Add CI for windows runner

[lalitpagaria]

Download archive from url without temp file along with adding CI for win & mac
To resolve #1445

Proposed changes:

• Sometime on windows python process doesn't have permission to create temp file hence extracting archive directly via in memory stream remove the need of temp file.
• Adding CI for Windows and Mac, but adding it in ci.yml is bit complex so will resolve issue step by step.

Status (please check what you already did):

• First draft (up for discussions & feedback)
• Final code
• Added tests
• Updated documentation

[lalitpagaria]

@oryx1729 @julian-risch Please review this PR.
Specially CI part. There are few changes required before making it operational for Win & Mac platform.

[lalitpagaria]

@tholor This PR is about to complete. As windows github env does not support Linux containers, we can't start Milvus, Weavite and GraphDB, Tika containers. Please suggest way to disable these test for windows platform (ie what command line argument need to pass?). Also not able to make crawler to work via web-driver.

[julian-risch]

@tholor This PR is about to complete. As windows github env does not support Linux containers, we can't start Milvus, Weavite and GraphDB, Tika containers. Please suggest way to disable these test for windows platform (ie what command line argument need to pass?). Also not able to make crawler to work via web-driver.

I think a clean way to exclude these tests would be to mark the tests or the individual parameter settings that don't work on windows. The pytest documentation tells me that "It is also possible to mark individual test instances within parametrize" https://docs.pytest.org/en/latest/parametrize.html Thereby, we can exclude them when running pytest.

For example,
@pytest.mark.skipif(condition, ..., *, reason=...): skip the given test function if any of the conditions evaluate to True. Example: skipif(sys.platform == 'win32') skips the test if we are on the win32 platform. See https://docs.pytest.org/en/stable/reference.html#pytest-mark-skipif
(taken from here: https://docs.pytest.org/en/latest/example/markers.html)

A slightly different option would be to mark all milvus tests (and parameter settings) and exclude them via something like pytest -m "not milvus"

[lalitpagaria]

Thanks @julian-risch
Yes first approach is clean but it is mainly due to windows GitHub runner doesn't support linux containers due to some licensing issue. Otherwise Milvus absolutely work correctly on my windows laptop. Also adding these annotations on each and every test will cause lot code changes on tests. Hence I would prefer second for this task. Please let me know if that is fine.

Also please let me know your view about creating separate windows-ci.yml instead of keeping it in same ci.yml.

[tholor]

For the document stores it should be sufficient to pass --document_store_type=faiss,memory,elasticsearch (see https://github.com/deepset-ai/haystack/blob/master/CONTRIBUTING.md#running-tests)

This leaves us with tika and graphdb - for both we already have markers so we should be able to exclude them via not tika and not graphdb. So in the end it should be something along these lines:

python -m pytest test_document_store.py --document_store_type=memory,faiss,elasticsearch -m "not tika and not graphdb"

[tholor]

Also please let me know your view about creating separate windows-ci.yml instead of keeping it in same ci.yml.

As we are not starting the same docker containers as in the linux CI, I think it would be simpler & cleaner to have it in a separate file. Do you see any downsides for it?

[lalitpagaria]

Also please let me know your view about creating separate windows-ci.yml instead of keeping it in same ci.yml.

As we are not starting the same docker containers as in the linux CI, I think it would be simpler & cleaner to have it in a separate file. Do you see any downsides for it?

Pros: It will be more clean
Cons:

• It will have separate file so if we add some changes to linux-ci then we would need to evaluate same for windows-ci

Also not sure to make tree based CI means first we will run type-check, then only we would like to run CI (linux and windows)

[lalitpagaria]

@tholor @julian-risch PR is ready for review. One query regarding when to run windows_ci?

As we are using one runner for each test file, most of the time our check will go to Queue mode. So I would suggest to run windows CI only on push to master instead on every PR. It will help in reduction in carbon footprint :)

[tholor]

So I would suggest to run windows CI only on push to master instead on every PR. It will help in reduction in carbon footprint :)

Yep, sounds good. Let's reduce carbon footprint a bit and only run for every push to master

[lalitpagaria]

@tholor Yes everything is complete in this PR. Please review it and merge if you are fine with the changes.

[tholor]

Thanks a lot! Looking good already! Left a few minor comments.

[lalitpagaria]

@tholor Addressed review comments.
This need you input for now CI is passing due to cache. Not sure how to fix properly.
#1458 (comment)

[tholor]

Looking good to me. Thanks a lot @lalitpagaria ! Will hopefully help us to spot more bugs for our windows friends