[ovirt-users] high number of IOPS from ovirt hosts

Andrea Fagiani andrea.fagiani at immobiliare.it
Tue Feb 14 15:08:57 UTC 2017


Correct,
coincidentally, there no instances of systemd-udevd going haywire either.

Andrea


On 14/02/2017 15:20, Yaniv Kaul wrote:
>
>
> On Tue, Feb 14, 2017 at 2:15 PM, Andrea Fagiani 
> <andrea.fagiani at immobiliare.it <mailto:andrea.fagiani at immobiliare.it>> 
> wrote:
>
>     Hi,
>
>     just a quick update on the issue. The high amount of IOPS was
>     indeed due to having the hosted-engine on the EQL storage; we
>     worked around the issue by moving the hosted-engine to a gluster
>     storage instead.
>
>
> And now there are no high amount of IOPS on the Gluster storage?
> Y.
>
>     Thanks to everyone involved,
>     Andrea
>
>
>     On 13/02/2017 13:24, Nir Soffer wrote:
>>     On Mon, Feb 13, 2017 at 12:32 PM, Andrea Fagiani
>>     <andrea.fagiani at immobiliare.it
>>     <mailto:andrea.fagiani at immobiliare.it>> wrote:
>>
>>         Hi Nir,
>>
>>         the amount of IOPS is pretty much constant, attaching the
>>         logs from the last hour or so.
>>
>>
>>     Andrea, can you send also logs from ovirt-hosted-engine-ha?
>>     (I'm not sure about the name of the service or the location of
>>     the logs)
>>
>>         Andrea
>>
>>
>>         On 13/02/2017 10:13, Nir Soffer wrote:
>>>         On Mon, Feb 13, 2017 at 11:05 AM, Andrea Fagiani
>>>         <andrea.fagiani at immobiliare.it
>>>         <mailto:andrea.fagiani at immobiliare.it>> wrote:
>>>
>>>             Thanks for the heads up, I'll take a look at Mark's thread.
>>>             We are presenting 2 storage LUNs to a single ovirt
>>>             cluster and yes, all hosts are hosted engine hosts.
>>>
>>>             Thanks,
>>>             Andrea
>>>
>>>
>>>         I think we already have hosted engine bug about doing to
>>>         many calls to vdsm
>>>         that cause reloads of lvm data; this can cause increased io.
>>>
>>>         Can you attach vdsm and hosted engine agent logs showing the
>>>         timeframe when
>>>         you have lot of iops on the server?
>>>
>>>         Nir
>>>
>>>
>>>
>>>
>>>             On 12/02/2017 18:27, Andrea Ghelardi wrote:
>>>
>>>                 Andrea,
>>>                 I don't want to point any fingers but Mark Greenall
>>>                 faced some issues with Ovirt 4.06/4.1 linked to
>>>                 equallogic as well.
>>>                 You may want to search posts in this ML with subject
>>>                 " Ovirt 4.0.6 guests 'Not Responding' " to see if
>>>                 you find any similarities.
>>>                 How many storages are you presenting to how many
>>>                 clusters?
>>>                 Also, are your nodes all Hosted Engine Hosts? In my
>>>                 old configuration, normal hosts read IOPS ranged
>>>                 from 50 to 500 while HE ones peaked even ~3000
>>>                 Cheers
>>>                 AG
>>>
>>>
>>>                 -----Original Message-----
>>>                 From: Andrea Fagiani
>>>                 [mailto:andrea.fagiani at immobiliare.it
>>>                 <mailto:andrea.fagiani at immobiliare.it>]
>>>                 Sent: Friday, February 10, 2017 5:56 PM
>>>                 To: Andrea Ghelardi <a.ghelardi at iontrading.com
>>>                 <mailto:a.ghelardi at iontrading.com>>; users at ovirt.org
>>>                 <mailto:users at ovirt.org>
>>>                 Subject: Re: [ovirt-users] high number of IOPS from
>>>                 ovirt hosts
>>>
>>>                 Hi Andrea,
>>>
>>>                 thanks for the insight; no I wasn't able to pinpoint
>>>                 the issue to a single host, the load seems to be
>>>                 evenly spread.
>>>                 We have another cluster running 4.0.6 on a
>>>                 compellent as well and we haven't experienced any
>>>                 similar issues.
>>>
>>>                 Thanks,
>>>                 Andrea
>>>
>>>
>>>                 On 10/02/2017 17:17, Andrea Ghelardi wrote:
>>>
>>>                     Hello Andrea,
>>>                     I saw that happening on my HE hosts connected to
>>>                     a COMPELLENT SC040
>>>                     when I was running Ovirt 3.5
>>>
>>>                     All HE hosts were hammering the SAN with 2~4k
>>>                     IOPS each
>>>
>>>                     Upgrading to 4.0.5 solved my case.
>>>                     Are you able to check if IOPS are generated from
>>>                     a particular host or rather spread among all nodes?
>>>
>>>                     Cheers
>>>                     AG
>>>
>>>
>>>                     -----Original Message-----
>>>                     From: users-bounces at ovirt.org
>>>                     <mailto:users-bounces at ovirt.org>
>>>                     [mailto:users-bounces at ovirt.org
>>>                     <mailto:users-bounces at ovirt.org>] On
>>>                     Behalf Of Andrea Fagiani
>>>                     Sent: Friday, February 10, 2017 12:47 PM
>>>                     To: users at ovirt.org <mailto:users at ovirt.org>
>>>                     Subject: [ovirt-users] high number of IOPS from
>>>                     ovirt hosts
>>>
>>>                     Hi,
>>>
>>>                     We've got a 8-host cluster running oVirt 4.1;
>>>                     each host is connected via iscsi to a EQL 4100
>>>                     storage (2x).
>>>                     Most of the VMs use barely any IOPS, however on
>>>                     our storage controller we've noticed a constant
>>>                     steam of reads, about 2000 IOPS; any idea what
>>>                     might cause the ovirt hosts to hammer on the
>>>                     storage so much?
>>>
>>>                     Looking at iotop, it seems that systemd-udevd is
>>>                     being activated every few seconds (and its IO%
>>>                     is usually fairly high) so I assume this might
>>>                     be somehow related.
>>>
>>>                     Thanks,
>>>                     Andrea
>>>                     _______________________________________________
>>>                     Users mailing list
>>>                     Users at ovirt.org <mailto:Users at ovirt.org>
>>>                     http://lists.ovirt.org/mailman/listinfo/users
>>>                     <http://lists.ovirt.org/mailman/listinfo/users>
>>>
>>>
>>>             _______________________________________________
>>>             Users mailing list
>>>             Users at ovirt.org <mailto:Users at ovirt.org>
>>>             http://lists.ovirt.org/mailman/listinfo/users
>>>             <http://lists.ovirt.org/mailman/listinfo/users>
>>>
>>>
>>
>>
>
>
>     _______________________________________________
>     Users mailing list
>     Users at ovirt.org <mailto:Users at ovirt.org>
>     http://lists.ovirt.org/mailman/listinfo/users
>     <http://lists.ovirt.org/mailman/listinfo/users>
>
>

-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://lists.ovirt.org/pipermail/users/attachments/20170214/6de1c7c3/attachment-0001.html>


More information about the Users mailing list