The core took some decisions regarding testing that may not be obvious at first sight. This chapter tries to answer some of the most frequent asked ones.

Why do you docker everything?

Executing tests in a containerized environment like docker has significant advantages over local execution. It takes away all the demanding setup needs of additional services like selenium, various database systems, different php versions in parallel on the same system and so on. It also creates a well defined environment that is identical for everyone: Extension authors rely on the same system dependencies as the core does, local test execution is identical to what the continuous integration system bamboo does. All dependencies and setup details are open sourced and available for everyone. Even Travis CI is forced to create the exact same environment in our examples, and it plays well with other dockerized solutions like ddev. In short: Using docker creates stable an easy to replay results.

Docker consumes too much hard disk

Yes. If using docker often and in various scenarios, it does not tend to be exactly economical when it comes to hard disk consumption. There are various maintenance commands that help, though: First, runTests.sh -u updates local typo3gmbh containers and removes old ones. This may help. Additionally, it is a good idea to check which local containers exist using docker images and remove obsolete ones once in a while using docker rmi. Also, some container setups tend to create many docker volumes and don't remove them after use. The command docker volume list gives an overview, and a (use with care, you have a backup right?) docker volume prune helps removing unused volumes.

Why do you need runTests.sh?

The script runTests.sh is a wrapper around docker-compose. While docker and docker-compose are great software, this stack has its own type of issues. In the end, runTests.sh just creates a .env file read by docker-compose to work around a couple of things. For instance, all tests mount the local git checkout of core or an extension into one or multiple containers. Executing tests then may write files to that volume. These files should be written with the same local user that starts the tests, otherwise local files are created with permissions a local user may not be able to delete again. Specifying the user that runs a container is however not possible in docker directly and a long standing issue about that is still not resolved. There are further issues which runTests.sh hides away. The goal was to have a dead simple mechanism to execute tests. runTests.sh does that.

Additionally, wrapping details in runTests.sh gives the core the opportunity to change docker-compose details without affecting the outer caller API: We can change things on docker level, adapt the script and nothing changes on the consuming side. This detail is pretty convenient since it reduces the risk of breaking consuming scripts like other CI systems.

Is a generic runTests.sh available?

No. Maybe later. The script runTests.sh is pretty young. It did not mature enough to make it generally available for example within the typo3/testing-framework to be used by extensions directly. At the moment each extension should maintain its own copy of runTests.sh and docker-compose.yml files on its own and adapt it to its extension specific needs. We'll see how this evolves and maybe deliver some more generic solution later.

Why don't you use runTests.sh in bamboo?

For TYPO3 core testing, bamboo has its own specification in Build/bamboo/src/main/java/core/PreMergeSpec.java (and a similar nightly setup in Build/bamboo/src/main/java/core/NightlySpec.java) and uses an own docker-compose file in Build/testing-docker/bamboo/docker-compose.yml. The main reason for that is the bamboo agents are not local processes on the given hardware directly, but are also executed in docker containers to have many of them in parallel on one host.

The bamboo agents are "stupid" and only contain the java agent but don't know about PHP versions or further services. To execute single test jobs, they start new "sibling" containers and call commands in them. For instance, they start an ad hoc postgres, a redis and a memcached container, then start a PHP 7.3 container assigned to the same network and run the functional test suite. With multiple agents on one host (to make good use of many CPU's on one hardware system), these per-agent specific ad-hoc networks and volume mounts need to be separated from each other. This requires some additional setup efforts runTests.sh does not reflect. Additionally, the ram disk, responsible shell user and home directory setup is different to speed up single runs and to efficiently cache network related stuff on the local host filesystem without agents colliding with each other.

So while bamboo and local test execution use the same underlying containers, the network, local cache and volume mounts are different and this forces us to separate that from the local "one thing at a time" tailored script called runTests.sh.

Why do you not use native PHP on Travis CI?

The documentation about extension and project testing using Travis CI sets up an environment that is identical to the local testing with the same containers used locally. This circumvents the PHP versions that Travis comes with and again runs everything in docker containers. It would also possible be to use the native Travis environments. This however requires additional fiddling and more knowledge of the testing internals, especially when it comes to functional and acceptance testing. Extension authors may very well decide to do that, but a setup like that is out of scope of this document. Torturing Travis CI to fetch and run the docker containers all the time is more time consuming but simplifies the setup a lot. And it gives stable results: If it fails on Travis, it will fail locally, too.

Functional tests set up own instances in typo3temp?

Yes. This is how it works. The functional (and acceptance) bootstrap create new, isolated instances within your project root's public directory (usually .Build/Web) in typo3temp/var/tests for each execution. Maybe later we make this path configurable and locate it elsewhere, but for the time being it is how it is.

Can I provide more hardware for bamboo?

Yes and no. We indeed consume quite some hardware to keep the TYPO3 core testing at a decent speed and we are always looking for more. Under normal operation, we currently consume 64 CPU's and half a terabyte of RAM. This is something. However, it is not trivial to help us: With the current system, we need root access to the host, a host needs 8 hardware CPU's with 64 GB of RAM, the system needs to be stable and always available, it needs to be wired into the TYPO3 GmbH internal VPN, we need a decent network connection and we need to trust the system. You probably do not want to load systems like that with additional tasks since the system load can grow quite high if the agent cluster is busy. If you can comply with this, please mail to Christian Kuhn for details, it is highly appreciated! In the future our system demands may hopefully shrink, maybe we're able to establish more dynamic scaling using kubernetes for instance, but we're not there yet.

Testing is slow on macOS

Yes. macOS has issues at least in this area. We're sorry. It is how it is. There are various issues that have not been solved in a satisfiable way when it comes to mounting bigger local directories as docker container volumes. github is full of issues with mac users complaining about poor docker performance. There may be some tricks to speed that up in our tiny TYPO3 testing world and we're open to persons having a look at this to reduce pain for mac users. Other than that, there is little we can do. Eventually, maybe macOS gets it's name spacing, file system layers and containerization right in the future? Windows is better and quicker in this area by now. Let that sink in. Please keep us posted.

Can i run tests with Windows?

Well. Maybe. We've had some successful reports using runTests.sh with WSL: Windows Subsystem for Linux but we did not get too much information and experience on this, yet. Please go ahead, push patches for core runTests.sh or the docker-compose.yml files and improve this documentation with hints about Windows. We already know there are some details to take care of, a starter can be found in a core patch commit message.