signalfx/splunk-otel-collector-chart splunk-otel-collector-0.150.0 on GitHub

This Splunk OpenTelemetry Collector for Kubernetes release adopts the Splunk OpenTelemetry Collector v0.150.0.

🛑 Breaking changes 🛑

chart: Default API and ingest endpoints changed from *.signalfx.com to *.observability.splunkcloud.com (#2364)
The default endpoints have been updated as follows:
- https://ingest.<realm>.signalfx.com → https://ingest.<realm>.observability.splunkcloud.com
- https://api.<realm>.signalfx.com → https://api.<realm>.observability.splunkcloud.com
  The new endpoints resolve to different IP addresses. Review the
  domain transition guide
  for the full list of network and security controls to update before upgrading,
  or the collector will fail to send data.
  To keep using the legacy *.signalfx.com endpoints, set these values explicitly:
```
splunkObservability:
  ingestUrl: "https://ingest.<realm>.signalfx.com"
  apiUrl: "https://api.<realm>.signalfx.com"
```

💡 Enhancements 💡

agent: Add maxUnmatchedBatchSize option to logsCollection.containers.multilineConfigs (#2379)
The chart now supports maxUnmatchedBatchSize in multilineConfigs, which controls how many
consecutive lines not matching firstEntryRegex are batched into a single log record.
gateway: Add topologySpreadConstraints support for gateway deployment (#2361)
The gateway deployment now supports gateway.topologySpreadConstraints to control how
gateway pods are distributed across topology domains (e.g., availability zones). When
combined with gateway.service.trafficDistribution (PreferSameZone on Kubernetes 1.35+,
or PreferClose on older versions), this enables zone-aware routing where clients prefer
sending traffic to gateways in the same zone. Together, these options ensure gateways are
evenly spread across zones while minimizing cross-AZ network traffic and costs.

🧰 Bug fixes 🧰

agent: Fix histogram bucket count for kubeproxy_network_programming_duration_seconds metric (#2328)
Splunk Observability Cloud has a limit of 32 buckets for histograms. By default this histogram
has more than 32 buckets, so the config will now merge buckets to meet the required limit
to avoid this metric being dropped entirely. This change is relevant when both the
useControlPlaneMetricsHistogramData feature gate and agent.controlPlaneMetrics.proxy
config option are both enabled.
chart: Add probes and scrapeconfigs RBAC permissions to Target Allocator ClusterRole (#2359)
The Target Allocator ClusterRole was missing permissions for probes and scrapeconfigs
resources in the monitoring.coreos.com API group. This caused the allocator to fail with
forbidden errors when attempting to list and watch these resources.
clusterReceiver: Remove resourcedetection processor from cluster-scoped event and object pipelines (#2369)
The cluster receiver's logs (k8s_events), logs/objects, and logs/events pipelines
previously used the resourcedetection processor, which added host and
cloud-provider attributes (host.id, host.name, cloud.provider, cloud.platform,
cloud.account.id, cloud.region, etc.) from the cluster receiver pod itself.
These host-level attributes were misleading because cluster-scoped events and objects are not
tied to any single host; they should reflect the cluster. These pipelines now only add k8s.cluster.name.
A host.name resource attribute is still set to the node name of the cluster receiver pod (K8S_NODE_NAME)
for use as the Splunk HEC host field.
gateway: Add profiling to gateway OTLP port enabled_for lists (#2365)
The gateway OTLP ports (4317, 4318) were missing profiling in their enabled_for
lists, unlike the agent ports which already included it. This meant the profiling signal
alone could not enable the gateway OTLP ports.
operator: Respect user-defined SPLUNK_PROFILER_ENABLED and SPLUNK_PROFILER_MEMORY_ENABLED in instrumentation.spec.env (#2365)
When splunkObservability.profilingEnabled is true, the chart auto-injects
SPLUNK_PROFILER_ENABLED=true and SPLUNK_PROFILER_MEMORY_ENABLED=true into
the Instrumentation CR. Previously, setting these to "false" in
instrumentation.spec.env to globally disable profiling while keeping the
pipeline active was silently overridden. The chart now correctly honors
user-defined values and skips injection when they are already present.