To determine which varnishes we consider "mobile" frontend varnishes
and need to be included into the mobile-sampled-100 stream, we
currently use [1] $active_nodes of from puppet's manifests/role/cache.

But it seems that variable holds only those caches that run both
frontend and backend varnishes. But when bringing new caches online,
it might be that they act only as frontend cache for some time
(manually through pybal), and hence are not yet added to
$active_nodes. Thereby, we loose logs of those
new-and-for-some-time-frontend-only machines.
(As it currently happens for bug 64145.)

Is it worth changing the way we detect mobile frontend varnish for
udp2log from puppet's $active_nodes to pybal?

(I guess no, but just wanted to have this documented somewhere. And as
what we currently do is a bug, hence using Bugzilla)

http://git.wikimedia.org/blob/operations%2Fpuppet.git/14357072e8d15e00a8a47306f92c698ab0a42515/templates%2Fudp2log%2Ffilters.oxygen.erb#L18

Version: unspecified
Severity: normal