I have observed that, when starting many nodes at once, the server can freeze for a significant amount of time before completing the handshake with all the nodes.

For instance, when starting 30 nodes, using node provisioning (so all on the same machine), the server freezes for about 50 seconds. If I start the 30 nodes in multiple steps, for instance 8 nodes then wait until handshake is done, then 8 more, etc..., then this issue does not occur.

Furthermore, turning on DEBUG-level logging in the server i observed that the freeze happens after the handshake completes for all clas loader connections, but before the handshake for the job channels starts. This needs to be investigated, as this could be very disruptive when hundreds of nodes are connecting at the same time, e.g. in the case of a driver restart.