danthegoodman1 / sqlgateway Goto Github PK

Need to be able to auth users accessing the nodes.

Probably best to do a Basic Auth in the form of a DSN like http(s)://USER:[email protected] like DBs do. That way a config can just be a single DSN rather than multiple parameters for things like future client libraries.

Similar to Firestore security rules, Supabase RLS rules, etc.

We could be given a JWT signing key, and a then write rules based on some JSON property of the decoded token, that allows operations like looking up tables with a (for example) WHERE user_id = {JWT.user_id}.

This is a bit out of scope but an interesting feature to add for allowing frontends to directly access this.

Extensive automated testing for single and multi-node environments that is run on any request to merge to main, but can also be run locally.

Let the transaction timeout be defined in the /begin request body.

Separate issue from #10 since that refers mostly to system-level timeouts.

Queries should have caching options for how long an individual statement will be cached for. Should explicitly enable caching, and only allow for SELECT statements.

If the cache TTL changes, then we should invalidate and revalidate. 0 always busts cache and deletes it.

Use mailgun's groupcache and use Redis for service discovery to coordinate nodes so the cache size is the collective cluster? Or just use Redis as the cache?

Also set a max response size of the cache so we don't store massive responses.

stale-while-revalidate is a good addition that might be scoped to a separate issue.

Under the /redis prefix. Would need to introduce Redis connection pooling and general handling. Pipelining and MULTI/EXEC would probably come in a second pass.

Vote with 👍 if you want to see this feature so interest can be measured!

Under the /mysql prefix. Would require abstraction of transaction handling.

Vote with 👍 if you want to see this feature so interest can be measured!

Definitey make an interface as mentioned in #11 , and make etcd an option for handling clustering coordination (transactions)

Will be it's own repo, but tracking initial work here.

Should resemble node-postgres in terms of connect with a client, then call query with parameter injections.

Have a multi method that allows you to send the array of queries for atomic execution.

Should have identical query interface for normal and transaction queries.

Right now pretty much everything is set to 30 seconds. They should be configurable via env vars.

For transactions, maybe we allow it to be specific at BEGIN time.

Will probably have to make an interface for this, but we should support redis cluster as well.

Cover different ENV vars, behavior when single node vs. clustered, etc.

Also API docs.

Maybe even include some example kubernetes manifests.