> For the complete documentation index, see [llms.txt](https://protocol-labs.gitbook.io/testground/llms.txt). Markdown versions of documentation pages are available by appending `.md` to page URLs; this page is available as [Markdown](https://protocol-labs.gitbook.io/testground/table-of-contents/analyzing-the-results/capturing-profiles.md).

# Capturing profiles

?> This feature was first made available in Testground v0.5.4 + sdk-go v0.2.8.

Profiling is a powerful way to understand how the code under test behaves under the hood. Different programming languages support different types of profiles. For example, Go supports these types:

* CPU profiles.
* Heaps and allocs profiles.
* Goroutine, threads, mutex and block profiles.

Refer to the [`pprof` package godocs](https://golang.org/pkg/runtime/pprof/#Profile) for more information.

Testground has first-class support for capturing profiles continuously throughout the test execution, and it's very versatile:

* You can choose to profile all test run instances, or just a subset of them (per group).
* You can choose which profile types to record.
* You can set the sampling interval or frequency of capture for snapshot-type profiles (e.g. goroutine, heap, etc. profiles in go).

?> **Success** Because profiling is a language feature, Testground delegates the capture to the corresponding SDK by sending the profiling requirements for each instance inside the `TEST_CAPTURE_PROFILES`� env variable.

The profiles are written to the [output path for each instance](/testground/table-of-contents/analyzing-the-results.md#testground-outputs-structure), and **are available in the output bundle** generated by `testground collect` or the `testground run --collect` flag.

### Enabling profiles

You can enable profiles by setting the profiling settings under the run configuration.

**To enable profiling globally**

**composition.toml**

```toml
...

[global.run]
# this enables a continuous CPU profile, a heap profile every 5 seconds,
# and an allocs profile every 10 seconds.
profiles = { cpu = "true", heap = "5s", allocs = "10s" }

...
```

**To enable profiling in a single group**

**composition.toml**

```toml
...

[[groups]]
id = "low-latency"
[groups.run]
# same settings as above
profiles = { cpu = "true", heap = "5s", allocs = "10s" }

...
```