5 ms·

Visualize MySQL query execution plans as interactive FlameGraphs

63 points by tanelpoder 8mo ago

vgrippa 8mo ago
Thanks for posting!
dveeden2 8mo ago
The video from the presentation on this: https://youtu.be/tdx9leN2kBg?si=uYa7xsOvpoonjaxB https://youtu.be/tdx9leN2kBg?si=uYa7xsOvpoonjaxB This was part of the MySQL Belgian Days that was organized in the days before FOSDEM.
- lofaszvanitt 8mo ago
  Any slides maybe?
  - vgrippa 8mo ago
    You can find the slides here: https://github.com/vgrippa/presentations/blob/main/2026_FOSDEM_BringingFlameGraphsMySQLQueryOptimization.pptx https://github.com/vgrippa/presentations/blob/main/2026_FOSD...
lofaszvanitt 8mo ago
Do you have slides from your FOSDEM presentation? The video has this funny angle :(.
- vgrippa 8mo ago
  Hi! The slides can be found here: https://github.com/vgrippa/presentations/blob/main/2026_FOSDEM_BringingFlameGraphsMySQLQueryOptimization.pptx https://github.com/vgrippa/presentations/blob/main/2026_FOSD...
  - lofaszvanitt 8mo ago
    Thanks for the placeholders ;)
javier2 8mo ago
Thanks for sharing!
hobs 8mo ago
Hah I did this for SQL Server execution graphs at one point (via profiler or extended events) and then I of course realized that anything that goes more than one or two events deep is just hellish SQL anyway, love to see the execution plan version that's much smarter.
- bonesss 8mo ago
  That feeling when your fancy graph points out the obvious, and ‘bad thing bad’ is the only takeaway. Data science <pew pew>!
ab_testing 8mo ago
Very cool but also more interesting is using Perl for any newish project. It is hard to find maintainers or developers knowing Perl nowadays. Especially in data science related projects as python has been the de-facto tool for that field for some time now.
- vgrippa 8mo ago
  Indeed, but it was easier to implement the logic since original flamegraphs are in Perl.
bmn__ 8mo ago
If I were in vgrippa's place, I would have not produced flamegraphs, but treemaps. It's the far superior choice for visualising profiling/query plans and the like, no matter along which property one would measure. Munging the mysql output into something that is delectable as input to kcachegrind is straight-forward. Screenshots: https://ddg.gg/?q=%21i+kcachegrind https://ddg.gg/?q=%21i+kcachegrind
- vgrippa 8mo ago
  Thanks for the suggestion. Actually I received some feedback to produce something like flamescope from netflix https://github.com/Netflix/flamescope https://github.com/Netflix/flamescope I will take a look at the treemaps and try to incorporate.

63 points by tanelpoder 8mo ago

Thanks for posting!

The video from the presentation on this: https://youtu.be/tdx9leN2kBg?si=uYa7xsOvpoonjaxB https://youtu.be/tdx9leN2kBg?si=uYa7xsOvpoonjaxB This was part of the MySQL Belgian Days that was organized in the days before FOSDEM.

lofaszvanitt 8mo ago
Any slides maybe?
- vgrippa 8mo ago
  You can find the slides here: https://github.com/vgrippa/presentations/blob/main/2026_FOSDEM_BringingFlameGraphsMySQLQueryOptimization.pptx https://github.com/vgrippa/presentations/blob/main/2026_FOSD...

lofaszvanitt 8mo ago

Do you have slides from your FOSDEM presentation? The video has this funny angle :(.

vgrippa 8mo ago
Hi! The slides can be found here: https://github.com/vgrippa/presentations/blob/main/2026_FOSDEM_BringingFlameGraphsMySQLQueryOptimization.pptx https://github.com/vgrippa/presentations/blob/main/2026_FOSD...
- lofaszvanitt 8mo ago
  Thanks for the placeholders ;)

javier2 8mo ago

Thanks for sharing!

hobs 8mo ago

Hah I did this for SQL Server execution graphs at one point (via profiler or extended events) and then I of course realized that anything that goes more than one or two events deep is just hellish SQL anyway, love to see the execution plan version that's much smarter.

bonesss 8mo ago
That feeling when your fancy graph points out the obvious, and ‘bad thing bad’ is the only takeaway. Data science <pew pew>!

ab_testing 8mo ago

Very cool but also more interesting is using Perl for any newish project. It is hard to find maintainers or developers knowing Perl nowadays. Especially in data science related projects as python has been the de-facto tool for that field for some time now.

vgrippa 8mo ago
Indeed, but it was easier to implement the logic since original flamegraphs are in Perl.

bmn__ 8mo ago

If I were in vgrippa's place, I would have not produced flamegraphs, but treemaps. It's the far superior choice for visualising profiling/query plans and the like, no matter along which property one would measure. Munging the mysql output into something that is delectable as input to kcachegrind is straight-forward. Screenshots: https://ddg.gg/?q=%21i+kcachegrind https://ddg.gg/?q=%21i+kcachegrind

vgrippa 8mo ago
Thanks for the suggestion. Actually I received some feedback to produce something like flamescope from netflix https://github.com/Netflix/flamescope https://github.com/Netflix/flamescope I will take a look at the treemaps and try to incorporate.