On 05/16/2013 04:14 PM, Andrey Vagin wrote: > CRIU requires a pid of the init. > > Signed-off-by: Andrey Vagin <ava...@openvz.org>
The way you coded it, it seems to me that we will always overwrite the pid file, which is fine: this way we won't run into the usual "pid file already exists" kinds of problem. When we reach this point, we already tested if the container is running... My only fear is that we do that by verifying if there is any tasks already in the cgroup. So can we race against other container creation? We put ourselves in the container before we call clone, so *at this point* the vps_is_run test would fail, but by now we have already tested it a while ago... Maybe vzctl has other mechanism against such races that I am not aware of, but could you explain me why we don't have this race? It is my only concern, the rest looks fine. Having the init pid is quite handy for other situations as well, like stop --fast. So far we hack that by killing everybody until no more tasks are found in the cgroup, but what I always wanted was really a mechanism to kill init. _______________________________________________ Devel mailing list Devel@openvz.org https://lists.openvz.org/mailman/listinfo/devel