Metoro runs a sampling profiler at 97Hz to understand exactly which container is running on each CPU on a host. This allows us to collect stacktraces of any program currently on the CPU, giving you detailed insights into where CPU time is being spent.

Metoro currently supports CPU profiling for C, C++, Rust, Golang, and Python. For other languages, we show their native runtime methods, though CPU time won't be attributed to the interpreted function itself.

Running the sampling profiler has minimal overhead, typically less than 0.5% of total CPU usage. This ensures that our profiling solution doesn't significantly impact your application's performance.

You can view CPU profiles for each service in its respective service page under the Profiling tab. This provides an aggregated view of all profiles taken across all pods associated with the service in your selected time period. You can filter by specific container in the pod or by cluster.

CPU profiling focuses on understanding how CPU time is spent in your application, whereas other types of profiling, such as memory profiling, focus on understanding how memory is allocated and used. CPU profiling is particularly useful for identifying performance bottlenecks and optimizing application performance.