[Openstack-operators] Distributed Filesystem

Razique Mahroua razique.mahroua at gmail.com
Wed Apr 24 16:21:15 UTC 2013


I feel you Jacob, 
Loring I had the exact same issue ! Using both Argonaut and Bobtail, on high I/O load the mount crashed the server - well the server wasn't crashing, the mount went crazy, and impossible to unmount the disk, kill the process, so I always ended up rebooting the nodes. What is interesting though is that the reason why it is not still considered as production-ready is because of the way metadata is currently implemented, rather than the code itself....


Razique Mahroua - Nuage & Co
razique.mahroua at gmail.com
Tel : +33 9 72 37 94 15



Le 24 avr. 2013 à 17:36, Lorin Hochstein <lorin at nimbisservices.com> a écrit :

> Razique:
> 
> Out of curiosity, what kinds of problems did you see with CephFS? I've heard it's not ready for production yet, but I haven't heard anybody talk about specific experiences with it.
> 
> Lorin
> 
> 
> On Sat, Apr 20, 2013 at 8:14 AM, Razique Mahroua <razique.mahroua at gmail.com> wrote:
> Hi Paras, 
> that's the kind of setup I've always seen myself. After unsuccessful tests with CephFS, I'll move to the following strategy:
> - GlusterFS as a shared storage for the instances (check the official doc, we wrote about its deployment for OpenStack)
> - Ceph cluster wit the direct RBD gateway from nova to RADOS
> - Ceph cluster as well the imaging service (Glance)
> 
> Some others use MooseFS as well the the stared storage (we wrote a deployment guide as well)
> Best regards,
> Razique
> 
> 
> Razique Mahroua - Nuage & Co
> razique.mahroua at gmail.com
> Tel : +33 9 72 37 94 15
> 
> <NUAGECO-LOGO-Fblan_petit.jpg>
> 
> Le 19 avr. 2013 à 17:05, Paras pradhan <pradhanparas at gmail.com> a écrit :
> 
>> Well I am not sure if we would like to do it since it is marked as deprecated. So this is what I am thinking. For shared storage, I will be using glusterfs and use cinder just for extra block disk on the instances. This what the Openstack operators doing typically ? 
>> 
>> Thanks
>> Paras.
>> 
>> 
>> On Fri, Apr 19, 2013 at 12:10 AM, Razique Mahroua <razique.mahroua at gmail.com> wrote:
>> More infos here:
>> http://osdir.com/ml/openstack-cloud-computing/2012-08/msg00293.html
>> 
>> But I'm not sure about the last updates - you can still use it at the moment 
>> Razique
>> 
>> Razique Mahroua - Nuage & Co
>> razique.mahroua at gmail.com
>> Tel : +33 9 72 37 94 15
>> 
>> <NUAGECO-LOGO-Fblan_petit.jpg>
>> 
>> Le 18 avr. 2013 à 17:13, Paras pradhan <pradhanparas at gmail.com> a écrit :
>> 
>>> Regarding block migration, this is what confuses me. This is from the Openstack operations manual
>>> 
>>> --
>>> Theoretically live migration can be done with non-shared storage, using
>>> a feature known as KVM live block migration. However, this is a littleknown feature in OpenStack, with limited testing when compared to live migration, and is slated for deprecation in KVM upstream.
>>> --
>>> 
>>> Paras.
>>> 
>>> 
>>> On Thu, Apr 18, 2013 at 3:00 AM, Razique Mahroua <razique.mahroua at gmail.com> wrote:
>>> Sure :)
>>> Great feedbacks around. Many technologies do pretty much everything on the paper - but I guess in the end it's more about if the tech. does the job and if it does it well.
>>> For such critical implementation, reliable solution is a must-have - ie that have proven through years they can be used and are stable enough for us to enjoy our week-ends :)
>>> 
>>> Razique
>>> 
>>> Le 18 avr. 2013 à 00:14, Paras pradhan <pradhanparas at gmail.com> a écrit :
>>> 
>>>> Thanks for the replies Razique. We are doing a test installation and looking for options for live migration. Looks like both cinder and shared file stirage are options. Among these two which one do you guys recommended considering the Cinder block will be typical lvm based commodity hardware.
>>>> 
>>>> Thanks
>>>> Paras.
>>>> 
>>>> 
>>>> On Wed, Apr 17, 2013 at 5:03 PM, Razique Mahroua <razique.mahroua at gmail.com> wrote:
>>>> Definitely, use the "--block_migrate" flag along the nova migrate command so you don't need a shared storage.
>>>> You can boot from Cinder, depending on which version of OPS you run 
>>>> 
>>>> Razique Mahroua - Nuage & Co
>>>> razique.mahroua at gmail.com
>>>> Tel : +33 9 72 37 94 15
>>>> 
>>>> <NUAGECO-LOGO-Fblan_petit.jpg>
>>>> 
>>>> Le 17 avr. 2013 à 23:55, Paras pradhan <pradhanparas at gmail.com> a écrit :
>>>> 
>>>>> Can we do live migration without using shared storage like glusterfs and using cinder to boot the volume from? 
>>>>> 
>>>>> Sorry little off topic 
>>>>> 
>>>>> Thanks
>>>>> Paras.
>>>>> 
>>>>> 
>>>>> On Wed, Apr 17, 2013 at 4:53 PM, Razique Mahroua <razique.mahroua at gmail.com> wrote:
>>>>> Many use either a proprietary backend or the good old LVM
>>>>> I'll go with Ceph for it since there is a native integration between cinder/ nova-volume and Ceph
>>>>> 
>>>>> Razique Mahroua - Nuage & Co
>>>>> razique.mahroua at gmail.com
>>>>> Tel : +33 9 72 37 94 15
>>>>> 
>>>>> <NUAGECO-LOGO-Fblan_petit.jpg>
>>>>> 
>>>>> Le 17 avr. 2013 à 23:49, Paras pradhan <pradhanparas at gmail.com> a écrit :
>>>>> 
>>>>>> What do people use for cinder?
>>>>>> 
>>>>>> Thanks
>>>>>> Paras.
>>>>>> 
>>>>>> 
>>>>>> On Wed, Apr 17, 2013 at 4:41 PM, Razique Mahroua <razique.mahroua at gmail.com> wrote:
>>>>>> I was about to use CephFS (Bobtail) but the I can't resize the instances without having CephFS crashing.
>>>>>> I'm currently considering GlusterFS which not only provides great performance, it's also pretty much easy to administer  :)
>>>>>> 
>>>>>> Le 17 avr. 2013 à 22:07, JuanFra Rodriguez Cardoso <juanfra.rodriguez.cardoso at gmail.com> a écrit :
>>>>>> 
>>>>>>> Glance and Nova with MooseFS.
>>>>>>> Reliable, good performance and easy configuration.
>>>>>>> 
>>>>>>> ---
>>>>>>> JuanFra
>>>>>>> 
>>>>>>> 
>>>>>>> 2013/4/17 Jacob Godin <jacobgodin at gmail.com>
>>>>>>> Hi all,
>>>>>>> 
>>>>>>> Just a quick survey for all of you running distributed file systems for nova-compute instance storage. What are you running? Why are you using that particular file system?
>>>>>>> 
>>>>>>> We are currently running CephFS and chose it because we are already using Ceph for volume and image storage. It works great, except for snapshotting, where we see slow performance and high CPU load.
>>>>>>> 
>>>>>>> 
>>>>>>> _______________________________________________
>>>>>>> OpenStack-operators mailing list
>>>>>>> OpenStack-operators at lists.openstack.org
>>>>>>> http://lists.openstack.org/cgi-bin/mailman/listinfo/openstack-operators
>>>>>>> 
>>>>>>> 
>>>>>>> _______________________________________________
>>>>>>> OpenStack-operators mailing list
>>>>>>> OpenStack-operators at lists.openstack.org
>>>>>>> http://lists.openstack.org/cgi-bin/mailman/listinfo/openstack-operators
>>>>>> 
>>>>>> 
>>>>>> _______________________________________________
>>>>>> OpenStack-operators mailing list
>>>>>> OpenStack-operators at lists.openstack.org
>>>>>> http://lists.openstack.org/cgi-bin/mailman/listinfo/openstack-operators
>>>>>> 
>>>>>> 
>>>>> 
>>>>> 
>>>> 
>>>> 
>>> 
>>> 
>> 
>> 
> 
> 
> _______________________________________________
> OpenStack-operators mailing list
> OpenStack-operators at lists.openstack.org
> http://lists.openstack.org/cgi-bin/mailman/listinfo/openstack-operators
> 
> 
> 
> 
> -- 
> Lorin Hochstein
> Lead Architect - Cloud Services
> Nimbis Services, Inc.
> www.nimbisservices.com

-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://lists.openstack.org/pipermail/openstack-operators/attachments/20130424/2e70c0e8/attachment-0001.html>
-------------- next part --------------
A non-text attachment was scrubbed...
Name: NUAGECO-LOGO-Fblan_petit.jpg
Type: image/jpeg
Size: 10122 bytes
Desc: not available
URL: <http://lists.openstack.org/pipermail/openstack-operators/attachments/20130424/2e70c0e8/attachment-0001.jpg>


More information about the OpenStack-operators mailing list