On Tue, Jan 02, 2018 at 03:19:38PM +0000, Dr. David Alan Gilbert wrote: > * Vladimir Sementsov-Ogievskiy (vsement...@virtuozzo.com) wrote: > > 28.12.2017 05:19, Peter Xu wrote: > > > On Wed, Dec 27, 2017 at 03:25:23PM +0300, Vladimir Sementsov-Ogievskiy > > > wrote: > > > > Hi all! > > > > > > > > Hmm, looks like leak is not fixed here: I've checked it while running > > > > iotest > > > > 181, that > > > > migration_instance_finalize is not called. > > > > > > > > If I understand correct, to call it we need unref current_migration > > > > object > > > > somewhere. > > > > > > > > Or, may be I'm missing something.. > > > I think you are right. > > > > > > It does not matter much though since we don't dynamically allocate > > > migration object (there is only one and it lives forever). Do you > > > want to post a patch? I guess the safest place to unref it is at the > > > end of main() to make sure no one will be using it any more. > > > > > > (Hmm, the incoming migration state is still static) > > > > > > Thanks, > > > > Ok, I'll send a patch. > > Be very very careful that it doesn't crash in cases like quit in the main > thread > while a migration is still running.
Agree. > > I have no problem with this object living forever and letting it just > die with exit(). Yes. But if better, I am thinking whether we should always make sure migration is completed/failed/cancelled before that point. If we quit QEMU with a working migration stream, logically we should stop it before leaving, as part of QEMU's cleanup path. -- Peter Xu
