Cloud Native Computing Foundation KubeCon + CloudNativeCon North America 2017 (Austin), 15 Dec 2017

Previous Meeting Next Meeting

⏯

youtube image

►

From YouTube: Block Volumes Support in Kubernetes [I] - Mitsuhiro Tanino, Hitachi Data Systems

Description

Block Volumes Support in Kubernetes [I] - Mitsuhiro Tanino, Hitachi Data Systems

Storage is an essential part of any computing systems. In current Kubernetes, user can utilize storage volume with filesystem in a container but can't be utilized volume without filesystem called raw block volume.

By adding a feature to enable raw block storage directly, for example, user can use the raw block volume for database applications such as MariaDB and this improves I/O performance.

In this session, I will explain current activity and feature plan of Block Volumes Support in Kubernetes.

About Mitsuhiro Tanino
Mitsuhiro Tanino is a software engineer who has been working for Hitachi since 2004 and a principal software engineer Hitachi Data systems since 2014. He has experience about development of virtual machine manager for heterogeneous cloud systems and RAS features for KVM virtual environments. Also he contributed OpenStack Cinder project for three years. His current working area is Kubernetes sig-storage, he is enhancing reliability and stability for iSCSI and FC drivers and also contributing Block Volumes Support feature developement.
Join us for KubeCon + CloudNativeCon in Barcelona May 20 - 23, Shanghai June 24 - 26, and San Diego November 18 - 21! Learn more at https://kubecon.io. The conference features presentations from developers and end users of Kubernetes, Prometheus, Envoy and all of the other CNCF-hosted projects.

A

Let's start my show, so my name is mr. Hirata Nino I'm software engineer at the Hitachi Pantera. So in this session, I had to introduce the new search feature role, block block bottom support. This provides a feature to use low block volume in inside of container instead of the hydro system volume.

A

So this is actually the agenda. So let me expand our three me: introduce myself quickly and also I've explained about above-ground. Why we contributing to that is future and also I, explained summary of the low block you Boreum support. What is about Roberto in support. There are shows detail usage of roblox volumes. They also I explained implementation on deep type about the overall volume. Their arrests are experienced about future work for the low block volume so that this rights is available for click on schedule size, so please feel free, download it and use it.

A

So rat makes introduce myself quickly so I mr. Canino doctor engineer to achieve antara previously studied at assistant, so I was in inaccessible service team for almost 80 years, and I was provided the accessible service to enterprise customers, especially related to the GPU.

A

Over time, our memory region and after that I started to contribute to the construction, the project almost three years, many bug fixes and added some future into the cinder and there, from this year, I started to contribute to the Cuban 86 trays and I'm now trying to enhance the durability and stability for I scary and, if she'd, drivers and also I'm contributing the block volumes about feature so I'm, usually in the kubernetes. Shing storage love child.

A

So if you have an equation related to the IFC I scuzzy or broke volumes, please pick me be as a slap China.

A

So next, let me explain was a background of this feature. So, as you do, you may know so, by using the kubernetes rich future user can create a eat. Volume with the airport frame pot with volume is created, switch traveling, creates volume and creates a file system on top of it. So therefore, are currently the this volume must be consumed by fire system inside the pot, so this figure shows the application pods. So, as you can see, the disability education pod can access mount class, which is a parameter as a persistent volume.

A

So this means user account. Consumer low blood volume such as the Philadelphia CEOs function, is that directory from the applications without filesystem it's a current status, so we think the benefit to suppose that all block volumes is enable the users to choose appropriate type of volume for their applications, such as Maria DB or higher DB. This is a she touches. The DBMS software, et cetera. Also, this feature enables provides consistent, I/o performance and relationship compared to for system volume, especially using the enterprise fibre channel storage.

A

Over I scuzzy storage, locally SSD disk, suitable low block youth will work use case also, we think at a production system. This low block feature is essentially a sensor functionality and also efficient at buoy 1.9 of the kubernetes container storage interface called she I speak. They find to support both for higher volume and applicable capabilities, and she is like dragons a lot of us that we were born. Nine.

A

So let me explain mod detail with the current program, so this figure shows kind workflow of to create a pot with five system, persistent Emporium. So, as you can see as a step, Wow use the request pot with persistent volume cream there out of step, two people buying their binds requested, passes in the volume cream and put a provision on the poshest in the podium dr.

A

step. Three switch problem attached its PB to create accurate note, then that PE is recognized as a solid waste da on Hubert. Novel in this case also probably creates first system on top of the syllabus the year. If fire system not exist, then the volume of thread of a CDA is mounted to the root, a wave on hewitt node.

A

After that, as a set for she read, start application port, then you are in the pod creation. The amount house mount point service. Mount point is passed into containers amount of us in this case the amount of assisted able as observer, live data. So in from inside report so application. The application pod noxious amount of us to live data choose to a DB application data.

A

So the Julia noticed so foreign components such as plastics in the boreal, persistent, the podium cream and the queue API server cubic controller manager, curate, also storage, dragging count 100. We look for the rock type of volume currently seeing Orion support file system type, otaku, bunnies version, 1.8.

A

Also, there is a workaround today so using the privileged option, so when user creates a pot with privileges, ocean or all block devices on cubed nodes exposed to a container, so by using this option, use a user application can use any block devices. On the curate note, however, you know that will use this reputation of shown other way you have to configure, as are all privileged record.

A

As when a lumen have to configure as an atom are over visit previous to I go through option for kubernetes cluster, so we think this is a decent recommended way since previously container may cause. She has security programs so from the container user. Application can anything to do the actuate nodes, so maybe the creator node may be destroyed, so it's a very dangerous option, so it's not recommended to use. So since there is a program, so we in order to throw this program, we contribute to add the new future low block poem.

A

So let me explain what low volume of support at version 1.9. So we introduce new feature row block podium, support to be 1.9 as alpha teacher, so the future has new it to new API supports. First, one is the volume mode body mode API is supported at assistant volume and also persistent the volume cream object, and second one is the volume divisive API. This API is for the proto definition, so I've explained about more detail with these parameters so parameter later.

A

Also, as was a problem part, we introduced fiber channel plug-in supports for low volume as a reference to driver. So please note internal fibre channel Pellegrini. Only both prakruti speak for the prohibition. The ashes in the podium already also, as was a dynamic provisioning support case, speak these three under discussions. So this is not concluded as a version 1.9 so bust. We confirmed our XG. I will challenge external provision era. This is not oasis, but this works together with updated internal airship problem. So maybe we can close. Maybe we can fix regards speak towards version.

A

One point: 10 I was the dynamic protein-protein support.

A

So this figure shows a new hub through a new role using the block parties in the podium. So there are four stay. Please, there are four steps at first step is a request pot with volume devices parameter and also PVC with poor. You, Moto e call Barak and after 53 people buying their binds the requested PVC and provisioned persistent volume, which has a body model called Brook, so the important sinks here is both species were in mode under P, with volume mode should have the same value.

A

To bind to bind not to bound to be bound understand. The 6-3 switch. Probably attached is persistent for you to create node and then persistent Boreum is recognized as radio is the a or something like that on the create note, since the volume mode eco broke in this case, plugin does not create 5 system on top of the database, the 8th so after that container a consumer device, other autograft device, so a two-step flow, creates that start application port.

A

Then, during pod creation block people is passed to the port as lateral CA or any user-defined device path. The device pass option can be used can be defined by the Posse persist, photo definition I explained after later, so as you can see in that figure, so the V application pot can see device path, so they be affiliates. So the DB application can issue IO to study with the a directory. In this case.

A

So let me explain about usage of the roadblock volumous feature, so, as I explained, this future is out of a future at Bowie 1.9. So, in order to use this feature, other mean is to configure future gates equal clock volume called for services such as Cuba, 8, chui, API server should be a controller manager, etc. To enable work, volume feature and the result using a future gate, configuration body, mode, parameter and volume devices parameter APF permit are not accepted. If lumens position, the Boreum claim were purchased in the volume definition.

A

So richmond shows more details that put persistent the volume definition, so this figure shows the persistent volume definition is volume model Colebrook. So, as you can see, a Domini can define in volume mode in the persistent volume definition, so the volume mode API field supports two volume modes. First, one is the fire system and the other one is broke so other we can define the expected usage of the podium through this podium mode parameter.

A

If body mode is not specified, the people is treated as a file system volume. This is the same same behavior currently existing.

A

So next one is the persistent volume cream definition. As you see, the volume cream definition is defined by user. So in this case also persistent volume cream supports persistent the body mode body mode parameter in this case, body mode is hit at the baroque. So also this volume mode supports two polymers. First lines of our system are and that Jesus is broke so purchasing body using the pushes in the bottom, correct user can define expected the body of model through the TV ship body mode and also, if body mode is not specified.

A

Pvc decreased volume with volume model code file system. This keeps compatibility to current holder current existing behavior.

A

South winds, proto definition, so as a forcible definition, we added a new API field volume devices, so volume device hub supports two parameters: first minds: name, second wines: the device bus. As for the name parameter, this is equals volume name and that's what the device bus this device bus is is used. This device bus is a actual device pass inside the container. So in this case, so the device pass is defined out. The floodway xvd, a so in the container user.

A

Application can consumes low block device by using a slot of X so that exp da dairy, free, also interpol, the definition user can define user can use a read-only parameter if this parameter is used and set as a true. The volume is positive, as read-only block polymers so use a application channel to light to the rock volume only need from the existing data from the block volume and also they're a solid combination, coffee limitation easier. So both the volume devices this parameter only supports a podium model.

A

Should the people I call broke and also when using the podium month's body motors should be for assisting. So if user defined the other cases, potent Eva finish to create.

A

So next I show the body mode binding rule for the revolution, particularly vision pushes in the volume disc. So in order to support road block volume, we added a new binding blue into the persistent, the podium contour area. So there are three conditions of the podemos. So first one is the sketch white. Second wine fire system and the cell lines broke. So if user isn't specialized volume mode, body mode is considered as a file system to complete compatibility with existing behavior.

A

If a user requests low block volume with the persistent volume cream mode field, then it can only bind to PP, which has the same volume mode. So this means in order to bind the PV and P bleep body. Your model should be how same VAR. You should have same value.

A

So this race shows a quick glance of the block volume. Persistent podium passes in the volume cream, so we also added a fix to sugar control command. So you that youth, quick control, this correct, described command to the persistent podium or rushes in the podium cream in the commander result. Body mode parameter is shown, so you can see you can confirm volume mode and using the quick on Cuba contour describe command. So please note this podium model is shown only if the future gate is enabled.

A

Otherwise the this field isn't shown in this command right result.

A

So next is finds describing to the detail described to the port definition. So in this case, new field devices is shown under the mounts point. So using the control described for pot user can see the devices tagged and also device paths. In this case, the reference is a Philadelphia, V da and also containers containers volume name is shown and volume name is connected to the passes in the podium cream. So, as you can see, the volume data is connected to the poshest in the podium cream broke, PVC, 0, 0, 1.

A

This one shows inside the container report, so fan support is running user confirm if the actually block device is attached to tooth inside the container. So this result shows areas area today and, as you can see, still a dbx PDA is shown under the stratum. Sila devil directory.

A

Also, since that this blood volume is writable use like young consumer DD, the use a key issue I owe using DD to the low block to block device. They actually like, like this.

A

Okay, so next I'd like to explains information deep type, so in this chapter, I'll explain about the three topics: first ones : its volume, traveling interface.

A

This is this is for struggling developers or firm implementing to the storage problem, sewage problem. You must house how many interface so I've explained about that. Also, second wines. The directories and accurate note, so fair pot with volume is started. On the accurate note there are. Two directories will be created on the cubed node. First, one is portable. You, military and second wines are probably actually I explained about the details. Each directories for usages south winds avoiding inside and the bottom is face replacement, defines the during development of this feature.

A

We noticed this program is happened, so I have explained in detail later.

A

So let me explain about kubernetes podium, frogging interface, so in order to implement the fridge plugin for quality storage, so at least probably have to have a mountain under a mountain top interface.

A

The interface is used for the function to mount an amount volume to the curate note and also into the container, and also plugging kibosh optionally, attach the touch interface. This interface, if probably needed, are needed, if probably has a crowd of provider attached functionality.

A

Also, provision and editor are needed, if probably provides the dynamic provisioning functionality and also recycle, as was a recycler. This is needed if problem provides, the volume scrubbing feature, recycling, features and forwarding to interface new interfaces. First, when the block volume, a PI interface and a block copolymer second whines block volume, a PI interface, this interface introduced for block podium, robe local body, must support.

A

So let me explain about current mount and unmount interface quickly, so this is for the file system volume usage. So this this this interface used to make data source, such as a blog volume block devices network, share et cetera, available as the every on Charette node Tibetan, no loss due to file system directory and the volume is mounted into a container by curate, contain at runtime interface.

A

So these are terrasphere methods are implemented, so these methods are always called on. The curator note by curator, binary, so mountain and the amount of interface requires some methods, such as the set up, art or art, etc. What was the feet apart? Functionality, this method executes, attach and mount Boreum to cuba, data node and also here down at these methods, execute amount and the attach volume. Hood of curator note.

A

So next one is block podium up and interface. This is new and force logo podium. So this interface is make data source, such as volume of a block device available as a probe device on cubed node, then the block device is possibly into a container by container runtime interface on the Quebec node, and there are these methods always quotes, also from accurate note on the cubed binary, so prop mapa, reporting, mapa and a map interface requires only two methods shown here. So first minds set up device and second one is a teardown device.

A

So as for the purpose of the setup device, this method do execute to attach Boreum to cuba, retinol and second, one here down device this method to execute, to add a touch body from cubed node. So, as you can see, there is a note. So if we are probably has a touch of a touch on the touch interface, these methods could be no operations because attach and detach operation done by persistent volume controller so attach and detach functionality not required inside the setup device over here and device.

A

So let me explain about dear Therese and as accurate note, so I suppose a lot of the directories. Another tribute note photo op or the podium directory, and this directory is used to mount a formatted volume per pod. Then this victory is mounted into the container so inside the container application access. This mounted mount point. Just was there data's and sequins are traveling literally. This victory is used to store Province basic data, specific configurations and then also this is body. This ball volume.

A

What the podium is mounted to this directory so, for example, ice cracking and RVD plugins towards a connection information to this problem directory. So using this information class Albany can find the Fitch block device is used on future application. Application pod for the other body, troubleshooting purpose.

A

So, however, about the block podium case, the volume is not home at it. So therefore they can't be mounted a trace. So, instead of mounting, we curate as well at the block volume case theaters to us symbolic link which is associated to the block device, such as slot ABCD x2, these greater ease. So as a result other way, you can find the physical block device responding to user pushes in the podium and pushes in the volume cream by checking the symbolic this nation, shimmery cream.

A

So let me explain more detail with these victories. So later he showed the mountain, the amount, our case so mountain the mountain case. So each pod has own port volume directory and that's a very accurate port for the UUID kubernetes io.

A

I watch on air for their and the ballroom is formatted and mounted, and Boyle name dear, such as very accurate port for the UID kubernetes IO. If she thrush purchased a volume name directory, so you have pod has multiple volumes which volume directories are created and put the volume directory so admin you can find physical block device by checking on these amount points and also broke Orion mapper and a mapper case. Each pod has own for the volume directory and that's very private parts for the UID volume devices.

A

Kubernetes io should I see so that the difference between mount our file system volume and volume system, so we added a volume devices directory and that for the UID directory, also symbolic link associated to the block device such as Radiesse DX, is stored under the total volume directory. This is visited. This symbolic link, associate tutors saludable seda OS is like a CD X and also, if you have pod, has multiple boards, multiple symbolic links with volume name throughout under that.

A

Yet to really so in this case, also, admin can find physical block device by checking such memory clink.

A

A

Yes, so in this case was a peewee named symbolic link associated to the Scylla dsdp. It won't shed a block device on the cubed note, so other mini cam finds the disc and physical disk on the given node for the troubleshooting.

A

So next one is probably in directory on cubed node, using the procedure to actually probably so I suppose a mounted case, so each volume has all travel in directory and thus very accurate, plugins cubed IO if she world why the name Adrian Adrian number in this case and the volume is formatted and mounted under the plug-in directory. It's the same. So in this case, II wave much reports use the same volume. Only one probably directory per volume is creative and also some storage configuration is to add another.

A

This victory, another block volume upper and upper case. Each volume has own plug-in directory underbody curate, plugins', cubase, AI, yo, f, xi, / volume devices worldwide name. Are you in 0? So in this case also we added the volume devices directory under the FC directory and also symbolic links associated to block device such as red, wavy, DX, to add under the drug injector e. So if much reports use the same volume in this case, multiple symbolic links with which has reported already named assad under the directory.

A

So a trust I'd like to explain about our silent volume replacement program. So during the development of this feature, we found a program that container runtime does not take a look to attach the block device, continual rain. So as a result, volume is the phrase silent or to another volume. The initial application is issuing I owe to rock volume inside this container. So this is the big problem for the user perspective.

A

So to avoid this program, we added a logic to open a device device file of rock polymer, such as a straight away, CX using a loose rock device file. So by using a rooftop device. This takes look using the file descriptor. Then the user can avoid device, high rental disgraceful, represent program. So, if you zoom in across all domain, you may notice that there are many those block device on cubed note. If user uses a dog block device out of rock bottom on the cubed, no.

A

No, actually, we saw under development, so we honed this program and we served before merging the disc feature. So there are no issue reported.

A

Ok, so I rust extremist future work. So our.

A

Experience so they make speculative dynamic provisioning is not completed, so she was bathroom with one point care we will discuss about the spec for that dynamic for German provisioning, the finalized that spec also in order to moving from the Alpha feature to be the feature we need to other more easily test cases or unit test cases to choose the local podium and also since country. Only one fibre channel flagging is supported.

A

We need to expand more bottom plugin support, so we are targeting local volumes, gcpd WB s, s kg r, BD raster, for but on one point team.

A

We are welcome to stop contribute, oh yeah yeah.

A

So yes, yes, please.

A

Any engineer from the CSS spec team to low block boring sample team.

A

So she's asking he's asking so he wants. You support a cs5 fragment at first except we 1.10 like.

A

Okay, so let me get we discussed at this meeting. I was that okay, yes,.

A

No, no previously are required.

A

So ret Amelia pubs decisions, so we introduce the new row problem which to version 1.9 as our future, so the I mentioned this in using this future is like use a low blood volume inside the container, without any previous talk show. Also, we added a new, a p5 field, customize the volume mode. Second wines, the volume devices also currently only five Italian flag, green suppose, dobro podium. Ok, ok, ok, that's all! Thank you very much.