Open Stack

On 15/03/2016 09:51, Mark Kirkwood wrote:
> On 15/03/16 22:21, Peter Brouwer wrote:
>
>> On 10/03/2016 05:14, Mark Kirkwood wrote:
>>> Hmm, I'm confused by the phrase 'partitions that point to the same 
>>> disk':
>
>> PArtitions is used in the swift context, i.e. the partitions scheme the
>> ring-builder uses.
>
> I'm sorry but that does not make sense. The ring builder lets you add 
> *devices*. Now a device could be a partition (e.g /dev/sdc1) as 
> opposed to a complete disk (/dev/sdc) but I'm at a loss to see where 
> you are going with this, as the ring builder is essentially partition 
> un-aware.
See https://ask.openstack.org/en/question/6766/what-is-a-swift-partition/

>
> I'm assuming a whole physical disk is used, i.e.
>> filesystem created on a disk using the whole physical disk.
>> So the ring structure provides a reference to a swift partition and a
>> disk location, right?
>
> Hmm...this partition word again...you are talking about a whole disk, 
> so there is no partition (or the 'partition is = the entire disk - you 
> have added a whole disk after all).
>
>> What happens if the disk it is pointing to is full, does swift returns
>> an error to the app/client or does it try a re-lookup in an attempt to
>> find space elsewhere?
>
> I think I answered that previously. Swift can survive *some* of the 
> disks being full, but eventually you'll get a PUT failure for 
> objects/containers (a 50x http error), and you will be unable to add 
> any more until you:
>
> - add more disks to existing servers (or)
> - add more servers and disks
>
> and amend the ring with these additional devices.
>
> regards
>
> Mark

-- 
Regards,

Peter Brouwer, Principal Software Engineer,
Oracle Application Integration Engineering.
Phone:  +44 1506 672767, Mobile +44 7720 598 226
E-Mail: Peter.Brouwer at Oracle.com