keble-keepa

The source line requires keble-helpers>=1.54.0 so exact Keepa token/call usage uses the same public contract in development and registry installs. Provider pricing remains outside this package.

Side effects if changes:

• Amazon provider services and Data Infra transport these usage events.
• Changing the helper floor requires lock and downstream raw API verification.

keble-keepa is Keble's Keepa API and cache package. It owns Keepa object schemas, sync/async API access, Mongo cache CRUD, batch loading, time conversion helpers, and pytest fixtures/fakes for downstream packages.

Verified State

Checked on 2026-07-18 against pyproject.toml, package exports, keble_keepa/, and tests.

• Package: keble-keepa 1.4.3
• Python: >=3.13,<3.14
• Critical deps: aiohttp, requests, keble-db, keble-helpers, keble-exceptions, python-dateutil
• Source roots: api/, crud/, schemas/, testing/, utils/
• Public exports: Keepa API objects, batch loader, cache helpers, time conversion helpers, and the keble_keepa.testing pytest plugin.

Install the current public release with pip install "keble-keepa>=1.4.3,<2". Version 1.4.3 requires the shared keble-db>=1.9.0 persistence contract and the shared upstream-failure plus exact-money vocabulary in keble-helpers>=1.54.0.

Keepa Price Integer Boundary

Keepa price histories are normalized internally to fixed hundredths of a major currency unit, including synthetic * 100 normalization for JPY. Product converts that provider-specific integer to exact Decimal major units before constructing canonical Money; the shared ISO exponent then stores USD cents, JPY yen, and other currencies correctly. Never pass a Keepa price integer directly to Money.amount.

Runtime Contract

• Sync Keepa calls use requests; async Keepa calls use aiohttp and should respect standard proxy environment variables when configured by callers.
• Bestseller boards expose both request_best_sellers and genuine-aiohttp arequest_best_sellers; the async method uses the same typed Redis/Mongo cache key and emits usage only after real upstream I/O.
• Every typed response retains Keepa's tokensConsumed field. Async cache misses emit that exact provider token count through UsageAccountingRecorderProtocol; missing or zero reports remain USAGE_ONLY and therefore billing-unknown rather than being inferred free.
• Cache hits emit no usage event even though cached response metadata is retained for traceability.
• Short-lived Redis response serialization uses Pydantic JSON mode so exact Decimal refill/capacity values survive successful live calls and cache replay.
• Every physical Keepa request is attempted once. The package does not sleep, retry, recursively split failed batches, or silently drop failed ASINs; provider exceptions reach Data Infra intact and durable retry belongs to the Platform worker.
• Rejections retain HTTP status, request ID, typed Keepa error data, and exact capacity evidence. refillIn is milliseconds, negative tokensLeft remains valid evidence, and gross refill is distinct from token-flow reduction.
• Failed async HTTP responses emit provider-reported tokensConsumed usage before the typed exception is raised; non-JSON failures remain usage-unknown rather than being inferred free.
• Cache reads and writes must stay explicit about key, marketplace, and freshness behavior. Do not hide broad cache scans behind convenience APIs.
• Downstream tests should use keble_keepa.testing fixtures/fakes before inventing package-local Keepa mocks.

Testing

uv sync --group test
RUN_INTEGRATION=1 RUN_REAL_DB=1 RUN_LOCAL_STACK=1 RUN_DB_STACK=1 \
  uv run pytest -q -m "not live and not container"
uv run pytest -q -m "not live and not slow and not eval and not local_stack and not db_stack and not container"
uv run pytest -q -m eval
npx --yes pyright .

The first command is the local-full Keble development proof and runs Mongo-backed integration tests with isolated cache collections. The second is the portable-offline fallback for no-DB machines. Selecting -m eval is the eval opt-in; there is no generic RUN_EVALS gate.

Real Keepa calls belong only in marked/env-gated live tests. Package-owned test helpers live under keble_keepa.testing; do not recreate root-level tests/config.py, root test packages, or legacy tests/test_* source folders.

Mongo/Redis/Keepa live settings load explicit process env first, then tests/.env, then the umbrella keble.backend/.env through keble_db.testing.resolve_test_env. Set KEBLE_BACKEND_ENV_FILE=/path/to/.env from unusual worktree or CI layouts. Do not copy backend secrets into this repo.

Live Keepa canaries should make one real provider call against an isolated cache collection. Do not add Keepa response-cache replay to live tests for speed; cache correctness belongs in integration tests with fake/shared fixtures.

Async cache tests receive a process-unique namespaced ExtendedAsyncRedis and delete all owned keys after every test. Keep this isolation when adding new cache consumers; an un-namespaced wrapper is a test and production leak risk.

Change Logs

Historical release notes and long AI change history live under logs/.

Product release metadata

Product.release_date_observation parses Keepa YYYY, YYYYMM, and YYYYMMDD values into a typed interval with retained precision. Consumers must use the interval for recency decisions; they must not invent a month or day for partial dates or substitute offer-history timestamps. original_language_code normalizes Keepa's finite language names to ISO codes and returns None when the source does not declare a recognized language.

Side Effect If Changes

• Backend report generation, product discovery, keble-keepa-analysis, and AMZ product-report workflows depend on this package for Keepa access.
• Cache query/index changes can affect production cache repair and live report correctness; verify query shape and cleanup.