Documentation Index
Fetch the complete documentation index at: https://docs.odigos.io/llms.txt
Use this file to discover all available pages before exploring further.
What is CPU Profiling?
CPU profiling continuously samples what your processes are spending time on. Every few milliseconds the profiler captures a snapshot of every running thread’s call stack. Aggregated over time, these snapshots form a flame graph — a visualization that makes it immediately obvious where CPU cycles are going, which functions are hot, and which library call is the unexpected bottleneck. Unlike traces, which tell you that a request was slow, a CPU profile tells you why: the exact code path consuming the most time. This makes profiling the natural next step after you’ve identified a latency problem in your trace data.
Why Profiling in Odigos?
Continuous profiling in Odigos is built directly into the same pipeline that handles your traces, metrics, and logs:- Zero-code changes. Profiling uses an eBPF kernel agent. No SDK, no agent library, no application restart required.
- Language-agnostic. Because collection happens at the OS level, every process on the node is profiled regardless of runtime — Java, Python, Ruby, PHP, Node.js, Perl, Erlang, .NET, Go, and Native executables (including C/C++, Rust, and Zig)
- No separate infrastructure. Profiles flow through the Odigos collector pipeline you already run. For quick investigation workflows the Odigos UI holds profiles in memory; for long-term storage you route them to a destination backend.
- Per-workload, on-demand. After enabling profiling, it works efficiently by only dynamically enabling in-memory buffer for workloads you are actively investigating, keeping Odigos resource usage predictable and under control.
How It Works
Odigos integrates the upstream OpenTelemetry eBPF Profiler as a receiver inside the Data Collector. Profiles are emitted as OTLP Profiles signals and travel through the same collector pipeline as your other telemetry. Odigos doesn’t require you to provision any data storage layer to view profiling data live, it does that by creating an efficient in-memory and per workload cache which maintains the lifecycle of profiling data based on cache size and TTL expiry.Learn More
Continuous Profiling in Odigos
Deep-dive blog post on how profiling was built into the Odigos pipeline.