►
Description
Join us for Kubernetes Forums Seoul, Sydney, Bengaluru and Delhi - learn more at kubecon.io
Don't miss KubeCon + CloudNativeCon 2020 events in Amsterdam March 30 - April 2, Shanghai July 28-30 and Boston November 17-20! Learn more at kubecon.io. The conference features presentations from developers and end users of Kubernetes, Prometheus, Envoy, and all of the other CNCF-hosted projects
Clusters as Cattle - Wrangling Clusters Not Just Nodes - Illya Chekrygin, Upbound
Kubernetes has fulfilled its promise at wrangling nodes and providing a self-healing, self-scaling, self-everything experience. Nodes can come and go and be automatically scaled using node scalers. But what if node scaling is not enough, and you need to scale clusters too? What if you want to separate types of workloads across different clusters for performance, isolation, geographic distribution, compliance or management reasons? Behold, the future is here! In this talk we’ll explore how Kubernetes can automatically deploy and manage other clusters dynamically and on as needed basis. We’ll show a demo of an application that is able to dynamically provision clusters across regions and schedule workloads across them. We’ll explore the benefits and challenges of this approach.
To Learn More: https://sched.co/GrWN
A
A
A
So
yeah,
so
there
is
no
secrets,
nothing
else
being
provisioned
so
far.
Well,
you
can
take
my
word
for
it.
I
also
have
a
couple
of
demo
helper
scripts,
which
will
help
me
to
list
daily
basis.
This
is
a
quick
hack.
I
am
running.
Multiple
CLI
commands
across
different
cloud
providers
to
see
that
I
have
databases,
running
and
I,
have
the
same
command
for
listing
in
clusters
and
I'll
be
using
them
throughout
presentation,
so
yeah
I
don't
have
any
cluster
for
now
provision
good.
A
Each
provider
contains
credentials
for
each
cloud
provider
and
two
resource
classes
will
see
in
presentation
how
we
leverage
in
the
resource
claim
as
abstract
resources
and
resource
classes
to
provision
dynamically
managed
services
with
that
said,
this
is
most
nerve-racking
part
is
to
kick-off
creation,
Cabrini's
clusters.
So
with
that
I
have
a
template
file
for
clusters.
This
is
kubernetes
cluster,
abstract
resource.
As
you
see
it
doesn't
have
any
specifications
in
it
all.
It
has
reference
to
a
resource
class.
We
look
in
the
source
classes
in
details
a
little
bit
later.
A
For
now
it
has
generic
word
cloud
and
what
I'll
do
is
to
replace
that
word,
not
delete
but
create
with
a
Tobias,
Asher
and
G
CPU,
and
so
hopefully
it's
all
created.
So
let's
I
can
at
least
make
various
clusters
in
control,
plane,
I,
see,
crossplane,
accepted
them
and
assigned
cluster
reference,
which
is
very
good
start
and
then
I
should
run
my
demo
script
I
mean
helper
script
to
list
the
clusters.
A
A
Over
the
last
decade,
we
witnessed
emergence
of
cloud
computing
as
a
predominant
IT
paradigm.
Cloud
computing
allows
organizations
to
focus
on
the
core
business
competencies
and
react
quickly
to
respond
to
changing
demands
without
expending
huge
resources
on
infrastructure
and
maintenance
organizations
can
instantly
take
advantage
of
set
of
world-class
services
across
multiple
infrastructures.
They
can
scale
the
businesses
globally
and
do
so
with
efficient
pay-per-use
model.
A
Yet,
despite
its
predominance
cloud,
computing
remains
completely
under
control
of
small
set
of
cloud
providers.
Amazon
Microsoft
Google
at
the
front
of
the
race
and
compete
aggressively
for
market
share
users
and
talent,
while
those
cloud
providers
themselves
heavy
adopters
of
open
source.
The
cloud
computing
remains
predominantly
proprietary
and
close
source.
Each
cloud
provider
offers
a
walled
garden
of
proprietary
services
that
are
designed
to
lock
in
the
customers
and
maintain
high
utilization
of
that
cloud
provider.
Infrastructure.
A
While
your
own
code
runs
in
VMs,
containers
functions
otherwise,
portable
more
frequent
enough,
it
depends
on
resources,
like
databases,
message,
queues,
analytics,
big
data,
machine
learning
and
others,
the
very
resources
that
cloud
providers
offers
managed
services
and
what
McMahan
service
is
so
attractive
that
you
need
to.
You
don't
need
to
worry
about
provisioning
scaling,
upgrading
patching
replicating,
in
other
words,
all
the
tasks
that
require
to
run
production,
great
systems.
A
A
So
as
an
alternative
to
many
services,
we
can
consider
not
to
use
them
at
all.
We
can
Decapolis
off
a
source
from
cloud
and
instead
stick
to
continuation
of
those
platform
elements.
We
need
a
number
of
community
appraise
available
today
which
allows
you
to
provision
stateful
workloads
like
my
sequel,
Kafka,
Redis,
spark
so
effectively.
You
can
pull
out
the
operator
installed
in
your
beliefs,
cluster
and
run
managed
service
alternative.
The
challenge,
in
my
opinion
and
I,
think
LC
also
echoed
it
earlier
that
installing
the
operator
in
Spanish
stateful
service
is
just
half
a
battle.
A
Cloud
providers
sometimes
offer
overlapping
set
of
services,
especially
if
those
are
adapted
from
open
source
domain.
However,
the
service
name
could
be
the
only
common
element.
Everything
from
provisioning
configuring
scaling
and
maintaining
it
could
be
entirely
different
depending
on
the
cloud
provider
clusters
accompanies
clusters
is
yet
another
resource
that
runs
on
cloud,
and
today
we
have
multiple
options.
One
comes
down
to
manage
kubernetes
clusters,
provision
them
easy.
However,
consistent
configuration
and
operation
is
hard,
just
like
with
any
other
managed
services.
A
A
With
those
goal
in
mind,
we
introduced
cross
plane.
The
open
source
project
that
was
started
at
a
bound
cross
plane
is
a
multi
cloud
control
plane.
A
single
cross
plane
enables
provisioning
a
full
provisioning
and
full
lifecycle
management
of
services
and
infrastructure
across
a
range
of
providers
and
regions.
A
A
Cross
plane
presents
a
declarative
management
style
API
that
will
cover
a
wide
range
of
portable
abstractions,
including
databases,
clusters
and
buckets
available
today
and
message
queues,
data
pipelines,
service
and
many
more
are
coming
is
based
on
declarative
resource,
a
model
of
kubernetes.
It
applies
the
same
concept
from
container
construction
to
multi-cloud
workloads
and
resource
orchestration.
A
Crossplane
presents
a
separation
concerns
between
application,
developer
and
cluster
or
cloud
administrator
when
developer
can
focused
on
the
applications
without
were
in
environment
policies
and
credentials,
and
administrators
can't
define
those
environments
and
shape
policies
and
apply
constraints.
The
separation
of
concerns
leads
to
high
degree
of
years
ability,
reusability
hardware
and
reduces
complexity.
A
Crossplane
extends
kubernetes
api
by
custom
resource
definitions
where
each
managed
resource
represented
by
dedicated,
CRD
and
processed
by
controller
with
resource
pacific
handling
functionality.
Today
crossplane
supports
three
cloud
providers:
AWS
and
DC
peanuts,
initial
support
and
who
plan
in
admin.
We
support
many
resources
like
databases,
my
sequel
and
Postgres
across
three
cloud
providers
who
support
buckets
and
manage
kubernetes
clusters.
A
A
Let's
take
a
look
at
the
cloud
provider
is
a
resource,
a
provider
comes
kind
of
with
two
pieces
information.
We
need
a
secret
reference
where
we
store
cloud
account
credentials,
so
we
can
communicate
with
the
crosspoint,
can
communicate
with
the
cloud
provider
and
has
arbitrarily
metadata,
which
could
be
specific
per
cloud
provider.
In
this
case,
this
is
AWS
provider
has
additional
information
metadata
for
region.
However,
those
additional
fields
could
very,
very
different
provider.
A
When
we
look
at
ministers
as
a
resource,
we
can
think
of,
as
as
I
mentioned
earlier,
it's
one
to
one
tire
assignment
between
customer
source
definition
and
managed
service,
and
it's
strong.
The
type
in
terms
of
each
spec
field
in
CRD
is
dedicated
for
this
concrete
resource.
This
is
example
of
my
sequel
server
from
asher.
As
you
can
see,
it
has
all
this
information
needed
to
operate,
stand
up
my
sequel,
server,
each
manager,
resource
CRD,
processed
by
controller,
which
responsible
for
provisioning,
generating
connections,
secrets
and
keep
track
status,
update,
continuous,
reconciling
and
check
another
resource.
A
A
resource
claim
is
a
request
for
source
by
user.
It
could
be
either
application
developer
or
could
be
also
administrator
user.
Just
like
it's
similar
to
paths
and
PVCs
in
communities.
Just
like
cloud
consumes.
Node
resources
and
PC
consumes
persistent
volumes.
Resource
claimed
consumed
can
create
resources,
just
like
part,
can
request
specific
levels
of
resource.
Like
give
me
this
node
with
CPU
such
CPU
and
such
memory,
so
does
the
resource
claim,
can
request
specific
resource
based.
A
Let's
say
it
was
specific
size
or
version,
and
while
resource
claim
allow
users
to
consume
abstract
resources,
it
is
common
that
they
actually
need
conclude
resources,
underneath
that
with
varying
properties
such
as
performance
like
for
different
problems
and
cluster
administrators,
are
responsible
to
provide
those
conclude
resources
either
stand
them
up
statically
or
by
dynamic
provisioning,
and
those
resources
need
to
defer
in
properties
other
than
performance
or
version
and
disk
size.
So
they
need
to
do
so
also
without
exposing
old
implementation
detail
of
each
conclude
resource
to
application
developer.
A
A
Just
an
example
of
my
sequel
instance,
as
abstract
resource
know
that
it's
very
similar
to
kubernetes
cluster
resource
claim
it's
similar
in
terms
of
it's
very
minimalistic.
In
the
definition
all
it
has
the
reference
to
a
class
and
the
request
to
give
to
provision
my
simple
with
the
database
engine
5.7.
A
Resource
class
provides
a
way
for
administrator
this
time
to
describe
classes
of
resources
they
offer
in
the
crossplane
different
classes,
map
to
different
resources
or
sometimes
map
to
the
same
resource,
but
with
the
different
quality
of
service
levels
or
arbitrarily
different
policies.
Underneath
that
cross
plane
itself
is
an
opinionated
what
resource
class
represent
and
each
resource
class
has
fallen
elements
as
a
reference
to
provisioner
the
component,
which
responsible
to
provision
conclude
resource.
It
has
a
provider
reference
indication
jefita.
Is
this
conclude
resource
to
given
cloud
provider?
A
This
is
example
right
here,
we're
looking
at
a
resource
class
for
what
is
it?
Yes,
as
your
a
key
s,
kubernetes
cluster
with
additional
properties
needed
to
stand
it
up,
and
we
can
also
imagine
that
resource
classes
can
be
further
customized
to
bootstrap
the
company's
cluster,
not
in
terms
of
simply
bringing
it
up,
but
also
applying
policies,
secrets
and
service
accounts.
So
it's
readily
can
be
consumed
by
applications.
A
A
This
brings
us
to
workload.
Crossplane
offers
workload
as
a
way
to
provide
application
developers
with
the
ability
to
write
a
portable
manifest
for
the
applications.
Web
application
developer
does
not
need
to
be
concerned
about
cloud
provider,
environment
or
policies.
Instead,
application
developers
focus
on
creating
abstract
resources
and
referencing
those
resources
in
the
workload
crossplane
workload
scheduler
will
find
matching
kubernetes
cluster.
What
to
propagate
this
workload,
and
it
does
so
either
by
selection
criteria
or
matching,
or
sometimes
even
by
name
assignment.
It's
very
similar
inspired.
A
How
we
find
pod
can
find
node,
schedulable
node
and
if
workload
scheduler
cannot
find
cluster,
it
actually
can
go
ahead
and
dynamically
create
one.
Crossplane
workload
specifies
kubernetes
deployment
and
service
as
a
payload.
It's
today
we
only
support
those
two
for
now.
So
that's
a
payload
which
will
be
pushed
to
target
clusters
and
will
also
specify
the
target
namespace
a
way
to
deploy
the
source
in
target
cluster.
A
All
right
so,
hopefully
our
kubernetes
clusters
are
up
and
running.
It
crossed
three
cloud
providers.
We
need
only
four,
yes,
ets
is
up
and
running
and
we
have
asier
and
a
key
s,
ng
key
cool.
So
let
me
I
have
another
demo
helper
script,
which
will
generate
my
cue,
config
context
file,
so
just
to
show
you
what
I
have
currently
have
I
have
only
single
context
for
micro
quieras.
A
That's
gonna
be
really
important:
that's
effective
what
I'm
doing
I'm
using
the
CLI
or
specific
club
specific
way
to
retrieve
the
configs
for
kubernetes
cluster
and
the
reason
why
I'm
doing
so
in
this
side,
I
have
three
panels:
each
represents
individual
kubernetes
cluster.
So
if
I
see
in
Ellis,
I
set
it
up
for
tab
being
a
KS
middle
eks
and
varmint
jiggy.
So
if
I
do
then.
A
A
A
Here
we
go
so
you
notice
that
six
items
have
been
created,
one
for
workload
and
one
for
my
sequel
instance
for
each
cloud
provider
and
now,
if
I
switch
back
to
target
clusters,
I
should
see
demo
namespace
being
created.
Americlean
we
should
see
is
my
sequel
connection.
Secret
will
be
propagated
here
and
as
well.
Workload
for
now
I
think
we're
still
waiting
for
creation
of
mythical
databases.
A
A
As
I
mentioned,
that
workload
gives
a
developer
stability.
Application
develops
application
developers,
ability
to
write
portable
application.
In
this
case,
I
use
the
same
Yama
file
to
provision
my
application
across
three
different
cloud
providers
in
three
different
companies
clusters
and
stand-up.
My
sequel
manage
my
sequel
database
in
each
cloud
provider
and
I
do
so
because
I'm
reference
used
in
abstract
reference
here
for
my
sequel
instance,
where
I'm
saying
that
I'm
pointing
to
specific
resource
class
and
telling
it
give
me
the
version
engine
version
5.7,
that's
all
I
need
to
know
as
application
developer
in
workload.
A
I
have
a
section
for
resources
which
say
for
me
to
successfully
deploy
and
operate.
I
need
following
list
of
resources.
In
this
case,
I
might
even
identifying
the
my
sequel
instance
I'm,
identifying
also
target
class
in
this
case.
I
use
a
name
assignment
in
it.
Propagate
just
Pacific
provider
target
cluster
and
I
have
a
payload
where
you
have
target
deployment,
which
is
standard
deployment
for
WordPress
application.
Its
standard
company's
deployment
with
the
secret,
providing
connection
information,
web
environment,
values
and
I
also
have
a
service
and
the
target
namespace.
A
So
now,
if
I
keep
looking
at
my
my
sequel
database
instances,
they
still
pending
and
I-
think
this
is
a
good
time
segue
to
describe
separation
of
concerns
a
little
bit
better
example.
So,
right
here,
I'm
acting
as
an
application,
developer
and
I'm
operating
in
the
default
namespace
in
default,
namespace,
let's
say
I
get
secrets
in
default:
namespace
I,
don't
have
any
information.
One
relates
to
compute
resources,
I,
don't
know
cloud
provider,
credentials,
I,
didn't
know
the
kubernetes
connection,
secrets,
I,
don't
know
even
instance,
connection
secrets
for
my
sequel.
A
So
far,
however,
if
I
compare
this
with
cross
play
in
the
system
namespace,
which
administrator
privileged
M
space,
you
can
see.
We
have
a
bunch
of
information
here.
So
we
have
a
cloud
arranged
providers.
We
have
cooperate
clusters
secrets
and
we
have
now
my
sequel
database
instance.
Secrets
are
coming
up
as
we
stand
in
the
map,
so
I
can
also
switch
to
sorry.
A
But
so,
as
you
can
see
now
already
propagating
my
sequel
connection
secret
to
Azure,
cluster
and
Jackie,
and
if
I
go
back
to
my
cross,
plane
and
I
list
again
my
my
sequel
instances,
you
can
see
that
I
already
have
Azure
in
GCP
instances
up
and
running
and
bound
and
now
I'm
just
waiting
for
the
workload.
So,
let's
take
a
look
at
workload,
we
still
don't
have
fully
provisioned
workload.
However,
if
I
go
back
to
minis
clusters,
here,
I
can
now
list
deployments.
A
And
I
should
see
yes,
I
already
have
a
deployment
in
two
of
the
classes.
It
guess
slide
the
leg
in
behind
I.
Guess
it
because
took
long
a
little
bit
to
stand
up
for
kubernetes
clusters,
the
Cambria
cluster,
but
we
should
see
it
shortly.
So
meanwhile,
I'll
keep
refreshing
the
workload
status.
Oh
cool,
so
I
already
have
an
external
IP
assigned
where's
my
mouse
yeah.
It
is
so
what
I
will
do?
I
just
grab
it
and
we're
gonna
kind
of
do
interactive
here
copy
in
my
browser-
and
hopefully
I
have
one
for
geeky
as
well.
A
Okay,
it's
coming
up.
Well,
it's
also
coming
up.
Let
me
show
you
quickly
how
the
resource
class
is
so
defined.
So
I
can
say
it
keeps
it
here.
I'll
get
I'm
just
wasting
time
here,
filming
you
yet
resource
classes
from
control,
crossplane
system
does
the
resource
classes
we
define
as
a
part
of
a
provision.
So
let's
take
a
look
at
the
specific
resource
class
in
more
detail.
Let's
pick
up
would
say
AWS
my
sequel
and
let's
look
at
the
mo
file.
A
So,
as
you
can
see,
it
has
a
little
bit
more
details
that
resource
claim
it
has
specific
information.
It
has
a
security
group,
for
example,
it's
getting
even
more
interesting
if
we
look
at
the
cluster
has
more
implementation,
details
like
security
group
worker
notes,
configurations,
EPCs
and
so
forth.
Those
are
exactly
implementation
details
which
cluster
administrator
wants
to
hide
from
application
developer.
Application
development
does
not
need
to
know
about
to
provision
workload,
so
hopefully
I
spend
enough
time
talking
about
that.
Let's
take
a
look
here:
okay,
we
have
them
three
and
that's
workload.
A
Here
we
go
so
now
we
have
source
external
hosting
I'm
gonna
grab
that
it
was.
Hopefully,
nobody
fast
enough
to
hack
my
WordPress
applications,
so,
okay,
I'm
gonna,
go
ahead
and
stop
provision
it
so
just
make
sure
that
we
are
and
I'm
looking
that
monitor,
which
is
super
small
I'm
gonna
use
this
password,
I
guess.
A
A
A
Tada,
so
just
to
kind
of
sum
up
what
we
do
in
here.
We
provisioned
three
kubernetes
clusters,
managed
companies,
clusters
and
three
different
cloud
providers,
those
as
they
offered
as
we
would
do
normally
in
a
KS
e
KS
ng
key.
Then
we
went
ahead
and
created
three
manage
databases.
In
this
case,
my
sequel.
We
create
RDS
in
a
Tobias
who
created
my
simple
server
in
Azure
and
created
cloud
SQL
in
DCP.
A
All
right
so
to
close
up
I
guess
what
did
I
have
here?
Oh
yes,
already
talked
about
separation
concerns
vision,
vision
is
important.
I
had
to
write
it
down
on
a
separate
piece
of
paper,
so
a
bond
mission
is
to
create
more
open
computing
platform,
which
offers
more
choices
and
less
lock-in
at
the
heart
of
every
control
plane
at
the
heart
of
every
cloud
provider.
This
control
plane
and
hope
that
cross
plane
can
become
a
control
plane
for
open
cloud
with
open
control
plane.
A
Anyone
can
add
new
api's
and
new
controllers
and
extend
it
to
manage
new,
manage
resources,
whether
it's
commercial,
open
source
well
at
the
very
beginning
of
our
exciting
journey.
We
just
open
sourced
cross
plane
last
week
and
community
response
was
overwhelming.
Please
check
us
out
if
you
haven't
done
so,
give
it
a
shot.
We
have
a
good
examples,
we'll
work
load,
I
just
demoed,
so
you
can
run
yourself
and
give
us
your
feedback.
Tell
us
what
you
think.
If
something
didn't
work
out,
please
create
an
issue
or
talk
to
me
directly.