[Gluster-devel] 1.4 AFR vs Roadmap Page

[Martin Fick]

Hi guys,

I would like to ask a question about the current 1.4 AFR implementation.  On 
the roadmap page:
  http://www.gluster.org/docs/index.php/GlusterFS_Roadmap

it says:

  #  AFR atomic writes - handle total power-loss, split-brains,.. (bullet-proof 
AFR)


While I know that there have been some improvements to AFR mentioned on this 
list, mainly file locking on append (or something like that), I really did not 
get the impression that any of the improvements were close to making AFR 
"bullet-proof" or anti-split brain.  Are there many changes in this regards 
that have not been explained here?  My impression was that the 1.4 AFR still 
couldn't handle the following simple split brain scenario yet:

Server A & B AFRed

1) Server B goes down
2) Server A has file FOO deleted on it
3) Server A goes down
4) Server B comes up

Will split brain be detected, or will server B gladly serve up file FOO still?  
Please, don't take my question as a criticism, I really am just trying to 
figure out whether the roadmap page is hyping up the capabilities of 1.4 AFR or 
not, and if so if perhaps the wording should be softened? :)

Thanks,

-Martin