1467850878 M * Bertl_oO off to bed now ... 1467850883 N * Bertl_oO Bertl_zZ 1467851188 Q * fstd Read error: Connection reset by peer 1467851194 J * fstd ~fstd@x4e310563.dyn.telefonica.de 1467856693 J * aj__ ~aj@x4db12279.dyn.telefonica.de 1467857134 Q * derjohn_mobi Ping timeout: 480 seconds 1467874227 J * Ghislain ~aqueos@adsl1.aqueos.com 1467875463 M * Ghislain hi 1467875502 M * Ghislain so the limit seems to print right. But if you put silly things like 250G ram+swp where ram=16G and swap=5G you print the 250G 1467875569 M * Ghislain when you put no limit it show the real max so that works 1467875660 Q * aj__ Ping timeout: 480 seconds 1467875879 M * Ghislain yep seems to work ok apart from the case where i put silly limits 1467875901 M * Ghislain so i would say if we suppose the admin do not stupid thing it works 1467877186 J * aj__ ~aj@2001:6f8:1337:0:1cd1:41b1:9143:5005 1467878926 M * Ghislain the dentry limit is rising up to infinity 1467878956 M * Ghislain DENT: 201442 0/ 201442 -1/ -1 0 1467878956 M * Ghislain i have no limits on it and i stopped sysbench and i still get a 4/s rise on it , no decrease 1467879085 M * Ghislain killing all process but cron do not help decrease the dentry 1467879470 M * Ghislain ok, so from my test ram/swap is ok if i do not put silly limits, cache/buffer still 0. For the dentry limits i am increasing. 1467879496 M * Ghislain now that is stopped all process but one i am stable but still at 204k+ dentry 1467882265 N * Bertl_zZ Bertl 1467882269 M * Bertl morning folks! 1467882306 M * Bertl so the dentry is missing a dec somewhere ... will check 1467882426 M * Ghislain i did not found any instance where dentry->d_lockref.count-- is missing a dec 1467882437 M * Ghislain but seems there is one missing yes 1467882438 M * Ghislain OR 1467882446 M * Ghislain an inc is where it should not 1467882473 M * Bertl yes, that is a possibility too 1467888169 M * Guy- Ghislain: I'm not sure the thing with the silly limits is a bug; after all you could add more swap and/or memory later 1467888245 M * Ghislain well the limit will be the physical ones so i think it should print real thing even but if this reduce performance then stay that way and hit the admin with a hard wood stick 1467888270 M * Ghislain printing the right value can prevent the problem of small error that are not shown 1467888524 M * Bertl I think it is a bad habit to correct mistakes silently 1467888550 M * Bertl and as it is not trivial to make the kernel reject certain settings, this is something left for userspace 1467888594 M * Bertl so userspace tools can do various checks on the plausibility of the configured data, but the kernel just takes the values and works with them 1467888660 M * Ghislain you got a point 1467888698 M * Ghislain do you made other change than the basic memory swap cgroup correction in this patch bertl ? 1467888737 M * Bertl I removed the repeated COW stuff tim reported 1467888747 M * Bertl and I adjusted the dentry accounting 1467888775 M * Ghislain and removed the locks ? 1467888783 M * Bertl the cgroup memory limit calculations were switched to page count 1467888825 M * Bertl in theory it is more complicated than to assume pages there, but in practice the Linux-VServer routines are called with a memory unit which equals the page size 1467888835 M * Bertl so that's fine for now 1467888850 M * Bertl the locks I removed, there is not much sense in accounting them it seems 1467888897 M * Ghislain ok 1467888951 M * Bertl so what needs to be tested is COW link breaking and once I found the missing dentry accounting that as well 1467888999 M * Ghislain ok cool thx for the roundup 1467889013 P * undefined 1467889069 M * Bertl I haven't been able to reproduce the issue with sleep, but I haven't tested the guest on a real system yet 1467889106 M * Ghislain the kernel is running with 4 sysbench all the night and till now no issue at all 1467889222 M * Ghislain test are fileio, cpu, threads and mutex tests in a while true loop 1467889916 M * Ghislain i think this is a good stress test for the kernel, if you have any idea of what to try i can test any other thing 1467894381 M * tim_ Bertl: thank you for taking care if this issue :). Follow up expected soon as I've got some more :) 1467895112 M * Bertl thanks for reporting it! 1467896490 J * undefined ~undefined@00011a48.user.oftc.net 1467897615 M * Bertl off for now ... bbl 1467897619 N * Bertl Bertl_oO 1467898223 M * Guy- Bertl_oO: could the sleep thing be caused by a misapplied patch? 1467904161 Q * jrklein Remote host closed the connection 1467904166 M * daniel_hozac yeah, but it seems unlikely. 1467904173 J * jrklein ~cloud@proxy.dnihost.net 1467914519 J * Gremble ~Gremble@cpc87179-aztw31-2-0-cust6.18-1.cable.virginm.net 1467915690 Q * aj__ Ping timeout: 480 seconds 1467917347 J * aj__ ~aj@x4db12279.dyn.telefonica.de 1467918817 Q * Gremble Quit: I Leave 1467923402 Q * Aiken Remote host closed the connection 1467928640 M * Bertl_oO off to bed ... 1467928641 N * Bertl_oO Bertl_zZ 1467930350 J * Aiken ~Aiken@d63f.h.jbmb.net 1467930422 Q * Aiken Remote host closed the connection 1467930684 J * Aiken ~Aiken@d63f.h.jbmb.net 1467933607 Q * Long_yanG Remote host closed the connection 1467933610 J * LongyanG ~long@15255.s.t4vps.eu 1467934465 Q * Ghislain Quit: Leaving.