Pattern
A burst of errors lasting under a minute or two, often connection resets (ECONNRESET) or timeouts, that stops by itself. Alerts with a one-minute window can fire and resolve within a few minutes.
What to do
- Check whether the error rate is already back to baseline. If it is, and nothing changed (no deploy, flag or config change), close the alert as transient and note it.
- If spikes repeat several times an hour, open a ticket; repeated resets usually mean a network or proxy problem.
- Do not roll back or restart because of a spike that has already recovered.