memcached does not wait until all items are drained when asked to shut down the engine (Data loss on shutdown)

In 1.8.1 we had this nailed. Shutdown was waiting for all dirty items to hit disk.

With latest 2.0 I'm not seeing this. Here's how I reproduced:

• load 3M items and shutdown all nodes while disk write queue is still > 0

• start nodes back and observe how many items are after warmup. In 2.0 it's always < 3M

Same thing works in 1.8.1.

Here's how I'm stopping all nodes for cluster_run environment: I'm POST-ing rpc:eval_everywhere(ns_bootstrap, stop, []). to /diag/eval. That's same API that our stop action of initscripts is doing.