Serve cached object without assembling ESI?

[Danila Vershinin]

Hi,

I'm trying to understand whether it's feasible to build a CDN of Varnish instances, whose primary origin(backend) is a Varnish instance which processes an ESI-capable app.
So basically:
USA visitor -> Varnish USA -> Varnish primary -> THE APP with ESI blocks
RUS visitor -> Varnish RUS -> Varnish primary -> THE APP with ESI blocks
CAN -> Varnish primary -> THE APP with ESI blocks

The APP emits HTML with ESI tags for different blocks, and sure enough, those have different cache policies.
If a request arrives to an edge Varnish, I want it to talk to the primary Varnish (so as to cache on the edge against a primary cache), but this "Varnish primary" serves the content with ESI already assembled. Thus there is no telling, how long we should cache that full page HTML on the edges.

Q: Is there a magic way to instruct the primary Varnish to not do ESI processing on the content that was already cached ??
Meaning keep things as usual if request arrives directly to primary Varnish (ESI on), but receive non-assembled HTML (with esi tags inside) otherwise, so that edges can do assembling themselves, and thus cache things on the edge with proper TTLs for blocks.

Or I just have to resort to completely disabling do_esi on the primary server? Which I could do, but that must mean that the primary Varnish cannot serve visitors directly (= loosing the edge location with lowest latency to nearby users).
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <https://www.varnish-cache.org/lists/pipermail/varnish-misc/attachments/20190915/0eb889b8/attachment.html>