Red Hat OpenShift Spain 2022 OpenShift Commons Gathering at KubeconEU, 17 May 2022

Previous Meeting Next Meeting

⏯

youtube image

►

From YouTube: Moving Everything to Container Storage Interface without Anyone Noticing -Jan Šafránek Red Hat OSCG

Description

Moving Everything to Container Storage Interface without Anyone Noticing
Speaker: Jan Šafránek Red Hat
Storage Update

OpenShift Commons Gathering Kubecon EU
May 17, 2022 Live from Kubecon EU in Valencia, Spain

Full Agenda here: https://commons.openshift.org/gatherings/OpenShift_Commons_Gathering_at_Kubecon_Europe_2022.html

Learn more at: https://commons.openshift.org

A

My name is janja franek. I work at hardhat and I am sick storage tech lead upstream and also I am a team lead of a small team in storage in red hat that manages openshift storage. So I wear many hats, partly upstream party downstream, and I will talk here about a recent feature that we have been working on in upstream, where we are removing code from kubernetes and moving in somewhere else and hopefully without nobody noticing. If we did our work right here down, you can find a link to the slides.

A

If you, if you find it useful, there are a couple of commands that you may need.

A

So before I start, let me show you brief intro into kubernetes history. It's long in july, 2015 we released kubernetes 1.0. I was there. I have my code there and our storage story. There was called volume plugins we had 10 of them. The name would suggest that there are plugins, you can plug them in. You can plug them out in runtime, but that's false. They are hard coded in kubernetes github repository.

A

You can't link them dynamically, so anybody wants to guess how many volume plugins we had uh two years later in september, 2015 17 any number.

A

Nobody come on uh how many volume plugins we had in 2017. We had 10 in 2015, so how many we had in 2017 18 close 20 25. We got 27.

A

So in two years we got 17 new volume, plugins for various storage, backends clouds bare metal software, defined storage, anything, and we knew that we can't go further.

A

We want something that is pluggable in runtime and we don't want to maintain in the kubernetes. So first attempt was were flex volumes. They were kind of clunky, both on the kubernetes side and on the storage provider side and in late 2017. We introduced container storage interface as alpha, and that was our future for storage in kubernetes.

A

As a carrot on the stick, so we lure everybody to use csi we implement it. We started implementing the new features like snapshots like cloning, whatever only as in csi. So if you want snapshots, you must use csi entry volume plugins, you will never get snapshot. Support csi got extremely extremely popular.

A

Yesterday I did the numbers I found 100 140 csi drivers that I know about. I'm pretty sure there are many other drivers that are proprietary and I don't know about the 140 drivers and, as we developed the csi drivers for the clouds and storage backends that we knew uh we ended up with the entry volume plugin that has some features and the csv driver. It has the same features plus the new features, and we ended up with two codes that we needed to maintain and uh early in 2018.

A

We decided we want to get off the entry code. The same applies to entry cloud providers. uh Again we don't want to maintain integral providers and external cloud providers, so entry must go away. That was 2018.

A

In november 2018 we had alpha of csm migration of the first volume plugins, and here we are three and a half years later. The csi migration is ga for azure disk and openstack cinder in kubernetes 124.

A

That means there is no entry cost involved in these two volume plugins, if you have kubernetes 124., so it took last, took us a free and for three and a half years to get there.

A

So what actually is csi migration? I already said we are moving code. We are not moving your data, the state data stays where it is. We are not changing the api. The api object stays exactly the same. So if you have a kubernetes or openshift cluster, we have entry volume plugins in three volume, pv storage classes, stickers, demand, sets everything and you upgrade to a version where csi migration is ga nothing changes.

A

You still have the same objects, the same pvs same storage classes, just under the hood. We are directing all the storage calls to the css drivers and they talk to the csi how to use they talk to the storage backend.

A

So in theory, nobody should ever notice that there is any csi migration happening right. We did our homework upstream. We did the test ci everything the same with openshift test, ci everything, and we will support it, so we got your covered, but on the other hand, I am a software engineer, and it makes me kind of uncomfortable to throw away.

A

Here is an example of amazon cloud provider and the volume plugin. We are throwing away 17 000 lines of code, including commands, including unit tests that is tested to death. It runs in production for many years. Everybody knows how it behaves. Everybody knows its corner cases and we are replacing it with some other code in the csi drivers. The css drivers they're running in production today again pretty well tested. We know how it behaves uh just that translation between the inter volume plugin and the css driver.

A

It's not really well tested like we did test it, definitely in our environment, but we know that many customers are crazy or maybe inventive they use kubernetes and openshift in a way that nobody would have ever imagined.

A

So if you are one of these crazy users, please test csi migration, and second thing is that csi migration is huge feature so in open shift. It is disabled by default when it's beta in kubernetes and it will get enabled by default only when it gets ga upstream.

A

That means, if you know kubernetes upstream, when a feature gets ga there, it is enabled- and you can't disable it. So if it, if the csi migration is broken, and you upgrade to a version where it is ga, you can't turn back, you can't turn it off. It's too late. You have a broken cluster.

A

So that's why I am here actually, with this lighting talks, I want to encourage everybody to test it. While it's tech preview in openshift, you can enable tech preview features very easily. You edit one cr, that we call feature gate.

A

You add the last line feature set tech preview, new upgrade. It will enable all tech preview features in your cluster, including synthetic migration, including all the other tech preview features, but there is a catch. The no upgrade suffix means that you can't upgrade upgraded that cluster. So please don't do that in production.

A

uh This is for ci for testing for your lab environments. So you can try new features in openshift before they get ga before you upgrade your clusters.

A

When you enable tech preview, no upgrade features, uh it will enable, like the feature gate in kubernetes, and when it's doing this it will drain your cube. It will drain your nose and restart the cubelets, so it will take some time if you have big cluster. This will take some time in small clusters. It takes. I don't know, 5-10 minutes.

A

Here is the timeline: uh what we do in both in upstream and in openshift and in the middle. There is a column uh when you can start testing csi migration in openshift.

A

So as I've said, azure disk and cinder are ga in kubernetes 124 in openshift 411. It will be ga very very soon. So please hurry with your tests. If you are using one of those.

A

Aws amazon cloud, the gc, google cloud and azure file will follow next in the next release: kubernetes 125 openshift for 12 and the last will be the sphere in the next release and the following release. So the timeline is pretty tight.

A

It's really the best time to start this thing before it's too late uh and let us know how it works.

A

This uh schedule is kind of optimistic. This is just the current upstream plan for the future releases. We have shifted the ga uh graduation a couple of times already, but I think right now it is the right. It looks like it's going ready to happen and, of course, uh please read: release notes, uh for example, for vsphere. We are deprecating support for old vsphere versions and kubernetes one sorry, openshift 413 will most probably require version 7.0.2, which is the current vsphere version. So if you have anything on 6.7 or 7.0.0 uh start thinking about upgrades.

A

uh Here is here are a couple of recommendations that I with my upstream head. I would recommend everybody if you, if you are starting new clusters uh and you have a csi driver available there, which is ga, then please forget about entry. Plugins use csi exclusively.

A

You will not exercise this csi translation, csi migration, layer and our lives and your lives could be simpler if you have existing cluster with existing volumes in three volumes and you have the possibility to move to csi, please move to csi. I know it's often well. In most cases, it's not possible because you need to back up all the data and restore them somewhere else.

A

So, but if you have the chance, please you start using csi and in both cases forget that entry ever existed. If you have a existing cluster with entry volumes and you can't move them away, we are here for you. uh We have this csm aggression exactly for you, so your cluster will work should work after upgrades and we will support uh the this entry pvs entry storage classes in foreseeable future, which is very short this time. But we are not planning to removing the support of these in three pvs into storage classes.

A

You can use them just you will exercise one additional layer that could bring complications and it life is better without that layer and again I encourage everybody. Please please, test csi migration before you upgrade to a version manage ga.

A

So, as a summary, I already told you we are not changing the data. We are not moving the data. We are not changing the api. You can use the api as it is. You can use the existing pvs existing storage classes, but please help us with the testing and for a petition.

A

Here is the schedule and I will be around. If you have any questions about storage or csi, I can answer anything okay. Thank you very much.

A