provisioning/provisioning-catalog
1
0
Fork 0
Code Issues Pull requests Projects Releases Packages Wiki Activity Actions
provisioning-catalog/components/MIGRATION_CONCERNS.md

15 KiB
Raw Permalink Blame History

ServiceConcerns Migration — Component Reference

This document describes the mechanical pattern for migrating each component under provisioning/extensions/components/<X>/nickel/ to declare a ServiceConcerns umbrella (ADR-008, ADR-009).

Status

Components migrated as of this writing — patrón aplicado y validado:

TLS endpoints / databases:

  • docker_mailserver — tls_endpoint_with_acme style; backup 'pending for mail-stack BackupGroup
  • zot — tls_endpoint_with_acme style; backup 'pending for registry-stack BackupGroup
  • postgresql — database; backup 'pending for workspace-level BackupPolicy with dump_strategy

Storage backends / cert infrastructure:

  • longhorn — infra_storage_managed; backup 'disabled (engine state via SystemBackupDef.longhorn_engine)
  • cert_manager — infra_storage_managed; backup 'disabled (cluster_resources via SystemBackupDef.tls_certmanager)

Stateless runtimes / OS / kernel layer:

  • containerd — stateless; container runtime
  • crio — stateless; container runtime
  • runc — stateless; OCI runtime
  • crun — stateless; OCI runtime
  • youki — stateless; OCI runtime
  • os — stateless; base OS configuration
  • kubernetes — stateless; kubeadm-managed CP/kubelet (state in etcd via SystemBackupDef.etcd, PKI via SystemBackupDef.k8s_certs)
  • k0s — stateless; data_dir via SystemBackupDef.host_configs when k0s is the active distribution
  • k8s-nodejoin — stateless; one-shot join operation
  • vol_prepare — stateless; block-device prep
  • longhorn_node_prep — stateless; node-level Longhorn prep

Infrastructure glue (controllers / CSI / network):

  • cilium — infrastructure_glue; CNI controller
  • hccm — infrastructure_glue; Hetzner cloud controller
  • hetzner_csi — infrastructure_glue; CSI driver
  • democratic_csi — infrastructure_glue; NFS dynamic provisioner
  • external_nfs — infrastructure_glue; NFS server
  • local_path_provisioner — infrastructure_glue; local-path provisioner
  • fip — infrastructure_glue; floating IP attachment
  • etcd — infrastructure_glue; data captured by SystemBackupDef.etcd
  • private_gateway — infrastructure_glue; Cilium gateway
  • buildkit_runner — infrastructure_glue; ephemeral by design (ADR-039)

DNS providers:

  • coredns — dns_provider; zone files via SystemBackupDef.external_coredns
  • external_dns — dns_provider; controller, records live in upstream provider
  • resolv — stateless; /etc/resolv.conf manager

TLS endpoints with ACME:

  • forgejo — tls_endpoint_with_acme; backup 'pending (Git+DB+actions)
  • woodpecker — tls_endpoint_with_acme; backup 'pending (DB+agent state)
  • stalwart — tls_endpoint_with_acme; backup 'pending (mail data+DB)
  • nats — internal cluster service; backup 'pending (JetStream state)
  • odoo — bespoke; backup 'pending for odoo-stack BackupGroup

Databases:

  • mariadb — database; backup 'pending (dump_strategy at workspace level)
  • surrealdb — database; backup 'pending (dump_strategy at workspace level)

VPN gateway:

  • wireguard — stateless; peer keys + config in git/SOPS

Total: 37 of 38 components migrated. Only backup_manager is excluded — it is the subsystem itself and does not consume its own ServiceConcerns schema.

No components pending migration with extensions/ directory.

Components without extensions/components/<name>/ directory (workspace-only):

  • prometheus.ncl, grafana.ncl, loki.ncl, vector.ncl — declare concerns directly in the workspace .ncl file
  • ops_controller.ncl, audit_mirror.ncl, radicle_seed.ncl, coredns_vpn.ncl, coredns_vpn_wrk1.ncl — workspace-declared, mostly reusing ext.make_coredns (which carries inherited concerns from coredns).

Reusable presets

provisioning/schemas/lib/concerns_presets.ncl exports 7 archetype presets that defaults.ncl can import directly to avoid duplicating the same 9-line block across components. Components that match an archetype exactly should use:

let _presets = (import "schemas/lib/concerns_presets.ncl").presets in

{
  <component> | default = {
    # ... existing fields ...
    concerns | default = _presets.<archetype>,
  },
}

Components using presets directly (8): containerd, crio, runc, crun, youki, os (stateless); cilium, hccm (infrastructure_glue).

Components using preset + override (_presets.<name> & { … | force = … }) (22): k8s-nodejoin, vol_prepare, resolv, wireguard, longhorn_node_prep, kubernetes, k0s (stateless variants); hetzner_csi, democratic_csi, external_nfs, local_path_provisioner, fip, private_gateway, buildkit_runner, etcd (infrastructure_glue); longhorn, cert_manager (infra_storage_managed); coredns, external_dns (dns_provider); postgresql, mariadb, surrealdb (database).

Components using parametric preset (_presets.tls_endpoint_with_acme {…}) (6): zot, docker_mailserver, forgejo, woodpecker, stalwart, odoo.

Of these, 4 also declare manifest_hooks in their defaults.ncl (standard gateway-based TLS lifecycle via catalog/lib/tls-hook-methods.sh): forgejo, wordpress_site, odoo, rustelo_website.

zot, docker_mailserver, stalwart use custom TLS lifecycle (own lib methods) and do NOT declare manifest_hooks — the preset does not carry hooks.

Components with full inline blocks (1): nats (internal cluster service, distinct from public TLS endpoints).

Total: 36 of 37 migrated components use presets (97%).

Mechanical pattern

For each provisioning/extensions/components/<X>/nickel/:

Step 1 — contracts.ncl

Add the import at the top of the file (above the outer {):

let _concerns_lib = import "schemas/lib/concerns.ncl" in

Inside the component schema record (before the trailing .. if it has one, or as the last field otherwise) add:

    # ServiceConcerns umbrella (ADR-008). See defaults.ncl for the populated value.
    concerns | _concerns_lib.ServiceConcerns | optional,

If the contract is a closed record (no ..), also add .. so legacy mode/operations/live_check fields are accepted by the contract.

Step 2 — defaults.ncl

Add a concerns | default = { ... } block before the closing } of the default record. Choose the variant matching the component's archetype.

Stateless preset (runtime/OS)

    concerns | default = {
      tls   = { kind = 'disabled, reason = "no TLS termination at this layer" },
      dns   = { kind = 'disabled, reason = "no DNS records owned by this component" },
      certs = { kind = 'disabled, reason = "no ACME issuer required" },
      backup = {
        kind   = 'disabled,
        reason = "stateless: configuration in git, no runtime data to capture",
      },
      observability = { kind = 'pending, reason = "ObservabilityImpl iteration deferred", backlog_ref = "OBS-001" },
      security      = { kind = 'pending, reason = "SecurityImpl iteration deferred", backlog_ref = "SEC-001" },
    },

Infrastructure glue (controllers/operators)

    concerns | default = {
      tls   = { kind = 'disabled, reason = "controller-level RBAC, not TLS endpoint" },
      dns   = { kind = 'disabled, reason = "no DNS records owned by this component" },
      certs = { kind = 'disabled, reason = "no ACME issuer required" },
      backup = {
        kind   = 'disabled,
        reason = "state in K8s API captured by SystemBackupDef.cluster_resources",
      },
      observability = { kind = 'pending, reason = "ObservabilityImpl iteration deferred", backlog_ref = "OBS-001" },
      security      = { kind = 'pending, reason = "SecurityImpl iteration deferred", backlog_ref = "SEC-001" },
    },

TLS endpoint with ACME (public service)

    concerns | default = {
      tls = {
        kind     = 'enabled,
        tls_impl = {
          secret_name = "<COMPONENT>-tls",
          issuer_ref  = "letsencrypt-prod",
          hostnames   = [],
        },
      },
      dns = {
        kind     = 'enabled,
        dns_impl = {
          internal = [],
          zone     = "",
        },
      },
      certs = {
        kind       = 'enabled,
        certs_impl = {
          acme_server = "https://acme-v02.api.letsencrypt.org/directory",
          email       = "",
          secret_ref  = "",
          provider    = 'cloudflare,
        },
      },
      backup = {
        kind        = 'pending,
        reason      = "BackupPolicy declared at workspace level",
        backlog_ref = "BACKUP-<COMPONENT>-001",
      },
      observability = { kind = 'pending, reason = "ObservabilityImpl iteration deferred", backlog_ref = "OBS-001" },
      security      = { kind = 'pending, reason = "SecurityImpl iteration deferred", backlog_ref = "SEC-001" },
    },

Database

    concerns | default = {
      tls    = { kind = 'disabled, reason = "internal cluster service, ingress-level TLS handled separately" },
      dns    = { kind = 'disabled, reason = "no public DNS records" },
      certs  = { kind = 'disabled, reason = "no ACME issuer required" },
      backup = {
        kind        = 'pending,
        reason      = "BackupPolicy with database scope + dump_strategy declared at workspace level",
        backlog_ref = "BACKUP-DB-<COMPONENT>-001",
      },
      observability = { kind = 'pending, reason = "ObservabilityImpl iteration deferred", backlog_ref = "OBS-001" },
      security      = { kind = 'pending, reason = "SecurityImpl iteration deferred", backlog_ref = "SEC-001" },
    },

Step 3 — Verify

cat > /tmp/check_<X>.ncl <<EOF
let mod = import "extensions/components/<X>/nickel/defaults.ncl" in
mod.<x>.concerns
EOF
nickel export --import-path provisioning /tmp/check_<X>.ncl

Expected output: a JSON object with tls, dns, certs, backup, observability, security keys, each with kind and matching payload.

Workspace-level overrides

After per-component defaults are populated, the workspace declarations in infra/<workspace>/components/<x>.ncl can override individual concerns. The most common override is to bring backup from 'pending to 'enabled with a real BackupPolicy referencing infra/<workspace>/lib/backup_policies.ncl presets.

Example — workspace activates backup for a component with full BackupPolicy:

let bp_lib = import "../../lib/backup_policies.ncl" in
let ext    = import "extensions/components/<X>/nickel/main.ncl" in

{
  <x> = ext.make_<x> {
    # ... existing fields ...
    concerns = ext.defaults.<x>.concerns & {
      backup = {
        kind        = 'enabled,
        backup_impl = {
          provider     = { name = "restic" },
          destinations = [bp_lib.destinations.hetzner_primary, bp_lib.destinations.b2_replica],
          encryption   = bp_lib.encryption.component_key "<x>",
          schedule     = { kind = 'cron, cron_expr = "0 3 * * *" },
          retention    = bp_lib.retention.gold,
          scopes       = [
            { kind = 'service_full, name = "main", paths = ["/var/lib/<x>"] },
          ],
          tag_strategy = bp_lib.tag_strategy.component "<x>",
        },
      },
    },
  },
}

Op-level manifest hooks (TLS gateway lifecycle)

Some cluster components require lifecycle steps that are structurally identical across instances — Cloudflare secret, ClusterIssuer, Certificate, ReferenceGrant, Gateway patch. Rather than repeating them in every manifest_plan.ncl, these components declare manifest_hooks inside their concerns record in defaults.ncl.

The Nu bundle builder (comp-build-cluster-bundle) merges hooks from two sources:

effective_pre  = concerns.manifest_hooks.op.pre  ++ manifest_plan.hooks.op.pre
effective_post = manifest_plan.hooks.op.post ++ concerns.manifest_hooks.op.post

Generated run-<op>.sh executes: effective_pre ++ plan_steps ++ effective_post.

Which components use manifest_hooks

Only components that deploy via the standard gateway-based TLS lifecycle (cert-manager ClusterIssuer + Gateway API HTTPRoute + Cloudflare DNS-01):

  • forgejo, wordpress_site, odoo, rustelo_website

Components with custom TLS lifecycle (docker_mailserver, stalwart, zot) use the preset for concerns compliance only — no manifest_hooks.

Required env vars for hook-based TLS components

Each component's env-<name>.j2 must export these vars so that catalog/lib/tls-hook-methods.sh can execute the hooks at deploy time:

TLS_HOOK_CF_SECRET_NAME=<cf-secret-name>       # e.g. dns-jesusperez-pro
TLS_HOOK_GATEWAY_NAME=<gateway-name>           # e.g. libre-wuji
TLS_HOOK_GATEWAY_NS=<gateway-namespace>        # e.g. kube-system
TLS_HOOK_LISTENER_NAME=https-<service-name>    # e.g. https-wordpress-diegodelgado
TLS_HOOK_DOMAIN=<primary-domain>              # e.g. diegodelgado.es
TLS_HOOK_TLS_SECRET=<tls-secret-name>         # e.g. wordpress-diegodelgado-tls
TLS_HOOK_NAMESPACE=<service-namespace>         # e.g. diegodelgado
TLS_HOOK_CLUSTER_ISSUER=<issuer-name>          # e.g. letsencrypt-prod-diegodelgado

These are populated from Tera variables that vars.nu must expose (no component prefix): cf_secret_name, gateway_ip, tls_secret, dns_zone.

Adding a new gateway-TLS component

  1. Declare concerns | default = (_presets.tls_endpoint_with_acme {...}) & { manifest_hooks = {...} } in defaults.ncl (copy from forgejo/defaults.ncl).
  2. Keep manifest_plan.ncl clean — no TLS steps (create-cf-secret, clusterissuer, certificate, referencegrant, patch-gateway-https, remove-gateway-https).
  3. Export TLS_HOOK_* vars in env-<name>.j2.
  4. Ensure vars.nu outputs cf_secret_name, gateway_ip, tls_secret (without component prefix).
  5. No changes to lib.sh — TLS methods live in catalog/lib/tls-hook-methods.sh.

Reference

  • ADR-008: Service Concerns Umbrella
  • ADR-009: Progressive Concerns Coverage
  • ADR-010: Backup Architecture
  • ADR-011: Multi-Destination Custody
  • Schema: provisioning/schemas/lib/concerns.ncl
  • Presets: provisioning/schemas/lib/concerns_presets.ncl
  • Hook methods: catalog/lib/tls-hook-methods.sh
  • Workspace defaults: workspaces/libre-wuji/infra/lib/concerns_defaults.ncl