[Gluster-devel] IMPORTANT: AFR Users

[Anand Babu Periasamy]

Brandon Lamb wrote:
> On Fri, May 9, 2008 at 2:53 PM, Anand Babu Periasamy <ab at gnu.org.in> wrote:
>> RACE CONDITION IN AFR:
>> ----------------------
>> * Who is affected?
>> If you are using AFR in a multi-user environment and your applications
>> are writing to common files without locks, there is a possibility
>> CORRUPTION.
>>
>> GlusterFS is expected to write atomically across replicated volumes
>> when the files are opened with O_APPEND mode, irrespective of locks.
>> Fix in works will ensure all writes to be atomic always. We will
>> make an announcement shortly.
>>
>> Thanks to Gordan Bobic and Martin Fick for identifying this bug.
>>
>> --
>> Anand Babu Periasamy
>> GPG Key ID: 0x62E15A31
>> Blog [http://ab.freeshell.org]
>> The GNU Operating System [http://www.gnu.org]
>> Z RESEARCH Inc [http://www.zresearch.com]
>>
>>
>>
>> _______________________________________________
>> Gluster-devel mailing list
>> Gluster-devel at nongnu.org
>> http://lists.nongnu.org/mailman/listinfo/gluster-devel
> 
> Is corruption the correct word here, or would it be more accurate to
> say "inconsistent"? I suppose it is a point of view.
> 
> It is just that to me corruption means the data in my files is
> unusable, which is slightly different than the data is "good" on both
> copies, they just are not the SAME. I could still at least pick one to
> use, whereas corruption to means I have no good data.
> 
> Unimportant maybe but corruption might freak people out a little more
> 
> =P

You are correct. No information is lost, but files get inconsistent.
This is still a very critical issue to us. Being in the file system
space, we should always be paranoid about reliability and correctness.

--
Anand Babu Periasamy
GPG Key ID: 0x62E15A31
Blog [http://ab.freeshell.org]
The GNU Operating System [http://www.gnu.org]
Z RESEARCH Inc [http://www.zresearch.com]