There could be a case when new shard connections initialization starts error out.
We need to track errors and slow it down or maybe completely stop for some time.
Connection initialization here probably should include the whole process starting from creating a socket up to connection getting into the pull.