Just back from the Juno summit, I attended most of the storage sessions and was extremely shocked how Ceph was avoided by storage vendors. However LVM, the reference storage backend for Cinder was always mentioned. Maybe, is it a sign that Ceph is taking over? Talking about LVM, the last OpenStack survey showed that it was the more used backend.
Six months have passed since Hong Kong and it is always really exciting to see all the folks from the community gathered all-together in a (bit chilly) convention center. As far I saw from the submitted and accepted talks, Ceph continues its road to the top. There is still a huge growing interest about Ceph. On Tuesday May 13th, Josh and I led a (3 hours long) session to discuss the next steps of the integration of Ceph into OpenStack. To be honest, back when we were in Hong Kong, I believe that we were too optimistic about our roadmap. So this time we decided to be a little more realistic and took a more step-by-step approach rather than “let’s add everything we can”. However, this does not mean that the Icehouse cycle was limited in terms of features, not at all! Indeed the Icehouse cycle has seen some tremendous improvements. I don’t know if you remember my last year article right after the Icehouse summit, there was a feature that I wanted so much: RADOS as a backend for Swift. And yes, we made it, so if you want more details you’d better continue the reading :).
Date: 15/05/2014 Video: Slides: Download the slides here.
A simple script to bootstrap a Ceph cluster and start playing with it. The script heavily relies on:
The final machine will contain:
For quite some time, Ceph has an admin API. This article demonstrates and gives some hints to monitor Ceph.
Following up this article.
This playbook was made to automate Ceph servers maintenance.
The typical use case is an hardware change.
By running this playbook you will set the noout flag on your cluster, which means that OSD can’t be marked as out of the CRUSH map, but they will be marked as down.
Thus the OSD will not receive any data.
Basically we tell the cluster to do not move any data since the operation will not last for too long.
A simple in-memory backend that stores objects in RAM is available since version 0.76 and will land with Firefly. Quick look at this new feature.
Managing an OpenStack public cloud can be tough and building it properly is even harder. You can not predict the workload of your platform, customers do what they want (yes they pay for this!). So yes, cloud performance are often unpredictable! Recent studies showed that while running a long-standing benchmark on several cloud platforms, they experienced a performance drop-down of 40% (crazy isn’t it?). However, there are some simple facilities in OpenStack that allow you to have a better control of the resources that you offer to your customers/users. This is what I am going to briefly explore in this article.
It has been two weeks now since we released the Ansible playbook for Ceph. This article is a little update concerning new features and roadmap.
Ceph: cow based deployment with Ansible :-).