Process memory model

process.memoryUsage() returns:

Field	Meaning
`rss`	Resident Set Size — total process memory in RAM
`heapTotal`	V8 heap allocated
`heapUsed`	V8 heap in use
`external`	C++ objects bound to JS (Buffers, etc.)
`arrayBuffers`	Memory for ArrayBuffers (Node 16+)

flowchart TB
  rss[RSS total process]
  heap[heapUsed / heapTotal]
  ext[external + arrayBuffers]
  native[Native code stacks etc]
  rss --> heap
  rss --> ext
  rss --> native

Healthy vs leaking server

Healthy: heapUsed sawtooths — spikes on work, dips after major GC.

Leaking: heapUsed and often rss stair-step upward over hours with steady traffic.

Logging snippet

function logMemory(label) {
  const m = process.memoryUsage();
  console.log(label, {
    rssMB: (m.rss / 1024 / 1024).toFixed(1),
    heapUsedMB: (m.heapUsed / 1024 / 1024).toFixed(1),
    externalMB: (m.external / 1024 / 1024).toFixed(1),
  });
}

Run every N seconds in staging to build intuition before Grafana.

Broken pattern

Monitoring only CPU — memory creep is invisible until OOM.

Fixed pattern

Alert on 7-day trend of heapUsed / rss, not single spikes. Pair with deployment markers.