1393288369 J * bzed_ ~bzed@bzed.netrep.oftc.net
1393288458 Q * bzed Ping timeout: 480 seconds
1393288462 N * bzed_ bzed
1393290637 Q * zerick Remote host closed the connection
1393298365 Q * Aiken Remote host closed the connection
1393298718 J * Aiken ~Aiken@2001:44b8:2168:1000:21f:d0ff:fed6:d63f
1393299573 N * Bertl_zZ Bertl
1393299586 M * Bertl back now ...
1393301656 Q * Aiken Remote host closed the connection
1393306549 Q * fosco Remote host closed the connection
1393306551 J * fosco fosco@marx.wirefull.org
1393309623 J * [Guy] ~korn@elan.rulez.org
1393309662 Q * Guy- Read error: No route to host
1393310326 J * Ghislain ~aqueos@adsl1.aqueos.com
1393314947 J * beng_ ~BenG@cpc29-aztw22-2-0-cust128.18-1.cable.virginm.net
1393316631 M * BlackPanx i have a weird issue. i have exact 2 vserver guests (other host server). same clone of same vm, same packages and running memcached. if i start memcached on one it binds to 0.0.0.0:11211, and on the other it binds only to it's 192.168.2.20 ip. it won't bind on 0.0.0.0. what could be different ? 
1393316714 M * BlackPanx one host is debian running vserver and one is ubuntu running vserver
1393316874 M * ard the one that is bound to 192.168.2.20 has default single ip turned on
1393316914 M * ard you should turn off single ip on that one
1393317599 J * virman ~oftc-webi@91.218.79.125
1393317646 Q * virman Remote host closed the connection
1393318292 J * Aiken ~Aiken@2001:44b8:2168:1000:21f:d0ff:fed6:d63f
1393318466 M * beng_ hi all
1393318477 M * beng_ I'm still trying to get the 3.13 kernel built
1393318485 M * beng_ trying 3.13.5 now
1393318500 M * beng_ Bertl, did you get a look at the config I sent?
1393318633 M * Bertl I'm on it (i.e. I will check and fix it in a few minutes)
1393321926 J * virman ~oftc-webi@91.218.78.160
1393324798 M * beng_ cheers Bertl 
1393324832 M * beng_ 3.13.5 produced a different error, so I'm going back to trying 3.13.1 to get the full build log on each archictecture
1393324877 M * Bertl do you have your config for 3.13.5?
1393324884 M * Bertl can you upload that one as well?
1393324919 M * beng_ I used the same config
1393326102 J * renihs ~arf@83-65-34-34.arsenal.xdsl-line.inode.at
1393326334 M * beng_ BIAB
1393326341 Q * beng_ Quit: I Leave
1393326705 N * l0kit Guest1344
1393326711 J * l0kit ~1oxT@0001b54e.user.oftc.net
1393327105 Q * Guest1344 Ping timeout: 480 seconds
1393328845 J * beng_ ~BenG@cpc29-aztw22-2-0-cust128.18-1.cable.virginm.net
1393329981 Q * renihs Quit: narf
1393330195 J * renihs ~arf@83-65-34-34.arsenal.xdsl-line.inode.at
1393330296 Q * ircuser-1 Ping timeout: 480 seconds
1393332835 J * ircuser-1 ~ircuser-1@35.222-62-69.ftth.swbr.surewest.net
1393333757 Q * Aiken Remote host closed the connection
1393335013 M * shaggy63 Thanks daniel_hozac
1393335636 M * shaggy63 Beng, thanks for the wheezy packages!
1393335833 M * shaggy63 250 vservers in testing, 450 mb each only using 9 gb of space :)
1393335914 M * Bertl nice :)
1393336168 M * shaggy63 People say that disk is cheap which is correct but they all run the same processes in testing and will in production.  Having them all hashified the entire system runs pretty damn nicely.
1393336218 M * Bertl yes, folks forget about the memory impact
1393336529 M * shaggy63 Well, keep up the good work.
1393336611 M * Bertl thanks, will do
1393336789 M * arekm is KSM helping vserver, too?
1393336803 M * Bertl not really
1393336881 M * Bertl of course, if setup properly, it could help
1393336906 M * Bertl but it adds the overhead of identifying identical pages and actively merging them
1393336918 M * arekm afaik it has no settings, just "works" (or not)
1393336975 M * Bertl you need madvise to activate it
1393337000 M * arekm ouh
1393337015 M * Bertl but you can easily check if there is any sharing happening with sysfs
1393337033 M * Bertl i.e. there is a pages_shared entry or so
1393337089 M * arekm so not useful at all with vserver (unless someone patches every app to do madvise)
1393337101 M * Bertl yup
1393337118 M * Bertl it was designed and is most useful for qemu/kvm
1393337118 M * shaggy63 I figured out the cgroups memory limit.  Thanks for that.
1393337366 M * daniel_hozac you could write an LD_PRELOAD library to do it for all of your allocations.
1393337424 M * daniel_hozac but i'm not sure how much that would actually help, if at all.
1393337501 M * Bertl in any case, unification is more resource efficient
1393337993 Q * virman Remote host closed the connection
1393339054 M * ard The memory impact is mood if you run java :-(
1393339098 M * Bertl nobody runs java, java runs you :)
1393339421 M * ard :-)
1393339532 M * ard Maybe putting the jvms into a kvm per vserver... That should get ksm working at least :-)
1393339725 M * Bertl good idea :)
1393339816 M * ard But the LD_PRELOAD is certainly not a bad idea... having a lot of the same type of test/development environment with java....
1393339894 M * Bertl yeah, for something like java it could work reasonably well
1393339955 M * daniel_hozac i don't know enough about the allocator to know how it aligns things.
1393340011 M * daniel_hozac it does seem like something that the free JVMs would have an option for though
1393340801 J * zerick ~eocrospom@190.187.21.53
1393341359 M * ard http://kerneldedup.org/ has a lot on it
1393341405 M * ard But it's just like the guys behine alibaba.com... Good code and very nice fixes, open source, and everything chinese...
1393341584 M * [Guy] arekm: there is an ld_preload library that madvise()s all memory as potentially shareable
1393341617 M * [Guy] arekm: also, ksm does have some settings (like how often to run and how many pages to check in one go)
1393341652 N * [Guy] Guy-
1393341706 M * Guy- I use the ld_preload hack on one box where I run dozens of copies of the same proprietary crap, and it's helping _some_, but not all that much
1393341715 M * Guy- but of course ymmv
1393341834 M * Bertl can you paste the sysfs entry showing the shared pages?
1393341840 M * Bertl (just curious)
1393342166 M * CcxCZ Guy-: I'd be also curious about your runit setup, can explain mine in return ;)
1393342404 Q * zerick Ping timeout: 480 seconds
1393342537 M * CcxCZ hmm, I want some duplicate file detection tool that uses some sane file/db backend and doesn't max out the RAM
1393342628 M * Guy- pages_shared:58
1393342629 M * Guy- pages_sharing:2490
1393342629 M * Guy- pages_unshared:2111
1393342629 M * Guy- pages_volatile:36
1393342641 M * Guy- Bertl: ^^^
1393342688 M * Bertl wow. 58 pages shared!!
1393342699 M * Guy- quite.
1393342743 M * Bertl with a page size of 8k, that is almost half a megabyte?
1393342775 M * Bertl or is that huge pages with 4M each?
1393342812 M * Guy- iiuc there are 2490 copies of the 58 pages being shared
1393342856 M * Guy- so I have about 42-43 copies of each, on average
1393342859 M * Bertl ah, okay, so that would be something
1393342917 M * Bertl with 8k page size, that's at least about 20M saved, yes?
1393342965 M * Guy- thereabouts, yes
1393342980 M * Guy- (but isn't the page size 4k?)
1393342994 M * Bertl probably, yes
1393343014 M * Bertl depends on the architecture, I assumed 8 because it is one of the larger sizes
1393343021 M * Bertl sparc IIRC, can do up to 16k
1393343024 M * Guy- this is on amd64
1393343088 M * Guy- CcxCZ: my runit setup is pretty straightforward... I replace init with runit and have a wrapper for start-stop-daemon that starts and stops runit services the runit way
1393343122 M * Guy- CcxCZ: in vservers, I use initstyle plain and otherwise do everything the same way I'd do on the host
1393343132 M * Guy- CcxCZ: is there something specific you'd like to know?
1393343162 M * CcxCZ how do you share startup code / parametrize the services perhaps
1393343209 J * zerick ~eocrospom@190.187.21.53
1393343233 M * Guy- CcxCZ: a paradigm I often use is to have a foo-template service; I then create /etc/sv/foo-instance1, symlink the run script of foo-template in it
1393343279 M * daniel_hozac CcxCZ: like hashify?
1393343284 M * Guy- CcxCZ: and the script sources /etc/default/foo-instance1 (which it figures out from the name of its directory)
1393343331 M * CcxCZ daniel_hozac: I guess :) how do I run it on arbitrary directory?
1393343387 M * Guy- CcxCZ: what is it you specifically want? just hardlinking, or also COW?
1393343398 M * Guy- (or maybe even block level deduplication?)
1393343423 M * CcxCZ hardlinking should be fine here really, cow is provided by rsync
1393343435 M * CcxCZ Guy-: how do you configure the differences then?
1393343462 M * Guy- CcxCZ: I have a script that hardlinks identical files together
1393343502 M * Guy- CcxCZ: http://sprunge.us/IeEL
1393343510 M * CcxCZ there's duff, fdupes and another one… they all afaik keep the metadata in ram
1393343557 M * Guy- ah, and that's a problem? I have another script that doesn't, but it ignores owners and permissions :)
1393343569 Q * undefined Quit: Closing object
1393343572 M * Guy- CcxCZ: re runit: what differences?
1393343736 M * CcxCZ well, I sometimes have bunch of services with similar setup, eg. one per VLAN interface. so I encode that information in the service name where possible. for the rest I now keep it in the svtab file where the proper part is sourced but that could be multiple files as well
1393343766 M * Guy- CcxCZ: yes, this is handled via /etc/defaults/foo-*
1393343771 M * Guy- CcxCZ: other script: http://sprunge.us/INZH
1393343803 M * Guy- CcxCZ: I don't want to encode too much info in the service name because it's ugly; if it's just a single setting, I do it sometimes
1393343841 M * CcxCZ yeah, when it's stuff like dhcpcd.eth0 it's quite clear. otherwise it gets nasty
1393343936 M * CcxCZ the thing I wanted to automate was creation of the service directories, since can be whole bunch of files and I'd like to be able to add services programatically
1393344029 M * Guy- I don't quite see the use-case for that
1393344030 M * CcxCZ as added bonus I now have conditions (shell functions) that check whether the service should run, which is pretty handy on desktops where there are many combinations so they don't fit into the runlevel approach
1393344054 M * Guy- I have a subversion repository with service directories
1393344075 M * Guy- I check it out on every box and just symlink the subdirs to /service/
1393344120 M * Guy- the scripts are configurable enough (in cases where this matters) to never need local modifications
1393344132 M * Guy- and local configuration must live somewhere anyway
1393344190 M * CcxCZ well, mostly made this when I was doing embedded data logging boxes where the startup procedure is nontrivial, handles carrier sense on eth0, checks whether ttyUSB0 is present, runs pppd accordingly and runs ssh client to send data out and ntpd when we have actual internet access
1393344225 M * Guy- I would have done that differently
1393344243 M * CcxCZ some daemons are finicky, eg. ntpd when it can't sync at start it won't perform the one-time large adjustment I need
1393344247 M * Guy- for example, have a pppd service that runs unconditionally but just sleeps if there is no ttyUSB0
1393344272 M * CcxCZ yeah, that's the common approach
1393344288 M * Guy- I'd put an ntpdate call into the ntp run script and restart the ntp service when ppp0 comes up
1393344303 M * Guy- (from ppp/ip-up)
1393344331 M * Guy- automatically adding and removing service directories seems like unnecessary complexity to me
1393344379 M * CcxCZ well, I want to be able to add servers it writes data to in a simple fashion
1393344421 M * CcxCZ though sure, here it is probably overkill
1393344458 J * undefined ~undefined@66-190-97-211.dhcp.unas.tx.charter.com
1393344609 M * Guy- I never had need for more than two "runlevels" either (and only use two on boxes that have partially encrypted disks and need to boot autonomously to some point; services that depend on the encrypted parts in nontrivial ways are not part of the default "runlevel" in this case)
1393344715 P * undefined 
1393345024 M * beng_ Bertl, apologies, the compile of 3.13.5 produces the exact same error 3.13.1 for me
1393345130 M * Bertl check the new patch
1393345848 M * beng_ oo, exciting...
1393346694 M * beng_ compiling now
1393347321 M * CcxCZ Guy-: my take btw http://wpr.cz/ccx/paste/2014-02-25/0/
1393347467 M * Guy- yikes, ${(qqq)metas[$meta]}
1393347522 M * Guy- it looks nicer than my quick'n'dirty scripts
1393347528 M * CcxCZ prettyprinting :P
1393347530 M * Guy- but mine don't have external dependencies :)
1393347596 M * CcxCZ also unnecessarily reads/hashes every file
1393347869 M * CcxCZ I wonder how stable are device numbers coming from stat()
1393348673 M * daniel_hozac depends on your underlying filesystems.
1393348681 M * daniel_hozac or well, devices.
1393348758 M * CcxCZ yeah. I guess LVM volumes over mdraid won't be all that stable. perhaps stable enough as long you don't add/remove LVs
1393349147 M * daniel_hozac definitely not something i'd rely on.
1393349322 M * CcxCZ indeed. though for the usage I had in mind it'd just fall back to other checks, so only harm would be unnecessary processing
1393351244 Q * beng_ Quit: I Leave
1393353384 J * SteeleNivenson ~SteeleNiv@41.76.216.194
1393357404 M * Bertl off for a nap ... bbl
1393357436 N * Bertl Bertl_zZ
1393357801 Q * SteeleNivenson Ping timeout: 480 seconds
1393358400 J * SteeleNivenson ~SteeleNiv@105-237-5-122.access.mtnbusiness.co.za
1393359891 Q * PowerKe Ping timeout: 480 seconds
1393360143 J * Aiken ~Aiken@2001:44b8:2168:1000:21f:d0ff:fed6:d63f
1393360273 Q * SteeleNivenson Ping timeout: 480 seconds
1393361138 J * PowerKe ~tom@94-227-28-213.access.telenet.be
1393361973 J * SteeleNivenson ~SteeleNiv@41.76.216.194
1393362223 J * virman ~oftc-webi@91.218.79.125
1393362298 M * virman hello - i yesterday asked but i got disconnected - is there a way to set new cgroup limits for guest so he sees only his available memory ? setting memory limits in bytes seems including even simple cache so copying large files causes process to die..
1393362317 M * virman or i'm doing sth wrong :)
1393363129 J * bonbons ~bonbons@2001:a18:205:7b01:998d:4682:ffe7:601c
1393363395 M * virman mhm
1393363822 M * ard you need to set cflags
1393363845 M * ard the cgroup is limited, but the stats are not virtualised
1393363876 M * ard algapi:/etc/vservers/shell3# cat cflags 
1393363876 M * ard VIRT_MEM
1393363876 M * ard VIRT_CPU
1393363876 M * ard VIRT_LOAD
1393363920 M * ard And soon that will be from my o-droid U2 ...
1393364081 M * virman hm
1393364109 M * ard you know where to set the limits, right?
1393364127 M * virman VIRT things i know
1393364146 M * ard algapi:/etc/vservers/shell3/cgroup# grep . *
1393364146 M * ard cpu.shares:128
1393364146 M * ard memory.limit_in_bytes:256M
1393364146 M * ard memory.memsw.limit_in_bytes:256M
1393364172 M * virman so you say it will work just as good old vlimit ?
1393364196 M * ard I thought that vlimit did not work anymore, so I did not use it
1393364210 M * virman well in old good days it worked :P
1393364211 M * ard cgroup has more settings depending on your kernel...
1393364257 M * virman well i latelly have installed new vserver machine giving it another chance - i'm using xenserver now mostly but i love vserver doe to much more utilized space and resources.
1393364269 M * virman also it is faster when you compile everything from scratch (gentoo)
1393364289 M * ard I tried xen, but xen crashes and most installations of xen do weird things with eth0
1393364299 M * ard kvm was more stable
1393364311 M * ard but vserver was the most reliable for me :-)
1393364311 M * virman well i'm using citrix xenserver
1393364344 M * virman for now it used to work with some HP diskless boxes and gentoo based FC array :P
1393364359 M * ard The best part of vserver is that you don't have a binary-blob-filesystem...
1393364368 M * virman true
1393364425 M * virman that's why i'm trying to get newest kernel with all good old options and once i'll master it i'll rebuild my management tool so it is working again.. i got last ancient server still running on old kernel with 1200 days uptime!
1393364455 M * virman so.. i said to myself- why not- i got mail server there and i don;t want to set it all up again :P
1393364483 M * virman okko - no stories - let's work :P
1393364507 M * ard 1200 only?... :-)
1393364530 M * shaggy63 Gah, xenserver..
1393364608 M * virman ard : giving lifecycles andupgrades it is waaayyy ok :P i never used to run boxes without reboot for more than a year
1393364624 M * ard ard@skyfire:~$ uptime 
1393364624 M * ard  22:43:20 up 1821 days,  6:54,  2 users,  load average: 0.48, 0.19, 0.25
1393364638 M * virman ok :D
1393364650 M * ard that's containing most of the office stuff, but we have older systems...
1393364669 M * ard the oldest setup is with a 2.6.19 kernel
1393364706 M * ard 1996 days with a 2.6.26 :-)
1393364717 M * virman :)
1393364739 M * ard important systems that cannot be turned off, but where no budget exists to upgrade or make it even high-available...
1393364743 M * virman not actually secure thing. but well
1393364761 M * virman hehe
1393364768 M * virman so it is available - quite high :D
1393364785 M * ard We use vserver also to easy the different debian versions and application versions....
1393364786 M * virman btw - i can;t find a way to set VIRT_MEM flag on runtime...
1393364804 M * ard I don't think you can set that on run-time...
1393364804 Q * bonbons Quit: Leaving
1393364826 M * ard cflags are during the start. but you can modify the cgroup runtime afterwards
1393364834 M * virman ;(
1393364857 M * virman well - it is not critical - i just set it up anyway.. well
1393364867 M * virman turning off phone.... :D
1393364944 M * ard the kvm has been up for 715 days, and the xen was last crashed 221 days ago
1393364954 A * ard uses kvm within vservers
1393365041 M * virman i'm trying to squeeze ever bit of performance out of boxes so that's why i like vserver
1393365072 M * ard well, with 3.10 I had some problems using shared disks and memory cgroups...
1393365082 M * virman i got one production machine where i do not care about memory limits and isolation so i just run 4 processes on bunch of ssds and there vserver shines most
1393365084 M * ard ext4 was deadlocked waiting for memory
1393365097 M * virman don;t scare me
1393365139 M * virman and i formatted with ext4 leaving reiserfs for old hw ......
1393365283 M * ard reiserfs was rock solid till 2.6 :-)
1393365330 M * ard the ext4 deadlock is pretty hard to get, you must have endless loops of OOM for a few days to actually hit it
1393365351 M * ard and you must have the same filesystem shared between multiple vservers
1393365501 M * virman well it is usualy the case with client's machines
1393365519 M * virman one filesystem for x users
1393365528 M * daniel_hozac you can use vattribute to set flags during runtime.
1393365537 M * daniel_hozac and inspect.
1393365551 M * virman hm - nice one - i'll try
1393365559 M * ard thanks! will remember that as part of "basic knowledge" we need to know at work :-)
1393365573 M * virman cgroups works :D
1393365580 M * virman nice and fine :]
1393365658 M * virman vattribute --xid 6047 --set --bcap VIRT_MEM ?
1393365676 M * daniel_hozac does it go in bcapabilities?
1393365688 M * virman well that i do not know
1393365700 M * ard cflags... ==?
1393365704 M * ard ccap?
1393365709 M * virman ok.
1393365719 A * ard is just guessing...
1393365721 M * daniel_hozac no, flag.
1393365725 M * ard ah
1393365743 M * virman ok worked :]
1393365766 M * virman i need to one day write blog about it.. but then.. well everyone would learn it and i loose my job :D
1393365772 M * virman hehe
1393365804 M * ard try putting it on linux-vserver.org ... :-)
1393365815 M * ard I also use that as a brain dump sometimes :-)
1393365853 M * virman probably - there is way too much outdated docs - so i guess.. once i earn my flat back i'll write elaborate :D
1393365876 M * ard (because me documenting or describing thins usually end up in rants or "good old days")
1393365893 M * virman hehe
1393365954 M * virman i was forced to prepare a lot of documantation for managements, clients,, so it sticked to me to write itiot-proof explanations - i like that things also - no one knows everything and 'm sometimes missing dumb things :)
1393365978 M * virman like setting 5000% inode reserve :D
1393366046 M * ard I usually set something up, explain it to a second person and a third person go and documents it because there was none to begin with. Works perfectly for years now 8-D
1393366069 M * virman :D
1393366094 M * virman ok thanks anyway it works like charm.
1393366105 M * virman now getting into java controlling app again ...
1393366135 A * ard is going to do something close to sleeping....
1393366143 M * ard or playing with my odroid...
1393366221 M * ard it's going to be a pain to get the vserver patches running, because I can only find a 3.8 or a 3.13 tree
1393366579 M * daniel_hozac why would that make it a pain?
1393366665 M * ard vserver is for 3.4 or 3.10, not 3.8 .
1393366688 M * daniel_hozac or 3.13
1393366708 M * ard the 3.13 is what I am testing now... no reply :-(
1393366719 M * ard pfff....
1393366725 A * ard needs to order a serial cable
1393366754 M * ard 1.8v with a not-common pin spacing
1393366798 M * ard the cable is $15, and the import vat processing cost is 25 euro :-(
1393367074 M * ard for now I have to try to enable kexec in a working kernel... everytime changing sd is not nice
1393368061 Q * SteeleNivenson Ping timeout: 480 seconds
1393370372 Q * Ghislain Quit: Leaving.
1393371461 Q * kiorky Ping timeout: 480 seconds
1393372202 Q * virman Quit: Page closed