Nebius Observability Agent for Kubernetes® changelog

​

Fixed

• Fixed a bug where the Helm chart generated an invalid configuration when the logs pipeline was disabled (for example, metrics-only or traces-only setups).

​

Added

• Configuration parameter config.metrics.labelAllowlist to control which labels from Pod, Service and Node objects are attached to scraped metrics. Each field (podLabels, serviceLabels, nodeLabels) is a regular expression matched against label names. Default: ".+" (all labels are attached, same as before). Use it to bound label cardinality: metrics with more than 33 labels are silently dropped, and clusters with many labels on Node objects (for example, added by Node Feature Discovery or GPU Operator) or on Service objects can exceed this limit. Example configuration that limits labels from Node objects to a curated allowlist:
  config:
    metrics:
      labelAllowlist:
        podLabels: ".+"
        serviceLabels: ".+"
        nodeLabels: "kubernetes_io_hostname|topology_kubernetes_io_.*|node_kubernetes_io_instance_type|nvidia_com_gpu_present|nebius_com_.*"
  

​

Fixed

• Removed the k8s.node_group.id label from kube-apiserver and etcd metrics. These metrics are cluster-wide, but each agent stamped them with its own node group ID, multiplying every apiserver_* and etcd_* time series by the number of node groups in the cluster and inflating the active time series count. Metrics local to a node keep the label.

​

Added

• Traces collection and forwarding to Nebius AI Cloud. For more details, see How to ingest traces with Nebius Observability Agent for Kubernetes.
• Configuration parameter config.traces.enabled to enable or disable trace collection. Default: true.
• OTLP endpoint for receiving traces from applications at nebius-observability-agent.<namespace>.svc.cluster.local:4317, where <namespace> is the Managed Service for Kubernetes namespace where the agent is installed (default: observability).

config:
  traces:
    enabled: true

​

Fixed

• Fixed a wrong image tag in the Helm chart.

​

Added

• Support for setting the metadata region in Helm values for Managed Service for Kubernetes clusters not running on Compute VMs.
• Configuration parameter metadata.region (for example, eu-north1) to specify the region when automatic detection is not available.
• Configuration parameter metadata.parent_id (for example, project-e00abcde...) to specify the parent project ID.
• Required setup commands for authentication:
  • nebius iam auth-public-key generate —service-account-id {targetSaId} —output credentials.json
  • kubectl create secret --namespace=observability generic iam-credentials-file --from-file=credentials.json
• The target service account (target-sa-id) must be granted the following permissions:
  • logging.logs.writer for sending logs to Observability Logs
  • monitoring.metrics.writer for sending metrics to Observability Metrics

config:
  iam:
    auth_scheme: "iam-credentials-file"
    credentials_file_secret_name: "iam-credentials-file"
    credentials_file_secret_key: "credentials.json"
  metadata:
    region: "eu-north1"
    parent_id: "project-e00abcde..."

​

Added

• Metrics collection capability for Managed Service for Kubernetes workloads.
• Metrics forwarding to Observability Metrics.
• Configurable settings to choose between logs only, metrics only or both.
• Parameter to exclude agent logs from collection (config.logs.collectAgentLogs: false).
• Parameter to exclude agent metrics from collection (config.metrics.collectAgentMetrics: false).
• Metrics enrichment with cluster ID, node group ID and other Managed Service for Kubernetes metadata.

​

Added

• Helm chart for Managed Service for Kubernetes.
• Out-of-the-box log collection from Managed Service for Kubernetes into Observability Logs with minimal configuration.
• Log enrichment (for example, cluster ID and metadata injection).
• Reliability features.