[Gluster-users] Self heal doesn't seem to work when file is updated

Mon Mar 21 17:30:30 UTC 2011

hi Mohit,
Please use these rules:    
1) The attributes with 0sAAAAAAAAAAAAAAAA tell whether it needs data-self-heal/ metadata self-heal or entry self-heal.
2) The other cases are different gfid on the same files.
3) File/Directory exists on one brick but not on the other brick in the replica.

you should not inspect/getfattr the files on the mount point, you should do it on individual bricks.

Pranith.

----- Original Message -----
From: "Mohit Anchlia" <mohitanchlia at gmail.com>
To: "Pranith Kumar. Karampuri" <pranithk at gluster.com>, gluster-users at gluster.org
Sent: Monday, March 21, 2011 10:25:05 PM
Subject: Re: [Gluster-users] Self heal doesn't seem to work when file is updated

Hi Pranith,

Will you have some time to send me the use cases so I can write the
script? Or get these scripts?

Thanks,
Mohit
On Wed, Mar 16, 2011 at 10:24 AM, Mohit Anchlia <mohitanchlia at gmail.com> wrote:
> Can you please let me know the cases so that I can get started with
> the script? Or if you are already planning on doing it then can you
> please let me know?
>
> Self healing seems to be of concern for majority of people and for me
> too. So atleast having some kind of monitoring will help reduce that
> concern :)
>
>
> On Mon, Mar 14, 2011 at 2:44 PM, Mohit Anchlia <mohitanchlia at gmail.com> wrote:
>> Thanks for this additional info. It really helps!
>>
>> If you give me the scenarios I can come up with the script and send it
>> to this user group. So far I have:
>>
>> 1. Different afr on node that was receiving updated
>> 2. Different gfid. Only way is to compare gfid is to run getfattr on
>> every replica node.
>>
>> BTW: I noticed that getfattr <mount>/file also heals. Don't understand
>> why but it does :)
>>
>> On Fri, Mar 11, 2011 at 5:59 PM, Pranith Kumar. Karampuri
>> <pranithk at gluster.com> wrote:
>>> hi Mohit,
>>>     I have sent a patch for the bug you have raised for self-heal not working in one case already.
>>>
>>> For a file, it should have either different extended attributes for client-0/1 (assuming two replica) and same gfid extended attribute.
>>> or different extended attribute for gfid.
>>> example:
>>> pranith @ /mnt/client
>>> 07:24:05 :) $ sudo getfattr -d -m "trusted*" /tmp/1/file1
>>> getfattr: Removing leading '/' from absolute path names
>>> # file: tmp/1/file1
>>> trusted.afr.vol-client-0=0sAAAAAAAAAAAAAAAA
>>> trusted.afr.vol-client-1=0sTwAAAgAAAAAAAAAA
>>> trusted.gfid=0s9JoVEu1WQQCykafEGcz74Q==
>>>
>>> pranith @ /mnt/client
>>> 07:24:34 :) $ sudo getfattr -d -m "trusted*" /tmp/2/file1
>>> getfattr: Removing leading '/' from absolute path names
>>> # file: tmp/2/file1
>>> trusted.afr.vol-client-0=0sAAAAAAAAAAAAAAAA
>>> trusted.afr.vol-client-1=0sAAAAAAAAAAAAAAAA
>>> trusted.gfid=0s9JoVEu1WQQCykafEGcz74Q==
>>>
>>> After self-heal completes they become:
>>> pranith @ /mnt/client
>>> 07:27:21 :) $ sudo getfattr -d -m "trusted*" /tmp/1/file1
>>> getfattr: Removing leading '/' from absolute path names
>>> # file: tmp/1/file1
>>> trusted.afr.vol-client-0=0sAAAAAAAAAAAAAAAA
>>> trusted.afr.vol-client-1=0sAAAAAAAAAAAAAAAA
>>> trusted.gfid=0s9JoVEu1WQQCykafEGcz74Q==
>>>
>>>
>>> pranith @ /mnt/client
>>> 07:27:26 :) $ sudo getfattr -d -m "trusted*" /tmp/2/file1
>>> getfattr: Removing leading '/' from absolute path names
>>> # file: tmp/2/file1
>>> trusted.afr.vol-client-0=0sAAAAAAAAAAAAAAAA
>>> trusted.afr.vol-client-1=0sAAAAAAAAAAAAAAAA
>>> trusted.gfid=0s9JoVEu1WQQCykafEGcz74Q==
>>>
>>> Thanks
>>> Pranith.
>>> ----- Original Message -----
>>> From: "Mohit Anchlia" <mohitanchlia at gmail.com>
>>> To: gluster-users at gluster.org
>>> Sent: Saturday, March 12, 2011 4:31:16 AM
>>> Subject: Re: [Gluster-users] Self heal doesn't seem to work when file is        updated
>>>
>>> Can somone from gluster developer community please tell respond how to
>>> find if self heal worked?
>>>
>>> Or better yet how to determine if a file needs to be healed? Please
>>> let me know as this very important aspect.
>>>
>>> On Thu, Mar 10, 2011 at 9:54 AM, William L. Sebok <wls at astro.umd.edu> wrote:
>>>> It also won't work well when the bricks are under use and might be changing.
>>>>
>>>> Bill Sebok      Computer Software Manager, Univ. of Maryland, Astronomy
>>>>        Internet: wls at astro.umd.edu     URL: http://furo.astro.umd.edu/
>>>>
>>>> On Thu, Mar 10, 2011 at 09:39:44AM -0800, Mohit Anchlia wrote:
>>>>> It doesn't look like a good way of checking when it involves millions
>>>>> of files and something you don't want to run every now and then. I was
>>>>> expecting something from gluster commands to straight away provide
>>>>> that info. Something that can then be monitored continuously. It's
>>>>> very important to have right version of files in production and only
>>>>> gluster knows which were healed.
>>>>>
>>>>> On Thu, Mar 10, 2011 at 1:40 AM, Marcus Bointon
>>>>> <marcus at synchromedia.co.uk> wrote:
>>>>> > On 9 Mar 2011, at 20:08, Mohit Anchlia wrote:
>>>>> >
>>>>> >> Asking my question again since it's very important to know how to find
>>>>> >> out if self healing worked.
>>>>> >
>>>>> > Do an rsync with dry-run option on the back-end stores? They should be identical.
>>>>> >
>>>>> > Marcus
>>>>
>>> _______________________________________________
>>> Gluster-users mailing list
>>> Gluster-users at gluster.org
>>> http://gluster.org/cgi-bin/mailman/listinfo/gluster-users
>>>
>>
>