[CINDER] - RBD backend reporting

Gorka Eguileor geguileo at redhat.com
Thu May 13 08:31:58 UTC 2021


On 12/05, Steven Relf wrote:
> Hey list.
>  
> Currently when using an RBD backend using the default settings, total_capacity_gb is reported as MAX_AVAIL + USED bytes, converted into GiB. This to mean seems a little odd, as I would expect that total_capacity_gb should report the total size in GiB of the backend cluster.
>  
> This of course can be fixed by adding "report_dynamic_total_capacity = false" into the cinder.conf section for the rbd backend.

Hi Steve,

That's the problem of having to keep backward compatibility, that even
if the driver is doing something non-standard and it's inconveniencing
some users [1][2], we cannot just change the behavior, as it could make
trouble for another group of users who are currently relying on the
current behavior.

That's why I had to set the default to true (keep old behavior) in the
fix.

[1]: https://bugs.launchpad.net/cinder/+bug/1712549
[2]: https://bugs.launchpad.net/cinder/+bug/1706057

> This works fine for ceph clusters where all pools consume from a single disk type/root. But in clusters where you have multiple root's or device types it does not work correctly.
>  

I don't currently have a system like that to check it, but I would
assume current code works as intended: It gets the stats and quota for
the pool and uses the most limiting value of the two.

As far as I know the stats should be returning the aggregate of the
different disks that form the pool.

I would like to better understand the difference between what is being
reported and what is expected in your environment.

Could you share the output of the following commands?:

  $ ceph -f json-pretty df

  $ ceph -f json-pretty osd pool get-quota <POOL-NAME>

Also what values is Cinder reporting, and what values it should be
reporting?

Thanks.


> Is this proving to be a pain point for anyone else, or is it just me, and if it is proving a problem for others, im happy to write a patch.
>  

I haven't heard anyone having that problem before, though that doesn't
mean there are not people suffering it as well.

Cheers,
Gorka.

> Im thinking something that gets the pools crushrule and works out the total_capacity_gb, based off the total available capacity of a pool based on its root/crushrules.
>  
> Rgds
> Steve.
>
> The future has already arrived. It's just not evenly distributed yet - William Gibson
>
>




More information about the openstack-discuss mailing list