The OpenTelemetry Collector Contrib contains everything in the opentelemetry-collector release, be sure to check the release notes there as well.

End User Changelog

🛑 Breaking changes 🛑

• pkg/fileconsumer: Move feature gate filelog.protobufCheckpointEncoding to beta and keep it enabled by default (#49387)

• receiver/oracledb: Enhanced Oracle SQL query plan details with runtime execution statistics. (#49329) This change requires the collector user to have access to V$SQL_PLAN_STATISTICS_ALL. Existing deployments that only grant access to V$SQL_PLAN may experience query plan collection failures until the appropriate privileges are granted.

  SQL query plan details are now retrieved from V$SQL_PLAN_STATISTICS_ALL instead of V$SQL_PLAN. Since V$SQL_PLAN_STATISTICS_ALL is a superset of V$SQL_PLAN, this change is backward compatible with respect to the emitted plan detail fields while enabling collection of additional runtime execution statistics, including OUTPUT_ROWS, LAST_OUTPUT_ROWS, LAST_ELAPSED_TIME, LAST_CR_BUFFER_GETS, LAST_CU_BUFFER_GETS, STARTS, and LAST_STARTS.

🚩 Deprecations 🚩

• connector/routing: Deprecate the request context in favor of otelcol.* OTTL paths. (#44762) The routing connector request context and request["key"] condition syntax are deprecated in favor of otelcol.client.metadata["key"][0] for HTTP/client metadata and otelcol.grpc.metadata["key"][0] for gRPC metadata. These paths work in all signal contexts and are documented in the OTelCol OTTL context. A warning is logged when the request context is still configured.

• exporter/datadog: The exporter.datadogexporter.metricremappingdisabled feature gate no longer has any effect when the serializer exporter path is active (the default). Use exporter.datadogexporter.DisableAllMetricRemapping to disable metric remapping. (#49320) See DataDog/datadog-agent#51232.

• receiver/aws_cloudwatch: rename to aws_cloudwatch with deprecated alias awscloudwatch (#45339)

🚀 New components 🚀

• processor/dynamic_sampling: Add dynamic sampling processor for adaptive trace sampling with W3C TraceState rate encoding. (#48898)

💡 Enhancements 💡

• cmd/opampsupervisor: Make the OpAMP Supervisor logger encoding configurable so it can emit console (plain-text) logs as well as JSON (#47532)

• connector/routing: Add OTTL context inference support to routing connector. (#38080) The routing connector now supports OTTL context inference, allowing users to write clearer routing conditions using context-qualified paths. Instead of relying on implicit context resolution, users can explicitly specify which context's attributes they want to access (e.g., resource.attributes["env"] or span.attributes["http.method"]). Unqualified paths like attributes["key"] continue to work and default to the resource context for backward compatibility.

• exporter/alibabacloud_logservice: Add security_token (STS) support alongside AK/SK; preserve ECS role/token-file fallback. (#48624) When security_token is provided in the exporter configuration the SDK is configured to use static STS credentials via the provider. The existing ecs_ram_role and token_file_path behavior remains the higher-priority authentication method and is unchanged.

• exporter/clickhouse: Add experimental support for the profiles signal (#49304)

• exporter/datadog: Emit otel.datadog_exporter.metrics.running.fargate{task_arn} for AWS ECS Fargate workloads so that each workload has its own metric. (#49042) Host-based workloads continue to use otel.datadog_exporter.metrics.running{host} unchanged. otel.datadog_exporter.metrics.running will no longer be tagged with task_arn. Fargate workloads will only use otel.datadog_exporter.metrics.running.fargate{task_arn}.

• exporter/datadog: Fargate workloads now emit a dedicated otel.datadog_exporter.metrics.running.fargate metric tagged with task_arn instead of sharing otel.datadog_exporter.metrics.running with host-based workloads. (#49320) Host-based workloads continue to emit otel.datadog_exporter.metrics.running{host} unchanged. See DataDog/datadog-agent#52203.

• exporter/elasticsearch: Support routing from data_stream attributes defined as scope attributes (#49306)

• extension/azure_auth: Promote the Azure Authenticator extension to beta stability. (#48521)

• extension/file_storage: Added max_size support to the filestorage extension to cap per-component database growth. (#38620)

• extension/sigv4auth: Add automatic service detection for CloudWatch OTLP metrics endpoint (monitoring..amazonaws.com) (#48738)

• pkg/translator/pprof: reduce allocations in sample conversion (#49452)

• processor/transform: Add exemplar context support to the transform processor, allowing metric_statements to read and modify exemplar fields on metric datapoints. (#49022)

• receiver/aws_lambda: Add support for AWS configuration overrides for static credentials (#49163)

• receiver/faro: Emit per-payload counters for ingested Faro logs, measurements, exceptions, and events (#48071) Adds otelcol_faro_log_ingested_total, otelcol_faro_measurement_ingested_total, otelcol_faro_exception_ingested_total, and otelcol_faro_event_ingested_total. Each counter is incremented by the number of entries in the corresponding payload section after the request body is parsed.

• receiver/mongodb: Add Query Sample collection to the MongoDB receiver, emitting a db.server.query_sample log event for each currently executing operation. (#48573) Samples are collected via $currentOp, with idle connections and administrative commands filtered out. Each event includes obfuscated query text and operation metadata. The logs signal is at development stability; attribute names may change until OTel db.server.query_sample conventions stabilize. The metrics signal remains at beta.

• receiver/oracledb: Add service.name and service.namespace opt-in resource attributes and allow overriding any resource attribute via override_value. (#47088)

• receiver/oracledb: Add workload analysis metrics related to scans, enqueue, LOB, parse, sort, cursor, and session activity (#48808) Adds fourteen new opt-in metrics (disabled by default):

  • oracledb.call.count, oracledb.call.recursive.cpu.time
  • oracledb.cursor.cache.hits, oracledb.cursor.cache.size, oracledb.cursor.open
  • oracledb.db.time
  • oracledb.enqueue.operations
  • oracledb.lob.operations
  • oracledb.parse.cpu.time, oracledb.parse.elapsed.time
  • oracledb.scan.count, oracledb.scan.table.rows
  • oracledb.sort.operations, oracledb.sort.rows No new SQL queries are issued; the existing v$sysstat scrape already returns all required rows, so the receiver adds no additional load on the monitored Oracle instance.

• receiver/oracledb: Add buffer cache and Database Writer (DBWR) metrics (#49061)

• receiver/oracledb: Add redo log metrics (#49060)

• receiver/postgresql: Add default service.name and service.namespace resource attributes (#47087) Both attributes default to enabled: false. When enabled, service.name defaults to unknown_service:postgresql and service.namespace defaults to an empty string. Additionally, override_value is enabled for all resource attributes, allowing users to set custom values via override_value in the collector configuration.

• receiver/postgresql: Adds blocking session and lock attributes to db.server.query_sample events for PostgreSQL blocking session detection. Blocking attributes are always emitted (empty string / 0 when not blocked). (#49028) New attributes under postgresql.blocking.*: postgresql.blocking.pids, postgresql.blocking.start_time, postgresql.blocking.wait_duration, postgresql.blocking.lock.mode, postgresql.blocking.lock.type, postgresql.blocking.lock.relation, postgresql.blocking.transaction.start_time. postgresql.blocking.pids contains the full array of blocking PIDs from pg_blocking_pids() (e.g. {5121,5122}). postgresql.blocking.start_time uses pg_locks.waitstart (PostgreSQL 14+) to record when the lock wait began. waitstart is stable for the full duration of the wait — it is set once when the lock wait starts and is not affected by blockers joining or leaving. postgresql.blocking.transaction.start_time provides a UTC timestamp (RFC3339) from xact_start representing when the transaction started, predating the lock contention itself. Idle-in-transaction blocker sessions (holding locks without an active request) are captured via pg_blocking_pids subquery in the WHERE clause. All blocking attributes are always present on every db.server.query_sample event; empty string / 0 values indicate no active blocking. Requires PostgreSQL 14 or later.

• receiver/prometheus: Promote receiver.prometheusreceiver.IgnoreScopeInfoMetric feature gate to beta. (#47312) The otel_scope_info metric is now ignored for scope attribute extraction by default. To temporarily restore the previous behavior, disable the feature gate with --feature-gates=-receiver.prometheusreceiver.IgnoreScopeInfoMetric.

• receiver/sqlserver: Add access methods and buffer pool metrics (#49182) Adds the following metrics, all disabled by default:

  • sqlserver.access.scan.rate
  • sqlserver.extent.operation.rate
  • sqlserver.ghost_record.skipped.rate
  • sqlserver.page.allocation.rate
  • sqlserver.page.compression.rate
  • sqlserver.page.read_ahead.rate
  • sqlserver.scan_point.revalidation.rate
  • sqlserver.worktable.cache.hit_ratio

• receiver/syslog: Add support for RFC6587 Octet Counting framing with RFC3164 message format. (#45216)

🧰 Bug fixes 🧰

• connector/routing: Fix duplicate routing table entry handling to correctly ignore duplicates instead of overwriting the original route (#44762) Previously, when duplicate routing table entries were encountered, the warning message indicated that the duplicate would be ignored, but in fact the original entry was being overwritten by the duplicate. This fix ensures that duplicates are properly ignored and the original route's consumer is preserved.

• exporter/awsxray: Use local service name for Consumer local root segment names, falling back to resource service name when aws.local.service is absent. (#43432) Consumer spans promoted to X-Ray Segments after #42633 were named using the span (operation) name instead of the local service name. The fix mirrors the existing Server span naming behavior.

• exporter/datadog: Compress the metric sketches payload with gzip in the legacy metric API client, matching the metric series payload. Previously sketches were sent uncompressed, increasing egress for distribution/histogram metrics. (#49313)

• exporter/datadog: Fix CPU core count fields (cpu_cores, cpu_logical_processors) missing from host metadata payloads for remote hosts in OTel gateway topologies. (#49320) CPU data arrives via system.cpu.physical.count / system.cpu.logical.count metrics and was never forwarded to the host metadata reporter in the serializer exporter path. See DataDog/datadog-agent#51893.

• exporter/datadog: Fix host metadata collection no longer aborts the entire batch when one resource fails; errors are accumulated and processing continues for remaining resources (#49153) See DataDog/datadog-agent#51893.

• exporter/datadog: Metrics from the Podman stats receiver now report the correct Datadog metric origin (opentelemetry_collector_podmanreceiver) instead of falling through to opentelemetry_collector_unknown. (#49320) See DataDog/datadog-agent#52594.

• exporter/file: Fix path traversal vulnerability on Windows when using grouping file exporter (#49194)

• exporter/opensearch: Preserve LogRecord EventName field in all log encoding modes. (#49314)

• extension/datadog: Skip cloud metadata source provider probe when hostname is already set in config, avoiding spurious GCP metadata server requests on non-GCP hosts. (#49241)

• pkg/translator/pprof: Wrong conversion of labels from pprof to OTLP profiles (#49449)

• processor/cumulativetodelta: Align reset detection and dropping on histograms and exponential histograms (#48278) Histograms and exponential histograms where there is a drop in total count or on any bucket counts are now treated as a reset and dropped.

• processor/cumulativetodelta: Fix non-deterministic order of valid metric types in error messages for invalid metric_types config values. (#49120)

• processor/gen_ai_normalizer: Fix OpenInference message normalization for flattened indexed attributes (#48421) Reconstruct gen_ai.input.messages and gen_ai.output.messages from OpenInference flattened indexed span attributes (llm.{input,output}_messages.N.message.*).

• processor/k8s_attributes: Fix cache key memory leak in k8sattributesprocessor when a Pod's IP is missing or cleared from the delete event (#48986) When a Pod is deleted, if its IP address is missing or already cleared from the delete event status, the processor now looks up the cached pod by its UID to retrieve the stored IP and correctly queue all associated keys for deletion.

• processor/resource_detection: Fix Docker resource detection failing to start when optional resource attributes are disabled, including container attributes when the hostname does not match a Docker container name. (#46275)

• processor/spanpruning: Fix non-deterministic depth of preserved outlier spans by grouping leaves and parents by tree depth, so same-named ancestors at different depths are no longer merged. (#49323)

• receiver/aws_cloudwatch: Fix logs being skipped when AWS pagination returns an empty page (#48577)

• receiver/github: Capture the datapoint timestamp per repository instead of once at scrape start, so per-repository metrics carry a timestamp close to when each repository was observed. (#48497)

• receiver/prometheus: Prometheus API server config/targets stay in sync with Target Allocator changing it if both are enabled. (#48040) Previously, the API server config and targets did not get the updates from the Target Allocator.

• receiver/tcp_log: Fix tcp input operator emitting partial log data after a connection read error when one_log_per_packet is enabled (#49199) Previously, if a TCP connection was reset mid-read while one_log_per_packet was set to true, the partial buffer was still processed and emitted as a log entry. Now the read error causes the message to be discarded instead.

API Changelog

🛑 Breaking changes 🛑

• all: handle breaking change in pdata/pprofile (#49373)
• pkg/ottl: Switch LambdaExpression arguments to be pointers. (#49420) These values are intended to be mutable and are passed by reference in normal use, so this keeps symmetry between the argument declaration and their use by the function.

💡 Enhancements 💡

• pkg/ottl: Add ottl.NewTestingLambdaExpression and internal helpers for implementing lambda-based OTTL functions. (#49180) Adds ottl.NewTestingLambdaExpression, a test helper that builds a LambdaExpression with a custom value body. The callback receives resolveBinding func(string) any to read formal parameters from the active local scope during evaluation.

  Also introduces internal helpers to support lambda-based OTTL functions:

  • funcutil.GetSliceOrMapValue coerces a getter result to pcommon.Slice or pcommon.Map.
  • funcutil.EvaluateBiFunction and funcutil.EvaluateBiPredicate bind two positional arguments, normalize them, and evaluate a LambdaActivation.
  • ottlcommon.NormalizeValue converts values, including pcommon.Value, into types used by OTTL comparisons.

• pkg/ottl: Add String() method to ottl.Statement, ottl.Condition, and ottl.ValueExpression that return the original OTTL text used during parsing. (#49415)

• processor/transform: Add exemplar context support to the transform processor, allowing metric_statements to read and modify exemplar fields on metric datapoints. (#49022)

We are thrilled to welcome our first-time contributors to this project. Thank you for your contributions @pgimalac, @gcacace, @avy252004, @yuminn-k, @nissessenap, @aniket866, @kathiehuang ! 🎉