app crashed - waiting for file changes before starting... nodemon crash state restart behavior

Direct answers

Backoff / crash threshold: No. Nodemon (through the 2.x line, e.g. 2.0.23) exposes no option for a restart backoff, retry budget, or maximum crash count. A child that exits non-zero puts nodemon into the crash state, and it stays there until a watched file changes. If the process crashes on every start, each file save triggers another immediate crash — there is nothing to configure away from that.

Crash vs. clean exit: The distinction is made purely on the child process's exit code. Exit code 0 is a clean exit; any non-zero code is a crash. This is the entire rule — nodemon does not inspect stderr, stack traces, or how the process died.

Delay interaction: --delay only applies to restarts triggered by file changes. It does not run while nodemon sits in the crash state. The sequence is: crash → suspend indefinitely → file change detected → wait the configured delay → restart attempt.

What this means in practice

The "app crashed - waiting for file changes before starting..." message is not an error in nodemon itself; it is nodemon telling you it refuses to restart-loop a broken process. This is intentional: without it, a syntax error would spin your CPU restarting forever.

Two consequences follow from the exit-code rule:

• If your app catches its own errors and calls process.exit(0) on failure, nodemon treats that as clean and (depending on config) may restart immediately or exit — you lose the crash suspension. Exit non-zero on real failures.
• Conversely, a graceful-shutdown handler that exits non-zero will be misread as a crash and stall your dev loop.

Working around the missing backoff

Since there is no built-in knob, the practical options are:

1. Fix the crash cause — the suspended state is doing its job by forcing you to look.
2. Narrow watch scope so stray writes (logs, temp files, build output) don't trigger restarts: nodemon --watch src --ext js,ts --ignore '*.log' app.js.
3. If you genuinely need retry-with-backoff semantics, wrap the start in your own script or use a process manager designed for it (e.g. pm2 with --exp-backoff-restart-delay), accepting that this moves you away from nodemon's dev-oriented model.

Verifying the behavior yourself

You can confirm the exit-code rule in a minute:

printf 'process.exit(1)\n' > crash.js
nodemon crash.js   # prints the crash message, then waits
touch crash.js     # saving triggers a restart attempt

And to check delay interaction: nodemon --delay 2s crash.js, then touch the file and observe the 2-second pause before the restart attempt. Behavior described here assumes nodemon 2.x; check your installed version with nodemon --version if results differ.