Use infra_probe to run a throwaway mini-pipeline before a destructive kzero reset or down. The goal is not to test your application workloads—it is to confirm that platform prerequisites still work right before you wipe real PVCs and core releases.

In live mode, a successful probe run is evidence that (for your chosen probe chart):

kzero orchestrates up → checks → down and optional gate before the main pipeline. It does not embed cloud-specific logic, registry URLs, or PVC cleanup rules—you maintain the probe chart under helm.workspace (or custom: scripts).

kzero ships an anonymous reference (public Bitnami Redis 8) in kzero-selfhosted:

Copy and edit those files, or ignore them entirely and wire infra_probe to whatever you prefer:

• MariaDB / PostgreSQL / RabbitMQ — another release.<ns>/<name> + values in your workspace
• Minimal busybox + PVC — custom: apply/delete scripts
• Same chart as production — usually avoid (probe should be disposable and isolated)

Align infra_probe.checks (pvc_bound, release_ready) with your chart’s PVC name and install semantics.

infra_probe:
  enabled: true
  before: ["reset"]          # add "down" to gate both commands
  fail_fast: true
  cache_ttl: 30m
  pipeline:
    up:
      - release.probe-ns/probe-storage
    down:
      - release.probe-ns/probe-storage
  checks:
    - pvc_bound: probe-ns/<your-pvc-name>
    - release_ready: true

run:
  mode: live
  probe_cache_dir: /var/lib/kzero/probe-cache   # optional; shared path for cron/CI

Place <release>.sh under helm.workspace (basename = release name in release.<ns>/<name>).

infra_probe:
  pipeline:
    up:
      - release.probe-ns/probe-redis
    down:
      - release.probe-ns/probe-redis
  checks:
    - pvc_bound: probe-ns/redis-data-probe-redis-master-0
    - release_ready: true

For distroless Jobs or hosts without /bin/sh, set run.execution: native (or auto) and use Helm SDK for probe release.* steps — no <release>.sh script.

1. Copy kzero-probe-redis-values.yaml and kzero-probe-redis.yaml (chart manifest) into helm.workspace — see docs/examples/infra-probe/ in the product repo and kzero-selfhosted run/examples/infra-probe/.
2. Merge kzero-infra-probe-native.sample.yaml (or selfhosted kzero-infra-probe-redis-native.sample.yaml).
3. Grant RBAC for Helm release secrets/configmaps, pods (checks), and optional pvc / exec in the probe namespace — see kzero-selfhosted run/in-cluster/.

Optional pvc.* on probe down or exec.* write/read checks can extend the mini-pipeline when chart retention does not delete PVCs automatically.

run:
  execution: native
  kubeconfig: ""   # in-cluster Job

infra_probe:
  pipeline:
    up:
      - release.probe-ns/probe-redis
    down:
      - release.probe-ns/probe-redis

kzero runs helm uninstall on probe down; it does not delete PVCs unless your chart/values arrange it.

Configure cleanup in the probe chart values you maintain, for example:

• Bitnami Redis / PostgreSQL / RabbitMQ (StatefulSet): PVC retention when the StatefulSet is deleted on uninstall:

master:
  persistentVolumeClaimRetentionPolicy:
    enabled: true
    whenScaled: Retain
    whenDeleted: Delete

• Generic Helm templates you control: helm.sh/resource-policy: delete where applicable (not always effective on StatefulSet PVCs).

• Custom manifests: retention policy or explicit PVC delete in your scripts.

Pick the pattern that matches your probe chart.

Dry-run main pipelines do not run the gate; use kzero probe with run.mode: dry-run to preview steps.

Recommended workflow:

1. kzero probe in dry-run — plan only
2. kzero probe in live — validate registry, credentials, volumes
3. Set infra_probe.enabled: true — gate reset / down in automation

After a successful probe in live mode, cache_ttl skips re-running probe steps when the fingerprint (cluster + pipeline + checks) matches. Set run.probe_cache_dir on shared hosts so cache survives process restarts.

• SPECIFICATIONS.md § infra_probe
• notifications.md — optional alerts on pipeline failure after probe passes