9. Regression Testing

The purpose of regression testing is summarized by the following Wikipedia quote [99]:

Regression testing (rarely non-regression testing) is re-running functional and non-functional tests to ensure that previously developed and tested software still performs after a change.

9.1. reggie2.0 Tool

PICLas is continuously tested by utilizing a Python based regression testing environment, which is run by a Gitlab Runner. Therefore, the tool reggie2.0 is used, which is found under https://github.com/piclas-framework/reggie2.0. Additionally, the different Analyze routines defined in the analysis.ini files that can be applied and the general structure of a regression test is described there. Different tests are executed on check-in, during nightly or weekly testing. These tests are defined in the file .gitlab-ci.yml that is located in the top level repository directory of PICLas. In this file, various tests are defined, which are found under regressioncheck and a summary of the different tests for PICLas are given here. The automatic execution by a Gitlab Runner can be performed on any machine that is connected to the internet and in the following sections, the setup of such a machine is described.

9.2. Local execution of reggie2.0

To quickly test regression checks locally, either to reproduce an error that has occurred during a GitLab pipeline or to test newly developed code, the reggie2.0 tool can be executed without the need to install a Gitlab Runner.

  1. Install reggie from GitHub as described here and run the tool with --help to get an overview of the available options

    reggie --help

  2. Build the required executable, e.g., piclas, either automatically using the reggie2.0 tool or configure cmake and compile the executable by hand.

    Building the executable automatically requires a directory under regressioncheck that contains a builds.ini file from which all the compilation flags are read automatically and is described in the next step because automatic compilation is always accompanied by also running the code and analysing the results.

    There is also a bash script to extract a single cmake command line containing all the compiler settings from the builds.ini to compile a single executable by hand. Navigate to a build directory and run the script

    cd ~/piclas/build
./../tools/cmake-builds-ini.sh ../regressioncheck/NIG_convtest_poisson/builds.ini

    The output of the script will look like this

     CMAKE_BUILD_TYPE ........................ Debug,Release
 LIBS_BUILD_HDF5 ......................... OFF
 PICLAS_POLYNOMIAL_DEGREE ................ N
 PICLAS_EQNSYSNAME ....................... poisson
 PICLAS_TIMEDISCMETHOD ................... RK3
 LIBS_USE_MPI ............................ ON,OFF
 PICLAS_NODETYPE ......................... GAUSS
 PICLAS_PARTICLES ........................ ON,OFF
 PICLAS_CODE_ANALYZE ..................... ON,OFF
 LIBS_USE_PETSC .......................... OFF,ON

Select the first set of options via

cmake ..  -DCMAKE_BUILD_TYPE=Debug -DLIBS_BUILD_HDF5=OFF -DPICLAS_POLYNOMIAL_DEGREE=N -DPICLAS_EQNSYSNAME=poisson -DPICLAS_TIMEDISCMETHOD=RK3 -DLIBS_USE_MPI=ON -DPICLAS_NODETYPE=GAUSS -DPICLAS_PARTICLES=ON -DPICLAS_CODE_ANALYZE=ON -DLIBS_USE_PETSC=OFF

    The last line can directly be copied into the terminal within a build directory to generate the make files for compiling PICLas with the first set of parameter options given in the builds.ini file (ignoring the nocrosscombination statements). Simply adjust the last line to have the correct flags set, execute cmake in the build directory and run make to compile:

    cmake ..  -DCMAKE_BUILD_TYPE=Debug -DLIBS_BUILD_HDF5=OFF -DPICLAS_POLYNOMIAL_DEGREE=N -DPICLAS_EQNSYSNAME=poisson -DPICLAS_TIMEDISCMETHOD=RK3 -DLIBS_USE_MPI=ON -DPICLAS_NODETYPE=GAUSS -DPICLAS_PARTICLES=ON -DPICLAS_CODE_ANALYZE=ON -DLIBS_USE_PETSC=ON
make -j

    Note that -DLIBS_USE_PETSC=ON has been adjusted in the above command. This will compile the piclas executable and it will be placed in the current directory under bin. Important note: Some regression tests build the PyHOPE meshes “on-the-fly”, hence, the pyhope tool is required additionally. pyhope needs to be pre-installed on the system and an installation guide can be found on GitHub - PyHOPE.

  3. Run the reggie2.0 tool either a) automatic mode or b) pre-compiled mode:

    This file is used to compile one or more piclas executables and the directories that accompany the builds.ini file will be used for testing. Note that not all executables might be used for those directories, as they might be excluded via the excludebuild.ini file.

    To run the automatic compilation and testing procedure, simply navigate the terminal to a location outside of the regressioncheck directory (which is found in the piclas repository). Switch to the home directory and run the reggie2.0 tool there via

    cd ~
reggie /path/to/piclas/regressioncheck/NIG_DSMC

    to start compiling and executing the resulting code. All output is placed under a new directory output_dir within the current directory. Never run the reggie2.0 tool from within the /path/to/piclas/regressioncheck/ //as the directory tree structure is copied from there and the source path and target path cannot be the same! This procedure will run all the example directories under NIG_DSMC.

    ls /path/to/piclas/regressioncheck/NIG_DSMC

2D_VTS_Distribution     builds.ini                  RotPeriodicBC                 SURF_PROB_DifferentProbs  VSS_VHS_SelfDiffusion
Ambipolar_Diffusion     Macroscopic_Restart         RotPeriodicBCMulti            SURF_PROB_MultiReac
Ambipolar_Diffusion_SF  MCC_BGG_Elec_XSec_Sampling  RotPeriodicBCMultiInterPlane  VirtualCellMerge

    To run the pre-compiled executable, navigate to the corresponding build directory and run the reggie2.0 tool there

    cd ~/piclas/build
reggie -e ./bin/piclas ../regressioncheck/NIG_DSMC/Ambipolar_Diffusion

    to run a specific test case, e.g., Ambipolar_Diffusion.

  4. The analysis.ini file within the Ambipolar_Diffusion directory lists the analysis that is performed after the successful execution of piclas. An overview of the available analysis functions can be found here. Also, look into the existing regressioncheck examples to get an idea how to construct a new regressioncheck setup or modify an existing one.

  5. Reference files: They can be created automatically with the command line arguments

    -z, --rc              Create/Replace reference files that are required for analysis. After running the program, the output files are stored in the check-/example-directory.
-i, --noMPI           Run program without "mpirun" (single thread execution).

    for analysis files, which use a reference file with which the output of a run is compared. Here, the flag -i is used to create the reference file with a single-core run, which is suggested as a best practice as the actual run might be performed with multiple cores and the output should ideally be the same. An example is given under regressioncheck/WEK_DSMC/ChannelFlow_SurfChem_AdsorpDesorp_CO_O2 where .h5 files are compared.

9.3. Compression of HDF5 reference files

When creating .h5 reference files that are used for comparison in regression tests, it is very beneficial to compress these to a minimum before committing and pushing them to the git repository. Consider the following example with an analysis.ini file containing

! hdf5 diff
h5diff_file            = TestRotatingWall_DSMCSurfState_000.00100000000000000.h5
h5diff_reference_file  = TestRotatingWall_DSMCSurfState_000.00100000000000000_reference.h5
h5diff_data_set        = SurfaceData
h5diff_tolerance_value = 40E-2
h5diff_tolerance_type  = relative
h5diff_var_attribute   = VarNamesSurface
h5diff_var_name        = Total_TorqueZ
h5diff_max_differences = 15

where the only field in the .h5 file that is used for comparison in the regression test is Total_TorqueZ. Everything else in the file in not required for the reggie. As described in the previous section, most reggie analysis functions can be run with the argument -z, --rc, which copies a file resulting from a piclas simulation as reference file directly into the regression test directory to be used in future runs without the need for copying the file and renaming it by hand

reggie -iz ../regressioncheck/WEK_DSMC/Torque_Output

which creates the file TestRotatingWall_DSMCSurfState_000.00100000000000000_reference.h5 in the directory /WEK_DSMC/Torque_Output. The argument -i is only used to create the reference file from a single-core execution as the test later uses more processes, but the same result is obviously required. Additionally, unnecessary parts of the .h5 file can be removed if these are not required for the reggie analysis. This can be achieved by modifying the .h5 file directly with, e.g., hdfview. Open the reference file with hdfview, remove all datasets that are not needed and resize the dataset SurfaceData so that it only contains one single property, name Total_TorqueZ. If most of the options, when right-clicking on the dataset are not available, make sure to Reload File As -> Read/Write first. After saving the file to the disk, it is necessary to remove the occupied disk space of the deleted datasets via

h5repack -i original.h5 -o compressed.h5

and

h5repack -v -f SHUF -f GZIP=9 original.h5 compressed-shuffled-GZIP9.h5
h5repack -v -f GZIP=9 original.h5 compressed-GZIP9.h5

by trying both commands and using the smaller resulting .h5 file. Note that shuffle by itself will do nothing to compress. It simply re-orders bytes in memory in hopes of making the resulting byte-stream easier to compress for something like GZIP than it would be able to do otherwise. This is because GZIP is a byte-level compressor. It doesn’t know about things like shorts or ints or doubles.

9.4. Running GitLab .gitlab-ci.yml Tests

The GitLab CI/CD tests can either be run locally or remotely and both methods are explained in the following. The tests are defined in the file .gitlab-ci.yml in the top-level directory of the piclas repository.

9.5. Remote Testing on Gitlab

Open a browser and go to the piclas gitlab pipelines website, where the latest pipeline jobs are displayed. To start a new pipeline, click the button Run pipeline and select the required branch name or tag, which should be tested. Then, activate the necessary inputs by selecting true, which are summarized in Table 9.1.

Table 9.1 Gitlab pipeline inputs

Property

Description

DO_CHECKIN

short tests that are also run, when new commits are pushed

DO_NIGHTLY

longer tests, executed every day

DO_WEEKLY

very long tests, executed once a week

DO_CODE_COVERAGE

Generate coverage data of piclas for all jobs in current pipeline

DO_REGGIE_COVERAGE

Generate coverage data of reggie2.0 itself for all jobs in current pipeline

DO_NODE_SPLIT

MPI: virtual CPU splitting for multi-node testing, where a specific number of cores/threads are grouped in separate nodes (default is 2)

DO_CORE_SPLIT

MPI: virtual CPU splitting for multi-node testing, where each core/thread resembles a separate node

DO_MPICH

MPI: Force compilation using MPICH instead of OpenMPI

Per default, DO_CHECKIN, DO_NIGHTLY, DO_WEEKLY, DO_NODE_SPLIT and DO_CORE_SPLIT are tested automatically for the branch master.dev. For details, see the Pipeline schedules section in Gitlab.

9.6. Code Coverage

Code coverage information can be inspected in multiple ways: on GitLab, by downloading GitLab artifacts and inspecting the data locally, or by running the tests locally in the first place. By default, all lines containing Call Abort() or Call CollectiveStop() are not considered when creating coverage reports.

9.6.1. Coverage on GitLab

To enable code coverage when running regression tests on GitLab, set the pipeline input DO_CODE_COVERAGE to true. This creates a separate stage, which is executed after all other stages. This way the code coverage stage is able to collect the coverage data from all (previous) jobs/runs of the current pipeline. This is done in the following way.

For each job (e.g., CHE_DSMC), the coverage data is collected and stored in a .json report file per build. These reports are named after the build and stored in the Coverage directory. If the build is reused for another job (e.g., same DSMC build for both CHE_DSMC and NIG_DSMC), the report file is updated. After all jobs in the other stages are finished, the coverage stage starts. The report files from all builds are combined into a single report containing the coverage data from all previously triggered runs/builds of the current pipeline. For example, if only the inputs DO_CODE_COVERAGE and DO_CHECKIN are set, the report will show coverage data from all builds with corresponding runs in the DO_CHECKIN case. The output of the coverage stage will also display which .json report files are found and therefore used to check whether the correct files and builds are considered.

9.6.1.1. Inspecting data on GitLab

The coverage stage will create a GitLab report artifact, which is used for the visualization. GitLab currently uses the coverage report of the latest pipeline of the current branch (if not expired yet) to display the coverage. Note that if the latest pipeline did not create coverage data, nothing will be displayed in the merge request difference view. This is done to prevent a false representation for new code evaluated with older coverage information. Moreover, displaying older coverage data would not be possible if the source code has changed. Keep in mind that new pipelines might change the coverage data if a different set of tests is run. The coverage report is shown in the merge request difference view/changes. A line that was tested is indicated by a green bar to its left, otherwise a red bar appears. This allows inspection of regression tests for new features directly on GitLab. For smaller features/tests, it is recommended to check the coverage locally first to avoid triggering unnecessary tests.

To generate a full coverage report for all available regression tests of PICLas, either:

  • Execute all tests in the same pipeline with DO_CODE_COVERAGE set to true, or

  • Combine separate reports manually

On GitLab the coverage is shown as a single number either in the output of the coverage job, on the right side when inspecting the job, or even in the merge request view. The displayed number is the line coverage (different coverage types in “Inspecting data locally”), which is set in .gitlab-ci.yml.

Currently the setup does not allow to test node/core split together, since the report can only be created per pipeline. Therefore, the reports have to be combined manually or inspected separately.

9.6.1.2. Inspecting data locally

In addition to the report artifact, the data is stored as a GitLab job artifact, which can be downloaded and inspected locally. This can be done on GitLab in the Pipeline overview on the right under Actions by choosing the Coverage artifact. To get a well-structured view of the data, it is recommended to use the Coverage/html/coverage.html file. Total coverage is displayed at the top right and split into three different categories: Lines, Functions and Branches. Lines checks how many of the compiled lines were actually executed. Lines that were covered by tests are highlighted in green, while untested lines appear in red. Yellow lines indicate some kind of branching, e.g. an if clause, where one condition was not tested. Functions works similarly, but only checks for function calls. So if a file contains only one function and it is called somewhere else, this will result in 100% coverage even if only 10% of lines were executed in this function. Lastly, Branches check how many different paths the program could take and then counts how many it did. So this mainly counts if/else statements or case select blocks.

9.6.2. Local coverage tests

Since reggie supports coverage outputs, it can also be generated locally. To enable code coverage information, each executable must be compiled with additional flags via the PICLAS_CODE_COVERAGE option. This generates additional .gcno and .gcda files per object file, which track all compiled lines and the number of calls per line. For example, create a build directory and compile the Poisson solver with the Leapfrom time integration scheme

mkdir build_poisson_code_coverage && cd build_poisson_code_coverage
cmake .. -DPICLAS_EQNSYSNAME=poisson -DPICLAS_TIMEDISCMETHOD=Leapfrog -DLIBS_USE_PETSC=ON -DPICLAS_CODE_COVERAGE=ON
make -j

To enable code coverage when using reggie locally, use the -o option. More information can be found in the reggie documentation. Note that due to the module names, e.g., “__mod_dsmc_MOD_dsmc_main”, gcovr must be run with --include-internal-functions, otherwise all functions will be excluded. This can be done with the additional reggie2.0 flag --gcovr_extra, e.g. --gcovr_extra \'--include-internal-functions\'.

To run the regression test and create the code coverage data for piclas

reggie -e bin/piclas ../regressioncheck/CHE_poisson_p_adaption/Laplace_h_N1_pAdaptionType0/ -o 1

and check if the report is created in the output_dir directory

find . -name *report.html

View the report with any browser

firefox ./output_dir/Coverage/combined_report/html/combined_report.html

9.6.2.1. Combining single reports

In some cases, it might be helpful to combine single reports of different reggie runs. This can be done using gcovr, which is the same tool that reggie2.0 uses itself. For this case, .json files are used. Separate reports can be combined with

gcovr --root <root_dir> --add-tracefile <json_file1> --add-tracefile <json_file2> --html-nested report_name.html

where --add-tracefile takes wildcard arguments as well. The --html-nested option generates the same output format as reggie2.0 on GitLab, which is nicely structured analogous to the src directory. --root specifies the root directory containing the source files and report_name.html the output file.

Note that <root_dir> must be the same directory as the one used for the gcovr call that originally created the single report files. E.g. all single reports are created with --root ~/some_dir, then the <root_dir> for combining the reports must also be --root ~/some_dir. Reggie2.0 usually tries to append src to the found root directory to exclude UnitTests in the coverage information for piclas. Therefore combining reports generated by reggie2.0 for piclas would be done with

gcovr --root /path/to/piclas/src --add-tracefile Coverage/*.json --html-nested report_name.html

For more information on gcovr and coverage report formats, see the gcovr documentation.

9.6.3. Reggie coverage

Besides generating code coverage reports of PICLas, it is also possible to generate a report of the reggie tool itself. To do this, set the pipeline input DO_REGGIE_COVERAGE to true, which wraps each reggie call with the Python coverage tool. This generates a coverage report of all used lines in the reggie module, which is stored as a GitLab artifact. The report can be inspected using the Coverage/reggie/index.html file.

9.7. Local Testing using gitlab-ci-local

To locally test the GitLab CI (including a YAML verification), gitlab-ci-local can be used. An installation guide can be found here. After a successful installation, you can view the available parameters through

gitlab-ci-local --help

To view the stages for the default check-in pipeline, execute in the main folder of piclas:

gitlab-ci-local --list

To view all stages and tests:

gitlab-ci-local --list-all

To execute the check-in pipeline locally (that is the jobs that were shown with the --list command), use

gitlab-ci-local --shell-isolation

to avoid errors due to parallel writing of the ctags.txt file. An alternative is to limit the concurrent execution to one job, which is analogous to the current configuration on the Gitlab Runner (requires gitlab-ci-local in version 4.42.0)

gitlab-ci-local --concurrency=1

It should be noted that currently the cache creation & utilization does not seem to represent the remote execution, meaning that some errors might only be recognized after a push to the remote. A specific job can be executed simply by reference its name, and to also consider the dependencies (i.e. the needs:), the following command can be utilized to execute, for example the DSMC check-in job:

gitlab-ci-local --needs CHE_DSMC

Another useful option to check the resulting configuration file is

gitlab-ci-local --preview preview.yml

which gives the expanded version of utilized extends: and <<: templates.

When running gitlab-ci-local on a system with a module environment, it is neccessary to pass the local modules that are used for compiling

DO_RUN_LOCAL="cmake/3.30.3   gcc/14.2.0   mpich/4.1.2/gcc/14.2.0    hdf5/1.14.0/gcc/14.2.0/mpich/4.1.2    petsc/3.21.6/gcc/14.2.0/mpich/4.1.2"
gitlab-ci-local --input DO_RUN_LOCAL=$DO_RUN_LOCAL

If multiple inputs are required add them to the command

gitlab-ci-local --input DO_RUN_LOCAL=$DO_RUN_LOCAL --input CHECK_WARNINGS=true

to envoke additional options of the pipeline.

9.7.1. Example

To run a specific reggie job, in this case a weekly reggie that depends on another job, the following parameters are passed

DO_RUN_LOCAL="cmake/3.30.3   gcc/14.2.0   mpich/4.1.2/gcc/14.2.0    hdf5/1.14.0/gcc/14.2.0/mpich/4.1.2    petsc/3.21.6/gcc/14.2.0/mpich/4.1.2"
gitlab-ci-local --shell-isolation --needs WEK_Radiation --input DO_RUN_LOCAL=$DO_RUN_LOCAL --input DO_WEEKLY=true

where the arguments are listed and explained in Table 9.2

Table 9.2 gitlab-ci-local inputs example

Parameter

Description

–shell-isolation

Avoid errors due to parallel writing of the ctags.txt file

–needs WEK_Radiation

Run WEK_Radiation, which also requires WEK_DSMC_Radiation

–input DO_RUN_LOCAL=$DO_RUN_LOCAL

Pass the locally installed and used modules

–input DO_WEEKLY=true

WEK_Radiation is a weekly reggie that requires the DO_WEEKLY to be passed

9.8. Regression Test Gitlab Runner Setup for self-hosted Servers

This section describes the necessary steps to install a Gitlab Runner on a Ubuntu system to run Gitlab Build Pipelines. In a first step, the required software packages for PICLas and reggie2.0 are installed on a new system. In a second step, the gitlab-runner program is installed and the setup of runner is described.

9.8.1. Prerequisites: Installation of Software on Clean Ubuntu Setup (18.04)

Latest tests on

  • Ubuntu (18.04), 3 Jul 2019

  • Ubuntu server (18.04.3 LTS), 19 Nov 2019

The following packages can be installed automatically by using the script located at ./tools/Setup_ModuleEnv/InstallPackagesReggie.sh. The system inquiries can be skipped by forcing yes as input via

yes | ./InstallPackagesRe

The script contains the following packages

# Check for updates
sudo apt-get update

# compiler
sudo apt-get install make cmake cmake-curses-gui gfortran g++ gcc

# python
sudo apt-get install python python-numpy python3 python3-numpy python3-matplotlib python-matplotlib

# editors
sudo apt-get install gedit vim vim-runtime vim-common vim-tiny gdb vim-gui-common

# git && versions
sudo apt-get install git gitg qgit subversion

# paraview
sudo apt-get install libpython-dev libboost-dev libphonon-dev libphonon4 libxt-dev mesa-common-dev
#apt-get install qt4-default qt4-dev-tools libqt4-dev qt4-qmake libqt4-opengl-dev
sudo apt-get install qttools5-dev libqt5x11extras5-dev qt5-default libgl1-mesa-dev

# Tecplot
sudo apt-get install libstdc++5

# tools
sudo apt-get install gzip gimp htop meld gnuplot gnuplot-x11 vlc okular ddd gmsh unzip
sudo apt-get install openvpn openssl openssh-client

# for FLEXI/PICLas
sudo apt-get install liblapack3 liblapack-dev zlib1g-dev exuberant-ctags

# for documentation
sudo apt-get install texlive-base
sudo apt-get install texlive-latex-extra

# hdf5-file viewer
sudo apt-get install hdfview

# Install libs for reggie
sudo apt-get install python-h5py

When no module environment is to be used on the server, the following packages are also required

# Further libs
sudo apt-get install hdf5-tools libhdf5-dev # this is maybe not required (do not install them if it works without these packages)


# openMPI
wget https://download.open-mpi.org/release/open-mpi/v3.1/openmpi-3.1.3.tar.gz
tar -xvf openmpi-3.1.3.tar.gz
cd openmpi-3.1.3
mkdir -p build
cd build
../configure --prefix=/opt/openmpi/3.1.3
sudo make all install
export PATH="/opt/openmpi/3.1.3/bin:$PATH"
export LD_LIBRARY_PATH="/opt/openmpi/3.1.3/lib:$LD_LIBRARY_PATH"
cd ../..
rm openmpi-3.1.3.tar.gz


#HDF5
wget https://support.hdfgroup.org/ftp/HDF5/releases/hdf5-1.10/hdf5-1.10.5/src/hdf5-1.10.5.tar.bz2
tar -xjf hdf5-1.10.5.tar.bz2
cd hdf5-1.10.5
mkdir -p build
cd build
cmake -DBUILD_TESTING=OFF -DHDF5_BUILD_FORTRAN=ON -DHDF5_BUILD_CPP_LIB=OFF -DHDF5_BUILD_EXAMPLES=OFF -DHDF5_ENABLE_PARALLEL=ON -DHDF5
sudo make && sudo make install
export HDF5_DIR=/opt/hdf5/1.10.5/share/cmake
cd ../..
rm hdf5-1.10.5.tar.bz2

otherwise a module environment can be installed at this point, see ~/piclas/tools/Setup_ModuleEnv/README.md, which is explained in detail in Chapter Developer Tools under Section Module Environment.

When no module environment is to be used on the server, the following commands must be places in the .gitlab-ci.yml file:

# Export the paths on new reggie2@reggie2 (no module env any more)
before_script:
  - ulimit -s unlimited
  - export PATH=/opt/openmpi/3.1.3/bin:$PATH
  - export LD_LIBRARY_PATH=/opt/openmpi/3.1.3/lib/:$LD_LIBRARY_PATH
  - export CMAKE_PREFIX_PATH=/opt/openmpi/3.1.3/share/cmake:$CMAKE_PREFIX_PATH
  - export CMAKE_LIBRARY_PATH=/opt/openmpi/3.1.3/lib:$CMAKE_LIBRARY_PATH
  - export HDF5_DIR=/opt/hdf5/1.10.5/share/cmake/
  - export PATH=/opt/hdf5/1.10.5/bin:$PATH
  - export LD_LIBRARY_PATH=/opt/hdf5/1.10.5/lib/:$LD_LIBRARY_PATH
  - export CMAKE_PREFIX_PATH=/opt/hdf5/1.10.5/:$CMAKE_PREFIX_PATH
  - export CMAKE_LIBRARY_PATH=/opt/hdf5/1.10.5/lib:$CMAKE_LIBRARY_PAT

otherwise, the correct environment must be loaded by adding the following in /etc/profile

```
# Default modules
module load gcc/9.2.0  cmake/3.15.3-d  openmpi/4.0.1/gcc/9.2.0  hdf5/1.10.5/gcc/9.2.0/openmpi/4.0.1
```

or by loading the modules directly in the gilab script file, e.g.,

```
module load XX/XX
```

NOTE: The stack size limit has been removed here by ulimit -s unlimited, which might be required by memory consuming programs

9.8.2. Installation of Gitlab Runners

Latest tests on

  • Ubuntu (18.04) with gitlab-runner 10.5.0 (10.5.0), 3 Jul 2019

  • Ubuntu server (18.04.3 LTS) with gitlab-runner 10.5.0 (10.5.0), 19 Nov 2019

  1. Install gitlab-runner from ubuntu packages (choose old version to avoid problems https://gitlab.com/gitlab-org/gitlab-runner/issues/1379) This creates the user gitlab-runner and a home directory (for 10.5 in /var/lib/gitlab-runner/)

    sudo apt-get install gitlab-runner

  2. Start the runner program

    sudo gitlab-runner start

  3. Register a runner using a shell executor (follow the information in the official guideline as it can vary slightly from version to version), on gitlab see Settings \(\rightarrow\) CI / CD Settings \(\rightarrow\) Runners (registration token during setup)

    sudo gitlab-runner register

  4. Restart the runner

    sudo gitlab-runner restart

  5. create ssh keys for normal user and set up password free access to gitlab (https://piclas.boltzplatz.eu)

    ssh-keygen -t ecdsa -b 521

    Add key to Enabled deploy keys. If multiple codes are on gitlab, add the key to one repository and select the key on the other repositories via Privately accessible deploy keys.

    Clone a code from each platform to create known hosts then

    sudo cp .ssh/*  /var/lib/gitlab-runner/.ssh

    Check rights in this folder. If necessary make gitlab-runner own the files with

    sudo chown -R gitlab-runner:gitlab-runner /var/lib/gitlab-runner/.ssh/

    If the runner is used to push to remote repositories, add the public key under deploy keys and execute, e.g.,

    sudo -u gitlab-runner git clone git@github.com:piclas-framework/piclas.git piclas_github

    to establish the first connection with the new repository and add the repo IP to the list of known hosts.

  6. Start pipeline in gitlab or github for testing of reggie

NOTE: Interesting information is found in /etc/systemd/system/gitlab-runner.service.

9.8.3. Configuration Files

The runner services can be adjusted by changing the settings in the file

/etc/systemd/system/gitlab-runner.service

in which the runner configuration file is specified:

[Unit]
Description=GitLab Runner
After=syslog.target network.target
ConditionFileIsExecutable=/usr/bin/gitlab-runner

[Service]
StartLimitInterval=5
StartLimitBurst=10
ExecStart=/usr/bin/gitlab-runner "run" "--working-directory" "/var/lib/gitlab-runner/" "--config" "/etc/gitlab-runner/config.toml" "--service" "gitlab-runner" "--syslog" "--user" "gitlab-runner"


Restart=always
RestartSec=120

[Install]
WantedBy=multi-user.target

The runner configuration settings can be edited by changing the settings in the file

/etc/gitlab-runner/config.toml

where the number of runners, the concurrency level and runner limits are specified:

concurrent = 2
check_interval = 0

[[runners]]
  name = "myrunner1"
  url = "https://gitlab.com/"
  token = "XXXXXXXXXX"
  executor = "shell"
  limit = 1
  [runners.cache]

[[runners]]
  name = "myrunner2"
  url = "https://gitlab.com/"
  token = "XXXXXXXXXX"
  executor = "shell"
  limit = 1
  [runners.cache]

[[runners]]
  name = "myrunner3"
  url = "https://gitlab.com/"
  token = "XXXXXXXXXX"
  executor = "shell"
  limit = 1
  [runners.cache]

9.8.4. Automatic Deployment to Other Platforms (GitHub)

  1. Add the required ssh key to the deploy keys on the respective platform (e.g. github)

  2. Clone a code from the platform to update the list of known hosts. Do not forget to copy the information to the correct location for the runner to have access to the platform

    sudo cp~/.ssh/.ssh/known_hosts /var/lib/gitlab-runner/.ssh/known_hosts

    This might have to be performed via the gitlab-runner user, which can be accomplished by executing the following command

    sudo -u gitlab-runner git clone git@github.com:piclas-framework/piclas.git piclas_github

  3. PICLas deployment is performed by the gitlab runner in the deployment stage

    github:
  stage: deploy
  tags:
    - withmodules-concurrent
  script:
    - if [ "$[[ inputs.DO_DEPLOY ]]" == "false" ]; then exit ; fi
    - rm -rf piclas_github || true ;
    - git clone -b master --single-branch git@piclas.boltzplatz.eu:piclas/piclas.git piclas_github ;
    - cd piclas_github ;
    - git remote add piclas-framework git@github.com:piclas-framework/piclas.git ;
    - git push --force --follow-tags piclas-framework master ;

    This script clones the master branch of PICLas and deploys it on github.