On (18/03/16 17:53), Lukas Slebodnik wrote:
>On (17/03/16 23:17), Jakub Hrozek wrote:
>>On Mon, Mar 07, 2016 at 04:12:00PM +0100, Lukas Slebodnik wrote:
>>> ehlo,
>>>
>>> simple patch is attached.
>>>
>>> LS
>>
>>Garbage-collecting easy patches untouched for a week or more..
>>
>>The patch looks good and according to:
>>
http://sssd-ci.duckdns.org/logs/job/39/61/rhel6/ci-build-debug/ci-make-ch...
>It should not be related to the patch.
>
I look deeply into the issue and it isn't realted to my patch.
It's caused by slow down caused by valgrind.
[==========] Running 3 test(s).
[ RUN ] test_run_as_root_fg
[ OK ] test_run_as_root_fg
[ RUN ] test_run_as_sssd_fg
[ OK ] test_run_as_sssd_fg
[ RUN ] test_run_as_root_daemon
killing 7393
[ OK ] test_run_as_root_daemon
[==========] 3 test(s) run.
[ PASSED ] 3 test(s).
PASS: server-tests
As you can see test passed here. Because test_run_as_root_daemon
check only creation of pid file and does not wait for full initialisation
of server.
(Thu Mar 17 07:49:13:826379 2016) [test_run_as_root_daemon] [ldb] (0x0020): Unable to open
tdb 'server_tests/config.ldb'(Thu Mar 17 07:49:13:899494 2016)
[test_run_as_root_daemon] [ldb] (0x0020): Failed to connect to
'server_tests/config.ldb' with backend 'tdb': Unable to open tdb
'server_tests/config.ldb'(Thu Mar 17 07:49:13:901723 2016)
[test_run_as_root_daemon] [confdb_init] (0x0010): Unable to open config database
[server_tests/config.ldb]
(Thu Mar 17 07:49:13:905037 2016) [test_run_as_root_daemon] [server_setup] (0x0010): The
confdb initialization failed
0x5 != 0
/var/lib/jenkins/workspace/ci/label/rhel6/src/tests/cwrap/test_server.c:154: error:
Failure!
[ FAILED ] test_run_as_root_daemon
[==========] 3 test(s) run.
[ PASSED ] 2 test(s).
[ FAILED ] 1 test(s), listed below:
[ FAILED ] test_run_as_root_daemon
The failure is in child process due to race condition caused by slow execution.
The parent process in server-tests sent SIGTERM signal to child process.
You can see a message "killing 7393" few lines before.
But child process tried to initialize confdb even though it isn't necessary
here for this case.
It isn't problem without valgrind becuase child process is fast enough.
One solution can be to send SIGKILL to child process as part of cleanup
in test_run_as_root_daemon. But that's not related to this patch :-)