crowdsec/test

What is this?

This directory contains scripts for functional testing. The tests are run with the bats-core framework, which is an active fork of the older BATS (Bash Automated Testing System).

With the addition of the ansible playbooks it is possible to use VMs to test the binary packages, service management and other CPU architectures.

cscli

Feature	Covered	Notes
cscli alerts	-
cscli bouncers	10_bouncers
cscli capi	01_base	status only
cscli collections	20_collections
cscli config	01_base	minimal testing (no crash)
cscli dashboard	-	docker inside docker 😞
cscli decisions	9[78]_ipv[46]*
cscli hub	dyn_bats/99_hub
cscli lapi	01_base
cscli machines	30_machines
cscli metrics	-
cscli parsers	-
cscli postoverflows	-
cscli scenarios	-
cscli simulation	50_simulation
cscli version	01_base

crowdsec

Feature	Covered	Notes
systemctl start/stop/restart	-
agent behavior	40_live-ban	minimal testing (simple ssh-bf detection)
forensic mode	40_cold-logs	minimal testing (simple ssh-bf detection)
starting without LAPI	02_nolapi
starting without agent	03_noagent
starting without CAPI	04_nocapi
prometheus testing	-

API

Feature	Covered	Notes
alerts GET/POST	9[78]_ipv[46]*
decisions GET/POST	9[78]_ipv[46]*
stream mode	`99_lapi-stream-mode

How to use it

pre-requisites

• git submodule init; git submodule update
• base64
• bash>=4.4
• curl
• daemonize
• jq
• nc
• openssl
• openbsd-netcat
• python3

Running all tests

Run make clean bats-all to perform a test build + run. To repeat test runs without rebuilding crowdsec, use make bats-test.

Debugging tests

See ./test/run-tests --help to run/debug specific tests.

Example: ./test/run-tests test/bats/02_nolapi.bats -f "cscli config backup" (the string is a regexp). You need to provide a path for a test file or directory (even if it's the full 'test/bats') to use the -f option.

How does it work?

In BATS, you write tests in the form of Bash functions that have unique descriptions (the name of the test). You can do most things that you can normally do in a shell function. If there is any error condition, the test fails. A set of functions is provided to implement assertions, and a mechanism of setup/teardown is provided at the level of individual tests (functions) or group of tests (files).

The stdout/stderr of the commands within the test function are captured by bats-core and will only be shown if the test fails. If you want to always print something to debug your test case, you can redirect the output to the file descriptor 3:

@test "mytest" {
   echo "hello world!" >&3
   run some-command
   assert_success
   echo "goodbye." >&3
}

If you do that, please remove it once the test development is finished, because this practice breaks the TAP protocol (unless each line has a '#' as first character, but really, it's better to avoid unnecessary output when tests succeed).

You can find here the documentation for the main framework and the plugins we use in this test suite:

• bats-core tutorial
• Writing tests
• bats-assert
• bats-support
• bats-file
• bats-mock

As it often happens with open source, the first results from search engines refer to the old, unmaintained forks. Be sure to use the links above to find the good versions.

Since bats-core is TAP (Test Anything Protocol) compliant, its output is in a standardized format. It can be integrated with a separate tap reporter or included in a larger test suite. The TAP specification is pretty minimalist and some glue may be needed.

setup and teardown

If you have read the bats-core tutorial linked above, you are aware of the setup and teardown functions.

What you may have overlooked is that the script body outside the functions is executed multiple times, so we have to be careful of what we put there.

Here we have a look at the execution flow with two tests:

echo "begin" >&3

setup_file() {
        echo "setup_file" >&3
}

teardown_file() {
        echo "teardown_file" >&3
}

setup() {
        echo "setup" >&3
}

teardown() {
        echo "teardown" >&3
}

@test "test 1" {
        echo "test #1" >&3
}

@test "test 2" {
        echo "test #2" >&3
}

echo "end" >&3

The above test suite produces the following output:

begin
end
setup_file
begin
end
 ✓ test 1
setup
test #1
teardown
begin
end
 ✓ test 2
setup
test #2
teardown
teardown_file

See how "begin" and "end" are repeated three times each? The code outside setup/teardown/test functions is really executed three times (more as you add more tests). You can put there variables or function definitions, but keep it to a minimum and don't write anything to the standard output. For most things you want to use setup_file() instead.

But.. there is a but. Quoting from the FAQ:

You can simply source .sh files. However, be aware that sourceing files with errors outside of any function (or inside setup_file) will trip up bats and lead to hard to diagnose errors. Therefore, it is safest to only source inside setup or the test functions themselves.

This doesn't mean you can't do that, just that you're on your own if the is an error.

Testing crowdsec

Fixtures

For the purpose of functional tests, crowdsec and its companions (cscli, plugin notifiers, bouncers) are installed in a local environment, which means tests should not install or touch anything outside a ./test/local directory. This includes binaries, configuration files, databases, data downloaded from internet, logs... The use of /tmp is tolerated, but BATS also provides three useful variables: $BATS_SUITE_TMPDIR, $BATS_FILE_TMPDIR and $BATS_TEST_TMPDIR that let you ensure your desired level of isolation of temporary files across the tests.

When built with make bats-build, the binaries will look there by default for their configuration and data needs. So you can run ./local/bin/cscli from a shell with no need for further parameters.

To set up the installation described above we provide a couple of scripts, instance-data and instance-crowdsec. They manage fixture and background processes; they are meant to be used in setup/teardown in several ways, according to the specific needs of the group of tests in the file.

• instance-data make

  Creates a tar file in ./local-init/init-config-data.tar. The file contains all the configuration, hub and database files needed to restore crowdsec to a known initial state. Things like machines add ..., capi register, hub update, collections install crowdsecurity/linux are executed here so they don't need to be repeated for each test or group of tests.

• instance-data load

  Extracts the files created by instance-data make for use by the local crowdsec instance. Crowdsec must not be running while this operation is performed.

• instance-crowdsec [ start | stop ]

  Runs (or stops) crowdsec as a background process. PID and lockfiles are written in ./local/var/run/.

Here are some ways to use these two scripts.

• case 1: load a fresh crowsec instance + data for each test (01_base, 10_bouncers, 20_collections...)

  This offers the best isolation, but the tests run slower. More importantly, since there is no concept of "grouping" tests in bats-core with the exception of files, if you need to perform some setup that is common to two or more tests, you will have to repeat the code.

• case 2: load a fresh set of data for each test, but run crowdsec only for the tests that need it, possibly after altering the configuration (02_nolapi, 03_noagent, 04_nocapi, 40_live-ban)

  This is useful because: 1) you sometimes don't want crowdsec to run at all, for example when testing cscli in isolation, or you may want to tweak the configuration inside the test function before running the lapi/agent. See how we use yq to change the YAML files to that effect.

• case 3: start crowdsec with the initial set of configuration+data once, and keep it running for all the tests (50_simulation, 98_ipv4, 98_ipv6)

  This offers no isolation across tests, which over time could break more often as result, but you can rely on the test order to test more complex scenarios with a reasonable performance and the least amount of code.

status, stdout and stderr

As we said, if any error occurs inside a test function, the test fails immediately. You call mycommand, it exits with $? != 0, the test fails.

But how to test the output, then? If we call run mycommand, then $? will be 0 allowing the test to keep running. The real error status is stored in the $status variable, and the command output and standard error content are put together in the $output variable. By specifying run --separate-stderr, you can have separated $output and $stderr variables.

The above is better explained in the bats-core tutorial. If you have not read it yet, now is a good time.

For convenience, the rune function is an alias for run --separate-stderr, which can be used in most cases. For example, you don't want extraneous log messages in the way when you check the output of a command.

The $output variable gets special treatment with the bats-support and [bats-assert][https://github.com/bats-core/bats-assert) plugins and can be checked with assert_* commands. The $stderr variable does not have these, but we can use run echo "$stderr" and then check $output with asserts.

Remember that run always overwrites the $output variable, so if you consume it with run jq <(output) you can only do it once, because the second time it will read the output of the jq command. But you can construct a list of all the values you want and check them all in a single step.

Note that <(output) is substituted with the file name of a file descriptor, so mycmd <(output) can become mycmd /dev/fd/23, mycmd /tmp//sh-np.hpc7Zs or mycmd /proc/self/fd/38 depending on the platform. To have it fed to standard input, use < <(output).

See the lib/*.sh and bats/*.bats files for other tricks we employ.

file operations

We included the bats-file plugin to check the result of file system operations: existence, type/size/ownership checks on files, symlinks, directories, sockets.

mocking external commands

The bats-mock plugin allows you to define a "fake" behavior for the external commands called by a package under test, and to record and assert which parameters are passed to it.

gotchas

• pay attention to tests that are not run - for example "bats warning: Executed 143 instead of expected 144 tests". They are especially tricky to debug.

• using the load command in teardown() causes tests to be silently skipped or break in "funny" ways. The other functions seem safe.

Testing with MySQL and Postgres

By default, the tests are run with the embedded sqlite database engine. This should be enough in most cases, since the database operations are abstracted via the ent ORM.

You can however easily test with a different engine.

Postgres

Run Postgres somewhere, version 10 or above - easy to do in a docker container.

You also need to install a postgresql-client package or equivalent, to provide recent pg_dump and pg_restore executables (not older than the PG version in the docker container).

$ sudo docker run --detach --name=postgres -p 5432:5432 --env="POSTGRES_PASSWORD=postgres" postgres:latest

The name of the container is not really important. If you are not using Docker, you may need to adjust the PGHOST/PGPORT/PGPASSWORD/PGUSER variables (defaults are 127.0.0.1, 5432, postgres, postgres).

An additional user and database both named crowdsec_test will be created.

Now you can build and run the tests (we skip bats-test-hub here, they really should not be affected by a change in DB).

$ export DB_BACKEND=postgres
$ make clean bats-build bats-fixture bats-test

or with the pgx driver:

$ export DB_BACKEND=pgx
$ make clean bats-build bats-fixture bats-test

The value of DB_BACKEND must not change between the build/fixture/test steps.

MySQL/MariaDB

Same considerations as above, with the following changes:

$ sudo docker run --cap-add=sys_nice --detach --name=mysql -p 3306:3306  --env="MYSQL_ROOT_PASSWORD=password" mysql
[...]
$ export DB_BACKEND=mysql
$ make clean bats-build bats-fixture bats-test

or for MariaDB

$ sudo docker run --cap-add=sys_nice --detach --name=mariadb -p 3306:3306  --env="MYSQL_ROOT_PASSWORD=password" mariadb

A mysql-client package is required as well.

troubleshooting

• CAPI is disabled, why? Most tests don't need it. Helper scripts are provided in test/enable-capi and test/disable-capi for interactive use, and two library functions config_enable_capi and config_disable_capi to call inside the tests. You still need to call cscli capi register after enabling it.

• My tests are hanging forever, why? See if you have a jq/yq or similar process waiting for standard input. Hint: you can pass a file from the result of the previous run command with <(output). This substitutes the expression with a file name, but if you really want it in standard input, you have to use < <(output). Bash is awesome but the syntax is often weird.

• I can't do X with jq. If you prefer you can use yq. It can parse and generate json, and it has a different syntax.

• I get "while parsing /tmp/....: yaml: line 5: mapping values are not allowed in this context" Check the heredocs (the <<EOT blocks). Each line must start with a hard TAB followed by spaces. You are probably missing some tabs.

gotchas

• Testing with Postgres or MySQL/MariaDB leads to (unpredictably) failing tests in the GitHub workflows, so we had to disable them by default. We do run these in a separate environment before doing releases. They should always pass if you run them in a development machine.