QUIC Socket Buffers

global_settings.quic_socket_buffer requests the UDP receive and send buffer size for each QUIC listener, outbound QUIC proxy endpoint and DNS-over-HTTP/3 socket. It defaults to 8388608 bytes (8 MiB); allowed values are 65536 through 16777216 bytes. This is per socket, not per QUIC connection, and is separate from QUIC flow-control windows and userspace buffer admission limits. Ordinary UDP forwarding sockets keep their existing buffer sizes.

- global_settings:
    quic_socket_buffer: 8388608

Requests are best-effort: if the OS clamps or rejects the size, startup continues with the available buffers. Shoes checks the resulting sizes and warns at most once per receive/send direction per process when the request is not satisfied or cannot be verified. Smaller buffers can limit throughput during bursts.

On Linux, administrators can allow the default request with net.core.rmem_max=8388608 and net.core.wmem_max=8388608; Shoes does not change sysctls or require elevated privileges for buffer tuning. Linux reports double the requested buffer size for accounting overhead; Shoes normalizes this when checking the result. The limits do not preallocate their maximum memory, but queued traffic consumes memory across all endpoints. Raising kernel ceilings does not resize existing sockets. Configuration reloads recreate QUIC listeners and can disconnect their active sessions; see Hot Reloading.