Connection pool does not scale up to minimum connections after a temporary network failure.

After a temporary network failure, all the connections are cleaned up. Any pending requests are requeued onto the send queue. As long as the number of queued messages are fewer than the backpressure setting, no new connection attempts are made and the requests will sit in the queue forever. E.g. the await GetClusterMap in the ConfigHandler poll loop will wait forever even after the network failure is resolved.