Debug your database with Flame Graphs

Flame Graphs are used as a debugging tool to identify latency and the part of the execution path that takes most of the CPU time. Use Flame Graphs when you:

  • Need to understand which Scylla code path/functions are using the most time. For instance, when you have latency issues.

  • Need to compare time spent in particular Scylla code paths/functions on different shards. For instance, when you have latency issues on one CPU but not on the other.

Run a Flame Graph


  1. Clone the repoistory

git clone
  1. Change into the directory where the flamegraph packages are located.

cd FlameGraph

Optionally if your are on Centos 8 or RHEL 8 run:

sudo dnf -y install perl-open.noarch
  1. Run the following perf commands, using Map CPU to Scylla Shards and Using the perf utility with Scylla for reference.

sudo perf record --call-graph dwarf -C <CPU on which you are onrecording>
sudo perf script | `pwd`/ | `pwd`/ > some_name.svg
  1. The result is an .svg file that is not just a picture but a dynamic diagram where you can search, zoom in, and zoom out. In order to enjoy a Flame Graph properly with all of its features you can open it in a browser such as Chrome or Firefox.



  • On the CPU you are recording, try to load Scylla to consume 100% of the CPU runtime. Otherwise you’ll see a lot of OS functions related to the idle time handling

  • Recording on all shards (e.g. using “perf record” -p parameter) may lead to confusing results recording the same symbol called from different threads (shards). This is not recommended.