github open-telemetry/opentelemetry-operator v0.160.0
Release v0.160.0
on GitHub

3 hours ago

0.160.0

🛑 Breaking changes 🛑

  • operator, collector, target allocator: Disable operator, collector, and target allocator network policies by default. (#5551, #5558, #5621, #5624, #5654)
    The operator.networkpolicy and operand.networkpolicy feature gates are moved back to alpha and are disabled by default.
    The generated network policies can block API server access on some clusters.
    To keep creating network policies, enable them with --feature-gates=+operator.networkpolicy,+operand.networkpolicy,
    or set spec.networkPolicy.enabled: true on individual collectors and target allocators.

🚩 Deprecations 🚩

  • auto-instrumentation: Remove autoinstrumentation-php image definition from the repository (#932)
    The PHP autoinstrumentation image will be hosted by the packaging SIG.
    The one released image (v0.1.0) of the image in the operator registry will
    eventually be deleted. This image was never used for any functionality shipped by the operator.

💡 Enhancements 💡

  • collector: Add RBAC support for extracting labels and annotations from CronJobs (and Jobs) via the k8sattributes processor. (#5527)
    The operator's own ServiceAccount must itself hold get/list/watch on batch/jobs and
    batch/cronjobs for this to take effect, since Kubernetes RBAC only lets a ServiceAccount
    grant permissions it already has. This PR also extends the operator's own ClusterRole
    accordingly (config/rbac/role.yaml).

  • collector, target allocator: Add enableServiceLinks to the OpenTelemetryCollector and TargetAllocator CRs, passed through to the pod spec of the generated workloads. It defaults to true, matching the Kubernetes Pod field. (#5569)

  • operator: Add support for Kubernetes 1.37 (#5650)

🧰 Bug fixes 🧰

  • auto-instrumentation: Honor Instrumentation CRD/API json tags when loading --config-file (no-CRD webhook configs). (#5618)
    Operator Config fields still load with gopkg.in/yaml.v3 (yaml tags). The
    instrumentations block is re-parsed with sigs.k8s.io/yaml so embedded API
    types keep their json tags (for example apacheHttpd, imagePullPolicy,
    resourceAttributes, valueFrom, volumeLimitSize, configPath).

  • collector: Preserve TCP and UDP protocols when generating NetworkPolicy ports. (#5611)

  • target allocator: Use the .svc FQDN for the TargetAllocator endpoint injected into the collector's Prometheus receiver so it works on clusters behind an HTTP proxy. (#5586)
    The bare service name (http://-targetallocator:80) does not match the standard NO_PROXY
    entries (.svc, .cluster.local), so Go routes the request through the proxy. Using
    -targetallocator..svc matches NO_PROXY=.svc while staying compatible with
    custom cluster domains (unlike the full .svc.cluster.local endpoint reverted in #3248).

  • target allocator: Restore promhttp_metric_handler_requests_total and promhttp_metric_handler_requests_in_flight on the /metrics endpoint when a custom gatherer is configured. (#5622)
    #5294 switched the /metrics handler to promhttp.HandlerFor when a custom gatherer
    is set, but omitted the InstrumentMetricHandler wrapper that registers these two metrics.
    The handler is now built once in NewServer (after options are applied) using
    InstrumentMetricHandler, restoring the missing metrics for both code paths.

  • target allocator: Watch and prune all the resources the TargetAllocator controller creates. (#5657)
    The controller now watches the resources it creates, so manual changes to them are reverted.
    It also deletes the resources it no longer needs, for example the NetworkPolicy after
    spec.networkPolicy.enabled is set to false, or the ServiceMonitor after
    spec.observability.metrics.enableMetrics is set to false.

  • collector: Fix string values that look like numbers, booleans, timestamps or special floats (e.g. "0e12", ".inf") being written unquoted to the collector ConfigMap, which made the collector read them as a different type and fail to start. (#4314)
    The ConfigMap is now rendered with the same YAML library the collector parses it with (go.yaml.in/yaml/v3),
    and the rendered document is checked to decode back to the values held by the CR before it is written.
    The target allocator rewrite of the Prometheus receiver no longer re-parses the rendered config, and the
    resulting ConfigMap uses the same formatting as configs without a target allocator (2-space indentation).

  • collector: Detect Gateway API availability by the HTTPRoute kind served in gateway.networking.k8s.io/v1 instead of the API group alone, so the operator no longer crash-loops on clusters with a partial Gateway API bundle (for example only GatewayClass). (#5571)

  • target allocator: Stop assigning targets to collector pods that are being deleted. A terminating pod keeps reporting Ready until the kubelet confirms the deletion, which never happens while its node is unreachable, so its targets were never reassigned. (#5576)

  • webhook: Configure the standalone OpenShift webhook with the environment variables required for cert-manager autodetection. (#5660)
    This allows TargetAllocator mTLS admission to discover cert-manager when the webhook is
    installed through the OpenShift/OLM deployment path.

  • opamp: Only require workload patch permissions when the AcceptsRestartCommand capability is enabled, and reject restart commands when it is not. (#5573)

Components

Check out latest releases or
releases around open-telemetry/opentelemetry-operator v0.160.0

Don't miss a new opentelemetry-operator release

NewReleases is sending notifications on new releases.

Get notifications