Files
computer-playbook/utils/cache/applications.py
T
kevinveenbirkenbachandClaude Opus 4.7 10cb02461e fix(inventory): bake only variants.yml overrides into matrix-deploy --vars
Since 576fad84f (2026-05-10) moved role images from
`defaults/main.yml` into `meta/services.yml`, matrix-deploy's variant
baker has been writing the FULL effective config (services.yml +
variants.yml deep-merged via `get_variants`) into host_vars through
the `--vars` JSON consumed by `apply_vars_overrides`. That populates
`applications.{role}.services.{svc}.image` (and `.version`, .cpus,
etc.) with the upstream value, which then silently disables
`apply_mirror_overrides` for every role with an explicit image
declaration: the default `if_missing` policy short-circuits when the
field is already non-blank, so the rendered compose / Dockerfile FROM
keeps pulling from the upstream registry rather than the GHCR mirror.

The regression went unnoticed because Docker Hub usually serves the
pull, but it eventually surfaced as a BuildKit 429 on
`docker.io/library/php:8.2-apache` during the suitecrm Dockerfile
build (job 77602110531 on run 26362676803). Nextcloud, moodle,
mariadb consumers, postgres consumers, simpleicons, phpmyadmin,
phpldapadmin, jenkins, baserow, fider, ... all share the same latent
defect — only suitecrm happened to hit the rate limit first.

Fix the root cause instead of papering over per-role with
`mirror_policy: force`: introduce `get_variant_overrides_only` next
to `get_variants` and switch payload.py to it. The new function
returns the raw `meta/variants.yml` entries per role — no merge with
`meta/services.yml`. matrix-deploy now bakes only the variant-
specific overrides into host_vars; the role's services.yml fields
flow through the normal Ansible lookup chain at deploy time, leaving
`image`/`version` blank in host_vars so `apply_mirror_overrides`
populates them from `mirrors.yml` under its default `if_missing`
policy.

Behavior preserved for the other `get_variants` consumers:
* utils/manager/inventory.py:104 (schema-driven credentials) keeps
  the full merged variant payload.
* planner.py:58 (variant-count only) is unaffected.
* legacy_resolver.py:42 (CombinedResolver runtime services_overrides)
  keeps the full merged view.

Coverage:
* 3 new tests in test_applications.py for the new helper (single
  empty variant when no meta/variants.yml, raw override without base
  merge, cache invariance).
* test_payload.py + test_builder.py mocks repointed from `get_variants`
  to `get_variant_overrides_only`.
* `make test` green across all 5 targets (lint, unit, integration,
  external, test-lint) — 188 tests total.

Side-effects assessed:
* Repo-wide grep confirms no template / task accesses
  `applications.X.services.Y.image` directly via hostvars — every
  consumer goes through `lookup('config', …)` which still deep-merges.
* Roles previously pulling upstream now resolve via the GHCR mirror
  whenever the image is already synced (the `images-mirror-missing.yml`
  workflow keeps the destination populated).

Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
2026-05-25 17:33:33 +02:00

381 lines
14 KiB
Python

"""Application-domain cache: variants, defaults, merged lookup.
Owns `_APPLICATIONS_DEFAULTS_CACHE`, `_VARIANTS_CACHE`, and
`_MERGED_APPLICATIONS_CACHE`. Public API: `get_application_defaults`,
`get_variants`, `get_merged_applications`. Strictly ansible-free at
import time so the GitHub Actions runner-host CLI path
(`cli.administration.deploy.development.init` -> `plan_dev_inventory_matrix` ->
`get_variants`) keeps working without ansible installed.
"""
from __future__ import annotations
import copy
from collections.abc import Mapping
from typing import TYPE_CHECKING, Any
from plugins.filter.merge_with_defaults import merge_with_defaults
from utils.roles.mapping import ROLE_FILE_META_VARIANTS
from .base import (
_RENDER_GUARD,
_cache_key,
_deep_merge,
_render_with_templar,
_resolve_override_mapping,
_resolve_roles_dir,
_stable_variables_signature,
)
from .yaml import load_yaml as _load_yaml_cached
from .yaml import load_yaml_any as _load_yaml_any_cached
if TYPE_CHECKING:
import os
from pathlib import Path
_APPLICATIONS_DEFAULTS_CACHE: dict[str, dict[str, Any]] = {}
_VARIANTS_CACHE: dict[str, dict[str, list[Any]]] = {}
_VARIANT_OVERRIDES_ONLY_CACHE: dict[str, dict[str, list[dict[str, Any]]]] = {}
_MERGED_APPLICATIONS_CACHE: dict[tuple, dict[str, Any]] = {}
# Every role's metadata lives under these `meta/<topic>.yml` files.
# The file root IS the value of `applications.<app>.<topic>` — there
# is NO wrapping key matching the filename.
_META_TOPICS: tuple[str, ...] = ("server", "rbac", "services", "volumes")
# `meta/info.yml` is descriptive role-level metadata. Loaded into
# `applications.<role>.info` like the other meta files, but does NOT mark a
# role as an application by itself — a metadata-only `info.yml` next to a
# bare `meta/main.yml` is just documentation, not config.
_META_INFO_TOPIC: str = "info"
def _extract_default_credentials(creds_node: Any) -> dict[str, Any]:
"""Walk a `meta/schema.yml` `credentials:` tree and return the subset of
leaves that carry a literal `default:` Jinja string.
The shape mirrors the schema tree: nested keys stay nested. The
literal string is preserved verbatim — no rendering, no validation.
Leaves WITHOUT `default:` are intentionally absent so the inventory's
apply_schema-generated values win the merge.
"""
if not isinstance(creds_node, Mapping):
return {}
is_leaf = any(
marker in creds_node
for marker in ("default", "algorithm", "validation", "description")
)
if is_leaf:
return {}
out: dict[str, Any] = {}
for key, value in creds_node.items():
if not isinstance(value, Mapping):
continue
leaf = any(
marker in value
for marker in ("default", "algorithm", "validation", "description")
)
if leaf:
if "default" in value:
out[key] = value["default"]
else:
nested = _extract_default_credentials(value)
if nested:
out[key] = nested
return out
def _has_application_metadata(role_dir: Path) -> bool:
"""Detect whether a role behaves as an application.
A role is an "application" iff at least one of its `meta/<topic>.yml`
files exists (plus `meta/schema.yml` and `meta/users.yml` for the
schema-only / users-only special cases). Replaces the legacy
`roles/<role>/meta/services.yml`-presence test.
"""
meta_dir = role_dir / "meta"
if not meta_dir.is_dir():
return False
for topic in _META_TOPICS:
if (meta_dir / f"{topic}.yml").is_file():
return True
if (meta_dir / "schema.yml").is_file():
return True
return bool((meta_dir / "users.yml").is_file())
def _build_role_base_config(
role_dir: Path,
roles_dir: Path,
) -> dict[str, Any]:
"""Assemble a role's effective `applications.<app>` payload from its
`meta/<topic>.yml` files.
`meta/server.yml` → `server`
`meta/rbac.yml` → `rbac`
`meta/services.yml` → `services`
`meta/volumes.yml` → `volumes`
`meta/info.yml` → `info` (descriptive role-level metadata)
`meta/schema.yml` → `credentials` (only the literal `default:` values;
non-default credentials are filled by the
inventory's apply_schema step and merged in by the
caller).
`meta/users.yml` → `users` (rewritten to `lookup('users', ...)` so the
user-domain cache stays the source of truth).
Empty role collapses to ``{}`` (no overrides applied).
"""
# Pure-Python GID resolver — does NOT pull ansible. The previous
# `ApplicationGidLookup().run([...])` call dragged
# `ansible.plugins.lookup.LookupBase` into this code path and broke
# `cli.administration.deploy.development init` on the GitHub Actions runner host
# (CI run 24935979190) where the runner Python ships without
# ansible. The split lives in plugins/lookup/application_gid.py:
# `compute_application_gid` is the pure helper, `LookupModule` is
# the ansible-facing wrapper.
from plugins.lookup.application_gid import compute_application_gid
application_id = role_dir.name
config_data: dict[str, Any] = {}
meta_dir = role_dir / "meta"
for topic in _META_TOPICS:
topic_data = _load_yaml_cached(meta_dir / f"{topic}.yml", default_if_missing={})
if topic_data:
config_data[topic] = topic_data
info_data = _load_yaml_cached(
meta_dir / f"{_META_INFO_TOPIC}.yml", default_if_missing={}
)
if info_data:
config_data[_META_INFO_TOPIC] = info_data
schema_data = _load_yaml_cached(meta_dir / "schema.yml", default_if_missing={})
if schema_data:
creds_defaults = _extract_default_credentials(
schema_data.get("credentials") or {}
)
if creds_defaults:
config_data["credentials"] = creds_defaults
users_meta = _load_yaml_cached(meta_dir / "users.yml", default_if_missing={})
if isinstance(users_meta, dict) and users_meta:
config_data["users"] = {
user_key: "{{ lookup('users', " + repr(user_key) + ") }}"
for user_key in users_meta
}
if not config_data:
return {}
config_data["group_id"] = compute_application_gid(application_id, str(roles_dir))
return config_data
def _iter_application_role_dirs(roles_dir: Path):
"""Yield application role directories in deterministic alphabetical order."""
for child in sorted(p for p in roles_dir.iterdir() if p.is_dir()):
if _has_application_metadata(child):
yield child
def _load_variants_overrides(path: Path) -> list[dict[str, Any]]:
"""Load a `roles/<role>/meta/variants.yml` variant list through the
shared YAML cache.
Each entry is a deep-merge override for the role's
`meta/services.yml`; ``null`` and ``{}`` are valid no-op entries.
Missing file or empty list collapses to a single empty variant so
the role keeps its pre-variant behaviour.
"""
if not path.exists():
return [{}]
raw = _load_yaml_any_cached(path)
if raw in (None, {}):
return [{}]
if not isinstance(raw, list):
raise TypeError(
f"{path} must contain a YAML list of override mappings (or be empty)."
)
if not raw:
return [{}]
normalised: list[dict[str, Any]] = []
for index, entry in enumerate(raw):
if entry is None:
normalised.append({})
elif isinstance(entry, Mapping):
normalised.append(dict(entry))
else:
raise ValueError(
f"{path}[{index}] must be a mapping (or null); got {type(entry).__name__}"
)
return normalised
def _build_variants(roles_dir: Path) -> dict[str, list[Any]]:
"""Return ``{application_id: [variant_0, variant_1, ...]}``.
Each variant is the role's assembled `meta/<topic>.yml` payload
deep-merged with the corresponding entry from
`roles/<role>/meta/variants.yml`. A missing/empty `meta/variants.yml`
collapses to a single empty variant so the role keeps its pre-variant
behaviour.
"""
variants: dict[str, list[Any]] = {}
for role_dir in _iter_application_role_dirs(roles_dir):
application_id = role_dir.name
base_config = _build_role_base_config(role_dir, roles_dir)
meta_path = role_dir / ROLE_FILE_META_VARIANTS
override_list = _load_variants_overrides(meta_path)
role_variants: list[Any] = []
for override in override_list:
if base_config:
role_variants.append(_deep_merge(base_config, override))
else:
# Role has no meta payload, but a variant list MAY still
# legitimately produce an override-only result. Fall back
# to a deep copy of the override so callers never observe
# shared mutable state.
role_variants.append(copy.deepcopy(override))
variants[application_id] = role_variants
return {key: variants[key] for key in sorted(variants)}
def _build_application_defaults(roles_dir: Path) -> dict[str, Any]:
"""Return ``{application_id: base_config}`` — each role's variant-free
``meta/<topic>.yml`` payload only. Variants stay accessible via
:func:`get_variants`. Using variant 0 here would leak its service
flags into variant-N consumers' runtime view via deep-merge."""
defaults: dict[str, Any] = {}
for role_dir in _iter_application_role_dirs(roles_dir):
defaults[role_dir.name] = _build_role_base_config(role_dir, roles_dir)
return {key: defaults[key] for key in sorted(defaults)}
def get_application_defaults(
*, roles_dir: str | os.PathLike[str] | None = None
) -> dict[str, Any]:
resolved_roles_dir = _resolve_roles_dir(roles_dir=roles_dir)
key = _cache_key(resolved_roles_dir)
cached = _APPLICATIONS_DEFAULTS_CACHE.get(key)
if cached is None:
cached = _build_application_defaults(resolved_roles_dir)
_APPLICATIONS_DEFAULTS_CACHE[key] = cached
return copy.deepcopy(cached)
def get_variants(
*, roles_dir: str | os.PathLike[str] | None = None
) -> dict[str, list[Any]]:
"""Return ``{application_id: [variant_0, ...]}`` cached per
``roles_dir``. Each variant is the role's effective configuration
after the corresponding `meta/variants.yml` override has been
deep-merged on top of `meta/services.yml`."""
resolved_roles_dir = _resolve_roles_dir(roles_dir=roles_dir)
key = _cache_key(resolved_roles_dir)
cached = _VARIANTS_CACHE.get(key)
if cached is None:
cached = _build_variants(resolved_roles_dir)
_VARIANTS_CACHE[key] = cached
return copy.deepcopy(cached)
def _build_variant_overrides_only(
roles_dir: Path,
) -> dict[str, list[dict[str, Any]]]:
overrides: dict[str, list[dict[str, Any]]] = {}
for role_dir in _iter_application_role_dirs(roles_dir):
meta_path = role_dir / ROLE_FILE_META_VARIANTS
overrides[role_dir.name] = _load_variants_overrides(meta_path)
return {key: overrides[key] for key in sorted(overrides)}
def get_variant_overrides_only(
*, roles_dir: str | os.PathLike[str] | None = None
) -> dict[str, list[dict[str, Any]]]:
"""Return ``{application_id: [override_0, ...]}`` — the raw
`meta/variants.yml` entries per role WITHOUT deep-merging
`meta/services.yml`.
Used by matrix-deploy's `--vars` bake so the host_vars payload
stays sparse: only the variant-specific overrides land there,
leaving `meta/services.yml` fields (notably `image`/`version`)
blank in host_vars so `apply_mirror_overrides` can populate them
from `mirrors.yml`.
"""
resolved_roles_dir = _resolve_roles_dir(roles_dir=roles_dir)
key = _cache_key(resolved_roles_dir)
cached = _VARIANT_OVERRIDES_ONLY_CACHE.get(key)
if cached is None:
cached = _build_variant_overrides_only(resolved_roles_dir)
_VARIANT_OVERRIDES_ONLY_CACHE[key] = cached
return copy.deepcopy(cached)
def get_merged_applications(
*,
variables: dict[str, Any] | None = None,
roles_dir: str | os.PathLike[str] | None = None,
templar: Any = None,
) -> dict[str, Any]:
# Late import: `get_merged_users` lives in the sibling `users` module
# and pulls user-domain machinery (token store, alias materialization,
# etc.) that this module's other entry points don't need. Importing
# at function scope keeps `import utils.cache.applications` cheap so
# the runner-host CLI path can use `get_variants` without paying for
# the user-domain transitive imports.
from .users import get_merged_users
variables = variables or {}
resolved_roles_dir = _resolve_roles_dir(roles_dir=roles_dir)
cache_key = (
_cache_key(resolved_roles_dir),
_stable_variables_signature(variables),
)
cached = _MERGED_APPLICATIONS_CACHE.get(cache_key)
if cached is not None:
return cached
defaults = get_application_defaults(roles_dir=resolved_roles_dir)
overrides = _resolve_override_mapping(variables, "applications", templar=templar)
merged = merge_with_defaults(defaults, overrides)
if getattr(_RENDER_GUARD, "applications", False):
# Re-entry via cross-lookup: return unrendered merged payload; the
# outer templar will resolve remaining Jinja at use-site.
return merged
_RENDER_GUARD.applications = True
try:
raw_users = get_merged_users(
variables=variables,
roles_dir=roles_dir,
templar=None,
)
rendered = _render_with_templar(
merged,
templar=templar,
variables=variables,
raw_applications=merged,
raw_users=raw_users,
)
finally:
_RENDER_GUARD.applications = False
_MERGED_APPLICATIONS_CACHE[cache_key] = rendered
return rendered
def _reset() -> None:
_APPLICATIONS_DEFAULTS_CACHE.clear()
_VARIANTS_CACHE.clear()
_VARIANT_OVERRIDES_ONLY_CACHE.clear()
_MERGED_APPLICATIONS_CACHE.clear()