Re: Key Update implementation priority

[Marten Seemann <[e-mail]>]

Key updates as defined by RFC 9000 are not an optional feature. Any
implementation that doesn't implement key updates is not compliant with the
RFC and has to expect that transfers will randomly break. Even worse,
ignoring a key update won't immediately kill the connection, but will
eventually result in a connection timeout.

quic-go is very aggressive and initiates the first key update after the
first 100 short header packets. Subsequent key updates are triggered every
100k packets.

On Sat, 22 Jul 2023 at 07:05, Nick Banks <nibanks=
[e-mail]> wrote:

> Hello,
>
> For MsQuic, we have a hook to force a key update that can be used by test
> code. Additionally, we have a knob to set a smaller threshold for the
> number of bytes transferred to trigger the update scenario. Feel free to
> look at our test code here
> <https://github.com/microsoft/msquic/blob/main/src/test/lib/HandshakeTest.cpp#L2196>
> .
>
> Thanks,
> Nick
>
> Sent from Outlook <http://aka.ms/weboutlook>
> ------------------------------
> *From:* QUIC <[e-mail]> on behalf of Kaushal Bhandankar <
> [e-mail]>
> *Sent:* Saturday, July 22, 2023 3:04 AM
> *To:* [e-mail] <[e-mail]>
> *Subject:* Key Update implementation priority
>
> You don't often get email from [e-mail]. Learn why this is
> important <https://aka.ms/LearnAboutSenderIdentification>
>
> I would like to know how popular QUIC implementations force key-update to
> evaluate the need to implement it.
>
> Chromium
> Firefox Browser (neqo)
> mvfst
> quiche
>
> Regards,
> Kaushal
>