open-telemetry/opentelemetry-collector-contrib v0.165.0

v1.4.0/v0.165.0

on GitHub

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

Unmaintained Components

These components are marked as Unmaintained and will eventually be removed from our community distributions. If you depend on these components we need your help to support them.

• exporter/logicmonitor
• receiver/riak

End User Changelog

🛑 Breaking changes 🛑

• all: Remove the deprecated SignalFx receiver. Use the OTLP receiver instead. (#51934)

• connector/service_graph: Remove deprecated feature gates connector.servicegraph.legacyLatencyMetricNames and connector.servicegraph.legacyLatencyUnitMs (#52246)

• exporter/signalfx: Stop producing disk_ops.total and disk_ops.pending metrics with the default translation rules. (#52729)
  Use system.disk.operations instead of disk_ops.total. It is sent by default as a cumulative counter, so the per-period
  number of operations on a host can be charted with SignalFlow, for example:

  data('system.disk.operations', rollup='delta').sum(by=['host.name'])
  

  Use system.disk.pending_operations instead of disk_ops.pending. Both are excluded by default, so replace
  disk_ops.pending with system.disk.pending_operations in include_metrics:

  exporters:
    signalfx:
      include_metrics:
        - metric_name: system.disk.pending_operations

• pkg/coreinternal: Promote internal.coreinternal.goldendataset.EmitV1DatabaseConventions and internal.coreinternal.goldendataset.DontEmitV0DatabaseConventions feature gates from Alpha to Beta. (#52690)
  The gates are now enabled by default, so goldendataset generates db.system.name (semconv v1.40.0)
  instead of the deprecated db.system (semconv v1.28.0). The migration remains reversible by disabling
  the gates.

• pkg/coreinternal: Promote internal.coreinternal.goldendataset.EmitV1NetworkConventions and internal.coreinternal.goldendataset.DontEmitV0NetworkConventions feature gates to Stable. (#52503)
  The gates can no longer be disabled. goldendataset now always generates network.local.address,
  network.peer.address, and server.address, and no longer generates the deprecated net.host.ip,
  net.peer.ip, http.host, and http.server_name attributes.

• pkg/coreinternal: Promote internal.coreinternal.goldendataset.EmitV1RPCConventions and internal.coreinternal.goldendataset.DontEmitV0RPCConventions feature gates to Stable. (#52219)
  The gates can no longer be disabled. goldendataset now always generates rpc.method and
  service.peer.name, and no longer generates the deprecated rpc.service and peer.service
  attributes.

• pkg/jaeger: Promote "pkg.translator.jaeger.EmitV1HttpConventions" and "pkg.translator.jaeger.DontEmitV0HttpConventions" feature gates to Beta. (#52146)
  The Jaeger translator now emits http.response.status_code instead of the deprecated http.status_code by default.

• pkg/zipkin: Promote "pkg.translator.zipkin.EmitV1CloudResourceConventions" and "pkg.translator.zipkin.DontEmitV0CloudResourceConventions" feature gates to Beta. (#52148)
  The Zipkin translator now emits cloud.resource_id instead of the deprecated faas.id by default.

• pkg/zipkin: Promote "pkg.translator.zipkin.EmitV1NetworkConventions" and "pkg.translator.zipkin.DontEmitV0NetworkConventions" feature gates to Stable. (#52250)
  The gates can no longer be disabled. The Zipkin translator now always emits network.local.address,
  network.peer.address, and service.peer.name, and no longer emits the deprecated net.host.ip,
  net.peer.ip, and peer.service attributes.

• processor/resource_detection: The system and docker detectors now set os.type to zos instead of the deprecated z_os on z/OS. (#52804)

• processor/transform: Remove the deprecated flatten_data config option. Use flatten.log instead. (#32080)

• receiver/file_log: Remove stable gate filelog.protobufCheckpointEncoding (#52387)

• receiver/kafka_metrics: Remove stable receiver.kafkametricsreceiver.UseFranzGo gate (#52241)

• receiver/prometheus: __scheme__ and __metrics_path__ labels on scraped samples are no longer dropped from datapoint attributes. (#52736)

🚩 Deprecations 🚩

• connector/routing: Stabilizing gate connector.routing.defaultErrorModeIgnore (#52352)
• processor/resource_detection: The system detector adds the os.build_id resource attribute and deprecates os.build.id. (#52806)
• processor/resource_detection: Deprecate the azure.resourcegroup.name resource attribute in favor of azure.resource_group.name, behind the processor.azuredetector.RenameResourceGroupAttribute feature gate (#49079)
  The new processor.azuredetector.RenameResourceGroupAttribute feature gate is alpha
  and disabled by default, so the azure detector continues to emit azure.resourcegroup.name and
  there is no change in behavior. Enabling the gate makes the detector emit
  azure.resource_group.name instead, matching the naming convention used by other semantic
  convention attributes such as cloud.availability_zone. The gate is intended to be promoted to
  beta, and enabled by default, in a future release, giving users time to migrate any
  configuration that filters or routes on azure.resourcegroup.name.

💡 Enhancements 💡

• connector/datadog: Use url.template for HTTP client span resource names when it is present. (#52553)
  From updated github.com/DataDog/datadog-agent modules: with the default resource name logic, HTTP client spans get resource names like GET /users/{id} from url.template, and otherwise just the method (DataDog/datadog-agent#54496).

• connector/span_metrics: Reduce consume-path allocations by reusing shared metric keys, avoiding per-event attribute map copies, and filtering resource key attributes without a full map copy. (#49860)
  When calls_dimensions and histogram.dimensions are both empty, the duration
  histogram reuses the calls metric key. Delta timestamp maps are allocated only
  for delta temporality. With events.enabled, dimensions are resolved from span,
  event, and resource attributes without building a merged map per event. When
  resource_metrics_key_attributes is set, only the configured attributes are
  copied into the resource key hash.

• exporter/azuredataexplorer: Restore the stability of the exporter from unmaintained to beta (#52282)

• exporter/clickhouse: Size the ClickHouse client connection pool from sending_queue.num_consumers by default. (#52350)
  The metrics exporter uses one connection per metric type per consumer.
  max_open_conns and max_idle_conns set in endpoint or connection_params still take precedence.

• exporter/clickhouse: Change the default column codec in the base table schemas from ZSTD(1) to ZSTD(3). (#52705)
  Only applies to newly created tables, existing tables are not altered.

• exporter/datadog: Use url.template for HTTP client span resource names when it is present. (#52553)
  From updated github.com/DataDog/datadog-agent modules: with the default resource name logic, HTTP client spans get resource names like GET /users/{id} from url.template, and otherwise just the method (DataDog/datadog-agent#54496).

• exporter/datadog: Map the instrumentation scope name and version on logs to otel.scope.name and otel.scope.version. (#52553)
  From updated github.com/DataDog/datadog-agent modules: the deprecated otel.library.* attributes are remapped to the new names (DataDog/datadog-agent#52945).

• exporter/datadog: Add the alpha feature gate datadog.serializerexporter.EnableGaugeInterval to use the datadog.gauge.interval attribute as the gauge reporting interval. (#52553)
  From updated github.com/DataDog/datadog-agent modules (DataDog/datadog-agent#58408).

• exporter/elasticsearch: Add cumulative temporality support for histograms and exponential histograms (#52547)
  Also adds a temporality field to serialized histograms and exponential
  histograms documents in OTel mode.

• exporter/load_balancing: Route logs without service.name by a hash of their resource attributes with routing_key: service, instead of sending all of them to one backend. (#52483)
  Logs without service.name now spread over several backends instead of one. Setups with stateful processing behind the exporter will see a one-time rebalance after the upgrade.

• exporter/opensearch: Add opt-in ISM rollover and custom index-template mapping support for otel-v1 mode. (#48585)
  mapping.ism.enabled provisions an Index State Management rollover policy and an initial
  write-aliased index on startup so otel-v1 indices roll over by size/age. mapping.index_template_file
  deep-merges a user-supplied JSON overlay over the built-in otel-v1 template, letting operators uplift
  attributes to typed fields or keep them un-indexed to avoid mapping explosion. Both options are
  best-effort and only valid with mapping.mode: otel-v1.

• exporter/prometheus: Add the pkg.prometheus.KeepIdentifyingResourceAttributes feature gate to preserve service.name, service.namespace, and service.instance.id on target_info. (#48890)

• exporter/prometheus_remote_write: Add the pkg.prometheus.KeepIdentifyingResourceAttributes feature gate to preserve service.name, service.namespace, and service.instance.id on target_info. (#48890)

• exporter/prometheus_remote_write: Add otelcol_exporter_prometheusremotewrite_wal_send_failed_batches and otelcol_exporter_prometheusremotewrite_wal_send_failed_samples to count batches the write-ahead log drops after the remote endpoint rejects them (#51710)

• exporter/pulsar: Enable FreeBSD/amd64 builds for the Pulsar exporter. (#15781)

• exporter/splunk_hec: Map eBPF CPU and allocation profiles to AlwaysOn Profiling and count dropped unsupported profile records. (#52218)

• extension/aws_iam_db_auth: This adds a config schema generated via schemagen to this component. (#52836)

• extension/aws_logs_encoding: Promote extension.encoding.awslogsencoding.DontEmitV0RPCConventions feature gate to beta. (#52432)
  The CloudTrail log unmarshaler no longer emits the deprecated rpc.system and rpc.service (semconv v1.38.0) attributes by default; only rpc.system.name (semconv v1.40.0) is emitted. Disable extension.encoding.awslogsencoding.DontEmitV0RPCConventions to restore the deprecated attributes.

• extension/azure_auth: Implement the database credential provider interface for Azure authentication extensions. (#50294)

• extension/bearertokenauth: Expose the configured token as an OAuth2 token so the extension can be used as a token source (#50574)
  This allows bearertokenauth to be referenced as the oauthbearer_token_source of the Kafka
  exporter and receiver, so a SASL/OAUTHBEARER token can be read from a file that is populated
  out of band, such as a Kubernetes projected service account token.

• extension/google_client_auth: Migrate credential detection to cloud.google.com/go/auth to support X.509 certificate-based Workload Identity Federation (WIF). (#52488)

• pkg/faro: Add a timestamp_micros key with the epoch microseconds to the log body when a Faro timestamp has sub-millisecond precision. (#52357)

• pkg/faro: Translate the new Faro mobile metadata fields in meta.device, meta.os, and meta.network to and from logs. (#52314)

• pkg/kafka/configkafka: Add producer::max_inflight_requests_per_broker to configure the number of produce requests kept in flight per broker. (#52780)
  Defaults to 1, preserving the previous behavior. Only takes effect with required_acks 0 or 1;
  values above 1 are rejected with required_acks: all.

• pkg/ottl: Use substring matching for literal-only patterns in IsMatch, delete_matching_keys, keep_matching_keys, replace_pattern and replace_all_patterns without changing regex semantics. (#52593)

• pkg/ottl: Allow converters and paths as keys in map literals, e.g. {ToUpperCase("key"): "value"}. (#52210)
  Enable the pkg.ottl.enableDynamicMapKeys feature gate to use this feature.

• pkg/ottl: Improve StandardStringLikeGetter performance by adding direct string conversion for int64 and bool values, avoiding unnecessary JSON marshaling. (#52596)

• pkg/ottl/contexts/xspanlink: Add ottlspanlink context, in the new experimental xspanlink module, exposing per-span-link fields (trace_id, span_id, trace_state, attributes, dropped_attributes_count, flags) for use in OTTL statements. (#26660)

• pkg/ottl/xottl: Add the experimental ToValidUTF8 converter to replace invalid UTF-8 byte sequences in a string with a replacement string (#51717)

• pkg/stanza: Add an otelcol operator that parses the collector's own self-logs, promoting ts, level, msg, and resource onto the log record's Timestamp, Severity, Body, and Resource. (#49525)

• processor/aws_ecs_attributes: Add the ECS metadata enrichment logic that attaches AWS ECS attributes to logs, metrics, traces and profiles. (#44476)

• processor/log_dedup: Add timestamp_mode configuration option (aggregated (default) or preserved) and new attributes first_event_timestamp and last_event_timestamp to the logdedup processor. (#43647)

  • aggregated sets the emitted log's timestamp to export time (default behavior) and adds first_event_timestamp to capture the original timestamp.
  • preserved retains the timestamp from the first log received in the aggregation window (omitting redundant first_event_timestamp).
  • last_event_timestamp records the timestamp of the last log received in the window in all modes.

• processor/redaction: Adds the masking_string configuration to allow for configurable masking (#50801)

• processor/redaction: Reject hash_function values md5 and sha1 at startup when the collector runs in FIPS 140-3 mode. (#52501)
  MD5 and SHA-1 are not FIPS 140-3 approved. A collector built with the Go FIPS module and run with
  GODEBUG=fips140=on would mark itself non-compliant on the first masked value and GODEBUG=fips140=only
  would panic. The processor now fails configuration validation in FIPS mode. Nothing changes outside FIPS mode.

• processor/resource_detection: Trimmed builds no longer link dependencies for resource detectors that are not compiled (#47218)

• processor/resource_detection: Keep Kubernetes and OpenShift clients out of resource detector configuration imports. (#47218)

• processor/resource_detection: Add opt-in host.id attribute to the k8s_api detector, set from the node's status.nodeInfo.machineID. (#52194)

• processor/resource_detection: Add the processor.resourcedetection.openshift.removeCloudNameRegion feature gate to stop the openshift detector from reporting cloud.region on Azure and OpenStack clusters. (#48159)
  The value was the Azure cloud environment name (e.g. azurepubliccloud) or the name of the OpenStack clouds.yaml entry, not a region. The openshift detector now uses the go-contrib SDK detector; while the gate is disabled the legacy value is still reported. It will be removed once the gate is stable.

• processor/rolling_span_latency: Add span_kinds and always_include_root_spans options to scope which spans are tracked and labeled. (#52061)

• processor/rolling_span_latency: This adds a config schema generated via schemagen to this component. (#52837)

• processor/transform: Add the processor.transform.enableExperimentalFunctions feature gate, which new transform processor functions will require while they are experimental. (#52523)

• processor/verify: Add cryptographic log record integrity verification to verifyprocessor. (#51383)

• receiver/flink_metrics: Add opt-in flink.job.id resource attribute to uniquely identify a job submission on job, task and operator metrics. (#49624)
  The value is the jid already returned by the Flink REST API. The attribute is disabled by default.

• receiver/google_cloud_spanner: Update operations_by_table_json_string label in top minute transaction stats from raw string to operations_by_table_json value_type so table operations are parsed and formatted as {table_name,insert_or_update_count,insert_or_update_bytes}. (#52267)

• receiver/google_cloud_spanner: Add user split points metrics (user_split_points/split_points_count) with table_name, index_name, initiator, split_key, and expire_time labels (including hide_split_stats_keys PII masking support for split_key) in googlecloudspannerreceiver. (#52151)

• receiver/google_cloud_spanner: To support both GoogleSQL and PostgreSQL dialects, the sample_lock_requests metric label now reads from the SAMPLE_LOCK_REQUESTS_JSON_STRING column in SPANNER_SYS.TOP_MINUTE_LOCK_STATS. (#51637)

• receiver/google_cloud_spanner: Add remaining labels (start_time, query_text_fingerprint, query_text, query_text_truncated, session_id, query_id, api_client_header, user_agent_header, server_region, priority, transaction_type) and truncate_text support to oldest_active_queries metrics to help identify and troubleshoot long-running active queries. (#52257)

• receiver/google_cloud_spanner: Add active partitioned DMLs metrics (num_partitions_total, num_partitions_complete, num_trivial_partitions_complete, rows_processed) to monitor the progress and execution of currently running Partitioned DML statements. (#52254)

• receiver/googlecloudmonitoring: Add support for GCP metrics with value type DISTRIBUTION and metric kind CUMULATIVE, converted to histograms with cumulative aggregation temporality. (#39600)

• receiver/host_metrics: Suppress warnings on Windows for not ready removable disks (#52723)

• receiver/kafka: Pass message metadata to opt-in message encoding extensions for all signals. (#51752)

• receiver/mongodb_atlas: Set the User-Agent header on MongoDB Atlas API requests to the collector name and version. (#52742)

• receiver/mysql: Add mysql.wait_event.type and mysql.wait_event.name attributes to db.server.query_sample, classifying the current wait as io, lock, synch, or CPU/User sleep/other. (#51432)

• receiver/oracledb: Add oracledb.wait.start_time attribute to db.server.query_sample event to capture the timestamp when the current wait started. (#52770)

• receiver/postgresql: Add the postgresql.backend.type attribute to the db.server.query_sample event. (#52615)

• receiver/postgresql: Extract the trace context of db.server.query_sample records from a SQLCommenter traceparent comment in the sampled query. (#52227)
  A traceparent in application_name keeps precedence.

• receiver/prometheus: Introduce alpha feature gate receiver.prometheus.UseTargetUrlForServerAddress to populate server.address, server.port, and url.scheme from post-relabeling __address__ and __scheme__ target labels instead of parsing instance and pre-relabeling service discovery labels. (#52565, #52570)

• receiver/prometheus_simple: Log a warning at start naming the tls settings that cannot be forwarded to the Prometheus scrape config when the target is scraped over https. (#52081)

• receiver/receiver_creator: Improve extensibility of receivercreator rules by allowing use of custom observer.EndpointDetails "types" (#50548)
  The receivercreator configuration validation no longer requires that the rule's type == <endpoint_details>
  prefix match one of the observer.EndpointDetails "types" defined in this repository. Now it only enforces the syntax requirements.
  The old enforcement didn't allow downstream distributions to create their own observer.EndpointDetails maps
  making harder to developer new observers that could be used with the receivercreator.

• receiver/sql_query: Support Microsoft Entra authentication for Azure SQL and Azure Database for PostgreSQL, including managed identities. (#50294)

• receiver/sql_server: Support database credential providers through db_auth. (#50294)

• receiver/sql_server: Add databases and exclude_databases options to limit the databases that metrics and events report on. (#52529)

• receiver/udp_log: Add resolve_names option to allow disabling reverse DNS lookups when network attributes are enabled. (#52603)

🧰 Bug fixes 🧰

• cmd/opampsupervisor: Stop losing the last lines of Collector output when passthrough logging is enabled. (#52675)
  The Supervisor read the Collector's stdout and stderr through pipes that exec.Cmd.Wait closes as soon
  as the process exits, so output still in flight, such as a final error line before a crash, could be dropped.

• connector/datadog: Update go-sqllexer to v0.2.6 for SQL obfuscation. (#52553)
  Fixes PostgreSQL strings containing backslashes being truncated, MySQL db . table identifiers being altered, and Db2 literals with backslashes (DataDog/datadog-agent#57039, DataDog/datadog-agent#58950).

• connector/datadog: Fix several APM stats computation issues. (#52553)
  From updated github.com/DataDog/datadog-agent modules:

  • traces.ignore_resources now matches as a regular expression, as documented, instead of by exact name (DataDog/datadog-agent#58619).
  • Spans whose stats were already computed by the tracer (_dd.stats_computed) are skipped, so they are no longer counted twice (DataDog/datadog-agent#54280).
  • Malformed spans are normalized before stats are computed (DataDog/datadog-agent#56205).
  • Spans with timestamps far in the future are counted in the current stats bucket instead of creating future buckets (DataDog/datadog-agent#56608).

• connector/span_metrics: Fix delta timestamp collisions between resources with matching dimensions, preventing downstream cumulative counters from resetting. (#52145)
  Metric names, attributes, and delta values are unchanged.
  When unset, metric_timestamp_cache_size now defaults to 10 times resource_metrics_cache_size, with a minimum of 1000 (10000 with default settings).
  Users who explicitly set metric_timestamp_cache_size may need to increase it because cache entries now include resource identity.
  The "Evicting cached delta timestamp" log level changes from Info to Debug.

• exporter/awss3: Wrap the AssumeRoleProvider in a credentials cache when s3uploader::role_arn is set, so STS AssumeRole is not called on every S3 request (#52109)

• exporter/azuredataexplorer: Propagate caller's context to Kusto ingestion submission so configured timeouts and cancellations are honored (#51872)
  Previously, ingestion calls used context.Background(), causing exports to ignore configured timeout deadlines.
  The default configuration (timeout: 0) is unchanged and still runs without a deadline; only an explicitly
  configured timeout is now enforced.

• exporter/datadog: Obfuscate SQL span attributes, keep the tracer's sampling decision maker, and let numeric span attributes take precedence over resource attributes. (#52553)
  From updated github.com/DataDog/datadog-agent modules:

  • db.statement and db.query.text on SQL and Cassandra spans are now obfuscated instead of sent raw, and sql.query is no longer added unless the tracer set it (DataDog/datadog-agent#57166).
  • The sampling decision maker _dd.p.dm set by a Datadog SDK is preserved instead of being overwritten with -4 (DataDog/datadog-agent#58471).
  • A numeric span attribute now overrides a resource attribute with the same key instead of both being emitted (DataDog/datadog-agent#58893).

• exporter/datadog: Fix histogram edge cases and serverless host attribution, and stop sending an empty host resource on hostless series. (#52553)
  From updated github.com/DataDog/datadog-agent modules:

  • Histogram and exponential histogram points with more than about 268M observations are dropped with a warning instead of exhausting memory, and a cumulative count that drops without a reset no longer reports huge .count values (DataDog/datadog-agent#57002).
  • Distributions made from OTLP histograms no longer have a minimum greater than their maximum (DataDog/datadog-agent#57109).
  • The serverless source is derived from cloud.resource_id, so valid serverless resources no longer fall back to the Collector hostname (DataDog/datadog-agent#57852).
  • Azure Container Apps detection now requires replica (or service.instance.id); incomplete resources fall back to hostname detection (DataDog/datadog-agent#57323).
  • Series without a hostname, such as otel.datadog_exporter.metrics.running.*, no longer carry an empty host resource (DataDog/datadog-agent#56455).

• exporter/datadog: Update go-sqllexer to v0.2.6 for SQL obfuscation. (#52553)
  Fixes PostgreSQL strings containing backslashes being truncated, MySQL db . table identifiers being altered, and Db2 literals with backslashes (DataDog/datadog-agent#57039, DataDog/datadog-agent#58950).

• exporter/elasticsearch: Fix failed document log fields accumulating across documents in a bulk response, which leaked previous documents' fields and input into later error logs. (#52714)

• exporter/elasticsearch: Add authentication credential support to the Elasticsearch Exporter version check (#52437)

• exporter/file: Fix race condition in fileExporter Shutdown causing nil pointer dereference panic when concurrent exports are in-flight (#46871)
  Calls to export telemetry after or during shutdown now return io.ErrClosedPipe
  instead of panicking, and repeated calls to Shutdown are idempotent.
  Shutdown also now waits for any active background flush to complete before closing the file.

• exporter/file: Log the first periodic flush error for each output file writer. (#52574)

• exporter/file: Report file close errors during shutdown when group_by is enabled. (#52587)

• exporter/prometheus_remote_write: Unblock the write-ahead log when the remote endpoint rejects the record at the head of the queue (#51710)

• exporter/signalfx: Replace queued entity event dimension updates with the latest entity state instead of merging them. (#52610)

• exporter/signalfx: Apply dimension_client::strip_k8s_label_prefix and the kubernetes_service_ property rename to dimension updates converted from entity events. (#52607)

• extension/aws_logs_encoding: Parse ELB access logs whose client and target addresses use unbracketed IPv6 host:port form. (#52532)

• extension/http_forwarder: Preserve all values of HTTP response headers, including multiple Set-Cookie headers. (#52453)

• extension/http_forwarder: Avoid false response truncation warnings for chunked and HEAD responses. (#52589)

• extension/opamp: Report health updates when only nested component statuses change. (#51862)

• internal/docker: Close underlying Docker sdk client transport on Client.Close() to prevent socket leak. (#51972)

• pkg/coreinternal: Fix strptime time parsing never returning for a format that repeats a directive, such as %H %H. (#52627)

• pkg/fileconsumer: Fix log loss on graceful shutdown under back pressure (#52067)
  Logs interrupted by shutdown are now re-read after restart instead of being skipped, so delivery is at-least-once and some logs may be duplicated.
  This only holds with the stanza.synchronousLogEmitter feature gate enabled; with the default batching emitter, buffered logs can still be lost on shutdown.
  For gzip files, the whole interrupted file chunk is re-read instead of being skipped.

• pkg/ottl: Make the MD5, SHA1 and CommunityID converters work when the collector runs in strict FIPS 140-3 mode (GODEBUG=fips140=only). (#52479)
  These converters hash telemetry data for non-security purposes. The digest is now computed with
  crypto/fips140.WithoutEnforcement, so a collector built with the Go FIPS 140-3 module and run with
  GODEBUG=fips140=only no longer fails or panics when a statement uses them. The algorithms themselves
  are still not FIPS approved and must not be used for any security function.

• pkg/prometheus: Correct UCUM time and byte unit conversions behind the pkg.translator.prometheus.CorrectUCUMUnits feature gate. (#52480)

• pkg/stanza: Preserve multiline message bodies in the syslog parser for RFC3164 records without octet counting. (#51554)

• processor/cardinality_guardian: Fix growth detection so a label producing a steady stream of new values each epoch is enforced after its first epoch, not only during it (#50403)

• processor/cumulative_to_delta: Return an error when an invalid regexp is configured in include or exclude instead of panicking on the first metric batch (#51963)

• processor/log_dedup: Flush pending aggregated logs on shutdown with a non-cancelled context so downstream consumers do not reject them (#52316)

• processor/redaction: Record the body key in the redaction summary attributes when a string log body matches an allowed value (#50417)
  Previously the processor returned early without recording the allowed value, so an
  allowed string log body was never reflected in redaction.body.allowed.keys /
  redaction.body.allowed.count.

• processor/redaction: Mask a whole multi-line value matched by blocked_key_patterns with a single mask or hash. (#50296)

• processor/resource_detection: The eks, k8s_api and kubeadm detectors no longer fail at startup when the Collector is not running in Kubernetes; they return no attributes, or fail detection when fail_on_missing_metadata is enabled. (#52552)

• processor/tail_sampling: Preserve resource and scope schema URLs when splitting incoming spans by trace ID. (#51524)

• receiver/awsecscontainermetrics: Prevent a panic in getResourceFromARN when the task ARN has fewer than 5 colon-separated segments. (#51810)

• receiver/azure_blob: Fix a panic in Event Hub mode when a blob is created in a container whose signal has no configured pipeline. (#52448)

• receiver/azure_event_hub: Report the number of log records, metric data points and spans in the receiver accepted and refused metrics, instead of the number of Event Hub events. (#52785)

• receiver/cloudflare: Apply max_request_body_size to the HTTP server so configured limits above 20MiB are honored. (#51897)

• receiver/docker_stats: Fix ContainerEnvToMap truncating env var values that contain = (e.g. JAVA_OPTS=-Xmx=512m, base64 strings, connection strings). Split on only the first = to preserve the full value. (#51948)

• receiver/flink_metrics: Record JVM garbage collector metrics for the JobManager and TaskManagers with both the G1 and Parallel collectors. (#52405)
  The receiver mapped only PS_* collector names for the JobManager and only G1_* names for the TaskManagers.
  With G1, the JVM's default collector, the JobManager reported no flink.jvm.gc.collections.count or
  flink.jvm.gc.collections.time data points.

• receiver/googlecloudmonitoring: Preserve distribution sums when converting Google Cloud Monitoring distributions to histograms. (#49409)

• receiver/googlecloudmonitoring: Skip metrics with unsupported value type and metric kind combinations instead of forwarding metrics with no data, which broke downstream exporters with errors like "metrics type is unset". (#39600)

• receiver/k8s_cluster: Don't emit container entity events for containers that don't have a container ID yet. (#52722)

• receiver/k8s_cluster: Use the pod spec image for container.image.name and container.image.tag, consistent with k8sattributesprocessor. (#46541)
  Previously the status image was used, which some runtimes report as a bare digest (sha256:<hex>).
  Values may change for images without a fully qualified registry (e.g. nginx instead of docker.io/library/nginx).
  Previously the status image was used, which some runtimes report as a bare digest (sha256:<hex>).
  Values may change for images without a fully qualified registry (e.g. nginx instead of docker.io/library/nginx).

• receiver/libhoney: Fix panics when event fields have an unexpected type, such as an integer duration_ms or a null trace.parent_id (#52072)
  Values from the request body were converted with unchecked type assertions, so a valid but unexpected type
  (for example an integer duration sent with msgpack, or a JSON null parent id) panicked the request handler and
  dropped the whole batch. Integer durations are now accepted, and fields with an unexpected type are ignored.

• receiver/libhoney: Enforce decompressed body size limit to prevent OOM (#51985)

• receiver/mongodb: Derive the UUID v5 service.instance.id outside FIPS 140-3 strict enforcement so the receiver no longer panics under GODEBUG=fips140=only. (#52539)
  The identifier is derived with uuid.NewSHA1, which is not a security use of SHA-1. The value is unchanged.

• receiver/mysql: Derive the UUID v5 service.instance.id outside FIPS 140-3 strict enforcement so the receiver no longer panics under GODEBUG=fips140=only. (#52539)
  The identifier is derived with uuid.NewSHA1, which is not a security use of SHA-1. The value is unchanged.

• receiver/postgresql: Derive the UUID v5 service.instance.id outside FIPS 140-3 strict enforcement so the receiver no longer panics under GODEBUG=fips140=only. (#52539)
  The identifier is derived with uuid.NewSHA1, which is not a security use of SHA-1. The value is unchanged.

• receiver/prometheus: Keep credentials in proxy_url and other URL fields on config reload instead of replacing the password with xxxxx. (#52337)

• receiver/prometheus_remote_write: Answer 400 Bad Request instead of 204 No Content when a native histogram is dropped for its bounds or buckets and Prometheus' own validation refuses it too. (#50340)

• receiver/prometheus_remote_write: Report zero in the X-Prometheus-Remote-Write-*-Written response headers when a request is refused as invalid. (#50340)

• receiver/prometheus_remote_write: Count the samples of dropped classic histogram and summary series in the X-Prometheus-Remote-Write-Samples-Written response header, count the dropped series in the new otelcol_receiver_prometheusremotewrite_dropped_series internal metric, and log each one at Debug level instead of Info. (#52224)

• receiver/prometheus_remote_write: Drop float flavored native histograms and rebuild the data point count from the translated buckets. (#50337)
  A count can be lower than the one Prometheus sent, because observations that
  no bucket represents are not counted. A custom bucket histogram whose bounds
  or buckets cannot be represented is now dropped and logged at error level.

• receiver/splunk_enterprise: Close the HTTP response body when setting the search job TTL (#51986)
  The response body of the search job TTL request was never closed on a successful response, or when reading it failed.
  This leaked a connection on every search scrape. The body is now always closed.

API Changelog

🛑 Breaking changes 🛑

• all: Remove the deprecated SignalFx receiver. Use the OTLP receiver instead. (#51934)
• processor/resource_detection: Change the type of the openshift detector token option to configopaque.String so it is redacted when the configuration is printed. (#52775)
• processor/transform: Remove the deprecated flatten_data config option. Use flatten.log instead. (#32080)

💡 Enhancements 💡

• extension/encoding: Add optional message decoding interfaces with explicit client metadata options for all signals. (#51752)
• pkg/ottl/contexts/xspanlink: Add ottlspanlink context, in the new experimental xspanlink module, exposing per-span-link fields (trace_id, span_id, trace_state, attributes, dropped_attributes_count, flags) for use in OTTL statements. (#26660)
• pkg/semconvtest: Pin the default semantic-conventions registry to a release instead of floating on main (#52122)
• receiver/receiver_creator: Improve extensibility of receivercreator rules by allowing use of custom observer.EndpointDetails "types" (#50548)
  The receivercreator configuration validation no longer requires that the rule's type == <endpoint_details>
  prefix match one of the observer.EndpointDetails "types" defined in this repository. Now it only enforces the syntax requirements.
  The old enforcement didn't allow downstream distributions to create their own observer.EndpointDetails maps
  making harder to developer new observers that could be used with the receivercreator.

We are thrilled to welcome our first-time contributors to this project. Thank you for your contributions @nikhilred2619, @ssijbabu, @chethankumblekar, @arxhr007, @gssbzn, @slashpai, @benjoseph-grafana, @SaifuddinM23, @cesan3, @atharvamhaske, @eugenekurasov, @bwplotka, @tanmaya-panda1, @dsyme, @nbiswal1, @sk-85, @robertrv, @shubhayu-dev, @ricardbejarano, @0x5t4l1n, @Akash-Kumar-Sinha, @shubhamrai9122-creator, @gezapeti, @thiagocarneiro90, @noctchillin, @smarri4, @vesari ! 🎉