refresh_slots fix dns resolve delay

[alexander-shabanov]

Temp fix: Change 'addresses to connections' (specifically get_socket_addrs) to be sequential instead of concurrent in order to workaround DNS query throttling

[nihohit]

why does DNS query throttling make it slower than sequential attempts?

[barshaul]

@nihohit We haven't dived deep into why it takes longer - it look like it's due to DNS throttling.
ATM we see that it takes 5 seconds when we run it concurrently VS ~500 milliseconds when we run it sequentially, in a large cluster test with 300 nodes.
We should RC it but at the moment lets have this fix and work to solve it later.
@alexander-shabanov have you opened an issue for this?

[barshaul]

@alexander-shabanov Can you test this change with a small cluster - lets say 3 shards, master&replica in each, to see what's the impact in terms of time to create the client?

[alexander-shabanov]

@nihohit @barshaul I have submitted an issue make DNS resolution and creation of connections to be concurrent to follow up this fix.

[alexander-shabanov]

@barshaul For small cluster (3 shards, 9 nodes) sequential DNS resolve takes 15 milliseconds. Client creation overall takes around 100 milliseconds.

[barshaul]

@alexander-shabanov Please pull latest changes

[alexander-shabanov]

@barshaul I pulled latest changes