Vishesh 'ironeagle' Bangotra
6c0d859a1e
- F11 PopulateRule field_name/collection_name corrections - F12 service adaptor update handles PopulatingRepository - F13 nested-references section; F14 authless /users/me guidance - F15 template ruff-clean; F16 stop vendoring .ruff_cache, ship .gitignore - document --template-style in build_recipe/spec_first_routing; refresh collected openapi-first + standards bundle (12 modules, 43 components) - add trial.md and trial_findings.md (F1-F16, re-trial A/B evidence)
8 lines
3.3 KiB
JSON
8 lines
3.3 KiB
JSON
{
|
|
"module": "infra_adaptors",
|
|
"content": {
|
|
"path": "infra_adaptors",
|
|
"docstring": "# Infra Adapters\n\n## Summary\n\nAll external systems are reached through adapters in `infra/`. Prefer an\nexisting internal library over writing a new adapter; where no library exists,\ncopy the Tier B scaffold from `templates/app`. See the Genericity Matrix for the\nTier A / B / C split.\n\n---\n\n## Database (MongoDB)\n\n- A single connection manager owns the client lifecycle.\n `setup_database(uri, db_name)` / `shutdown_database()`.\n- A repository **factory** selects the repository flavour:\n\n ```python\n repo_factory(collection, document, engine=None,\n populate_rules=None, cache=None)\n ```\n\n - base repository — plain CRUD;\n - populating repository — resolves referenced documents (foreign keys);\n - cached repository — an LRU/TTL cache in front of reads.\n\n- Referenced documents are described declaratively:\n\n ```python\n PopulateRule(field_name=\"owner\", collection_name=\"users\",\n nested_rules=[], max_depth=2, projection=None)\n ```\n\n- Transactions use a session context manager provided by the library.\n\n**Tier A — Library (Mongo Ops).** Do not write raw Motor calls in the domain;\nconsult the registry and the library's MCP before touching the driver.\n\n## HTTP response cache\n\n**Tier B — template.** Copy from `templates/app`; not a standalone library.\n\nA generic GET cache sits behind a decorator.\n\n```python\n@cache_response(ttl=60)\nasync def list_items(request): ...\n```\n\n- Keyed on path + query + user.\n- No-op when no store is wired.\n- `invalidate(*resources)` bumps a generation counter for write-triggered\n invalidation.\n\n## Event / progress bus\n\n**Tier B — template.** Copy from `templates/app`.\n\nA port with an in-memory default adapter (`publish`, `get_events`). Swap for a\ndurable broker in production (see Worker & Progress).\n\n## External clients (ingestion)\n\nExternal fetchers (files, mail, HTTP) are adapters that wrap ingestion\nlibraries. Construct them in the composition root and inject into\norchestrators.\n\n**Tier A — Libraries:** Omniread (formats) and Mail Intake (email).\n\n## LLM clients\n\n**Tier B — template.** Copy the pool scaffold from `templates/app`.\n\nWhen a pipeline calls a model provider, route through a pool adaptor that\nhandles provider selection, retries, and client lifecycle:\n\n```python\npool = setup_async_pool(provider=settings.llm_provider)\nresult = await pool.complete(model=\"...\", messages=[...])\n```\n\nProviders are selected by `LLM_PROVIDER` and configured by environment.\n\n## Key-value store\n\n**Tier B — template.** Copy from `templates/app`.\n\nFor report/snapshot storage, provide a pluggable store port with a Redis\nadapter and an in-memory fallback.\n\n## Rule\n\n- **Tier A (libraries: DB, DI, ingestion, mail, auth, docs, pipelines)** — use\n the library; writing a private copy is a defect. Query `find_library` first.\n- **Tier B (cache, progress bus, LLM pool, KV store, composition shape, worker,\n tests)** — these are template code, not libraries: copy them from\n `templates/app` rather than searching the registry.\n- **Tier C (models, business logic, pipeline families, rules)** — author them\n yourself.\n",
|
|
"objects": {}
|
|
}
|
|
} |