/node-crashit

A simple way of handling application crashing and normal termination the exact same way.

Primary LanguageJavaScriptMIT LicenseMIT

crashit

npm version Build Status Maintainability Test Coverage Known Vulnerabilities

A simple way of handling application crashing and normal termination the exact same way.

Crash Only and Crash First

Crash-only apps are applications that are coded in such a way that shutting them down is performed just by crashing the app. Crash-first apps are crash-only apps that also include a correct way of being shut down, which is generally better and preferred, but still respond well and can work correctly if they are just crashed every time.

Crash here just means "end abruptly". For node, this means calling process.exit. For a Unix process, it means killing a process with SIGKILL. Either way, the result is the same: The process is stopped with little to no chance of performing any clean up.

This library thus provides a single function that allows you to crash your application. It lets you choose the method of crashing and allows you to define crash hooks. It also allows you to completely bypass the hooks or to set a timeout after which the process is just crashed.

crashit.crash([reason[, runHooks[, timeout]]])

Crashes the current process with the given reason.

  • reason: The reason of the crash. This can be one of the following:

    • An integer, which will be passed to process.exit directly.
    • A signal name.
    • An exception object.

    When anything is passed as a reason other than 0, the process will exit with an exit code different from 0. If unspecified by the reason, this exit code will be 170.

    Its default value is 0.

  • runHooks: Whether to run crash hooks or not. Note that you should test your application by crashing it with and without running crash hooks or you risk your application being neither crash-only nor crash-first.

    Its default value is true.

  • timeout: The time to wait, in milliseconds, for hooks to end before definitely crashing the application. If hooks are not allowed to run, this value is completely ignored. If the application exits due to a timeout, it will exit with exit code 171.

    Its default value is 5000, that is, five seconds.

The process will be ended by using process.exit, and will be passed a value that depends on the reason:

  • If reason was an integer, it's passed to process.exit.
  • If reason was a string, it's interpreted as a signal name and process.exit receives 128 + signalNumber, unless the string was not the name of a signal, in which case it receives 166.
  • If reason was an error object, process.exit is passed the value 166.
  • If the execution of the hooks times out, process.exit is called with the value 171.

crashit.addHook(hook)

Adds a new crash hook. Crash hooks are global, which means that multiple version of the library loaded by different modules will the same crash hooks.

  • hook: A function to run when the application crashes, with the following signature: function (reason[, callback]).

    • reason the crash reason used when the crashit.crash function was called.
    • callback: An optional callback argument in case the function is asynchronous. The callback follows the common pattern of having an error as its first argument. It doesn't accept any more arguments.

    If the hook function does not have a callback argument, it is assumed that all it had to do was completed when it returns unless it returns a Promise. If either a promise is returned or a callback accepted, the hook is considered asynchronous and will be waited for before finishing the crash (unless, of course, it times out).

It is recommended to use promises rather than callbacks, but either way is fine. Errors during both synchronous or asynchronous hooks are completely ignored.

There's one guarantee about hooks: They will never run twice.

crashit.handleSignals(signals[, runHooks[, timeout]])

In order to easily work with signals such as SIGINT and SIGTERM and handle them as direct crashes, the crashit.handleSignals function is provided.

To use it, just pass one or more signal names to it. When the process receives said signals, the crash method will be automatically called with the passed arguments and the signal name as the reason.

  • signals: A signal name or array of signal names to watch and crash on.
  • runHooks: Whether to rook crash hooks before crashing.
  • timeout: Time, in milliseconds, to wait for crash hooks.

Note that there might be multiple callers for the same signal. That is, however, completely idiotic, as signal handlers should be configured just once in an application. For that reason, nothing is done to prevent it or to define what happens. Most likely, all handler are called at once and the fastest wins. In any case, crash hooks are never run twice.

crashit.handleExceptions([runHooks[, timeout]])

Sets up an uncaught exception handler that instantly performs a crash with the offending exception as reason. The handler will print a stack trace to the screen if and only if the runHooks option is set to false. This ensures a trace is written even if the handler is configured to not run any hooks.

Note that this means that you should print the stack trace yourself in a hook when setting runHooks to true.

crashit.handleRejections([runHooks[, timeout]])

Sets up an unhandled rejection handler that instantly performs a crash with the offending exception as reason. The handler will print a stack trace to the screen if and only if the runHooks option is set to false. This ensures a trace is written even if the handler is configured to not run any hooks.

Note that this means that you should print the stack trace yourself in a hook when setting runHooks to true.

crashit.handleUncaught([runHooks[, timeout]])

Calls handleExceptions and handleRejections with the same arguments.