Gluster users - Aug 2008 - Split-brain scenario

I apologise for taking this out of the original thread, but I didn't
feel my questions were really relevant to the original "Add/remove new
server volumes" subject
> > 2. Second, when using "files replicating strategy"
(scheduler), is it
> >possible to remove a server node  without stopping the hole cluster
> >(ex. hardware maintaining reasons, add more disk/ram to the node ...)?
> if you have multiple servers AFR'ed, if one server is down, the rest
> will continue after the timeout period.  when that servers back on
> line it will coninue AFRing to it.
> Remember, AFR does healing when files are accessed, so the server
> that's offline may be out of sync until any updated files are accessed
again.
Does this mean that the temporarily offline server A will be unaware
of changes to files made on server B until they are accessed on server
A, or on server B?

How does this extend to files that are created on B when A is offline
- at what point would A become aware of them?  On a "ls", or only when
an explicit attempt is made to open a file that's known to exist (even
though it doesn't yet exist)?

Finally, what sort of conflict resolution is in Gluster AFR?  If,
rather than A going offline, the communication between A and B is
unavailable, and different changes are made to the same file on A and
B, what happens?  I realise this is a highly undesirable situation,
but it'd be helpful to understand how Gluster / AFR deals with it.

I hope everyone has a great weekend,


Rob

On Sat, Aug 16, 2008 at 12:59 AM, Rob S <rtshilston at gmail.com>
wrote:
> I apologise for taking this out of the original thread, but I didn't
> feel my questions were really relevant to the original "Add/remove new
> server volumes" subject
>
>> > 2. Second, when using "files replicating strategy"
(scheduler), is it
>> >possible to remove a server node  without stopping the hole cluster
>> >(ex. hardware maintaining reasons, add more disk/ram to the node
...)?
>
>> if you have multiple servers AFR'ed, if one server is down, the
rest
>> will continue after the timeout period.  when that servers back on
>> line it will coninue AFRing to it.
>> Remember, AFR does healing when files are accessed, so the server
>> that's offline may be out of sync until any updated files are
accessed again.
>
> Does this mean that the temporarily offline server A will be unaware
> of changes to files made on server B until they are accessed on server
> A, or on server B?
Correct.
>
> How does this extend to files that are created on B when A is offline
> - at what point would A become aware of them?  On a "ls", or only
when
> an explicit attempt is made to open a file that's known to exist (even
> though it doesn't yet exist)?
On "cd" 0 bytes files will be created, however on open the file will
be sync'ed
( you could experiement it in your setup )
>
> Finally, what sort of conflict resolution is in Gluster AFR?  If,
> rather than A going offline, the communication between A and B is
> unavailable, and different changes are made to the same file on A and
> B, what happens?  I realise this is a highly undesirable situation,
> but it'd be helpful to understand how Gluster / AFR deals with it.
As of now split-brain case is not being handled, but we will be checking
in the code to handle it very soon. What we will do is we will mark that
a write is being done on the other child and unmark it after write is
successful. During open if we know that marks for both the subvols
are set, then we know that its a split brain case.
>
> I hope everyone has a great weekend,
>
>
> Rob
>
> _______________________________________________
> Gluster-users mailing list
> Gluster-users at gluster.org
> http://zresearch.com/cgi-bin/mailman/listinfo/gluster-users
>