s3 website ignore --error-document argument

[quiver]

When both --index-document and --error-document options are supplied,
--error-document argument was ignored.

How to reproduce

$ aws s3 mb s3://b0e1.example.com
make_bucket: s3://b0e1.example.com/
$ aws s3 website s3://b0e1.example.com --index-document  test.html --error-document error.html
$ aws s3api get-bucket-website --bucket b0e1.example.com
{
    "RedirectAllRequestsTo": {},
    "IndexDocument": {
        "Suffix": "test.html"
    },
    "ErrorDocument": {}, # <- ignored!
    "RoutingRules": []
}

[jamesls]

I'm seeing an error on this line. Do you see this as well?

======================================================================
ERROR: test_create_website_configuration (tests.integration.customizations.s3.test_plugin.TestWebsiteConfiguration)
----------------------------------------------------------------------
Traceback (most recent call last):
  File "aws-cli/tests/integration/customizations/s3/test_plugin.py", line 746, in test_create_website_configuration
    self.assertEqual(parsed['ErrorDocument']['Key'], 'error.html')
KeyError: 'Key'
-------------------- >> begin captured logging << --------------------

[quiver]

I got the same error in a newly created environment and I think I found the culprit.

The problem stems from how integration test is executed.

As you can see in tests/integration/__init__.py line #21-#24, integration tests are executed by running CWD/bin/aws command, and when they are run, module awscli is imported not from CWD/awscli but from system default module search path.

That's why new test case(added to demonstrate the current problem) fails.

Following hackish code forces integration tests to be executed using CWD/awscli module.

$ diff --git a/awscli/clidriver.py b/awscli/clidriver.py
index 6a7da73..3e20f96 100644
--- a/awscli/clidriver.py
+++ b/awscli/clidriver.py
@@ -535,3 +535,6 @@ class CLIOperationCaller(object):
             output = self._session.get_variable('output')
         formatter = get_formatter(output, args)
         formatter(operation, response)
+
+if __name__ == '__main__':
+    main()
diff --git a/tests/integration/__init__.py b/tests/integration/__init__.py
index fdde492..d8cca93 100644
--- a/tests/integration/__init__.py
+++ b/tests/integration/__init__.py
@@ -82,6 +82,7 @@ def aws(command, collect_memory=False, env_vars=None,
         aws_command = os.environ['AWS_TEST_COMMAND']
     else:
         aws_command = 'python %s' % AWS_CMD
+        aws_command = 'python -m awscli.clidriver'
     full_command = '%s %s' % (aws_command, command)
     stdout_encoding = _get_stdout_encoding()
     if isinstance(full_command, six.text_type) and not six.PY3:

With this patch, I can run tests without error:

$ nosetests tests/integration/customizations/s3/test_plugin.py:TestWebsiteConfiguration
.
----------------------------------------------------------------------
Ran 1 test in 4.730s

OK

[jamesls]

Upon further debugging it looks like this is because we're using the same bucket for this test, and the second command that adds the error document takes a while to propogate. If I had a time.sleep(2) between the first and second command, the test always passes for me.

Another option is to just split this into two separate tests.

[quiver]

I had a time.sleep(2) between the first and second command, the test always passes for me.

It didn't work for me.

Another option is to just split this into two separate tests.

Deleting the original --index-document only test part or separating into two tests didn't work for me.

For me, the root cause seems to be integration tests' module search path.

[quiver]

Sorry, my fault. Even solving search path oddity, I still got key error.

I had a time.sleep(2) between the first and second command, the test always passes for me.
Another option is to just split this into two separate tests.

Both options fixed key error. It looks like reusing botocore's session is causing trouble. Following @jamesls advice, I modified the original test to two separate tests.

[jamesls]

Thanks!