Every request slows down at once, then recovers
How you'll know: Response times spike for all users together, often while someone runs a report, an import or a PDF export.
Fix: Something is blocking the event loop. Node runs your JavaScript on one main thread, so heavy CPU work or a synchronous file or crypto call holds up every other request. Move that work to a worker thread or a background job queue, and keep request handlers short.
Memory climbs for days until the process dies
How you'll know: The service is quick after a restart, gets slower and slower, and finally crashes, usually overnight.
Fix: A leak: a cache that never evicts, listeners added on every request, or objects piling up in a module-level array. Take heap snapshots a few hours apart under load and compare what's growing. A scheduled restart only hides it.
One bad request takes the whole service down
How you'll know: A single customer's odd input, or one failed call to another API, knocks every user offline for a moment.
Fix: Current Node versions end the process on an unhandled promise rejection. Handle errors where each async call is made, return a proper error to that one request, and let the process manager restart cleanly if anything still slips through.
A package nobody chose directly is now a risk
How you'll know: npm audit lists advisories, or a dependency several layers down has been abandoned or taken over by someone new.
Fix: Commit the lockfile, review audit output on a schedule, and drop packages that do one small job you could write in ten lines. Supply-chain attacks through npm packages are real and keep recurring.
The Node version is past end of life
How you'll know: The server runs a release line the Node.js project no longer patches.
Fix: Move to a current long-term support release, tested on a staging copy first. Older native modules are what usually breaks, so check those before the switch rather than after it.