Skip to content

Scaleway Kapsule infrastructure live acceptance — 2026-09-15

Status: PASS for the bounded infrastructure-only cell. This record does not certify Magento runtime health, multi-zone HA, backup/restore, regional DR, observability collection, or edge behavior. It replaces the 2026-08-13 cell (kept in git history).

Cell

  • Provider project: redacted
  • Region and zone: nl-ams, nl-ams-1 (fr-par Kapsule was in provider-side shortage on 2026-09-15; admission correctly refused before mutation)
  • Kapsule: 1.36.4, one DEV1-L worker
  • Managed database: MySQL 8, db-dev-s, not HA
  • Managed cache: Redis 8.6.6, RED1-micro, cluster size 1
  • Network: one disposable /22 VPC/private network
  • Artifact: public immutable NGINX digest, used only as a pullable infrastructure smoke image; it is not Magento-compatible
  • Harness: scripts/k8s-acceptance-local.sh scaleway, deploy --infra-only
  • Ownership prefix: scw915
  • Catalog spec: certification-scaleway
  • Cell: nl-ams / nl-ams-1, Kapsule 1.36.4, RDB not HA, Redis cluster size 1, digest is NGINX infra-only (not Magento)
  • TTL: 3600s; Pulumi state and passphrase were local to the disposable acceptance work directory

Result

PASS:

  1. Dependency, authenticated-profile, project, region, one-cell-catalog, and immutable-image preflights passed before provider mutation.
  2. The run created 23 resources: VPC/private network, RDB instance and database, Redis cluster, Kapsule cluster and pool, Kubernetes provider, generated secrets, Deployments, and a LoadBalancer Service.
  3. Provisioning completed in 13m12s. The RDB instance, Redis cluster, and Kapsule worker pool were the material asynchronous waits.
  4. The infrastructure-only output stage passed and returned the owned cluster, network, private-subnet, database-secret, encryption-key-secret, and service identities. Runtime health was not claimed because the immutable NGINX image cannot run Magento.
  5. Teardown deleted all 23 resources in 3m0s.
  6. The harness's independent cleanup assertion returned scaleway assert_clean ok.
  7. Fresh direct owning-service inventories after the harness exited returned zero resources with the exact scw915 prefix for Kapsule clusters, RDB instances, Redis clusters, VPCs, private networks, and load balancers.

Corrections discovered by the live cell

  • The generated managed-service passwords did not guarantee every character class (Special:true only widens the pool), and Scaleway now rejects passwords missing a digit, uppercase, lowercase, or special character. The cache and database generators now pin minima for all four classes; the mock-graph test pins the guarantee.
  • Stale catalog defaults moved to the live values: Redis 8.6.6, RED1-micro, RDB db-dev-s, Kapsule 1.36.4.
  • One attempt was interrupted mid-deploy by the operator; the orphaned cluster, RDB instance, Redis cluster, and VPC were removed with ordered provider deletes and re-verified empty before the passing run.

Spend

About 16 minutes of minimum SKUs (one DEV1-L worker, db-dev-s, RED1-micro, one Kapsule control plane): well under €1, inside the shared $50 own-money cap with vendor margin intact.

Explicit non-claims

  • --infra-only skipped Magento migration, cutover, and runtime health.
  • One zone and one worker do not prove Kapsule multi-zone HA or node-failure recovery.
  • No backup, restore, known-content integrity, regional DR, fencing, failback, collector delivery, native Cockpit workload integration, or edge traffic test was run.
  • The result does not promote Scaleway to a production-certified target; it closes only this deterministic fake-client plus disposable Kapsule infrastructure lifecycle evidence for the single-zone cell.