1641255616 J * daniel_hozac ~daniel@217-211-243-90-no536.tbcn.telia.com 1641269059 M * Bertl_oO off to bed now ... have a good one everyone! 1641269060 N * Bertl_oO Bertl_zZ 1641281023 J * Ghislain ~ghislain@adsl1.aqueos.com 1641286553 M * Ghislain well my lxcfs with a vserver kernel or a normal debian kernel all fail 1641294809 N * Bertl_zZ Bertl 1641294811 M * Bertl morning folks! 1641294824 M * Bertl Ghislain: so lxcfs is not a suitable replacement then? 1641294875 M * Ghislain it seems not in the actual state and as i dont know anything about programming it seems i cannot try to patch it. 1641294886 M * Ghislain i tried to understand the code but i really dont 1641294955 M * Ghislain in theory it should be really great for us but i have no mean to see why it fails if there is no pid namespace 1641295256 M * Bertl and the author(s)/developers couldn't (or didn't want to) help there either I presume? 1641295273 M * Ghislain i left them a message but they have not answered yet 1641295298 M * Ghislain https://github.com/lxc/lxcfs/issues/502. 1641295476 M * Ghislain tried to test pid namespace by cgexec -g *:testlxcfs unshare --pid -r chroot /vservers/testlxcfs /bin/bash 1641295476 M * Ghislain but it fails too (to see if we can use it if adding namespace to vserver instead) 1641295535 M * Bertl did you test it with lxc/lxd? does that work as expected? 1641295549 M * Ghislain in lxd it works fine 1641295633 M * Bertl what if you compare a simple memory test between lxd and chroot e.g. just with strace on the fuse code? 1641295686 M * Ghislain cgexec -g *:testlxcfs unshare --pid -r -f chroot /vservers/testlxcfs /bin/bash works but the shell show still the 32Go ram 1641295744 M * Ghislain lxd process do so manything i dont think the strace could be readable 1641295783 M * AlexanderS Ghislain: Which version of lxcfs are you using for your tests? 1641295812 M * Bertl well, the main question here seems to be: is the lxcfs triggered by your chroot tests at all? 1641295825 M * Ghislain 4.0.0 1641295829 M * Bertl then, if that is the case, where does it get the data from? 1641295852 M * Bertl and if that looks sound, the final question would be: why is the data still incorrect? 1641296026 M * AlexanderS Ghislain: Ok, 4.0.0 seems to support "shared pid namespaces": https://github.com/lxc/lxcfs/commit/c9236032f87c3d8f4be9a76e29fe1fd0c623bba2 (So it should work without pidns.) 1641296164 M * Ghislain yes i know do you want acdess to my test machine to see what i did wrong ? 1641296712 M * Ghislain i use this script to test the chroot 1641296714 M * Ghislain https://pastebin.com/raw/CnKcQGqw 1641296721 M * Ghislain i must go eat , brb 1641298485 M * Ghislain it says 4.0.0 but this is compiled from source of git clone so it should be 4.0.11 at least 1641305355 M * Bertl off for now ... bbl 1641305356 N * Bertl Bertl_oO 1641305651 M * Ghislain i tried 4.0.11, funny build is make instead of meson for the git.. anyway same result 1641305719 M * Ghislain so i must miss something( i checked the patch you linked is there) because it do not work like it should 1641307572 M * AlexanderS Ghislain: You could try to set "initpid = fc->pid;" unconditially at the patched locations. So we could eliminate potential errors in is_shared_pidns(..). 1641307663 M * AlexanderS If this is working, we can try to idenify the problem with the pidns check. 1641307707 M * Ghislain is_shared_pidns i chnage this to return allways true ? is that what you tell me ? 1641307894 M * AlexanderS Ghislain: I thought of something like this: https://pastebin.com/LV1sQVfA 1641307957 M * AlexanderS But "return true" at top of is_shared_pidns should work, too. 1641308004 M * Ghislain the patch you sent is on which file ? 1641308372 M * Ghislain ok i see 1641308517 M * AlexanderS This was just an example. This would be the full patch: https://pastebin.com/seHBFN7p 1641308657 M * Ghislain i failed with the return true on is_shared_pidns 1641308681 M * Ghislain make clean make make install, restart the process and still 32Gb memory in the chroot 1641308898 M * AlexanderS Can you strace the lxcfs process while you execute free? 1641309104 M * Ghislain https://pastebin.com/raw/SPZy5cEf 1641309309 M * Ghislain looks like hell , just a little better than a java config file 1641309617 M * AlexanderS Ghislain: Could you strace lxcfs while running just "cat /proc/meminfo" inside the chroot? 1641309793 M * Ghislain https://pastebin.com/raw/beu4kYLv 1641309888 M * Ghislain as i said i can give access if you want to test :) 1641311474 M * Ghislain if i remove -d of the lxcfs the strace is a lot smaller 1641311499 M * Ghislain but obvioulsy you have less informations 1641311549 M * Ghislain but you avoid all the read write of the debug information in the strace 1641312833 M * Ghislain alexanderS private conversation do not work well on my system, you can send it to gadnet@aqueos.com ;p sorry for the issue my otr plugin mess with private conversation that are not crypted 1641312964 M * AlexanderS Ghislain: Ok, I will send you an email. 1641313488 M * Ghislain key should be ok, ip is in the mail 1641313590 N * emcepe mcp 1641319324 M * Ghislain is it working ? 1641322203 M * AlexanderS Ghislain: Sorry, was in a conference. Yes it is working. 1641323797 M * AlexanderS Ghislain: get_hierarchy in https://github.com/lxc/lxcfs/blob/d7195f5fa3591b47c085dd3015e5c17a75fd0811/src/cgroups/cgroup.h#L207 seems to return NULL. It seems, that lxcfs cannot detect the mounted cgroups. 1641325655 M * Ghislain cgroup2 on /sys/fs/cgroup/unified type cgroup2 (rw,nosuid,nodev,noexec,relatime) on lxc there is cgroup2 mounted 1641325664 M * Ghislain perhaps it need to be added ? 1641325776 M * AlexanderS On a debian buster I have something like this: https://pastebin.com/raw/n2Ep9xVM 1641325894 M * Ghislain yes its not just cgroup mounted it a false cgropup as tmpfs with all sub directory mounted one by one 1641325918 M * Ghislain its lxd or you have a fstab or something else ? 1641326025 M * Ghislain i do not touch anything :) 1641326034 M * Ghislain or i will step on your foot 1641327468 M * AlexanderS Ghislain: It's systemd. 1641327478 M * Ghislain oh 1641327495 M * AlexanderS Ghislain: Can I reboot this maschine? 1641327502 M * Ghislain well is there something systemd do not do 1641327506 M * Ghislain yes of course 1641329073 M * Ghislain one thing if you recompile lxcfs my scriupt use the lxc /usr/local/share/lxcfs/lxc.mount.hook and if you recompile this must be changed and you must replace /usr/local//var/lib/lxcfs with /var/lib/lxcfs in it 1641329307 M * AlexanderS Ghislain: It is working now. There is an error: "/usr/local/share/lxcfs/lxc.mount.hook: 50: [: unexpected operator" but free is showing the correct values. 1641329324 M * AlexanderS Ghislain: Use testchroot2-per_ss.txt 1641329366 M * Ghislain ok so it was not the system but just that it ask for a specific mounting of cgroup as per the systemd sect 1641329375 M * Ghislain thanks a lot for the work ! 1641329379 M * AlexanderS Ghislain: I touched /etc/vservers/.defaults/cgroup/per-ss to update the mount stuff. 1641329559 M * Ghislain oh woha, so for one chroot it nedd 33 mounts 1641329764 M * Ghislain i did not known that per-ss existed lol 1641329923 M * Ghislain for me the cgexec -g cpuset,memory:${GUEST} chroot /vservers/${GUEST} free 1641329923 M * Ghislain is failling 1641330069 M * AlexanderS Ghislain: lxcfs was stopped when i logged out 1641330082 M * Ghislain oh stupid me ! 1641330084 M * Ghislain ok 1641330756 M * Ghislain not working now but i am too tired, will try again tomorow. Tahnks for the help ! 1641336968 Q * Romster Ping timeout: 480 seconds 1641340593 Q * Ghislain Ping timeout: 480 seconds