- read up on
context
package, maybe reuse context in several requests and/or use incoming request context
- separate authenticator into read and write authenticators, read for frontend (poll requests) and write for scene modification requests
- look into using rpc code in returned errors
- tls, musthave
- re-poll for new drawings immediately after getting non-empty response to get the whole scene faster