Random(?) test_crawler.py failure on Travis

[elacuesta]

Every now and then the Travis build fails on tests.test_crawler.CrawlerRunnerTestCase.test_deprecated_attribute_spiders, I'm not sure why.
Some examples:

• https://travis-ci.org/scrapy/scrapy/jobs/582184413#L407
• https://travis-ci.org/scrapy/scrapy/jobs/580207692#L419
• https://travis-ci.org/scrapy/scrapy/jobs/584701732#L415
• https://travis-ci.org/scrapy/scrapy/jobs/584762868#L418

PS: Sorry, I feel bad for ignoring the bug report template, but I felt like it didn't apply in this case.

[s-sanjay]

for some reason github search did not give me any result when I searched test_deprecated_attribute_spiders so I opened the other issue. hoping this comment gets indexed by github

[elacuesta]

@s-sanjay That's ok, I only included the full path in the description after seeing your issue. Thanks!

[elacuesta]

A different error on tests.test_spider.CrawlSpiderTest.test_process_request_instance_method: https://travis-ci.org/scrapy/scrapy/jobs/593353901#L414

[s-sanjay]

let me try to run in the same environment using the same HASHSEED

[s-sanjay]

the flakiness is happening due to this

{message : ResourceWarning('unclosed file <_io.BufferedReader name=30>',), category : 'ResourceWarning', filename : '/Users/sanjay/nobackup/scrapy/.tox/py36/lib/python3.6/copy.py', lineno : 239, line : None}
{message : ResourceWarning('unclosed file <_io.BufferedReader name=25>',), category : 'ResourceWarning', filename : '/Users/sanjay/nobackup/scrapy/.tox/py36/lib/python3.6/copy.py', lineno : 239, line : None}
{message : ScrapyDeprecationWarning('CrawlerRunner.spiders attribute is renamed to CrawlerRunner.spider_loader.',), category : 'ScrapyDeprecationWarning', filename : '/Users/sanjay/nobackup/scrapy/tests/test_crawler.py', lineno : 172, line : None}

These warnings ResourceWarning('unclosed file <_io.BufferedReader name=30> are the reason the tests are failing. Every run, the filename is changed and I still have to figure out who is opening these files and everytime it blames deepcopy module.

[Gallaecio]

Should we use https://pypi.org/project/flaky/ to have these tests re-run a few times if they fail? (discovered through scrapinghub/frontera#379)

[s-sanjay]

@Gallaecio so when tests fail due to some other issue re running them is great but shouldn't we as much as possible prevent flakiness from being introduced.. as a last resort we can re run them but if we re run would it not also delay fixing the reasons for flakiness ?

[Gallaecio]

@s-sanjay I was suggesting that as a temporary measure, to avoid new contributors from getting hit by this and us having to explain to them that the issue is in Scrapy and not in their changes.

In any case, #4095 should have fixed the actual issue.