* If the client connects to an overloaded Simple Push node, the node should return a `"redirect"` response containing the URL of a peer node in the cluster. * If the entire cluster is full, the node should redirect the client to a node in a different cluster. * Simple Push nodes should publish their connection count (or threshold) to a shared etcd cluster every minute. * etcd fetches should be staggered (perhaps every 5 minutes) and cached in memory to avoid excessive queries.
Is this specifically Simple Push or Loop Push /me head is exploding...
Loop Push (I wrote "Simple Push" out of habit, since they use the same codebase, but only Loop Push needs redirects enabled). Apologies for the confusion.
:oremj and I agreed this should land in both clients, too. Unlikely to make it in until at least Firefox 36.
Note that this feature is probably of limited utility, at least for Loop, for reasons I describe in Bug 1094433 comment 4.
No longer valid with the new Push server.