[Gluster-users] Tools for the admin

Thu Jul 14 22:01:54 UTC 2011

On 07/14/2011 04:35 PM, John Mark Walker wrote:
> Joe - thanks for taking the time to write this up. It sounds like the
> issue this is designed to fix is related to the GFID mismatch issue
> that we released a preventive fix for today.

Hi John

   We need to get this into testing soon.

>
> The sanity checks could be useful, though. Does today's release
> change anything with respect to your tools?

Don't know yet.  If we can get access to more of the internals 
monitoring, we should be able to craft better tools for this.  Our plan 
is to eventually get something like a parallel scan and "fix" tool that 
does best practices for the fix.  Not quite an fsck (thats at a 
different level).  But a real sanity checker.

I'd imagine it would be useful for the CloudFS group too ...

>
> -John Mark Gluster Community Guy
>
>
> ________________________________________ From:
> gluster-users-bounces at gluster.org [gluster-users-bounces at gluster.org]
> on behalf of Joe Landman [landman at scalableinformatics.com] Sent:
> Wednesday, July 13, 2011 9:15 PM To: gluster-users Subject:
> [Gluster-users] Tools for the admin
>
> Hi folks
>
> We have run into a number of problems with missing files (among
> other things).  So I went hunting for the files.  Along the way, I
> came up with some very simple sanity checks and tools for helping to
> correct situations.  They will not work on striped data ... sorry.
>
> Sanity check #1: conservation of number of files
>
> The sum of the number of files on your backing stores (excluding
> links and directories) should equal (with possible minor variance due
> to gluster internals) the sum of the number of files (excluding links
> and directories) in your gluster volumes.
>
> If you have say, 6 bricks, each with nearly 1M files, and a dht
> volume built from those bricks, you really ... REALLY ... shouldn't
> have only 1.8M files in your volume.  If you do, then some files are
> missing from the volume (really).  You can tell what these files are,
> as they have no xattr.  Yeah.  Really.
>
> How can you enumerate what you have?
>
> Simple.  Meet file_accounting.pl  (available at
> (http://download.scalableinformatics.com/gluster/utils/)
>
> This handy utility will tell you important things about your file
> system.
>
> [root at jr4-1 temp]# /data/tiburon/install/scan/file_accounting.pl
> --bspath=/data/brick-sdc2/dht/ Number of entries: 944604 Number of
> links  : 6711 Number of dir    : 102825 Number of files  : 834794
>
> --bspath is the "backing store path", where the files reside.  It
> works just as well on your gluster volume, which allows you to
> inspect your sanity with appropriate sums.
>
> So you need to copy these into the volume.  And move them out of the
> way first  before copy in.
>
> Which leads to tool #2 and #3.  First, you need to scan your backing
> store file system for the files.
>
> Tool #2: scan_gluster.pl
>
> /data/tiburon/install/scan/scan_gluster.pl
> --bspath=/data/brick-sdc2/dht
>> /data/brick-sdc2/temp/sdc2.data
>
> will grab lots of nice info about the file, including the
> attributes. You can now use grep against the sdc2.data file and look
> only for 'attr=,' and those will be things gluster knows varying
> degrees about in your file system.  Some things, specifically files
> with this condition, yeah, those are missing files.  The ones I was
> trying to find.
>
> If you have a user who notes that files occasionally go missing,
> yeah, this can help you find them if they exist on the backing store.
> Which they probably do.
>
> The next tool is dangerous.  So far all we have done is to scan the
> backing store.  Now we are going to make changes.  No, don't worry,
> its actually ... almost ... safe.  We do a file move to another
> location (preferably on the same device/mount point in the backing
> store), then a copy into gluster volume (yes you need to mount it on
> your brick nodes). The danger is in modifying a gluster file system
> backend.  Don't do this.  Ever.  Unless 3/4 of your files go
> missing.
>
> And, by the way, we have a handy dandy --md5 switch on there, if you
> want the scan to take forever.
>
> Tool #3: data_mover.pl
>
> This will do the dirty work.  It parses the output of scan_gluster,
> and makes changes.  There is a --dryrun option for those who want to
> try it, and a -T number   option to specify the number of changes to
> make to the file.  Allows you to try it (hence the T ... for TRY) on
> some number of files.  It will preserve ownership and permission mask
> (ohhh ahhh ... shiny!).  The --tmp option happily sets your temporary
> directory. Verbose and debug should be obvious.
>
> nohup ./data_mover.pl --data sdd2-nomd5.data --debug --verbose
> --tmp `pwd`/tmp  -T 2000000>>  out 2>&1&
>
>
> Note:  all of these tools currently use /opt/scalable/bin/perl as
> the interpreter.  This is because our Perl build (5.12.3) includes
> all the bits we need to make this work.  If you want to use them, you
> are welcome to change /opt/scalable/bin/perl to /usr/bin/perl, and
> they you will have to install a few modules
>
> cpan Getopt::Lucid File::ExtAttr
>
> If you have an issue with either, please let me know.
>
> We can turn these into binaries if someone needs.  Source is at
>
> http://download.scalableinformatics.com/gluster/utils/
>
> Let me know (offline) if you run into problems if you decide to give
> them a try.  Note, they are GPL2 (no license tag on them), no
> warranty, and data_mover.pl will MOST DEFINITELY DESTROY DATA.  We
> aren't liable for any damages if you use it.  Caveat Emptor.  Let the
> admin beware. Did I mention that data_mover.pl WILL DESTROY YOUR
> DATA?   I am not sure if I did.  So here it is again.  data_mover.pl
> WILL DESTROY YOUR DATA.
>
> Don't use these unless you have a backup.  Especially data_mover.pl.
> Because IT WILL MOST DEFINITELY DESTROY YOUR DATA.  Might even bite
> your dog, egg your house, and do all sorts of other nastiness.  It
> will increase entropy in the universe.
>
> But if you are staring at the rear end of 3.8M missing files,
> wondering WTF, mebbe ... that data lossage thing doesn't sound so
> bad.  Especially if you can reverse it.
>
> So feel free to look them over.  I plan to hone them over time, and
> refine them.  Add documentation even.  If they prove useful enough
> for people to use, please let me know.  And they are GPL v2.
>
> The site is http://download.scalableinformatics.com/gluster/utils/ .
>
> -- Joseph Landman, Ph.D Founder and CEO Scalable Informatics, Inc.
> email: landman at scalableinformatics.com web  :
> http://scalableinformatics.com
> http://scalableinformatics.com/sicluster phone: +1 734 786 8423 x121
> fax  : +1 866 888 3112 cell : +1 734 612 4615
> _______________________________________________ Gluster-users mailing
> list Gluster-users at gluster.org
> http://gluster.org/cgi-bin/mailman/listinfo/gluster-users

-- 
Joseph Landman, Ph.D
Founder and CEO
Scalable Informatics Inc.
email: landman at scalableinformatics.com
web  : http://scalableinformatics.com
        http://scalableinformatics.com/sicluster
phone: +1 734 786 8423 x121
fax  : +1 866 888 3112
cell : +1 734 612 4615