Killing

The steady-state "pet killer". Every tick the module collects pods matching the configured selectors, picks one at random, and removes it — respecting a minAvailable floor so it never drains a workload below a safety threshold.

When to use it

• You want to continuously validate that your workload recovers from a random pod loss.
• You care about graceful disruption — the default strategy: evict goes through the Eviction API and respects PodDisruptionBudget.
• You need a safety floor. minAvailable prevents over-killing.

For mass kills, see GorillaKill. For graceful rollout-restart-style churn, see Rollout.

Minimal config

kind: Killing
name: kill-my-app
metadata:
  namespace: default
scenario:
  interval: 60s
  minAvailable: 1
  matchers:
    labels:
      app: my-app

Full reference

kind: Killing
name: kill-my-app                       # required — log correlation id
metadata:
  namespace: default                    # required — target namespace
scenario:
  interval: 60s                         # required — Go duration (mutually exclusive with cron)
  # cron: "*/5 * * * *"               # alternative to interval — standard 5-field cron
  wait: 10s                             # optional — first-tick delay, < interval (ignored with cron)
  minAvailable: 1                       # required — won't kill below this floor
  dryRun: false                         # optional — log only, no API mutation
  strategy: evict                       # optional — "evict" (default) | "delete"
  matchers:                             # required — at least one selector
    labels:
      app: my-app
    # podName: my-pod-abc123
    # deploymentName: my-deployment
    # daemonsetName: my-daemonset
    # statefulsetName: my-statefulset

Field details

Field	Type	Default	Notes
scenario.interval	duration	—	Cadence between ticks. Mutually exclusive with cron. One is required.
scenario.cron	string	—	Standard 5-field cron expression. Mutually exclusive with interval.
scenario.wait	duration	0	Delay before the first tick. Must be < interval. Ignored when cron is set.
scenario.minAvailable	integer	0	Skip the tick if len(pods) - minAvailable <= 0.
scenario.dryRun	bool	false	When true, logs the intended victim without calling the API.
scenario.strategy	enum	evict	evict uses the Eviction API (PDB-aware). delete is a hard delete.
scenario.matchers	object	—	See Matchers. Must be non-empty.

Behavior

1. On each tick, collect the union of pods matching every configured selector.
2. If no pods match, emit a killing skipped: no_match warning and exit.
3. Compute killable = len(pods) - minAvailable. If <= 0, emit a killing skipped: min_available_floor warning and exit.
4. Pick a uniform random victim from the pool.
5. Log the intended action (always, including in dry-run).
6. If dryRun: false, call the API:
   • strategy: evict → CoreV1().Pods(ns).EvictV1(...)
   • strategy: delete → CoreV1().Pods(ns).Delete(...)

Failure modes (e.g. the API call returns an error) are logged but do not stop the module — the next tick runs as scheduled.

RBAC

Strategy	Required verbs
evict	pods → list, get; pods/eviction → create
delete	pods → list, get, delete

Plus whatever read access matchers need (see Matchers RBAC).

See RBAC reference for a ready-to-apply Role.