On Thu, Apr 17, 2014 at 10:19:02AM -0600, Kevin Fenzi wrote:
The first patch here looks fine to me... but the second one isn't going to work I don't think.
Those services use memcached, but they use our existing memcached03/04 instances that are still in puppet, not local memcached.
Not all. Nuancier seems to use localhost : http://infrastructure.fedoraproject.org/cgit/ansible.git/tree/roles/nuancier...
So does pkgdb : http://infrastructure.fedoraproject.org/cgit/ansible.git/tree/roles/pkgdb2/t...
( unless i missed something and pkgdb2 and nuancier run on a memcache server )
But indeed, badges do use memcached03, I wonder how I missed that :/
I guess we should think about how we want to use memcached moving forward some. Do we want dedicated instances? Or do we want to prefer using local caches on each applications servers? Using dedicated servers means we could share more, but local might still be good enough...
Shared servers would help if more than 1 server use it, like 2 redundant applications.
But having local memcache permit to know how much memory each service use, and to restart them selectively. It would also reduce the network trafic and likely improve latency. ( depending on the topology, I didn't really look at it )