Develop Reference

How to use pytest-custom_exit_code plugin

Need help!
I have a job on Gitlab ci, that runs tests and reruns failed ones. If there are no failed tests, job fails with exit code 5, that means that there are no tests for running. I found out that there is plugin "pytest-custom_exit_code", but I don't know how to correctly use it.
I need just to add command 'pytest --suppress-no-test-exit-code' to my runner.sh?
It looks like this now:
#!/bin/sh
/usr/local/bin/pytest -m test
/usr/local/bin/pytest -m --last-failed --last-failed-no-failures none test

Assumption here is that plugin is installed first using
pip install pytest-custom_exit_code
command like option pytest --suppress-no-test-exit-code should work after that.
If configuration file like .pytest.ini is used , following lines should be added in it
[pytest]
addopts = --suppress-no-test-exit-code

Local GitLab runner freezes while Shared GitLab.com runner succeeds

EDIT: As Rekovni pointed out, using a GitLab runner with Docker on a Windows machine is a problem. Installing the runner in a Linux-based virtual machine solved the problem.
I am developing a Python program using a conda environment. It is hosted on GitLab.com and I am using GitLab-CI to generate the documentation.
I configured the following .gitlab-ci.yml file for it:
image: continuumio/miniconda3:latest
before_script:
# Update conda and create environment, which is then activated.
- conda update -vvv -y -c conda-forge conda
- conda env create -f helpers/NAME.yml
- source activate NAME
# Correct installation.
- conda install -q -y gsl=2.2.1
pages:
script:
# Install make.
- apt-get update
- apt-get install -q -y build-essential
# Install Spinx-related packages.
- conda install -q -y sphinx sphinx_rtd_theme
# Create documentation.
- cd REPO/doc
- sphinx-apidoc -o source/ ../REPO --force --separate
- make html
# Transfer documentation to public pages folder.
- mv build/html/ ../../public/
artifacts:
paths:
- public
# only:
# - master
Running this script with a shared GitLab runner that is supplied with GitLab.com works and the documentation is generated and placed in the public folder.
For future unit tests (which take much longer), I want to provide a local runner on a Win 10 machine in my network. For this, I installed the gitlab-runner.exe and Docker Desktop. I successfully registered the runner with the project on GitLab.com.
The runner is using the following config.toml configuration file:
concurrent = 1
check_interval = 0
log_level = "info"
[session_server]
session_timeout = 1800
[[runners]]
name = "NAME"
url = "https://gitlab.com"
token = "TOKEN"
executor = "docker"
[runners.custom_build_dir]
[runners.docker]
tls_verify = false
image = "alpine:latest"
privileged = false
disable_entrypoint_overwrite = false
oom_kill_disable = false
disable_cache = false
volumes = ["/cache"]
shm_size = 0
[runners.cache]
[runners.cache.s3]
[runners.cache.gcs]
The problem is now that the local runner freezes during the execution of the above script without producing any error messages and I am at a loss on how to debug it. What I have is
The log of the script that is shown on the Job page on GitLab.com; and
The console output of the gitlab-runner.exe on the local machine.
Regarding 1., I see
[0KRunning with gitlab-runner 11.10.0 (3001a600)
...
[32;1mChecking out COMMIT_HASH as BRANCH_NAME...[0;m
...
[0K[32;1m$ conda update -vvv -y -c conda-forge conda[0;m
DEBUG conda.gateways.logging:set_verbosity(148): verbosity set to 3
...
...
...
TRACE conda.gateways.disk.update:rename(52): renaming /opt/conda/share/doc/openssl/html/man3/OSSL_STORE_LOADER_new.html => /opt/conda/share/doc/openssl/html/man3/OSSL_STORE_LOADER_new.html.c~
TRACE conda.core.path_actions:execute(1041): renaming share/doc/openssl/html/man3/OSSL_STORE_LOADER_set_close.html => share/doc/openssl/html/man3/OSSL_STORE_LOADER_set_close.html.c~
TRACE conda.gateways.disk.update:rename(52): renaming /opt/conda/share/doc/openssl/html/man3/OSSL_STORE_LOADER_set_close.html => /opt/conda/share/doc/openssl/html/man3/OSSL_STORE_LOADER_set_close.html.c~
TRACE conda.core.path_actions:execute(1041): renaming share/doc/openssl/html/man3/OSSL_STORE_LOADER_set_ctrl.html => share/doc/openssl/html/man3/OSSL_STORE_LOADER_set_ctrl.html.c~
where it abruptly stops without reaching the - conda env create -f helpers/NAME.yml line.
Regarding 2., I see
C:\GitLab-Runner>gitlab-runner.exe --debug run
Runtime platform arch=amd64 os=windows pid=14116 revision=3001a600 version=11.10.0Starting multi-runner from C:\GitLab-Runner\config.toml ... builds=0
Checking runtime mode GOOS=windows uid=-1
Configuration loaded builds=0
...
Feeding runners to channel builds=0
Checking for jobs... nothing runner=TOKEN
Feeding runners to channel builds=0
Checking for jobs... received job=203033130 repo_url=REPO_URL.git runner=TOKEN
...
Attaching to container HASH ... job=203033130 project=6249897 runner=TOKEN
Starting container HASH ... job=203033130 project=6249897 runner=TOKEN
Waiting for attach to finish HASH ... job=203033130 project=6249897 runner=TOKEN
Waiting for container HASH ... job=203033130 project=6249897 runner=TOKEN
Appending trace to coordinator... ok code=202 job=203033130 job-log=0-10348 job-status=running runner=TOKEN sent-log=1801-10347 status=202 Accepted
Appending trace to coordinator... ok code=202 job=203033130 job-log=0-19445 job-status=running runner=TOKEN sent-log=10348-19444 status=202 Accepted
...
Appending trace to coordinator... ok code=202 job=203033130 job-log=0-933150 job-status=running runner=TOKEN sent-log=241860-933149 status=202 Accepted
Submitting job to coordinator... ok code=200 job=203033130 job-status= runner=TOKEN
Submitting job to coordinator... ok code=200 job=203033130 job-status= runner=TOKEN
where it seems that the switch from Appending trace to coordinator to Submitting job to coordinator happens around the time when it gets stuck.
After this, 1. is not updated with any further information and 2. is stuck in a Submitting job to coordinator loop.
Does anyone know:
What the reason for the failure with a local runner could be (when the same script works with a shared runner)?
What I could do to debug this problem?
Thanks and all the best,
Thomas

GitLab CI doesn't currently offer a solution for using its runner with Docker on a Windows environment, however there is an epic at the moment which is tracking progress for this.
In one of the issues of the epic, a contributer has managed to get a working version of a gitlab-runner which uses Docker for Windows, with which more details can be found here.
A more common (and potentially easier) way of using Docker in a Windows environment, would be to install the gitlab-runner as a Shell runner, and call the Docker commands manually to run your tests.
Conversely, if you just want to keep using the same CI script, you could install a Linux VM on your Windows 10 machine, and have that host the docker runner!

Conditional Commands in tox? (tox, travis-ci, and coveralls)

tl;dr:
I'm setting up CI for a project of mine, hosted on github, using tox and travis-ci. At the end of the build, I run converalls to push the coverage reports to coveralls.io. I would like to make this command 'conditional' - for execution only when the tests are run on travis; not when they are run on my local machine. Is there a way to make this happen?
The details:
The package I'm trying to test is a python package. I'm using / planning to use the following 'infrastructure' to set up the tests :
The tests themselves are of the py.test variety.
The CI scripting, so to speak, is from tox. This lets me run the tests locally, which is rather important to me. I don't want to have to push to github every time I need a test run. I also use numpy and matplotlib in my package, so running an inane number of test cycles on travis-ci seems overly wasteful to me. As such, ditching tox and simply using .travis.yml alone is not an option.
The CI server is travis-ci
The relevant test scripts look something like this :
.travis.yml
language: python
python: 2.7
env:
- TOX_ENV=py27
install:
- pip install tox
script:
- tox -e $TOX_ENV
tox.ini
[tox]
envlist = py27
[testenv]
passenv = TRAVIS TRAVIS_JOB_ID TRAVIS_BRANCH
deps =
pytest
coverage
pytest-cov
coveralls
commands =
py.test --cov={envsitepackagesdir}/mypackage --cov-report=term --basetemp={envtmpdir}
coveralls
This file lets me run the tests locally. However, due to the final coveralls call, the test fails in principle, with :
py27 runtests: commands[1] | coveralls
You have to provide either repo_token in .coveralls.yml, or launch via Travis
ERROR: InvocationError: ...coveralls'
This is an expected error. The passenv bit sends along the necessary information from travis to be able to write to coveralls, and without travis there to provide this information, the command should fail. I don't want this to push the results to coveralls.io, either. I'd like to have coveralls run only if the test is occuring on travis-ci. Is there any way in which I can have this command run conditionally, or set up a build configuration which achieves the same effect?
I've already tried moving the coveralls portion into .travis.yml, but when that is executed coveralls seems to be unable to locate the appropriate .coverage file to send over. I made various attempts in this direction, none of which resulted in a successful submission to coveralls.io except the combination listed above. The following was what I would have hoped would work, given that when I run tox locally I do end up with a .coverage file where I'd expect it - in the root folder of my source tree.
No submission to coveralls.io
language: python
python: 2.7
env:
- TOX_ENV=py27
install:
- pip install tox
- pip install python-coveralls
script:
- tox -e $TOX_ENV
after_success:
- coveralls

An alternative solution would be to prefix the coveralls command with a dash (-) to tell tox to ignore its exit code as explained in the documentation. This way even failures from coveralls will be ignored and tox will consider the test execution as successful when executed locally.
Using the example configuration above, it would be as follows:
[tox]
envlist = py27
[testenv]
passenv = TRAVIS TRAVIS_JOB_ID TRAVIS_BRANCH
deps =
pytest
coverage
pytest-cov
coveralls
commands =
py.test --cov={envsitepackagesdir}/mypackage --cov-report=term --basetemp={envtmpdir}
- coveralls

I have a similar setup with Travis, tox and coveralls. My idea was to only execute coveralls if the TRAVIS environment variable is set. However, it seems this is not so easy to do as tox has trouble parsing commands with quotes and ampersands. Additionally, this confused Travis me a lot.
Eventually I wrote a simple python script run_coveralls.py:
#!/bin/env/python
import os
from subprocess import call
if __name__ == '__main__':
if 'TRAVIS' in os.environ:
rc = call('coveralls')
raise SystemExit(rc)
In tox.ini, replace your coveralls command with python {toxinidir}/run_coveralls.py.

I am using a environmental variable to run additional commands.
tox.ini
commands =
coverage run runtests.py
{env:POST_COMMAND:python --version}
.travis.yml
language: python
python:
- "3.6"
install: pip install tox-travis
script: tox
env:
- POST_COMMAND=codecov -e TOX_ENV
Now in my local setup, it print the python version. When run from Travis it runs codecov.

Alternative solution if you use a Makefile and dont want a new py file:
define COVERALL_PYSCRIPT
import os
from subprocess import call
if __name__ == '__main__':
if 'TRAVIS' in os.environ:
rc = call('coveralls')
raise SystemExit(rc)
print("Not in Travis CI, skipping coveralls")
endef
export COVERALL_PYSCRIPT
coveralls: ## runs coveralls if TRAVIS in env
#python -c "$$COVERALL_PYSCRIPT"
In tox.ini add make coveralls to commands

Best practices for debugging vagrant+docker+flask

My goal is to run a flask webserver from a Docker container. Working on a Windows machine this requires Vagrant for creating a VM. Running vagrant up --provider=docker leads to the following complaint:
INFO interface: error: The container started either never left the "stopped" state or
very quickly reverted to the "stopped" state. This is usually
because the container didn't execute a command that kept it running,
and usually indicates a misconfiguration.
If you meant for this container to not remain running, please
set the Docker provider configuration "remains_running" to "false":
config.vm.provider "docker" do |d|
d.remains_running = false
end
This is my Dockerfile
FROM mrmrcoleman/python_webapp
EXPOSE 5000
# Install Python
RUN apt-get install -y python python-dev python-distribute python-pip
# Add and install Python modules
RUN pip install Flask
#copy the working directory to the container
ADD . /
CMD python run.py
And this is the Vagrantfile
Vagrant.configure("2") do |config|
config.vm.provider "docker" do |d|
d.build_dir = "." #searches for a local dockerfile
end
config.vm.synced_folder ".", "/vagrant", type: "rsync"
rsync__chown = false
end
Because the Vagrantfile and run.py work without trouble independently, I suspect I made a mistake in the Dockerfile. My question is twofold:
Is there something clearly wrong with the Dockerfile or the
Vagrantfile?
Is there a way to have vagrant/docker produce more
specific error messages?

I think the answer I was looking for is using the command
vagrant docker-logs
I broke the Dockerfile because I did not recognize good behaviour as such, because nothing really happens if the app runs as it should. docker-logs confirms that the flask app is listening for requests.

Is there something clearly wrong with the Dockerfile or the Vagrantfile?
Your Dockerfile and Vagrantfiles look good, but I think you need to modify the permissions of run.py to be executable:
...
#copy the working directory to the container
ADD . /
RUN chmod +x run.py
CMD python run.py
Does that work?
Is there a way to have vagrant/docker produce more specific error messages?
Try taking a look at the vagrant debugging page. Another approach I use is to log into the container and try running the script manually.
# log onto the vm running docker
vagrant ssh
# start your container in bash, assuming its already built.
docker run -it my/container /bin/bash
# now from inside your container try to start your app
python run.py
Also, if you want to view your app locally, you'll want to add port forwarding to your Vagrantfile.

How to convince python tox to run tests only for the available python interpreters?

I am using python tox to run python unittest for several versions of python, but these python interpreters are not all available on all machines or platforms where I'm running tox.
How can I configure tox so it will run tests only when python interpretors are available.
Example of tox.ini:
[tox]
envlist=py25,py27
[testenv]
...
[testenv:py25]
...
The big problem is that I do want to have a list of python environments which is auto-detected.

As of Tox version 1.7.2, you can pass the --skip-missing-interpreters flag to achieve this behavior. You can also set skip_missing_interpreters=true in your tox.ini file. More info here.
[tox]
envlist =
py24, py25, py26, py27, py30, py31, py32, py33, py34, jython, pypy, pypy3
skip_missing_interpreters =
true

First if you don't have tox : pip install tox.
Use this command : tox --skip-missing-interpreters , it skips for the compilers which are not available locally and just runs for the available versions of python

tox will display an Error if an interpreter cannot be found. Question is up if there should be a "SKIPPED" state and making tox return a "0" success result. This should probably be explicitely enabled via a command line option. If you agree, file an issue at http://bitbucket.org/hpk42/tox .