[Date Prev][Date Next]   [Thread Prev][Thread Next]   [Thread Index] [Date Index] [Author Index]

Re: Persistent disks in cartridges


Depends on what you mean by "crashed". But we'll back this up a little: in Online we use a raid array comprised of a combination of ephemeral and EBS drives. Should a node have issues we have the potential to lose ephemeral disks but we still have the EBS volume. In our case we can reattach the EBS volumes to a new instance, rebuild the array and move on. You could do this exclusively with EBS volumes if you wanted as well.

With respect to clustered file systems I'm talking specifically about things like glusterfs, ceph, pnfs (or nfs for that mater though not clustered), etc. There's a number of expectations that OpenShift makes about the underlying storage that may or may not work with clustered file systems, and there may be performance concerns in using them right now.

I wouldn't call EBS a clustered file system (it's not even a file system properly, it's a block device), and things like SANs and iSCSI, etc should be fine to use with OpenShift.

- John

On 11/07/2013 09:27 PM, Arunabha Ghosh wrote:
Hi John,
             Thanks for the clarification. I was specifically interested
in the node crashes. In OpenShift Origin, what would happen if a Node
crashed ? I'm guessing that all on disk data for the app would be lost ?
You mentioned that Online can recover from crashed nodes. Can you
elaborate on how that's done ? Also you mentioned that you don't
recommend running OpenShift with a clustered file system. Did you mean
things like Amazon EBS ?


On Thu, Nov 7, 2013 at 4:52 PM, John Hawley <jhawley redhat com
<mailto:jhawley redhat com>> wrote:

    On 11/07/2013 12:40 AM, Arunabha Ghosh wrote:

               Is there any way to provide persistent storage for
        cartridges ?
        Does the disk storage for a cartridge disappear when the cartridge's
        gear is moved to another node ? What about Node crashes ?

    Storage on OpenShift is already persistent, and if a gear gets moved
    to a new node, it's files have to get moved with it.  Please don't
    confuse persistent with things like clustered or redundant storage
    though, those are completely different things.

    In Online we have multiple levels of redundancy on the storage, and
    should a node crash there is the ability to move the storage from a
    crashed node to a new one.  If you are running OpenShift on your
    own, it all depends on how you setup your storage and what you have
    available that will define what kind of redundancy and portability
    you have available.  Right now I can't recommend running OpenShift
    with a clustered file system for the gear storage, even though it
    may be possible.

    - John 'Warthog9' Hawley

users mailing list
users lists openshift redhat com

[Date Prev][Date Next]   [Thread Prev][Thread Next]   [Thread Index] [Date Index] [Author Index]