A reconnect loop that resolved the hostname once, before the retry loop started, keeps retrying that first IP forever — even long after a DNS failover moved the record somewhere reachable. The process never crashes. It just retries a connection that can never succeed again, and looks exactly like "still down," indefinitely, until something restarts it.
function naiveReconnectLoop(resolve, maxAttempts) {
const ip = resolve('chat.example.com'); // resolved ONCE, before the loop
for (let i = 0; i < maxAttempts; i++) {
if (connect(ip)) break; // always the same ip
}
}That reads like a reconnect loop and behaves like a retry-forever loop against one fixed address. The fix is one line of scope — move the resolve call inside the loop — but it's easy to miss, because the code compiles, the loop runs, and "it's retrying" looks like the system doing its job. We've seen this exact shape: a reconnect loop that logged "reconnecting..." forever, healthy-looking, because it never stopped trying — it just never tried the right address again.
node check.js --demoZero dependencies, no network — models a DNS failover with a fake resolver that returns a dead IP for the first 3 lookups, then a live one, and runs a naive loop and a fixed loop against the identical scenario.
The naive loop runs out of attempts still failing against a dead IP that DNS stopped using three lookups ago. The fixed loop recovers the moment it asks again.
Against a real hostname (built-in dns module only, still zero deps):
node check.js --host chat.example.com [--n 5] [--interval-ms 1000]This can't prove your own reconnect code is safe — only reading that code can — but it tells you whether DNS for that host is actually changing during the window you're testing, which is the precondition for this bug ever mattering.