[SSSD] [PATCH] SSSD: Add a new command diag_cmd

Jakub Hrozek jhrozek at redhat.com
Wed Nov 11 08:32:37 UTC 2015


On Wed, Nov 11, 2015 at 09:12:52AM +0100, Petr Cech wrote:
> On 11/10/2015 04:20 PM, Jakub Hrozek wrote:
> >On Tue, Nov 10, 2015 at 01:22:54PM +0100, Lukas Slebodnik wrote:
> >>>On (10/11/15 13:15), Jakub Hrozek wrote:
> >>>> >On Mon, Nov 09, 2015 at 11:32:30AM +0100, Petr Cech wrote:
> >>>>> >>On 11/04/2015 11:24 AM, Jakub Hrozek wrote:
> >>>>>> >> >Hi,
> >>>>>> >> >
> >>>>>> >> >I created this patch to try to diagnose an issue where sssd would
> >>>>>> >> >randomly restart on any of machines in a VM cluster without giving too
> >>>>>> >> >much advise why. I think it might be useful to merge in general.
> >>>>> >>
> >>>>> >>Hi Jakub,
> >>>>> >>
> >>>>> >>I reviewed the patch. Code looks good to me.
> >>>>> >>CI tests passed:http://sssd-ci.duckdns.org/logs/job/32/25/summary.html
> >>>>> >>
> >>>>> >>Then I tried to test new functionality.
> >>>>> >>
> >>>>> >>Man pages are right, I found diag_cmd in sssd.conf.
> >>>>> >>
> >>>>> >>And I really got the right message when I kill sss_pam:
> >>>>> >># (Mon Nov  9 04:30:47 2015) [sssd] [svc_child_info] (0x0040): Child [25767]
> >>>>> >>terminated with signal [9]
> >>>>> >>
> >>>>> >>I would like to see output of pstack, but I don't know, how to get the right
> >>>>> >>state of SSSD. Can you help me, please?
> >>>> >
> >>>> >I tested the patch by setting a low 'timeout' in the 'domain' section
> >>>> >and then setting the diag_cmd:
> >>>> >    [domain/foo]
> >>>> >    timeout = 2
> >>>> >    diag_cmd = pstack %p
> >>>> >
> >>>> >then I stopped the back end:
> >>>> >    # kill -STOP $(pidof sssd_be)
> >>>> >
> >>>> >You should see the pstack output in /var/log/sssd/sssd.log, also the
> >>>> >debug_level must be increased in the [sssd] section. You might also need
> >>>> >to set SELinux to Permissive, otherwise sssd might not be able to fork
> >>>> >an exec pstack..
> >>>So in this case I would prefer if this opton was not documented.
> >>>or it should be documented issues with SELinux
> >>>
> Hi Jakub,
> 
> I just sent the patch to the CI tests and they passed
> http://sssd-ci.duckdns.org/logs/job/32/63/summary.html
> 
> >Then I would prefer undocumented. It matches how we (don't) document the
> >"command" option.
> I agree with little exception. I think it could be more clear if we write
> little documentation to commit message or to the code near the new option.
> But, how everybody wrote, we could leave man page without documentation.

Good idea, I can move the documentation that was previously in the man
page to the commit message, would that work for you?


More information about the sssd-devel mailing list