Calling a remote service which is down freezes the caller

[debbabi]

When a remote peer is gone, a time interval is needed to detect its departure. In between, other peers could call its services as the proxies are always present. In this case, the caller will remains blocked until the a timeout is expired (normally this is a big interval of time).

To overcome this problem, all RPC calls should wait two messages from the target peer :
(1) an ack that the request is received by the target peer
(2) the answer of the target peer.

This should be handled transparently so that developers have not to worry about this details.
When the caller does not receive the first message (1) before a short time_out_1 interval, it directly exists with an exception. However, If the first message (1) is received before time_out_1, the caller remains blocked until it receives the answer (2) before a time_out_2.

Values of time_out_1 and time_out_2 should be given by the user as a startup configuration for each node (default value are used if no user-specifc value is given), as it depends on the network overlays and average time of processing the caller requests.