Apache Druid

if you have some thoughts on how people should think about `somaxconn` &amp; other parameters, that'd be a great addition to the docs

yea i can verify for your first bulletpoint we managed to replicate that scenario on one of our larger clusters handling around 4K-5K tasks

disabling syncookies + upping the jetty queue resolved it for us

The lazy connection initialization would help. btw don't you see similar connection saturation on the brokers?

when syncing metadata state we saw similar issues occur, but fixed it using the above approach