Well, Nextone for instance has a database the keeps most of the state of calls, and it’s replicated between the two nodes. (I seem to recall the database was GNU dbm, but I might be mistaken.) However, as of 4.3 anyways, the CDRs still get truncated when there’s any kind of switchover.

BTW in Nextone v4.0.x the GNU db is used for storing configuration data like storing routes & other bits which is then loaded into memory. Nextone 4.3 and above uses postgres for this configuration data. The actual call state information is stored in memory and replicated to the standby box via some custom network protocol.

Stateful call migration would be a very useful feature in FS but I imagine its way down the roadmap.

But as to the original question of live upgrades, having some form of load balancing proxy and then bleeding off traffic from the box you want to upgrade is the most feasible approach, as others have mentioned.

Steve