On 2/28/19 9:41 AM, Glen Baars wrote:
> Hello Wido,
>
> I have looked at the libvirt code and there is a check to ensure that
> fast-diff is enabled on the image and only then does it try to get the real
> disk usage. The issue for me is that even with fast-diff enabled it takes
> 25min to get the space usage for a 50TB image.
>
> I had considered turning off fast-diff on the large images to get around to
> issue but I think that will hurt my snapshot removal times ( untested )
>
Can you tell a bit more about the Ceph cluster? HDD? SSD? DB and WAL on SSD?
Do you see OSDs spike in CPU or Disk I/O when you do a 'rbd du' on these
images?
Wido
> I can't see in the code any other way of bypassing the disk usage check but I
> am not that familiar with the code.
>
> -------------------
> if (volStorageBackendRBDUseFastDiff(features)) {
> VIR_DEBUG("RBD image %s/%s has fast-diff feature enabled. "
> "Querying for actual allocation",
> def->source.name, vol->name);
>
> if (virStorageBackendRBDSetAllocation(vol, image, &info) < 0)
> goto cleanup;
> } else {
> vol->target.allocation = info.obj_size * info.num_objs;
> }
> ------------------------------
>
> Kind regards,
> Glen Baars
>
> -----Original Message-----
> From: Wido den Hollander <w...@42on.com>
> Sent: Thursday, 28 February 2019 3:49 PM
> To: Glen Baars <g...@onsitecomputers.com.au>; ceph-users@lists.ceph.com
> Subject: Re: [ceph-users] Mimic 13.2.4 rbd du slowness
>
>
>
> On 2/28/19 2:59 AM, Glen Baars wrote:
>> Hello Ceph Users,
>>
>> Has anyone found a way to improve the speed of the rbd du command on large
>> rbd images? I have object map and fast diff enabled - no invalid flags on
>> the image or it's snapshots.
>>
>> We recently upgraded our Ubuntu 16.04 KVM servers for Cloudstack to Ubuntu
>> 18.04. The upgrades libvirt to version 4. When libvirt 4 adds an rbd pool it
>> discovers all images in the pool and tries to get their disk usage. We are
>> seeing a 50TB image take 25min. The pool has over 300TB of images in it and
>> takes hours for libvirt to start.
>>
>
> This is actually a pretty bad thing imho. As a lot of images people will be
> using do not have fast-diff enabled (images from the past) and that will kill
> their performance.
>
> Isn't there a way to turn this off in libvirt?
>
> Wido
>
>> We can replicate the issue without libvirt by just running a rbd du on the
>> large images. The limiting factor is the cpu on the rbd du command, it uses
>> 100% of a single core.
>>
>> Our cluster is completely bluestore/mimic 13.2.4. 168 OSDs, 12 Ubuntu 16.04
>> hosts.
>>
>> Kind regards,
>> Glen Baars
>> This e-mail is intended solely for the benefit of the addressee(s) and any
>> other named recipient. It is confidential and may contain legally privileged
>> or confidential information. If you are not the recipient, any use,
>> distribution, disclosure or copying of this e-mail is prohibited. The
>> confidentiality and legal privilege attached to this communication is not
>> waived or lost by reason of the mistaken transmission or delivery to you. If
>> you have received this e-mail in error, please notify us immediately.
>> _______________________________________________
>> ceph-users mailing list
>> ceph-users@lists.ceph.com
>> http://lists.ceph.com/listinfo.cgi/ceph-users-ceph.com
>>
> This e-mail is intended solely for the benefit of the addressee(s) and any
> other named recipient. It is confidential and may contain legally privileged
> or confidential information. If you are not the recipient, any use,
> distribution, disclosure or copying of this e-mail is prohibited. The
> confidentiality and legal privilege attached to this communication is not
> waived or lost by reason of the mistaken transmission or delivery to you. If
> you have received this e-mail in error, please notify us immediately.
>
_______________________________________________
ceph-users mailing list
ceph-users@lists.ceph.com
http://lists.ceph.com/listinfo.cgi/ceph-users-ceph.com