For sending, we should require that a client stop sending more 0-RTT before they hit the limit (this means EOED needs to be sent within the limit, so they need to reserve some space). We might note that most of the current limits for writing are larger than the largest value for max_early_data, so this is unlikely to be necessary.
For receiving, we should require that a server track failed decryptions against the per-connection limit, noting that this only applies to DTLS.
See https://github.com/tlswg/dtls13-spec/issues/260
For sending, we should require that a client stop sending more 0-RTT before they hit the limit (this means EOED needs to be sent within the limit, so they need to reserve some space). We might note that most of the current limits for writing are larger than the largest value for max_early_data, so this is unlikely to be necessary.
For receiving, we should require that a server track failed decryptions against the per-connection limit, noting that this only applies to DTLS.