Skip to main content

Flame Charts with CPU profiling data

The Flame Chart is a visualization method that allows you to see the CPU usage of your application over time. In other words, it is similar to a flame graph, but has an additonal time dimension. This means that you can see how the CPU usage of your application changes over time, and how it is affected by different events.

Screenshot of Flame Chart with CPU profiling data: Flame Chart with CPU profiling data

Prerequisite​

Before you can use Flame Chart with CPU profiling data, you need to be collecting the CPU profiling data with cpu or thread_id label, or any other label that will never have samples with overlapping timestamps, threads or CPUs can't have two things running on them at the same time so they are well suited for this purpose.

You can do that by adding the following section to the parca agent configuration if you don't have it already:

  relabel_configs:
    - source_labels: [__meta_thread_id]
      target_label: thread_id
    - source_labels: [__meta_cpu]
      target_label: cpu

This will add the cpu and thread_id labels to the CPU profiling data, which can be then used for Flame Chart visualization. More information about the relabeling can be found in the relabeling documentation.

How to use Flame Chart with CPU profiling data​

Once you're on the Polar Signals cloud, you can use Flame Chart by selecting the Flame Chart option in the visualization type dropdown. Flame Chart with CPU profiling data

info

Note: Since the Flame Chart is right now in alpha, you will have to specifically enable Flame Charts in the Preferences section. you can do this by clicking on the Preferences button near the visualization type selector and then toggling the Enable Flame Charts on.

It is always recommended to use flame chart with the following query conditions:

  • Add a node filter to the query to limit the data to a specific node.
  • Add either cpu or thread_id label to the Sum by section to group the data by CPU or thread ID.