QA (public) » openQA Project (public)

probably unsolvable with Inline::Python ...

see https://github.com/niner/inline-python-pm/blob/8a2ec302d5e83cb84167da418432f26d37e1a403/Python.xs#L215-L220

on another side, it looks like Inline::Pythonon isn't completely dead ( some commit in last two years)

mkittler wrote:

Without any modifications to os-autoinst I'm actually getting a useful error message:

[2021-09-09T17:41:52.047 CEST] [debug] ||| starting search tests/openQA/search.py
[2021-09-09T17:41:52.149 CEST] [info] ::: basetest::runtest: # Test died: FileNotFoundError: [Errno 2] No such file or directory: 'foobar' at line 

The message is also recorded as step detail.

I've been provoking the error by putting an open("foobar") in the Python code. Apparently that's not enough to reproduce the issue. What kind of error was it in your case which only lead to the generic error message?

It was missing Python dependencies which resulted in this job which has the reason Reason: tests died: unable to load /tests/openQA/search.py, check the log for the cause (e.g. syntax error) but no specifics in the logs - that's what the ticket description shows.

Then I'm getting the reason tests died: unable to load tests/openQA/search.py, check the log for the cause (e.g. syntax error) and the log actually contains a Python traceback and error message:

[2021-09-10T15:32:08.328 CEST] [debug] scheduling login /tests/openQA/login.pm
Traceback (most recent call last):
  File "<string>", line 3, in <module>
ModuleNotFoundError: No module named 'foobar'
[2021-09-10T15:32:08.423 CEST] [warn] !!! autotest::loadtest: error on /tests/openQA/search.py: Error -- py_eval raised an exception at /usr/lib/perl5/vendor_perl/5.32.1/x86_64-linux-thread-multi/Inline/Python.pm line 177.
  BEGIN failed--compilation aborted at (eval 146) line 5.

error on /tests/openQA/search.py: Error -- py_eval raised an exception at /usr/lib/perl5/vendor_perl/5.32.1/x86_64-linux-thread-multi/Inline/Python.pm line 177.
BEGIN failed--compilation aborted at (eval 146) line 5.
Compilation failed in require at /hdd/openqa-devel/repos/os-autoinst/OpenQA/Isotovideo/Utils.pm line 211.

The error message is not in line with our usual logging but I find it quite acceptable.

I'm wondering what I do different than you. I'm also testing with the openQA-in-openQA test using the very same test module you've been working on. In case you were using Python 2 I don't see how it would make a difference (https://docs.python.org/2.7/c-api/exceptions.html#c.PyErr_Print).

Oh, and hacks like sys.stderr = open('autoinst-log.txt', 'a') shouldn't really be required because autoinst-log.txt is actually the file sys.stderr ends up in.

https://openqa.opensuse.org/tests/1922451

In production the traceback output is missing, indeed:

[2021-09-17T18:47:10.792 CEST] [debug] scheduling login /tests/openQA/login.pm
[2021-09-17T18:47:10.902 CEST] [warn] !!! autotest::loadtest: error on /tests/openQA/search.py: Error -- py_eval raised an exception at /usr/lib/perl5/vendor_perl/5.26.1/x86_64-linux-thread-multi/Inline/Python.pm line 177.
  BEGIN failed--compilation aborted at (eval 130) line 5.

error on /tests/openQA/search.py: Error -- py_eval raised an exception at /usr/lib/perl5/vendor_perl/5.26.1/x86_64-linux-thread-multi/Inline/Python.pm line 177.
BEGIN failed--compilation aborted at (eval 130) line 5.
Compilation failed in require at /usr/lib/os-autoinst/OpenQA/Isotovideo/Utils.pm line 209.

On the production worker the same version of perl-Inline-Python is installed as on TW (via https://build.opensuse.org/package/show/devel:openQA:Leap:15.2/perl-Inline-Python) so this shouldn't make a difference. Maybe the Perl/Python version makes a difference. On TW /usr/lib/perl5/vendor_perl/5.32.1/x86_64-linux-thread-multi/auto/Inline/Python/Python.so links against libpython3.8.so.1.0 and on Leap 15.2 /usr/lib/perl5/vendor_perl/5.26.1/x86_64-linux-thread-multi/auto/Inline/Python/Python.so links against libpython3.6m.so.1.0. However, I doubt that Python's PyErr_Print function has changed a lot.

The following works just fine on the worker host as well:

martchus@openqaworker7:~> perl -e "use Inline Python => 'import foobar';"
Traceback (most recent call last):
  File "<string>", line 1, in <module>
ModuleNotFoundError: No module named 'foobar'
Error -- py_eval raised an exception at /usr/lib/perl5/vendor_perl/5.26.1/x86_64-linux-thread-multi/Inline/Python.pm line 177.
BEGIN failed--compilation aborted at -e line 1.

So there must be something within our code which suppresses/redirects Python's traceback somehow in some environments.

I can reproduce the problem locally when I use the container setup which is using Leap 15.2. However, I still don't know what makes the difference. When I test this outside of isotovideo but in the same way as autotest.pm uses Inline Python the error message is even shown within the container:

4068c8b89349:/ # perl -I/usr/lib/os-autoinst -e "use base 'basetest'; use Mojo::File qw(path); use Inline Python => 'import sys; sys.path.append(\"/usr/lib/os-autoinst\")'; use Inline Python => path('/var/lib/openqa/share/tests/openqa/tests/openQA/search.py')->slurp;"
Traceback (most recent call last):
  File "<string>", line 3, in <module>
ModuleNotFoundError: No module named 'foobar'
Error -- py_eval raised an exception at /usr/lib/perl5/vendor_perl/5.26.1/x86_64-linux-thread-multi/Inline/Python.pm line 177.
BEGIN failed--compilation aborted at -e line 1.

can you try https://blog.famzah.net/2010/04/08/auto-flush-both-stdout-and-stderr-in-perl/ ?

If this does not work please update the ticket description with clear steps to reproduce, e.g. based on the container based command lines you gave, plus a workaround, unassign, move the ticket to "future" with low prio.

PR which should fix the issue: https://github.com/os-autoinst/os-autoinst/pull/1781

The PR has been merged. I'll trigger a job again on o3 after the deployment.

https://openqa.opensuse.org/tests/1940977

even in color :)