#ceph IRC Log


IRC Log for 2011-10-12

Timestamps are in GMT/BST.

[0:48] * aliguori (~anthony@cpe-70-123-132-139.austin.res.rr.com) has joined #ceph
[1:09] * aliguori (~anthony@cpe-70-123-132-139.austin.res.rr.com) Quit (Quit: Ex-Chat)
[1:29] * lxo (~aoliva@lxo.user.oftc.net) Quit (Remote host closed the connection)
[1:29] * lxo (~aoliva@lxo.user.oftc.net) has joined #ceph
[1:48] * Meths_ (rift@ has joined #ceph
[1:54] * Meths (rift@ Quit (Ping timeout: 480 seconds)
[2:00] * cp (~cp@ Quit (Quit: cp)
[2:11] * Dantman (~dantman@S01060023eba7eb01.vc.shawcable.net) Quit (Ping timeout: 480 seconds)
[2:53] * adjohn (~adjohn@ Quit (Quit: adjohn)
[3:02] * yoshi (~yoshi@p9224-ipngn1601marunouchi.tokyo.ocn.ne.jp) has joined #ceph
[3:17] * Dantman (~dantman@S010600259c4d54ff.vs.shawcable.net) has joined #ceph
[3:55] * jojy (~jojyvargh@ Quit (Quit: jojy)
[5:03] * Dantman (~dantman@S010600259c4d54ff.vs.shawcable.net) Quit (Remote host closed the connection)
[5:13] * Dantman (~dantman@S010600259c4d54ff.vs.shawcable.net) has joined #ceph
[6:03] * adjohn (~adjohn@50-0-92-177.dsl.dynamic.sonic.net) has joined #ceph
[8:04] * adjohn (~adjohn@50-0-92-177.dsl.dynamic.sonic.net) Quit (Quit: adjohn)
[10:48] * andret (~andre@pcandre.nine.ch) Quit (Quit: Verlassend)
[10:48] * andret (~andre@pcandre.nine.ch) has joined #ceph
[11:52] * yoshi (~yoshi@p9224-ipngn1601marunouchi.tokyo.ocn.ne.jp) Quit (Remote host closed the connection)
[13:02] * julienhuang (~julienhua@pasteur.dedibox.netavenir.com) has joined #ceph
[13:18] * julienhuang (~julienhua@pasteur.dedibox.netavenir.com) Quit (Quit: julienhuang)
[13:48] * danpb (~berrange@ has joined #ceph
[13:56] <danpb> i'm new to Ceph and trying to get RBD working with QEMU. I have a ceph cluster initialized with mkcephfs, and can mount the ceph FS successfully on my hosts. I get stuck with using RBD though:
[13:56] <danpb> # rbd create demo1 --size 100
[13:56] <danpb> 2011-10-12 12:54:42.100820 7fd53f768740 librbd: failed to assign a block name for image
[13:56] <danpb> create error: Input/output error
[13:58] <danpb> (packages ceph-0.31-4.x86_64 on kernel )
[13:59] <darkfader> can you run "touch" in the mounted fs?
[14:00] <darkfader> or does that also fail? :)
[14:00] <darkfader> well touch newfile
[14:00] <danpb> yep, i can create a file on one node and read on the other node
[14:01] <darkfader> ok :( then you'll have to wait for the elder gods to wake up, i don't have used the qemu/rbd things
[14:02] <danpb> in case its relevant, my ceph.conf is http://pastebin.test.redhat.com/64347
[14:33] <danpb> darkfader: ah ha, found the problem - it looks for libcls_rbd.so in /usr/lib/rados-classes, and since I'm on a 64-bit host, they're actually in /usr/lib64/rados-classes
[15:04] * gregorg (~Greg@ has joined #ceph
[15:06] * aliguori (~anthony@cpe-70-123-132-139.austin.res.rr.com) has joined #ceph
[16:28] * MattBenjamin (~matt@aa2.linuxbox.com) has joined #ceph
[16:29] * MattBenjamin (~matt@aa2.linuxbox.com) Quit ()
[16:30] * MattBenjamin (~matt@aa2.linuxbox.com) has joined #ceph
[16:34] <MattBenjamin> Merging changes for our Ganesha FSAL. We had exposed some types in libceph.h->libcephfs.h. Don't think what we did is attractive. I'm ready to receive guidance. :)
[16:38] * MattBenjamin (~matt@aa2.linuxbox.com) has left #ceph
[16:38] * MattBenjamin (~matt@aa2.linuxbox.com) has joined #ceph
[17:03] * adjohn (~adjohn@50-0-92-177.dsl.dynamic.sonic.net) has joined #ceph
[17:31] * gregorg_taf (~Greg@ has joined #ceph
[17:31] * gregorg (~Greg@ Quit (Read error: Connection reset by peer)
[18:18] <sagewk> mattbenjamin: where can we look at the patches/changes? is there a git tree somewhere?
[18:19] <sagewk> danpb: we need to make the default in common/config_opts.h suck in that default location from the configure libdir or whatever
[18:21] <danpb> sagewk: yep, that would seem best
[18:21] * ajm- (adam@adam.gs) has left #ceph
[18:21] * ajm- (adam@adam.gs) has joined #ceph
[18:25] * andret (~andre@pcandre.nine.ch) Quit (Ping timeout: 480 seconds)
[18:25] * andret (~andre@pcandre.nine.ch) has joined #ceph
[18:28] * jojy (~jojyvargh@75-54-231-2.lightspeed.sntcca.sbcglobal.net) has joined #ceph
[18:33] * joshd (~joshd@aon.hq.newdream.net) has joined #ceph
[18:44] * Tv|work (~Tv|work@aon.hq.newdream.net) has joined #ceph
[18:56] <sagewk> danpb: pushed a patch to master
[18:59] <gregaf> sagewk: joshd: I'm running teuthology and crashing an OSD and the log's backtrace doesn't include line numbers…any idea why?
[19:00] <sagewk> the one in the log never does/did
[19:00] <sagewk> only gdb does that
[19:00] <gregaf> hrmm, I coulda sworn it did
[19:00] <sagewk> the backtrace() call only gives you byte offsets in the function
[19:00] <danpb> sagewk: thanks, I'll get it pulled into Fedora
[19:00] <sagewk> assert() does...
[19:00] <sagewk> danpb: great
[19:01] <gregaf> ah, that's what it is, okay
[19:01] <sagewk> danpb: hey, have you looked at the recent ceph.spec changes?
[19:02] <danpb> sagewk: you mean spec changes in Fedora, or in upstream GIT ?
[19:02] <sagewk> trying to get the .spec file in to shape and cleanly building on fedora, rhel, sles, and opensuse, but i don't really know what i'm doing
[19:02] <sagewk> upstream git
[19:02] <sagewk> ideally they would be one and the same :)
[19:03] <danpb> its at least reasonable to have one spec that works with fedora and rhel - we do that for libvirt already
[19:03] <danpb> when you add in sles it gets a bit more hairy, because they have different policies to fedora in some cases
[19:03] <sagewk> danpb: the upstream one is failing in fedora 14+15 i586 on build.opensuse.org, for instance: https://build.opensuse.org/project/show?project=home%3Aliewegas
[19:04] <sagewk> danpb: yeah, there's lots of %if %{defined suse_version} in there now
[19:05] <danpb> sagewk: you mean the
[19:05] <danpb> /home/abuild/rpmbuild/BUILD/ceph-0.36/src/common/simple_spin.cc:42: undefined reference to `__sync_bool_compare_and_swap_4'
[19:05] <danpb> error ?
[19:06] <danpb> co-incidentally i got this very same error this morning, when trying to build on F14 i686 hosts - x86_64 was fine, only i686 failed for me
[19:45] * noahdesu (~nwatkins@kyoto.soe.ucsc.edu) has joined #ceph
[20:03] <sagewk> danpb: yeah.. ig uess that's not a packaging error per se. weird, though... the code is doing __sync_bool_compare_and_swap on a __u32
[20:03] <sagewk> maybe it should just be unsigned or something?
[20:03] * sagewk hates autoconf
[20:08] <danpb> sagewk: some web pages suggest its a problem caused by -march=i386
[20:08] <danpb> sagewk: which is something that's added by default on Fedora i $RPM_OPT_FLAGS
[20:09] <Tv|work> oh heh, i386
[20:09] <Tv|work> can't you settle on, like, i486 ;)
[20:09] <danpb> sagewk: you might want to try doing s/i386/i486/ in the $RPM_OPT_FLAGS and seeing if that solves it
[20:09] <danpb> gcc is supposed to fallback to heavier function calls on i386, but clearly its not doing that properly, so using i486 is probably easiest
[20:10] <Tv|work> why would you specify what setting *that* low? http://gcc.gnu.org/onlinedocs/gcc/i386-and-x86_002d64-Options.html
[20:10] <Tv|work> i mean, generic is pretty much i686, show me the person who doesn't have at least a pentium...
[20:10] <sagewk> danpb: how would i do a substition on RPM_OPT_FLAGS from inside the .spec?
[20:10] <danpb> Tv|work: i can't remember exactly why - its in email threas somewhere, but Fedora does -march=i386 -mtune=i686
[20:11] <Tv|work> danpb: because somebody had an early pentium machine that they really really loved?
[20:12] <Tv|work> some googling tells me fedora switched from i386 to i586 default in 2009: http://lists.fedoraproject.org/pipermail/devel-announce/2009-March/000378.html
[20:12] <Tv|work> http://fedoraproject.org/wiki/Features/ArchitectureSupport
[20:12] <danpb> sagewk: before this line of ceph.spec.in
[20:13] <danpb> try doing something like
[20:13] <danpb> export RPM_OPT_FLAGS=`echo $RPM_OPT_FLAGS | sed -e 's/i386/i486/'`
[20:13] <Tv|work> http://fedoraproject.org/wiki/Features/F12X86Support says i686 for F12 -- etc
[20:15] <sagewk> danpb: oh right, this section is just shell
[20:20] <danpb> Tv|work: hmm, yes, it could be the suse builders don't have a redhat-rpm-config installed to set the opts correctly then
[20:25] <danpb> Tv|work: sagewk ah yes, i confirmed the machine i had this problem on was missing 'redhat-rpm-config' package. installing that fixes the compile flags to -march=i686 -mtune=atom
[20:28] * danpb (~berrange@ Quit (Quit: Leaving.)
[21:36] * votz (~votz@pool-108-52-121-23.phlapa.fios.verizon.net) has joined #ceph
[23:19] <sagewk> does this usage seem reasonable to everyone? https://github.com/NewDreamNetwork/ceph/commit/0f0793e36250d5e87a94b87ad41f0f681410815f
[23:20] <Tv|work> sagewk: no but that's a bigger rant ;)
[23:21] <Tv|work> sagewk: (tl;dr of the bigger rant: you're imposing your mental model as the architect on users, who don't need/want the same stuff)
[23:21] <sagewk> let's hear it
[23:21] <Tv|work> sadly i don't have time for it now :(
[23:21] <Tv|work> but the "monitor commands" are on my list of things ceph would be better without
[23:21] <Tv|work> sure, have the same functionality, but the interface is miserable
[23:22] <sagewk> the purpose is for issuing low level commands to specific compoents for the purposes of debugging and disaster recovery
[23:22] <Tv|work> 1) can be done on the node itself, 2) needs way way more automation than "tell osd.42 to foo"
[23:23] <gregaf> these changes allow us to give the user feedback from the nodes; it sure as hell isn't ideal but it's closer to good than what we have right now; ergo it's reasonable and should go in
[23:23] <gregaf> ;)
[23:23] <sagewk> this is something you would never automate
[23:24] <Tv|work> if it's not meant to be automated, see 1) ;)
[23:24] <sagewk> requiring it to be run on the node itself (say, interacting via the admin socket) might be an option.. it'll be may more tedious though
[23:25] <Tv|work> sagewk: i consider ssh to be nicer than "oh there's another network service that'll run near-arbitrary actions, and btw i'm not 100% convinced the authentication works right"
[23:25] <sagewk> this is for things like "benchmark the low level object store", "turn on heap profiling", "change debug log level"
[23:25] <Tv|work> 1
[23:26] <Tv|work> 1)
[23:26] <sagewk> any other opinions?
[23:26] <Tv|work> i prefer "ssh mynode bar" over "ceph tell mynode bar" any day; there's no real value added in my mind
[23:27] <Tv|work> and off to interview
[23:27] * cp (~cp@ has joined #ceph
[23:31] <sagewk> in this case, it's "ssh `some command to figure out host that osd.123 runs on` newdaemontool $path_to_admin_socket/osd.123 foo"

These logs were automatically created by CephLogBot on irc.oftc.net using the Java IRC LogBot.