Currently wdqs is defined directly in varnish as a pool of randomized backend hostnames. There should be a real service hostname for the internal service like wdqs.svc.eqiad.wmnet, which is defined in LVS with pybal controlling the pooling of the 3x backends, and then varnish's configuration should be updated to use that hostname rather than enumerating the backends directly. This involves a handful of complex puppet changes and LVS new-service deploys are always a bit "special" (requiring careful manual restarts).
Related Gerrit Patches:
|operations/puppet : production||wdqs - configure varnish to use LVS service as backend|
|operations/puppet : production||wdqs - add icinga check for LVS services|
|operations/puppet : production||wdqs - LVS configuration|
|operations/dns : master||wdqs LVS DNS entries|