definitely
|
@assignUser we are happy with this. Hopefully upon the next release, we can elicit some feedback when the RC is voted upon. LMK if you want to review as well. Otherwise I think this is good to merge. |
assignUser
changed the title
After #65 I realized that it would be simple to add JavaScript benchmarks (which we've not visualized before). Because those are run in the same machines as the R and python benchmarks I've added a column that distinguishes between the macro/micro benchmarks:  Otherwise this PR is just some display tweaks to better show labels and updating some text to better explain the microbenchmark explorer.
This PR implements an interface to work with the microbenchmarks, include the C++ and Java ones. Additionally, because there are so many, I've enabled the user to be able to filter for conbench detected regressions and improvements. I've followed the style of the other plots in this report though these one are built in Javascript.
I have also added a link so users can go to conbench to directly view the compare page.
@assignUser I would like to request a review from Ed (https://github.com/alistaire47) but I'm not actually able to tag him here. Are you able to do that?
Attached is a zip file containing a rendered copy using the last release:
performance-release-report.html.zip. Reminder that this needs to be served up via http because of js libraries.