On Mon, 07 Nov 2016 00:20:12 -0500, JackDrogon <jack.xsuperman@gmail.com>
wrote:
> Hi, All:
> I need varnish to return data directly and update cache at every time
> when the backend is ok. I also need varnish to return the last normal
> beresq data from >the cache when the backend is down. How shuold I do
> with varnish?
>
>> Thanks.
Hi Jack,
There are a couple of ways to approach this (as there almost always are
with a tool as flexible as Varnish). If traffic is generally light, you
could get away with something like this:
in vcl_backend_response, set
beresp.ttl = 1ms;
beresp.keep = 0s;
beresp.grace = 1h; # or as long as you would want to serve from cache
during an outage
The very short TTL should translate into no hits as long as you don't get
a lot of requests within a 1ms window. But this might not be exactly the
behavior you want.
If you have a heavily loaded scenario where you would get multiple hits in
this period, then you could simply put some logic into vcl_hit that looks
at the backend health, and forces a cache miss if the backend is healthy,
and otherwise honors the TTL/grace periods. In this scenario it wouldn't
matter as much what the TTL value is, but beresp.keep should stay set to
0s.
in vcl_recv,
if ( std.healthy(req.backend_hint) ) {
set req.http.allow-caching = "no";
} else {
set req.http.allow-caching = "yes";
}
in vcl_hit
if ( req.http.allow-caching == "no" ) {
return (fetch);
}
//... otherwise, carry on with the rest of the normal vcl_hit comparisons
of ttl, grace, and keep...
Of course these examples don't take into account the behavior of what to
do with set-cookie and other headers on the response, but for simplicity
I'm assuming here that it's ok to cache the responses as-is. You could
always strip headers out during vcl_deliver on the basis of the
req.http.allow-caching header, or more generically on the obj.hits
counter, if necessary.
Best,
Mark