enhance load balancer for conn to graph

[lee-qiu]

considering 1 situation:
when one graph shutdown, then sessions connected to this graph will reconnect to others. if session pool is never updated, this graph will never be used although it maybe be recovered lately. and this case is regular when upgrading graph.

if this is true, maybe we should enhance the load balancer in session pool or connection pool. for example, we can build a map with key=address, val=sessions to it, define min/max session size to each address and then update session to each address timely?

[wey-gu]

cc @Aiee @yixinglu what do you think, please?

[dutor]

Sorry about the late response.

We have noticed this issue for quite a while and proposed some solution similar to yours, but this issue remains due to our limited resource.

We are glad to receive a patch to address this if you would like to contribute.