hull metrics

hull metrics samples a release’s pods over time and prints per-container CPU and memory statistics — optionally with suggested requests and limits.

When to use it

To right-size a release: watch what its containers actually use before you set resources.requests and resources.limits.
To spot a container that spikes far above its average, so you catch it before it gets OOM-killed or throttled.
Add --recommend to get a ready-to-paste values block instead of reading the numbers yourself.

Sample over a window that spans real traffic — a 30-second window during a quiet period will undersize. The cluster needs metrics-server installed; without it the first sample fails with a clear message.

What happens

You name a release. hull finds every pod that belongs to it (matched by the names of the workloads in the release manifest) and prints which prefixes it is sampling.
It polls the metrics.k8s.io API every --interval for --duration, building a usage history.
It prints a table: one row per container with the sample count and the min / avg / p50 / p95 / max for CPU (in millicores) and memory.
With --output json the same statistics print as JSON instead.
With --recommend it adds a resources: block: requests sized from p50 and limits from p95, each with headroom. Treat these as starting points.

Usage

hull metrics <release-name> [flags]

Flags

Flag	Type	Default	Description
`--duration`	duration	`30s`	how long to sample overall — make it span real traffic
`--interval`	duration	`5s`	how often to take a sample within the window
`-o, --output`	string	`table`	render as `table` or `json`
`--recommend`	—	`false`	also print a suggested `resources.requests`/`limits` block

Persistent flags inherited from `hull`

Flag	Type	Description
`--debug`	—	enable debug output
`--kube-context`	string	Kubernetes context to use
`--kubeconfig`	string	path to kubeconfig file
`-n, --namespace`	string	Kubernetes namespace

Worked example

Sample the web-api release for ten minutes and recommend resources:

hull metrics web-api --duration 10m -n prod --recommend

Output:

Sampling web-api pods every 5s for 10m0s (matching prefixes: [web-api])

CONTAINER                          SAMPLES    CPU(min/avg/p50/p95/max, m)              MEM(min/avg/p50/p95/max)
api                                   120         12 /     41 /    38 /    92 /   140        70Mi / 118Mi / 112Mi / 210Mi / 240Mi

# suggested resources block (paste into values.yaml):
resources:
  # container: api (over 120 samples)
  requests: {cpu: 50m, memory: 144Mi}
  limits:   {cpu: 150m, memory: 320Mi}

Get the raw statistics as JSON for another tool:

hull metrics web-api --duration 5m -n prod -o json | jq '.[] | {container, cpuP95, memP95}'

If no pods match the release, hull prints:

no samples collected (no pods matching the release labels?)