[Gluster-devel] RFC - "Connection Groups" concept

Joe Landman landman at scalableinformatics.com
Thu Jun 27 14:37:23 UTC 2013


On 06/27/2013 10:30 AM, Stephan von Krawczynski wrote:
> On Wed, 26 Jun 2013 11:46:36 -0400
> Jeff Darcy <jdarcy at redhat.com> wrote:
>
>> On 06/27/2013 09:37 AM, Stephan von Krawczynski wrote:
>>> On Wed, 26 Jun 2013 11:04:07 -0400 Jeff Darcy <jdarcy at redhat.com>
>>> wrote:
>>>
>>>> [Jeff on UUIDs]
>>>
>>> I generally vote against using UUIDs and for IPs. In runtime I can
>>> easily switch an IP in a replacement situation, but can I switch a
>>> UUID in the same easy manner?
>>
>> I don't see why that would be problematic.  The UUIDs we're talking
>> about aren't tied to hardware.  They're essentially big random numbers
>> we assign ourselves.  IIRC they're just stored in a file, so they can be
>> trivially copied from a system to its replacement.  The problem is
>> precisely that DNS names and IP addresses aren't good *system*
>> identifiers.  For one thing, they refer to interfaces rather than
>> systems (which might have many interfaces).  For another, even that
>> association is too transient.  Such IDs are convenient for referring to
>> a system *at a specific point in time* but not permanently, and a
>> permanent ID for the whole system is something we really need.  It sure
>> would be nice if the networking community would stop ****ing around when
>> it comes to multi-homed or mobile hosts, but they don't seem inclined to
>> so the rest of us have to fall back on other established patterns for
>> identifying hosts separately from their addresses.
>
> Now the exact reason why IPs are always better than your idea of UUIDs is
> because they are linked to interfaces. You _want_ your fs node to use exactly
> _this_ interface and not _some_ interface because you then have a real chance
> to direct your traffic like you want it and not like glusterfs currently
> thinks could be best.

No.  One of the largest issues we and our customers have been having for 
years has been tightly tying gluster volume creation to single IP 
addresses.  This makes multihomed usage, well, problematic, at best. 
Worse than this, is the use of the DNS name (or other name) which, 
exactly as Jeff indicates, tightly ties the brick/mount point to a 
particular interface.

What you want is an abstraction which represents a group of connections 
on a brick or set of bricks.  UUID does this.

FWIW: we'd been using split horizon/multi-horizon DNS to try to work 
around these issues for multihomed systems, and inevitably, they break 
some aspect of the file system.

UUID with connection groups, and even better, primary/secondary, or even 
weights on each connection, would be quite nice ... a breath of fresh air.

-- 
Joseph Landman, Ph.D
Founder and CEO
Scalable Informatics, Inc.
email: landman at scalableinformatics.com
web  : http://scalableinformatics.com
        http://scalableinformatics.com/siflash
phone: +1 734 786 8423 x121
fax  : +1 866 888 3112
cell : +1 734 612 4615




More information about the Gluster-devel mailing list