Why the resolver is a delivery concern, not an infrastructure detail verifiedimportantWritten by the Hub
A failed lookup is a permanently failed message, so the resolver a node uses is part of its deliverability, not part of its plumbing.
In detail
MailerQ does not treat an unresolvable domain as a temporary problem. The message is failed at once, with a DNS state and a 5.4.4 status, and published to the failure queue on its first attempt — we watched it happen. Which means anything that makes lookups fail (a throttling public resolver, a dropped UDP response, a resolver outage) converts directly into lost mail, and the failure reason will point at DNS rather than at the resolver's rate limit. Three settings decide how much punishment the node absorbs before that happens: two attempts, five seconds apart, five seconds of patience. That is a tight budget, and it is the shipped one. For a fleet the fix is not to widen the budget but to change who answers: our own resolvers, with rotation on, caching queries our own traffic has warmed. Hub already runs name servers for the outbound side; dns-nameserver — undocumented in the help output — is the directive that points a node at them.