signalfx/splunk-otel-collector-chart splunk-otel-collector-0.149.0

on GitHub

This Splunk OpenTelemetry Collector for Kubernetes release adopts the Splunk OpenTelemetry Collector v0.149.0.

🛑 Breaking changes 🛑

• chart: Rename all k8sattributes processor references to k8s_attributes (#2326)
  The k8sattributes alias has been deprecated in favor of k8s_attributes in the chart-generated configuration.
  Any Helm values or overrides that reference k8sattributes (for example, *.config.processors.k8sattributes
  or pipelines that list k8sattributes as a processor) must be updated to use k8s_attributes.
  The chart will fail to be installed or upgraded if the deprecated alias is still referenced.
• chart: Rename all otlp exporter references to otlp_grpc (#2326)
  The otlp alias has been deprecated in favor of otlp_grpc in the chart-generated configuration.
  Any Helm values or overrides that reference otlp (for example, *.config.exporters.otlp
  or pipelines that list otlp as an exporter) must be updated to use otlp_grpc.
  The chart will fail to be installed or upgraded if the deprecated alias is still referenced.
• chart: Rename all otlphttp exporter references to otlp_http (#2326)
  The otlphttp alias has been deprecated in favor of otlp_http in the chart-generated configuration.
  Any Helm values or overrides that reference otlphttp (for example, *.config.exporters.otlphttp
  or pipelines that list otlphttp as an exporter) must be updated to use otlp_http.
  The chart will fail to be installed or upgraded if the deprecated alias is still referenced.

💡 Enhancements 💡

• agent, clusterReceiver, gateway: Add OpenShift resource detection support (#2330)
  When distribution is set to "openshift", the openshift resource detector is now enabled
  in the resourcedetection processor. Previously, OpenShift clusters only had cloud-provider
  detectors (azure for cloudProvider=azure, gcp for cloudProvider=gcp) but no
  OpenShift-specific detector, and no detector at all for cloudProvider=aws (see the
  ec2 bug fix in this release). The openshift detector auto-discovers k8s.cluster.name,
  cloud.provider, cloud.platform, and cloud.region via the OpenShift API
  (config.openshift.io/v1/infrastructures). The cloud.platform value reflects the
  actual hosting environment (e.g., aws_openshift, gcp_openshift, azure.openshift).
  Cloud-provider-specific detectors remain enabled alongside the openshift detector to
  collect instance-level metadata: ec2 on AWS, gcp on GCP, and azure on Azure. The
  openshift detector is listed first to ensure the correct cloud.platform is set before
  cloud-provider detectors run.
  The clusterName Helm value is now optional for OpenShift (same as EKS and GKE). When
  not set, the openshift detector auto-discovers it from the cluster API. When explicitly
  set, the static value takes precedence.
• chart: Support all standard K8s probe fields (timeoutSeconds, periodSeconds, failureThreshold, successThreshold) in readinessProbe and livenessProbe configuration (#2341)
  Previously only initialDelaySeconds was applied from readinessProbe and livenessProbe
  values. All other fields like timeoutSeconds were ignored. The probe templates
  now pass through all user-provided fields via toYaml.
• chart: Add configurable resource requests and limits for the Target Allocator container (#2349)
  Users can now set targetAllocator.resources in values.yaml to configure CPU/memory
  requests and limits for the Target Allocator container.
• chart: Bump the OBI subchart to upstream opentelemetry-ebpf-instrumentation 0.6.0 (#2300)
  Updates the vendored OBI dependency and rendered eBPF instrumentation example
  manifests to the upstream 0.6.0 chart release.
• operator: Bump operator to 0.109.0 in helm-charts/splunk-otel-collector/Chart.yaml (#2347)

🧰 Bug fixes 🧰

• agent, clusterReceiver, gateway: Fix resource detection for non-EKS Kubernetes clusters on AWS (#2330)
  For non-EKS Kubernetes clusters on AWS (cloudProvider=aws with distribution unset or
  set to openshift), EC2 instance metadata (host.id, cloud.account.id,
  cloud.availability_zone, etc.) was never collected. Two issues contributed to this:
  1. The chart used the eks detector, which only works on actual EKS clusters — it checks
     for EKS-specific signals (IRSA/Pod Identity token paths, OIDC issuer, Kubernetes
     version string) and returns an empty resource when those are absent. Switched to the
     ec2 detector, which works on any AWS instance with IMDS access.
  2. This code path only matched when distribution was unset (empty string), so OpenShift
     clusters on AWS (distribution=openshift, cloudProvider=aws) were excluded entirely.
     Broadened the condition to also match the openshift distribution.
• clusterReceiver: Set dnsPolicy to ClusterFirstWithHostNet for cluster receiver when clusterReceiver.hostNetwork is enabled (#2348)
  When clusterReceiver.hostNetwork is enabled (e.g. for EKS Auto mode), the cluster receiver
  pod was unable to resolve in-cluster DNS names. Pods running with
  hostNetwork fall back to the node's DNS configuration unless dnsPolicy is explicitly set
  to ClusterFirstWithHostNet. This matches the existing daemonset behavior.