Previously, a transaction panicking in the handle() call would cause the
transaction to be indefinitely logged as "running" in the transaction
handler. This means subsequent transactions from the sending server
would be met with the "You're still sending me a txn!" 429 response,
forever, effectively causing defederation between the two servers, until
continuwuity is fully restarted.
I could just fix the known panics in the handle function and internal
calls, however there's so many subsystems that could reasonably panic
that it is just better to have a safer approach and ensure that the
handle call is infallible.
As lk-jwt-service is hosted in a bridge network, the previous
`extra_hosts` using 127.0.0.1 would just loop back to the container
itself. Using `network_mode: host` is unnecessary, and the
`host-gateway` approach works well for both Podman and Docker
Host loopback issues are prevalent in non-Livekit scenarios too: push
gateways is another example. Considering moving these to a dedicated
Docker section of the troubleshooting page.
The token endpoint already refused grant types the client had not
registered, but reported it as `invalid_grant`. RFC 6749 section 5.2
reserves `invalid_grant` for an authorization grant which is "invalid,
expired, revoked, does not match the redirection URI used in the
authorization request, or was issued to another client", and defines
`unauthorized_client` for a client which "is not authorized to use this
authorization grant type".
Report the condition with the error code the specification assigns to it,
matching the device authorization endpoint.
The device authorization endpoint issued a device code to any registered
client, without checking that the client had registered the
`urn:ietf:params:oauth:grant-type:device_code` grant type. Such a client
could therefore start a device authorization flow and have the server
show an approval prompt to a user, even though the subsequent token
request was always going to be refused.
RFC 8628 section 3.2 states that, in the event of an error such as an
invalidly configured client, the device authorization endpoint responds
in the same way as the token endpoint specified in RFC 6749 section 5.2,
which defines `unauthorized_client` as "the authenticated client is not
authorized to use this authorization grant type".
Apply the same grant type check the token endpoint already performs, and
add the `unauthorized_client` error code it requires.
timedout