Mailing List Archive

Hi James,

James Penick <jpenick@gmail.com> wrote:
>Hey folks,
> Does anyone have experience using zookeeper or redis to handle HA failover
>in cinder clusters?

I'm guessing you mean failover of an active/passive cinder-volume
service?

>I know there's docs on pacemaker, however we already
>have the other two installed and don't want to add yet another component to
>package and maintain in our clusters.

I'm afraid I don't, but if you make any progress on this, please let
me know as it would be great to document:

- how this would work
- any pros and cons vs. Pacemaker

and maybe I can help with that.

One particular question: if the node running the service becomes
unreachable, is it safe to fail it over straight away, or is fencing
required first? (I'm pretty sure I've asked this same question
before, but I can't remember the answer - sorry!)

_______________________________________________
OpenStack-operators mailing list
OpenStack-operators@lists.openstack.org
http://lists.openstack.org/cgi-bin/mailman/listinfo/openstack-operators

On 9/9/2018 10:58 PM, Adam Spiers wrote:
> Hi James,
>
> James Penick <jpenick@gmail.com> wrote:
>> Hey folks,
>> Does anyone have experience using zookeeper or redis to handle HA
>> failover
>> in cinder clusters?
>
> I'm guessing you mean failover of an active/passive cinder-volume
> service?
>
>> I know there's docs on pacemaker, however we already
>> have the other two installed and don't want to add yet another
>> component to
>> package and maintain in our clusters.
>
> I'm afraid I don't, but if you make any progress on this, please let
> me know as it would be great to document:
>
>  - how this would work
>  - any pros and cons vs. Pacemaker
>
> and maybe I can help with that.
>
> One particular question: if the node running the service becomes
> unreachable, is it safe to fail it over straight away, or is fencing
> required first?  (I'm pretty sure I've asked this same question
> before, but I can't remember the answer - sorry!)
James,

I echo Adam's input.  I have only heard of people implementing with
pacemaker but there is no reason that this couldn't be tried with other
HA solutions.

If you are able to try it and document it would be great to add
documentation here:  [1]

Also, Gorka Eguileor is a good contact as he has been doing much of the
work on HA Cinder though his focus is on Active/Active HA.

Let us know if you have any further questions.

Thanks!
Jay


_______________________________________________
OpenStack-operators mailing list
OpenStack-operators@lists.openstack.org
http://lists.openstack.org/cgi-bin/mailman/listinfo/openstack-operators

Additionally if you require some resources to test this against OpenLab is
a great resource - https://openlabtesting.org provides more info -
https://github.com/theopenlab/resource-requests/issues/new - is where you
can skip having to go through the site to do so

On Mon, Sep 10, 2018 at 3:19 PM Jay S Bryant <jungleboyj@gmail.com> wrote:

>
> On 9/9/2018 10:58 PM, Adam Spiers wrote:
> > Hi James,
> >
> > James Penick <jpenick@gmail.com> wrote:
> >> Hey folks,
> >> Does anyone have experience using zookeeper or redis to handle HA
> >> failover
> >> in cinder clusters?
> >
> > I'm guessing you mean failover of an active/passive cinder-volume
> > service?
> >
> >> I know there's docs on pacemaker, however we already
> >> have the other two installed and don't want to add yet another
> >> component to
> >> package and maintain in our clusters.
> >
> > I'm afraid I don't, but if you make any progress on this, please let
> > me know as it would be great to document:
> >
> > - how this would work
> > - any pros and cons vs. Pacemaker
> >
> > and maybe I can help with that.
> >
> > One particular question: if the node running the service becomes
> > unreachable, is it safe to fail it over straight away, or is fencing
> > required first? (I'm pretty sure I've asked this same question
> > before, but I can't remember the answer - sorry!)
> James,
>
> I echo Adam's input. I have only heard of people implementing with
> pacemaker but there is no reason that this couldn't be tried with other
> HA solutions.
>
> If you are able to try it and document it would be great to add
> documentation here: [1]
>
> Also, Gorka Eguileor is a good contact as he has been doing much of the
> work on HA Cinder though his focus is on Active/Active HA.
>
> Let us know if you have any further questions.
>
> Thanks!
> Jay
>
>
> _______________________________________________
> OpenStack-operators mailing list
> OpenStack-operators@lists.openstack.org
> http://lists.openstack.org/cgi-bin/mailman/listinfo/openstack-operators
>


--
Kind regards,

Melvin Hillsman
mrhillsman@gmail.com
mobile: (832) 264-2646

Ah ok so this is a case of no ones documented it, but it's do-able.

If anyone out there has done it we'd be happy to take your notes! Otherwise
we'll figure it out and upstream the process.

thanks!
-James

On Mon, Sep 10, 2018 at 2:18 PM Jay S Bryant <jungleboyj@gmail.com> wrote:

>
> On 9/9/2018 10:58 PM, Adam Spiers wrote:
> > Hi James,
> >
> > James Penick <jpenick@gmail.com> wrote:
> >> Hey folks,
> >> Does anyone have experience using zookeeper or redis to handle HA
> >> failover
> >> in cinder clusters?
> >
> > I'm guessing you mean failover of an active/passive cinder-volume
> > service?
> >
> >> I know there's docs on pacemaker, however we already
> >> have the other two installed and don't want to add yet another
> >> component to
> >> package and maintain in our clusters.
> >
> > I'm afraid I don't, but if you make any progress on this, please let
> > me know as it would be great to document:
> >
> > - how this would work
> > - any pros and cons vs. Pacemaker
> >
> > and maybe I can help with that.
> >
> > One particular question: if the node running the service becomes
> > unreachable, is it safe to fail it over straight away, or is fencing
> > required first? (I'm pretty sure I've asked this same question
> > before, but I can't remember the answer - sorry!)
> James,
>
> I echo Adam's input. I have only heard of people implementing with
> pacemaker but there is no reason that this couldn't be tried with other
> HA solutions.
>
> If you are able to try it and document it would be great to add
> documentation here: [1]
>
> Also, Gorka Eguileor is a good contact as he has been doing much of the
> work on HA Cinder though his focus is on Active/Active HA.
>
> Let us know if you have any further questions.
>
> Thanks!
> Jay
>
>

James,

Sorry, I forgot to include the link to our HA documentation in the
earlier e-mail:
https://docs.openstack.org/cinder/latest/contributor/high_availability.html

Jay


On 9/10/2018 3:39 PM, James Penick wrote:
> Ah ok so this is a case of no ones documented it, but it's do-able.
>
> If anyone out there has done it we'd be happy to take your notes!
> Otherwise we'll figure it out and upstream the process.
>
> thanks!
> -James
>
> On Mon, Sep 10, 2018 at 2:18 PM Jay S Bryant <jungleboyj@gmail.com
> <mailto:jungleboyj@gmail.com>> wrote:
>
>
> On 9/9/2018 10:58 PM, Adam Spiers wrote:
> > Hi James,
> >
> > James Penick <jpenick@gmail.com <mailto:jpenick@gmail.com>> wrote:
> >> Hey folks,
> >> Does anyone have experience using zookeeper or redis to handle HA
> >> failover
> >> in cinder clusters?
> >
> > I'm guessing you mean failover of an active/passive cinder-volume
> > service?
> >
> >> I know there's docs on pacemaker, however we already
> >> have the other two installed and don't want to add yet another
> >> component to
> >> package and maintain in our clusters.
> >
> > I'm afraid I don't, but if you make any progress on this, please let
> > me know as it would be great to document:
> >
> >  - how this would work
> >  - any pros and cons vs. Pacemaker
> >
> > and maybe I can help with that.
> >
> > One particular question: if the node running the service becomes
> > unreachable, is it safe to fail it over straight away, or is fencing
> > required first?  (I'm pretty sure I've asked this same question
> > before, but I can't remember the answer - sorry!)
> James,
>
> I echo Adam's input.  I have only heard of people implementing with
> pacemaker but there is no reason that this couldn't be tried with
> other
> HA solutions.
>
> If you are able to try it and document it would be great to add
> documentation here:  [1]
>
> Also, Gorka Eguileor is a good contact as he has been doing much
> of the
> work on HA Cinder though his focus is on Active/Active HA.
>
> Let us know if you have any further questions.
>
> Thanks!
> Jay
>