Two settings decide whether a Node collector runs well, and neither of them is the proxy address.
Keep connection reuse on
Without it every request re-establishes the connection. On a long run that is the single largest avoidable cost.
Set it explicitly; the defaults have changed between versions and assuming is unwise.
Pick the agent deliberately
The modern client stack has its own proxy agent, and it behaves differently from the older wrappers. Mixing them produces requests that quietly ignore the address.
Verify with one call to an address-reporting service after any change here.
Write in batches
A thousand rows at a time. Per-row writing turns the disk into the bottleneck well before the network is saturated.
Bound the concurrency
Unbounded parallel requests will exhaust the pool limit in seconds. Use a queue with a fixed width.
const agent = new ProxyAgent(PROXY)
const res = await fetch(url, { dispatcher: agent })
