num_replicas works when server goes down but not when it comes back up

[Michael Verrilli]

I'm using pylibmc but I think this is just a wrapper for functionality in libmemcached.

The situation is:

1. create a connection with num_replicas:1 on two servers (server_1, server_2)
2. add a key to this connection (note the this connection does show this new value, server_1 directly does not, server_2 directly does)
3. stop server_2
4. check the main connection for the key. Connection error received.
5. check the main connection again for the key, key is found so replicas did work (checking server_1 directly also shows the key)

-- all good so far --

Now, bring back up server_2.
check the main connection for the key. Not found!
check server_2 directly for the key, not found.
check server_1 directly for the key, found!

It seems to me that in this situation, the main connection should find the key still, regardless of server_2 coming back up. I am imagining situations where people have an autostart in case the service goes down.