Skip to content

testing.md

Latest commit

 

History

History
85 lines (66 loc) · 3.61 KB

testing.md

File metadata and controls

85 lines (66 loc) · 3.61 KB
Raw

Testing

English | 中文

Running the suite

The tests need a reachable PostgreSQL server. They never touch your real application database — tests/conftest.py creates a uniquely-named rpa_test_* database for the session, runs every test against it, and drops it at teardown.

export PG_TEST_URL=postgresql://user:pass@127.0.0.1:5432/postgres
make test          # equivalent to: pytest tests/ -q

PG_TEST_URL only needs to point at a database server you have credentials for (it connects to postgres as an admin database to issue CREATE DATABASE/DROP DATABASE) — it is never used as the database tests actually run against. If PG_TEST_URL isn't set, RAP_DATABASE_URL is used for the same purpose, and finally a postgresql://rpa:rpa@127.0.0.1:5432/rpa fallback.

Safety rails baked into conftest.py:

  • _db_url() refuses to build a connection string for any database name that doesn't start with rpa_test_ (or isn't the admin postgres db).
  • The teardown's DROP DATABASE call re-checks the rpa_test_ prefix before dropping anything.
  • Between tests, an autouse fixture truncates every data table — but only after confirming the active database is an rpa_test_* one.

This means pointing PG_TEST_URL at a shared/staging Postgres server is safe: the worst that can happen is a throwaway rpa_test_<uuid> database being created and dropped.

What the suite covers

  • ETL: platform detection, normalization, de-duplication on re-ingest (test_full_row_uniqueness.py, test_multiplatform.py, test_platform_raw_tables.py)
  • API endpoints: auth, uploads, analytics, SQL console, admin (test_api_endpoints.py, test_upload_validation.py, test_upload_background.py, test_sql_console.py)
  • Self-media ingestion: WeChat, 小红书, 知乎 (test_xhs.py, test_xhs_accounts.py, test_zhihu.py)
  • Caching, rate limiting, structured logging, content-impact analytics

The synthetic dataset

Several e-commerce tests assert against exact aggregate numbers (order counts, revenue totals, top SKU). Rather than depending on a real exported order file, those tests use a small, fully fabricated dataset in tests/sample_dataset.py:

from sample_dataset import synthetic_youzan_df

df = synthetic_youzan_df()   # pandas DataFrame, raw 有赞 (Youzan) column headers

It has known, documented properties tests assert against directly — e.g. SAMPLE_CUSTOMER_PHONE, SAMPLE_CUSTOMER_JULY_ORDER_COUNT, SAMPLE_CUSTOMER_JULY_TOTAL are exported alongside the dataset so assertions don't need magic numbers scattered through test files. If you add a test that needs different aggregate shapes, prefer extending this module over inlining another ad-hoc CSV string.

Tests that are skipped by default

tests/test_media_upload.py covers a WeChat xlsx-upload code path that's disabled in this codebase (WeChat data arrives via the official API sync instead — see WeChat auto-sync); the whole module is marked pytest.mark.skip. Remove the module-level pytestmark line to re-enable it if you bring that upload path back into use.

There is intentionally no test fixture that depends on a real customer export file shipped in the repo — data/ is gitignored, and every test that needs order-shaped input uses the synthetic dataset above instead.

Fast password hashing in tests

conftest.py swaps in a trivial password helper (RAP_TEST_FAST_PASSWORDS=true, set by default for the test session) so tests don't pay real bcrypt cost on every login. This only takes effect when that env var is set — it never affects a normal run of the app.