1125878404 M * daniel_hozac that runs 5 tests or so, IIRC. 1125878410 M * hillct ah 1125878416 M * daniel_hozac s/5/9/ 1125878418 M * hillct the output I saw doesn't say 1125878424 M * daniel_hozac run it manually after the build has failed. 1125878448 M * hillct will try again shortly, but whenever I run it manually, all succeeds 1125878466 M * hillct still wavn't been able to work out the differences between the manual run and the spec file run 1125878500 M * daniel_hozac so rpmbuild -ba util-vserver.spec; cd /util-vserver-0.30.208; make check succeeds? 1125878534 M * hillct hang on 1125878583 M * hillct fails same way, same output 1125878595 M * hillct I meant when I make by hand 1125878606 M * hillct not using the RPM build at all 1125878624 M * hillct can't quite reproduce the donditions of the RPM build in that scenerio 1125878626 M * daniel_hozac ok, so run lib/testsuite/fmt 1125878653 M * hillct [root@vmaster util-vserver-0.30.208]# ./lib/testsuite/fmt 1125878653 M * hillct Killed 1125878653 M * hillct [root@vmaster util-vserver-0.30.208]# 1125878658 M * hillct hmm 1125878681 M * hillct could it be the test is being interupted by the running vservers? 1125878689 M * hillct It appears to be an integer format test though 1125878692 M * hillct should be simple 1125878759 M * daniel_hozac if you run it through gdb or similar, can you tell where it's getting killed? 1125878782 M * hillct gotta recall gdb syntax 1125878811 M * daniel_hozac just gdb lib/testsuite/fmt; run 1125878857 M * daniel_hozac bt once it's killed. 1125878951 M * hillct No Stack 1125878976 M * daniel_hozac oh, of course. sorry, brain malfunctioning. 1125879015 M * daniel_hozac do you get anything in your kernel logs? i don't see why it would be getting killed. 1125879078 M * hillct nothing at all 1125879306 M * hillct what's this configure option about? 1125879308 M * hillct --with-initrddir= 1125879327 M * daniel_hozac where to put init scripts. 1125879356 M * hillct makes sense 1125879376 M * hillct the others being used are: 1125879378 M * hillct --enable-release 1125879392 M * hillct er, that's it 1125879619 M * daniel_hozac do the utils created work, even though the format test fails? 1125879633 M * hillct duno 1125879643 M * hillct I was afraid to test them lest they blow up my vservers 1125879789 M * hillct where in the built tree are the binaries? 1125879799 M * daniel_hozac src/ 1125879801 M * hillct they're not in an obvious location 1125879802 M * hillct ah 1125879817 M * hillct guess they get installed directly from there 1125879885 M * hillct they seem to work 1125879895 M * hillct I mean, I guess I could just disable the tests 1125879903 M * hillct , build and install 1125879908 M * hillct but that seems wrong 1125879976 M * daniel_hozac try using vlimit. 1125880046 M * hillct [root@vmaster src]# ./vlimit -a 1125880047 M * hillct vc_get_rlimit(): No such process 1125880047 M * hillct RSS N/A N/A 0x407e0 1125880052 M * hillct for vlimit -a 1125880059 M * hillct not a good sign 1125880087 M * daniel_hozac try vlimit -a --xid 1125880137 M * hillct [root@vmaster src]# ./vlimit -a --xid 5 1125880137 M * hillct RSS N/A N/A inf 1125880137 M * hillct NPROC N/A N/A inf 1125880137 M * hillct NOFILE N/A N/A inf 1125880137 M * hillct MEMLOCK N/A N/A inf 1125880138 M * hillct AS N/A N/A inf 1125880138 M * hillct vc_get_rlimit(): [unknown error] 1125880141 M * hillct LOCKS N/A N/A 0x40020c 1125880141 M * hillct 18 N/A N/A inf 1125880142 M * hillct [root@vmaster src]# 1125880164 M * hillct hmm 1125880181 M * hillct I wonder if I compiled the kernel without the required legacy support 1125880217 M * daniel_hozac you should only need legacy support if you have legacy guests, or actually using legacy tools. 1125880231 M * hillct I'm not and I don't 1125880234 M * daniel_hozac legacy networking support is all you need if using new style guests and recent tools. 1125880245 M * daniel_hozac (in terms of legacy support, that is) 1125880246 M * hillct everything was setup initially usin 1.9.5 1125880278 M * hillct and tools 206 1125880315 M * hillct what I did was move from FC2 to FC4 on the host and rebuild the kernel going from VS 1.95 to 2.01 1125880340 M * daniel_hozac 2.0.1-pre2? or what? 1125880360 M * hillct yup 1125880395 M * hillct and I have active vservers running 1125880403 M * hillct using this kernel 1125880543 M * hillct the tools build fine for i386 right? 1125880562 M * daniel_hozac they do for x86_64 as well. 1125880565 M * hillct I remember hearing of someone using FC4 on i385 host 1125880573 M * daniel_hozac but yeah, i have i386 RPMs for FC4 working good.. 1125880576 M * hillct er 386 1125880596 M * daniel_hozac (with Bertl's patch and one of my patches) 1125880610 M * hillct and I can get the 208 tools built for x86_64 using FC2 1125880651 M * hillct but it's something about FC4 x86_64 1125880667 M * hillct and or the way the spec file handles that architecture 1125880683 M * daniel_hozac there's nothing architecture specific in the spec, IIRC. 1125880705 M * hillct not that I saw 1125880715 M * hillct yeah, it's really odd 1125880726 M * daniel_hozac could you try something like %define optflags -g -O2 -Wall in the spec file? 1125880766 M * hillct sure, but not at the moment. something just came up 1125880775 A * hillct copies the flags down 1125880790 M * daniel_hozac any flags will do, just override the defaults. 1125881146 N * Bertl_oO Bertl 1125881153 M * Bertl evening folks! 1125881161 M * hillct hey 1125881165 M * daniel_hozac evening! 1125881190 M * Bertl hillct: is the vlimit issues you showed already resolved? 1125881240 M * daniel_hozac the problem itself is that the testsuite for the formatting functions fail. 1125881301 M * hillct Bertl nope 1125881308 M * hillct as daniel_hozac says 1125881338 M * hillct now if I add something like --with-optflags= to the configure args, will that pass them along properly? 1125881350 M * daniel_hozac no, %configure will do that for you. 1125881362 M * daniel_hozac (i hope, i believe optflags is the correct one) 1125881366 M * hillct og optfalgs is pulled in automtically 1125881374 M * hillct er oh 1125881377 M * hillct K 1125881508 M * Bertl okay, so _what_ do we know by now? (short explanation of the issue) 1125881538 M * hillct the fmt test suite fails 1125881543 M * daniel_hozac well, not much, lib/testsuite/fmt gets killed for some reason. 1125881549 M * hillct because the test is killed 1125881555 M * hillct yeah 1125881569 M * Bertl on _what_ kernel and _distro_? 1125881601 M * hillct kernel 2.2.6.13vs2.0-vs2.0.1-pre2 1125881603 M * hillct FC4 1125881611 M * hillct x96_64 1125881618 M * hillct x86_64 1125881635 M * hillct same result with optflags 1125881655 M * Bertl okay, which gcc for kernel and tools? 1125881697 M * hillct gcc version 4.0.1 20050727 (Red Hat 4.0.1-5) 1125881743 M * hillct kernel is successfully running two vservers 1125881747 M * hillct at the moment 1125881755 M * Bertl what gives 'showattr /proc/stat' ? 1125881779 M * hillct [root@vmaster SPECS]# showattr /proc/stat 1125881779 M * hillct Awh-ui- /proc/stat 1125881779 M * hillct [root@vmaster SPECS]# 1125881803 M * Bertl and the testme.sh runs fine? (I assume older tools, no)? 1125881825 M * hillct earlier tools were 30.206 1125881831 M * hillct yes 1125881858 M * hillct although I can't recompile them under FC4 either 1125881875 M * hillct they're using the binaries compiled under FC2 1125881952 M * Bertl could you downgrade the compiler? 1125881974 M * hillct I guess 1125881984 M * hillct I'd really rather not if I can avoid it though 1125881999 M * Bertl then let's try that ... you can upgrade again if that doesn't help 1125882022 M * hillct what shall I downgrade to? 1125882052 M * Bertl 3.3.x if possible, 3.4.x if not :) 1125882057 M * daniel_hozac gcc 3.2 is readily available ;) 1125882084 M * hillct well, I can't install 3.3 on HC4 1125882088 M * hillct on FC4 1125882102 M * hillct but I have an FC2 x86_64 box 1125882108 M * hillct where I know the compile succeeds 1125882121 M * hillct gcc version 3.3.3 20040412 (Red Hat Linux 3.3.3-7) 1125882127 M * Bertl yeah, good, but that doesn't help us to narrow it down, no? 1125882149 M * Bertl (unless you update that one to FC4 without touching the gcc :) 1125882158 M * hillct well, it's got the exact same version of dietlibc 1125882182 M * hillct let me see what it'll take to downgrade GCC 1125882190 M * hillct there may be RPMs available 1125882201 M * hillct I'm trying to keep this box RPM-only 1125882221 M * daniel_hozac as i said, you have gcc 3.2 in the form of compat-gcc-32. 1125882233 M * hillct ah 1125882239 M * hillct sorry missed that 1125882935 Q * ntrs_ Remote host closed the connection 1125882970 J * ntrs_ ~ntrs@Dardeene-68.188.50.87.charter-stl.com 1125883039 M * Bertl wb ntrs_! 1125883289 M * hillct K. Got the compatibility compiler 1125883296 M * hillct I'll be back to this after dinner 1125883302 M * Bertl k, cya! 1125883355 M * hillct back in an hour or so 1125883925 M * Aiken setattr --iunlink does not work on nfs, this is with a cow kernel both client and server 1125883992 M * Bertl morning Aiken! good point! 1125884013 M * Aiken hi 1125884013 M * Bertl both kernels are vserver patched I assume? 1125884029 M * Aiken both 2.6.13-vs2.1.0-pre5 1125884049 M * Bertl in the .config, is the nfs tagging support enabled (vserver specific) 1125884053 M * Aiken cow works on each machine 1125884095 M * Aiken this is the option? [*] Tag NFSD User Auth and Files 1125884161 M * Bertl yep 1125884214 M * Bertl could you try a 'modified' testfs.sh on some nfs dir? 1125884239 M * Aiken just turned the alpha back on to check, 700m round trip is just too far to walk to turn a machine on 1125884241 M * Aiken yes 1125884246 M * Aiken 70 not 700 1125884291 M * Aiken both machines have that option set 1125884299 M * Bertl is it easy to mount/umount the/a nfs share? 1125884315 M * Bertl (in your setup) 1125884332 M * Aiken yes 1125884346 M * Bertl okay, let me 'adjust' the testfs.sh for nfs testing then :) 1125884373 M * Aiken ----ui- test 1125884373 M * Aiken ----ui- test 1125884387 M * Aiken they are the before and after of setattr --iunlink test 1125884431 M * Bertl NFS from the local host (with a host IP, not 127.0.0.1) is working with a vserver guest? 1125884449 M * Bertl (if you haven't tried yet, could you give it a try?) 1125884467 M * Bertl (not regarding unification, just 'normal' operation) 1125884547 M * Aiken I had been using a guest on nfs with no problem before I started playing with the cow setup 1125884583 M * Bertl yep, but not mounted from the same machine, I gues? 1125884585 M * Bertl *guess 1125884627 M * Aiken no 1125884661 M * Bertl okay, so let's export some dir to your local machine (host ip) and see if that works for a guest (not unified) 1125884674 M * Aiken locally mounting nfs and setattr still fails to set the flags 1125884694 M * Bertl yep, I assumed that, but the rest is supposed to work ... 1125884726 M * Bertl i.e. I want to know if a local NFS mount is compareable to your setup, because if so, we can test by changing only one kernel :) 1125885155 M * Aiken I now have a vserver running on a local nfs share 1125885192 M * Bertl okay, so that works (except for the now known issues :) 1125885245 M * Aiken it seems to work fine 1125885497 M * Bertl okay, what mount options would you suggest for NFS? 1125885581 M * Aiken I normally use the defaults which give me rw,v3,rsize=32768,wsize=32768,hard,udp,lock,addr=fred 1125885591 M * Aiken have not noticed any difference between tcp and udp 1125885648 M * Bertl okay, vers=3,hard,intr,tcp,lock,sync should work then? 1125885734 M * Aiken rw,sync,v3,rsize=32768,wsize=32768,hard,intr,tcp,lock,addr=fred 1125885737 M * Aiken it mounted 1125885739 M * Bertl could you try if a tcp nfs mount from 127.0.0.1 works with the guest too? 1125885822 M * Aiken once I added the appropriate line to exports it worked 1125885835 M * Aiken rw,sync,v3,rsize=32768,wsize=32768,hard,intr,tcp,lock,addr=127.0.0.1 1125885925 M * Bertl okay, so we can use that for testing too, great! 1125886314 Q * litage Ping timeout: 480 seconds 1125886435 M * Bertl Aiken: http://vserver.13thfloor.at/Stuff/SCRIPT/testfs.sh-0.08 (not sure it already works as expected, see -h for options, you have to specify -F nfs 1125886451 A * Bertl is now compiling a kernel with nfs support :) 1125886689 M * Aiken bash testfs.sh-0.08 -F nfs -M /vservers/test/ -N 127.0.0.1:/opt/exports/fred 1125886696 M * Aiken it still tried to mount /dev/zero 1125886711 M * Bertl hmm, right! 1125886719 M * Aiken 13870 pts/7 S 0:00 bash testfs.sh-0.08 -F nfs -M /vservers/test/ -N 127.0.0.1:/opt/exports/fred 1125886719 M * Aiken 13888 pts/7 R 0:04 mount -o vers=3,hard,intr,tcp,sync,rw /dev/zero /vservers/test/ 1125886733 M * Bertl sec 1125886791 M * Aiken I just chnaged to -N to -D 1125886794 M * Aiken got many errors 1125886827 M * Bertl let's try the following change: 1125886833 M * Bertl ( DEV="$NFS" do_test nfs "$nfsopt,$mntopt" ) 1125886838 M * Aiken http://pastebin.com/354899 1125886866 M * Bertl right below the 'nfs)' 1125886906 M * Bertl yes, looks as expected, xattr not supported there ... 1125886970 M * Aiken same results as before 1125887031 M * Bertl okay, so that change works then, what is the base filesystem? 1125887044 M * Aiken ext3 1125887045 J * litage ~nick@203.201.98.84 1125887090 M * Aiken CONFIG_EXT3_FS=y 1125887090 M * Aiken CONFIG_EXT3_FS_XATTR=y 1125887124 M * Bertl wb litage! 1125887164 M * Bertl Aiken: okay, if you try to use chattr +i on some NFS file, does that work? and does it work over an umount/mount cycle? 1125887247 M * Aiken (root@fred) chattr -i test 1125887248 M * Aiken chattr: Inappropriate ioctl for device while reading flags on test 1125887261 M * Aiken (root@fred) chattr +i test 1125887262 M * Aiken chattr: Inappropriate ioctl for device while reading flags on test 1125887275 M * Bertl looks like we found the source of this issue :) 1125887335 M * Aiken limitation of nfs 1125887352 M * Bertl not sure yet, maybe some options are required 1125887367 M * Bertl maybe NFSv4? 1125887409 M * Bertl but, you can try to set the COW flags on the underlying FS 1125887437 M * Aiken cow on the underlying filesystem works 1125887449 M * Bertl yes, but over NFS? the breaking? 1125887476 M * Aiken but not over nfs 1125887554 M * Bertl the breaking fails over NFS or just the setting? 1125887619 M * Aiken the setattr fail over nfs, without those flags set I get no debugging messages from the cow stuff so I assume the link breaking code is not even called 1125887633 M * lilo (hi Bertl!) 1125887647 M * Bertl Aiken: the COW would kick in on the server ... 1125887654 M * Bertl hey lilo! 1125887749 A * lilo waves 1125887766 M * Aiken and no debugging messages from the server 1125887788 M * Bertl okay, let's try the following 'test' setup 1125887806 M * Bertl - NFS loopback on localhost 1125887826 M * Aiken yes 1125887827 M * Bertl - create two hard linked files on the underlying FS 1125887848 M * Bertl - set the properties (iunlink+immutable) on the underlying FS 1125887868 M * Bertl - verify them on the underlying FS 1125887885 M * litage Bertl: wb? 1125887893 M * Aiken yes 1125887899 M * Bertl - write (best with echo >>) to one of them on the NFS mount 1125887904 M * Bertl litage: welcome back :) 1125888011 M * litage ah =P 1125888025 M * Aiken (root@fred) echo zxcv > test 1125888025 M * Aiken su: test: Permission denied 1125888033 A * litage 's setting up linux-vservers for the first time 1125888064 M * Bertl Aiken: okay, I guess I know what happens here, the NFS only 'transmits/checks' the immutable flag ... 1125888092 M * Bertl litage: and? everything working so far? 1125888100 M * Aiken and the host is able to change one of the files and the link breaks 1125888192 M * Aiken once the link is broken the files can then be changed via the nfs mount 1125888233 M * litage Bertl: yup. had a small problem installing the rebuilt kernel on my 2nd machine, but so far all is well 1125888430 M * Bertl Aiken: hmm, but v3 seems to support get/setattr ... 1125888434 M * litage Bertl: when you said "wb litage!" my first thought was "white balance? shwa?" cuz i've been playing with my digicam lately 1125888446 M * Bertl litage: lol! 1125888461 M * Bertl litage: yeah, those funny acronyms :) 1125888498 A * litage lives by acronymfinder.com 1125888580 M * Bertl Aiken: there are NFS debug options (via sysfs) 1125888621 M * Bertl dprintk("NFS call setattr\n"); 1125888635 M * Bertl we are looking for something like this ... 1125888705 M * litage i'm following the instructions at http://deb.riseup.net/vserver/create-instance/ to create my first vserver. does the first step mean to add a new entry to /etc/hosts ? 1125888756 M * Bertl guess that is optional 1125888770 M * Bertl Aiken: seems to be the 'PROC' facility 1125888775 M * litage Bertl: but that's what it's referring to, right? 1125888778 M * Aiken http://pastebin.com/354920 1125888806 M * Bertl Aiken: yeah, looks like it's there ... 1125888818 M * Bertl #ifdef RPC_DEBUG 1125888825 M * Bertl so RPC_DEBUG is required ... 1125888842 M * Bertl #ifdef CONFIG_SYSCTL 1125888842 M * Bertl #define RPC_DEBUG 1125888866 M * Aiken (root@fred) cat /proc/sys/sunrpc/rpc_debug 1125888867 M * Aiken 0 1125888878 M * Bertl yes, and /proc/sys/sunrpc/nfs_debug 1125888886 M * Bertl (as well as the nfsd_debug) 1125888912 M * Bertl forget about the rpc one for now, but enable the nfs(d) ones by 1125888915 M * Aiken so echo 255 to those files? 1125888924 M * Bertl echo 65535 >/proc/sys/sunrpc/nfs_debug 1125888951 M * Bertl litage: yes, seems they add the guest to DNS first 1125888963 M * Aiken that is going to produce lots of data 1125888985 M * Bertl try to keep the NFS I/O low (minimal) 1125889002 M * Bertl best, just do the lsattr/chattr 1125889061 M * Bertl litage: but the steps look sane (create, start, enter) 1125889115 M * Aiken http://pastebin.com/354924 1125889154 M * Aiken http://pastebin.com/354925 1125889191 M * Bertl hmm ... but the chattr failed? 1125889226 M * Bertl I don't see a setattr there ... hmm ... 1125889226 M * Aiken chattr: Inappropriate ioctl for device while reading flags on test 1125889239 M * Bertl maybe we killed it somehow :) 1125889252 A * Bertl is checking the source now ... 1125889272 M * Bertl Aiken: do you have a vanilla kernel (2.6.13) to compare with? 1125889281 M * Aiken no 1125889286 M * Bertl ok, np 1125889323 M * Aiken my latest non vserver kernel is a 2.6.10 with himem 1125889341 M * Bertl hehe, that's the spirit! 1125889370 M * Aiken I tried compiling with nfs4 support and the compile failed :( 1125889382 M * Bertl ah? vserver related? 1125889443 M * Aiken http://pastebin.com/354931 1125889487 M * Bertl hmm, hmm, probably BME related 1125889490 M * Aiken won't take long, I run up a std 2.6.13 1125889502 M * Bertl okay, great! 1125889529 M * Aiken the annoying bit will be reinstalling the nvidia module when I reboot 1125889557 M * Bertl why not use fb? 1125889582 M * Aiken opengl support 1125889591 M * Bertl i.c. 1125889607 M * Aiken what framerate does fb run at? it is setable? 1125889634 M * Bertl yes, you can specify that ... IIRC something like video=1024x768@75 1125889710 M * Aiken last time I used fb was when the voodoo banshee was current, I don't remember the refresh being setable then 1125889727 M * Aiken 60Hz refresh, the headaches :( 1125889737 M * Bertl yeah, I remember too ... 1125889765 J * douglas ~douglas@douglas.user.oftc.net 1125889771 M * Bertl welcome douglas! 1125889772 M * douglas hey hey 1125889776 M * douglas hows it going bertl? 1125889784 M * Bertl fine thanks, and for you? 1125889786 M * douglas wanna help me some more? :) 1125889788 M * douglas hehe 1125889797 M * Bertl sure, always :) 1125889826 M * litage when building a new vserver, what should go after the "/" for --interface?: --interface eth0:172.168.1.101/ 1125889852 M * mugwump hmm, strange - I've got a vserver that isn't depleting sched tokens on 2.6.12.5-vs2.0 / 0.30.208 1125889855 M * Bertl the 'prefix', i.e. /24 for 24 bit netmask (255.255.255.0) 1125889868 M * litage thanks Bertl 1125889875 M * Bertl mugwump: any prio_* flag set? 1125889883 M * Bertl mugwump: and good morning :) 1125889888 M * mugwump which /proc file has those flags? it's set in /etc/vservers/flags.. 1125889908 M * mugwump a good morning to you, too! 1125889932 M * mugwump /etc/vservers/foo/flags of course 1125889966 M * mugwump /proc/virtual/XXX/status? 1125890076 M * Bertl yep 1125890079 M * mugwump Flags: 0000000203070254 # hmm, not set 1125890372 M * mugwump these is no VC_VXF_* macro for bit 1! so you can't set it?! 1125890380 M * mugwump (in the vserver.h in util-vserver, that is) 1125890465 M * Bertl you can set it via ^1 IIRC 1125890884 M * mugwump is there an API call to change those flags on-line? 1125890906 M * Bertl mugwump: sure, vattribute 1125890913 M * Aiken I have a better look after I have rebooted but the std kernel compiled fine with nfs4 support 1125890935 M * Bertl Aiken: okay! 1125890939 Q * Aiken Quit: Leaving 1125891156 J * Aiken ~james@tooax6-165.dialup.optusnet.com.au 1125891164 M * Bertl wb Aiken! 1125891173 M * Aiken so far so good :) 1125891188 M * douglas bertl do you recommend setting hard limits? 1125891202 M * Bertl hard cpu scheduler (sched_hard)? 1125891210 M * douglas yea 1125891224 M * Bertl depends on your usage pattern (and the guests of course) 1125891243 M * Bertl in a potentially hostile environment, it is suggested to do so 1125891267 M * douglas and setting sched_hard is just editing flags file and putting sched_hard in? 1125891274 M * Bertl if you know what the guest will be running, the sched_prio is more resource efficient 1125891298 M * Bertl douglas: yes, and restarting the guest (or alternatively setting it via vattribute) 1125891427 M * litage i'm building my first vserver, but a strange error is occuring: http://rafb.net/paste/results/fefsaQ85.html 1125891461 M * litage n/m, got it! :) 1125891494 M * mugwump strange ... still no tokens being used 1125891516 M * Bertl mugwump: testme.sh output? 1125891578 M * mugwump all succeeded... 1125891585 M * Bertl first 5 lines? 1125891606 M * douglas ok bertl I set that up and the cpu is still taking 100% 1125891615 M * douglas the guest is leeching 100% of the cpu 1125891624 M * mugwump http://nopaste.snit.ch:8001/4118 1125891636 M * Bertl douglas: could you also run the testme.sh and upload that somewhere? 1125891646 M * mugwump is that the right testme.sh? 1125891655 M * Bertl looks good! 1125891685 M * douglas http://pastebin.com/354963 1125891701 M * douglas brb real quick 1125891751 M * Bertl douglas: you have not enabled the hardcpu scheduler in the kernels .config 1125891793 M * Bertl mugwump: what does the flags section show/list now? 1125891818 M * mugwump nproc ulimit namespace ^1 sched_prio virt_mem virt_uptime virt_cpu hide_mount hide_netif 1125891830 M * mugwump perhaps some of those conflict? 1125891844 M * Bertl nproc and ulimit is obsolete 1125891924 M * mugwump ok, I've turned those off via vattribute 1125891946 M * Bertl how do you check the TB? 1125891955 M * mugwump /proc/virtual/xid/sched 1125891962 M * Bertl and do you ahve something running inside? 1125891968 M * mugwump yep, a 1 while 1 1125891990 M * Bertl okay, give me a minute, I'll try to recreate this here ... 1125892082 M * Aiken I get the same ioctl errors with a std 2.6.13 1125892125 M * Bertl Aiken: okay, great! could you strace a chattr +i over NFS? 1125892126 M * litage when a vserver is being built and it says "I: Retrieving exim4", does that mean the exim4 package is being downloaded? 1125892135 M * litage or is it just being installed from the host? 1125892149 M * Bertl yes, it uses the 'usual' debootstrap pathes 1125892164 M * Bertl so if you have a local cache/repository that will be used 1125892181 M * douglas wheres that located bertl? 1125892207 M * Bertl CONFIG_VSERVER_HARDCPU=y 1125892210 M * litage Bertl: and what if the only repos in /etc/apt/sources.list are Debian CDs 1-3, and security.debian.org? 1125892217 M * douglas where in make menuconfig I mean hehe 1125892238 M * douglas or is it not? 1125892254 M * Bertl press '/' and enter 'HARD' 1125892306 M * Bertl Depends on: EXPERIMENTAL 1125892320 M * Bertl (guess we should remove that dependancy soon :) 1125892328 A * mugwump sighs at the gentoo-specificness of libvserver 1125892369 M * douglas I'm in linux_vserver where it says but I dont see that at all 1125892381 M * litage is there a way to tell debootstrap to use CDs rather than an internet repo? 1125892399 M * mugwump douglas: so, you have to go back up to the "general" section and enter "yes" to the experimental stuff question 1125892401 M * Bertl douglas: yes, you have to enable EXPERIMENTAL to get it shown 1125892407 M * douglas http://pastebin.com/354967 1125892417 M * douglas experimental where? 1125892429 M * Bertl litage: you have to ask the debian folks about that ... 1125892469 M * mugwump litage: you should be able to use file:///cdrom/ or something like that 1125892471 M * Bertl douglas: Code maturity level options 1125892487 M * Bertl Prompt for development and/or incomplete code/drivers 1125892513 M * douglas if I enable hard_cpu should I enabel the idle task one as well? 1125892539 M * litage mugwump: but can you specify multiples? 1125892547 M * Bertl not really required, unless you want to see excellent test results, in exchange for more overhead :) 1125892574 M * mugwump litage: I doubt it. you'd probably need to combine the CDs on your hard drive somewhere 1125892584 M * douglas well it says marginally, so not much over head then that it adds 1125892598 M * mugwump but you only need disc 1 for debootstrap 1125892603 M * douglas do I have to enable anything to limit the memory? 1125892610 M * Bertl nope 1125892634 M * douglas ok and the schedule file I created had the memory limits in there or so that another file? 1125892698 M * Bertl the memory limit goes to the other 'rlimitsÄ 1125892705 M * Bertl s/Ä/'/ 1125892711 M * douglas ahh I see 1125892722 M * Bertl http://linux-vserver.org/Resource+Limits 1125892746 M * douglas yea I see 1125892746 M * Bertl the page size on x86 is 4k and 16k on x86_64 IIRC 1125892790 M * douglas so to limit to 128 megs would be what? 4/1280000 ? 1125892820 M * Bertl 128*1024*1024/4096 1125892827 M * Bertl or 32768 :) 1125892847 M * douglas ahh K and the as should be what 65000? 1125892870 M * douglas or more? like 80000? 1125892871 M * Bertl depends on what you want for your guests ... 1125892889 M * Bertl a 'normal' linux system has no such limit :) 1125892908 M * douglas AS would be like swap space right? 1125892924 M * Bertl the best approach is to test with a guest, and look at the 'current' and 'max' usage ... 1125892947 M * Bertl then, if you see what is used, you double or triple that, and set it as limit 1125892984 M * douglas memlock should be one? 1 in the file? or does that file actually contain a real number? I'm a bit confused heh :) 1125892996 M * Bertl you can say that AS >= swap+ram >= ram >= RSS 1125893021 M * Bertl memlock is in pages too 1125893041 M * douglas what does memlock actually do? 1125893062 M * Bertl userspace might decide to 'lock' pages into memory (i.e. make them unswapable) 1125893078 M * Bertl this is an upper limit to the number of such pages 1125893091 M * douglas locks pages into memoryb ut if I define RSS and AS do I really care how many pages can be locked into memory? 1125893098 M * douglas in a hostile system would I define this as well? 1125893144 M * Bertl a normal linux system has something like RSS/3 or so for MEMLOCK 1125893147 M * mugwump Bertl, which part of kernel/sched.c#scheduler_tick() actually takes the tokens? 1125893179 M * douglas great, I'm recompiling the kernel to support hard schedule. 1125893368 M * douglas so once this is all done, I could technically run a stress program to grab as much memory as possible from the guest and it should only be allowed to grab rss + as correct? 1125893399 M * Bertl nope, it will fail at RSS (unless the memory is swapped out) 1125893488 M * Bertl mugwump: kernel/sched.c 2625 vx_need_resched() 1125893491 M * douglas so it should only be able to grab 128 megs of ram? what is AS used for then? maybe I'm confused 1125893507 Q * Aiken Quit: Leaving 1125893516 M * Bertl AS is the address space available (in sum) to the guest 1125893525 M * mugwump Bertl, so, CONFIG_VSERVER_HARDCPU needs to be defined for any tokens to be deducted at all. :) 1125893534 M * Bertl mugwump: no, why? 1125893545 M * mugwump include/linux/vs_sched.h#vx_need_resched() 1125893558 M * douglas so if I config RSS to 128, then as should be what exactly? 128? if all I want to let it have is 128 megs of phsyical memory? 1125893570 M * mugwump that vx_consume_token() call is inside the #ifdef 1125893581 M * Bertl ah, right! 1125893593 M * douglas that to me? 1125893626 M * Bertl no, usually I prefix if not answering to the previous line :) 1125893655 J * Aiken ~james@tooax6-165.dialup.optusnet.com.au 1125893684 M * Bertl mugwump: maybe we should add a separate TOKEN_BUCKET option? 1125893687 M * douglas ok so I'm confused what as should be? heh 1125893695 M * douglas err AS should be 1125893705 M * Bertl douglas: on a normal linux system it's 3GB 1125893720 M * Bertl douglas: you get the same if you leave it unlimited 1125893758 M * douglas so what does AS allow the guest to do? throttle? I mean if RSS is 128megs whats the purpose of AS? telling the machine how much actual memory there is? 1125893796 M * Bertl for example, java will allocate address space, then fill this space with actual memory pages on demand 1125893856 M * Bertl if you allow java to get 1GB memory, but limit the RSS to 128M, then there might be a situation where java did allocate 1GB space (address wise) but fails after instatiating more than 128MB 1125893875 M * Bertl (this is known as the OOM case) 1125893906 M * douglas ahh 1125893909 M * douglas gotcha 1125893962 M * Bertl btw, you can also set the AS/RSS limits on a per process basis with the ulimits 1125894037 J * lilo_ ~lilo@lilo.usercloak.oftc.net 1125894131 M * douglas I gotcha 1125894157 M * douglas now will the vserver use swap after 128megs of phsyical memory? or is RSS not care if its phsyical or swap space? 1125894204 M * Bertl RSS is just physical pages, swap will be used when the host decides to swap 1125894232 M * litage when building a vserver and wanting to specify a mirror for debootstrap, where would "file:///cdrom/debian" go? 1125894234 M * litage i don't see how the args that vserver accepts allows you to specify the mirror, since debootstrap takes args like this: "options suite target mirror" but you can't specifically give vserver a mirror arg 1125894284 M * Bertl http://linux-vserver.org/alpha+util-vserver 1125894299 M * Bertl (see Examples section) 1125894305 M * litage looking.. 1125894309 M * Bertl * -- -d sarge -m ftp://ftp.debian.org/debian 1125894477 Q * lilo Ping timeout: 480 seconds 1125894567 M * douglas so rss doesn't care, that kinda tough, cuz lets say I only want to give the guest 128megs of phsyical ram but 512megs of swap space 1125894577 M * douglas doesn't quite work that way 1125894593 M * Bertl yes, because it would not make sense for the resource sharing part :) 1125894623 M * douglas gotcha 1125894624 M * Bertl just consider a guest which has 64MB limit (RSS) but 128MB 'swap' 1125894650 M * douglas I'm considering what would you put in the RSS and AS files then? 1125894653 M * Bertl it would bring down the host easily by using 128MB and swapping all the time 1125894661 M * douglas true 1125894669 M * douglas hmm 1125894674 M * Bertl that's not what you want with linux-vserver 1125894679 M * douglas its a tough situation either way you go. 1125894692 M * douglas cuz I dont want to give full memory to the vserver cuz it can do the same thing then 1125894703 M * Bertl if the host is short on real memory, it will swap out host and guests 1125894714 M * douglas it does this automaticly? 1125894725 M * Bertl (this swap decision requires some optimizations in the future, but yes, it is done automatically) 1125894743 M * Bertl for example, we will prefer guests over a soft limit for swapout 1125894762 M * Bertl (or to be precise, pages of guests over a soft limit :) 1125894819 M * Bertl you might also overbook the RSS-max it will lead to automatic swapout when required 1125894843 M * douglas rss file has a min and max number? 1125894846 M * douglas I though its just one number 1125894865 M * Bertl actually it has min, soft and max, but for now, only max is honored 1125894869 M * douglas so instead of 36000 (the number you said earlier) maybe make it 50000 even? 1125894903 M * douglas which technically is overbooking the RSS 1125894914 M * Bertl as I said at the beginning, best check with a real guest, and see how much it actually uses 1125894968 M * litage thanks Bertl, that helped a ton! 1125894976 M * douglas roger 1125894981 M * Bertl litage: you're welcome! 1125895076 M * litage first vserver being built right now. i'm seeing a lot of pre-dependency problems during pkg configuration. is that normal? 1125895154 M * Bertl my last guest build went without any unusual events, but it is plain debootstrap/debian stuff so if that is normal or not, I can not answer 1125895233 M * litage hrmm, building died. any suggestions?: http://rafb.net/paste/results/FhpeVW46.html 1125895241 M * Bertl mugwump: I guess once we split up the token buckets per cpu we add an option ENABLE_TOKEN_BUCKET(S), which does the accounting ... 1125895279 M * Bertl litage: why did unpacking of those two packages fail? 1125895301 M * litage Bertl: i don't know. how can i find out? is there a log i can check? 1125895323 M * Bertl well, the dpkg output probably gives some clues, but I only see the tail ... 1125895346 M * Bertl (maybe scrolling up might help :) 1125895398 M * Bertl Aiken: I fixed the nfsv4 compile issues, they are not hard to fix, just add ,NULL to each vfs_* call there 1125895425 M * Bertl (each vfs_* call listed as error, that is :) 1125895518 M * Aiken I gave up trying to get nfs4 working with the std kernel 1125895525 M * Bertl okay :) 1125895550 M * Aiken I am missing something and to day I just am not seeing what it is 1125895572 M * Bertl k, let's focus on v3 for now 1125895598 M * Aiken the ioctl errors with chattr and lsattr are the same with a vserver kernel and a std kernel 1125895613 M * Bertl yes, but maybe we can 'improve' that :) 1125896210 M * mugwump hmm, odd. a vserver has (1/4) CPU, is using 100% cpu but tokens are fluctuating near full 1125896269 M * mugwump this isn't some weird variable HZ is it? 1125896307 Q * Hollow Ping timeout: 480 seconds 1125896313 M * Bertl mugwump: no idea ... 1125896329 J * Hollow ~Hollow@home.xnull.de 1125896334 M * Bertl mugwump: you're the scheduler expert, no? :) 1125896356 M * mugwump clearly not 1125896393 M * mugwump I thought that the feature was being used a bit more 1125896426 M * mugwump Hollow, I've got problems trying to compile your libvserver 1125896436 M * mugwump want me to nopaste them? 1125896474 M * Bertl mugwump: all the folks use is the hard scheduler ... 1125896510 M * mugwump surely the hard scheduler must suffer from the same problem here. 1125896535 M * Bertl well, check it out ... it works quite fine in my tests ... 1125896557 M * Bertl (worked to be precise, didn't test it for a few weeks now) 1125896626 M * mugwump the tokens fluctuate, but are replenished far too quickly. so, the effect is the same no matter which scheduler is used. 1125896629 A * mugwump hmms 1125896652 M * litage "I: Base system installed successfully." was printed, but a lot of pre-dependency problems like this occured: http://rafb.net/paste/results/0QfUHM47.html 1125896655 M * Bertl maybe some x86_64 specific issue? 1125896658 M * litage is the new vserver safe/stable to use? 1125896700 M * Bertl I guess so, just fire it up and see if everything works ... 1125896751 M * litage heh 1125896788 M * litage hrmmm, /etc/network/interfaces wasn't setup. is that normal? 1125896821 M * Bertl litage: yes, because networking happens on the host 1125896838 M * Bertl (basically all hardware related setup happens on the host) 1125896874 M * litage Bertl: not sure what you mean by "happens on the host" 1125896908 M * litage Bertl: as in the host brings up or shuts down a vserver's NIC? 1125896954 M * Bertl yep, you can look at it this way ... as a guest has no dedicated nic (at least not usually) 1125897111 M * litage danke 1125897225 M * Bertl bitte (you're welcome :) 1125897353 M * litage aber naturlich 1125897398 M * litage i setup my vserver's /etc/network/interfaces file as i would on a normal machine. should i undo that? 1125897429 M * litage (interfaces/0/dev and /ip and /prefix are setup on the host) 1125897474 M * Bertl you should avoid all hardware related setups, but they will not succeed anyway, so except for a bunch of errors, it will work with or without 1125897499 M * Bertl you configure the guest IP(s) from outside 1125897524 M * Bertl (either with the config or at build time with --interface.. 1125897569 M * litage Bertl: i configured it with --interface, but pinging the vserver fails 1125897584 M * Bertl did you start it? 1125897600 M * litage Bertl: yes, and entered it 1125897610 M * Bertl and from where do you ping it? 1125897634 M * litage Bertl: a remote machine 1125897680 M * Bertl okay, can you upload the /etc/vserver//interfaces files? 1125897697 M * Bertl (or comment/paste) them here 1125897775 A * mugwump puts some debugging into kernel/sched.c and sighs at how much must be rebuilt 1125897807 M * Bertl litage: find /etc/vservers/test102/interfaces/ -print -type f -exec cat {} \; 1125897819 M * Bertl replace test102 by your name ... 1125897832 M * litage dev: eth0 ip: 172.168.1.101 prefix: 24 1125897854 M * Bertl this is in interfaces/0, right? 1125897857 M * litage yes 1125897874 M * Bertl okay, does 'ip addr ls' show that ip? 1125897890 M * Bertl litage: btw, maybe 172->192? 1125897909 M * litage Bertl: yes: http://rafb.net/paste/results/NSwnGQ40.html 1125897927 M * litage Bertl: i'm using 172.168.1.0/24 in this lan 1125897943 M * Bertl aha, public IPs? 1125897969 M * Bertl okay, and a ping to that doesn't work on your other machine? 1125898002 M * Bertl this means that your ping gets blocked somewhere ... use tcpdump to verify that (on both machines) 1125898006 M * Aiken 172.168.1.101 belongs to aol 1125898038 M * litage Bertl: correct. pinging 172.168.1.101 from 172.168.1.15 fails 1125898048 M * litage Aiken: ?? 1125898071 M * Bertl litage: use tcpdump -vvnei eth0 icmp (on both machines the host and the remote machine) 1125898106 M * Bertl Aiken, litage: yeah, I assume it should be 192.168 but some error was propagated and institutionalized :) 1125898111 M * litage Bertl: ooh wait a sec. i've got iptables running on the vserver host 1125898135 M * Aiken I like the 172.16.x.x - 172.31.x.x range 1125898164 J * dddd44 ~dhb55@tor-irc.dnsbl.oftc.net 1125898165 M * Aiken litage the ip 172.168.1.101 is in range that is allocated to AOL 1125898175 M * Bertl welcome dddd44! 1125898414 M * litage Bertl: heh, my bad. firewall shenanigans =) 1125898481 M * litage Aiken: that's not correct. see section 3: http://www.faqs.org/rfcs/rfc1918.html 1125898496 M * Aiken yes it is 1125898509 M * litage Aiken: sorry, you're right. hahha oops again 1125898521 M * Aiken 10.0.0.0 - 10.255.255.255 (10/8 prefix) 1125898522 M * Aiken 172.16.0.0 - 172.31.255.255 (172.16/12 prefix) 1125898522 M * Aiken 192.168.0.0 - 192.168.255.255 (192.168/16 prefix) 1125898525 M * litage Aiken: ah well, i don't need aol anyways 1125898792 M * Aiken time to straddle a 1000cc motor, I am out of here for an hour 1125898834 M * Bertl I'm off to bed soon, have fun! 1125898851 M * Aiken oh I will :) 1125898853 M * Bertl Aiken: and thanks for the testing! 1125898891 M * Hollow morning 1125898902 M * Bertl hey Hollow! 1125898922 M * Hollow hey Bertl 1125898967 M * Hollow mugwump: still there? 1125898971 M * litage thanks for your help, Bertl 1125898991 M * Bertl litage: you're welcome! 1125899070 Q * dddd44 Quit: Leaving 1125899322 M * mugwump hi 1125899348 M * mugwump Bertl, I think that there is 32bit vs 64bit badness going on 1125899362 M * mugwump in particular in sched_def.h 1125899369 M * Bertl mugwump: could be, could be ... 1125899416 M * litage when this vserver starts, it says "secure-mount: chdir("/vserver/camel/cdrom"): No such file or directory" when trying to mount host:/mnt/cdrom0 into vserver:/cdrom, but /vserver/camel/cdrom exists 1125899446 M * Bertl litage: probably not inside the guest :) 1125899494 M * Bertl litage: forget it, it's simpler: /vserver/camel/cdrom -> /vservers/camel/cdrom 1125899524 M * Bertl (spot the _s_) 1125899935 M * litage Bertl: actually, i named it /vserver/ because that's more standard than /vservers/ 1125899972 M * Bertl aha ... okay :) 1125900007 J * lilo debian-tor@lilo.usercloak.oftc.net 1125900025 M * Bertl well, I'm off to bed now ... will be back when I get up again ... 1125900044 M * Bertl so have a good whatever everyone and cya later ... 1125900050 M * litage Bertl: vserver complains "chdir("/vserver/camel/mnt/cdrom0"): No such file or directory" buuuut: `ls -l /vserver/camel/mnt/` returns "cdrom0/" 1125900060 M * litage hah thanks Bertl. cya later 1125900083 N * Bertl Bertl_zZ 1125900105 Q * lilo_ Ping timeout: 480 seconds 1125900153 M * mugwump hey, I have a vserver called camel too 1125900163 M * mugwump have you haxx0red my box litage ? :) 1125900185 M * Hollow litage: bay you try to access /vservers/camel/mnt/cdrom0 inside your guest..? 1125900205 M * Hollow s/bay/maybe/ 1125900206 M * Hollow :) 1125900285 M * Hollow mugwump: did you solve your problems with libvserver? 1125900372 M * mugwump well, I think the main problem is that getting the right headers file is problematic 1125900416 M * mugwump ie, it couldn't find things like uint32_t, which I think are new between 2.6.0 (which are in /usr/include on my system) and 2.6.12 (only in /lib/`uname -r`/build) 1125900447 M * mugwump as standard best practice is *not* to deliver these files to /usr/include 1125900480 M * mugwump unless I make the good old /usr/include/linux symlink 1125900492 M * mugwump but this is a nasty hack 1125900577 A * mugwump reboots into a new kernel... brb 1125900679 M * litage Hollow: /media/cdrom0/ is mounted on the host. /vserver/camel/media/cdrom0/ exists (camel is the vserver) 1125900718 M * Hollow mugwump: try to get libvserver from svn, it's solved there 1125900724 Q * Hollow Remote host closed the connection 1125900758 J * Hollow ~Hollow@home.xnull.de 1125900762 M * mugwump Hollow: svn url? 1125900766 M * Hollow that's odd 1125900772 M * Hollow i suddenly got logged out 1125900774 M * Hollow :P 1125900783 M * Hollow http://dev.croup.de/repos/libvserver/trunk/ 1125900813 M * Hollow you have to run "make -f Makefile.cvs" first 1125901301 J * michal ~michal@michal.usercloak.oftc.net 1125901329 Q * monrad Quit: Leaving 1125901597 M * litage Hollow: when i start the vserver, despite the paths existing on the host and vserver, it says "chdir("/vserver/camel/media/cdrom0"): No such file or directory" 1125901650 M * Hollow litage: the question is where it tries to access /vserver/camel/media/cdrom0, if it tries in the guest, for sure there won't be such a patch, if it tries on the host, than that's odd ;) 1125901668 M * Hollow s/patch/path/ 1125901913 M * litage Hollow: why wouldn't vserver:/media/cdrom0 exist for guest? 1125901949 M * Hollow that would exist, but not /vservers/camel/media/cdrom0... where did you specifiy this mountpoint? 1125901980 M * litage i created /vserver/camel/media/cdrom0/ on the host 1125902042 M * litage this is the vserver's fstab entry: /media/cdrom0 /vserver/camel/media/cdrom0 auto rbind 0 0 1125902073 M * Hollow yep, this tells util-vserver to mount the hosts /media/cdrom on the guests /vservers/camel/media/cdrom0 1125902079 M * Hollow but the guest has no /vserver/ 1125902092 M * Hollow the entry should look like: 1125902103 A * Hollow is playing ♫ Global Underground #001 ♫ by Tony De Vit (Tel Aviv) [amaroK] 1125902106 M * Hollow uh? 1125902113 M * Hollow i found some macro 1125902115 M * Hollow *gg* 1125902121 M * Hollow /media/cdrom0 /media/cdrom0 none rbind 0 0 1125902159 M * litage Hollow: zing! works! thanks dude 1125902164 M * Hollow np :) 1125902178 M * Hollow i was a bit confused to of that behaviour ;) 1125902268 M * Hollow let's see where this amarok macro comes from.. 1125902508 M * litage Hollow: do you know what the rbind option does? 1125902522 M * Hollow recursive bind 1125902539 M * eyck what is 'rbind' ? 1125902549 M * Hollow s.o. 1125902552 M * Hollow :) 1125902603 M * eyck ... The entire file hierarchy including submounts, ok, that explains it. 1125902849 M * litage so the dvd-rom is mounted on the host, and the vserver can read the contents (/media/cdrom0) 1125902957 M * litage however, when i run apt-setup within the vserver, selecting cdrom or filesystem and pointing it to /media/cdrom0 fails 1125903026 M * litage any idea how i can fix this? 1125903158 M * Hollow litage: any error message? 1125903404 M * litage Hollow: think i just got it working with apt-cdrom rather than apt-setup. says it's scanning the disc now.. 1125903416 M * Hollow k 1125903487 M * litage awesome! works! no idea why apt-setup didn't like /cdrom though.. 1125903589 M * Hollow maybe normal and cd repos are different at debians? 1125903593 M * Hollow i dunno ;) 1125903936 M * mugwump right. it was 32 vs 64 bit badness, after all 1125903949 M * mugwump you'd only notice it after calling vsched, I think 1125903972 M * mugwump Bertl_zZ, ... oh, you're asleep 1125904067 M * Hollow mugwump: which 32/64 madness? 1125904133 M * mugwump in the sched code ... lots of `int', `long', etc, and the odd `uint32_t' just to make sure that you get corner cases 1125904137 A * mugwump reboots again 1125905831 A * mugwump sighs 1125906347 M * mugwump now my silly pc won't start X. 1125906357 A * mugwump curses nvidia and their proprietary drivers 1125906400 M * litage if the firewall on the vserver host is setup to not forward packets and you try to ssh into the vserver, what happens? 1125906443 M * mugwump forwarding isn't a problem, it's all local traffic 1125906509 M * mugwump maybe if you are also using NAT it is considered forwarding. I long stopped caring, after finding fwbuilder :) 1125906564 M * litage thanks mugwump 1125906605 A * mugwump lathers, rinses, repeats 1125908350 J * Aiken_ ~james@tooax7-119.dialup.optusnet.com.au 1125908665 Q * Aiken Ping timeout: 480 seconds 1125908906 J * prae ~prae@gut75-1-81-57-27-189.fbx.proxad.net 1125909599 A * Hollow loves fwbuilder too, great app 1125909897 J * dddd44 ~dhb55@tor-irc.dnsbl.oftc.net 1125909909 Q * dddd44 Remote host closed the connection 1125909931 J * dddd44 ~dhb55@218.111.177.154 1125909931 Q * dddd44 Quit: 1125912374 Q * Topper Quit: Leaving 1125913444 Q * cryo Remote host closed the connection 1125913588 J * cryo ~say@212.86.243.154 1125915157 M * Vudumen Hollow: are you there? 1125915205 J * Neubix ~brian@p54B03578.dip.t-dialin.net 1125915224 J * Elch ~root@vserver2.toppoint.de 1125915227 M * Elch Hi 1125915275 M * Elch I will try the new 2.6 step by step guide for my new vserver :-) 1125915655 M * Neubix Hi Elch .. I have write a step by step HowTo in german 1125915743 M * Elch Neubix cool, where can I find it? 1125915748 M * Neubix http://www.neubix.de/vserver/ 1125915834 M * Elch Ah and yes I want more then 8 vserver :/ 1125915850 M * Neubix no Problem .. 1125915866 M * Elch Na its more work 1125915901 M * Neubix read the Howto .. you can setup more then 8 .. I have 20 vserver current running on one Host 1125915969 M * Elch Nice, I do that for my club (Buergernetzverein) and thing more then 8 ppl want one 1125916013 M * Elch Can I use 12.5 ? 1125916018 M * daniel_hozac Elch: you shouldn't IRC as root ;) 1125916035 M * Elch daniel I know .. :( 1125916106 M * eyck why not? 1125916462 J * brian_ ~brian@p54B044C0.dip.t-dialin.net 1125916541 M * Elch Neubix? 1125916583 M * brian_ yes elch ?? 1125916596 M * brian_ ups .. 1125916599 M * brian_ new Name .. 1125916613 M * brian_ my internet connectin was down .. 1125916626 M * brian_ I'm Neubix ... 1125916646 M * Elch ah ehh I have a problem with your step by step guide 1125916677 M * brian_ problem ?? 1125916703 M * brian_ mom, telephone ... 1125916707 M * Hollow Vudumen: yep 1125916809 M * Elch brian yeah there is a list after make oldconfig ... what does that mean? 1125916879 A * HostingGeek hugs Neubix 1125916879 Q * Neubix Ping timeout: 480 seconds 1125917531 Q * Aiken_ Quit: Leaving 1125918031 M * brian_ sorry, I'm back 1125918101 M * brian_ type make oldconfig with your current .config file and a patched kernel source .. the "make oldconfig" will ask you this .. 1125918314 M * Elch no the table after that 1125918335 M * Elch and I have no oldconfig, new Debian on that system 1125918759 M * brian_ you arb meening the images .. this is the screen of "make menuconfig" if you don's have a .config file you need this command 1125918785 Q * brian_ Quit: Verlassend 1125918798 J * brian_ ~brian@p54B044C0.dip.t-dialin.net 1125918824 Q * brian_ Quit: 1125918846 J * Neubix ~brian@p54B044C0.dip.t-dialin.net 1125918918 M * Elch brian/neubix aha 1125918951 M * Elch and I did a make deb-pkg is that ok? 1125918978 M * Neubix yes, back as Neubix :-) 1125919020 M * Neubix I don't test it .. try it .. 1125919616 M * Elch Oki.. 1125919659 M * Elch btw why I have do write down the vroot in grub? Without that I cant start more then 8 vserver? 1125920454 M * Hollow Elch: because per default you only have 8 vroot devices 1125920469 J * menomc ~amery@200.75.27.30 1125920495 M * Hollow and it seems you're using quota/disk limit on more than 8 vservres ;) 1125920577 Q * mnemoc Ping timeout: 480 seconds 1125920577 N * menomc mnemoc 1125921166 M * Elch Hollow since I never ever used quota I wont use it with vservers. 1125921191 M * Hollow well, you won't need vroot then anyway 1125921242 M * Elch Ah good. So I can make 253 Vservers in my /24 ? 1125921268 M * Hollow yup 1125921318 A * HostingGeek hugs Neubix 1125921342 M * Neubix Hugs Geek :) 1125921355 M * HostingGeek * HostingGeek hugs Neubix 1125921356 M * HostingGeek <-- Neubix has quit (Ping timeout: 480 seconds) 1125921370 M * HostingGeek like a few seconds after I huged you 1125921374 M * HostingGeek you timed out 1125921389 M * Neubix yes, my internet connection was down ... 1125921457 M * HostingGeek my huging power release the power of timing out 1125921462 A * HostingGeek hugs lilo 1125921466 M * HostingGeek muwahahaha 1125921537 M * Vudumen Hollow: you work on this libvserver. i think it would be better if you name it libvserver2 because util-vserver contains a libvserver already 1125921560 M * Vudumen the other thing that am i right that i've read that you are working on a new vserver script? 1125921574 M * Hollow the library already has a different version number, so you should be able to use both 1125921601 M * Hollow define "new vserver script" 1125921628 M * Vudumen i mean the script what is used to start/stop/create/etc vservers 1125921640 M * Hollow right 1125921654 M * Vudumen is it available for the public? :) 1125921666 M * Hollow http://dev.croup.de/proj/vserver-utils/ 1125921672 M * Hollow no releases yet, use svn 1125921677 M * Vudumen okay. thanks :) 1125921690 M * Hollow but the tools work quite well 1125921697 M * Vudumen thanks :) 1125921716 M * Hollow but i think you won't be able to do much with them though ;) 1125921718 M * Vudumen ooo just a question. if i use your libvserver and the utils included. do i have to enable legacy networking? 1125921750 M * Hollow Vudumen: you can't even start a vserver with these utils 1125921758 M * Hollow well, you can... 1125921768 M * Hollow but only when you really know what you're doing ;) 1125921803 M * Vudumen Hollow: well my problem currently that when i use enricos util-vserver my debian systems are debootstrapped in the default context with default hostname etc... 1125921812 M * Vudumen so now i'm looking for other solution 1125921841 M * Hollow Vudumen: svn only includes basic tools (vcontext, vnamespace etc) there is no distro pkg management, no start/stop/enter no everything 1125921886 M * Vudumen Hollow: fine :) if i can figure it out how to start a vserver maybe i try to write an application for this :) 1125921910 M * Hollow again, with the source in svn you won't be able to start a vserver 1125921944 M * Hollow you can create contexts, and scheduler settings etc, but you won't be able to start a vserver like you are used to it ;) 1125921946 M * Vudumen ok :) 1125921968 M * Vudumen Hollow: one more quiestion. how the hell can i checkout this? :)))) i cannot find here 1125921977 M * Hollow the tools are merely a reference implementation of libvserver 1125922012 M * Hollow ah well... the vserver-utils wiki doesn't contain svn info yet... http://dev.croup.de/repos/vserver-utils/trunk is the svn uri 1125922061 M * Vudumen thanks :) 1125922091 M * Hollow in my local copy i have some changes to make vserver semi-start, but it's just playing around.. want a diff? 1125922119 M * Vudumen would be fine :) 1125922160 M * Hollow http://home.xnull.de/misc/vserver-utils-vcontext-fifo.patch 1125922301 M * Hollow the fifo approach is for eliminating the need to chain every syscall 1125922489 J * OD-100 od100@192.116.240.34 1125922517 M * Hollow off for now, back later 1125922522 M * OD-100 helo 1125922540 M * OD-100 how do i list files in directory in linux? 1125922548 M * OD-100 (similar to dir in windows) 1125922549 M * OD-100 ? 1125922579 M * Hollow sure you're on the right channel? ;) 1125922591 M * Hollow ls 1125922878 Q * maharaja Read error: Connection reset by peer 1125922912 M * OD-100 Hollow. :P i was just trollin 1125922993 M * OD-100 whats the lastestr util-vserver? 1125923566 Q * Loki|muh Ping timeout: 480 seconds 1125923752 J * Loki|muh loki@satanix.de 1125924093 M * Elch should I use the debian vservertools? 1125924184 M * Neubix my Howto is only for the new version 2.0 and the current vserver util util-vserver-0.30.208.tar 1125924243 Q * lilo Remote host closed the connection 1125924265 J * maharaja maharaja@ip52.ipax.at 1125924521 M * Elch Neubix I currently use the old one and there is a newvserver script... how do I make a new vserver then? 1125924698 M * Neubix I don't know .. I begin with the new version. try --help :) 1125924767 M * Elch oki bbl now I will reboot ... 1125924781 Q * Elch Quit: I hope my vserver will work 1125925009 M * Vudumen Hollow: the following lines should be written to linux/vserver/*_cmd.h as is see. is there any other way for this? i don't think that these lines should appear in every file instead of one 1125925013 M * Vudumen #include 1125925016 M * Vudumen #include 1125925046 M * Vudumen (except namespace_cmd.h) 1125925093 M * HostingGeek ya use windows? 1125925095 M * HostingGeek what the point of a reboot? 1125925107 Q * flock Ping timeout: 480 seconds 1125925364 M * Neubix Elch need a very long reboot :) 1125926119 Q * OD-100 Quit: 1125927020 Q * BobR_oO Quit: leaving 1125927182 Q * Neubix Quit: Verlassend 1125927484 J * BobR ~georg@212.16.62.52 1125927558 J * Blissex ~Blissex@82-69-39-138.dsl.in-addr.zen.co.uk 1125927862 N * BobR BobR_afk 1125927895 M * Vudumen Hollow: i've updated the libvserver debian packages to 0.3 1125927935 M * Vudumen Hollow: and the renaming of the lib was suggested because of this: 1125927936 M * Vudumen dpkg: error processing libvserver_0.3-1_i386.deb (--install): trying to overwrite `/usr/lib/libvserver.so', which is also in package util-vserver 1125928618 N * BobR_afk BobR 1125928724 M * HostingGeek just --force it 1125928727 M * HostingGeek and report a bug 1125929064 M * Greek0 HostingGeek: libvserver isn't in the official debian archive AFAICS, but a Conflict: line with util-vserver would probably be a good idea anyway 1125929085 M * HostingGeek Greek0: no 1125929098 M * HostingGeek Greek0: just make one of them include the file 1125929100 N * BobR BobR_oO 1125929106 M * HostingGeek Greek0: and make the other depend on it 1125929119 M * HostingGeek Greek0: or make a new package with the lib files 1125929128 M * HostingGeek Greek0: and make both of them depend on it 1125929156 M * micah huh, odd... vserver-stat did not show my vserver 'admin' as running, but if I tried to start it with vserver admin start it said it was running 1125929169 M * micah so I stopped it and started it, and its there now 1125929176 M * Greek0 HostingGeek: uhm. this is not the same libvserver.so file AFAICS 1125929214 M * Greek0 the libvserver.so file in libvserver is the main product of the project. whereas libvserver.so is an implementation detail in enrico's util-vserver. 1125929245 M * Greek0 unless I've missed something (i.e. enrico switched to using hollow's libvserver, which I doubt) 1125929262 M * HostingGeek Greek0: rename the file 1125929273 M * HostingGeek Greek0: many ways of getting past this 1125929316 M * Greek0 HostingGeek: yep. I know. but I guess ola and/or enrico wouldn't be too happy if they'd have to change the name of their .so just because some other libvserver project came along 1125929340 M * Greek0 and I guess hollow isn't really all to keen on renaming libvserver to something else just to get past that conflict 1125929365 M * Greek0 IMHO a Conflicts: util-vserver in the libvserver*.deb would be the easiest way to go 1125929412 M * HostingGeek Greek0: but can they both be installed and used at the same time 1125929418 M * HostingGeek Greek0: read the policy 1125929430 M * HostingGeek Greek0: it isn't about what is easiest 1125929439 A * SiD3WiNDR sighs at HostingGeek :p 1125929442 M * HostingGeek Greek0: its about what the policy says 1125929513 M * Greek0 HostingGeek: this has nothing to do with the policy as libvserver*.deb isn't an official debian package, i.e. it's not in the archive 1125929521 M * Greek0 it's just some .deb someone hacked together 1125929557 M * HostingGeek Greek0: so... you still failed to explain why it doesn't have to follow the policy 1125929566 M * Greek0 and renaming the lib there would _really_ suck, since everything that needs the library would find the _wrong_ .so 1125929568 M * HostingGeek Greek0: there is no reason why it shouldn't 1125929586 M * Greek0 it would fine the util-vserver DSO, and not the libvserver one. unresolved symbols included. 1125929663 M * Greek0 IMHO this is where a Conflicts: is in order, unless one of the projects changes its soname 1125929711 M * Greek0 they just contain the same file, and these don't have a compatible interface. 1125929794 M * Greek0 renaming the library only in the libvserver*.deb, but not upstream would be a PITA IMHO, since you would have to adjust every application that uses libvserver.so to use libvserver-new.so (or whatever) instead 1125929869 M * Greek0 a diversion also doesn't work, since the 2 .so's don't have the same ABI 1125929966 M * Vudumen HostingGeek amd Greek0: i'd like to make this package installable with util-vserver too 1125930035 N * Bertl_zZ Bertl 1125930037 M * Greek0 Vudumen: it would be nice, but IMHO it would be quite a annoying to make sure the right .so is used for every app that wants to use libvserver 1125930047 M * Bertl morning folks! 1125930051 M * Greek0 moin 1125930070 M * Vudumen Greek0: but if it's called libvserver2 instead of libvserver then it doesn't have any annoying effects :) 1125930074 M * Vudumen hi Bertl :) 1125930103 M * Greek0 Vudumen: well, if you do it upstream, then yes. or libvserver-ng.. 1125930126 M * Greek0 but doing it just in the .deb means that stuff built for other distros wouldn't be compatible with that .deb 1125930127 M * HostingGeek or 1125930128 M * Bertl the main questions IMHOP are: do we want both libraries to coexist? 1125930134 M * Vudumen well i think libvserver2 would be better because it supports vserver 2.0 :) and ng could be a bit confusing because of ngnet 1125930139 M * HostingGeek libvserver-other.so 1125930141 M * HostingGeek or 1125930148 M * HostingGeek libvserver-lamer.so 1125930154 M * Greek0 Vudumen: yep, 2.so sounds nice 1125930175 M * HostingGeek libvserver-someotherversionwithadifferentname.so 1125930188 M * Vudumen Bertl: i think that until we don't have the vcontext, vkill, etc. utils built with the new libvserver it should i think 1125930224 M * Bertl Vudumen: at least on RPM based systems, you do not install the lib by default! 1125930273 M * Vudumen Bertl: ooooooo so on RPM based systems is the utils linked statically? 1125930291 M * Greek0 well, the current situation is that the 2 projects have an soname clash, which makes coexisting on the same system a pain. 1125930321 M * Greek0 I think it would be best if this problem was resolved, so there's at least the possibility to install both at the same time 1125930345 Q * Snow-Man Ping timeout: 480 seconds 1125930698 M * Bertl Vudumen: of course, they are built with dietlibc, no? 1125930714 M * Vudumen Bertl: currently not 1125930744 M * Bertl Vudumen: yes, they are (maybe not on debian :) 1125930772 M * Vudumen well now it's built with standard libc 1125930774 M * Vudumen (the deb packages) 1125930790 M * Bertl Vudumen: but we already had a bunch of reports where rebuilding with dietlibc did fix reboot/shutdown issues 1125930816 M * Bertl (and nothing else was changed, so I'd go for dietlibc) 1125930835 M * Vudumen Bertl: hmm, interesting. i'll try to build one with dietlibc 1125930865 M * Bertl Vudumen: alternatively you could fix the libc issues, including the security implications enrico is always talking about ... and feed that back to mainline :) 1125930894 M * Vudumen :))))) 1125930944 M * Greek0 Bertl: what .config do you use for the test-with-qemu kernels you build? 1125931205 M * Bertl sec, I'll upload the recent for 2.0.1* 1125931307 M * Greek0 I finally came around to setting up a vm with debian in qemu.. it was actually easier than I thought 1125931482 M * Bertl http://vserver.13thfloor.at/Stuff/QEMU/ 1125931537 M * Greek0 merci 1125931650 M * Greek0 do you know what the current status of COW links in mainline is? obviously it would be nice if this got merged sometime soon so we could use it 1125931727 M * Vudumen well i updated the packages for dietlibc 1125931900 M * Vudumen hmmhmm, interesting. as ldd shows it still depends on libc.so.6 1125931920 M * Vudumen and /lib/tls/libc.so.6 instead of /usr/lib/diet/lib/libc.so 1125931952 M * Bertl # ldd `which chcontext` 1125931961 M * Bertl not a dynamic executable 1125931970 M * Bertl # ll `which chcontext` 1125931982 M * Bertl -rwxr-xr-x 1 root root 5762 2005-08-06 01:39 /usr/sbin/chcontext 1125932050 J * lilo ~lilo@lilo.usercloak.oftc.net 1125932118 M * Vudumen well currently i built shared objects too which depend on libc :/ 1125932163 M * Bertl wb lilo! 1125932202 M * Bertl Vudumen: btw, the tools are on x86_64 (debian sarge) 1125932223 M * Bertl (just compiled from the source + fix02) 1125932241 M * Vudumen Bertl: well now i'm working on an i386 box... and now i'm packaging hollows libvserver 1125933050 J * eugenesan eugenesan@bzq-219-189-203.cablep.bezeqint.net 1125933066 M * eugenesan hi all 1125933129 Q * ag- Ping timeout: 480 seconds 1125933364 M * Bertl welcome eugenesan! 1125933374 M * eugenesan how are you? 1125933384 M * Bertl fine, thanks! and you? 1125933395 M * eugenesan Fine too. 1125933471 M * eugenesan Is there patch ready for 2.6.13? 1125933511 M * Bertl http://vserver.13thfloor.at/Experimental/patch-2.6.13-vs2.0.1-pre2.diff.bz2 1125933557 M * eugenesan it pre version, when do you thunk we could call it stable? 1125933588 M * Bertl it is vs2.0 plus some fixes, not all fixes yet, so not 2.0.1 but 2.0.1-pre2 :) 1125933618 M * Bertl i.e. it is a prerelease of the _next_ stable release 1125933638 M * eugenesan ok, will test it today. 1125933659 M * eugenesan BTW: I was browsing news and found that: http://openvirtuozzo.org/documentation/tech/ :-) 1125933677 M * Bertl with a little luck, we will have 2.0.1 in a week or two ... 1125933710 M * Bertl eugenesan: ah, interesting, so they finally decided to release the GPL parts? 1125933784 M * Bertl LOL "Virtuozzo -- The Only True VPS" (from their comparison page :) 1125933792 M * Vudumen :DD 1125933893 M * Bertl Hollow: might be interesting to look at the tools, seems they are QPL (Trolltech) 1125933987 M * eugenesan no, it''s GNU GPL and QPL 1125934012 M * eugenesan their tools and network stack should really interesting 1125934068 M * eugenesan I wonder what can cause company like SWSOFT do that??? 1125934110 M * Bertl well, I'd say there are many reasons: 1125934134 M * Bertl - a pending lawsuit regarding the kernel modifications 1125934175 M * Bertl - the competition of linux-vserver (which they tried to buy out :) 1125934205 M * Bertl - the idea of recruiting open source developer for free ... 1125934240 M * Bertl but maybe I'm wrong, and they are just a bunch of friendly guys which discovered the right path :) 1125934249 M * eugenesan :-) 1125934271 M * eugenesan i think there is only one guy that makes decisions there. 1125934279 M * Bertl interesting is the forum :) they already have one message there :) 1125934388 M * Bertl hmm, the networking doesn't sound that different to me: http://forum.openvirtuozzo.org/index.php?t=msg&th=3&start=0& 1125934736 M * Bertl hmm, seems 2.6.8 is the latest kernel ... 1125934763 M * eugenesan yes, it is. 1125934797 M * eugenesan but they tell they patch includes all backports from RHEL4 1125934817 M * Bertl from their FAQ: (a little self consumed :) 1125934822 M * Bertl Q. What are highlights of Open Virtuozzo technology? 1125934822 M * Bertl A. In short, Open Virtuozzo is the only highly scalable virtualization technology with near-zero overhead, strong isolation and rapid customer provisioning that's ready for production use right now. 1125934931 Q * Blissex Remote host closed the connection 1125934936 A * Hollow loughs 1125934962 M * Bertl but to some point they are very realistic: 1125934981 M * Bertl Q. What does 021stab018 in Open Virtuozzo kernel version mean? 1125934981 M * Bertl A. "stab" series kernels are more or less stable kernel releases which are close to be released or are in production already. 1125935002 M * Nicoli well the real question is, can u build ur own OS templates or do u ahve to use theirs ? 1125935026 M * Bertl Nicoli: guess you could try it and report back? 1125935032 M * Nicoli meh 1125935036 M * Nicoli i dont really want to heh 1125935043 M * Nicoli maybe lateron this week if i ahve time 1125935059 M * Nicoli caus ei ahve used the commercial version 1125935061 M * Bertl I for sure would appreciate feedback from folks testing/comparing both ... for sure will look at the GPL-ed parts ... 1125935064 M * Nicoli and their templates blow 1125935079 M * Bertl yes? how so? 1125935085 M * Nicoli they are al old 1125935087 M * Nicoli all 1125935098 M * Bertl well, 2.6.8 isn't that new either ... 1125935101 M * Nicoli so far we only ahve suse 9, debian 3.0 rh9 and fc1 1125935108 M * Nicoli thast it 1125935111 M * Nicoli thats 1125935138 M * Nicoli u know wats even funnier u can no longer use the plesk auto insatller on the virtuozo vps's 1125935153 M * Nicoli it will give and error that is isnt supported anymore 1125935158 M * Bertl lol 1125935222 M * Bertl they have a document called OS template cache creation ... so I guess you need to prepare the templates somehow 1125935336 M * Nicoli yea i already looked @ that 1125935343 M * Nicoli it talked about uising their pre-created ones 1125935349 M * Nicoli nothing about actually screating your own 1125935417 M * eugenesan it looks like there is no template creation tools :-( 1125935486 M * Bertl hey they have an irc channel too ... 1125935495 M * Nicoli lol 1125935554 M * Nicoli hrm i think u technically can create a template 1125935560 M * Nicoli they jsut amdeit a pain in the ass 1125935570 M * Greek0 Hollow: have you noticed our libvserver.so-filename-clash discussion above? 1125935589 M * Hollow no 1125935602 M * Nicoli u just gtotta take their metadate template and change it for whatever os u want and then create the template from that 1125935623 M * Greek0 started around 16.00 (if you happen to be in the same timezone as I am (MEST) ;) 1125935627 M * Bertl hmm, seems I'm the only one @ #openvirtuozzo :) 1125935637 M * Hollow lol 1125935638 M * Nicoli what network ? 1125935647 M * Bertl freenode.net 1125935664 M * Nicoli im kinda surprised, iw oudla thought they woulda made thir own 1125935670 M * Bertl Anyone with the operator status is usually an Open Virtuozzo developer. (ah, now they got me lol :) 1125935686 M * Nicoli heh 1125935709 M * Greek0 mm. bertl switched to the dark side ;) 1125935728 M * Nicoli nah hes spying 1125935752 M * micah howdy 1125935758 M * Nicoli u can only run about 40 vps's on a dual opterong with 10gb of ram 1125935766 M * Greek0 Bertl: what additions do I actually want to -pre5 to get cowlinks working? 1125935767 M * Nicoli afte that it starts to run like crap 1125935784 M * Bertl well, I've been trying to get the swsoft-folks to cooperate for a long time now ... because I think both sides could benefit from a development cooperation 1125935813 M * Bertl Greek0: just enable them ... that's enough ... 1125935826 M * Nicoli Bertl: didn't they try to buy u out ? 1125935833 M * Greek0 ah, so you have all the fixes in -pre5 already? 1125935870 M * Greek0 I didn't watch too closely what was going on at the cow-links-more-or-less-broken front 1125935898 M * Bertl Nicoli: well, kind of, but I'm not very vengeful ... 1125935909 M * Bertl Greek0: let me verify that .. will only take a second 1125935957 M * Bertl Greek0: okay, I'll do a pre6 release for 2.6.13 1125935958 M * Hollow Greek0: well, my libvserver is libvserver.so.1 enricos is libvserver.so.0 so can't they exist in paralell? 1125936019 M * Greek0 the clash is probably in the libvserver.so symlink 1125936054 M * Greek0 hmm. which would probably mean that only the -dev packages are conflicting 1125936102 M * Bertl yes, but debian has no devel package, no? 1125936116 M * Greek0 there is no official debian package for libvserver yet. 1125936132 M * Bertl I'm talking about util-vserver here ... 1125936157 M * Greek0 no, not really. probably because util-vserver's libvserver isn't used by anything else 1125936216 M * Greek0 hmm. we could bug ola to take that .so (not the .so.0) out of the util-vserver package. 1125936331 M * Bertl would make sense to beak up the debian package ... 1125936335 J * frz ~frzzzz100@jaim.at 1125936340 P * frz 1125936370 M * Greek0 Bertl: util-vserver or libvserver? 1125936393 M * Bertl Greek0: both, yes :) 1125936464 M * Greek0 well, I'm not sure if a split of util-vserver is needed, as I don't know if anyone actually uses the devel stuff util-vserver provides 1125936484 M * Greek0 the only reverse dependency it has (aside from the kernel patch) is vserver-debiantools 1125936541 M * Greek0 and the vserver-debiantools seems to be only shell scripts or simmilar (at least arch-independent, so it can't use the .so anyway) 1125936549 M * Bertl well, and the vserver-debiantools are depreciated (at least on my side :) 1125936579 M * Bertl they should be integrated into util-vserver or whatever tool package to come ... 1125936605 M * Greek0 I guess it's pretty much up to Ola. but I think the .so isn't needed anyway, so it could just as well be removed from util-vserver without breaking stuff 1125936619 M * Hollow Bertl: do you think it would be good to force the relation xid and nid? 1125936624 M * Hollow +of 1125936634 M * Bertl preferable 1:1 1125936636 M * Hollow Greek0: wrong 1125936646 M * Hollow apis are very different 1125936661 M * Hollow Bertl: sure, but should we force it, or make it a common policy? 1125936669 M * Greek0 Hollow: I'm talking about the libvserver.so symlink to libvserver.so.0.0.0 1125936673 M * Greek0 not about the library as such 1125936677 M * Hollow ah the symlink.. k 1125936685 M * Bertl Hollow: I'm not sure if we will have the nid with ngnet :) 1125936701 M * Bertl so I'd just make it common practice now and maybe drop it later :) 1125936703 M * Hollow i.e. it would be better to force it, and don't even record an extra id? 1125936707 M * Hollow k 1125936719 M * Greek0 is it ok if I bring up this topic on the ML, ask ola for his opinion and suggest just removing the symlink? 1125936732 M * Bertl Hollow: you can also stay with the dynamic nids as util-vserver does now 1125936762 M * Hollow i'd prefer to just set the network context and don't care about the id.. ;) 1125936802 M * Bertl good, so we'll revise that once we know how ngnet does look like, yes? 1125936809 M * Hollow wouldn't it be even more accurate to remove the ncontext things and just add or remove addresses from normal contexts? 1125936815 M * Hollow yeah 1125936815 M * Bertl s/revise/revisit/ 1125936971 M * Hollow Bertl: i looked at secure-mount but tbh i'm not able to follow what it does... can you explain how to mount things so only the guest context "sees" it? i was horribly lost with this one.. 1125937054 M * Bertl first, secure mount and guest only are different things 1125937057 M * Hollow especially as mounting is done before any vnamespace operation is done 1125937081 M * Bertl (at least in my concepts) 1125937103 M * Hollow yeah, basically i'm talking about the concept, dunno if util-vserver follows it.. ;) 1125937139 M * Bertl no, I don't think so ... (now that I think of it :) 1125937145 M * Hollow heh 1125937157 M * Hollow so how should i/we do it correctly? 1125937162 M * Bertl I introduced a flag for 'secure' mounting inside a guest (a capability actually) 1125937175 M * Hollow so the guest itself can mount? 1125937176 M * Bertl this is what I understand when I hear secure mount 1125937192 M * Bertl yes, it can, but the mount will be done with nodev 1125937221 M * Bertl (to make it secure) 1125937229 M * Hollow ok.. i'm talking about the secure-mount app in util-vserver, it does all the mounting _before_ any context or namespace is created 1125937246 M * Bertl for the namespace and mounting, the following applies: 1125937265 M * Bertl - if you create a new namespace (with the clone syscall) 1125937279 M * Bertl then this automatically is a copy of the current namespace 1125937308 M * Bertl - if you now do mounting there, it will only apply to the new namespace (also valid for umount, btw) 1125937338 M * Bertl - there you should do the guest mounts and umounting of removeable mounts from the host (cleanup) 1125937370 M * Bertl - after that, the magic --rbind happens and the new namespace can be 'registered' for the guest 1125937397 M * Hollow what will be rbind-ed? 1125937401 M * Bertl (if you want to allow for host style maintainance, you have to register before the --rbind 1125937412 M * Bertl /vservers/guest -> / 1125937446 M * Bertl AFAIK that's how util-vserver does it right now 1125937449 M * Hollow mhm... i.e. i don't need chroot? 1125937473 M * Bertl not really, that's the advantage of namespaces :) 1125937490 M * Hollow mhm 1125937517 M * Bertl you should have a talk with Doener, he did extensive namespace research some time ago 1125937539 M * Hollow ok 1125937540 M * Greek0 why rbind and not pivot_root? 1125937542 M * Bertl Greek0: test compiling now, patch will be available shortly 1125937549 M * Greek0 thanks Bertl 1125937568 M * Bertl Greek0: because the pivot requires the old / to go somewhere :) 1125937596 M * Greek0 yep, but it could be umounted then, not? 1125937614 M * Greek0 hmm, but rbind skips this step.. well 1125937625 M * Bertl except for that, they are equivalent ... 1125937654 M * Bertl IIRC, there was some refcounting issue which prevented unmounting after a pivot_root 1125937664 M * Doener Greek0: the tools are required to be available in the new namesapce, so pivot_root would have to happen after the namespace is registered 1125937666 M * Bertl (maybe dig through the archives if you are really interested) 1125937671 M * Hollow Bertl: register means set namespace? 1125937673 M * Doener the rbind happens before the namespace is registered 1125937686 M * Doener just the final chroot happens after the namespace has been registered 1125937699 M * Greek0 but I thought pivot_root did some magic that made everything work (i.e. running apps survive it without problems and see the new root or something) 1125937721 M * Doener the special thing about the rbind is that you then do "chroot /" instead of "chroot /vservers/foo" IIRC 1125937724 M * Bertl evening Doener! 1125937741 M * Doener evening Bertl, #vserver (short visit) 1125937741 M * Hollow but why would i cahnge my root to my current root? 1125937762 M * Doener Hollow: the rbind binds the /vservers/foo to / 1125937779 M * Doener and IIRC the chroot takes the rbinded / 1125937808 M * Bertl yep, would be possible, because the running app keeps the old namespace, root, whatever 1125937838 Q * ntrs_ Remote host closed the connection 1125937841 M * Greek0 ah, so that the running vserver tools also see the new root, and not the old one 1125937842 M * Hollow but if i clean the new namespace and make /vservers/foo > / i can't access anything above the / can i? 1125937848 M * Bertl that's why removing a temporary file after opening it works ... 1125937886 M * Greek0 ok, this was the magic that I thought pivot_root did. you just do it manually with rbind + chroot.. 1125937997 M * Doener Hollow: you keep being inside the old / until you chroot 1125938007 M * Hollow after rbind? 1125938013 M * Doener yep 1125938038 M * Hollow mhm 1125938057 M * Doener doener b # mount -o bind ../a ../b 1125938057 M * Doener doener b # ls 1125938057 M * Doener doener b # cd .. 1125938057 M * Doener doener ~ # cd b 1125938057 M * Doener doener b # ls 1125938058 M * Doener test 1125938076 M * Doener oops, sorry, missed the browser window... 1125938180 M * Doener anyways, the bind only gets 'active' if you re-enter the directory 1125938200 M * Doener as long as you're in the "old" directory, you stay there and the bind mount doesn't affect you 1125938244 M * hillct I need a quick refresher on spec files. 1125938254 M * hillct Tried to specify CC=gcc32 1125938265 M * hillct but aparently I'm not doing it properly 1125938350 M * Bertl depending on the distro, you should have something called alternatives 1125938397 M * Bertl if not, then use the -V option to specify the version, try with gcc -V 3.2.x -v 1125938415 M * Bertl (you can then add the -V to the CFLAGS) 1125938452 M * Hollow Doener: wouldn't chdir() be enough then? 1125938454 M * hillct K 1125938478 M * Bertl Greek0: http://vserver.13thfloor.at/Experimental/patch-2.6.13-vs2.1.0-pre6.diff.bz2 1125938487 M * Doener Hollow: / keeps being your parent 1125938500 M * Bertl I'm off for a few hours now ... but will be back later ... 1125938508 M * Hollow bye Bertl 1125938508 N * Bertl Bertl_oO 1125938540 M * Doener Hollow: it might be, that the chroot is still to /vservers/foo. but the usual "chdir ../../../../../../../" won't work cause then you'd leave your current / and end up in the new / 1125938564 M * Doener which is the same as the directory you'd been chrooted into 1125938587 Q * cryo Quit: ðÏËÉÄÁÀ 1125938598 J * cryo ~say@212.86.243.154 1125938606 M * Doener my brain is filled with too much math stuff atm to answer that questio for sure 1125938612 M * Hollow :) 1125938728 M * Hollow so we'd need the following: CLONE_NEWNS, vx_namespace_cleanup, chroot /vservers/foo, mount things, rbing, chroot / 1125938730 M * Hollow correct? 1125939173 M * Greek0 thanks Bertl_oO 1125939227 M * Greek0 Doener: chdir `pwd`? 1125939285 M * Greek0 Hollow: why that first chroot? 1125939325 M * Hollow dunno, Doener talked about the chroot to /vservers/foo so i thought it's necessary 1125939355 M * Doener Hollow: the last step is either chroot / _or_ chroot /vservers/foo 1125939363 M * Doener i simply don't remember which it was ;) 1125939385 M * Hollow but after we rbinded /vservers/foo to / we won't have a /vservers/foo anymore, do we? 1125939394 M * Hollow or does the current process still have it? 1125939751 M * Doener the current process still has it... 1125940062 M * Doener Hollow: ok, it's _not_ chroot / 1125940438 M * Greek0 If I had more knowledge of the Linux FS stuff (inodes/dentries) I'd probably to articulate this a bit better, but are you sure that you can't break out of that chroot then? i.e. that the outer stuff (old /, old /path/to/vservers/, ..) is really gone? 1125940988 J * renihs ~renihs___@193.170.52.70 1125941140 Q * SiD3WiNDR Ping timeout: 480 seconds 1125941603 J * nayco ~nayco@lns-vlq-21-nan-82-255-73-178.adsl.proxad.net 1125941613 M * nayco hello, all !! 1125941676 M * nayco Bertl ? 1125941723 M * HostingGeek _o0 == far far away 1125941735 M * HostingGeek you scarde him away 1125941740 M * HostingGeek way to go! 1125941743 M * nayco ;) 1125941759 M * nayco Ok, I gonna mail him. 1125941973 Q * prae Quit: Execute Order 69 ! 1125943069 Q * douglas Ping timeout: 480 seconds 1125944322 J * dddd44 dhb55@218.111.177.154 1125944351 Q * dddd44 Read error: Connection reset by peer 1125944373 J * dddd44 ~dhb55@tor-irc.dnsbl.oftc.net 1125944780 J * monrad ~monrad@213.83.190.134 1125945441 J * prae ~benjamin@sherpadown.net 1125945730 Q * dddd44 Ping timeout: 480 seconds 1125945744 N * pg|cereal _pg|cereal 1125945886 J * dddd44 ~dhb55@tor-irc.dnsbl.oftc.net 1125946379 J * ag- ag@caladan.roxor.cx 1125947485 M * Greek0 Hollow: ping 1125947490 M * Hollow pong 1125947562 M * Greek0 http://greek0.net/~greek0/patches/libvserver-symbol-versioning.diff 1125947634 M * Greek0 this adds symbol versioning support to libvserver 1125947734 M * hillct OK, so we know the compiler isn't the problem in building usil-vserver 208 RPMs on fc4 x86_64 1125947772 M * hillct because I can duplicate exactly, the configure and Make commands used 1125947793 M * hillct ourside the RPB build and it works fine 1125947812 M * hillct could be a path issue I suppose 1125947847 M * Greek0 hillct: can you provide build logs please? 1125947857 M * hillct sure 1125947863 M * Greek0 or something else that gets me up to date on that issue? 1125947980 M * hillct http://pastebin.com/355473 1125948017 M * hillct building inside the RPM build process results in failure of the fmt test 1125948034 M * hillct building identically, outside the RPM build process succeeds 1125948062 M * Greek0 are you sure the log isn't truncated? 1125948089 M * hillct hmm 1125948099 M * hillct looks like pastebin doesn't like things that big 1125948119 M * hillct lets try something else... 1125948282 M * mnemoc http://42.vg/ ? 1125948299 M * Greek0 hillct: go to http://greek0.net 1125948311 M * Greek0 there is an upload thingie where you can drop files up to 50mb to 1125948322 M * mnemoc :) 1125948329 M * Greek0 "File upload area" 1125948417 M * hillct http://42.vg/77184 1125948423 M * hillct seems to have worked 1125948518 Q * lilo Remote host closed the connection 1125948647 M * hillct Greek0 it's also in your file upload area 1125948653 M * Greek0 thx 1125948704 M * mnemoc *g* 1125948723 M * hillct kernel is 2.6.13vs2.0-vs2.0.1-pre2 1125948765 M * hillct gcc version 4.0.1 1125948779 M * hillct for kernel build and util-vserver build 1125948825 M * hillct works fine when built on FC2 x86_64 1125948856 J * lilo U2FsdGVkX1@lilo.usercloak.oftc.net 1125948885 M * hillct so there's gotta be something slightly broken about the RPM enviroment 1125948986 M * Greek0 what I do not fully understand is how that test is supposed to work on 32 bit archs 1125949013 M * hillct It's greek to me 1125949023 M * hillct it's just an integer length test right? 1125949058 M * Greek0 no, it tests a function that converts 64 bit ints into their hex representation as string 1125949075 M * hillct but the error returned when I run the test as built by the failed RPM build is 'Killed' 1125949089 M * Greek0 TEST(68719476736ul, "1000000000"); <-- that one bugs me, since 68719476736 doesn't fit into an ul on 32 bit archs 1125949105 M * hillct ah 1125949119 M * Greek0 oh, strange 1125949122 M * Greek0 overlooked that 1125949158 M * hillct but again, I can execute the exact same configure and make lines, by hand, and it works like a champ 1125949159 M * hillct er 1125949174 M * hillct with the exception of _smp_flags 1125949179 M * hillct gotta check what those are 1125949181 M * Greek0 can you go to the rpm built tree and execute the test by hand? 1125949189 M * hillct yup 1125949198 M * hillct and result is 'Killed' 1125949217 M * Greek0 can you do it in gdb and provide a backtrace? 1125949236 M * hillct generates an empty stack 1125949272 M * hillct so we went in and tried some of the tools 1125949285 M * Greek0 "an empty stack"? 1125949285 M * hillct which complained of missing functions 1125949290 M * hillct yeah 1125949299 M * hillct hamn on let me set this up again 1125949707 M * hillct http://42.vg/77192 1125949756 M * hillct hmm 1125949785 M * Greek0 use ulimit -c 10000000 to enable coredumping 1125949796 M * Greek0 i.e. ulimit -c 10000000000; gdb ... 1125949813 M * hillct I didn't rebuild dietlibc when I upgraded from FC2 to FC4 1125949825 M * hillct K 1125949829 M * hillct first ulimit 1125949855 M * Greek0 aeh yes. just enable coredumping with that ulimit command, then run gdb -c core fmt 1125949929 M * hillct bah! 1125949936 M * hillct where did it put my core file... 1125949937 M * hillct hmm 1125949947 Q * Nicoli Ping timeout: 480 seconds 1125949948 M * Greek0 usually to the cwd 1125949960 M * hillct yeah 1125949962 M * hillct not there 1125949977 M * Greek0 what does ulimit -c say? 1125949997 M * hillct no error 1125950002 M * Greek0 what output? 1125950006 M * Greek0 infinite? 0? 1125950006 A * hillct tries again 1125950105 M * hillct the process seems to be getting killed off 1125950111 M * hillct before it can generate a core 1125950123 M * hillct or provide any debugging trace 1125950124 P * jazzanova 1125950140 M * Greek0 well, actually the kernel should produce the core dump actually 1125950158 M * Greek0 (well, think of one "actually" as being invisible) 1125950160 M * hillct t'aint happening 1125950197 M * Greek0 hmm. makes sense. sigkill 1125950221 M * Greek0 hmm. no idea really. 1125950244 M * hillct well, I'll rebuild dietlibc 1125950262 M * Greek0 no idea where that sigkill is coming from 1125950522 M * hillct actually, the dietlibc thing makes no sense 1125950529 M * hillct since it works when built by hand 1125950648 M * Greek0 hmm. I'd ensure that I've got a clean build environment in the RPM build thing. 1125950699 M * Greek0 that it builds fine outside of the RPM build thing (and also on debian amd64) seems strange 1125950789 M * hillct clean as far as I can tell 1125950989 M * Greek0 no idea then 1125951123 M * daniel_hozac are you duplicating the RPM build environment exactly? 1125951134 M * daniel_hozac setting the same CFLAGS etc.? 1125951266 M * hillct as best I can 1125951303 M * hillct but I'm obviously missing something 1125951347 M * daniel_hozac could you use the same %build script as RPM? (as in, stealing it from %{_tmpdir}) 1125951373 M * hillct sure 1125951470 M * hillct well 1125951472 M * hillct no 1125951482 M * hillct because it gets deleted after the build 1125951497 M * hillct can tweak the spec file so it's not deleted 1125951498 M * daniel_hozac put an exit 1 at the end of %build 1125951503 M * hillct yeah 1125952147 M * hillct ok, well that's a good sign 1125952162 M * hillct using the RPM build script I can reproduce the problem for testing 1125952213 M * hillct rather embarrased this testing strategy didn't occur to me, but anyway... 1125952466 M * hillct http://42.vg/77203 1125952858 M * daniel_hozac could you try removing the -Wp,-D_FORTIFY_SOURCE=2? 1125952933 M * hillct any idea what it's for 1125952938 M * hillct never seen it used before 1125952944 A * hillct removed it 1125952954 M * hillct er will momenterily 1125953155 M * hillct still fails 1125953165 M * hillct any other candidates? 1125953218 M * daniel_hozac -O2? 1125953225 M * daniel_hozac -fexceptions? 1125953246 M * hillct -O2 I'd hardly believe would cause this problem 1125953251 M * hillct exceptions maybe 1125953305 M * daniel_hozac you are doing all the test runs in brand new trees, right? 1125953314 M * daniel_hozac just extracted, that is. 1125953519 M * hillct after make distclean 1125953595 J * douglas ~douglas@c-67-160-174-224.hsd1.or.comcast.net 1125953598 P * douglas 1125953617 J * douglas ~douglas@douglas.user.oftc.net 1125954362 M * daniel_hozac did -fexceptions cure it? or -O2? 1125954399 M * hillct neither 1125954415 M * daniel_hozac tried -pipe? 1125954415 M * hillct just as a sanity check I built by hand again 1125954429 M * hillct I'm going to yank everything out 1125954434 M * hillct just for kicks 1125954563 J * Aiken ~james@tooax6-190.dialup.optusnet.com.au 1125954570 M * hillct well 1125954588 M * hillct turns out editing the cflags at the top doesn't impact the script 1125954591 M * hillct d'oh 1125954597 M * daniel_hozac hehe 1125954607 M * hillct I should have noticed earlier 1125954983 M * hillct progress 1125955001 Q * wibble Read error: Connection reset by peer 1125955002 J * wibble ~tim@sophie.wobb1e.co.uk 1125955025 M * hillct removed all the cflafs, cxxflags and fflags and it passed 1125955038 M * hillct now putting them back in one by one 1125956032 Q * brc Ping timeout: 480 seconds 1125956081 M * Greek0 Hollow: did you have time to look at my patch? 1125956125 M * Hollow Greek0: i'm still looking.. (while watching tv) but i don't understand it yet, reading symbol versioning docs though 1125956208 M * hillct progress has been made! 1125956234 M * Greek0 I found http://www.google.com/url?sa=t&ct=res&cd=1&url=http%3A//people.redhat.com/drepper/dsohowto.pdf&ei=VLocQ4mbCo2WiAK1o-m3Dg very useful, and I got that HIDDEN stuff from http://people.debian.org/~vorlon/dependency-hell/img9.html 1125956363 M * Greek0 hmm. well. for the HIDDEN stuff http://lists.debian.org/debian-mentors/2005/09/msg00088.html is useful 1125956383 J * SiD3WiNDR luser@bastard-operator.from-hell.be 1125956499 M * Greek0 Hollow: the configure.ac stuff is obviously only logic to enable symbol versioning if --with-versioned-symbol is passed. 1125956511 M * Hollow yeah.. 1125956535 M * Hollow i understand what the options do with the library, but i don't understand what the library does with the versioned symbols ;) 1125956554 M * Greek0 Hollow: the -export-dynamic flag is pretty unrelated actually. it only makes it possible for the library to be used with dlopen() [apparently needed on some platforms] 1125956585 M * Greek0 and --version-script tells ld to uhm. use the version-script to do symbol versioning. 1125956598 M * Hollow yup 1125956605 M * Greek0 symbol versioning is nice since you can do ABI changes without having to change the soname of your library 1125956648 M * Greek0 binaries compiled against the old version have the proper symbol version hardcoded in them, and your library still provides the old interface. but if you recompile a binary against the library it can use the newer interface 1125956701 M * Greek0 it's just enables you to do ABI changes without having to recompile every single application that links against the library 1125956716 M * Greek0 drepper's pdf explains this really well IMHO 1125956726 M * Hollow well, my library doesn't support "old versions" 1125956740 M * Hollow i'll take a look, but not today, bit tired already 1125956778 M * Greek0 hmm. it just enables you to change the interface of your library without completely breaking every application that links against it 1125956804 M * Hollow hm, i thought it solves the "both libvservers are installed"-issue 1125956815 M * Greek0 Hollow: no 1125956826 M * Greek0 I've sent a mail to the ML about that, I hope it comes through 1125956850 M * Hollow k, got that wrong then 1125956865 M * Greek0 I've looked a bit deeper and nothing except util-vserver uses it's libvserver. So the nicest way would be to move libvserver.so away from /usr/lib to /usr/lib/util-vserver 1125956868 M * Hollow so this is basically a new feature ;) 1125956897 M * Greek0 more or less. I hope it isn't a bug ;) 1125956912 M * Hollow we'll see, thx anyway 1125956942 M * Greek0 somehow I don't trust in lists.linux-vserver.org.. 1125956947 M * Hollow Greek0: interested in libvserver development? 1125956950 M * Greek0 I hope that damn mail got through 1125956970 M * Hollow (read: help with it?) 1125956971 M * Hollow *g* 1125956980 M * Greek0 hehe possibly :) 1125957012 M * Hollow ok, just let me know if you want svn access or something 1125957077 M * Greek0 mm yes. my attention is just kind'a jumpy, so I want to work on a project for some more time, just to prove to myself that I stay interested 1125957102 M * Hollow heh 1125957111 M * Greek0 and linux-vserver has kept me interested for quite some time now actually :) 1125957119 M * Hollow yeah, same here ;) 1125957527 Q * wibble Read error: Connection reset by peer 1125957532 J * wibble ~tim@sophie.wobb1e.co.uk 1125957541 M * Hollow ok, off to bed, cya 1125957557 M * Greek0 cu 1125957616 M * nayco Greek0: 1125957629 M * nayco Do you use netatalk ? 1125957714 M * Greek0 no, sorry 1125957860 M * nayco ok. 1125957879 M * nayco Er, you work with libvserver ? 1125957927 M * Greek0 I've looked at it a bit 1125957978 M * nayco Could you telle me what is its purpose ? (Ok, I can guess...) ? Which languages does it support ? 1125958041 M * Greek0 It is a library interface to the vserver stuff the kernel provides 1125958058 M * Greek0 it's a c library, I've been thinking about looking into swig to generate python bindings 1125958074 M * Greek0 there are also tools that use it, although I've lost their URL :-/ 1125958088 M * Greek0 but they are in early alpha stage afaik 1125958125 M * Greek0 the goal is to replace util-vserver one day, mostly because the util-vserver code is a mess (IMHO) 1125958153 M * nayco thanks. so, it could be used to rewrite (It's an exemple, to understand, not a critisism) util-vserver in C, or code an interface to administer vservers, right ? 1125958170 M * nayco ok. 1125958252 M * Greek0 util-vserver is partly c, partly shell scripts 1125958288 M * Greek0 I'm not sure what hollow is planning for his utils 1125958318 M * nayco I was thinking about libvserver for some kind of inteface to create and manage vservers, but as I think of it, it's too low-level. I mean, for example, making a web interface would better use utils-vserver (however they will be coded) intead ... 1125958340 M * nayco Hollow is coding utils ? 1125958348 M * Hollow Greek0: probably C+bash too... 1125958500 M * Hollow watching tv again, just ping me... 1125958538 M * Greek0 Hollow: you better watch out, in the end you also develop such a wake/sleep cycle like Bertl ;) 1125958558 M * Hollow i would... 1125958583 M * Greek0 nayco: yep, libvserver is a low-level thing. and there is also some web interface for util-vserver 1125958604 M * Greek0 http://www.openvps.org/Plone IIRC 1125958633 M * nayco ok. 1125958638 M * nayco Gonna have a look 1125958702 M * nayco seems nice. Sniffle, no screenshots ;-) 1125958746 M * hillct the problematic CFLAG appears to be -mtune=nocoma 1125958789 M * daniel_hozac really? 1125958796 M * nayco ok, so libvserver could be used to develop a fast specialised tool, if I understand correctly 1125958797 M * hillct about to look it up but can anyone give me a rundown on it's use in util-vserver? 1125958807 M * daniel_hozac not at all. 1125958812 M * daniel_hozac it's a compiler thing. 1125958819 M * daniel_hozac optimize for nocoma. 1125958826 M * hillct yeah 1125958830 M * hillct which is what/ 1125958832 M * Hollow nayco: you can do that yeah ;) 1125958883 M * daniel_hozac Intel's x86_64, IIRC. 1125958893 M * hillct ah 1125958897 M * hillct that'd do it 1125958907 M * daniel_hozac it shouldn't. 1125958921 M * hillct since AMD <> Intel 1125958925 M * daniel_hozac tune is just ordering of instructions. 1125958946 M * hillct well, it seems to make all the difference 1125958992 M * daniel_hozac i wonder what the default -mtune is for x86_&4. 1125958997 M * daniel_hozac s/&/6/ 1125959022 M * hillct looking now 1125959087 Q * nayco Quit: Bonne nuit ! 1125959402 N * Gonzo^away [MUPPETS]Gonzo 1125959644 Q * Doener Ping timeout: 480 seconds 1125959648 J * Doener ~doener@p54875446.dip.t-dialin.net 1125959973 J * nayco ~nayco@lns-vlq-21-nan-82-255-73-178.adsl.proxad.net 1125959983 Q * nayco Quit: 1125960937 M * hillct of course I should have realized this days ago, but the proper mtune value is of course (in my case) opteron 1125960949 M * hillct which works fine 1125961000 M * hillct but it appears to be an issue with my RPM build configuration in general 1125961030 M * hillct it's just that the util-bserver code responded more drastically to the slight misconfiguration 1125961081 Q * prae Quit: Pwet 1125962200 M * mugwump dammit, I hate the way when you change anything in kernel/, it recompiles absolutely everything on `make bzImage' 1125962599 N * Bertl_oO Bertl 1125962602 M * Greek0 mugwump: normally the linux 2.6 build system is quite smart about rebuilds 1125962605 M * Bertl mugwump: huh? 1125962619 M * mugwump hey Bertl 1125962646 M * mugwump maybe it was the other changed file, in include/linux/vserver/sched_def.h 1125962704 M * Bertl unlikely ... are you sure your time/date is correct? 1125962724 M * Bertl I had that once, when a build machine was stuck in the past 1125962793 M * mugwump hmm possibly 1125962857 Q * tchan Quit: leaving 1125963160 J * tchan ~tchan@c-24-13-81-164.hsd1.il.comcast.net 1125963181 M * Bertl mugwump: okay, so you want to track down the TB issue? 1125963220 M * mugwump sure. 1125963239 M * Bertl I have a pretty good idea what might happen, but I have to verify it first ... 1125963244 M * mugwump I'm currently booted into a kernel with the problem. I'm having extremely frustrating issues with testing 1125963278 M * mugwump eg, after booting into a new kernel, it'll moan about format problems of unix.ko, etc 1125963298 M * mugwump so I'm currently looking at ~30min test cycles 1125963335 M * Bertl hmm, how so? 1125963500 M * mugwump ok, so I built and installed a kernel, 2.6.12.5-vs2.0, using `make bzImage && make modules && make install && make modules_install' 1125963508 M * mugwump then, I applied that little patch 1125963515 M * mugwump `make bzImage && make install', reboot 1125963526 M * mugwump the system refuses to use the new kernel and the old modules 1125963546 M * Bertl wrong config, but okay 1125963573 M * mugwump am I doing something obviously wrong there? 1125963576 M * Bertl # CONFIG_MODVERSIONS is not set 1125963579 M * Bertl # CONFIG_MODULE_SRCVERSION_ALL is not set 1125963615 M * mugwump those are both =y, they should be off? 1125963625 M * Bertl if you don't want checking, yes :) 1125963723 M * Bertl okay, let's aim for something like 2.6.13-vs2.0.1-* 1125963935 M * Bertl mugwump: you are running on a 64bit SMP system, with sched_prio but not sched_hard, right? 1125963995 M * mugwump I have both the config options now. That no tokens are deducted at all without the CONFIG_VSERVER_HARDCPU set is another bug 1125964007 M * mugwump I have no patch for that 1125964023 M * Bertl well, its not a bug, strictly speaking 1125964044 M * Bertl but it's unexpected, I grant you that ... 1125964050 M * mugwump you're catching the core dev's bug denial disease Bertl ! :) 1125964066 M * Greek0 *argl* there seems to be no sane way to create .deb packages of the kernel when you build out-of-tree 1125964074 M * Bertl mugwump: nah, I just forgot why we did not support it ... 1125964131 M * Bertl mugwump: but as I said, we will split the HARD_SCHED config option in two (soon) 1125964160 M * Bertl one only enabling the token bucket(s) the other also enabling the hold queues 1125964238 M * hillct so, the problem is -mtune=nocona which in my case should be -mtune=opteron which seems to be a function of the system configuration of rpmbiold. The util-vserver code seems to be especially sensitive to these optimizations 1125964264 M * Bertl mugwump: the point is, enabling the token consumption is the part which introduces overhead in the scheduler, not the hold queues ... 1125964267 M * hillct I havn't worked out where rpmbuild pulls the default optimization flags from though 1125964283 M * mugwump Bertl: say what? 1125964286 M * Bertl hillct: IIRC, there is a macros file for rpm build? 1125964289 M * Greek0 hillct: sounds like a gcc bug IMHO. -mtune shouldn't break code 1125964307 A * Greek0 is quick in blaming gcc in the last time ;) 1125964312 M * mugwump The TBF calcs are like a few instructions every tick 1125964339 M * hillct there are many macro files for RPMbuild and the docs say I should simply be able to place my own file in /etc/rpm 1125964345 M * mugwump even the recalculations is only a handful of instructions 1125964354 M * hillct but that doesn't seem to do the trick 1125964362 M * Bertl mugwump: and they flush the cacheline and cause inter cpu issues :) 1125964435 M * mugwump ah, right. That should go away with context affinity? 1125964446 A * mugwump reboots quickly to see if this MODVERSIONS change worked 1125964456 M * Bertl well, it will go away with per cpu buckets :) 1125964674 M * mugwump ok. A simple way to fix that problem would be only to account tokens at resched time rather than every tick 1125964689 Q * eyck Remote host closed the connection