On Jan 14, 2012, at 11:07 AM, Poul-Henning Kamp wrote:
>> Couple of points.
>>
>> I agree with all, except I think they should be delivered through =
>> vcl_deliver, or you have to duplicate header logic in both of them.
>
> Presently we allow limited obj.* access in vcl_deliver{}, but that's
> actually limited to the obj.hits and obj.lastuse variables which are
> under dispute for causing write-backs to object storage.
>
I would love to keep the obj.hits and obj.lastuse in the struct obscure.
> So if we eliminate those two, using vcl_deliver{} should be feasible.
>
>> Regarding ESI, as long as an ESI can have an fragment be a synth reply, =
>> I am happy.
>
> It can, but the synth reply cannot be ESI parsed.
>
> I did consider the alternative, where you call synth() from vcl_recv{}
> and then go from there to vcl_fetch{}, in essense making synth{} a
> virtual backend.
>
> Would that be better ?
What happens in an error and you want to go vcl_synth?
Artur
_______________________________________________
varnish-dev mailing list
[email protected]
https://www.varnish-cache.org/lists/mailman/listinfo/varnish-dev
