HTTP caching and ETAG header - caching

Question

After a lot of unsuccessful research, I'm wondering if it's possible to use nginx or varnish cache function to check data freshness it contains using etag attribute in the header.
This scenario is easier to explain with a diagram, please see below :
http://image.noelshack.com/fichiers/2014/49/1417684577-cache-nginx.png
Do you think this use case is feasible ?
Thank you and sorry for my english :s

Answer 1

You can do something similar with varnish, stale-while-revalidate, but not quite the same. The idea with the front caches is that when the backend sends an object to the cache it sets an Expires header that will allow the cache to serve this content without bothering the backend. If the cache would check every request with the backend then there wouldn't be much speed to gain, as you would end up waiting for the backend on each request.
To solve this, you should set your Expires headers accordingly, and/or purge/ban the urls if suddenly you need to get rid of stale content before Expires. That way the cache would send the content from its own cache, or get from the backend if it doesn't have it or it's stale.
Edit:
I have tested with Varnish 4.0.1, it supports conditional requests to the backend.
My test setup has been an Apache virtual hosts with ExpiresByType text/html "access plus 5 seconds"
The result:
* << BeReq >> 65539
- Begin bereq 65538 bgfetch
- Timestamp Start: 1418403624.141695 0.000000 0.000000
- BereqMethod GET
- BereqURL /prueba.html
- BereqProtocol HTTP/1.1
- BereqHeader User-Agent: lwp-request/6.03 libwww-perl/6.05
- BereqHeader X-Forwarded-For: 127.0.0.1
- BereqHeader Host: localtest
- BereqHeader Surrogate-Capability: key=ESI/1.0
- BereqHeader Accept-Encoding: gzip
- BereqHeader If-Modified-Since: Thu, 01 Dec 2011 10:37:55 GMT
- BereqHeader If-None-Match: "3e8be5-46-4b30571f606c0"
- BereqHeader X-Varnish: 65539
- VCL_call BACKEND_FETCH
- VCL_return fetch
- Backend 17 vdir server1(127.0.0.1,,80)
- Timestamp Bereq: 1418403624.141869 0.000174 0.000174
- Timestamp Beresp: 1418403624.143205 0.001510 0.001336
- BerespProtocol HTTP/1.1
- BerespStatus 304
- BerespReason Not Modified
- BerespHeader Date: Fri, 12 Dec 2014 17:00:24 GMT
- BerespHeader Server: Apache
- BerespHeader ETag: "3e8be5-46-4b30571f606c0"
- BerespHeader Expires: Fri, 12 Dec 2014 17:00:29 GMT
- BerespHeader Cache-Control: max-age=5
- TTL RFC 5 -1 -1 1418403624 1418403624 1418403624 1418403629 5
- BerespProtocol HTTP/1.1
- BerespStatus 200
- BerespReason OK
- BerespHeader Last-Modified: Thu, 01 Dec 2011 10:37:55 GMT
- BerespHeader Content-Type: text/html
- VCL_call BACKEND_RESPONSE
- TTL VCL 5 21600 0 1418403624
- VCL_return deliver
- Storage malloc s0
- ObjProtocol HTTP/1.1
- ObjStatus 200
- ObjReason OK
- ObjHeader Date: Fri, 12 Dec 2014 17:00:24 GMT
- ObjHeader Server: Apache
- ObjHeader ETag: "3e8be5-46-4b30571f606c0"
- ObjHeader Expires: Fri, 12 Dec 2014 17:00:29 GMT
- ObjHeader Cache-Control: max-age=5
- ObjHeader Last-Modified: Thu, 01 Dec 2011 10:37:55 GMT
- ObjHeader Content-Type: text/html
- BackendReuse 17 server1(127.0.0.1,,80)
- Timestamp BerespBody: 1418403624.143465 0.001770 0.000260
- Length 70
- BereqAcct 289 0 289 181 0 181
- End
Here you can see that I ask for a stale object, so varnish goes to fetch it, and request it this way:
- BereqURL /prueba.html
- BereqHeader If-Modified-Since: Thu, 01 Dec 2011 10:37:55 GMT
- BereqHeader If-None-Match: "3e8be5-46-4b30571f606c0"
The backend answers with:
- BerespStatus 304
- BerespReason Not Modified
- BerespHeader Date: Fri, 12 Dec 2014 17:00:24 GMT
- BerespHeader ETag: "3e8be5-46-4b30571f606c0"
- BerespHeader Expires: Fri, 12 Dec 2014 17:00:29 GMT
- BerespHeader Cache-Control: max-age=5
So no body bytes are transferred, and varnish refreshes it's object with the newly updated headers.
Perhaps that's the behavior you wanted.
Keep in mind that a GET request that does validate the conditions (If-Modified-Since and If-None-Match) would only return a 304 Not Modified, some headers and no body (exactly the same as if it would be a HEAD request with the same conditions), if it doesn't validate it will return the headers and the new body. But a HEAD requests that does not validate would return 200 OK, and NO body, forcing you to re-request it with a GET.
HTTP/1.1 Method Definitions RFC:
The HEAD method is identical to GET except that the server MUST NOT return a message-body in the response. The metainformation
contained in the HTTP headers in response to a HEAD request SHOULD be
identical to the information sent in response to a GET request.