Copy as Markdown[Open in ChatGPT](https://chatgpt.com/?q=Read%20https%3A%2F%2Fcoralogix.com%2Fdocs%2Fopentelemetry%2Finstrumentation-options%2Febpf-auto-instrumentation%2Frelease-notes.md%20and%20help%20me%20with%20my%20question%20about%20this%20Coralogix%20documentation%20page.)[Open in Claude](https://claude.ai/new?q=Read%20https%3A%2F%2Fcoralogix.com%2Fdocs%2Fopentelemetry%2Finstrumentation-options%2Febpf-auto-instrumentation%2Frelease-notes.md%20and%20help%20me%20with%20my%20question%20about%20this%20Coralogix%20documentation%20page.)

# OBI changelog

Behavior changes in each OBI ([OpenTelemetry eBPF Instrumentation](https://github.com/open-telemetry/opentelemetry-ebpf-instrumentation)) release that may affect existing dashboards, queries, span filters, and log parsers. Review the relevant version before upgrading. For current behavior, see the [OBI overview](https://coralogix.com/docs/opentelemetry/instrumentation-options/ebpf-auto-instrumentation/overview.md) and [Exported metrics](https://coralogix.com/docs/opentelemetry/instrumentation-options/ebpf-auto-instrumentation/exported-metrics.md).

Sep 4, 2026

OBI v0.13.0

* **`service.name` and `service.namespace` are no longer default metric labels.** They remain resource attributes, available through `target_info` and the OTLP resource, but they no longer decorate metric series by default. Dashboards, queries, and alerts that group or filter metrics by `service.name` or `service.namespace` must now select them explicitly through the `attributes->select` configuration. See [Exported metrics](https://coralogix.com/docs/opentelemetry/instrumentation-options/ebpf-auto-instrumentation/exported-metrics.md#attribute-selection-for-obi-metrics).
* **`error.type` values are unified across spans and metrics.** The attribute name is unchanged, but its values are now consistent between spans and metrics. Queries and span filters that match specific `error.type` values may need updating.
* **Python, JVM, and Node.js runtime metrics are expanded.** With the `application_runtime` feature enabled, Python services export CPython garbage-collection metrics, JVM services add class, thread, and CPU metrics alongside the existing memory metrics, and Node.js services add V8 garbage-collection and heap-space metrics. See [Exported metrics](https://coralogix.com/docs/opentelemetry/instrumentation-options/ebpf-auto-instrumentation/exported-metrics.md#application-metrics).
* **Database and messaging metrics carry additional attributes.** Database duration metrics gain `db.namespace`, `db.response.status_code`, and `server.port`; messaging spans gain `messaging.operation.name`.
* **Context-propagation and protocol reliability fixes.** A group of memory-safety fixes stops the HTTP header injector from writing a `traceparent` into an unrelated TLS stream (which could reset the connection), bounds message-buffer reads so unrelated kernel memory cannot appear in span payloads, and invalidates stale message buffers. Application-supplied `traceparent` headers are preserved in Go HTTP/2, TCP client and server roles are classified by the local port (fixing swapped `network.tcp.handshake.role` values for same-namespace clients), established idle connections are no longer counted in `obi.stat.tcp.failed.connections`, and DNS spans follow the OpenTelemetry conventions with IPv6 reverse-DNS support.

[Full release notes for OBI v0.13.0](https://github.com/open-telemetry/opentelemetry-ebpf-instrumentation/releases/tag/v0.13.0).

Aug 22, 2026

OBI v0.12.2

* **Log enrichment is limited to standard output and standard error.** Earlier versions intercepted every pipe write from an instrumented process, which could blank out shell command substitutions and, in some cases, stall shells. OBI now enriches (and suppresses the original of) only writes to a tracked process's standard output or standard error; any other pipe (a command substitution, application IPC, or a pipe on another file descriptor) passes through untouched. See [Trace-log correlation](https://coralogix.com/docs/opentelemetry/instrumentation-options/ebpf-auto-instrumentation/trace-log-correlation.md).
* **`application_span` and `application_span_sizes` span-metric features are deprecated.** They keep working and still emit the Grafana-convention `traces_spanmetrics_*` metrics, but OBI logs a deprecation warning at startup. Move to `application_span_otel`, which emits the OpenTelemetry-convention `traces_span_metrics_*` metrics. `application_span_sizes` has no direct replacement. The closest metrics are `http.server.request.body.size` and `http.server.response.body.size`. Enabling a legacy span-metric feature together with `application_span_otel` is rejected, and the `*` (all) selector keeps only the OTel version. See [Exported metrics](https://coralogix.com/docs/opentelemetry/instrumentation-options/ebpf-auto-instrumentation/exported-metrics.md).
* **Reliability and accuracy fixes.** HTTP/2 headers are decoded in capture order, preventing HPACK dynamic-table drift from attaching the wrong header values to a span; TLS connections that use in-memory BIOs are correlated again; and Aerospike responses that span multiple TCP reads can be reassembled, if you see missed or truncated Aerospike spans, raise the capture buffer with `ebpf.buffer_sizes.aerospike`.

[Full release notes for OBI v0.12.2](https://github.com/open-telemetry/opentelemetry-ebpf-instrumentation/releases/tag/v0.12.2).

Aug 20, 2026

OBI v0.12.1

OBI v0.12.0 was never published, v0.12.1 supersedes it and carries its changes.

* **Server-side database operations report `db.server.operation.duration`.** Redis, Memcached, and SQL measured on the server side now record `db.server.operation.duration` (`db_server_operation_duration_seconds` in Prometheus) instead of `db.client.operation.duration`. Update dashboards and alerts that read server-side database latency from the client metric. See [Exported metrics](https://coralogix.com/docs/opentelemetry/instrumentation-options/ebpf-auto-instrumentation/exported-metrics.md#application-metrics).
* **Four internal eBPF metrics dropped an incorrect counter suffix.** In OTLP, `obi.bpf.map.entries_total`, `obi.bpf.map.max_entries_total`, `obi.bpf.network.packets.total`, and `obi.bpf.network.ignored.packets.total` become `obi.bpf.map.entries`, `obi.bpf.map.max_entries`, `obi.bpf.network.packets`, and `obi.bpf.network.ignored.packets`. In Prometheus, the two map gauges lose the `_total` suffix (`obi_bpf_map_entries`, `obi_bpf_map_max_entries`); the packet counters keep it. These belong to the opt-in eBPF metrics family. See [Exported metrics](https://coralogix.com/docs/opentelemetry/instrumentation-options/ebpf-auto-instrumentation/exported-metrics.md#internal-obi-metrics).
* **Node.js event-loop runtime metrics.** With the `application_runtime` feature enabled, Node.js services export event-loop utilization, delay statistics (min, max, mean, standard deviation, and percentiles), and busy/idle time. See [Exported metrics](https://coralogix.com/docs/opentelemetry/instrumentation-options/ebpf-auto-instrumentation/exported-metrics.md#application-metrics).
* **Node.js manual spans can be captured without an SDK.** When a Node.js process creates spans through `@opentelemetry/api` but registers no OpenTelemetry SDK, set `nodejs.manual_spans: true` (env `OTEL_EBPF_NODEJS_MANUAL_SPANS`) to export them. It stays inactive when an SDK is already present.
* **Java service-name discovery follows the OpenTelemetry Java conventions.** For a Java process without an explicit service name, OBI derives one from the Spring Boot application name, then the JAR manifest `Implementation-Title`, then the JAR file name, before falling back to `java`.

[Full release notes for OBI v0.12.1](https://github.com/open-telemetry/opentelemetry-ebpf-instrumentation/releases/tag/v0.12.1).

Aug 17, 2026

OBI v0.11.0

* **Plain-text log enrichment is enabled by default.** Every service selected by the log enricher now receives space-separated `trace_id=… span_id=…` fields on non-JSON writes, which previously passed through unchanged. This can break downstream parsers for structured non-JSON log formats. To keep the previous behavior, set `ebpf.log_enricher.plain_text.enabled: false` before upgrading. JSON enrichment is unaffected. In Configuration v2 the same setting is `extensions.obi.correlation.log_trace_annotation.plain_text.enabled`. See [Trace-log correlation](https://coralogix.com/docs/opentelemetry/instrumentation-options/ebpf-auto-instrumentation/trace-log-correlation.md).
* **Direct OTLP consumers receive OpenTelemetry dot-notation metric names.** `target_info` is now `target.info`, `traces_target_info` is now `traces.target.info`, `traces_host_info` is now `traces.host.info`, `traces_span_metrics_calls_total` is now `traces.span.metrics.calls`, and `traces_span_metrics_duration` is now `traces.span.metrics.duration`. Consumers that scrape through a collector's Prometheus exporter are unaffected, because it maps the dots back to underscores.
* **Prometheus metric names are now derived from their OTLP definitions**, which corrected several of them. `obi_stat_tcp_retransmits` is now `obi_stat_tcp_retransmits_total` and `obi_stat_tcp_failed_connections` is now `obi_stat_tcp_failed_connections_total`; the CUDA GPU metrics lost an incorrect suffix, so `gpu_cuda_kernel_grid_size_total` is now `gpu_cuda_kernel_grid_size`, `gpu_cuda_kernel_block_size_total` is now `gpu_cuda_kernel_block_size`, and `gpu_cuda_memory_copies_bytes_total` is now `gpu_cuda_memory_copies_bytes`. Update Prometheus queries, dashboards, and alerts that key on the old names. See [Exported metrics](https://coralogix.com/docs/opentelemetry/instrumentation-options/ebpf-auto-instrumentation/exported-metrics.md).
* **JVM runtime metrics moved to the shared `application_runtime` feature.** The deprecated `application_jvm` feature alias and the `jvm_runtime_metrics.enabled` flag were both removed. Enable JVM and Go runtime metrics through `application_runtime` instead. The OBI-specific `obi.jvm.heap.used` metric (`obi_jvm_heap_used_bytes` in Prometheus) was dropped; read heap usage from `jvm.memory.used` filtered on `jvm.memory.type=heap`. See [Exported metrics](https://coralogix.com/docs/opentelemetry/instrumentation-options/ebpf-auto-instrumentation/exported-metrics.md).
* **Configuration v2 is available for standalone OBI and the OBI Collector receiver, and Configuration v1 remains supported.** `obi config validate` and `obi config migrate` check and convert existing configurations, failing closed on settings they can't preserve. Native Configuration v2 authoring uses a default-on capture policy: omitting `capture.policy.default_action`, or setting it to `include`, captures every process that no rule excludes. Canary the generated configuration before rolling it out, and see the [Config v1 to v2 migration guide](https://github.com/open-telemetry/opentelemetry-ebpf-instrumentation/blob/v0.11.0/devdocs/config/version-2.0/migration.md). The Helm chart values shown throughout these pages use Configuration v1.
* **Span names and span counts changed for SQL and Kafka.** SQL span names follow the OpenTelemetry semantic conventions more closely, SQL type is detected on Go SQL spans, and a multi-topic Kafka `Produce` or `Fetch` request now emits one span per topic instead of one span per request. Aggregations that count Kafka spans or group on SQL span names may shift.
* **Go Trace API auto-instrumentation is activated.** Go applications that use the global `otel.Tracer` API without registering an SDK are instrumented automatically. Go runtime telemetry also expands with CPU, memory, scheduler, GC, and histogram metrics. See [Go Trace API auto-instrumentation](https://coralogix.com/docs/opentelemetry/instrumentation-options/ebpf-auto-instrumentation/overview.md#go-trace-api-auto-instrumentation).
* **gRPC and HTTP/2 context propagation covers connections that were already open when OBI started.** Huffman-encoded `traceparent` values are decoded on ingress (which needs Linux kernel 5.17+), HTTP server context is preserved through response bodies, and Go `net/http` HTTP/1 clients no longer inject a duplicate `traceparent`. See [Distributed tracing](https://coralogix.com/docs/opentelemetry/instrumentation-options/ebpf-auto-instrumentation/distributed-tracing.md#how-trace-context-propagation-works).
* **Protocol, runtime, and AI/LLM coverage expanded.** Aerospike client calls are instrumented; Deno processes are identified as a JavaScript runtime and emit metrics and spans (trace-context propagation for Deno isn't available yet); and Ollama's native `/api/chat` and `/api/generate` endpoints and OpenAI-compatible gateways (LiteLLM, vLLM, LocalAI, OpenRouter, Ollama `/v1/`) are instrumented, with Gemini SSE streaming responses captured.

[Full release notes for OBI v0.11.0](https://github.com/open-telemetry/opentelemetry-ebpf-instrumentation/releases/tag/v0.11.0).

Jun 30, 2026

OBI v0.10.0

* **gRPC/HTTP2 network-level context propagation is now supported.** OBI propagates the W3C `traceparent` header on gRPC and HTTP/2 requests via HPACK header injection, without requiring memory-level writes. See [Distributed tracing](https://coralogix.com/docs/opentelemetry/instrumentation-options/ebpf-auto-instrumentation/distributed-tracing.md#how-trace-context-propagation-works).
* **GenAI attributes aligned with OpenTelemetry semantic conventions.** Message shapes for OpenAI, Anthropic, Google Gemini, Qwen, and AWS Bedrock have been normalized. Dashboards or queries keyed on prior attribute names may need updating.
* **`url.query` is now emitted by default on HTTP spans when present, and client `url.full` includes the query string.** Sensitive query parameters are automatically redacted. The built-in redaction list covers AWS SigV4 (`X-Amz-Signature`, `X-Amz-Credential`, `X-Amz-Security-Token`), legacy AWS signed URLs (`AWSAccessKeyId`, `Signature`, `SecurityToken`), Google (`X-Goog-Signature`, `sig`), and common token names (`token`, `access_token`, `refresh_token`, `id_token`, `jwt`). Extend the list with `OTEL_EBPF_SENSITIVE_QUERY_PARAMS_ADD`, or trim it with `OTEL_EBPF_SENSITIVE_QUERY_PARAMS_REMOVE`.
* **GraphQL documents are no longer exported by default.** Enable them explicitly with `payload_extraction.http.graphql.enabled: true` (env: `OTEL_EBPF_HTTP_GRAPHQL_ENABLED`).
* **Java route harvesting is disabled by default** while JVM-blocking edge cases are investigated. The `http.route` attribute on Java spans falls back to the raw URL path until harvesting is re-enabled upstream.
* **`ip` context-propagation value is deprecated; the undocumented `http` alias has been removed.** Valid values for `OTEL_EBPF_BPF_CONTEXT_PROPAGATION` are `disabled`, `headers`, `tcp`, and `all`. If you set `ip`, OBI warns and does not enable propagation. Use `headers` instead.
* **AI/LLM coverage expanded.** Tool-call extraction is added for OpenAI, Anthropic, Gemini, and Qwen; MCP tool-call arguments and results are extracted with independent attribute-selection controls; SSE streaming is captured for OpenAI-compatible, Anthropic, and Qwen responses; and vector-database retrieval spans are emitted for Pinecone, Qdrant, Milvus, Zilliz, Chroma, and Weaviate.
* **New protocol and platform support.** SunRPC/ONC RPC detection is added; automatic OpenShift cluster-name detection is added; JDK 26 and 27 are supported.

[Full release notes for OBI v0.10.0](https://github.com/open-telemetry/opentelemetry-ebpf-instrumentation/releases/tag/v0.10.0).

May 11, 2026

OBI v0.9.0

* **Database error response reporting is now opt-in.** OBI no longer captures database error response bodies by default. Enable it explicitly if you previously relied on those fields appearing on SQL spans.
* **Generic Go TLS support.** TLS visibility extends beyond OpenSSL to custom Go protocols. See [Distributed tracing](https://coralogix.com/docs/opentelemetry/instrumentation-options/ebpf-auto-instrumentation/distributed-tracing.md#supported-languages-for-context-propagation).

[Full release notes for OBI v0.9.0](https://github.com/open-telemetry/opentelemetry-ebpf-instrumentation/releases/tag/v0.9.0).

Apr 3, 2026

OBI v0.7.0+

* **SQL spans now use Server kind.** Database spans for SQL instrumentations (MySQL, PostgreSQL, MariaDB) are emitted as `SERVER` spans instead of `CLIENT`. Update any trace or span filters that key on span kind for SQL operations.
* **DNS excluded from span and service-graph metrics by default.** DNS calls no longer appear in span metrics or in the service graph. If you relied on DNS visibility there, expect those entries to disappear after upgrading.
* **Ordinal port guessing disabled by default for network metrics.** OBI no longer infers a service identity from ordinal source/destination ports on network flows. Existing queries built on guessed ports may return different results.
* **HTTP header enrichment available as an opt-in feature.** You can now capture selected HTTP request and response headers as span attributes. Configure under `ebpf.payload_extraction.http.enrichment` (env: `OTEL_EBPF_HTTP_ENRICHMENT_ENABLED`), disabled by default.

[Full release notes for OBI v0.7.0](https://github.com/open-telemetry/opentelemetry-ebpf-instrumentation/releases/tag/v0.7.0) and [v0.7.1](https://github.com/open-telemetry/opentelemetry-ebpf-instrumentation/releases/tag/v0.7.1).