Broken CI

[yajo]

This repo is broken.

The tests exhibit some kind of race condition that sometimes makes builds fail.

You can see the failures in https://travis-ci.org/OCA/maintainer-quality-tools/builds/637822768, which are unrelated to the PR.

I just rebuilt one job in the master branch and it failed with the same failure: https://travis-ci.org/OCA/maintainer-quality-tools/jobs/623695176

The failure seems to come from around the combination of these code chunks, and has been probably introduced by @hugo-ad in #434:

maintainer-quality-tools/travis/self_tests

Lines 255 to 257 in f565184

	# On TravisCI tests, force stopping the alive server
	odoo_pid = open('/tmp/odoo.pid').read()
	subprocess.call(['kill', '-9', odoo_pid])

maintainer-quality-tools/travis/test_server.py

Lines 409 to 454 in ae6dfd2

	for command, check_loaded in commands:
	if db_odoo_created and instance_alive:
	# If exists database of odoo test
	# then start server with regular command without tests params
	rm_items = [
	'coverage', 'run', '--stop-after-init',
	'--test-enable', '--init', None,
	'--log-handler', 'openerp.tools.yaml_import:DEBUG',
	]
	command_call = [item
	for item in commands[0][0]
	if item not in rm_items] + \
	['--pidfile=/tmp/odoo.pid']
	else:
	command[-1] = to_test
	# Run test command; unbuffer keeps output colors
	command_call = (["unbuffer"] if unbuffer else []) + command
	print(" ".join(cmd_strip_secret(command_call)))
	pipe = subprocess.Popen(command_call,
	stderr=subprocess.STDOUT,
	stdout=subprocess.PIPE)
	with open('stdout.log', 'wb') as stdout:
	for line in iter(pipe.stdout.readline, b''):
	stdout.write(line)
	print(line.strip().decode(
	'UTF-8', errors='backslashreplace'
	))
	returncode = pipe.wait()
	# Find errors, except from failed mails
	errors = has_test_errors(
	"stdout.log", database, odoo_version, check_loaded)
	if returncode != 0:
	all_errors.append(to_test)
	print(fail_msg, "Command exited with code %s" % returncode)
	# If there are no errors,
	# adds an error when returcode!=0
	# because it's actually an error.
	if not errors:
	errors += 1
	if errors:
	counted_errors += errors
	all_errors.append(to_test)
	print(fail_msg, "Found %d lines with errors" % errors)
	if not instance_alive and odoo_unittest:
	# Don't drop the database if will be used later.
	subprocess.call(["dropdb", database])

I have no idea on how to fix it, this is all the info I could get 😕

[pedrobaeza]

@moylop260 you are one of the masters of this repo. Can you help?

[moylop260]

I @yajo
Thank for your issue

FYI I see the error here:

• https://travis-ci.org/OCA/maintainer-quality-tools/jobs/623695176#L756

I have created the following PR in order to verify it:

• [REF] install: pin coveralls version and supporting broken_uninstallable module #627

[yajo]

Just in case it helps, let me explain how I got to the conclusion I said in #626 (comment).

AFAICS, comparing the ❌ job with the latest ✔️ one, the main difference is here:

Job that failed:

test_modules_to_preinstall (__main__.MainTest) ... ok
test_pylint_check (__main__.MainTest)
Testing empty paths and pylint_run fix of: ... skipped 'Set LINT_CHECK'
test_travis_helpers (__main__.MainTest) ... ----------
ok

Job that succeeded:

test_modules_to_preinstall (__main__.MainTest) ... ok
----------
test_pylint_check (__main__.MainTest)
Testing empty paths and pylint_run fix of: ... skipped 'Set LINT_CHECK'
test_travis_helpers (__main__.MainTest) ... ('\x1b[1;31mFAIL\x1b[0;m', 'Command exited with code -9')
('\x1b[1;31mFAIL\x1b[0;m', 'Found 1 lines with errors')
Module test summary
('\x1b[1;31mFAIL\x1b[0;m', 'broken_module')
Expected 2 errors, found 1!
ok

I have searched and there seems to be nothing that exits with -9, but the closest one seems to be:

maintainer-quality-tools/travis/self_tests

Lines 255 to 257 in f565184

	# On TravisCI tests, force stopping the alive server
	odoo_pid = open('/tmp/odoo.pid').read()
	subprocess.call(['kill', '-9', odoo_pid])

It's a weird way to kill the server. It would have made more sense to start it and stop it in the same test, or in the setup/teardown methods. Also, it seems like the server is expecting 1 more error than what it finds. That's why I said it seems a race condition (after all, nothing has changed recently in those parts of the code).

I hope it helps.

[moylop260]

Fixed #627