Have one Dockerfile instead of multiples

[valentin-seitz]

As discussed in precice/precice#1700 this PR introduces a small, but working protoype of the ansatz described in precice/precice#1700 using just one Dockerfile

To execute the test use python systemtests.py --suites=openfoam-adapter-release
or python systemtests.py --suites=calculix_test

It currently supports the following adapters:

• fenics
• nutils
• openfoam
• calculix

But this approach is more flexible as we can easily "install" a certain precice/precice version by using the PRECICE_REF build argument without the need of publishing a tag to a docker registry first.

[MakisH]

Looks good generally, how/what should I test?

What is the problem with CalculiX? The steps look very similar to the VM provisioning script, which works: https://github.com/precice/vm/blob/develop/provisioning/install-calculix.sh

I guess you probably stumbled upon the GCC-related issue, since you are basing on Ubuntu 22.04, and the VM on 20.04: https://precice.org/adapter-calculix-get-adapter.html#compiling-with-gcc-10-or-newer

Here is how we solve this when preparing Debian packages: https://github.com/precice/calculix-adapter/blob/9057470c6b7018a28895c24c964f65364a7fbd04/.github/workflows/ubuntu_build.yml#L16

[MakisH]

Closes precice/openfoam-adapter#184

[valentin-seitz]

@MakisH if you could maybe have a quick look if it also runs smoothly on your machine?

Mainly python systemtests.py --suites=calculix_test
and maybe python systemtests.py --suites=fenics_test

[MakisH]

@MakisH if you could maybe have a quick look if it also runs smoothly on your machine?

Mainly python systemtests.py --suites=calculix_test and maybe python systemtests.py --suites=fenics_test

The calculix_test seems to be a false negative (it fails after a long time, but it seems to work actually):

~/repos/precice/tutorials [chonky-dockerfile]$ python tools/tests/systemtests.py --suites=calculix_test
Using log-level: INFO
INFO: About to run the following systemtest in the directory /home/gc/repos/precice/tutorials/runs:
 [Perpendicular flap (fluid-openfoam, solid-calculix)]
INFO: Started running Perpendicular flap (fluid-openfoam, solid-calculix),  0/1
CRITICAL: Systemtest Perpendicular flap (fluid-openfoam, solid-calculix) had serious issues executin the docker compose command about to kill the docker compose command. Please check the logs! Command '['docker', 'compose', '--file', 'docker-compose.tutorial.yaml', 'up', '--build']' timed out after 360 seconds
CRITICAL: Docker Compose failed, skipping fieldcompare and writing stdout.log and stderr.log
INFO: Running Perpendicular flap (fluid-openfoam, solid-calculix) took 925.802237306998 seconds
ERROR: Failed to run Perpendicular flap (fluid-openfoam, solid-calculix)

even though the run completes and both solvers exit with zero:

stderr.log
stdout.log

The fenics_test succeeded:

~/repos/precice/tutorials [chonky-dockerfile]$ python tools/tests/systemtests.py --suites=fenics_test
Using log-level: INFO
INFO: About to run the following systemtest in the directory /home/gc/repos/precice/tutorials/runs:
 [Flow over heated plate (fluid-openfoam, solid-fenics)]
INFO: Started running Flow over heated plate (fluid-openfoam, solid-fenics),  0/1
INFO: Running Flow over heated plate (fluid-openfoam, solid-fenics) took 269.88722395299556 seconds
INFO: Success running Flow over heated plate (fluid-openfoam, solid-fenics)

However, in contrast to the failed case, I don't see any stderr.log and stdout.log files in the archive. These would always be useful (to check that indeed the test did what it was supposed to), especially since we now only print the overview.

[valentin-seitz]

The calculix_test seems to be a false negative (it fails after a long time, but it seems to work actually):

Yes and no :)
The script will also fail once it times out. Since you probably had to built the calculix image from scratch it probably timed out.
The newest version of the systemtests addresses it such that the actual solver time is the one that has a timeout as a deadlock is way more common there. (if one participant segfaults etc)

However, in contrast to the failed case, I don't see any stderr.log and stdout.log files in the archive. These would always be useful (to check that indeed the test did what it was supposed to), especially since we now only print the overview.

Is added, we now write always the logs into the .log files

[MakisH]

Works now, great job!

~/repos/precice/tutorials/tools/tests [develop]$ python systemtests.py --suites=calculix_test
Using log-level: INFO
INFO: About to run the following systemtest in the directory /home/gc/repos/precice/tutorials/runs:
 [Perpendicular flap (fluid-openfoam, solid-calculix)]
INFO: Started running Perpendicular flap (fluid-openfoam, solid-calculix),  0/1
INFO: Running Perpendicular flap (fluid-openfoam, solid-calculix) took 191.29625335900005 seconds
INFO: Success running Perpendicular flap (fluid-openfoam, solid-calculix)
+-------------------------------------------------------+---------+-------------------+-----------------+-----------------------+
| systemtest                                            | success | building time [s] | solver time [s] | fieldcompare time [s] |
+-------------------------------------------------------+---------+-------------------+-----------------+-----------------------+
| Perpendicular flap (fluid-openfoam, solid-calculix)   |    1    |       71.39       |     104.51      |         15.22         |
+-------------------------------------------------------+---------+-------------------+-----------------+-----------------------+