On Tue 11-12-18 13:22:27, osalvador@xxxxxxx wrote: > On 2018-12-11 11:18, Michal Hocko wrote: [...] > > The main question here is. Do we want to migrate as much as possible or > > do we want to be conservative and bail out early. The later could be an > > advantage if the next attempt could fail the whole operation because the > > impact of the failed operation would be somehow reduced. The former > > should be better for throughput because easily done stuff is done first. > > > > I would go with the throuput because our failure mode is to bail out > > much earlier - even before we try to migrate. Even though the detection > > is not perfect it works reasonably well for most usecases. > > I agree here. > I think it is better to do as much work as possible at once. This would be great to mention in the changelog. Because that is the real justification for the change IMHO. -- Michal Hocko SUSE Labs
