Double the number of neighbours we gossip to, for improved coverage.
Previously a 3-node cluster would send to 1 peer each time, and a 7-node cluster to 2, so information could take a while to get everywhere.
This calculation maintains the spirit that the number grows slowly in larger clusters - in a 100-node cluster each node will gossip to 12.
Slow down accepting new connections, to reduce peak memory usage.
When 100 connections arrive at once, each one sends the whole topology and we read it all into memory at once then try to relay new information out again.
These new parameters mean we will take 4 seconds to accept 100 connections, and 9 seconds to accept 200, so the system has time to process each message without spiking memory.
Double the number of neighbours we gossip to, for improved coverage.
Previously a 3-node cluster would send to 1 peer each time, and a 7-node cluster to 2, so information could take a while to get everywhere.
This calculation maintains the spirit that the number grows slowly in larger clusters - in a 100-node cluster each node will gossip to 12.
Slow down accepting new connections, to reduce peak memory usage.
When 100 connections arrive at once, each one sends the whole topology and we read it all into memory at once then try to relay new information out again.
These new parameters mean we will take 4 seconds to accept 100 connections, and 9 seconds to accept 200, so the system has time to process each message without spiking memory.