[openstack-dev] [Fuel] Ceph Public Network Setting

Zhou Zheng Sheng / 周征晟 zhengsheng at awcloud.com
Thu Jun 25 03:36:48 UTC 2015


Hi,

I also notice that in latest ceph-deploy source code, it allows us to
define public_network and cluster_network argument when calling
'ceph-deploy new'. It also checks if monitor IP address is in either of
the two networks.

https://github.com/ceph/ceph-deploy/blob/master/ceph_deploy/new.py#L158

I'm not sure that Ceph monitor operating in the public network is just
an artificial requirement, or it's an inherent precondition when they
implement Ceph and breaking the precondition might lead to some strange
problems. If it's the latter case, when we change Ceph public network,
we have to change Ceph monitor IP address configuration at the same time.

on 2015/06/25 11:01, Zhou Zheng Sheng / 周征晟 wrote:
> Hello Andrew,
>
> I did some similar modifications in Fuel 6.0 as [1][2]. However in [3]
> it says Ceph monitor have to run on the public network.
>
> In the file "deployment/puppet/ceph/manifests/conf.pp", we can see it
> calls 'ceph-deploy new ${::hostname}:${::internal_address}'.
> In "deployment/puppet/ceph/manifests/mon.pp", it calls 'ceph-deploy
> mon create ${::hostname}:${::internal_address}'.
>
> After we set Ceph public network to OpenStack storage network, Ceph
> monitor still runs on management network. I can see in QEMU command
> line, it connects to Ceph via management network IP address. Though
> this setup does not follow the rule suggested in [3], it probably
> works because Ceph monitor only provides PG location information, and
> then QEMU should talk to OSDs directly using the storage network
> addresses.
>
> That's why I didn't take your patches as-is. I meant to setup a Fuel
> 6.1 environment and test the modifications before I can give an
> accurate comment.
>
> [3]
> http://ceph.com/docs/master/rados/configuration/network-config-ref/#monitor-ip-tables
>
> on 2015/06/25 03:20, Andrew Woodward wrote:
>> Zhou,
>>
>> As mentioned, Please review [1][2]. This is the interface we will
>> support as we implement the advanced networking parts. In your case,
>> just fip the role map in your patch to nailgun and the library
>> interface will remain the same.
>>
>> [1] https://review.openstack.org/#/c/194434/
>> [2] https://review.openstack.org/#/c/194438/
>>
>> On Wed, Jun 24, 2015 at 3:21 AM Stanislav Makar <smakar at mirantis.com
>> <mailto:smakar at mirantis.com>> wrote:
>>
>>     Hello
>>     My five cents :)
>>     I also very proud of our 6.1 release
>>     Unfortunately we still do not have this separation for ceph
>>     public and cluster networks in 6.1.
>>
>>     One little hint
>>     In 6.1 we have granular deployment
>>     (https://docs.mirantis.com/fuel/fuel-master/release-notes.html#granular-deployment-based-on-pluggable-tasks)
>>     and all tasks which are connected with ceph deployment are in
>>
>>     https://github.com/stackforge/fuel-library/tree/stable/6.1/deployment/puppet/osnailyfacter/modular/ceph
>>
>>     To have your patch working in 6.1 you should hack all these files now
>>
>>     If you have question feel free to ask.
>>
>>     Thanks.
>>
>>
>>     -- 
>>     All the best,
>>     Stanislav Makar
>>     skype: makar_stanislav 
>>     irc: stamak
>>
>>     On Wed, Jun 24, 2015 at 12:04 PM, Sergii Golovatiuk
>>     <sgolovatiuk at mirantis.com <mailto:sgolovatiuk at mirantis.com>> wrote:
>>
>>         Hi Zhou,
>>
>>         Try Fuel 6.1 where we have a lot of very nice features
>>         including 'advanced networking'. Feel free to read release
>>         notes [1]. It's one of the most significant releases with
>>         many many features and improvements. I am very proud of it.
>>
>>         [1] https://docs.mirantis.com/fuel/fuel-master/release-notes.html
>>
>>         --
>>         Best regards,
>>         Sergii Golovatiuk,
>>         Skype #golserge
>>         IRC #holser
>>
>>         On Wed, Jun 24, 2015 at 11:42 AM, Zhou Zheng Sheng / 周征晟
>>         <zhengsheng at awcloud.com <mailto:zhengsheng at awcloud.com>> wrote:
>>
>>             Thank you guys!
>>
>>             I'm reading the advanced networking spec and think it's
>>             good. I'll try
>>             to comment it. It's too epic for me for now. I wrote a
>>             patch to move
>>             Ceph public network to storage network. It's based on
>>             Fuel 6.0, so I'm
>>             trying to rebase it to lastest master then maybe ask you
>>             for a review.
>>
>>             on 2015/06/23 19:21, Igor Kalnitsky wrote:
>>             > Hello,
>>             >
>>             > That makes sense to me. Still, I want to point that
>>             we're going to
>>             > implement advanced networking and with this feature
>>             you'll be able to
>>             > assign every single network role to any network.
>>             >
>>             > That means, you'll be able to assign ceph network role
>>             to storage,
>>             > management or  whatever-you-want network. Sounds cool,
>>             ha? :)
>>             >
>>             > Feel free to read a design spec [1].
>>             >
>>             > Thanks,
>>             > Igor
>>             >
>>             > [1]: https://review.openstack.org/#/c/115340/
>>             >
>>             > On Tue, Jun 23, 2015 at 1:13 PM, Zhou Zheng Sheng / 周征晟
>>             > <zhengsheng at awcloud.com
>>             <mailto:zhengsheng at awcloud.com>> wrote:
>>             >> Hi!
>>             >>
>>             >> I notice that in OpenStack deployed by Fuel, Ceph
>>             public network is on
>>             >> management network. In some environments, not all NICs
>>             of a physical
>>             >> server are 10Gb. Sometimes 1 or 2 among the NICs on a
>>             machine may be
>>             >> 1Gb. Usually on this type of machine we assign
>>             management network to 1Gb
>>             >> NIC, and storage network to 10Gb NIC. If Ceph public
>>             network is with
>>             >> management network, the QEMU accesses Ceph using
>>             management network, and
>>             >> the performance is not optimal.
>>             >>
>>             >> In a small deployment, cloud controller and Ceph OSD
>>             may be assigned to
>>             >> the same machine, so it would be more effective to
>>             keep Ceph client
>>             >> traffic separated from MySQL, RabbitMQ, Pacemaker
>>             traffic. Maybe it's
>>             >> better to place Ceph public network on the storage
>>             network. Agree?
>>             >>
>>             >> --
>>             >> Best wishes!
>>             >> Zhou Zheng Sheng, Software Developer
>>             >> Beijing AWcloud Software Co., Ltd.
>>             >>
>>             >>
>>             >>
>>             >>
>>             >>
>>             __________________________________________________________________________
>>             >> OpenStack Development Mailing List (not for usage
>>             questions)
>>             >> Unsubscribe:
>>             OpenStack-dev-request at lists.openstack.org?subject:unsubscribe
>>             <http://OpenStack-dev-request@lists.openstack.org?subject:unsubscribe>
>>             >>
>>             http://lists.openstack.org/cgi-bin/mailman/listinfo/openstack-dev
>>             >
>>             __________________________________________________________________________
>>             > OpenStack Development Mailing List (not for usage
>>             questions)
>>             > Unsubscribe:
>>             OpenStack-dev-request at lists.openstack.org?subject:unsubscribe
>>             <http://OpenStack-dev-request@lists.openstack.org?subject:unsubscribe>
>>             >
>>             http://lists.openstack.org/cgi-bin/mailman/listinfo/openstack-dev
>>
>>             --
>>             Best wishes!
>>             Zhou Zheng Sheng / 周征晟  Software Developer
>>             Beijing AWcloud Software Co., Ltd.
>>
>>
>>
>>
>>             __________________________________________________________________________
>>             OpenStack Development Mailing List (not for usage questions)
>>             Unsubscribe:
>>             OpenStack-dev-request at lists.openstack.org?subject:unsubscribe
>>             <http://OpenStack-dev-request@lists.openstack.org?subject:unsubscribe>
>>             http://lists.openstack.org/cgi-bin/mailman/listinfo/openstack-dev
>>
>>
>>
>>         __________________________________________________________________________
>>         OpenStack Development Mailing List (not for usage questions)
>>         Unsubscribe:
>>         OpenStack-dev-request at lists.openstack.org?subject:unsubscribe
>>         <http://OpenStack-dev-request@lists.openstack.org?subject:unsubscribe>
>>         http://lists.openstack.org/cgi-bin/mailman/listinfo/openstack-dev
>>
>>
>>     __________________________________________________________________________
>>     OpenStack Development Mailing List (not for usage questions)
>>     Unsubscribe:
>>     OpenStack-dev-request at lists.openstack.org?subject:unsubscribe
>>     <http://OpenStack-dev-request@lists.openstack.org?subject:unsubscribe>
>>     http://lists.openstack.org/cgi-bin/mailman/listinfo/openstack-dev
>>
>> -- 
>>
>> --
>>
>> Andrew Woodward
>>
>> Mirantis
>>
>> Fuel Community Ambassador
>>
>> Ceph Community
>>
>>
>>
>> __________________________________________________________________________
>> OpenStack Development Mailing List (not for usage questions)
>> Unsubscribe: OpenStack-dev-request at lists.openstack.org?subject:unsubscribe
>> http://lists.openstack.org/cgi-bin/mailman/listinfo/openstack-dev
>
> -- 
> Best wishes!
> Zhou Zheng Sheng / 周征晟  Software Developer
> Beijing AWcloud Software Co., Ltd.

-- 
Best wishes!
Zhou Zheng Sheng / 周征晟  Software Developer
Beijing AWcloud Software Co., Ltd.

-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://lists.openstack.org/pipermail/openstack-dev/attachments/20150625/0f1aed5a/attachment.html>


More information about the OpenStack-dev mailing list