Camelia, the Perl 6 bug

IRC log for #gluster, 2013-05-25

| Channels | #gluster index | Today | | Search | Google Search | Plain-Text | summary

All times shown according to UTC.

Time Nick Message
00:18 recidive joined #gluster
00:58 mohankumar__ joined #gluster
02:23 recidive joined #gluster
02:23 badone joined #gluster
02:34 sjoeboo_ joined #gluster
02:38 sjoeboo_ left #gluster
03:10 hjmangalam1 joined #gluster
03:13 portante joined #gluster
03:21 badone joined #gluster
03:56 hagarth joined #gluster
04:06 hjmangalam1 joined #gluster
04:13 Tobarja joined #gluster
04:21 Tobarja I tried to replace-brick from an existing drive to a new drive in the same box. It seems to have failed (volume replace-brick: failed: Commit failed on localhost.), but is hung up as a task that I can't kill. Any ideas how to make it go away?
05:35 edong23 joined #gluster
05:50 badone joined #gluster
06:12 nightwalk joined #gluster
06:24 lnxsix` joined #gluster
06:58 nightwalk joined #gluster
06:59 piotrektt joined #gluster
09:08 hagarth joined #gluster
09:21 lalatenduM joined #gluster
10:11 ekuric joined #gluster
10:14 ni291187 joined #gluster
10:15 ni291187 left #gluster
10:16 ni291187 joined #gluster
10:16 ni291187 left #gluster
10:17 ni291187 joined #gluster
10:58 chirino joined #gluster
11:02 hagarth joined #gluster
11:09 duerF joined #gluster
11:19 rb2k joined #gluster
11:36 duerF joined #gluster
11:48 recidive joined #gluster
12:32 isomorphic joined #gluster
12:46 recidive_ joined #gluster
12:58 vrturbo joined #gluster
13:04 duerF joined #gluster
13:19 lnxsix joined #gluster
13:36 chirino joined #gluster
13:50 mriv joined #gluster
13:52 samppah @yum repo
13:52 glusterbot samppah: The official glusterfs packages for RHEL/CentOS/SL are available here: http://goo.gl/s077x
13:52 samppah @latest
13:52 glusterbot samppah: The latest version is available at http://goo.gl/zO0Fa . There is a .repo file for yum or see @ppa for ubuntu.
14:06 mriv joined #gluster
14:19 rb2k joined #gluster
14:23 mriv joined #gluster
15:00 hjmangalam1 joined #gluster
15:03 ekuric joined #gluster
15:15 ingard_ joined #gluster
15:31 chirino joined #gluster
15:50 chirino joined #gluster
15:54 ni291187 joined #gluster
15:54 ni291187 left #gluster
15:59 ujjain joined #gluster
15:59 mohankumar__ joined #gluster
16:03 ni291187 joined #gluster
16:03 ni291187 left #gluster
16:04 ni291187 joined #gluster
16:04 ni291187 left #gluster
16:05 ni291187 joined #gluster
16:06 ni291187 left #gluster
16:12 aliguori joined #gluster
16:12 aliguori_ joined #gluster
16:36 chirino joined #gluster
16:51 chirino joined #gluster
16:53 Tobarja left #gluster
17:25 chirino joined #gluster
18:00 nightwalk anyone have any favorite oss imap server software that they use on top of gluster? dovecot seems to be the best-supported, but several crippling bugs (ex: randomly attempting to double-bind ports, causing a fatal error at startup) keep it from being useful in production
18:01 nightwalk I don't see anything regarding cyrus-imapd, but they argue fervently against nfs-based mail, so I'm guessing that would also be a bad choice
18:07 GabrieleV joined #gluster
18:29 JoeJulian I use cyrus, but not on GlusterFS. Their 1 file per email storage doesn't seem like it would be efficient, though I haven't actually tried.
18:29 JoeJulian It
18:29 JoeJulian It's possible it wouldn't be that bad since it uses a bdb index thus might not have to do directory listings.
18:59 nightwalk well, I was thinking courier would be the best bet, but it's not available on centos/rhel/etc and I don't like maintaining my own rpms if I can help it
19:00 nightwalk I'd be tempted to go with database-backed mail storage rather than using gluster, but that doesn't lend itself well to active-active configurations (and everything else is a waste)
19:01 edong23 joined #gluster
19:01 samppah i have been happy with dovecot but i'm not really using it with gluster
19:02 NuxRo i wouldnt like to put my maildirs on glusterfs ... :)
19:02 NuxRo i'm the kind of guy who's on many lists and never deletes anything ..
19:02 samppah would be fun to try dovecot's object storage functions
19:03 NuxRo i would have thought cyrus' one file per mailbox is much more gluster-friendly
19:03 nightwalk dovecot has nfs options which have to be configured to make it cluster-capable across nfs-based spools, but other bugs that have been present for years that have supposedly been fixed creep up every once in a while. Not really production ready if they can't even fix simple bugs like I mentioned earlier and KEEP them fixed
19:04 nightwalk NuxRo: mail is one of the few places where it'd be the *safest* to put it on gluster, actually...
19:05 nightwalk the one mail per file thing combined with filenames that couldn't possibly conflict (i.e. they use the receiving mail servers' hostname in them) make the actual message storage as safe as can be
19:05 NuxRo nightwalk: but but but, glusterfs doesn't like many (small) files
19:05 nightwalk even *if* gluster screwed itself over, a simple rsync from one to the other would transfer whatever files (read: emails) didn't exist on one server to the other
19:05 nightwalk lol. gluster doesn't really like *anything*, does it? :)
19:06 NuxRo it's okay for VM images#
19:06 nightwalk LOL
19:06 nightwalk that's the thing gluster has handled the *worst*
19:06 NuxRo well, that has not been everyone's experience
19:06 samppah it would definetly be good for gluster to see some real use cases
19:06 NuxRo ymmv
19:07 nightwalk many small, random writes to super-huge files is difficult for ANY active-active clustered filesystem
19:07 nightwalk you basically need lockable byte ranges, and iirc, that's only materialized in gluster within the past couple minors
19:07 samppah nightwalk: at least they have been working for it.. there are lots of performance improvements in 3.4 and qemu alreay has support for libgfapi
19:08 NuxRo I've had good small file performance on loop mounted raw images from gluster rather than on glusterfs directly
19:08 nightwalk I'd still be tempted to use something actually designed more for vm storage for that sort of thing *shrug*
19:08 nightwalk ceph, clvm, etc...
19:09 NuxRo my philosophy is that they all have their problems
19:09 samppah +1
19:09 NuxRo if you stick with one at least you get good at fixing and working well with it
19:09 * nightwalk is waiting for HAMMER2 code to materialize ;)
19:09 NuxRo lol
19:09 NuxRo it's been how many years now?
19:09 NuxRo plus, it would require dragonfly to run ..
19:10 nightwalk they announced it before they even had any designs drawn up. haven't heard anything since
19:10 samppah lol
19:10 nightwalk and it doesn't really matter which BSD it'd run on
19:11 NuxRo we'll see, there is definitely room for more clustered filesystems
19:11 NuxRo for now i like glusterfs, with all its bedside manners#
19:11 nightwalk as long as it's !linux, !windows, and !osx I'd be happy
19:12 NuxRo lol
19:12 NuxRo Haiku!
19:12 nightwalk :)
19:13 nightwalk seriously though. It doen't matter how good the filesystem is if the host OS is so buggy that it can't stay running for five minutes
19:13 NuxRo have you been running Win98 again?..
19:14 nightwalk pick a windows version, and it inevitably blue screens...a LOT
19:14 NuxRo i hate windows and ms generally, but technically they got a lot better, windows 2008r2 is quite stable for example
19:15 stickyboy Yep. :)
19:15 nightwalk and yet, it still blue screens if you get a driver does anything at all unexpected
19:15 stickyboy And iwlwifi on Linux has been crap from 3.4 - 3.9... meh.
19:15 NuxRo you will still hit ms moronisms :)
19:15 NuxRo stickyboy: i run iwlwifi on 3.9 right now, seems ok
19:16 nightwalk stickyboy: that's because you're running 2.6.39+. anything up to 2.6.38 was pretty stable
19:16 nightwalk then they removed the bkl, introduced pv_ops code, and, well...we have a mass of highly experimental code being pushed on the masses now :)
19:16 samppah about bsod: http://www.theregister.co.uk​/2012/01/13/tieto_emc_crash/
19:16 glusterbot <http://goo.gl/zTe6c> (at www.theregister.co.uk)
19:17 nightwalk NuxRo: ms moronisms == not following the standards (SUS, posix, etc) ;)
19:18 NuxRo standards are for pussies
19:18 NuxRo ms and google make their own :)
19:18 stickyboy NuxRo: You're lucky. :\  This is on newer Centrino chipsets coming with new-ish SNB / IVB systems.  Not sure what you're running?
19:18 JoeJulian The "many small files" thing, wrt file-per-message storage, would only matter if it's pulling directory listings. Since cyrus stores the list of emails in a bdb table, theoretically it would almost never pull a directory listing. As long as it's always referencing the pre-existing file directly there /shouldn't/ be much overhead.
19:19 samppah JoeJulian: yeah, dovecot does lots of caching and indexing aswell
19:19 NuxRo JoeJulian: thanks for the explanation, I'll keep it in mind should i ever move my inbox over..
19:19 NuxRo stickyboy: 02:00.0 Network controller: Intel Corporation Centrino Advanced-N 6230 (rev 34)
19:20 stickyboy NuxRo: Heh, close to mine.  01:00.0 Network controller: Intel Corporation Centrino Advanced-N 6235 (rev 24)
19:20 nightwalk JoeJulian: that's part of what you have to disable to make dovecot cluster-capable over nfs/gluster, I believe
19:20 JoeJulian But I still think an object-store with a distributed lookup model would be much more resilient and able to scale across datacenters better.
19:20 stickyboy NuxRo: We should probably stop the off topic... but I'm curious what distro you're on?
19:20 badone joined #gluster
19:21 NuxRo stickyboy: both Centos 6 kernel and latest mainline works well - well, almost, after a number of suspends wifi is dead and need to reboot
19:21 samppah nightwalk: you can always direct users to specific servers.. not sure if that's a good way though ;)
19:21 NuxRo JoeJulian: that would be interesting for imap ... it would certainly scale
19:21 * JoeJulian considers starting a project that he knows he doesn't have time for...
19:21 NuxRo hehe
19:21 nightwalk JoeJulian: the closest we have to that is database storage via sql atm. nosql and others *might* work if someone hacked out a bit of code, but there isn't any in the oss ecosystem currently afaik
19:22 NuxRo stickyboy: i run centos 6
19:22 JoeJulian There's not. But really object store + elasticsearch would be fairly easy to implement.
19:22 nightwalk samppah: the idea is to use lvs to eliminate *all* points of failure
19:22 JoeJulian Plus it would offer a very easy full-text backend search capability.
19:22 NuxRo hm, search on objects.. i imagine that would be slow
19:23 NuxRo unless you index it separately, like Zimbra does etc
19:23 JoeJulian Right, elasticsearch for the index
19:23 nightwalk full text searching is overrated
19:23 JoeJulian It's auto-scaling and rack-aware.
19:24 NuxRo heh
19:24 nightwalk I'd rather err on the side of caution (read: privacy) and performance/efficiency
19:24 JoeJulian Heh, I use the full-text search capability of cyrus-imap all the time.
19:24 NuxRo i'd like to look one day into mail in MySQL/MariaDB + Galera
19:24 nightwalk I mean, if you don't know the thread you discussed something in, you're already well on your way to Alzheimer's :)
19:24 NuxRo lol
19:25 JoeJulian Yeah, but the thread 2 years ago?
19:25 nightwalk not really a problem for me *shrug*
19:26 NuxRo well, it's important to be able to find stuff
19:26 JoeJulian Then there's the business issues that pop up. "Why don't we do X" ... search emails for the explanation you gave last time this question came up...
19:26 nightwalk Then there's the matter of searching being something that should really be happening client-side, but that's a whole other discussion in and of itself...
19:27 JoeJulian I highly disagree with client-side searches.
19:27 * NuxRo is out, ttyl &
19:27 nightwalk servers aren't as scalable as clients
19:27 JoeJulian client = cell phone
19:28 nightwalk there can be thousands of clients, but only one server. so thousands of searches server-side would kill the server whereas tens of thousands of searches client-side would probably not make it sweat very much
19:28 nightwalk yes, yes, I realize that mail is i/o-bound, of course
19:29 nightwalk it still doesn't take many clients doing intensive searches to kill even a moderately well-spec'ed server though :)
19:30 JoeJulian That's an engineering problem.
19:30 nightwalk It's a practical problem, actually
19:31 nightwalk cost of increases in processing power grow exponentially beyond a certain point
19:31 stickyboy nightwalk: True (killing a moderately-spec'd server).  It reminds me of the discussion a few days ago about kernel.org's git mirroring.
19:31 stickyboy nightwalk: Each client eats ~1GB of RAM + net... whew!!
19:31 nightwalk stickyboy: DBAs share my view too ;)
19:32 nightwalk all of those fusion io devices and ramsans and so forth are ungodly expensive
19:32 nightwalk and they add little to overall processing power and stability when everything is running from a single (or pair) of beefy servers
19:33 JoeJulian Is data transfer that much cheaper? I have 8 imap clients. Client-side search means that I have to store the same data on all 8 clients that I have on the server. On my cell phone, that's not even a possibility.
19:33 nightwalk you're leveraging storage on the clients for redundancy purposes (some offsite) that they'd otherwise not be using for productive purposes. I don't see the problem there
19:36 JoeJulian One of the great things about philosophical discussions is that "you" are always right and "they" are always wrong (regardless of which side of the argument you're on). :D
19:36 nightwalk in fact, when google decided to become super-hostile to tor, client-side caching allowed me to keep data that google had otherwise cut me off from. I'm sure it's not difficult to imagine other, similar benefits for clients as well :)
19:36 nightwalk JoeJulian: not really. everyone is right, just working from different contexts ;)
19:37 JoeJulian :D
19:37 JoeJulian I'm off. I'm going to reboot into the unholy OS for the sole purpose of playing Firefall (the only thing I have installed).
19:38 stickyboy JoeJulian: Game? :)
19:38 JoeJulian Yep
19:38 stickyboy Cool.  Night!
19:38 stickyboy Err... bye. :)
19:38 JoeJulian After saying I didn't have time to make a new mail server, I'm going to waste time playing a game. :D
19:39 JoeJulian ... the irony...
19:39 nightwalk you COULD use that time to set it up under wine
19:39 nightwalk ;)
19:39 JoeJulian Tried. I forget what's missing, but something is.
19:40 nightwalk did you try winetricks?
19:40 JoeJulian yep
19:40 nightwalk some games need one of the directx libs, but that's all I usually have to install *shrug*
19:40 JoeJulian It was some .net component
19:41 nightwalk ah...probably an issue like I had with DDO then
19:41 nightwalk With DDO, it needed .net for its launcher but the game worked fine without. So everyone just uses the oss pylotro wrapper instead :)
19:42 nightwalk the wine folk are getting pretty close to getting Mono support up-to-par with .net though. I'd be surprised if they didn't have win7-level .net support to 90% or better within 2 years
19:43 nightwalk anyway...have fun with that unholy OS. Hope it doesn't crash TOO many times for you :)
19:43 JoeJulian Thanks. :)
19:56 glusterbot joined #gluster
20:55 ni291187 joined #gluster
20:55 ni291187 left #gluster
20:56 ni291187 joined #gluster
20:56 ni291187 left #gluster
20:58 ni291187 joined #gluster
21:20 recidive joined #gluster
21:26 ricky-ticky joined #gluster
22:15 badone_ joined #gluster
22:54 yinyin joined #gluster
23:16 atrius_ joined #gluster
23:56 joelwallis joined #gluster
23:58 hagarth joined #gluster

| Channels | #gluster index | Today | | Search | Google Search | Plain-Text | summary