depends-on

Manages the external services that your tests depend on. Think of it as async.auto with process control for integration tests.

examples

dependencies.json is a file describing your dependencies. Keep it next to your tests.

… with one dependency

test-thing.js:

var ready = require('depends-on')('redis');
var test = require('tape');

test('init dependencies for ' + __filename, ready);

test('test that uses redis', function(t) {
  …
});

dependencies.json:

{
  "redis": {
    "cmd": ["/usr/sbin/redis-server"],
    "wait_for": {
      "port": 6379
    }
  }
}

ready() is a function that takes a callback (or a tape test object). It will call that callback when your dependencies are ready. They'll be stopped automatically when your tests are done.

… with multiple dependencies

Dependencies can have dependencies. This is the typical use case, where multiple services must start before your tests can run.

Building on the previous example, say you want to clear all values from (or load some fixtures into) Redis after it starts, but before your tests run.

test-other-thing.js:

var ready = require('depends-on')('fresh & clean redis');
var test = require('tape');

test('start fresh redis', ready);

test('test that uses redis', function(t) {
  …
});

dependencies.json:

{
  "redis-server": {
    "cmd": ["/usr/sbin/redis-server"],
    "wait_for": {
      "port": 6379
    }
  },
  "fresh & clean redis": {
    "cmd": ["./bin/flushall.sh"],
    "depends": ["redis-server"],
    "wait_for": {
      "exit_code": 0
    }
  }
}

If multiple tests are require()'d and share dependencies, depends-on will share them across the test files, each dependency only being started once. When node exits, all dependencies will stopped (by default with SIGTERM but the signal is configurable).

dependencies.json

dependencies.json is a file containing a json object mapping dependency names to objects that describe each dependency.

dependency fields

wait_for fields

You can wait for either a socket to be accepting connections or for a process to exit, so one of port or exit_code is required to use wait_for.

    "wait_for": {
      "host": "127.0.0.1",
      "port": 22181,
      "timeout": 30
    }

By default, depends-on will throw an error if one of the processes it starts exits before your tests are done. Use an exit_code in your wait_for block if a process is expected to complete and exit prior to tests running (maybe you're loading a db schema).

    "wait_for": {
      "exit_code": 0,
      "timeout": 120
    }

graph dependencies

Just for fun, bin/graph-dependencies.js can graph the dependencies in your project with graphviz.

For example, graphing the dependencies in this repo's fixtures: dot -Tpng -odeps.png <(./bin/graph-dependencies.js) && open deps.png ->

notes

using with tape

Be sure to require('depends-on') before you require('tape'). Like this:

var ready = require('depends-on')('something');
var test = require('tape');

Both depends-on and tape have process.on('exit', …) handlers, but tape calls process.exit in its process.on('exit', …) handler, so if tape's handler runs first, depends-on's handler will never run (and child processes won't be cleaned up). Handlers run in the order they are added, so depends-on must be required first.

why not

… use a bash script or Makefile?

I like to be able to run integration tests individually like $ node tests.js without running anything else or relying on some external service coincidentally being on.