Test tiers

pytest and make test run the standard merge suite. It is deliberately limited to quick correctness coverage: it should finish in minutes and stay within roughly 8 GiB total RSS on a 16 GiB machine. Full-depth parameter cases are marked extended and are skipped by this default profile.

Run make test-extended after larger implementation work. It includes every test case and has a twenty-minute timeout by default. Both targets run under the python -m httk.core.memguard process-group guard supplied by httk-core; HTTK_TEST_MAX_RSS_GB can override its limit when diagnosing a failure.

ClickHouse is an opt-in test arm. The exact server/client memory arithmetic, the 7 GiB server allowance, the 4.5 GB allocator cap, setup, required KeeperMap bootstrap DDL, and recovery procedures are in the ClickHouse testing guide. Start it with make clickhouse-dev-server and stop it with make clickhouse-stop.

PostgreSQL is another opt-in test arm. Set HTTK_TEST_POSTGRES_URI to a reachable admin URI (postgresql+psycopg:// driver) and the parameterized backend suites include it; leave it unset and every PostgreSQL test skips. The PostgreSQL testing guide covers the server setup and the one known bulk-ingest limitation. Start a local server with make postgres-dev-server and stop it with make postgres-stop.

MongoDB tests require HTTK_TEST_MONGODB_URI to name a reachable replica set, for example mongodb://127.0.0.1:27127/?replicaSet=rs0. Install httk-store[mongodb] and enable enableTestCommands=1 on a dedicated test server to exercise transaction fault injection. The MongoDB CI job starts this isolated service and runs both profiles. Without the URI, MongoDB tests skip; these skips do not verify the backend. See the MongoDB guide for its transaction and deployment requirements.

Performance investigations are separate from correctness testing. make benchmarks runs the opt-in harnesses in benchmarks/; benchmark code is not collected by pytest and is never invoked by test, check, or ci.