This PR brings over most of the test suite improvements I made
[here](https://github.com/payloadcms/enterprise-plugins/pull/249) to our
payload monorepo.
- Replaces mongodb-memory-server with an actual mongo db using the
mongodb-community-server docker image. This unblocks the vitest
migration PR (https://github.com/payloadcms/payload/pull/14337).
Currently, debugging does not work in that PR - this is due to the
global setup script that has to start the mongo memory db.
- Just like postgres, all mongodb databases now support vector search.
mongodb-atlas-local supports it natively, and for
mongodb-community-server, our docker compose script installs `mongot`,
which unlocks support for vector search. This means we could add [vector
storage/search tests similar to the ones we have for
postgres](https://github.com/payloadcms/payload/blob/main/test/database/postgres-vector.int.spec.ts)
- Int tests now run against both mongodb adapters: mongodb
(mongodb-community-server) and mongodb-atlas (mongodb-atlas-local)
- Adds docker scripts for mongodb, mongodb-atlas, and postgres,
documented in README.md. Updates default db adapter URLs to
automatically pick up databases started by those scripts. This makes it
easier for people cloning the repo to get started with consistent
databases matching CI - no complicated manual installation steps
- Simplified db setup handling locally in CI. In CI, everything is
scoped to `.github/actions/start-database/action.yml`. Locally,
everything is scoped to `test/helpers/db`. Each database adapter now
shares the same username, password and db name
- Use consistent db connection string env variables, all ending with
_URL
- Updates the CONTRIBUTING.md with up-to-date information and adds a new
database section. We now recommend everyone to use those docker scripts
Follow up to #12404.
Templates include a custom route for demonstration purposes that shows
how to get Payload and use it. It was intended that these routes are
either removed or modified for every new project, however, we can't
guarantee this. This means that they should not expose any sensitive
data, such as the users list.
Instead, we can return a simple message from these routes indicating
they are custom. This will ensure that even if they are kept as-is and
deployed, no sensitive data is leaked. Payload is still instantiated,
but we simply don't use it.
This PR also types the first argument to further help users get started
building custom routes.
Alessio Gravili