1438214424 J * fstd ~fstd@xdsl-84-44-226-23.netcologne.de
1438216179 Q * fstd Ping timeout: 480 seconds
1438216631 J * fstd ~fstd@xdsl-87-78-143-218.netcologne.de
1438222613 Q * jrklein_ Remote host closed the connection
1438222630 J * jrklein ~cloud@proxy.dnihost.net
1438228453 Q * arekm_ Read error: Connection reset by peer
1438228466 J * arekm ~arekm@ixion.pld-linux.org
1438235812 J * ivanF ~ivan@69-165-160-244.dsl.teksavvy.com
1438236340 M * ivanF the Distro Specific Notes for building vserver guests on Jessie mention that systemd needs to be prevented from installing and they have a --include=sysvinit-core,sysvinit, sysvinit-utils section and a --exclude=systemd. After examining the packages, I think it would make more sense to do --include=sysvinit-core,systemd-shim --exclude=systemd-sysv.
1438237161 M * Bertl_oO well, did you test that this works? if so, why not add a comment?
1438237605 M * ivanF Bertl_oO: I haven't tried it yet, but if it works I will add a comment.
1438237775 Q * derjohn_mob Ping timeout: 480 seconds
1438238011 J * derjohn_mob ~aj@p578b6aa1.dip0.t-ipconnect.de
1438239549 Q * derjohn_mob Ping timeout: 480 seconds
1438239666 J * Ghislain ~aqueos@adsl1.aqueos.com
1438240130 Q * Aiken Quit: Leaving
1438240134 Q * Ghislain Read error: Connection reset by peer
1438240150 J * Aiken ~Aiken@d63f.h.jbmb.net
1438240190 J * Ghislain ~aqueos@adsl1.aqueos.com
1438241244 J * derjohn_mob ~aj@fw.gkh-setu.de
1438243020 J * Ghislain1 ~aqueos@adsl1.aqueos.com
1438243050 J * Ghislain2 ~aqueos@adsl1.aqueos.com
1438243433 Q * Ghislain Ping timeout: 480 seconds
1438243503 Q * Ghislain1 Ping timeout: 480 seconds
1438243798 J * wicope ~wicope@0001fd8a.user.oftc.net
1438245805 Q * Ghislain2 Read error: Connection reset by peer
1438245863 J * Ghislain ~aqueos@adsl1.aqueos.com
1438246769 M * Bertl_oO off for a nap ... bbl
1438246777 N * Bertl_oO Bertl_zZ
1438246789 M * Ghislain good nap
1438247470 J * nikolayK ~nkichukov@199.91.137.248
1438255372 Q * AlexanderS Quit: WeeChat 1.1.1
1438256439 J * AlexanderS ~Alexander@home.zedat.fu-berlin.de
1438257061 Q * Jb_boin Ping timeout: 480 seconds
1438257234 J * Jb_boin ~dedior@proxad.eu
1438257560 Q * fstd Remote host closed the connection
1438257571 J * fstd ~fstd@xdsl-81-173-184-149.netcologne.de
1438258122 J * Gremble ~Gremble@cpc29-aztw22-2-0-cust128.18-1.cable.virginm.net
1438258868 Q * arekm Read error: Connection reset by peer
1438258890 J * arekm ~arekm@ixion.pld-linux.org
1438261275 N * Bertl_zZ Bertl
1438261277 M * Bertl back now ...
1438262103 M * Ghislain fresh and ready for everything ;p
1438262120 Q * Aiken Remote host closed the connection
1438262140 M * Ghislain i don't knwo if you had time to hit on 4.1 this week but if any testing is necessary i am here
1438262440 M * Bertl what is the current status of the testing? is the accounting working without the cow link breaking?
1438262452 M * Bertl daniel_hozac: did you have some time to look into that?
1438262598 M * Ghislain on the last test we had to disable it to prevent a crash, do you want me to try it with it on again ?
1438263058 M * Bertl we disabled the cow link breaking yes, but that still leaves the accounting on
1438263083 M * Bertl the question is, does the accounting work now or is it still "one off" on overwrites?
1438263188 M * Ghislain the testfs show different errors with  plenty of "out of space" ones and inode differences like "        verify 300: space -16 = -16
1438263188 M * Ghislain         verify 300: inode 0 = 5
1438263464 M * Ghislain each time i do a vserver xx enter and just exit the inode count  is +2
1438263553 M * Ghislain removing did change correctly to -1 but as we saw overwritting a file increase the inode count
1438263873 M * Bertl okay, good, so that is still an open issue, I suggest we tackle that first maybe daniel_hozac will find some time to look at the cow linkbreaking issues
1438263874 M * Ghislain creating 2 10mb files in the guest show the increase and decrease in space and inodes, overwriting a file does show the correct size
1438263885 M * Ghislain yep
1438263927 M * Bertl I should have some time in a few hours, I'll prepare a patch for testing
1438264084 M * Ghislain ok the quota seems also to have issue
1438264169 M * Ghislain if i do dd if=/dev/urandom of=/home/accesclient/my_10mb_file6 bs=10240 count=1024; multiple times seems only the first one is counted
1438264183 M * Ghislain oh wait
1438264220 M * Ghislain no it seems to catch up later ok there is a delay
1438264475 M * nikolayK Ghislain, you may want to use /dev/zero instead of /dev/urandom ... though you are just creating a very small file...
1438264685 M * Ghislain i love to deplete te entropy pool ;p
1438268084 M * Ghislain Bertl: ok so the little test i done seems to indicate that quota are ok inode and size wise, i created/deleted 150 random size files several times
1438268292 M * Ghislain and overwriting too so quota is ok
1438268340 M * Ghislain bertl: is it normal that i see /dev/hdv2 AND /dev/mapper/pool-wheezy641 as /var in my df -h ?
1438268348 M * Ghislain should it not mask the second one ?
1438268392 M * Bertl I presume the second one was mounted that way, only / is faked so far
1438268446 M * Bertl of course, with the (cgroup) device mapping, you should be able to "remap" the block device
1438268456 M * Ghislain i have a / that is a shared partition for all the guest so ican share inodes for library etc.. that have dlimits on it, then i have a /var that is with quota and is a partition dedicated to the guest
1438268490 M * Bertl so I don't think that changed between the last kernel and this one
1438268503 M * Ghislain not at all but i allway wondered
1438268550 M * Ghislain is it doable to hide the device linked to hdvXX ?
1438268775 M * Bertl if you are actually using the vroot device, it should be that which shows up
1438268845 M * Bertl but I presume you have a direct mount of that lvm 
1438268866 M * Ghislain the fstab is /dev/mapper/vserverpool-wheezy641, you mean i should put dev/hdv2 directly ?
1438268876 M * Ghislain well i really screw things up then
1438268922 M * Bertl I don't think that you can actually mount it
1438268944 M * Bertl haven't looked at the code for some time now, but I don't think we implemented any I/O
1438268971 M * Ghislain ok so i have no alternate choice the  fstab must have it and then it appear on df 
1438269017 M * Bertl the question is, what does /proc/mounts show?
1438269026 M * Bertl (inside the guest)
1438269102 M * Ghislain i tried /dev/hdv2 in fstab and i am hang now ;p let me a second to recover
1438269264 M * Ghislain i have some  reboot -d -f -i  process hanged i cannot kill in the context
1438269415 M * Ghislain humm seems i need to reboot, i cannot kill them
1438269585 M * Ghislain vkill --xid wheezy641 -s 9 -- 15517, is that not supposed to kill it ? it was in R state
1438269741 M * Ghislain hum i have some rcu_preempt detected stalls on CPUs/tasks, the reboot do not want to go with a vwait process at 100%cpu
1438269867 M * Ghislain bertl: FYI shutdown was not working i had to "reboot" directly
1438269882 Q * nikolayK Ping timeout: 480 seconds
1438269944 M * Bertl do you have kernel traces?
1438270003 M * Ghislain Bertl: /proc/mounts in the guest show /dev/root, the proc/dev/tmp and the var/tmp partition that is a bind mount (nosuid noexec etc..)
1438270019 M * Ghislain so the proc/mounts do not show /dev/hdv2 at all
1438270119 M * Ghislain http://pastebin.com/raw.php?i=wRfRusfk
1438270262 M * Ghislain bertl: and the proc mount http://pastebin.com/raw.php?i=AfNnrhDz
1438270721 M * Bertl yeah, I expected so mcuh
1438270723 M * Bertl *much
1438270807 M * Bertl the thing is, it would be rather simple to 'hide' or 'remap' this information, but the question is if it gives any actual benefit
1438270952 M * Ghislain the issue is that webmin for exemple try to launch commands on the device because it show thta is has quota
1438270987 M * Bertl and that would be better if it showed another name there?
1438271012 M * Ghislain i have to use a wrapper that filter things to make it work on all quota commands. No it would be better if only dev/hdvXX showed i bet
1438271044 M * Bertl in what way?
1438271061 M * Ghislain i have searched about the remap but i cannot find any information on how cgroup could help :)
1438271076 M * Bertl i.e. it would "try" to launch commands on a non existant /dev/hdvXX then, no?
1438271098 M * Ghislain for exemple if you do repquota /var it fails because it try repquota on the dev/mapper and not on dev/hdvxx
1438271122 M * Bertl so you do use the hdvxx for the devices then?
1438271149 M * Bertl i.e. there is a hdvxx configured for /var ?
1438271169 M * Ghislain i am not sure i understand what you mean, i use hdvXX for the var partition yes
1438271204 M * Ghislain it does not show in /proc/mounts but show in df and i can use the quota commands on it
1438271224 M * Ghislain but the system seems to prefer the /dev/mapper entry for commands so most fails
1438271258 M * Ghislain like repquota /var => 
1438271258 M * Ghislain UEST:wheezy641:/root%(root)> repquota /var
1438271258 M * Ghislain repquota: Mountpoint (or device) /var not found or has no quota enabled.
1438271272 P * undefined 
1438271309 M * Bertl an with repquota /dev/hdvXX it works?
1438271321 M * Ghislain on a non vserver system it works but the code here see the non hdvx line and try this one instead of the dev/hdvx for the device :). Yes it works on /dev/hdvx
1438271358 M * Ghislain this is what my wrapper does it try to map the directory to the dev/hdvX and remap the repquota to the vserver device
1438271406 M * Ghislain so it "works" but not naturaly you have to use wrappers and all
1438271589 M * Ghislain anyway the inode issue is the master thing rigth now as i moved around this issue, would be nice if it was more natural but anyway that works more or less
1438271664 M * Bertl recent kernels support a show_devname function per superblock/filesystem
1438271681 M * Bertl so it might be possible to set the device name with a mount option (in the future)
1438271710 M * Ghislain could be usefull :)
1438271754 Q * ggherdov__ Remote host closed the connection
1438271959 M * Bertl what was the last version which worked fine regarding inode allocation?
1438272057 M * Ghislain on my side  i think 3.18, previous ones had vspace issues hten from 3.18 the issue shifted to inodes, i thnink
1438272062 J * undefined ~undefined@00011a48.user.oftc.net
1438272097 M * Ghislain so seems the issue is between > 3.14.38
1438272135 M * Ghislain i mean after 3.14.38 (i have it on 3.18 and beng signaled it was not present on 3.14.38
1438272181 M * Bertl okay, tx
1438272257 Q * CcxCZ Ping timeout: 480 seconds
1438272388 M * Ghislain to be noted we have space error before and it was corrected
1438272400 M * Ghislain so perhaps this is just the same to be done on the inode part
1438272441 M * Ghislain 3.13 had the vspace errors but not the inode ones, 3.18 is the reverse, not any vspace issue but inode issue
1438272473 M * Bertl please try to comment out lines 1628 and 1629 in fs/quota/dquot.c
1438272477 M * Bertl         if ((ret = dl_alloc_inode(inode)))
1438272477 M * Bertl                 return ret;
1438272503 M * Bertl i.e. just add a /* */ around them or // on both lines
1438272642 M * Ghislain ok,  kernel is building, will take some time
1438272671 M * Bertl np, thanks!
1438272689 M * Ghislain do you think the dcache.c crash i had could be related to this ?
1438272705 M * Bertl doesn't look like
1438272721 M * Ghislain ok too bad i like to hit several birds with one stone
1438272950 J * bonbons ~bonbons@2001:a18:201:f001:4c31:ecaa:d40a:e93d
1438273000 M * Ghislain is there a reason for the hdvxx system to be seen as UFS intead of the real FS underneath ?
1438273003 J * ggherdov__ sid11402@id-11402.ealing.irccloud.com
1438273056 M * Bertl yes, because some of the quota tools have been "too smart" in the past and decided to go for direct device access on e.g. ext2/3/4
1438273076 M * Bertl which in turn caused the quota tools to fail miserably
1438273187 M * Ghislain this quota system seems really a legacy horror. Now BRTFS is shiny new and it handle it better : it is not even able to tell the space used by files and the space left lol
1438273279 M * Ghislain well it can if you ask it but then it has to reread all the trees to find out , perhaps they improved since last i heard about all the quota nigthmare with it
1438273373 M * Bertl I think that improved a lot, problem is, with a filesystem that allows for snapshots and copy on write as well as raid natively, it is not so simple to tell free space
1438273416 M * Ghislain oh i get it completly but this is rather funny, imagine: your disk is full !! why ?? no way to know ! LoL
1438273438 M * Ghislain that sound like a funny dialog with a customer no ?
1438273485 M * Bertl to know why all you need is 'du' which will work the same on ext3/4 and btrfs
1438273541 M * Bertl we've seen what tricks are required to get a sensible quota display working, and as you probably know, it only works because it keeps track of changes
1438273566 M * Bertl if we would calculate the quota correctly, it would take ages on ext2/3/4
1438273571 M * Ghislain yes all of this is really a mess. It is still needed for me to prevent stupid to fillup disk
1438273592 M * Ghislain xfs seems to handle it inside the FS should be more accurate
1438273874 M * Ghislain dam the machine is locked again
1438273888 M * Bertl yeah, that's why we initially started work on xfs, because of the quota support
1438273924 M * Bertl but xfs is still a mess after all those years (code wise)
1438274007 M * Ghislain could not tell, ext4 works fine so , i had some hope for xfs for quota and project quota but this was not working on vserver mixed with tag
1438274022 M * Ghislain http://pastebin.com/raw.php?i=jW1V7HT2, i powercycle because the guest cannot be stopped
1438274121 M * Bertl it looks like vwait is still waiting for the guest to shut down
1438274136 M * Ghislain apart from the inode there seems to have another issue i cannot stop guest they lock the system
1438274137 M * Ghislain yes
1438274175 M * Bertl I think you need to enble vserver debugging to investigate this further
1438274196 M * Ghislain in the kernel or systctl ? if sysctl tell me how
1438274200 M * Bertl i.e. it might be that a reference to the context is still held or similar
1438274243 M * Bertl CONFIG_VSERVER_DEBUG=y
1438274258 M * Bertl in the config, then you can enable it via sysfs after startup
1438274301 M * Ghislain CONFIG_VSERVER_DEBUG=y
1438274319 M * Ghislain so.. i put how many in vserver.debug_xxx
1438274325 M * Ghislain 1 or 255 or ?
1438274654 M * Ghislain Bertl: well i tried 255 and rsyslog exploded to 100% cpu
1438274739 M * Ghislain -rw-r----- 1 root adm 298M Jul 30 18:44 /var/log/kern.log
1438274739 M * Ghislain in a few seconds lol
1438274768 M * Bertl lol, yeah, not unexpected
1438274918 M * Ghislain well inodes number do not move at all when creating files
1438274980 M * Ghislain humm
1438275063 M * Ghislain no i said silly thing
1438275108 M * Ghislain i was on /var not on /
1438275115 M * Bertl okay :)
1438275162 M * Ghislain hum seems ok now
1438275221 M * Bertl okay, so next problem: cow link breaking I guess?
1438275226 Q * arekm Quit: leaving
1438275232 M * Ghislain space_used=1103300
1438275232 M * Ghislain inodes_used=77236
1438275232 M * Ghislain i create 150 file then overwrite them then delete them
1438275232 M * Ghislain space_used=1086452
1438275232 M * Ghislain inodes_used=77236
1438275232 F * FloodServ +q *!*@adsl1.aqueos.com
1438275269 M * Bertl if you do the same test again, how does the space usage change?
1438275282 M * Bertl (because at the moment you gained space)
1438275330 J * CcxCZ ~ccxCZ@asterix.te2000.cz
1438275416 J * arekm ~arekm@ixion.pld-linux.org
1438275999 Q * derjohn_mob Ping timeout: 480 seconds
1438276622 M * Bertl off for now ... bbl
1438276626 N * Bertl Bertl_oO
1438277353 Q * Gremble Quit: I Leave
1438278837 F * FloodServ -q *!*@adsl1.aqueos.com
1438281503 J * opuk_ ~kupo@h-1-5.a176.priv.bahnhof.se
1438281609 Q * opuk Ping timeout: 480 seconds
1438282589 N * opuk_ opuk
1438288596 Q * bonbons Quit: Leaving
1438288902 J * Aiken ~Aiken@d63f.h.jbmb.net
1438292288 Q * Ghislain Quit: Leaving.
1438292843 J * Ghislain ~aqueos@adsl1.aqueos.com
1438292997 Q * Ghislain 
1438293436 J * Ghislain ~aqueos@adsl1.aqueos.com
1438293567 Q * Ghislain 
1438294057 J * Ghislain ~aqueos@adsl1.aqueos.com
1438294167 Q * Ghislain 
1438294659 J * Ghislain ~aqueos@adsl1.aqueos.com
1438298654 Q * wicope Remote host closed the connection
1438299753 Q * Ghislain Quit: Leaving.
1438300059 J * Ghislain ~aqueos@adsl1.aqueos.com
1438300237 Q * Ghislain 
1438300760 Q * fstd Remote host closed the connection
1438300771 J * fstd ~fstd@xdsl-87-78-16-240.netcologne.de