Hi Marco, While i totally agree with you on the design and resiliance to failures of newly developed apps, I disagree with you on the necessity to have a vm snapshot feature with kvm. and let me explain why.
At the moment, I am running a small acs + kvm + ceph cluster with about a hundred or so vms. Currently, it is a huge problem for me to perform vm snapshots in a reasonably good state, especially if vm has multiple volumes. The problem with ACS implementation (which is not present in openstack for example) is that there is no way I can keep the snapshots on the primary storage (using fast interconnect). The snapshots are copied over a slow link to the nfs secondary storage. It takes ages to perform any operation with the secondary storage due to its silly design (even though I see the reason for its existence in some cases). Now, consider that I have to take a daily snapshot of every volume and some volumes have to snapped on an hourly basis. This totally overloads the network + nfs storage. Imagine that you have to recover a few volumes from snapshots or create new volumes or templates from them while the daily/hourly snapshot cycle is taking place. You will have to wait hours before you get anywhere even on smaller size volumes. However, when I was testing the XenServer setup with the VM snapshot capability, the snapshot creation and roll back was pretty quick. and it worked for all disk volumes in that vm. I don't remember seeing much network traffic or load on the nfs server either. This feature is a must imho regardless of the application design. Besides, as far as I remember, the KVM have that capability, so, why not implement it within ACS, just like it's done for vmware and xenserver? Andrei ----- Original Message ----- > From: "Marc-Aurèle Brothier" <ma...@exoscale.ch> > To: "users" <users@cloudstack.apache.org> > Sent: Monday, 19 December, 2016 08:31:26 > Subject: Re: KVM Live VM Snapshots > Hi Asai, > > In my opinion, doing a VM snapshot is making a step in the wrong direction. > Your applications/system running inside your VMs should be designed to > handle an OS crash. Then a new VM, freshly installed, should be able to get > back into your application setup so that you have again an appropriate > number of healthy nodes. > > Marco > > On Mon, Dec 19, 2016 at 4:34 AM, Asai <a...@globalchangemusic.org> wrote: > >> Greetings, >> >> Is it correct that currently there is no support in Cloudstack for KVM >> live VM snapshots? I see that Volume snapshots are available for running >> VMs, but that makes me wonder what everyone is doing to get a disaster >> recovery backup of a KVM based VM? I did ask this question a few weeks >> back, but only one person responded with one solution, and I am really >> trying to figure out what the best solutions are here. >> >> Has anybody seen this script? https://gist.github.com/ringe/ >> 334ee88ba5451c8f5732 >> >> What is the community's opinion of scripts like this? And also, big >> question, if this script is good, why isn't it integrated into Cloudstack? >> >> Thanks, >> Asai >>
