In a previous post Profiling Software Using perf and Flame Graphs, we took a look at how to generate Flame Graphs from perf outputs. In this short follow-up, we will use Flame Graphs to process outputs from pt-pmp. Flame Graphs were not originally thought of for this purpose, but since pt-pmp outputs are similar to the folded samples (the result of using stackcollapse-perf.pl), we can take advantage of its functionality.

There are mainly two differences in the pt-pmp outputs:

• The first column is the number of threads with that same backtrace, and
• The function names are separated by a comma instead of a semicolon

Additionally, pt-pmp will print the date in the first line, so we’ll need to trim that part, too.

Lastly, before …

Please join Percona Principal Support Engineer Marcos Albe as he presents his talk “Flame Graphs 201″ on Thursday, August 22th, 2019 at 11:00 AM PDT (UTC-7).

Register Now

Visualizing profiling information can be a very powerful tool for performance diagnostics, especially if we zoom right into the problem. Flame graphs were developed for this purpose and we use them on a daily basis at Percona to successfully solve complex performance issues. In this presentation, attendees will learn advanced techniques for profiling with performance and visualizing those profiles with flame graphs, as well as other assorted tricks that the flame graph scripts allow us to do.

If you can’t attend, sign up anyways we’ll send you the slides and recording …