If a datagram is too large, the Butler will split it into multiple datagram fragments. It is up to the client to identify this condition, request those fragments, and reassemble the original datagram.
Currently, this logic isn't implemented. Datagram consumers so far (namely Milo) basically discard partial datagrams as errors with the assumption that no datagram will actually be big enough that it needs splitting.
However, this logic should be implemented in a central place and installed in datagram consumers so they can handle this perfectly valid case.
Comment 1 by bugdroid1@chromium.org
, Sep 22 2016