bug#29676: Guix test failure on tests/store.

[Roel Janssen <roel@gnu.org>]

Ludovic Courtès writes:

> Roel Janssen <roel@gnu.org> skribis:
>
>> Ludovic Courtès writes:
>>
>>> Hi Roel,
>>>
>>> ludo@gnu.org (Ludovic Courtès) skribis:
>>>
>>>> Roel Janssen <roel@gnu.org> skribis:
>>>
>>> [...]
>>>
>>>>>>> actual-error:
>>>>>>> + (srfi-34
>>>>>>> +   #<condition &nix-protocol-error [message: "path `dtmp/guix-tests/store/462z3fnl7bs44vp9s97jyg1z74nsfvly-tar' is not in the Nix store" status: 1] 1071090>)
>>>>>>> result: FAIL
>>>>>>
>>>>>> The “dtmp” bit (instead of “/tmp”) looks fishy and would suggest a
>>>>>> memory corruption, presumably in the daemon.
>>>>>
>>>>> The daemon used in the test, or the daemon used to do the package build?
>>>>
>>>> The daemon under test (within the build environment).
>>>>
>>>>>> I suppose the failure random, isn’t it?
>>>>>
>>>>> I ran it again, and I've got the same error:
>>>>>
>>>>> actual-error:
>>>>> + (srfi-34
>>>>> +   #<condition &nix-protocol-error [message: "path `dtmp/guix-tests/store/462z3fnl7bs44vp9s97jyg1z74nsfvly-tar' is not in the Nix store" status: 1] 1754ed0>)
>>>>>
>>>>> So that's either a very funny coincidence, or it's a structural problem.
>>>>
>>>> It’s better if it’s not random.  :-)
>>>>
>>>>> Do you have any suggestions for how I can debug this problem?
>>>>
>>>> Assuming the failure also happens when you run “make check” outside the
>>>> build container (in the failed build tree), can you add ‘valgrind’ in
>>>> ‘test-env’?  Specifically, in ‘test-env’, look for the line that invokes
>>>> ‘./pre-inst-env guix-daemon’ and change it to ‘./pre-inst-env valgrind
>>>> guix-daemon’.
>>>
>>> Did you manage to gather more info?
>>
>> No.  When I run "make check -k" many tests fail.
>
> In a fresh checkout?  If many tests fail, then there’s probably
> something wrong with the environment, such as guix-daemon failing to
> start or something.  Could you check the logs?
>
> Thanks,
> Ludo’.

Indeed.  I found out that guix-daemon couldn't start because I didn't
set --localstatedir properly.  I "make check" again and that yields the
same issue with tests/store.scm.

However, upon changing test-env to include valgrind, more tests fail
again because of a problem accessing the daemon-socket.

So I manually started the guix-daemon with --disable-chroot and
prepended valgrind --leak-check=full to it.  Then I ran:
$ guile -s tests/store.scm

Which yielding many more failures.
Then:
$ make check

Which also yielded more test failures.  But at least I have a valgrind
log now, which I uploaded here:
https://www.roelj.com/guix-daemon-valgrind.log.gz

I don't think the memory problem is in the log, but at least it shows
that we should have a look at the memory allocation of guix-daemon.
Sometimes it seems to leak memory.

Kind regards,
Roel Janssen