signalfx/splunk-otel-collector v0.152.0

on GitHub

This Splunk OpenTelemetry Collector release includes changes from the opentelemetry-collector v0.152.0
and the opentelemetry-collector-contrib v0.152.0 releases where appropriate.

🛑 Breaking changes 🛑

• (Contrib) connector/span_metrics: Promote connector.spanmetrics.includeCollectorInstanceID feature gate to beta. (#40400)
  Adds a collector.instance.id attribute to all metrics emitted by the spanmetrics connector.
• (Contrib) processor/tail_sampling: Stabilize disableinvertdecisions feature gate. (#47650)
• (Contrib) receiver/kafka_metrics: Promote the receiver.kafkametricsreceiver.UseFranzGo feature gate to stable (#41480)
  The franz-go client is now the only implementation; the gate is now permanently enabled and will be removed in v0.154.0.
  The Sarama-based implementation has been removed.

🚩 Deprecations 🚩

• (Contrib) processor/metrics_transform: Rename processor type from metricstransform to metrics_transform (#45339)
• (Contrib) exporter/prometheusremotewrite: add_metric_suffixes is deprecated. Use translation_strategy: UnderscoreEscapingWithoutSuffixes if you are setting this to false. (#33661)
• (Contrib) receiver/cloud_foundry: Rename receiver type from cloudfoundry to cloud_foundry (#45339)
• (Contrib) receiver/kafka_metrics: Rename kafkametrics receiver to kafka_metrics with deprecated alias kafkametrics (#45339)
• (Contrib) receiver/kubelet_stats: Rename receiver type from kubeletstats to kubelet_stats (#45339)
• (Contrib) receiver/tcp_check: Rename tcpcheck receiver to tcp_check with deprecated alias tcpcheck (#45339)

💡 Enhancements 💡

• (Splunk) packaging: Update Splunk OpenTelemetry Java agent to v2.27.0 (#7483)
• (Splunk) packaging: Update JMX metrics gatherer to v1.55.0 (#7447)
• (Splunk) packaging: Update Splunk OpenTelemetry Node.js agent to v4.5.0 (#7448)
• (Core) pkg/exporterhelper: Add otelcol_exporter_in_flight_requests metric to track the number of export requests currently in-flight per exporter. (#15009)
  This UpDownCounter increments in startOp and decrements in endOp, allowing operators to monitor
  concurrent export activity and detect when an exporter is saturating its worker pool.
• (Contrib) exporter/file: Allow usage of append mode with zstd compression (#44382)
• (Contrib) exporter/kafka: Allow record_headers to accept multiple headers with the same key. (#48092)
• (Contrib) exporter/loadbalancing: Add stable attribute routing key encoding for traces and metrics in the loadbalancing exporter (#46094, #46095)
  Routing keys now encode attributes as name=value| segments, including explicit markers for missing attributes.
  Non-string attribute values are deterministically stringified and used consistently across traces and metrics.
• (Contrib) extension/bearertokenauth: Update token file parsing to ignore everything after the first whitespace, allowing for inline comments. (#46100)
• (Contrib) extension/docker_observer: Add include_all_containers option to emit a port-less endpoint for every running container, including those with no exposed ports. (#48252)
  When enabled, the observer emits a container endpoint with no port
  information for every running container, alongside any per-port endpoints.
  This allows receiver_creator rules of type == "container" to attach
  receivers to every container regardless of whether it exposes ports.
  Defaults to false for backwards compatibility.
• (Contrib) processor/attributes: Added support for default values in the attributes processor. (#45352)
  This enhancement allows users to specify default values for attributes in the attributes processor.
  If the primary value source (e.g., environment variable, attribute, or context value) is not available,
  the default value will be used. This ensures that the pipeline doesn't fail due to missing configuration.
• (Contrib) processor/k8s_attributes: Improve deployment name extraction heuristic when deployment_name_from_replicaset is enabled (#44831)
  When deployment_name_from_replicaset is true and the ReplicaSet informer is not used for deployment names only,
  the processor derives k8s.deployment.name using the pod-template-hash label and ReplicaSet naming rules.
  When a ReplicaSet informer is running (for example for k8s.deployment.uid), API-backed metadata takes precedence,
  independent of the deployment_name_from_replicaset setting.
• (Contrib) processor/k8s_attributes: Add watch_sync_period config option to configure informer cache resync period. (#48111)
  The watch_sync_period config option defaults to 5m to match the previously hardcoded behavior.
• (Contrib) processor/k8s_attributes: Use PartialObjectMetadata for non-Pod informers (#47389)
  Switch Namespace, Node, Deployment, StatefulSet, DaemonSet, and Job informers
  from full typed objects to PartialObjectMetadata via the metadata client.
  These resources only need labels, annotations, UID, name, and owner references—all available
  in object metadata—so fetching full spec/status is unnecessary overhead.
  Pods continue using full objects since they require spec/status
  fields (pod IP, node name, containers, host network).
• (Contrib) processor/resource: Added support for default values in the resource processor. (#45352)
  This enhancement allows users to specify default values for attributes in the resource processor.
  If the primary value source (e.g., environment variable, attribute, or context value) is not available,
  the default value will be used. This ensures that the pipeline doesn't fail due to missing configuration.
• (Contrib) receiver/azure_event_hub: Expose the Prefetch option for Azure Event Hub receiver (#48038)
• (Contrib) receiver/k8s_cluster: Enable the re-aggregation feature for the k8s_cluster receiver (#46361)
• (Contrib) receiver/k8s_events: Simplified RBAC example in README to only include required events resource permission. (#48192)
  A namespace-scoped Role can also be used when the receiver is configured with specific namespaces.
• (Contrib) receiver/kafka: Add support for custom consumer-group partition-assignment strategies via extensions that implement kgo.GroupBalancer. Set group_rebalance_strategy to the component ID of a registered extension to use a custom balancer. (#48096)
  The four built-in strategies (range, roundrobin, sticky, cooperative-sticky) continue to work unchanged.
  Any other value for group_rebalance_strategy is now resolved as an extension component ID at runtime.
• (Contrib) receiver/kafka: Cache OTel metric attribute sets in broker hook callbacks to reduce per-read allocations (#47395)
  OnBrokerRead previously rebuilt attribute sets on every call. The computed
  MeasurementOption is now cached per (nodeID, outcome) key and evicted on
  broker disconnect. Growth is bounded by 2 × number-of-brokers.
• (Contrib) receiver/kubelet_stats: Add optional system container metrics (cpu.time, cpu.usage, memory.working_set) to kubeletstats receiver. (#3531)
• (Contrib) receiver/memcached: Enables dynamic metric reaggregation in the Memcached receiver. This does not break existing configuration files. (#46364)
• (Contrib) receiver/mysql: Added mysql.query_plan to query sample records collected (#47413)
• (Contrib) receiver/mysql: Fix the query hash generation for MySQL queries when the query is not explainable. (#48059)
• (Contrib) receiver/prometheus_remote_write: Add debug logging when dropping histograms with invalid schema values (#48027)
• (Contrib) receiver/sqlserver: Adds idle blocking sessions to db.server.query_sample by appending sleeping blocker rows (with sqlserver.command=IDLE_BLOCKER) when blocker session IDs are missing from active requests. (#47120)
  Idle blocker enrichment is best-effort: a targeted secondary idle-session query runs only for missing blocker session IDs, and failures are logged as warnings without failing the scrape.

🧰 Bug fixes 🧰

• (Core) pkg/confighttp: Close the original request body after reading block-format Content-Encoding: snappy requests. (#15262)
• (Core) pkg/confighttp: Recover from panics in decompression libraries, return HTTP 400 instead of 500. (#13228)
• (Core) pkg/confighttp: Enforce max_request_body_size on Content-Encoding: snappy requests before the decoded buffer is allocated. (#15252)
• (Core) pkg/otelcol: Stop emitting verbose gRPC transport messages at WARN during normal client disconnect. (#5169)
  grpc-go gates chatty per-RPC notices (e.g. "HandleStreams failed to read frame:
  connection reset by peer") behind LoggerV2.V(2). zapgrpc.Logger.V conflates
  grpclog verbosity with zap severity, so V(2) returns true whenever WARN is
  enabled and these messages emit at WARN. Wrap the installed grpclog.LoggerV2
  with a corrected V() that compares against a fixed verbosity threshold,
  matching grpclog's intended semantics. See uber-go/zap#1544.
• (Core) pkg/pdata: pcommon.Value.AsString no longer HTML-escapes <, >, and & inside ValueTypeMap and ValueTypeSlice values, matching the behavior already used for ValueTypeStr. (#14662)
• (Core) pkg/service: Fix Prometheus config defaults mismatch when host is explicitly set in telemetry configuration. (#13867)
  When users explicitly configured the telemetry metrics section (e.g. to change the host),
  the Prometheus exporter boolean fields (WithoutScopeInfo, WithoutUnits, WithoutTypeSuffix)
  defaulted to nil/false instead of true, causing metric name format changes compared to the
  implicit default configuration. This fix applies the correct defaults during config unmarshaling.
• (Core) pkg/service: Return noop tracer provider when no trace processors are defined (#15135)
• (Contrib) exporter/kafka: Fix collector hanging indefinitely on shutdown when the Kafka broker is unreachable (#48140)
  Re-enables the exporterhelper timeout sender (previously disabled with a zero timeout) so that each send attempt is bounded by the configured timeout (default 5s). The per-attempt context is now passed through to ProduceSync, allowing in-flight sends to unblock when the broker is unreachable. FranzSyncProducer.Close also cancels the kgo client context before calling client.Close, ensuring shutdown completes promptly.
• (Contrib) extension/health_check: Fix deadlock when the collector run context is cancelled while the healthcheck extension is present. (#47591)
• (Contrib) extension/opamp: Decorrelate service.instance.id and OpAMP instance_uid (#46495)
  Previously:
  • the service.instance.id reported in the AgentDescription was based on the OpAMP instance UID
  • the instance UID was typically set based on the service.instance.id from the Collector resource attributes
  • it could be overridden using the instance_uid configuration of the OpAMP extension
    This meant that the reported service.instance.id did not always match the Collector resource attributes,
    which is a problem for correlation, and that server implementations got used to the typical case of
    service.instance.id and instance_uid matching, despite there being no guarantee of this.
    Now:
  • the reported value of service.instance.id always matches the Collector resource attributes
  • the instance UID is either taken from the instance_uid configuration or generated randomly.
    This means that the two values can never be expected to match, unless both configurations are explicitly set to the same value.
    That is what the OpAMP supervisor does, which means its behavior is unaffected.
• (Contrib) processor/k8s_attributes: Fix a data race when reading the kube-system namespace from the client cache to populate the k8s.cluster.uid attribute. (#47910)
  The ClusterUID rule in extractPodAttributes accessed the shared Namespaces
  map without acquiring the client's read lock, which could race with writes from
  the namespace informer. The access now goes through the locked GetNamespace
  helper.
• (Contrib) processor/k8s_attributes: Fix container.image.tags config check (#47113)
• (Contrib) processor/tail_sampling: Fix sampling inconsistencies when using span-ingest sampling strategy (#47476)
  1. Drop policies were not working if the initial batch did not contain a span matching the drop policy.
  2. If max_spans or upper_threshold_ms were respectively set on span_count or latency policies they may not have been respected.
• (Contrib) receiver/host_metrics: Fix duplicate rootPath prefix in filesystem mountpoint translation (#47083)
  When gopsutil falls back to reading /proc/self/mountinfo, the reported
  mountpoints already contain the rootPath prefix. This caused
  translateMountpoint to add it a second time, resulting in incorrect paths
  like /hostfs/hostfs/data.
• (Contrib) receiver/mysql: Strip leading SQL comments before EXPLAIN check so queries prefixed with block or line comments are correctly identified as explainable. (#46587)

Docker Image Manifests:

• Linux (amd64, arm64, ppc64le) and Windows (2019 amd64, 2022 amd64):
  • quay.io/signalfx/splunk-otel-collector:0.152.0
  • digest: sha256:8f1e074d9935e0a5dff36fae8cd84d235917de72be38362c09df79292098234a