[I strongly suspect this whole email thread went way out of scope of the issue really deserves] I didn't want to participate any further but let me clarify one thing because I can see how the discussion could generate some confusion. On Tue 23-10-18 10:37:38, Petr Mladek wrote: [...] > My understanding is that this situation happens when the system is > misconfigured and unusable without manual intervention. If > the user is able to see what the problem is then we are good. Not really. The flood of _memcg_ oom report about no eligible tasks should indeed happen only when the memcg is misconfigured. The system is and should be still usable at this stage. Ratelimit is aimed to reduce pointless message which do not help to debug the issue itself much. There is a race condition as explained by Tetsuo that could lead to this situation even without a misconfiguration and that is clearly a bug and something to deal with and patches have been posted in that regards [1] The rest of the discussion is about how to handle printk rate-limiting properly and whether ad-hoc solution is more appropriate than a real API we have in place and whether the later needs some enhancements. That is completely orthogonal on the issue at hands and as such it should be really discussed separately. [1] http://lkml.kernel.org/r/20181022071323.9550-1-mhocko@xxxxxxxxxx -- Michal Hocko SUSE Labs
