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
- cmd
- array of command and args for child_process.spawn
- depends
- array of names of dependencies which this dependency depends on
- cwd
- The cwd to pass to child_process.spawn
- stdout
- path to log dependency's stdout to (default: your stdout). This can be relative to either the `cwd` value you specify or `__dirname` of `dependencies.json`.
- stderr
- path to log dependency's stderr to (default: your stderr). This can be relative to either the `cwd` value you specify or `__dirname` of `dependencies.json`.
- signal
- signal to use to stop the dependency when your tests end (default: `SIGTERM` )
- wait_for
- lets you specify a timeout and either an exit code or ip:port to wait for socket availability before considering a dependency started
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
}
- host
- default: localhost
- port
- if port is set, will wait for a TCP connection to be accepted on this port
- exit_code
- if exit_code is set, will wait for the process to exit with this code before proceeding
- timeout
- seconds to wait before considering a dependency to have failed starting up (default: 30)
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.