W3C home > Mailing lists > Public > ietf-http-wg@w3.org > January to March 2015

Re: Alt-Svc alternative cache invalidation (ext#16)

From: Mark Nottingham <mnot@mnot.net>
Date: Fri, 20 Feb 2015 15:24:37 +1100
Cc: Martin Thomson <martin.thomson@gmail.com>, "Julian F. Reschke" <julian.reschke@gmx.de>, HTTP Working Group <ietf-http-wg@w3.org>
Message-Id: <38E70447-193C-4F7A-8722-9019B6B20BC8@mnot.net>
To: Erik Nygren <erik@nygren.org>
Reading the thread again -- AIUI the intent is for invalidation to be scoped to a single discovery mechanism (the frame, a header, whatever).

If that's the case, the use cases below will work, because they both use different mechanisms.

So, I'm OK with this. We will need to be *very* careful to scope the invalidations, however.

Cheers,


> On 25 Aug 2014, at 10:30 am, Mark Nottingham <mnot@mnot.net> wrote:
> 
> So, to be clear, you're suggesting that both the Alt-Svc header field and the ALTSVC frame type have the side effect of cache invalidation?
> 
> Personally -- I'm not sure that's a good idea. 
> 
> For example, imagine a http:// service that a) wants to use Opp-Sec and b) the alternate wants to do some load balancing, etc.
> 
> The http:// service sets an Alt-Svc header field with a very long lifetime, so that Opp-Sec is as sticky as possible.
> 
> The alternate, OTOH, uses a fairly short lifetime for load balancing.
> 
> With cache invalidation, the alternate doing load balancing is going to clear the cache of the Opp-Sec hint, thereby forcing the client to go back to the http:// origin once the (short lifetime) load balancing policy expires.
> 
> Without invalidation, it'd fall back to the original Opp-Sec alternative.
> 
> Likewise for the SNI segmentation use case. 
> 
> Regards,
> 
> 
> On 24 Aug 2014, at 11:30 am, Erik Nygren <erik@nygren.org> wrote:
> 
>> On Fri, Aug 22, 2014 at 7:50 PM, Martin Thomson <martin.thomson@gmail.com> wrote:
>> On 22 August 2014 14:53, Erik Nygren <erik@nygren.org> wrote:
>>> but does not define anything similar for the ALTSVC frame.  Aligning the
>>> frame and the
>>> header would allow this to apply to both.
>> 
>> I think that we would want to move the Origin field up to the header
>> with Max-Age.  Logically, you store alternatives for different origins
>> separately, so requiring different frames makes sense there.  It also
>> removes any potential for duplication.
>> 
>> Also 8 bits of length is not sufficient for an HTTP origin if the name
>> is maximum size.  I'd assume that the same applies to authority.
>> 
>> 
>> Agreed on both counts.  What about this, then:
>> 
>>  0                   1                   2                   3
>>  0 1 2 3 4 5 6 7 8 9 0 1 2 3 4 5 6 7 8 9 0 1 2 3 4 5 6 7 8 9 0 1
>> +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+
>> |                          Max-Age (32)                         |
>> +---------------+---------------+-------------------------------+
>> | Origin-Len (16)               |         Origin? (*)         ...
>> +---------------------------------------------------------------+
>> |Num-Alt-Auth(8)|
>> +---------------+---------------+-------------------------------+
>> | Proto-Len(8)  |        Protocol-ID (*)                        |
>> +---------------+-----------------------------------------------+
>> | Alt-Auth-Len (16)             |        Alt-Auth (*)         ...
>> +---------------+-----------------------------------------------+
>> |                        Ext-Param? (*)                       ...
>> +---------------------------------------------------------------+
>> 
>> where Origin-Len=0 would be used in the case where this was part of a Stream != 0
>> and Num-Alt-Auth>=1.  The {Proto-Len, Protocol-ID, Alt-Auth-Len, Alt-Auth} would be
>> repeated Num-Alt-Auth times.  Alt-Auth is a string such as "server.example.com:443"
>> 
>> 
>> 
>> 
>> 
>> 
> 
> --
> Mark Nottingham   https://www.mnot.net/
> 
> 

--
Mark Nottingham   https://www.mnot.net/
Received on Friday, 20 February 2015 04:25:10 UTC

This archive was generated by hypermail 2.4.0 : Friday, 17 January 2020 17:14:43 UTC