Frequently Asked Questions

API/Design

Why have a separate engine for server and client? Surely traffic could be differentiated as much as it needs to be internally in one engine?

The traffic cannot be differentiated for gQUIC versions Q046 and Q050. This is because in these versions, the server never includes a connection ID into the packets it sends to the client. To have more than one connection, then, the client must open a socket per connection: otherwise, the engine would not be able to dispatch incoming packets to correct connections.

To aid development, there is a LSQUIC_FORCED_TCID0_VERSIONS that specifies the list of versions with 0-sized connections. (If you, for example, want to turn them off.)

Once gQUIC becomes deprecated in the future, there will remain no technical reason why a single engine instance could not be used both for client and server connections. It will be just work. For example, the single engine settings lsquic_engine_settings will have to be separated into client and server settings, as the two usually do need to have separate settings.

Example Programs

http_client does not work with www.google.com, www.facebook.com, etc.

Check the version. By defaut, http_client will use the latest supported version (at the time of this writing, “h3-31”), while the server may be using an older version, such as “h3-29”. Adding -o version=h3-29 to the command line may well solve your issue.

There is an outstanding bug where lsquic client does not perform version negotiation correctly for HTTP/3. We do not expect this to be fixed, because a) this version negotiation mechanism is likely to become defunct when QUIC v1 is released and b) version negotiation is not necessary for an HTTP/3 client, because the other side’s version is communicated to it via the Alt-Svc HTTP header.