Check MongoDB Migration Readiness
When to use
Use this skill before migrate-mongodb-to-percona. Nothing in this skill
mutates the cluster.
The procedure it gates is documented in
docs/runbooks/mongodb-bitnami-to-percona-migration.md (the source of truth).
Inputs
NVSENTINEL_NAMESPACE(defaultnvsentinel)NVSENTINEL_RELEASE(defaultnvsentinel)MIGRATION_PVC_SIZE_GI(default8): the volume size the Percona install will request. Set it to the value from the operator's planned values file so the storage check validates the real request.
Setup
export NVSENTINEL_NAMESPACE="nvsentinel"
export NVSENTINEL_RELEASE="nvsentinel"
export MIGRATION_PVC_SIZE_GI="8"
Steps
Run the preflight script and show the operator the full table:
scripts/mongodb-migration/preflight.shInterpret the verdict:
- Exit
2(BLOCKED): stop. Explain each FAIL row and propose the fix, but do NOT execute mutations inside this skill (it is read-only by contract); the operator applies fixes, then re-run the preflight. Typical blockers: a failed/pending Helm release, both backends present (mixed state; see the runbook troubleshooting table), cert-manager missing, or the default StorageClass minimum above the requested volume size (OCI block volumes have a 50Gi minimum; the Percona operator wedges before replica set init when the provisioned volume is larger than requested). - Exit
0with REVIEW rows: each REVIEW row is an operator decision, not yours. Do not proceed until each is acknowledged in conversation.
- Exit
Capture the decisions the migration skill needs:
- Data handling: the preserve path (dump and restore) is the DEFAULT: document IDs are preserved, node annotations and remediation resources stay valid, and in-flight fault handling resumes after the restore. It requires scaling fault-quarantine, node-drainer, and fault-remediation to zero before the dump, so no references get created for events the archive will not contain; capture that as part of the plan. The clean path (no dump, all health event data lost) is the opt-out: one-time faults such as GPU XIDs are never re-detected on it, so the operator must explicitly choose it.
- Quarantined nodes: record the list from the REVIEW row. With the clean path the operator must decide per node: remediate first, keep cordoned for manual review, or return to service.
- GitOps: ask whether ArgoCD/Flux manages the installation. If yes, reconciliation must be suspended before the migration and the desired state in git updated before resuming (runbook steps 2, 4a and 6a).
- Chart reference: record which chart and version the installation
currently runs (from the Application spec, HelmRelease, or
helm list), so the migration redeploys the SAME chart version with new values instead of silently upgrading NVSentinel.
Output
Report to the operator:
- verdict (
READY/READY with review items/BLOCKED) - the chosen data-handling path
- the recorded quarantined-node list (or "none")
- whether GitOps suspension is required
- the values changes the install will need (backend flags, volume size, scheduling keys)
- the chart reference and version to redeploy (
CHART_REFfor the migration skill)
Next skill to run
migrate-mongodb-to-percona(only onREADY, with the decisions above captured)
References
| Topic | Reference |
|---|---|
| Runbook (source of truth) | docs/runbooks/mongodb-bitnami-to-percona-migration.md |
| Backend configuration | docs/configuration/mongodb-store.md |
| Scripts | scripts/mongodb-migration/ |