Sébastien Han

One major use case for operators is to be able to reserve a certain amount of memory in the hypervisor. This is extremely useful when you have to recover from failures. Imagine that you run all your virtual machines on shared storage (Ceph RBD or Sheepdog or NFS). The major benefit from running your instances on shared storage is that it will ease live-migration and evacuation. However, if a compute node dies you want to make sure that you have enough capacity on the other compute nodes to relaunch your instances. Given that the nova host-evacuate call goes through the scheduler again you should get an even distribution.

But how to make sure that you have enough memory on the other hypervisors? Unfortunately there is no real memory restriction mechanism. In this article I will explain how we can mimic such behavior.

Read On...

This feature has been around for quite a while now, if I remember correctly it was introduced in the Grizzly release. However, I never really got the chance to play around with it. Let’s assume that you use NFS to store Glance images, we know that the default booting mechanism implies to fetch the instance image from Glance to the Nova compute. This is basically streaming the image which involves network throughput and makes the boot process longer. OpenStack Nova can be configured to directly access Glance images from a local filesystem path. This is ideal for our NFS scenario.

Read On...

Libvirt has the ability to configure a watchdog device for QEMU guests. When the guest operating system hangs or crashes the watchdog device is used to automatically trigger some actions. The watchdog support was added in OpenStack Icehouse.

Read On...

Doing backups in the Cloud is not an easy task. In this article I will try to answer some frequently asked questions.

Read On...

Self promotion ahead :) For the next OpenStack summit I have submitted two talks.

Read On...

A while back, I wrote an article about taking consistent snapshots of your virtual machines in your OpenStack environment. However this method was really intrusive since it required to be inside the virtual machine and to manually summon a filesystem freeze. In this article, I will use a different approach to achieve the same goal without the need to be inside the virtual machine. The only requirement is to have a virtual machine running the qemu-guest-agent.

Read On...

Only Magic Card player might recognize that post picture :) (if you’re interested)

I have been waiting for this for quite a while now. Discard, also called trim (with SSD), is a space reclamation mechanism that allows you to reclaim unused blocks on a disk. RBD images are sparse by default, this means that the space they occupy increase the more you write data (opposite of preallocation). So while writing on your filesystem you might end up to the end of your device. On the Ceph side, no one knows what is happening on the filesystem, so we actually end up with fully allocated blocks… In the end the cluster believes that the RBD images are fully allocated. From an operator perspective, having the ability to reclaim back the space unused by your running instances is really handy.

Read On...

What’s happening under the hood while playing with Cinder and Ceph? Answer table :-).

Read On...

For operationnal reasons, you might not want to automatically make your compute node available. With the following flag, during its first bootstrap the compute node will register itself to the service list. However it will be disabled, so virtual machines can not be scheduled on it:

enable_new_services=False

Read On...

Just a reminder for me :D.

Read On...