[Pkg-samba-maint] Bug#563844: samba restart fails with large smbd

[Ross Boylan]

On Tue, 2010-01-05 at 14:03 -0800, Steve Langasek wrote:
> On Tue, Jan 05, 2010 at 11:20:11AM -0800, Ross Boylan wrote:
> > invoke-rc.d samba restart
> 
> invoke-rc.d is not an admin interface.  In lenny, the supported interface is
> '/etc/init.d/samba restart'; in squeeze, it's 'service samba restart'.
What is the significance of it being an admin interface?  Given my
runlevel, I think invoke-rc.d samba and /etc/init.d.samba are the same
thing.
> 
> > results in no running smbd process.
> 
> > This arises when I have an smbd process that has grown outlandishly
> > (to 1G in the most recent case).  After I execute the restart command
> > the terminal shows the usual stop and start messages for smbd and
> > nmbd.  However, ps still shows the old process running.  After awhile,
> > that process dies, leaving me with no running smbd.
> 
> > The "real" problem is the huge footprint of smbd, reported in Debian
> > bug 558453 (my report is message 24 and subsequent) and upstream at
> > https://bugzilla.samba.org/show_bug.cgi?id=7020.  However, this seems
> > to indicate the restart logic isn't quite right, probably assuming
> > that after n seconds the stop has killed all the processes.
> 
> Yes, which is the only way we have to handle such restarts, until such time
> as we switch to an init system such as upstart that handles process
> supervision.
I thought some packages monitored process to see when they disappear.
That's not very elegant, of course, and I guess it's vulnerable to
problems if other programs are invoked with the same command line.

> 
> So while this bug isn't "wontfix", resolving it is dependent on other
> systemic changes in Debian.  Adding an extra delay to the restart command
> isn't a reasonable workaround.
> 
The delay in my case is so long that bumping the wait time is clearly
unreasonable.

Ross