On October 30, 2026, the OpenTelemetry project released version 1.0.0 of OpenTelemetry eBPF Instrumentation (OBI) – marking a milestone: observability for running applications without any code changes, without SDK integration, and without restarts. For DevOps teams, this is more than just a release: OBI 1.0 finally provides a stable, predictable foundation for instrumenting those services that have remained invisible – legacy applications, third-party binaries, or old Go services whose source code no one wants to touch.
From Beyla to OBI: The path to 1.0
OBI started as Grafana Beyla, which Grafana Labs donated to OpenTelemetry in 2025. After the first alpha at the end of 2025, the public beta followed at KubeCon EU 2026 in Amsterdam by Splunk – a pace that underscores the demand for zero-code instrumentation. An analysis by the CNCF Observability TAG from the first quarter of 2026 confirms this: according to it, 67 percent of production Kubernetes clusters already use at least one eBPF-based observability tool.
The idea behind it: OBI attaches eBPF probes directly to running processes and generates spans as well as RED metrics (Rate, Errors, Duration) for HTTP/S, HTTP/2, gRPC, Kafka, and database calls – without SDK, agent injection, or restart in the target service. The telemetry goes as OTLP to the OpenTelemetry Collector, so it can be integrated into existing backends like any other OTel data source.
What 1.0 specifically promises: Stability instead of features
The central promise of 1.0 is predictability. Configuration becomes a documented interface via Config v2: with a generated schema and the commands obi config validate and obi config migrate. Config v1 remains usable for compatibility reasons but is frozen and will no longer receive fixes. A documented compatibility agreement defines what to expect during upgrades and which environments are supported.
In addition, there is a formalized approach to telemetry evolution: OBI describes its data in a Weaver-compatible schema registry; distinguishing stable semantic conventions from stable data and fixing bugs in conformant telemetry is part of the migration promise. For teams operating dashboards and alerts, this means: upgrade paths can be planned instead of reacting to unannounced changes. Additionally, release processes have been hardened – signed artifacts, checksums, and software bills of materials are now standard.
More protocols, better traces, more context
Over the releases leading to 1.0, instrumentation coverage has grown significantly: new additions include MQTT, NATS, AMQP 1.0, MSSQL, Couchbase, Memcached, and Aerospike – a win for mixed stacks because database and messaging activity becomes visible through a single instrumentation approach. Improved HTTP/2 and gRPC propagation handles multiplexed traffic and stream contexts more cleanly, and route detection for frameworks like Django, FastAPI, Flask, Rails, .NET, and PHP distinguishes endpoints without turning every URL into its own metric series.
Practical for incident analysis is trace-log correlation: OBI can add trace and span IDs to emitted logs at runtime. Prerequisites: active trace context, elevated privileges, and for the usual write path a Linux kernel 6.0 or higher. Additionally, OBI provides runtime metrics for Go, Java, .NET, Node.js, and Python, as well as network measurements like TCP retransmissions – though runtime metrics are still considered experimental.
AI instrumentation without code
A unique selling point of OBI 1.0 is GenAI coverage: model providers, embeddings, reranking, vector retrieval, and MCP interactions can be observed, including tool calls and OpenAI-compatible gateways as well as native Ollama support. This makes dependencies on LLM services visible in the same traces as the rest of the application. Important: detailed payload analyses are deliberately not activated automatically – they require explicit enablement.
Where OBI belongs – and where it doesn't
OBI is not a replacement for SDKs, but a complement. It quickly closes blind spots: a baseline service map for the entire cluster in days instead of quarters, especially for legacy and third-party services. SDKs, in turn, provide business context, custom spans, and attributes – most mature environments operate hybrid. Two pitfalls to be aware of: context propagation (traceparent) is off by default and must be explicitly activated, otherwise distributed traces break apart. And OBI requires elevated privileges – typically a privileged DaemonSet with hostPID – as well as a kernel with BTF support. Processes that already emit their own OTel telemetry are skipped by OBI by default.
Conclusion
With OBI 1.0, OpenTelemetry closes a central gap in many observability strategies: teams no longer have to choose between "instrument everything" and "see nothing." To get started, a representative service or a legacy application whose traces were previously missing is enough – exported into the existing Collector, OBI takes effect immediately. At the same time, the SIG is already preparing the roadmap for 2027; anyone missing protocols or runtimes can influence prioritization. OBI 1.0 is thus a strong argument to reassess your own observability strategy: kernel-based instrumentation as a standard baseline, SDKs where business depth matters.
Sources
- OpenTelemetry eBPF Instrumentation reaches v1.0.0 – OpenTelemetry Blog, 30.10.2026
- Zero-code observability with OBI: OpenTelemetry's eBPF option – asksumanta.com, 25.10.2026
- eBPF Instrumentation (OBI) – OpenSearch Observability-Dokumentation
- OpenTelemetry eBPF Instrumentation (OBI) – The Complete Guide (KubeCon EU 2026 Beta, 1.0-Roadmap) – dev.to/ManoIT, 23.04.2026
Top comments (0)