A summary of our discussion and presentations during our first-ever DoK Ecosystem Day.
DoK is a vendor-neutral space, but we also know organizations rely on vendors for different parts of their stack. In January, the Data on Kubernetes Community gave vendors an opportunity to be shameless for its inaugural Ecosystem Day event!
Each vendor was given 5 minutes to present a lightning talk.
The following provides a brief overview, video, and transcript of the event.
Ecosystem Day
Presented by:
- Cindy Ho, Senior Technical Product Manager, Dell Technologies
- Robert Hodges, CEO, Altinity
- Peter Shuurman, Software Engineer, Google
- Dean Steadman, Senior Product Manager, NetApp
- Alvaro Hernandez, Founder and CEO, OnGres
- Edith Pucilla, Technology Evangelist, Percona
- Matt LeBlanc, Senior System Engineer, Avesha
In this town hall, various DoK community members and industry leaders in the DoK landscape demonstrate how they provide value to the community and end users in 5-minute lightning talks.
Cindy Ho – Dell Technologies: Crossing the Storage Chasm
- How Dell delivers enterprise-class Kubernetes data storage capabilities through Dell APEX Navigator for DevOps teams that need storage capabilities that go beyond standard Container Storage Interface (CSI) drivers as they adopt a multicloud approach.
Robert Hodges – Altinity: Helping Users Build Fast, Cheap, Modern Analytic Stacks
- How Altinity enables customers to operate high-performance ClickHouse on any Kubernetes cluster and account and includes support for on-prem operation.
Peter Shuurman – Google: Cost Effective Availability with GKE Stateful HA Controller
- How Google balances cost and availability for stateful apps on Kubernetes using GKE Stateful High Availability Controller. Peter covers two stateful architectures where the GKE Stateful HA Controller can deliver the sweet spot on the cost/availability curve.
Dean Steadman – NetApp: Intelligent Data Infrastructure for Kubernetes
- How NetApp Astra simplifies the way to protect, move, and store Kubernetes workload across hybrid multi-cloud environments.
Alvaro Hernandez – OnGres: Sharding Postgres on Kubernetes
- How to use K8s operators to create sharded Postgres clusters–one of the most complex Postgres deployments–a breeze with a dozen lines of YAML (or the Web Console!) and create production-ready sharded clusters with coordinators, workers with high availability, connection pooling, and more.
Edith Puclla – Percona: Automating Database Operations with Percona Kubernetes Operators
- How to automate database operations using Percona Kubernetes Operators, a solution free from vendor lock-in. Learn about automation examples that significantly reduce the complexity and time involved in deploying, scaling, and managing databases, ensuring efficient and reliable infrastructure.
Matt LeBlanc – Avesha: Breakthrough Data Gravity with KubeSlice
- Learn how Avesha’s KubeSlice enables easy connection between multicloud K8s clusters and discusses two use cases: migration and burst/partial migration.
Rob Reid – Cockroach Labs: Mission-Critical Applications in Kubernetes with CockroachDB
- How CockroachDB is built for the cloud and uses distributed SQL that survives any outage, scales horizontally, and ensures data consistency whether running in a single Kubernetes cluster or multiple globally distributed clusters across cloud providers.
Watch the Replay
Ways to Participate
To catch an upcoming Town Hall, check out our Meetup page.
Let us know if you’re interested in sharing a case study or use case with the community.
Data on Kubernetes Community
Website | Slack | LinkedIn | YouTube | Twitter | Meetups
Operator SIG
#sig-operator on Slack | Meets every other Tuesday
Transcript
Unknown Speaker 0:02
Go ahead and let the people in. Okay, great
Speaker 1 0:16
All right, welcome everybody. We’re gonna it’s going to, we are going to be on a
bit of a tight timeline, want to make sure everybody has the chance to speak but
we’re just going to wait a couple more minutes before we get into some community
information and then we’ll get right to the topics
Unknown Speaker 0:43
wait till 1002 And I’m going to skip put my slides up tackiness?
Speaker 1 1:25
You’re just hopping in welcome. Happy to have you here. Our first ever D, okay,
ecosystem day.
Speaker 2 1:45
All right, well, 1002, I think we should just kick this off. Hi, my name is
Paul. I’m the community manager for the UK community. Today’s our first DMA
ecosystem day, where we’re gonna start hearing from industry leaders in the UK
landscape, see how they’re providing us value to the community members and end
users. We want to thank our DoK community sponsors, our gold committee sponsors,
our Google Cloud and Percona. A couple of announcements for the events that are
taking place have a do K track at scale. And that’s March 14 through the 17th.
And then also DoK day at KubeCon in Paris on March 19. So I’ll be there.
Hopefully, I’ll get to meet all of you as well. If you’re interested in getting
involved in the community, you can scan this QR code, I’ll take you to the place
to do that. And then if you’re also interested in becoming a sponsor, you can do
that as well. Just a few announcements about the format. This is these are five
minute lightning talks. And we have eight people on there’s gonna be a little
bit of introduction and in between time, so we’re gonna limit questions to the
chat. So if you put your questions in the chat, we’ll then ask our presenters to
answer those questions. And then we’ll, we’ll add those to a blog post their
responses along with the video of this event. So feel free to let the questions
flow in the chat. And yeah, we won’t have time to get to them during the actual
event. But we’ll make sure we get those questions answered. And then also, if
you want to be connected, I’ll show this slide again, to any of these of our
presenters, you can fill out this form, you can use that link there that will
give us your information. And we’ll connect you with with our presenters. And
this is not as it says here, not a sign up for endless emails. It’s just a way
for us to introduce you to to our different presenters. All right. That being
said, we have eight wonderful presenters, and the first of them being Cindy
Howell from Dell Technologies. So I’ll let her take it away.
Speaker 2 4:13
Thanks, Paul. Everybody, quickly share.
Unknown Speaker 4:21
I’ll start your timer right now. Can everyone see ya?
Speaker 2 4:28
Oh great, everybody. I’m a product manager at Dell Technologies. I’m also joined
here by Brian who some of you may also know our lead developer advocate who’s on
standby for any questions. Get started. So today I’ll be talking about about how
Dell is helping our customers across the storage chasm, with their Kubernetes
workloads. This is specific to our Kubernetes data storage software. So Dell
primarily works with mid large sized enterprises providing services As storage
for data data intensive workloads, as many of you are aware of, although a
majority of these customers are VM based Kubernetes is becoming more and more of
a priority, there’s more exploration in it, which is why we’re progressing
progressing our Datastore software to make our primary storage capabilities more
Kubernetes native. And this, this started with the container storage interface,
dealt with amended the universal specification to create our own CSI drivers to
allow data persistence with our block and file storage, and more recently, our
object object storage as well. Just a couple a couple of months ago, we
introduced the container object storage interface, cosy drivers. So that’s been
exciting. But with all of this, and from experience talking to customers, these
are just table stakes. For most of these customers, there are a lot of data
management problems that CSI does not currently tackle. So as for our next step,
we decided to use these CSI drivers as the foundation to build out more advanced
capabilities. And these are called the container storage modules. There you can
see them as an enhanced layer on top, further accelerating deployment testing,
for even faster deployment cycles. So today, we have seven of these modules,
they have different functionalities on top of the storage arrays, and they
intend for some of them, they introduce open source tools, and technologies that
you guys know very well, Prometheus, Grafana, opentelemetry, rustic just to name
a few. For some others, they take these enterprise class storage capabilities
that already exist in the storage arrays, and then make them Kubernetes. So
Kubernetes native, so we do the reverse there and kind of combined the best of
both worlds. For example, our replication module leverages existing replication
capabilities of this Dell storage arrays that weren’t originally actually
designed for Kubernetes. Customers can pick and choose which modules they they
would like to use, and for which clusters. And among these authorization is the
most popular allows Kubernetes storage admins to apply our back rules and
instantly and automatically restrict cluster attendance usage of storage
resource. This, lets just a really quick overview. But we’ve seen that
Kubernetes popularity continues to grow, we have more and more customers using
these modules. And one question that we’ve asked ourselves and tried to focus in
on is how to make deployment operations of our software even more simple,
efficient, and scalable. Many of our customers today are new to Kubernetes,
especially in the storage realm, a lot of them, you know, are just exploring
right now. And they and then we also have staff just with a lot of different
priorities, like what do they want to use our tool for. So making the entry
point to our software as seamless as possible, has become a very top priority.
And with this, this is why we are introducing a new platform, we are developing
a unified user experience and user interface. So it’s tying all those pieces
together and adding on another layer on top, built specifically to simplify
multi cloud multi site Kubernetes data storage management. This is called the
apex navigator for Kubernetes coming out later this year, and integrates
installation and deployment of our Kubernetes data soft storage software, and
provides an intuitive GUI that allows staff who aren’t Kubernetes experts, not
Kubernetes savvy at all, to easily complete jobs without even needing to use
command line or having any baseline knowledge. So we’re excited for what this
offer means. This year will be very exciting for us. And we’re also you know,
this is just part of our continuation to build our entire portfolio to better
enable data and Kubernetes streamlined experience over and over at simplicity
and performance. So I know that was a lot in a couple of minutes. Thank you all
for any questions. Feel free to use the chat. I don’t want to take up any any
more time.
Unknown Speaker 9:09
Paul, you’re on mute if you’re speaking.
Speaker 1 9:13
Thank you. Thank you. Yeah. Thank you. You came in, just under. So great job.
Thanks, Cindy. Yeah, again, if you have questions, feel free to add them in the
chat. And we’ll try and get those answered in the blog format. Next up, we have
Robert Hodges from alternity and not take it away, Robert.
Unknown Speaker 9:37
Oh, Robert, you are also on mute if you didn’t know that.
Speaker 3 9:43
Tricky. Thank you very much, Paul. I’m here to talk about building real time
analytics with Kubernetes and alternity. Alternative that’s the company that’s
- I’m the CEO of the company. Let me tell you a little bit about us. So we are
database geeks. I’ve been working on databases for 40 years. We have 45 people
in the company worldwide decades of experience in databases, and particularly
analytic applications, the folks we typically talk to, and I think a lot of you
in the audience are app developers, you’re looking in particular to build
analytics, real time, solve business problems, and you love Kubernetes. So to
wait, so we do support and services for clique, so real time analytic database.
I’ll be talking about some of the pieces. But the way we got into this was we
were the develop the operator for clique house that is widely used in 1000s of
deployments worldwide. And we built an entire cloud on it. Let’s talk a little
bit about that. I’m going to start with clique house, excuse me, it is a real
time analytic database. That means is designed to read vast amounts of data and
spit back answers in a second or less, by vastly mean trillions of data, or
trillions of rows of data. In many cases. It’s kind of like a combination of
MySQL, an open source database that runs anywhere, and a traditional data
warehouse like Vertica, which is optimised for high performance in and reads,
and you can read some of the features here. So what can you do with a database
like this? Well, it enables you to answer questions on data that is rapidly
arriving, and where you need to get answers very quickly, either because you’re
giving information to a piece of software, perhaps it’s rendering a page and
needs to add something to it, or you’re an analyst. For example, in financial
services, you may be trying to figure you may be a trader and trying to figure
out the best trading strategy to follow. You have an input stream millions of
rows per second of market pricing data, you’re iterating through your different
visualisations are different projections of your asset positions, depending on
your trading strategies, looking at things like bid ask spreads, that’s shown
graphically and allows you to set your trading strategies. So that’s a typical
use case for click house, there are many others. So let me just hold on this for
a second. Where we come into the picture is if you decide that you’re going to
use click house, we can help you build the application no matter where it runs.
So think you know, frame the problem, do the performance analysis, optimise
schema, this is support, we do it on any platform that you choose to run on. But
what’s even better for most people is if we run it for you, and for that, we
have a alternative cloud, which runs clusters on Kubernetes. For every customer,
we spin up a separate environment one or more, you can run them in our V PCs, so
on on Amazon, Google and Azure. Or you can run in your own Kubernetes clusters
that run in the cloud or even on prem, we have secure connectivity that allows
us to come into any Kubernetes, cluster and manage clusters there. And those
clusters can then live alongside your applications. What does it look like when
you’re managing these? Well, it doesn’t really matter where you are, we’re going
to present the same UI, which will allow you to do all the types of operations
that you need to do to run high performance analytics. This is just one example
where we allow people to scale the clusters dynamically, click OS, support
sharding. It supports replication between those between copies of the shards, we
can switch machine types, we can increase storage, all of these things we can do
with just a few clicks. And as much as possible, do them, do them while your
applications are running. So big thing in this type of in this type of system is
to be fast, but also to be able to make changes with zero downtime. What’s
inside alternative cloud? Well, we’re on Kubernetes. Yep. So was that a ball was
that what was that symbol?
Unknown Speaker 14:03
That was the one minute warning.
Speaker 3 14:04
Other women are warning. Okay, thanks. What’s inside a lot of open source
software, I won’t go into details, you can actually run this yourself.
Everything that we that we build into your stack is open source. And what we do
is we help you make it super fast on Kubernetes. So we wire together things like
CSI, networking, let you choose instance types. These are just examples of the
performance that we get such showing sub second response on Kubernetes. In fact,
we’ve never seen over the years of using Kubernetes. We’ve never seen any real
performance impact from running on that platform. Instead, it’s mostly just big
advantages. So I if you want to, if you’re doing analytics, building real time
systems, you’ve picked clockhouse. Come talk to us. We can help you in three
different ways. We can run it for you on our alternative cloud cloud platform,
we can help you build a Kubernetes cluster clusters yourself and manage them
yourself. And then we can give you software and enterprise support anywhere you
choose to run. That’s us. So hope to talk to you. And you can ping me on the
DLQI slack if you want to talk to me directly. I’m out there and watching for
for messages. Thank you very much, Paul.
Speaker 1 15:17
All right. Thanks, Robert. Next step, we have Peter Sherman, Google. Take it
away, Peter.
Speaker 4 15:28
Hey, everyone, Selfridge here on Google, working on storage infrastructure.
We’ll be talking about the GK staple ha controller new tool that can help you
balance cost and availability for block storage applications on Kubernetes. So
let’s talk about availability. I think we can all agree availability can be
expensive. And why is that? data replication. So if you want your data to be
available during a zonal outage, you need to move your data across zone
boundaries at the application layer. All major cloud providers charge requests
or egress. And this replication can rack up depending on the data throughput
your application happens. The other challenge is extra compute. In order to
replicate your data you need to compute running in multiple zones, do you have
to replicate the data at the application layer. However, for many applications,
a single replica properly scaled is potentially capable of handling all your
reading rights. In this case, the extra compute capacity is functionally only
necessary to provide availability for your data. In order to replicate it. What
you really want to be paying for is just the storage replication and not the
compute required to replicate it. Let’s go to the other end of the spectrum,
single replica apps that aren’t highly available. Kubernetes does have some
automated failover capabilities and rescheduling if there’s infrastructure
failures, but even this is, has conservative defaults and cluster wide defaults.
Failure detection depends on Cuba reporting and healthy replications tolerance
for node unavailability takes. In the case of block storage, the volume forced
to detach interval. The other challenge is data availability. So in the case of
his own failure, your data may be durable. For example, GCS disk, has five nines
of durability, but what you really care about is the availability of that data.
In the event that the data is not accessible, or services are unavailable in a
particular zone, your data is effectively lost for that period of time. So, cost
availability are correlated, you can either have low cost, low availability, or
high cost high availability with multiple replicas. I ideally you’d be in the
left top left quadrant here, where you’d have low cost high availability. And
the stateful, he controller can deliver a balance of that. If your application
architecture can converge to a specific model that didn’t match that domain. For
single replica apps, that gives you Multi Zone data availability, and
application specific rescheduling speeds. So you can put an upper bound on how
quickly your application can work on the failover period for your application.
And for multi replica apps, you can achieve a significant reduction in cost by
moving to a single zone. But still having data availability in the event isn’t
failure. So how does this work? Let’s talk about what it how it’s built. It’s
built on regional persistent disks. So this regional persistent disk is GCPs
synchronised only replicated block storage products. And the nice thing here is
you only pay for the additional storage space. So effectively, your storage cost
is doubled. But you’re not paying the costs of replicating that data. So if you
have I crossed network egress traffic, that cost line goes away. The other
benefit or the other building blocks here is the staple H A controller is
running in GK is dedicated control plane is able to detect node failure, it’s
able to evict your staple replicas from a failed node and reschedule the
application on an alternate alternate node in the zone. Working alongside the
Korea scheduler, it can mature applications regional Persistent Disk fat volumes
failover, within a bounded period of time, to your new node. So you get to
control how quickly your application replicas get rescheduled after no failure.
So let’s talk about two case studies where we can take advantage of this. So
first example Kafka. It’s a standard three replica application. So three zones
running through it through a book as running at three zones. It was really
originally developed by LinkedIn, or Crasto network was effectively free on prem
if you own a network, network stack. But Kafka can produce vast amounts of cross
zone network traffic that occurs egress costs. The other challenges are to work
like a See, the other thing to consider here is RTL. If you move all your nodes
to one zone, you get the same RTO of a single node fails, but there is a in the
event of a zone failure, all your brokers will need to be rescheduled to an
alternate zone. So the recovery is kind of a trade off here, we got quite a
significant cost reduction for this pricing pricing model here 3%. But the the
worst case fell over time does increase. So that is a bit of a trade off to
consider. And then the other case study is a standard single replica relational
database, you may have multiple reasons to run a single replica database
certification cost optimization, but stateful, stateful stateful, HJ can take
the existing architecture and give you an upper bound and recovery time, also
giving you data durability, cost. So if you want to learn a little bit more, we
have a blog post out you can search for GK staple which a controller, there’s a
link to the preview there, you can test out installing your own cluster NRG is
coming soon in a number of weeks. Thanks for your time.
Speaker 1 21:04
All right. Thank you, Peter. Appreciate it. All right. Next up, we have Dean
Stedman from NetApp.
Speaker 5 21:12
Thanks for having me. My name is Dean Steadman, I’m a product manager here on
the NetApp team. And for folks that aren’t familiar with NetApp, we are a 30
year old storage company, and data services company, we really started our data
journey with our customers really focusing on, you know, helping folks, you
know, manage both structured and unstructured data. And so over the years, we’ve
evolved as the market has, and, you know, most recently, you know, beating,
being able to offer up our customers, our storage services as a first party
service and all three hyper scalars really allowing customers to build out true
multi cloud hybrid solutions. And then as we look to the future, we’re looking
at adding in more and more intelligence into the way that folks manage and
maintain data. Now, rather than walk through kind of the boring marketing slides
that we’ve got here, I found that for these lightning talks, adding a little bit
of humour into it makes this a little bit more memorable for folks. So what I’d
like to do is I’m going to introduce three concepts that our team really focuses
on when we talk about managing data for Kubernetes. And so to do that, I’m going
to, you know, like I said, use a little bit of humour here, use some memes that
hopefully will lock into your brain, make you think in that app when you run
into any of these challenges. So the first theme that we kind of take a look at
is making sure that we allow customers to manage applications and data together,
and making sure that that linkage between those two is built into everything we
- So that as you’re rolling out applications, you automatically know you know,
how they’re going to store, how they’re going to be backed up, how they’re going
to be protected, and what their performance attributes are going to be. And the
more of that that we can bake into everything from the get go easier everyone’s
life is. So in all of our solutions, we take that application first mindset and
make sure that apps and data live together for easy use. Second thing here is
that if I’ve been in the storage industry for 20 years, I can tell you, it’s not
the most exciting thing in the world to manage a bunch of ones and zeros. So the
more we can do to help customers to automate their workflows, and to build best
practices for data management into applications, so that every application is
born and deployed automatically with the right performance characteristics, the
right amount of capacity, the right replication layers, the right data
protection capabilities, the more of that, that you can inject into the left
hand side of the the deployment lifecycle, the happier everyone’s going to be.
And you know, as a as an IT guy. Quality of Life of IT people is something we
don’t talk about a lot. But getting rid of the boring parts of our jobs and
having the automation workflows, pays dividends. So keep that in mind. Our third
attribute here that we really make sure is that the mere existence of data
implies that you’re going to need some type of data protection, and whether
that’s ransomware, or whether that’s backups or whether that’s disaster
recovery. All of those different use cases have different requirements. from a
technology standpoint, we offer different solutions at different layers for all
of these things, really making it easy for our customers to implement data
protection in for their applications at the right level as easy as possible. So
if you’ve got data protected, super simple. Last year is just a little bit of a
look at our portfolio. I’ve kind of called out three different pillars that our
team really focuses on. The first is our ONTAP storage operating system. This is
an operating system that was born on premises and has Now migrated as I said, is
a first party service to all three hyper scalars gives our customers the same
capabilities, the same tool sets all regardless of where they’re deploying. We
add into that products called Astra and trident. Trident is our CSI Driver Set.
And then Astra is a data management and data protection platform that we have
that augments that adding in additional capabilities. And then finally, we have
a set of solutions from spot, which are our observability. And data management
tools. Really helping customers focus on REITs are using both from a resource
perspective, but then also a cost perspective. And NetApp, we always try to make
things super simple to get up and running. We allow customers to start with our
Astra product suite, you can manage 10 namespaces for free forever. Super simple
to get going with the products. Give it a test drive. And if you have any
questions, I’m available in the doc Slack channel. So nice to meet everyone.
Thank you very much.
Speaker 1 26:08
All right. Thanks so much, Dean. Yeah, we’re moving well on track. So we may
have time for questions at the end. But what if we keep moving to the right clip
to next we have Alberto Hernandez, Iris.
Speaker 6 26:22
Okay, hello, everybody. Let me share my screen here. All right, the screen.
Okay, thanks. Okay, so I’m going to talk to you about sharding and sharding on
Kubernetes. And I’ll just dive in directly, let me briefly introduce myself
first. I am the founder and CEO of a company called Congress. Congress is the
short for on Postgres. So you can imagine what we do for people that know me,
and the post response response was painful. If you call me three times, I will
pop up anywhere you are. I’ve been working for passwords for quite a long time,
or even 20 years already. And I like to work on r&d research and development
trying to come up with stupid ideas. Some of them become something tangible, the
software, like, for example, staggers, which is a software I’m gonna be slightly
talking about today that we have developed as a fully open source project for
running clusters on Kubernetes. I have done a lot of tech talks around 130, or
close to 104 years of today, they’re all online on my website, ah, t.es, quite
short. So you’re going to check them out. There’s a lot of talks at the Dlk
also. So feel free to find me over there. And I’d also like to do some
nonprofits. So I run a prosperous nonprofit foundation. And I’ve been elected
also as an Amazon hero 2019. Let’s talk about sharding today. So very briefly,
what is sharding sharding for horizontal scaling is basically splitting the
workload, the data that I have from our potentially large database into multiple
writer instances. Most relational databases, for example, have a typical
architecture of a single primary and multiple writers sorry, a single right or
no and multiple reads. But if you really want to scale this writer node, despite
that normally scaled very, very well. But if you want to scale this writer node,
you need to split the data into multiple chunks, and direct each of those chunks
to a single writer instance, this is what it means sharding and horizontal
scaling. It allows you to do basically two things to scale the writes,
obviously, but also to reduce the blast radius. If one of your writer instances,
for example, fails on even the HA mix mechanism fails also, or there’s some data
corruption, you will affect only a percentage of your users. So it’s also very
good for security and availability purposes. Now, a typical sharding
architecture with a relational database like Postgres will look like this, you
will have at the bottom, all the shards, like where all the data is split into
this chunk that will go to a specific server, which potentially will have a
replica. So here we have primaries and replicas for high availability purposes.
And on top of that, you have coordinators routing, or transaction routers,
they’re called differently in different technologies. That word obviously
coordinate the queries receive the query serve as the entry point, and then
spin, sorry, send queries to the appropriate shard or shards and resolve the
queries, they may also be highly available. So this is an architecture that is
easy to understand. But if you think about how to deploy this architecture is
non trivial. You need to deployment one server but n plus one. But then if you
want high availability with n number of instances, you multiply that then you
really need to go out typically at least for Posterous, connection pooling and
specific configurations and potentially specific extensions, and then functions
to build the clusters and what about distributed backups. So operationally it
becomes very complex. So what we have done at stacker is thanks by running
Kubernetes things about by the power of CRDs is to create a custom CRD that
makes deploying this whole architecture with tuning with high availability with
almost everything that I mentioned here. As simple as typing this channel. Can
you type is gentle, then you’ve got that sorted cluster immediately. And it’s
very high club, it just talks about number of instances, the size, the versions.
And that’s pretty much it, then you will get on the web console. If you use
Cypress for these, you will get a nice UI where you can see all the status of
your coordinators, the charts and all the characteristics of it. So just to
name, the features that this this supports already is, first of all, is
supporting the site to site use the sign extension for Postgres, which guess
what it does sharding. And we just take all the power of cycles and make it
extremely easy to orchestrate. It supports both highly available high
availability for both coordinators and restaurants. It does integrate connection
pooling, which is very important for situs. It supports distributed backups, so
you get a consistent backup across all your shards. And also, and this is very
unique in the industry now is not even present on Asier, at least yet. It will
use charts, you know that distribution of data may not be homogeneous, and one
chart needs to be bigger than the other one, we can achieve this with a java
file by overriding the specification for a particular service or service, and
also automated operations for restarting and restarting a cluster. So that’s all
that I wanted to talk about today. I’m also available on Dlk, Slack, Twitter,
LinkedIn anywhere, ping me for any questions you may have. Thank you.
Speaker 1 31:26
All right, thank you. Barbara came in just under five minutes. Perfect. All
right. Next we have Edith bleah. From Percona. Take it away.
Speaker 7 31:39
Yeah, let me share my screen. Okay. Okay, is this baseboard? Good to go? Use a
second, I’ll put the timer yet. I will set everything good. And, okay. I’m
ready. Big. Yeah. Hello, everyone. Thank you for having me. This is Ed Bucha. I
work in the side of the community at Percona. And today, I’m excited to talk
about personas contributions to the open source community, especially focusing
on the data on Kubernetes. But before starting, let’s see how this is related
with they don’t give this an operator see. They don’t give unless operator seek
discusses gaps in information around Kubernetes operators for industry and CO
creates projects to fill those gaps. But Ghana is a member of the donkey rom
this community as a database solution company Percona works with different
databases like MongoDB, PostgreSQL, and MySQL. And when we use databases, we
talk about Stateful applications, the need to keep safe and secure our data with
Kubernetes. with Kubernetes, we deployed our database, and it became
challenging. Because we know that Kubernetes was initially designed it for the
stateless applications, and not for a stateful application like databases. We
call the things gaps with solutions. But extending the API of Kubernetes and
developing Kubernetes operators, we were able to fulfil the majority of this
gaps. Let’s explore some of the use cases. Our operator can create clusters of
open source database ready to use it means that deployment complexity is
eliminated. And we include in this issue such a configuration errors both notice
starting also notes that are failing to yo yoing into the cluster. With Percona
operators, we can automatically scale backups, restore and upgrade the database.
There is no need to worry about a storage, over provisioning networking traffic
spikes, applications and services though times or data loss. We can also
integrate Percona operator with your existing infrastructure as a code tool like
continuous integration or Continuous Delivery pipelines and automate the one and
a two operations in the Kubernetes. Applications lifecycle. That always is on
Kubernetes are commonly considered complex. But with with broken operators, you
can simplify database management enabled easy migrations to Kubernetes or
respond to demand spikes in a flexible manner. It also provides a user interface
and an API that hides the complexity that Kubernetes brings from for the user.
We will see that in a moment. We put on operator we can run databases anywhere
on premise on Cloud multi cloud hybrid cloud environments are also in 1000s of
IoT devices. Also, you can move data quickly from one cloud to another cloud
without users restrictions or vendor locking. Here are our Percona As you can
find in all on GitHub, we have Percona xtradb cluster operator we have. We have
also operators for MongoDB, PostgreSQL and MySQL. What is next for Percona in
the Cloner DB space? But Coronavirus is our next step is to not ever this is an
open source solution that enables you to create a prepaid database as a service
in your infrastructure, wherever you are a platform engineer, DevOps specialist
or developer. Ares allows you to deploy and manage database clusters with a
friendly user interface, hiding the complexities of Kubernetes YAML conflicts
files and hubs configurations. areas also utilises Kubernetes operators to
deploy database cluster, you can benefit for functionalities, such as scale
esterase, restore backup, and advanced monitoring capabilities. What are our
unique benefits in our in our Percona database operators, we are completely open
source and free from vendor locking. Our operator are supported by Percona, and
the open source community. And you can find those in our Percona community forum
that is always available to help us in our website. Thank you.
Speaker 1 36:20
All right. Thank you so much, Edith. All right, we have two more speakers. Next
up is Matt LeBlanc from the vicia.
Speaker 8 36:36
Always practice, right. How you doing? My name is Matt LeBlanc. I’m a senior
Systems Engineer here at avecia. Today I’m going to talk about cube slice and
how you can use it to break vendor lock, move your your applications to any
cloud infrastructure, we’re gonna be focusing on that one hybrid cloud migration
use case. But it does apply to a whole bunch of others, which we’re not really
going to cover today. You know, one of the big challenges once you choose the
vendor, you’re more likely to be stuck in there, whether you’re originally a
Gmail user, and you end up using all of Google’s ecosystems, or you started off
with an often office 365 licence ended up in Azure, or you just started with
AWS, often customers are stuck in those spaces. And the reason why is once you
start there, that data is created, all the applications are designed to be close
to the source, making it harder to move. And eventually makes it easy to break
that vendor lock. And we basically use cute slice to move your start moving your
workloads to another cloud somewhere else, any flavour in Kubernetes. And here’s
an example of that, you know, we have this basic application that’s running, we
have our boutique online webshop. It’s a front end, storefront server versus the
right. And then we also have our payment services, the other cloud vendor out
there, and you’re stuck there, right, you’re gonna sue every thing inside
Google. But here’s how we break that boundary, your application connectivity for
fleets of clusters without changes to the application. This is a zero trust
isolation solution. We’re doing that by basically connecting your namespaces we
call that a cube slice or a slice for short. So how do we do that? Very simple.
We’re going to create a disk boutique cube, slice create that slice, we’re going
to add our clusters to it. In this case, we’re going to connect our g k e,
Google Cloud cluster to our Oracle or OCI cluster, we’re going to add that
namespace to that slice. And now once we’ve created that slice, and created that
secure connection, it is a is encrypted is reused role based access control. And
you have the ability to really isolate that applic application isolation
information. Now, once you have created that slice, you can then deploy that
application over in the Oracle environment that that new cluster. And once that
is up and running, you can retire that old front end service. Now the key here
is we didn’t move the data, we just redeployed parts of those front end services
elsewhere, that’s going to allow you to start a partial migration or hybrid
migration where you can start moving. So let’s see how that works once we have
done that movement here. So first, we’ll add a couple items to our cart here.
We’ve got the candle holder got five items, and now we have $103 in our shopping
cart. We commit that to our purchase orders complete. And now we can look in the
backend or the look at the cluster itself. If you look at that front end, that
is still that is now running on that Oracle cluster and we can check the logs
and basically see that same $103.94. Now, just quickly conclude here, you can
choose where you want your apps to run, don’t let data gravity drag you down. If
you have some use cases where you want to maybe start leaving your original
vendor, we want to move to someone else, or Russia is the solution for you. If
you have any questions, please feel free to contact me my email address is
[email protected], where you can scan the QR code down below. Thanks.
Speaker 1 40:33
All right. Thanks, Matt. All right, we are to our last speaker, which is Rob
Reed from cockroach labs. Hello,
Speaker 9 40:42
let me just share my screen with you all right.
Speaker 9 40:56
Right, thank you very much. I’m Rob Reed technical evangelist at cockroach labs.
And today I’d like to talk to you about running mission critical applications in
Kubernetes on top of cockroach dB, so it’ll make sense to start with distributed
SQL. And fundamentally, cockroach DB is a distributed SQL database. That means
it, it brings together a lot of the different database paradigms that we’ve seen
historically. For example, that means the reliability, consistency and
familiarity of databases like Postgres, and MySQL, the traditional relational
database management systems. RDBMS is, together with the no SQL databases, the
scalability, flexibility and resilience of those databases. Historically, those
two database paradigms have an overlap if we’re talking about a Venn diagram,
but that’s what cockroach DB brings to the table cockroach is a distributed
relational database, and it provides everything you’d want and expect from that
kind of database. It provides referential integrity, normalisation and the like.
It also provides serialisation by default, so you’re guaranteed that whatever
you write into the database, all of your consumers of that database will see
out. And by adding nodes to it, you horizontally scale, not just for reads, but
also for writes, every node in cockroach DB can handle reads and writes. So
scaling is quite a seamless process. And there’s no and there’s no master or
primary node in cockroach dB, as you see with a lot of databases. So once scaled
cockroach DB will rebalance the data across the nodes. Essentially, you can
think of cockroach DB as being a monolithic key value store, which looks and
feels just like Postgres, we chop up the data in each table. And then those
chunks are called ranges. And we distribute those ranges to different nodes
within the cluster for resilience and scalability. And scaling might be required
across regions. In this example, we’re scaling across California, Ohio and North
Virginia, you might also need to scale across clouds, cockroach DB has you
covered there as well, it doesn’t really matter, cockroach DB ultimately boils
down to a single binary that you can, you can we can either host for you in a
managed service, or you can run self hosted yourself. So it also being a
horizontally scalable database with no master nodes, it’s really good at
surviving multiple kinds of outage scenarios, up to including regional outages,
and even whole entire cloud outages if that’s if that’s your your fancy. So for
example, users in this demo in this site, they might have been talking to region
three, North Virginia, but that let’s say that whole region went out. And now
we’re in the realm of some of the most severe outages in history. The the worst
you’re likely to see in a cockroach setup, perhaps is a missing in flight
request that can be retried. And but, but probably more likely, the worst case
scenario is you’re gonna see slightly increased latencies until that region
comes back online and and consumers near that region can talk to him that
closest region. Scaling might also be required across continents, you can hide
if you can harness different database topology patterns, in order to get great
performance no matter where your users are. So for example, you can lock data to
specific regions, which I’ll cover in a minute, by using the regional by row
topology, that’s great for providing low Read and Write latencies to users near
that, that region, you can also use the global tables topology pattern, which
gives fast low latency writes or reads sorry, to everyone, regardless of where
they are. And we’ve got other products as well, I’m talking, I’m gonna talk
about serverless at the moment, because we run that on Kubernetes. We run
dedicated another offering on Kubernetes, as well, but I’ll talk about
serverless. And it’s a great one to try if you’re looking at dipping your toes
into the world of distributed SQL. So in this example, we see one entry point, a
separation of SQL and storage, the isolation between tenant pods, a warm pool of
unassigned pods that are ready to be associated to a tenant. And it’s also
running across multiple easy’s although in more you’re more likely You’d be
running it across multi region, that’s exactly what we tend to do. And let’s
just focus on tenant one for the time being. I see there’s a raised hand, pull.
Gianna, do you have a question?
Unknown Speaker 45:10
That was the one minute warning.
Speaker 9 45:14
Oh, my word. So what, what I’ll do, I’ll blaze through this. So if 10 one needs
more space and more capacity, we bring in a pod from the hot one. If they, if
one that need goes away, we remove or remove it, and it goes back into the hot
pod pool, if the overnight usage for Telemon completely drops off, so to the
pods, and it comes back online when they do. So we’ve got an example of a
database running a global application, this took about 10 minutes to create. And
I’m running in the UK with low latency. And with great British pounds GBP. I’m
hitting a Ireland database. And my application is running in London. If I then
hit, a German user would experience slightly higher latency, they’re still
talking to the Ireland database. And they see everything in euros, United States
high latency, but you’re not going to talk to a United States database from the
- So to simulate a United States user, I flick my VPN over and I’m now getting
really low latency. And it doesn’t matter what language you’re talking in the
localization table is completely global. So users, regardless of their language
can get a great experience from the app. And this is the entire database
definition for that whole app, I create a database, I enable super regions and
into those super regions I have. So I’m pinning data to the US and the EU, I
create a table called product. And whenever I see a market of Germany or France,
I asked for that data to go to the EU central Frankfurt note. And then the
internationalisation table is global. And I enable these with just one single
line of text. So it’s really easy to get a fast user experience and a global
database for for your users. And we can run on Kubernetes just as just as
easily. That’s it. Sorry, I probably went over a little bit there. Your one
minute morning spooked me, but I’m really glad I got it.
Speaker 1 46:59
Yeah, I’m sorry. I gave you a little extra there. Because I because the the one
minute warning, spooking. Thank you. All right. Well, yeah, thank you,
everybody, for for your presentations. These are all seem like useful tools. And
we actually do have a few minutes for questions that people did have questions.
I know, we heard a lot of information in a very short amount of time. But if
people did want to ask questions, we can make a little time for that. Otherwise,
I do have a couple more announcements I can make. I guess if anybody has
questions, please raise your hand. We will make this or this will be available
on YouTube. And then we’ll also do a follow up blog post as well. Well, if there
aren’t any questions, I’m going to share my screen and you can still ask
questions that they come up
Speaker 1 48:06
my, okay, this is basically a slide here. If you want to connect with any of
these, our speakers, you can scan that QR QR code, which will take you to a form
that just goes to us the DNA community. And we’ll pass your information along if
you do want to connect or ask questions to any of the speakers today. Since we
have just a little bit of time, I did want to just mention, also the new
website. So we have relaunched the website, we’ve tried to make it a resource
for the go to resource for the Okay, information. So it’s a new look and new
feel. We have a new resource library where you can access various resources. And
then also we’re we have an events portal where we’re going to list all the
events including events like these. And you can access all that information
here. We’re always also looking for new resources or use cases. So if you have
something like that you can always reach out to me on Slack or LinkedIn. And we
can get out to the site if it’s if we think it’s something that will be
beneficial to the community. See, okay, well, I think that’s it unless anybody
else has anything. Any last questions to make? Say? Ask. Okay, well, really
appreciate all of your time, both for our speakers and also for our attendees.
Hopefully this was useful to you all. And yeah, thanks a lot for attending. And
you know, as always, we Have these meetings once a month. And this is a little
bit of a different format. So first of our ecosystem days, but this is normally
the time for our town hall event. So yeah, thanks everybody for coming. And it’s
we have, you can have a little bit of your your, your day back.
Unknown Speaker 50:21
Thanks so much, Paul. All right. Thank you.
Unknown Speaker 50:27
Bye. Bye.
Unknown Speaker 50:28
Thank you. Bye
