[Users] forced shutdown with client agent
I have run into a scenario after installing the client agent. If a VM is shutdown, the client agent calls the shutdown command with a 1 minute timeout. Dummy-2::INFO::2013-03-28 14:05:21,892::vdsAgentLogic::138::root::Shutting down (timeout = 30, message = 'System Administrator has initiated shutdown of this Virtual Machine. Virtual Machine is shutting down.' Since the shutdown command is called with time parameter the VM sets the /etc/nologin file. When the VM is forced down the /etc/nologin file is not cleared and when it comes back up only root can login until the /etc/nologin file is cleared. Is their some some reason the shutdown time is set to 30 seconds (rounded up to 1 minute in the code)? Are there any know issues with setting this to 0? Is this the right way to change it to 0? psql engine postgres -c "update vdc_options set option_value = '0' where option_name = 'VmGracefulShutdownTimeout';"
On 20 Apr 2013, at 22:42, Itamar Heim <iheim@redhat.com> wrote:
possibly yes
notmally the grace period is there for other users to react befor shutdown starts. I wonder if it's that much relevant here.., if we assume single user most of the time 0 (i.e. "now") should be ok
On 04/20/2013 10:52 PM, Michal Skrivanek wrote:
Sounds pretty much like a bug to me
-- Regards, Vinzenz Feenstra | Senior Software Engineer RedHat Engineering Virtualization R & D Phone: +420 532 294 625 IRC: vfeenstr or evilissimo Better technology. Faster innovation. Powered by community collaboration. See how it works at redhat.com
----- Original Message -----
The idea was to allow a logged in user to orderly close his work right? For that I'm not sure that even 30 seconds is enough. So this shutdown grace period really depend if there are users connected to the VM or not. The question is how to easily distinguish - maybe depend on a spice session connection? That may be too heuristic. But in any case the timeout should be per configuration and not be up-to-doubled in the code.
On Apr 22, 2013, at 12:56 , Simon Grinberg <sgrinber@redhat.com> wrote:
For that typically sysadmins use 5 or 10 minutes. 30s is worthless, no one would be able to react that fast…so to me this is the same as initiating shutdown right away. It's graceful anyway and all common apps would save the workspace on SIGTERM anyway. Well, does it even do anything on Windows? That's why I'd vote for 0. And keep is configurable in case you want to use it, sure.
yes, that's wrong
----- Original Message -----
From all the said above seems that indeed default shut-down time should be 0 But we need to be able to set shut-down policy per VM/Cluster/Something that will override the global config. One time fits all does not sound reasonable.
I would say that for Desktop use case you'll probably want to allow the User some time to finish up things, while for Server you won't care. Etc....
On Tue, Apr 23, 2013 at 8:58 AM, Michal Skrivanek wrote:
BTW: I tried today with ovirt 3.2.1 and it is not possible to interrupt shutdown, at least with "shutdown -c" command With agent inside a CentOS 6.4 VM the comamnd run is: [root@c2 ~]# ps -ef|grep shut root 15767 1581 0 23:39 ? 00:00:00 rhev-shutdown -h +1 "System Administrator has initiated shutdown of this Virtual Machine. Virtual Machine is shutting down." root 15768 15767 0 23:39 ? 00:00:00 [shutdown] <defunct> root 15799 15781 0 23:40 pts/0 00:00:00 grep shut If you try to cancel it with shutdown -c you get shutdown: Cannot find pid of running shutdown I don't know if Michal meant instead to kill the pid processes themselves (15767 and and 15768 in my case).... Gianluca
participants (8)
-
Gianluca Cecchi -
Itamar Heim -
Michal Skrivanek -
Michal Skrivanek -
Omer Frenkel -
Simon Grinberg -
Thomas Scofield -
Vinzenz Feenstra