This required a ton of changes including updating error handling and separating out the models into intermediate representations so that fields that are marked non-null in the database are not `Option` in the final model.
The update allows using `query_as!` in `interior_ref_list` and `merchandise_list` and model functions to specify a generic `Executor` param that can take either a db pool connection, transaction, or plain db connection. This should allow me to impl my old `Model` trait again.
Also compile times are magically 20x faster?!?
Uses `tokio::spawn` to delay updating the cache while the server responds to the request.
Because `tokio::spawn` can run on another thread, references need to be static, so I initialized the cache in `lazy_static`.
Now the client can opt out of receiving the whole JSON body if it hasn't changed since they last requested.
Right now, only `ETag` and `If-None-Match` headers are implemeted which isn't very RFC-spec compliant but it's all I need so I don't care.
Creates the transaction record and updates the merchandise quantity in one db transaction.
Managed to do the merchandise update in one UPDATE query, but the error that's thrown when an item to buy is not found is pretty confusing, so I convert it to a 404.
I also added some DB indexes.
Use `refinery_cli` against a folder of `.sql` migrations.
I got tired of commenting out my code when I just wanted to rerun the initial migration.
Plain SQL is a lot more flexible than the `barrel` syntax.
Caches responses of each GET handler in a separate capacity-limited cache (as a
custom clone-able `CachedResponse` struct). Subsequent requests will build a
`Response` from the cached bytes instead of re-querying the database and
re-serializing the JSON. This greatly speeds up the list endpoints and
`get_interior_ref_list`.
Also caches the api-key-to-id mapping for `Owner`s in order to speed up frequent
authentications.
Each create handler clears the entire list response cache. Each delete handler
also clears the entire list response cache and deletes the cached response for
that key. Deleting an owner also deletes their entry in the
`owner_ids_by_api_key` cache.
Ran into some limitations of sqlx while trying to bulk create interior_refs. I
also discovered how slow creating hundreds of rows in postgres is and I'm
planning on saving interior_refs data in a jsonb column instead which seems to
be much faster.
Pretty comfortable with the choice of crates now so it's time to start
committing.
Currently the API only returns errors, but throwing good errors is important.
