cloudstack-users mailing list archives

Site index · List index
Message view « Date » · « Thread »
Top « Date » · « Thread »
From Simon Weller <swel...@ena.com.INVALID>
Subject Re: Problems with KVM HA & STONITH
Date Thu, 02 Nov 2017 13:57:22 GMT
James,


Ceph is a great solution and we run all of our ACS storage on Ceph. Note that it adds another
layer of complexity to your installation, so you're going need to develop some expertise with
that platform to get comfortable with how it works. Typically you don't want to mix Ceph with
your ACS hosts. We in fact deploy 3 separate Ceph Monitors, and then scale OSDs as required
on a per cluster basis in order to add additional resiliency (So every KVM ACS cluster has
it's own Ceph "POD").  We also use Ceph for S3 storage (on completely separate Ceph clusters)
for some other services.


NFS is much simpler to maintain for smaller installations in my opinion. If the IO load you're
looking at isn't going to be insanely high, you could look at building a 2 node NFS cluster
using pacemaker and DRDB for data replication between nodes. That would reduce your storage
requirement to 2 fairly low power servers (NFS is not very cpu intensive). Currently on a
host failure when using a storage other than NFS on KVM, you will not see HA occur until you
take the failed host out of the ACS cluster. This is a historical limitation because ACS could
not confirm the host had been fenced correctly, so to avoid potential data corruption (due
to 2 hosts mounting the same storage), it doesn't do anything until the operator intervenes.
As of ACS 4.10, IPMI based fencing is now supported on NFS and we're planning on developing
similar support for Ceph.


Since you're an school district, I'm more than happy to jump on the phone with you to talk
you through these options if you'd like.


- Si


________________________________
From: McClune, James <mcclunej@norwalktruckers.net>
Sent: Thursday, November 2, 2017 8:28 AM
To: users@cloudstack.apache.org
Subject: Re: Problems with KVM HA & STONITH

Hi Simon,

Thanks for getting back to me. I created one single NFS share and added it
as primary storage. I think I better understand how the storage works, with
ACS.

I was able to get HA working with one NFS storage, which is good. However,
is there a way to incorporate multiple NFS storage pools and still have the
HA functionality? I think something like GlusterFS or Ceph (like Ivan and
Dag described) will work better.

Thank you Simon, Ivan, and Dag for your assistance!
James

On Wed, Nov 1, 2017 at 10:10 AM, Simon Weller <sweller@ena.com.invalid>
wrote:

> James,
>
>
> Try just configuring a single NFS server and see if your setup works. If
> you have 3 NFS shares, across all 3 hosts, i'm wondering whether ACS is
> picking the one you rebooted as the storage for your VMs and when that
> storage goes away (when you bounce the host), all storage for your VMs
> vanishes and ACS tries to reboot your other hosts.
>
>
> Normally in a simple ACS setup, you would have a separate storage server
> that can serve up NFS to all hosts. If a host dies, then a VM would be
> brought up on a spare hosts since all hosts have access to the same storage.
>
> Your other option is to use local storage, but that won't provide HA.
>
>
> - Si
>
>
> ________________________________
> From: McClune, James <mcclunej@norwalktruckers.net>
> Sent: Monday, October 30, 2017 2:26 PM
> To: users@cloudstack.apache.org
> Subject: Re: Problems with KVM HA & STONITH
>
> Hi Dag,
>
> Thank you for responding back. I am currently running ACS 4.9 on an Ubuntu
> 14.04 VM. I have the three nodes, each having about 1TB of primary storage
> (NFS) and 1TB of secondary storage (NFS). I added each NFS share into ACS.
> All nodes are in a cluster.
>
> Maybe I'm not understanding the setup or misconfigured something. I'm
> trying to setup an HA environment where if one node goes down, running an
> HA marked VM, the VM will start on another host. When I simulate a network
> disconnect or reboot of a host, all of the nodes go down (STONITH?).
>
> I am unsure on how to setup an HA environment, if all the nodes in the
> cluster go down. Any help is much appreciated!
>
> Thanks,
> James
>
> On Mon, Oct 30, 2017 at 3:49 AM, Dag Sonstebo <Dag.Sonstebo@shapeblue.com>
> wrote:
>
> > Hi James,
> >
> > I think  you possibly have over-configured your KVM hosts. If you use NFS
> > (and no clustered file system like CLVM) then there should be no need to
> > configure STONITH. CloudStack takes care of your HA, so this is not
> > something you offload to the KVM host.
> >
> > (As mentioned the only time I have played with STONITH and CloudStack was
> > for CLVM – and I eventually found it not fit for purpose, too unstable
> and
> > causing too many issues like you describe. Note this was for block
> storage
> > though – not NFS).
> >
> > Regards,
> > Dag Sonstebo
> > Cloud Architect
> > ShapeBlue
> >
> > On 28/10/2017, 03:40, "Ivan Kudryavtsev" <kudryavtsev_ia@bw-sw.com>
> wrote:
> >
> >     Hi. If the node losts nfs host it reboots (acs agent behaviour). If
> you
> >     really have 3 storages, you'll go clusterwide reboot everytime your
> > host is
> >     down.
> >
> >     28 окт. 2017 г. 3:02 пользователь "Simon Weller"
> > <sweller@ena.com.invalid>
> >     написал:
> >
> >     > Hi James,
> >     >
> >     >
> >     > Can you elaborate a bit further on the storage? You say you're
> > running NFS
> >     > on all 3 nodes, can you explain how it is setup?
> >     >
> >     > Also, what version of ACS are you running?
> >     >
> >     >
> >     > - Si
> >     >
> >     >
> >     >
> >     >
> >     > ________________________________
> >     > From: McClune, James <mcclunej@norwalktruckers.net>
> >     > Sent: Friday, October 27, 2017 2:21 PM
> >     > To: users@cloudstack.apache.org
> >     > Subject: Problems with KVM HA & STONITH
> >     >
> >     > Hello Apache CloudStack Community,
> >     >
> >     > My setup consists of the following:
> >     >
> >     > - Three nodes (NODE1, NODE2, and NODE3)
> >     > NODE1 is running Ubuntu 16.04.3, NODE2 is running Ubuntu 16.04.3,
> > and NODE3
> >     > is running Ubuntu 14.04.5.
> >     > - Management Server (running on separate VM, not in cluster)
> >     >
> >     > The three nodes use KVM as the hypervisor. I also configured
> primary
> > and
> >     > secondary storage on all three of the nodes. I'm using NFS for the
> > primary
> >     > & secondary storage. VM operations work great. Live migration works
> > great.
> >     >
> >     > However, when a host goes down, the HA functionality does not work
> > at all.
> >     > Instead of spinning up the VM on another available host, the down
> > host
> >     > seems to trigger STONITH. When STONITH happens, all hosts in the
> > cluster go
> >     > down. This not only causes no HA, but also downs perfectly good
> > VM's. I
> >     > have read countless articles and documentation related to this
> > issue. I
> >     > still cannot find a viable solution for this issue. I really want
> to
> > use
> >     > Apache CloudStack, but cannot implement this in production when
> > STONITH
> >     > happens.
> >     >
> >     > I think I have something misconfigured. I thought I would reach out
> > to the
> >     > CloudStack community and ask for some friendly assistance.
> >     >
> >     > If there is anything (system-wise) you request in order to further
> >     > troubleshoot this issue, please let me know and I'll send. I
> > appreciate any
> >     > help in this issue!
> >     >
> >     > --
> >     >
> >     > Thanks,
> >     >
> >     > James
> >     >
> >
> >
> >
> > Dag.Sonstebo@shapeblue.com
> > www.shapeblue.com<http://www.shapeblue.com>
[http://www.shapeblue.com/wp-content/uploads/2017/06/logo.png]<http://www.shapeblue.com/>

Shapeblue - The CloudStack Company<http://www.shapeblue.com/>
www.shapeblue.com
Rapid deployment framework for Apache CloudStack IaaS Clouds. CSForge is a framework developed
by ShapeBlue to deliver the rapid deployment of a standardised ...



> [http://www.shapeblue.com/wp-content/uploads/2017/06/logo.png

[http://www.shapeblue.com/wp-content/uploads/2017/06/logo.png]

]<
> http://www.shapeblue.com/>
[http://www.shapeblue.com/wp-content/uploads/2017/06/logo.png]<http://www.shapeblue.com/>

Shapeblue - The CloudStack Company<http://www.shapeblue.com/>
www.shapeblue.com
Rapid deployment framework for Apache CloudStack IaaS Clouds. CSForge is a framework developed
by ShapeBlue to deliver the rapid deployment of a standardised ...



>
> Shapeblue - The CloudStack Company<http://www.shapeblue.com/>
[http://www.shapeblue.com/wp-content/uploads/2017/06/logo.png]<http://www.shapeblue.com/>

Shapeblue - The CloudStack Company<http://www.shapeblue.com/>
www.shapeblue.com
Rapid deployment framework for Apache CloudStack IaaS Clouds. CSForge is a framework developed
by ShapeBlue to deliver the rapid deployment of a standardised ...



> www.shapeblue.com<http://www.shapeblue.com>
[http://www.shapeblue.com/wp-content/uploads/2017/06/logo.png]<http://www.shapeblue.com/>

Shapeblue - The CloudStack Company<http://www.shapeblue.com/>
www.shapeblue.com
Rapid deployment framework for Apache CloudStack IaaS Clouds. CSForge is a framework developed
by ShapeBlue to deliver the rapid deployment of a standardised ...



> Rapid deployment framework for Apache CloudStack IaaS Clouds. CSForge is a
> framework developed by ShapeBlue to deliver the rapid deployment of a
> standardised ...
>
>
>
> > 53 Chandos Place, Covent Garden, London  WC2N 4HSUK
> > @shapeblue
> >
> >
> >
> >
>
>
> --
>
>
>
> James McClune
>
> Technical Support Specialist
>
> Norwalk City Schools
>
> Phone: 419-660-6590
>
> mcclunej@norwalktruckers.net
>



--



James McClune

Technical Support Specialist

Norwalk City Schools

Phone: 419-660-6590

mcclunej@norwalktruckers.net

Mime
  • Unnamed multipart/alternative (inline, None, 0 bytes)
View raw message