Support for testing Taskcluster components.
This module contains a number of utilities that facilitate testing Taskcluster components. It is typically installed as a devDependency, so it is not used in production code.
See the source for detailed documentation.
A sticky loader is a thin wrapper around taskcluster-lib-loader to support
dependency injection. It "remembers" each value it has returned and will return
it again on the next call; it can also have a dependency injected. Use it like
this in helper.js:
const {stickyLoader} = require('taskcluster-lib-testing');
const load = require('../src/server');
exports.load = stickyLoader(load);
exports.load.inject('profile', 'test');
exports.load.inject('process', 'test');The load.inject(component, value) method sets a loader overwrite without
attempting to load it. There is a corresponding load.remove(component) to
remove a component.
In test scripts:
const {load} = require('./helper');
suite('SomeTable', function() {
suiteSetup(async function() {
load.save(); // save the state of the loader to restore in tearDown
await load('cfg'); // load the cfg so we can edit it
load.cfg('azure.accountName', 'inMemory'); // edit the cfg in-place
const SomeTable = await load('SomeTable');
await SomeTable.ensureTable({ /* ... */ });
});
suiteTeardown(function() {
load.restore(); // restore the state of the loader
});
test(async function() {
const component = await load('some-component');
// some-component will be loaded with the same cfg and with
// the same instance of SomeTable that we set up above
});
});The load.save() and load.restore() methods push and pull loader states in a
stack, and are best used in setup/teardown methods to ensure that one suite
does not "pollute" the loader state for the next.
The load.cfg(path, value) method edits the cfg component in place, using a
dotted path to specify the config value. The save and restore methods are
careful to deep-copy cfg so that these in-place modifications affect only
the current loader state.
If cfg is not loaded, the load.cfg() method will not work, so generally (as
in the example above) a bare load('cfg') is used to ensure its presence.
This class handles getting secrets for tests, and easily determining what
secrets are available. It integrates with taskcluster-lib-config. Set it up by
in test/helper.js:
const {Secrets} = require('taskcluster-lib-testing');
exports.secrets = new Secrets({
secretName: [
'project/taskcluster/testing/taskcluster-foo',
'project/taskcluster/testing/taskcluster-foo/main-only',
],
// (optional) provide a stickyLoader instance for use in mockSuite
load,
secrets: {
pulse: [
// env - the environment variable by which this secret is set in the config (if any)
// cfg - dotted path to the config value containing this secret (if any)
// name - name for the secret (used for programmatic access in tests; defaults to env)
// mock - value to provide if secret is not set (for mock runs only)
{env: 'PULSE_USERNAME', cfg: 'pulse.username', name: 'username', mock: 'dummy'},
{env: 'PULSE_PASSWORD', cfg: 'pulse.password', name: 'password'},
],
aws: [
{env: 'AWS_ACCESS_KEY_ID', cfg: 'aws.accessKeyId'},
{env: 'AWS_SECRET_ACCESS_KEY', cfg: 'aws.secretAccessKey'},
],
},
});If a secret is defined in the loaded configuration, that value will be used even if the env key is also set.
Secrets should not have any value set in config.yml (although !env is OK), or this class will not function properly.
If the system you are testing does not use taskcluster-lib-config, simply do not specify the cfg properties to the constructor.
You can also leave out load in this case.
You can then call await secrets.setup() to set up the secrets (reading from cfg if necessary).
This must be called during Mocha's runtime, so either in a setup function or a test.
It short-circuits multiple calls, so it's safe to call it all over the place.
In fact, mockSuite (below) will call it for you.
In CI (when $TASK_ID is set), the setup method will attempt to fetch the secrets named in secretName from the secrets service.
It expects the fetch value to be a map from environment variable name to value.
If a fetch fails, it is considered equivalent to fetching an empty map.
This allows, for example, secrets that can only be fetched on pushes to the main branch, and not pull requests.
The secrets object has a few useful methods, all of which can only be called after setup, and thus only in a setup function or a test:
secrets.have(name)-- true if the given secret is availablesecrets.get(name)-- returns an object containing the secret values by name, or throws an error if not avaialble
The secrets.mockSuite function abstracts away the most common case: running the same tests in a mock and real environment, skipping the real tests if secrets are not available.
It is called as secrets.mockSuite(title, [secrets], async function(mock, skipping) { .. }) in the same location you might call Mocha's suite(..).
The secrets is an array of secret names required to run this suite in a real environment.
The given function should define the suite, and can include setup, suiteSetup, and so on.
The mock parameter is true for the mock version, and false for the real version.
If $NO_TEST_SKIP is set, mockSuite will throw an error when secrets are not available.
Note that Mocha continues to run setupSuite and teardownSuite functions even after a suite has been skipped.
Mocha does not provide any way to determine if a suite has been skipped.
Use skipping() to determine if the suite is currently skipping, and avoid doing initialization that will fail.
Note, too, that all modern versions of Mocha have a bug causing nested suites to run anyway, even when the parent suite is skipped. A quick (but unfortunate) way to work around this bug is
secrets.mockSuite('mySuite', [..], function(mock, skipping) {
suiteSetup(function() {
if (skipping()) {
this.skip();
}
});
});// helper.js
const {Secrets, stickyLoader} = require('taskcluster-lib-testing');
const load = stickyLoader(require('../src/main'));
const secrets = new Secrets({
secretName: 'project/taskcluster/testing/taskcluster-ping',
secrets: {
pingdom: [
{name: 'apiKey', env: 'PINGDOM_API_KEY', cfg: 'app.pingdom.apiKey'},
],
taskcluster: [
{name: 'clientId', env: 'TASKCLUSTER_CLIENT_ID', cfg: 'taskcluster.credentials.clientId'},
{name: 'accessToken', env: 'TASKCLUSTER_ACCESS_TOKEN', cfg: 'taskcluster.credentials.accessToken'},
],
},
load,
});
exports.secrets = secrets;
exports.load = load;// some_test.js
const {secrets, load} = require('./helper');
// for testing by passing secrets to the subject..
secrets.mockSuite('pingdom updates', ['pingdom'], function(mock, skipping) {
let pingdomUpdater, pingdomComponent;
suiteSetup(async function() {
// use secrets.get(..) in the real case
pingdomUpdater = new PingdomUpdater({apiKey: mock ? 'pretendKey' : secrets.get('pingdom').apiKey});
if (mock) {
nock('https://pingdom.com:443', ..); // mock out Pingdom API
}
});
suiteTeardown(function() {
if (mock) {
nock.clearAll();
}
});
test('updates once', function() { .. });
});
// for testing a loader component..
secrets.mockSuite('Floobits', ['taskcluster'], function(mock, skipping) {
let Floobits;
suiteSetup(async function() {
if (mock) {
// set the special accountName that will cause azure-entities to use its fake version;
// otherwise, the loader component will use the taskcluster secrets to get access
// to the a Azure table
helper.load.cfg('azure.accountName', 'inMemory');
}
if (!skipping()) {
Floobits = await helper.load('Floobits');
await Floobits.ensureTable();
}
});
test('create', async function() {
await Floobits.create(..);
// ..
});
});The test output for the first suite will contain something like
pingdom updates (mock)
✓ updates once
pingdom updates (real)
- updates once
Note that even in cases where no secrets are required, mockSuite is still
useful for providing the mock, skipping values required by other components
of taskcluster-lib-testing.
Test schemas with a positive and negative test cases.
The method should be called within a suite, as it will call the mocha test
function to define a test for each schema case.
schemasetOptions- {} // options to pass to the taskcluster-lib-validate constructorcases- array of test casesbasePath- base path for relative pathnames in test cases (defaultpath.join(__dirname, 'validate'))
Each test case looks like this:
{
schema: 'https://tc-tests.localhost/svcname/v7/frobnicate-foo.json', // JSON schema identifier to test against
path: 'test-file.json', // Path to test file (relative to basePath)
success: true || false // true if validation should succeed; false if it should fail
}A fake for the auth service to support testing APIs without requiring production credentials, using Nock.
This object intercepts requests to the auth service's authenticateHawk method
and returns a response based on the given clients, instead. Note that
accessTokens are not checked -- the fake simply controls access based on
clientId or the scopes in a temporary credential or supplied with
authorizedScopes.
To start the mock, call testing.fakeauth.start(clients, {rootUrl}) in your suite's
setup method. The first argument has the form
{
"clientId1": ["scope1", "scope2"],
"clientId2": ["scope1", "scope3"],
}The auth service on the cluster identified by rootUrl will be faked. When
used to test an API in a microservice, this is same as the root URL for the
fake web server -- http://localhost:1234 or something of that sort.
Call testing.fakeauth.stop() in your test suite's teardown method to stop the HTTP interceptor.
In cases where you must test that HTTP requests without credentials are properly authenticated, for example, to test a webhook. To achieve this, temporarily set the anonymous scopes with
testing.fakeauth.withAnonymousScopes(['some-scope'], async () => {
..; // make the credential-free call
});This function is intended for use with usual configuration for Postgres databases.
It sets up a "real" database using $TEST_DB_URL, accessed with a user corresponding to the given serviceName.
If $TEST_DB_URL is not set, it will exit during test setup.
The database is upgraded to the latest version at the beginning of the suite.
The resulting database is injected into the taskcluster-lib-loader as db and also available as helper.db.
It is up to the test suite implementation to reset the contents of the database between tests.
Ideally this is done via helper.db.fns methods.
If a client that can execute SQL directly is required, use await helper.withDbClient(async client => { .. }).
That method will use a connection from the per-service DB pool, meaning that it is limited by the service's DB permissions.
The similar await helper.withAdminDbClient(async client => { .. }) will create a new connection using the administrative credentials.
Note that this is intended to operate against a temporary Postgres server such as one running in a docker container.
It's not a good idea to run this against a "real" Postgres server.
In particular, it will create a bunch of users with names beginning with test_ (global to the server) and reset their passwords and access.
The function is typically used like this:
// helper.js
exports.secrets = new Secrets({
// ...
});
exports.withDb = (mock, skipping) => {
withDb(mock, skipping, exports, 'my-service');
};// some_test.js
helper.secrets.mockSuite(testing.suiteName(), [], function(mock, skipping) {
helper.withDb(mock, skipping);
...
});Note that, while it takes mock, skipping like other utilities in this library, withDb always uses a real Postgres DB.
There is also a utility function, resetTables, which will truncate a list of tables.
This is typically used in a setup function to start each test with a clean slate.
const {resetTables} = require('taskcluster-lib-testing');
exports.resetTables = (mock, skipping) => {
setup('reset tables', async function() {
await resetTables({tableNames: [
'some_table',
'another_table',
]});
});
};Finally, to completely reset the DB to an empty state (but with per-service users defined), call resetDb:
const {resetDb} = require('taskcluster-lib-testing');
// this automatically uses TEST_DB_URL.
await resetDb();This function helps test applications that publish pulse messages.
It is typically set up in test/helper.js like this:
const {withPulse} = require('taskcluster-lib-testing');
exports.withPulse = (mock, skipping) => {
withPulse({helper, skipping, namespace: 'taskcluster-someservice'});
};It assumes that helper.loader is set up as a sticky loader, and that there is a loader component named pulseClient.
It replaces this client with a fake version, for which libPulse.consume() and exchanges.publisher() will produce fake consumers and publishers, respectively.
The fake client can be identified by having a truthy isFakeClient property.
For fake publishers, it offers the following:
onPulsePublish(callback)- call the given function when a pulse message is published. This is useful to raise exceptions on publishing. The callback is called with (exchange, routingKey, payload, CCs).assertPulseMessage(exchange, check)- assert that a matching message has been sent to pulse. The optionalexchangeis a suffix of the expected exchange (just the portion after/v1/); all messages will match if omitted. The optionalcheckfunction is called with each message on that exchange and should return true for matching messages.assertNoPulseMessage(exchange, check)- the opposite ofassertPulseMessageclearPulseMessages- clear the accumulated pulse messages. This is useful when you expect duplicte messages: asert that the first one was sent, then clear, then assert that the second was sent.
Messages are reset before each test case.
To simulate receipt of a pulse message by a consumer, call await helper.fakePulseMessage({exchange, routingKey, routes, payload}).
The message will be routed to consumers with matching bindings.
A consumer for which bindings are changed at runtime, using amqplib functions bindQueue and unbindQueue, can be faked by calling consumer.setFakeBindings(bindings).
All services should call testing.withMonitor(helper) to set up the taskcluster-lib-monitor loader component for testing.
Call this method at the module level, such as within helper.js, not as a part of each test suite.
The function does the following:
- Set up the default MonitorManager instance to log via the
debugmodule in a human-readable fashion, and to record messages (mock=True). - Fail if there are any messages at the ERROR level or higher at the completion of each test case. Tests that generate errors should check for and remove them.
- Reset the list of stored messages after each test.
Tests should import the MonitorManager instance from ../src/monitor.js to get access to the messages and modify the message list
Libraries can use this function as
withMonitor(exports, {noLoader: true});The sleep function returns a promise that resolves after a delay.
NOTE tests that depend on timing are notoriously unreliable, and suggest poorly-isolated tests. Consider writing the tests to use a "fake" clock or to poll for the expected state.
When testing functionality that involves timers, it is helpful to be able to simulate the rapid passage of time.
The testing.runWithFakeTime(<fn>, {mock, maxTime, ...}) uses zurvan to do just that.
It is used to wrap an argument to a Mocha test function, avoiding interfering with Mocha's timers:
test('noun should verb', runWithFakeTime(async function() {
...
}, {
mock,
maxTime: 60000,
}));The maxTime option is the total amount of simulated time to spend running the test; it defaults to 30 seconds.
The mock option is for use with mockSuite and can be omitted otherwise.
Fake time is only used when mocking; in a real situation, we are interacting with real services and must use the same clock they do.
Any other options are passed directly to zurvan.
The poll function will repeatedly call a function that returns a promise
until the promise is resolved without errors.
await poll(
maybeFunc, // function to be called
11, // max times to try it
100); // delay (ms) between tries