Simplify the logic of compute added and removed addresses in routing table

[zhenlineo]

Terms:

• ConnectionPools which is a map from addresses to connection pools
• RoutingTable which contains addresses for routers, writers, and readers

For a routing driver, once an address is removed from the current RoutingTable, the address is no longer accessable. a.k.a. no new connection will be created in the corresponding connection pool.

When updating routing table, we also need to signal the addresses in ConnectionPools to be active if they are newly added into the current routing table or passive if they have already been removed from the routing table. For the pools connected to addresses that have been removed, when a connection is free, the connection should be terminated rather than reused. When there is no connection in the pool, the pool could be safely removed from ConnectionPools.

So the logic that need to be changed to:

• When a new RoutingTable is available, compute
  added_addr = distinct_addr_in(new_routingTable) - distinct_addr_in(pre_routingTable)
removed_addr = distinct_addr_in(pre_routingTable) - distinct_addr_in(new_routingTable)
• Mark all addresses in set added_addr in ConnectionPools to be active connection pools
• Mark all addresses in set removed_addr in ConnectionPools to be passive connection pools
• Remove passive connection pools if no connection is inUse (all connections are idle)
• When returning a connection to a passive connection pool, terminate the connection directly, [and remove the connection pool if no connections is InUse]