connection_minmessages verifiedimportantWritten by the Hub
Messages before opening another connection
How much work an existing connection must be given before the engine decides to open a second one.
In detail
The engine's restraint mechanism, and a non-obvious default at ten. It means the node prefers to queue messages onto a connection it already has rather than open another — which is exactly what a receiver wants to see, and exactly why a node can look slower than its configured limits allow. Together with domain_maxconnecting: 1 it produces the shipped behaviour: connections are opened one at a time, and only when the existing ones are genuinely busy. Polite, and worth understanding before someone "fixes" throughput by raising the message rate, which will not help.
What happens if you get it wrong
Raised high on a node with few messages per domain.A second connection is never opened, because no single connection ever accumulates enough work; parallelism collapses to one.inference · inference