[Ocfs2-users] ocfs2 - disk usage inconsistencies

Fabio Corazza fabio at newbay.com
Sun Sep 24 13:18:58 PDT 2006 

I'm getting the same here. 4GB used in reality but 8GB displayed by df.
This is a 100% more of space consumed despite the real usage of the file
system. Initially I addressed the problem to the actual cluster size and
block size (64k and 4k), since we have a LOT of directories and a LOT of
small files, but still the 100% of usage in respect of the real usage is
too much. The filesystem is being used by a web application which does a
lot of read/writes since it stores photos on users albums, and this will
cause huge SAN space problems in the future (when we will reach 600GB of
space this will require 1.2TB).

I've opened a ticket internally to reformat the filesystem with 4k of
cluster size and 2k of block size, but I think that this will not solve
completely the problem, seeing what the other users report to this
mailing list.

With GFS we never got those problems, but I wanted to get rid of it
since I wanted to get away from the RedHat cluster suite.

Being said this, we should probably discuss more about this issue.



Fabio

Matthew Flusche wrote:
> I unmounted the file system from each node and ran the fsck.ocfs2 on it.
> That cleaned up the orphan_dirs but I did not get my 17 GB of space
> back.   Any other suggestions?  Am I going to have to
> backup,format,restore to fix this?
> 
> -----Original Message-----
> From: Sunil Mushran [mailto:Sunil.Mushran at oracle.com] 
> Sent: Wednesday, September 20, 2006 2:31 PM
> To: Matthew Flusche
> Cc: Ocfs2-users at oss.oracle.com
> Subject: Re: [Ocfs2-users] ocfs2 - disk usage inconsistencies
> 
> Yes.
> 
> Matthew Flusche wrote:
>> When I stat the only file listed in //orphan_dir:0002 it shows 0
>> clusters.  There are only directories listed in //orphan_dir:0000 and
>> none of the directories have contents.  So my next courses of action
>> (since I don't know what has these unlinked files opened) would be to
>> stop every process accessing this file system?  Possibly umount/mount
>> the file system? would fsck.ocfs2 clean this up?
>>
>> -----Original Message-----
>> From: Sunil Mushran [mailto:Sunil.Mushran at oracle.com] 
>> Sent: Wednesday, September 20, 2006 1:52 PM
>> To: Matthew Flusche
>> Cc: Ocfs2-users at oss.oracle.com
>> Subject: Re: [Ocfs2-users] ocfs2 - disk usage inconsistencies
>>
>> Another node or that node itself.
>>
>> As far as the filesize goes, "ls -l" does not give the ondisk size.
>> Do "stat <inodenum>" on the unlinked files and see the "Clusters".
>>
>> Matthew Flusche wrote:
>>   
>>> There has been a lot of file system activity recently.
>>>
>>> I have files in orphan_dir:0000 and orphan_dir:0002.  But that
> doesn't
>>> seem to account for the 17 GB missing.  The truncate logs seem clean.
>>> So having files in orphan_dir:0000 is telling me that the node in
> slot
>>>     
>> 0
>>   
>>> deleted files and another node(s) still has the file open, correct?
>>>
>>> Matt
>>>
>>> debugfs: ls -l //orphan_dir:0000
>>>         16              drwxr-xr-x  13     0     0          774144
>>> 10-Sep-2006 00:08 .
>>>         10              drwxr-xr-x   6     0     0            4096
>>> 2-May-2006 16:11 ..
>>>         3052182         drwxrwxrwx   0   501   500            4096
>>> 19-Jul-2006 17:50 00000000002e9296
>>>         8234094         drwxrwxrwx   0   501   500            4096
>>> 19-Jul-2006 17:50 00000000007da46e
>>>         13063783        drwxrwxrwx   0   501   500            4096
>>> 19-Jul-2006 17:50 0000000000c75667
>>>         7869995         drwxrwxrwx   0   501   500            4096
>>> 22-Aug-2006 13:27 000000000078162b
>>>         3741473         drwxrwxrwx   0   501   500            4096
>>> 22-Aug-2006 13:29 0000000000391721
>>>         3351057         drwxrwxrwx   0   501   500            4096
>>> 19-Jul-2006 17:50 0000000000332211
>>>         7842503         drwxrwxrwx   0   501   500            4096
>>> 19-Jul-2006 17:50 000000000077aac7
>>>         2056493         drwxrwxrwx   0   501   500            4096
>>> 5-Sep-2006 08:53 00000000001f612d
>>>         7861894         drwxrwxrwx   0   501   500            4096
>>> 5-Sep-2006 08:53 000000000077f686
>>>         1487817         drwxrwxrwx   0   501   500            4096
>>> 5-Sep-2006 08:53 000000000016b3c9
>>>         1702439         drwxrwxrwx   0   501   500            4096
>>> 5-Sep-2006 08:53 000000000019fa27
>>>
>>> debugfs: ls -l //orphan_dir:0002
>>>         18              drwxr-xr-x   2     0     0           94208
>>> 5-Jul-2006 17:13 .
>>>         10              drwxr-xr-x   6     0     0            4096
>>> 2-May-2006 16:11 ..
>>>         4301446         -rw-r--r--   0   503   500               0
>>> 12-Aug-2006 10:40 000000000041a286
>>>
>>> -----Original Message-----
>>> From: Sunil Mushran [mailto:Sunil.Mushran at oracle.com] 
>>> Sent: Wednesday, September 20, 2006 12:32 PM
>>> To: Matthew Flusche
>>> Cc: Ocfs2-users at oss.oracle.com
>>> Subject: Re: [Ocfs2-users] ocfs2 - disk usage inconsistencies
>>>
>>> Did you remove some large files recently? If so, check the orphan_dir
>>> and truncate_log for all the slots.
>>>
>>> 1. Start debugfs:
>>> # debugfs.ocfs2 /dev/sdX
>>>
>>> 2. List system directory:
>>>  > ls -l //
>>>
>>> 3. List files in all orphan_dir(s):
>>>  > ls -l //orphan_dir:0000
>>>
>>> If there are files, means some process in the cluster is still using 
>>> that file.
>>>
>>> 4. stat all trancate_log(s):
>>>  > stat //truncate_log:0000
>>>
>>> I will be surprised if you see any bits here. If there are, do 
>>> "sync;sync;sync;" on the
>>> appropriate node.
>>>
>>> 5. You can find the appropriate node by dumping the slotmap:
>>>  > slotmap
>>> Find the slot-to-nodenum mapping. Do the sync on that node.
>>>
>>> For this and more, refer to the on-disk format support guide.
>>>
>>>     
> http://oss.oracle.com/projects/ocfs2/dist/documentation/03-disk_format.p
>>   
>>> df
>>>
>>> Matthew Flusche wrote:
>>>   
>>>     
>>>> Hi all.
>>>>
>>>> I have a 50 GB OCFS2 file system. I'm currently using ~26GB of space
> 
>>>> but df is reporting 43 GB used. Any ideas how to find out where the 
>>>> missing 17GB is at?
>>>>
>>>> The file system was formatted with a 16K cluster & 4K block size.
>>>>
>>>> Thanks,
>>>>
>>>> Matt
>>>>
>>>>
>>>>     
>>>>       
> ------------------------------------------------------------------------
>>   
>>>   
>>>     
>>>> _______________________________________________
>>>> Ocfs2-users mailing list
>>>> Ocfs2-users at oss.oracle.com
>>>> http://oss.oracle.com/mailman/listinfo/ocfs2-users
>>>>   
>>>>     
>>>>       
>>> _______________________________________________
>>> Ocfs2-users mailing list
>>> Ocfs2-users at oss.oracle.com
>>> http://oss.oracle.com/mailman/listinfo/ocfs2-users
>>>   
>>>     
> 
> _______________________________________________
> Ocfs2-users mailing list
> Ocfs2-users at oss.oracle.com
> http://oss.oracle.com/mailman/listinfo/ocfs2-users

More information about the Ocfs2-users mailing list