Optimize schema check storage of composite sdl and supergraph

[n1ru4l]

Right now we store the composite SDL and supergraph on every record within the schema_checks table.

When a schema check runs that has no active changes within the schema, we store a lot of redundant data within our database. E.g. bots like renovate cause a lot of these runs. Some GraphQL schemas are quite huge.

A lot of schema_checks are linked to the schema_versions table, which also stores the composite SDL and supergraph.
In case a schema check has no active changes (read as the same checksum), we could omit to store the composite SDL and supergraph within the schema_checks table and instead retrieve them from the schema_versions table when consuming the field via the GraphQL API.

A further optimization could be to even store schemas and their checksums in a completely new table, e.g. checksum_schemas, that is a checksum to SDL mapping. That way we could furthermore reduce the redundancy of large SDL strings.

Note: If we do the latter, we also need to make sure the cleanup happens when the checksum entry is no longer referenced by any schema_checks entry, which would furthermore complicate the database schema.

[andriihrachov]

would like to take it

[n1ru4l]

Hey @andriihrachov, feel free to send a pull request! I would be happy to review it ☺️

[kamilkisiela]

@andriihrachov before touching code, do you mind sharing with us when do you plan to work on it and how you want to implement it?

[andriihrachov]

@kamilkisiela sorry I missed your message, but opened a PR with first steps (migration, recording schema SDL and checksum in separate table, etc,) - it's small enough to kill if I took wrong way, pls feedback :)