Pip install Postgres – no Docker/Brew/apt

Summary

Simple Postgres helpers for testing with Python - no docker, brew, apt, etc - uses postgresql-binaries. The interface is simple, but close enough to the metal that you can use it to eg. have a new database per testing thread/use a fresh database from a TEMPLATE or archive. pip install 'postgresql-testing[15]' Then to use with pytest, eg: import postgresql_testing @pytest.fixture(scope="session") def db() -> Iterator[str]: config = postgresql_testing.DatabaseConfig.default("testing-db") postgresql_testing.initdb(config.directory, on_existing="use") with postgresql_testing.serve(config): postgresql_testing.ensure_user(config) postgresql_testing.create_database(config, on_existing="replace") yield config.dsn There are various useful flags and things - the source code is short enough to just dive in. You can spin up a fresh db for immediate testing with: postgresql-testing-serve postgres://testing:@localhost:8421/my-db Or if you want a throwaway Postgres from anywhere: uvx --from 'postgresql-testing[16]' postgresql-testing-serve Bundled for good measure is a local-first copy of explain.dalibo.com Get your query plan by: EXPLAIN (ANALYZE, COSTS, VERBOSE, BUFFERS, FORMAT JSON) SELECT ... Then copy and: pbpaste | postgresql-testing-explain There are a couple of helpers for creating/using template dbs and archives. I have some vague long term plan for some kind of "Docker layers for migrated databases" with clever caching (or not), but I'm not quite sure what it looks like yet. Some benchmarking: number of tables create from template dump to archive create from archive 100 80ms 80ms (0.2MB) 150ms 1000 500ms 300ms (2MB) 1100ms On macos using a ramdisk, it is slightly quicker: number of tables create from template dump to archive create from archive 100 70ms 70ms 120ms 1000 350ms 300ms 1000ms See claims/advice from planet Go.