I would also add weights so that each varnish prefers to ask its sibling
say 250 times more often than asking the app-server.
This will ensure that both varnishes will be as filled and as equal as
possible.
I don't know if 250 could be replaced by a larger number but this works
for me.
Regards,
Martin
On 02/10/2011 02:46 PM, Laurence Rowe wrote:
On 8 February 2011 23:05, Ben Dodd<[email protected]> wrote:
Given this functionality, does anyone know what the VCL is required to
implement this dual server Varnish setup scenario?
http://varnish-cache.org/trac/wiki/VCLExampleHashIgnoreBusy
Something like this should work:
backend appserver { ... }
backend other_varnish { ... }
# Server 1
sub vcl_recv {
set req.backend = other_varnish;
if (req.http.X-Via-Varnish || !req.backend.healthy) {
# Request is coming from the other server.
# Ignore busy objects to avoid race condition.
set req.hash_ignore_busy = 1;
set req.backend = appserver;
} else {
set req.http.X-Via-Varnish = "true";
}
}
# Server 2
sub vcl_recv {
set req.backend = other_varnish;
if (req.http.X-Via-Varnish || !req.backend.healthy) {
# Request is coming from the other server.
set req.backend = appserver;
} else {
set req.http.X-Via-Varnish = "true";
}
}
Laurence
_______________________________________________
varnish-misc mailing list
[email protected]
http://www.varnish-cache.org/lists/mailman/listinfo/varnish-misc
_______________________________________________
varnish-misc mailing list
[email protected]
http://www.varnish-cache.org/lists/mailman/listinfo/varnish-misc
