Test runner

The test runner script is used to compile MongooseIM and run tests.

The help command prints a list of supported options.

./tools/test-runner.sh --help

Test runner examples

Usage example:

./tools/test-runner.sh --db redis --preset internal_mnesia

The command runs both big (feature) and small (unit) tests.

To view more examples, run:

./tools/test-runner.sh --examples

Test runner completion

Test runner supports shell TAB completion.

To enable completion in bash or zsh, run:

source tools/test-runner-complete.sh

To view completion examples, run:

./tools/test-runner.sh --examples-complete

Viewing test reports

To view test execution results, run:

./tools/test-runner.sh --show-big-reports
./tools/test-runner.sh --show-small-reports

Rerun big tests

Very often we want to restart a specific suite when some test failed.

For example, some test has failed in mam_SUITE. The command was used to execute tests:

./tools/test-runner.sh --skip-small-tests --db mysql --preset mysql_mnesia --skip-stop-nodes

--skip-stop-nodes is optional here, because if any big test fails, then nodes would be still running.

We can just execute the same command, but it would rebuild nodes and start them.

The command can be used instead:

./tools/test-runner.sh --rerun-big-tests -- mam

--rerun-big-tests expands into --skip-small-tests --skip-setup-db --dev-nodes --test-hosts --skip-cover --skip-preset.

And mam is used to run mam_SUITE suite only.

Unit tests (a.k.a. "small tests")

These test suites are aimed at testing various modules and libraries standalone, without launching a MongooseIM instance. They are very useful for developing/debugging libraries.

The test suites are located in test/ directory. To run all of them, use ./rebar3 ct; to run just a selected suite, use ./rebar3 ct --suite test/my_selected_SUITE. Rebar recompiles all the code automatically, there is no need for a separate compilation step.

If all the tests pass, you wll get no output and summary log will be available in ct.log. If any of the tests fail the summary log is printed to stdout.

Detailed test results in a nice HTML format are saved in

_build/test/logs/ct_run.[something][datetime]/

Unit test running example using test runner:

# Run all small tests, show progress
./tools/test-runner.sh --skip-big-tests --verbose

# Run sha_SUITE without cover
./tools/test-runner.sh --skip-big-tests sha --skip-cover

# Run reload_cluster group in ejabberd_config_SUITE, show progress
./tools/test-runner.sh --skip-big-tests ejabberd_config:reload_cluster --verbose

End-to-end tests (a.k.a. "big tests")

Using test runner

Most important options are preset and database:

# Runs privacy_SUITE and private_SUITE with MySQL
./tools/test-runner.sh --skip-small-tests --db mysql --preset mysql_mnesia -- privacy private


# Runs MAM tests for MUC light with MySQL and Postgres
./tools/test-runner.sh --skip-small-tests --db mysql pgsql --preset mysql_mnesia pgsql_mnesia -- mam:odbc_muc_light

# Runs rdbms_SUITE with MSSQL
# Inits a single MongooseIM node (works for some tests only)
# Disables cover
./tools/test-runner.sh --skip-small-tests --db mssql --preset odbc_mssql_mnesia --test-hosts mim --dev-nodes mim1 -- rdbms --skip-cover

TL;DR

In shell #1:

$ cd $MONGOOSEIM
$ ./rebar3 compile
$ make devrel

In shell #2:

$ cd $MONGOOSEIM/_build/mim1/rel/mongooseim
$ ./bin/mongooseimctl live

In shell #3:

$ cd $MONGOOSEIM/_build/mim2/rel/mongooseim
$ ./bin/mongooseimctl live

In shell #4:

$ cd $MONGOOSEIM/_build/mim3/rel/mongooseim
$ ./bin/mongooseimctl live

In shell #5:

$ cd $MONGOOSEIM/_build/fed1/rel/mongooseim
$ ./bin/mongooseimctl live

In shell #6:

$ cd $MONGOOSEIM/_build/reg1/rel/mongooseim
$ ./bin/mongooseimctl live

Back to shell #1:

$ cd big_tests/
$ make quicktest

Wait for the tests to finish and celebrate (or wallow in despair and grief)!

One-liner alternative for tmux users:

./rebar3 compile
make devrel
tmux new-window -n mim1 '_build/mim1/rel/mongooseim/bin/mongooseimctl live'
tmux new-window -n mim2 '_build/mim2/rel/mongooseim/bin/mongooseimctl live'
tmux new-window -n mim3 '_build/mim3/rel/mongooseim/bin/mongooseimctl live'
tmux new-window -n fed1 '_build/fed1/rel/mongooseim/bin/mongooseimctl live'
tmux new-window -n reg1 '_build/fed1/rel/mongooseim/bin/mongooseimctl live'
_build/mim1/rel/mongooseim/bin/mongooseimctl started
_build/mim2/rel/mongooseim/bin/mongooseimctl started
_build/mim3/rel/mongooseim/bin/mongooseimctl started
_build/fed1/rel/mongooseim/bin/mongooseimctl started
_build/reg1/rel/mongooseim/bin/mongooseimctl started
make -C big_tests quicktest

Start a new tmux and paste the commands.

Step-by-step breakdown

make devrel builds four server nodes, preconfigured for a wide range of features covered by end-to-end tests.

  • $MONGOOSEIM/_build/mim1/rel, for most test SUITEs
  • $MONGOOSEIM/_build/mim*/rel, in order to test cluster-related commands;;
  • $MONGOOSEIM/_build/fed1/rel, in order to test XMPP federation (server to server communication, S2S).
  • $MONGOOSEIM/_build/reg1/rel, in order to test global distribution feature.

In general, running a server in the interactive mode (i.e. mongooseimctl live) is not required to test it, but it's convenient as any warnings and errors can be spotted in real time. It's also easy to inspect the server state or trace execution (e.g. using dbg) in case of anything going wrong in some of the tests. To run the server in the background instead of the interactive mode, use mongooseimctl start && mongooseimctl started.

The quicktest configuration is a relatively comprehensive one, giving good overview of what does and what doesn't work in the system, without repeating tests. Why would we want to ever repeat the tests? In order to test different backends of the same parts of the system. E.g. a message archive might store messages in MySQL/PostgreSQL or Riak KV - the glue code between the XMPP logic module and database is different in each case, therefore repeating the same tests with different databases is necessary to guarantee a truthful code coverage measurement.

Testing a feature in development / TDD

The whole suite takes a significant amount of time to complete. When you develop a new feature, the speed of iterating is crucial to maintain the flow (who doesn't like the feeling?!) and not lose focus.

In $MONGOOSEIM/big_tests/ we have:

$ tree big_tests/ -L 1 -F
big_tests/
├── Makefile
├── README.md
├── default.spec
├── test.config
├── tests/
└── ...

tests/ is where the test suites reside.

*.config files are the suite configuration files - they contain predefined XMPP client specifications, server addresses and XMPP domains to use, and options required by test support libraries (i.e. Escalus).

*.spec files are the test specifications - they define the configuration file to use, the suites, test groups or individual test cases to run or skip, and some less important things.

default.spec is the default when running make quicktest, but it can be overridden with a TESTSPEC variable:

# make sure we're in $MONGOOSEIM/big_tests/
cd $MONGOOSEIM/big_tests/
make quicktest TESTSPEC=my-feature.spec

To speed up the development cycle, developers usually create a .spec file for each feature (or each project, if you're cloning away) and only enable the suites / test groups they are working on. The allows testing only the parts of the system that are actually being changed. It's worth running default.spec once in a while to check for regressions.

Consult the default.spec file to see how to run only selected tests/groups/cases.

If you're sure that none of the test dependencies have changed and you only edited the test suites and/or MongooseIM code, it's possible to speed up the tests by skipping the Rebar dependency and compilation checks by providing PREPARE= (i.e. an empty value):

make quicktest PREPARE=

Consult the big_tests/Makefile to see how it works.

Applying code changes

When working on a feature or a bug fix you often modify the code and check if it works as expected. In order to change the code on dev nodes that are already generated (mim* and fed*) recompile the code for a specific node. For example, to update the code on mim1 node all you have to do is:

./rebar3 as mim1 compile

A similar command applies to other nodes, the important thing being rebar3's profile.

When the above command finishes, the code can be reloaded on the server by either reloading changed module(s) in the node's shell, e.g. l(mongoose_riak), or restarting the node.

Reading test reports

When finished, the test engine writes detailed html reports into a directory:

big_tests/ct_report/ct_run.[gobbledygook][datetime]/

Each run is saved into a new directory. This snippet:

#!/bin/bash

lst=$(ls -rt ct_report | grep ct_run | tail -n 1)
rm ct_report/lastrun
ln -s $lst ct_report/lastrun

can be of some help.

Checking coverage

If you want to check how much of the code is covered by tests, run:

make cover_quicktest

Note: You need all the mim nodes (mim1, mim2 and mim3) up and running, even if you only run some of the tests. If any of the nodes is down, the test will crash.

This command will recompile and reload the code on dev nodes with coverage enabled and run test suites as defined in the spec. Coverage statistics will be available in big_tests/ct_report/cover.html and coverage subdirectory.

Advanced topics

There are many more options available. One of them is sequentially testing a number of preset configurations - we do it every day on Travis, testing MongooseIM with various OTP versions and database backends. Altogether, we have eight preset configuration.

If you want to dig deeper, consult .travis.yml and tools/travis-test.sh, everything we do is there.

Gathering test reports from Travis tests

If you test your MongooseIM fork on Travis, you might want to access test reports (which also include node logs and crash dumps) that are created by the test runner.

Uploading reports to S3

Our script uses AWS CLI to upload test results to an S3 bucket. Simply set relevant environment variables in your repository settings on Travis (at least AWS_ACCESS_KEY_ID and AWS_SECRET_ACCESS_KEY have to be set), and enjoy test reports landing straight into your bucket (AWS_BUCKET variable should store the bucket's name).

Uploading reports to Google Drive

To store test results in Google Drive you need to create a new project and obtain service account credentials. You must also add Google Drive API to your project - to do this, navigate to APIs & Services in your project console and find & add Google Drive API in the Library tab. Once downloaded, encode the credentials file with base64 (e.g. cat serviceCreds.json | base64) and use the result as GDRIVE_SERVICE_ACCOUNT_CREDENTIALS environment variable in your Travis repository settings.

Saving reports on your personal account

The uploaded files will belong to the project that you created, i.e. will not be immediately visible from your personal Google Drive UI. To be able to upload files to your personal account, you can share the reports' directory with the project account. First, note the ID of the project's user that you created to gain the service account credentials (e.g. test-123@fair-smile-123456.iam.gserviceaccount.com). You can see this on the Service Accounts tab of the project console. Now, create a directory on your Google Drive that will serve as the test root directory. Go into the directory's sharing options and paste in the project's user ID, granting it write access. Click to expand the advanced sharing options and note the ID of the shared directory that's displayed in the share link (e.g. if the link is https://drive.google.com/drive/folders/1234567890abcdef?usp=sharing, the directory's ID is 1234567890abcdef). Finally, set GDRIVE_PARENT_DIR environment variable of your Travis build to the directory ID that you noted in the previous step.

Load testing

Alongside CI, we do also CLT (Continuous Load Testing). We have our own load testing infrastructure, called Tide, which is triggered after every successful test run, and gives us a feedback on changes to MongooseIM performance.

Test results are publicly available on the Hello Tide! page.