OpenTelemetry Profiles Reaches Public Alpha, Establishing Profiling as the Fourth Observability Signal
The OpenTelemetry Profiling SIG has promoted its Profiles signal to public alpha, adding vendor-neutral continuous profiling alongside traces, metrics, and logs. Built on Elastic's donated eBPF agent, the new signal supports over a dozen language runtimes with minimal overhead.
Overview
The OpenTelemetry Profiling Special Interest Group announced on March 26 that the Profiles signal has officially entered public alpha, marking the first time the project offers a vendor-neutral standard for continuous production profiling. Profiles joins traces, metrics, and logs as the fourth core observability signal within the OpenTelemetry framework, a CNCF incubating project with more than 28,000 contributors across over 5,300 organizations.
The milestone follows roughly two years of development after Elastic donated its proprietary eBPF-based continuous profiling agent to the project, which now serves as the reference implementation within the OpenTelemetry ecosystem.
What the Profiles Signal Provides
Continuous profiling captures CPU and memory allocation data from running applications in production, enabling engineering teams to identify performance bottlenecks, reduce computation costs, and troubleshoot incidents without requiring code changes or service restarts. Unlike traditional profiling tools that sample on demand, continuous profiling runs persistently with minimal overhead.
The alpha release introduces a unified data format with deduplicated stack representations and dictionary tables for efficient encoding. The specification supports lossless round-trip conversion between OTLP Profiles and the widely used pprof format, ensuring backward compatibility with existing profiling toolchains.
Architecture and Runtime Support
At the core of the Profiles signal is the eBPF profiling agent, which operates at the Linux kernel level to capture stack traces across processes without requiring per-application instrumentation. The agent now functions as an OpenTelemetry Collector receiver, integrating directly with existing Collector processing pipelines for metadata enrichment and data transformation.
The alpha release supports a broad set of language runtimes including C/C++, Go, Rust, Python, Java, Node.js, .NET, PHP, Ruby, and Erlang/Elixir (BEAM). Notable improvements in this release include on-target Go symbolization, ARM64 support for Node.js V8, .NET 9 and 10 compatibility, and enhanced Ruby unwinding.
Three Collector components ship alongside the agent: a pprof receiver for importing existing profile data, a Kubernetes attributes processor for infrastructure metadata enrichment, and OTTL support for custom transformation rules.
Industry Backing
The Profiles alpha reflects contributions from engineers at Google, Datadog, Elastic, Grafana Labs, Red Hat, and Shopify, among more than 20 contributing organizations. Elastic characterized the release as the culmination of its continuous commitment to open-source profiling, noting that the donated agent now benefits from community-driven improvements beyond what any single vendor could deliver.
OpenTelemetry itself remains one of the highest-velocity open-source projects in the CNCF ecosystem, with a health score of 89 and an estimated software value of $332.7 million.
What Comes Next
The Profiling SIG’s roadmap toward beta and general availability includes signal correlation mechanisms that would allow engineers to jump directly from a trace span or log entry to the corresponding CPU profile, standardized symbolization APIs, and process-level context sharing between agents. Teams interested in evaluating the alpha can deploy the eBPF profiler with OTel Collector v0.148.0 or later.