Thread: [pgadmin4][patch] Use pytest test runner for unit tests

Hey all,

Attached is a rather large patch that attempts to introduce the Pytest test runner and Grappah test matcher into our code base. The patch replaces all of the previous python unit tests from the previous design. This is a follow-up from our previous proof of concept and discussion around the idiosyncrasies of our existing test suite.

We are motivated to submit this change for the following reasons:

1. Having a highly customized test suite increases the learning curve towards contributing to the code base. Pytest is a mature stable python test framework that outside developers are more likely to be familiar with.

2. Pytest supports a lot of the features that we've built for our use-cases - out of the box, and with extensive documentation on the matter. 

3. The lack of ability to run tests individually has become a pain point. Especially when it comes to debugging long running feature tests.      

Test can be ran with the following command:

yarn test:unit

​

Please let us know what you think and if there are any issues!

Note: The feature tests are not yet completed. We expect our CI to fail as a result of this patch. We will complete this step soon in a follow-up patch! 

https://docs.pytest.org/en/latest/

https://github.com/grappa-py/grappa

Here’s a followup patch with the relevant README and Makefile changes. To be clear, both patches need to be applied in succession to run the tests.

The error that you were running into was because the appropriate PYTHONPATH environment variable was not set. We updated the README to reflect that, but haven’t done anything to the code for that

As part of the development environment we do not see the reasoning behind not add PYTHONPATH to the environment variables, specially because this looks like the way pytest was invisoned.

However please try the following patch instead. We've changed the pytest invocation to assume the relevant dir as part of the directories to load, as well as the docs and Makefile

Hello Hackers,
We have good news, we just finished porting all the tests to pytest/grappa. Please see the attached diff with the multiple commits for this. Bare with us in this HUGE patch. We split it as follow:

0001 - Convert all the Unit Tests to pytest/grappa(Patch previously sent, with some tests that were not running previously
0002 - Change of README, makefile and package.json to use the new syntax
0003 - Convert all the Feature Tests to pytest/grappa

From now on, we can launch 1 test at a time if we want to by using pytest. (In this gist https://gist.github.com/kwmiebach/3fd49612ef7a52b5ce3a there are some examples on how to use it) Because we are using the pytest CLI all the options are made available to use, the community also as a ton of plugins if we ever need something else.

We updated all README’s with the following information as well.

As we discussed previously in order to use the pytest binary directly you need to point the PYTHONPATH to $PGADMIN_SRC/web
If you do not want to do that then you can use it like this web $ python -m pytest -q -s pgadmin

We also changed the feature test, they now are under regression/feature_tests as they do no longer need to be in the application path in order to run.
To address the password issue we added to the makefile

python -m pytest --tb=short -q --json=regression/test_result.json --log-file-level=DEBUG --log-file=regression/regression.log pgadmin

the --tb=short options is the one responsible for removing a chunk of the traceback, making it smaller and hiding the variable printing

Known issues:

• Python 2.7, the library we are using for assertions (Grappa) is failing while trying to assert on strings. We created a PR to the library: https://github.com/grappa-py/grappa/pull/43 as soon as this gets in all the tests should pass

• We found out that the tests in pgadmin/browser/tests where not running because they needed SERVER_MODE to be True. We ported these tests but 2 tests are still failing TestResetPassword.test_success and TestChangePassword.test_success. They are currently failing for 2 different reasons, but when we went back to master branch they were also failing. So we kept them with the mark xfail, this mark allow us to run the tests and do not report the failure. You can read more one the topic at (https://docs.pytest.org/en/3.6.0/skipping.html). Unfortunately we were not able to correct the issues, if someone could look into these test it would be great.

• Jenkins server need a change. Because we now run tests for a single database at a time the Jenkins flow need to change. Our proposal for this is to isolate each database in its own task something similar to the pipeline that we currently use internally:

​

The boxes that we are pointing with the teal arrow represent 1 Python Version Against 1 Database. This can be accomplished using Jenkins as well. In order to accomplish that we are going to generate e Jenkinsfile(https://jenkins.io/doc/book/pipeline/jenkinsfile/) to do this in a more DSL manner, making it easier for us to deploy the pipeline in a more consistent way. The LTS version of jenkins is 2.107.3 the version that we have in CI should be ok, but is always good to be in a LTS version of Jenkins or newer because the community is pretty active.
The idea would be to replace the 5 tasks that we have there and start using a pipeline that will spawn docker containers to isolate the testing between version/database. That is what we do with concourse as we depicted in the previous image.
Does anyone have any thoughts about this?

Known issues:

• Python 2.7, the library we are using for assertions (Grappa) is failing while trying to assert on strings. We created a PR to the library: https://github.com/grappa-py/grappa/pull/43 as soon as this gets in all the tests should pass

Any guesses as to the ETA? Given that most of our dev, and our Windows and Mac packages both run on 2.7 at the moment, it's clear that this is a  required fix before we can proceed.

Attached you can find the patch that bumps grappa to version 0.1.9 that support Python 2.7 

• Jenkins server need a change. Because we now run tests for a single database at a time the Jenkins flow need to change. Our proposal for this is to isolate each database in its own task something similar to the pipeline that we currently use internally:

​

The boxes that we are pointing with the teal arrow represent 1 Python Version Against 1 Database. This can be accomplished using Jenkins as well. In order to accomplish that we are going to generate e Jenkinsfile(https://jenkins.io/doc/book/pipeline/jenkinsfile/) to do this in a more DSL manner, making it easier for us to deploy the pipeline in a more consistent way. The LTS version of jenkins is 2.107.3 the version that we have in CI should be ok, but is always good to be in a LTS version of Jenkins or newer because the community is pretty active.
The idea would be to replace the 5 tasks that we have there and start using a pipeline that will spawn docker containers to isolate the testing between version/database. That is what we do with concourse as we depicted in the previous image.
Does anyone have any thoughts about this?

Well the current public Jenkins system is going away soon, and the new one has had a ton of jobs created that assume the tests run in series automatically against each database server (and is completely different from the current system). I do plan to change that in the longer term, but it requires a change to the way I've been using (and know how to use) Jenkins.

 Do you have a link to the new CI that you can share with us so we can take a pick. We can give some advises on what we believe it is a a good flow for the Continuous Delivery pipeline.

We have some examples from our pipeline that we can share:

Script that we use to run the UT + Feature tests on a docker image that has python+yarn+selenium+postgres installed on it:

https://github.com/greenplum-db/pgadmin4-ci/blob/master/tasks/run-postgres-tests/run.sh

This type of scripts can be added to the Jenkinsfile to create a pipeline step. A good practice in a reproducible pipeline is to use Docker to ensure that every test runs in a clean and predictable environment, this make it easy to reproduce a problem found in testing.

However, the bigger problem for me is that I often run the tests against multiple DB servers on my laptop, without Jenkins. How would that workflow look now?

 

We understand that and it looks like an interesting use case and it could be accomplished by creating a script of something similar that launch the tests multiple times. That is something that we can schedule as a future steps after this patch is reviewed and committed.

Known issues:

• Python 2.7, the library we are using for assertions (Grappa) is failing while trying to assert on strings. We created a PR to the library: https://github.com/grappa-py/grappa/pull/43 as soon as this gets in all the tests should pass

Any guesses as to the ETA? Given that most of our dev, and our Windows and Mac packages both run on 2.7 at the moment, it's clear that this is a  required fix before we can proceed.

Attached you can find the patch that bumps grappa to version 0.1.9 that support Python 2.7 

Hi Dave,

 

No, because it's firewalled to the nines inside our network. There's no chance I'm making production build machines internet-accessible.

For a Open Source project, if the community cannot see the place where the tests are running it looses a huge part of the process. We believe that removing this capability will have a negative impact on the development, especially because we do not have a CI/CD.  Before the code is merged we will never know if master is broken or not.

 

 

We have some examples from our pipeline that we can share:

Script that we use to run the UT + Feature tests on a docker image that has python+yarn+selenium+postgres installed on it:

https://github.com/greenplum-db/pgadmin4-ci/blob/master/tasks/run-postgres-tests/run.sh

This type of scripts can be added to the Jenkinsfile to create a pipeline step. A good practice in a reproducible pipeline is to use Docker to ensure that every test runs in a clean and predictable environment, this make it easy to reproduce a problem found in testing.

Docker is of little use to us here, as 2 of the 4 build platforms cannot be run in Docker (macOS and the Docker container), and the 3rd would be extremely difficult to run that way (Windows)

The docker files that we are talking about here is to run the tests, and we believe that the tests are all running in a Linux environment.

Hi Dave,

Attached are four different patches that should be applied in order.  Included is a fix for the failing test above.