Re: [libvirt] [PATCH] Add APIs for killing off processes inside a cgroup

On Wed, Feb 23, 2011 at 03:26:30PM +0800, Daniel Veillard wrote: > On Tue, Feb 22, 2011 at 11:17:05AM -0700, Eric Blake wrote: > > On 02/22/2011 10:55 AM, Daniel P. Berrange wrote: > > > The virCgroupKill method kills all PIDs found in a cgroup > > > > > > The virCgroupKillRecursively method does this recursively > > > for child cgroups. > > > > > > The virCgroupKillPainfully method does a recursive kill > > > several times in a row until everything has really died > > > > Do we really need three APIs, or can we just have: > > > > virCgroupKill(..., VIR_CGROUP_KILL_RECURSIVE | VIR_CGROUP_KILL_PAINFUL) > > > > I haven't reviewed the body of the patch, yet. > > Agreed, I was surpized by this at first. > > Went though the patch my only remarks are > rc = killedAny ? 1 : 0; > is overkill since killedAny values seems only 0 or 1 anyway > and the loop for virCgroupKillPainfully looks a bit arbitrary, > sending 15 signals to all looks a lot, and the delay for the loop > a bit short, if there is like a hundred processes in that group that > could be a bit heavy, isn't it ? The apporach & timings I'm following here are the same as that used in systemd, and the only way a process could not die after the SIGKILL is if the kernel had it in an uninterruptable sleep. So I think this is sufficient for our needs - it certainly is dramatically more reliable than the code it is replacing in LXC :-)