Agent Configuration

All values below live under the agent key in your values.yaml or can be set with --set agent.<path>=<value>.

Authentication

Value	Type	Default	Description
`config.token`	string	`""`	Agent token from the Kubeadapt dashboard. Required (unless `existingSecret` is set).
`config.existingSecret`	string	`""`	Name of an existing Kubernetes Secret containing the token. The Secret must have a key named `token`. When set, `config.token` is ignored and no Secret is created by the chart.

Tip

If you use a GitOps controller or store values in Git, reference an existing Secret via existingSecret so you don't have to keep the token as plain text in your repository.

Data Collection

The agent collects cluster metrics at 1-minute intervals by default. This granularity can be overridden by Kubeadapt Cloud for your subscription tier. The agent handles retries and recovery automatically.

Compression

Value	Type	Default	Description
`config.compressionLevel`	int	`3`	zstd compression level for data sent to Kubeadapt Cloud. Min: 1 (fastest). Max: 4 (smallest payload).

The agent uses zstd streaming compression. Level 3 is a good balance for most clusters. Increase to 4 for larger clusters (3000+ nodes) where payload size matters.

Resource Sizing

Default resources:

yaml

agent:
  resources:
    requests:
      cpu: 100m
      memory: 128Mi
    limits:
      cpu: 1000m
      memory: 1Gi

The agent automatically tunes GOMAXPROCS and GOMEMLIMIT based on the container limits you set.

Tip

The agent is CPU-light. If the pod gets OOMKilled, bump the memory limit. Track agent health and latency in the dashboard under the Connectivity tab.

Cluster Size	Memory Request	Memory Limit
Up to 1000 nodes	128Mi (default)	1Gi (default)
1000+ nodes	512Mi	2Gi

GPU Metrics

The agent auto-discovers NVIDIA DCGM Exporter pods and collects GPU utilization and memory metrics. No configuration needed in most cases.

Value	Type	Default	Description
`config.gpuMetricsEnabled`	bool	`true`	Enable GPU metrics collection from DCGM Exporter.
`config.dcgmPort`	int	`9400`	DCGM Exporter port.
`config.dcgmNamespace`	string	`""`	Restrict DCGM discovery to a specific namespace. Empty means search all namespaces.
`config.dcgmEndpoints`	string	`""`	Comma-separated list of DCGM Exporter endpoints (IPs or hostnames). Use this if auto-discovery fails.

If you do not have NVIDIA GPUs in your cluster, the agent silently skips GPU collection. No need to disable it.

Note

GPU monitoring works at the node level for shared GPU configurations (time-slicing, MPS). DCGM Exporter does not expose per-container utilization counters in shared mode. See GPU Monitoring for details.

Scheduling

Value	Type	Default	Description
`tolerations`	list	`[]`	Pod tolerations.
`nodeSelector`	object	`{}`	Node selector constraints.
`affinity`	object	`{}`	Pod affinity rules.
`topologySpreadConstraints`	list	`[]`	Topology spread constraints.

Full Values Reference

The sections above cover what you need for capability enablement and successful scheduling. For other configurations, see the full values.yaml.