Re: [Webpush] Use Case related to subscription sets

[Martin Thomson <[e-mail]>]

As Ben says here, I think that the question of limits is something
that is specific to a service.  The concurrent stream limit is a valid
way of limiting load, certainly.

However, I don't think that it makes the problem go away.  It's a
second order solution in the sense that it attempts to address the
consequences of bad behaviour.  I would rather have a service reject
attempts to create unlinked subscriptions.

An explicit link between subscriptions isn't that hard to do.  I've
created a pull request that illustrates how simple it would be to
accommodate Hervé's request.

https://github.com/webpush-wg/webpush-protocol/pull/67

On 23 November 2015 at 09:31, Benjamin Bangert <[e-mail]> wrote:
> On Mon, Nov 23, 2015 at 1:56 AM, Hervé Ruellan <[e-mail]>
> wrote:
>>
>>
>> On 20/11/15 18:49, Brian Raymor wrote:
>>>
>>>
>>> On Fri, Nov 20, 2015 at 8:43 AM, Benjamin Bangert
>>> <[e-mail]> wrote:
>>> > Since most push servers, as far as I know, will be using SSL, I
>>> don't know how a proxy could even function
>>> > since it would need to MITM the secured connection which is
>>> obviously not good.
>>>
>>> The mobile phone ("proxy") acts as the user agent in this case with a
>>> HTTPS connection to the push service. A different protocol than HTTP
>>> may be required between phone and the camera. Sounds similar to a
>>> field gateway scenario?
>>
>>
>> Yes, the HTTPS connection is between the mobile phone and the push server.
>> And another protocol is used between the camera and the phone.
>>
>>> > Either way, this situation and use-case seems accounted for by the
>>> spec as it is, since those wishing to handle
>>> > this scenario may run their own Push Server without a low concurrent
>>> stream max, and unsecured so that a
>>> > proxy may function.
>>>
>>> Regarding "unsecured", Section 9 requires HTTPS - This protocol MUST
>>> use HTTP over TLS [RFC2818].
>>
>>
>> For our scenario, we would like to be able to use a generic push server. I
>> agree that the spec supports the use-case, but we would also like the
>> implementations to support the use-case.
>> A short mention of the use-case in the spec would be really helpful. The
>> consequences for an implementation would be to allow a user agent to create
>> a *few* subscription sets, and not limit it to *1*, and also to allow a user
>> agent to open a *few* streams to subscription resources.
>
>
> Implementation-wise, the only reason a client using our service is
> restricted to one active subscription set at a time is due to the concurrent
> stream limit setting. It is however, just that, a setting.
>
> Someone could use our eventual implementation and set the stream limit
> setting to whatever they want to support at scale. Our specific deployment
> will restrict our clients to 2 concurrent streams via this setting.
>
> I'm assuming there will be multiple implementations out there to choose from
> if you don't want to implement your own, and many/most of them will likely
> implement the ability to restrict subscription sets in a similar manner...
> merely by setting a config flag.
>
> Cheers,
> Ben
>
> _______________________________________________
> Webpush mailing list
> [e-mail]
> https://www.ietf.org/mailman/listinfo/webpush
>