Flamegraphs

Every quasar profile run generates a JSON profile and starts a local flamegraph server on port 7777:

  my_program  2,816 CU
     ...
  flamegraph  http://127.0.0.1:7777/?program=my_program

The server forks into the background and auto-shuts down after 30 seconds of inactivity. If port 7777 is already in use, the profiler reuses the existing server.

Reading the Flamegraph

• Width = cost — wider bars consumed more CU
• Vertical stacking = call depth — bottom is the entrypoint, each bar above is called by the bar below
• Click a bar to zoom in, click root or breadcrumb to zoom out
• Hover for function name, CU count, and percentage

What to look for:

• Wide bars at the top — leaf functions burning the most CU. Primary optimization targets.
• Plateaus — a function as wide as its parent means the parent does nothing besides calling that child.
• Many thin bars — small calls that add up. Consider inlining.

Diff View

quasar profile --diff my_program

Overlays two flamegraphs, highlighting regressions in red and improvements in green. Runs in the foreground -- press Ctrl-C to stop.

Profile History

All profiles are stored as timestamped JSON files in target/profile/profiles/:

my_program__2025-03-15-14-30-00-123.profile.json

The viewer can load any profile from this directory. Each file contains the program name, version, binary hash (SHA-256), binary size, and a hierarchical frame tree with CU values.