open-telemetry/opentelemetry-operator v0.161.0

Release v0.161.0

on GitHub

0.161.0

🛑 Breaking changes 🛑

• opamp: Use the Kubernetes cluster ID to compute standalone OpAMP Bridge agent instance IDs. (#5689)

  • Existing standalone agents will report new instance IDs after upgrade.
  • The bridge now requires the "get" permission on the "kube-system" namespace to read the cluster ID.

• collector: Remove the deprecated --create-rbac-permissions CLI flag. RBAC permission creation is now exclusively autodetected at startup. (#2588)
  The flag was deprecated in v0.88.0. The operator now always autodetects whether its service account can create ClusterRoles and ClusterRoleBindings.

💡 Enhancements 💡

• collector: Add revisionHistoryLimit field to OpenTelemetryCommonFields to control the number of old ReplicaSets retained for Deployment and StatefulSet workloads, and for the TargetAllocator Deployment. (#0)

• collector: Skip the automatic GOMEMLIMIT and GOMAXPROCS environment variables when the collector config enables the cgroup_runtime extension. (#5651)
  The extension sets both variables from the cgroup limits and does nothing when they are already present.
  The deprecated cgroupruntime type name is recognized as well.
  Each variable is skipped only while its gomemlimit.enabled or gomaxprocs.enabled setting is true.

• collector, target allocator: Add imagePullSecrets to the OpenTelemetryCollector and TargetAllocator CRs, passed through to the pod spec of the generated workloads. A target allocator embedded in an OpenTelemetryCollector uses the collector's imagePullSecrets. (#4376)

• collector: Register the webhook_event receiver (alias webhookevent) so the operator derives its Service port from endpoint, which the receiver requires, and rejects a configuration that omits it. (#5631)

🧰 Bug fixes 🧰

• collector: Fix the collector defaulting webhook panicking when receivers or exporters is missing. (#5735)
  The request is now rejected by CRD validation, as it already was when no component defaults apply.

• target allocator: Honor the cumulative temporality preference for counters bridged from the Prometheus registry in OTLP self-telemetry. (#5700)
  Bridged counters, such as the Prometheus service discovery metrics, were converted to delta regardless of
  temporality_preference. They are now converted only when the preference selects delta for counters
  (delta or low_memory), and are exported as cumulative otherwise.

• target allocator: Keep the Kubernetes API server addresses in the target allocator NetworkPolicy up to date. (#5551)
  The operator now watches the EndpointSlices of the default/kubernetes Service and updates the target allocator
  NetworkPolicy when they change, instead of using the addresses discovered once at operator startup.
  Addresses which disappear from the EndpointSlices remain allowed for 5 minutes, to tolerate transient changes.
  The NetworkPolicy also allows egress to the kubernetes Service ClusterIP, and uses correct CIDRs for IPv6 addresses.
  The operator now requires the watch permission on endpointslices in the discovery.k8s.io API group.

• operator: Fix the --tls-min-version CLI flag being ignored because it was matched against the wrong flag name. (#3565)

• sidecar: Prevent sidecar injection into invalid pods with zero regular containers (#5737)
  Pods without containers are invalid in Kubernetes. Skipping injection allows Kubernetes to reject the invalid Pod rather than mutating it into a valid one.

Components

• OpenTelemetry Collector - v0.161.0
• OpenTelemetry Contrib - v0.161.0
• Java auto-instrumentation - v2.32.0-1
• .NET auto-instrumentation - v1.17.0-2
• Node.JS - v0.78.0-6
• Python - v0.66b1-2
• Go - v0.24.0
• ApacheHTTPD - 1.0.4-2
• Nginx - 1.0.4-2