►
Description
While Kubernetes has become a de facto standard for running the Cloud Native workloads, the platform on which Kubernetes runs remains pretty diverse. Kubernetes multi-cloud story is about providing the platform and infrastructure engineers of an organization the freedom to create and manage production grade conformant Kubernetes clusters on infrastructure of their choice.
There are several projects that have come up to solve the challenges around managing the Kubernetes Lifecycle Management, with Cluster API becoming a standard way to tackle the problem.
This workshop will start with an introduction of the challenges around Kubernetes Lifecycle management from Day-0 to Day-n and deep dive into the Cluster API projects that tackle them in a declarative way. As an example, we will demonstrate how to use CAPI providers to manage Kubernetes Clusters on AWS and bare metal.
A
This
is
cap
ecosystem.
If
you
can
see
on
the
diagram
how
it
works
is
you
know
you
declare
everything
in
an
yaml
file,
and
that
is
where
we
are
calling
it
as
in
declarative
config.
You
declare
what
keyword
in
this
version.
You
want
how
many
worker
nodes
you
want
and
how
you
want
your
current
test
cluster
to
look
like.
A
And
you
know
you,
you
create
a
you,
create
a
cluster.
Let's
say
we
call
that
cluster
as
a
bootstrap
cluster,
you,
that
is
a
simple
plain
kind:
cluster
you
create
a
bootstrap
cluster
and
you
install
install
capi
and
provider
components
on
it.
Those
components
are
nothing
but
simple
parts
that
are
running
on
your
kubernetes
cluster
and
you
apply
this
declarative
yaml
and
according
to
the
configuration
it
is
going
to
create
your
target
kubernetes
cluster.
You
know
it
could
be
google
cloud
or
aws
or
as
or
any
sort
of
cloud
like
that.
A
A
couple
of
keywords
I
want
to
highlight
here
like
bootstrap
cluster
management,
cluster
and
workload
cluster.
So
when
you
go
through
this
cap
ecosystem,
these
are
the
keywords
that
you're
bound
to
see
so
bootstrap
cluster
is
a
cluster
where
we
install
capi
and
other
tooling,
and
then
you
you
could
utilize
the
the
capi
apis
to
provision
a
kubernetes
cluster
on
the
target
cloud,
and
this
bootstrap
cluster
is
pretty
straightforward
and
simple
and
we
generally
use
kind
cluster.
A
This
is
done
to
actually
onboard
users
very
quickly
and
you
know
without
any
difficulty,
but
then
this
bootstrap
cluster,
you
know,
is
not
like
into
or
you
know,
production
grade,
so
it
is
not
recommended
for
production
use
cases.
So
what
typically
we
do
is
like.
We
create
a
kubernetes
cluster
on
a
choice
of
our
cloud
provider
by
using
bootstrap
cluster.
It
could
be
azure
or
google
cloud,
and
then
you
know
that
cluster
that
we
created
on
a
choice
of
infrastructure.
A
We
marked
that
as
a
management
cluster
and
from
that
management
cluster,
we
do
the
same
thing
that
we
didn't
did
on
bootstrap
cluster.
We
inst
we
have
all
the
cappy
and
other
cappy
tooling,
and
you
know
we
can
start
creating
clusters
from
that
management
cluster
to
different
cloud
providers,
as
you
can
see
on
the
screen.
So
the
the
difference
between
bootstrap
and
management
cluster
is
that
you
know
management
cluster
is
an
actual
cluster,
it's
in
production
grade
cluster
and
we
just
use
capi
itself
to
create
the
management
cluster.
A
As
the
name
suggests,
management
cluster
is
to
manage
your
workloads,
a
workload
cluster
across
you
know
cloud
providers.
Now
I
mentioned
about
workload
cluster,
so
workload.
Cluster
are
the
clusters
that
you
create
using
management
cluster,
and
this
is
the
cluster
where
you
want
to
deploy
your
actual
workloads
or
applications.
A
For
example,
you
want
to
deploy
some
app
or
some
gpu
workloads.
According
to
your
name,
you
could
you
know,
deploy
that
on
your
workload
cluster.
So
this
is
it
about
bootstrap
management
and
workload
cluster.
I
want
to
pause
for
30
seconds
and
see
you
know
if
anybody
has
any
question
or
want
to
ask
anything
cool.
B
Yeah,
thank
you
for
setting
that
context,
so
clustered
provider
for
bring
your
own
host,
so
so
I
saw
in
the
poll
that
most
of
you
are
like.
You
have
been
using
kubernetes
for
quite
some
time.
So
I'm
sure
you
must
have
heard
about
cluster
api
by
now
and
if
not
ashitas
gave
like
a
very
good
overview.
So
there
are
like
different
infrastructure
providers
that
you
know
do
their
own
life
cycle
management,
as
provided
by
capi
like
we
have
vsphere.
B
We
have
cluster
api
provider
for
aws,
azure
digital
ocean
gcp,
there's
a
very
long
list.
So
now
you
might
ask
plus
api
provider
for
bring
your
own
host.
Is
it
just
yet
another
provider?
Well,
it
is.
The
answer
is
yes
and
no?
Yes,
because
it
is
an
infrastructure
provider
wherein
you
can
deploy
your
kubernetes
workloads
and
no
because
there's
a
slight
difference
between
byoh
and
rest
of
the
providers,
and
that
is
what
we'll
be
concentrating
today.
B
Can
you
move
please
before
that,
like
there
are
like
few
terms
that
are
specific
to
byoh?
So
what
is
a
host?
A
host
is
a
running
computer
system.
It
could
be
physical
or
virtual.
The
expectation
is
that
it
has
some
kernel
and
some
base
operating
system
installed
on
it
and
a
b
bio
host.
A
bio
host
is
something
that
is
managed
outside
of
cluster
api.
B
So,
like
ashitos
mentioned,
cluster
api
takes
the
responsibility
of
the
life
cycle,
management
of
right
from
your
machine
to
the
kubernetes
node
layer,
but
in
biohost
this
host
is
something
that
is
managed
outside
of
cathy,
and
next
we
have
the
bih
capacity
pool.
Since
it
is
bring
your
own
host,
you
are
allowed
to
create
a
capacity
pool
of
sorts,
as
in
you
can
add
any
number
of
hosts
to
that
pool,
and
you
can
take
out
any
number
of
hosts
from
that
pool,
and
then
we
have
kubernetes
node.
B
So
this
is
the
node
that
runs
on
top
of
a
host,
that
is
to
say,
when
you
run
things
like
cubadium,
that
is
where
it
will
bootstrap
your
host
as
a
kubernetes
node
and
this
node
provisioning
and
the
provisioning.
That
is
the
life
cycle.
Management
of
kubernetes
node
is
done
by
the
brioh
provider,
and
then
we
have
the
host
components.
So
kate's
host
components
are
nothing
but
things
like
cube.
Adm
cubelet
or
container
d,
everything
that
is
needed
to
run
kubernetes
on
your
host.
B
By
that
I
mean
when
you
say
you
want
to
create
a
say,
a
three
node
cluster,
then
the
infrastructure
provider
looks
at
okay.
I
need
a
three
node
cluster.
It
means
I
need
to
create
three
vms.
I
need
to
scale
down
cluster
by
two
nodes.
Then
I
need
to
delete
two
vms,
so
the
entire
lifecycle
management
of
this
stack
is
done
by
the
infra
provider,
but
in
case
of
bring
your
own
host,
as
you
can
see,
the
bottom
two
layers
are
user,
managed
that
is,
you
bring
your
own
host.
B
So
if
there's
like
one
takeaway
that
you
all
want
to
have
as
to
how
big
is
different,
it
is
this
concept.
Okay,
next
we'll
look
at
the
high
level
architecture.
We
will
refer
to
this
diagram
back
once
we
have
once
we
get
into
the
workshop
mode,
so
it
will
be
easier
to
relate
better,
but
just
to
give
an
overview.
We
have
a
management
cluster
that
is
in
the
top
half
of
this,
and
then
we
call
it
edge
location,
but
it
could
be
anywhere.
B
It
could
be
in
your
dc
as
well,
wherein
you
have
some
spare
servers
as
capacity
so
in
the
management
cluster
we
will
install
the
infra
provider
and
then
on
every
host.
If
you
see
there
is
a
green
box
called
host
agent,
so
this
is
again
something
that
is
very
special
to
the
byuh
provider.
So
this
agent
is
nothing
but
a
simple
golang
binary
that
you
have
to
start
as
a
process
on
every
host.
B
So
please
register
me
into
your
capacity
pool
and
whenever
there's
a
request
to
create
a
cluster
or
scale
up
a
cluster,
you
can
use
me
to
you
know:
provision
nodes.
So
that's
how
it
registers
itself
and
whenever
there's
a
request,
the
bih
info
providers
looks
into
this
capacity
pool
whether
are
there
any
hosts
in
the
capacity
pool.
If
there
are,
it
goes
ahead
and
provisions
the
kubernetes
node.
So
this
is
the
high
level
architecture
of
pyh.
B
I
know
it
could
be
a
lot
to
take
in
at
once,
but
probably
we'll
get
to
the
depth
of
it
when
we
go
to
the
workshop.
So
if
there
are
any
questions
I'll
take
a
pause,
it's
cappy,
ga
for
production,
yes,
cappy
is
definitely
gfo
production.
I
think
ashitosh
can
give
the
latest
versions
of
cappy
information.
A
C
Can
we,
you
know
people
who
are
running
in
the
production
right
like
if
you
have?
What
is
the
biggest
challenge
that
you
face
that
you
you
know?
Maybe
it
would
be
worth
like
posting
that
on
the
slack
and
we
can
see
if,
like
cluster
api
as
part
of
the
life
cycle,
management
actually
addresses
some
of
those
pencils.
B
A
B
I'm
going
to
answer
for
that
or
I
don't
because
I
can't
think
of
limitation
as
such,
so
your
we
use
cubadium
to
bootstrap
our
nodes.
So
cubedium
has
some
bare
minimal
prerequisites
that
it
should
be
two
cpu
cores,
8gb
memory,
so
the
same
applies
to
byoh
as
well
and
limitations.
I
wouldn't
say
limitations,
but
currently
b
is
a
very
new
provider
by
the
way
it's
just
six
months
old.
So
we
support
only
ubuntu
2004
and
I
think
only
122
and
123
versions
of
kubernetes
are
not
before
that.
A
D
Yeah
thanks
ashley
stan,
shannon
so
yeah
pc,
so
the
tc
stands
for
tenzoo
community
edition,
so
it's
open
source
kubernetes
that
form
provided
by
vmware,
so
it
launched
in
october
last
year,
and
currently
it
have
more
than
one
case-
starts
on
its
repo
and
2700
plus
downloads
on
different
platforms
like
mac
linux
windows.
So
if
I
talk
about
what
pc
specifically,
is
that
it's
a
cli
which
provides
lots
of
features,
plugins
packages
and
also
easy
to
install
so,
and
you
can
also
install
it
on
it.
D
D
Can
we
please
move
to
next
slide?
Yeah
thanks.
So
this
is
a
complete,
wholesome
architecture
of
dc,
so
everything
which
you
are
seeing
in
this
architecture
that
all
those
things
they
see
include
in
it.
So
there
is
a
cloud
provider,
the
runtimes
all
the
above
applications
are
present
in
tc
in
the
form
of
packages
you
can
install
them,
avail
them
use
them
while
deploying
any
application
or
by
using
pc.
So
a
thought
is
coming
to
your
mind,
okay,
so
this
is
tc
so
how
to
use
it
where
to
use
it.
D
So
for
that
we
have
some
transducer
clusters,
so
the
one
like
management,
cluster
and
workload
cluster.
What
ashutosh
had
discussed
earlier
so
similar
thing
we
have
for
pc
also,
so
you
can
create
a
management
and
a
workload
cluster
and
install
dc.
There
use
all
the
features
plugins
packages,
what
tc
provide.
So
I
am
going
to
discuss
unmanaged
cluster
in
the
next
slide.
D
Yeah
thanks
so
what's
unmanaged
cluster
and
how
it
is
different
from
our
management
and
the
workload
cluster.
So
unmanned
cluster
is
a
single
node
tanzu
cluster.
If
you
are
a
beginner,
you
are
a
learner
or
you
are
exploring
a
kubernetes.
So
this
is
the
best
thing
to
start
with
why
I
said
that
thing
that
it
consumes
or
it
use
both
the
cube,
ctl
cli,
as
well
as
a
tensor
cli
in
it.
So
you
can
use
your
net
sledding
inside
this
cluster
also
and
can
also
explore
the
this
tc.
D
So
there
is
an
amazing
getting
started
guide
which
we
have
and
if
you
once
started
that
reading
that
guide,
there's
a
no
go
back
thing
so
yeah,
that's
that
and
so
yeah
the
unmanned
cluster
as
it's
a
single
node
cluster.
So
there
is
no
like
life
cycle
management
of
a
cluster
thing
there,
that
is
for
managed
cluster
and
workload
cluster
it.
It
is
there
for
them.
So
there
is
a
hands-on
which
we
are
going
to
do
in
a
while.
So
my
requirement
request
is
that?
D
Can
you
please
just
pull
out
this
image?
Just
do
docker
pool
and
this
image
so
that
the
installment
of
like
unmanned
cluster
will
be
faster
and
we
can
give
time
to
the
exploring
the
application
installing
the
application
on
an
unmanaged
cluster,
so
that
exciting
thing
can
take
time
instead
of
just
this
pulling
image.
D
A
Thanks
that
was
very
good
high
level
overview.
Thank
you
everyone.
So
if
you
have
any
last
minute,
questions
feel
free
to
ask,
but
this
is
enough:
slides
as
per
anusha
and
usha.
Don't
want
to
have
too
much
of
slides
and
thanks
for
bringing
up
this
beautiful
content
here
and
we
can
jump
into
workshop
right
away.
Kiran.
Do
you
have
anything
to
say
here.
C
Also
like
from
the
audience
perspective
right,
can
you
just
post
in
the
chat
really
quickly
before
we
get
started
into
the
workshop?
What's
the
type
of
provider
that
you
are
using,
you
know
we
just
heard
from
fibo
about
gcp.
Similarly,
like
you
know,
others
like,
can
you
just
post?
What
are
the
providers
that
you
are
looking
for,
their
micro
or
any
other
cloth
providers
I'll.
A
A
Yeah,
you
could
do
that
so
that
you
know.
If
I
get
that
right,
you
want
to
deploy
like
clusters
across
az
right.
You
could
do
that.
It's
just
that
the
diagram
says
you
know
depicts
okay
on
the
cloud
provider
high
level
story
that
you
could
potentially
deploy
in
aws.
You
could
have
your
own.
You
know
strategy
and
you
know
topology
how
you
want
to
do
that
and
it
is
quite
possible
and
it
does
vary
depending
on
like
azure
cl
as
or
aws,
depending
on
different
cloud
provider.
E
No
so
so
my
use
case
is
that
so
right
now
we
have
multiple
regions,
and
then
we
have
a
single
kubernetes
cluster
on
in
each
region,
with
the
separate
control
plane
and
the
work,
the
workers
and
everything.
So
my
question
is:
if
at
all,
we
can
use
cappy
for
something
like
that.
Would
each
region
be
a
separate,
entirely
separate
cluster,
which
you
manage
through
capi
or
the
control
plane
would
be
the
I
mean
sorry,
the
management
cluster
manages
just
the
workers
on
each
region.
A
Management
cluster
manages
you
could
potentially
do
both
ways.
You
could
have
a
you
could
manage.
You
know
your
workload
clusters
distributed.
You
know,
let's
say
you
have
aws,
you
could
have
thousands
of
kubernetes
cluster
each
into
different
reasons,
but
then
you
can
still
have
a
single
workload
cluster.
You
know
with
this
topology
spread.
You
could
do
that
easy
thing
in
the
in
a
particular
workload.
Cluster
itself.
F
A
C
A
Eks
or
aks,
where
you
don't
have
too
much
access
or
visibility
of
control
planes,
so
management
cluster
is
a
you
know
altogether
different
kubernetes
cluster,
just
that
its
job
is
to
manage
other
workload
clusters,
so
we
term
it
as
an
management
cluster.
If
you,
if
you
look
at
the
cluster
in
terms
of
you,
know,
components
or
you
know,
kubernetes
component,
it
all
are
just
same.
You
know
so
management
cluster
itself
has
control
plane,
as
well
as
worker
nodes.
A
A
A
Okay,
so
I'll
go
to
this
is
the
repo
that
I
have
cloned
and
you
know
in
a
different
screen.
I've
opened
that
repo
to
copy
the
instructions
so
that
I
don't
have
to
swipe.
My
screens
already
have
this.
You
know
prequest
met,
but
just
for
the
sake
of
completeness
I'll,
just
do
and
first
part,
so
I
have
like
azure
cli
done
and
if
you
see
that
in
the
in
the
instructions
doc,
if
you
are,
if
you
are
using
a
azure
cloud
account
for
the
first
time
you
want
to
do
couple
of
registrations.
A
A
G
A
Now
I'm
going
to
create
a
contributor
our
back
and
that
I'm
not
going
to
kind
of
display
on
the
screen,
because
I'll
get
my
password
for
that.
So
just
makes
just
if
you
can
follow
the
instruction
that
if
I
run
this
command
azp
create
for
our
back.
That
is
going
to
give
me.
You
know
an
output
and
I'll
just
copy
and
paste
that
into
the
env,
where
files
that
I've
created.
So
I'm
going
to
open
a
different
terminal
that
you
all
can't
see
and
do
that.
A
H
A
A
So
that
is
done,
and
you
know
now
like
I
have
all
the
credentials
set
up.
So
this
is
an
environment
variable
where
you
have
like
as
a
secret
and
app
ide
registered
and
then
set
up
and
the
next
is
like
you
know.
If
you
see
on
this
instructions,
you
know
I've
just
done
this
step
and
now
I'm
going
to
create
a
secret.
So
this
is
the
secret
that
will
be
used
for
authentication.
A
The
secret
is
created,
and
now
I
will
do
so.
This
step
is
a
cluster
ctl
edit.
So
what
it
does
it
is
going
to
install
cluster
api
component
and
I'm
using
cluster
api
provider
azure.
So
azure
is
the
component
that
deals
with
create
resource
creation
on
azure
infrastructure,
so
this
is
going
to
install
capi
plus
the
azure
components
on
my
client
cluster,
and
this
is
how,
when
we
do
this,
we
make
this
kind.
Cluster,
bootstrap
cluster.
A
So
if
you,
if
you
can
see
you
know,
these
are
the
parts
that
I
was
talking
about.
So
this
is
like
the
cap
g
controller
manager,
so
it
deals
with
infrastructure
on
azure
cloud.
This
is
the
core
cappy
this.
This
is
like
the
central
core
that
plug-ins
to
different
providers.
Right
now
it
is
plug-in
with
cap
g
in
a
single
q,
a
bootstrap
cluster.
You
can
have
different
plugins
right.
A
You
can
also
have
capzi
existing
with
aws
providers,
so
you
can
have
capsi
as
well
as
cap
a
but
for
the
sake
of
simplicity
and
this
works
out.
We
only
have
the
azure
provider
okay
now
so
now.
What
I'm
trying
to
do
is
next
step.
Is
you
know
I'll,
create
an
cluster
configuration,
and
for
that
let
me
show
you
what
it
is.
A
A
This
is
done
and
in
the
next
step
you
know
I'm
going
to
generate
this,
so
when
I,
when
I
run
this
command,
so
this
command
actually
utilizes
those
env
variables
to
generate
a
cluster
yaml
right.
So,
for
example,
we
mentioned
kubernetes
version
and
all
of
this
stuff
like
how
many
control
plane
and
worker
node
we
want
to
have.
A
It
is
done
just
to
take
a
quick
peek
of
what
it
is,
so
it
is
just
a
couple
of
yamls
right,
so
you
can
see
this
part
there's
a
clustering
ml
and
then
you
know
other
yamls
on
infrastructure
without
getting
into
too
much
into
this
thing.
Let
me
just
apply
this
ammo.
So,
let's
see
what's
the
next
step
next
step
is
to
apply
this
one,
so
I'm
going
to
apply.
A
This
has
been
created,
and
now
I've
also
mentioned,
like
couple
of
commands,
that
you
can
use
to
analyze
the
state
of
creation
where
we
are
in.
We
could
do
cube,
ctl
get
clusters
and
we
can
see
that
kcd
demo
is
in
provisioning
page.
We
can
also,
you
know,
execute
cluster
ctl
describe
cluster
question
demo,
okay,
so
this
failed
demo
gods-
and
this
is
because
of
you
know,
reasons
sometime.
You
know
because
this
stacia
I've
used
that
particular
vc
view.
A
A
H
A
H
B
Okay,
so
I
specifically
wanted
to
show
this
yaml
file
and
walk
through
this,
because
this
is
going
to
be
applicable
to
any
infrastructure
provider
or
any
provider
that
you
choose.
So
we
see
here
this
is
kind
cluster.
Now
this
custom
resource
is
specific
to
cluster
api.
Similarly,
cluster
api
has
other
constructs
like
cluster
machine
deployment,
cubadium
control
plane.
B
So
every
cluster
is
a
generic
class
repair
object
and
if
you
see
there's
a
infrastructure
reference
here,
this
is
where
you
say
what
kind
of
infra
provider
you
want
either.
If
you
want
azure
cluster,
you
give
azure
cluster
reference.
If
you
want
vsphere,
you
give
vsphere
cluster.
Similarly,
in
b
by
h,
we'll
see
we'll
be
adding
a
b
by
oh
cluster
here
and
then
you
have
to
write
the
specification
of
what
each
kind
is
here.
You
write
everything
about
azure
cluster.
These
are
the
required
fields
that
it
needs
to
bring
up
an
azure
cluster.
B
Before
we
go
more
into
this
file,
there's
a
control,
plane
ref
also,
so
you
can
say
which
control
plane
provider
you
want
to
use
to
bring
up
your
control
planes,
and
that
is
where
we
are
using
kubernetes
control
plane
and
if
we
scroll
down
further,
we
have
definition
for
cubidium
control.
Plane
like
these
are
the
bunch
of
things
that
it
needs.
B
So
this
is
the
init
configuration
and
joint
configuration.
Since
this
is
a
cubanium
control
plane,
it
will
use
cubadium
under
the
hood
to
bootstrap
your
clusters.
Init
configuration
is
something
the
very
first
control
plane.
Node
in
your
cluster
is
going
to
use
and
thereafter
from
the
second
control
plane
to
any
other
worker
node,
they
are
going
to
use
the
join
configuration.
B
So
if
there
are
like
any
specific
things
like
you
want
to,
I
don't
know
give
special
parameters
for
your
cubelet.
You
give
it
in
terms
of
cubelet
extra
arcs,
any
other
host
specifications.
Okay,
you
could
give
a
bunch
of
those
as
well
and
then
we
have
the
machine
template.
So
each
infra
provider
has
something
called
a
machine
template.
B
You
have
azure
machine
template,
vsphere
machine
templates,
wherein
you
say
that
every
vm,
every
azure
vm
or
a
vsphere
vm
that
is
spawned
what
should
be
the
specifications
of
that
and
then
again
machine
deployment
is
a
generic
construct
yep.
I
think
we
are
at
the
end
of
that,
so
this
yaml
file
will
have
a
mix
of
cluster
api
constructs,
as
well
as
specific
to
your
infrastructure
as
well.
G
B
Yeah,
so
I
think
if
asitos
can
do
cube,
ctl
get
pods.
You
can
see
a
bunch
of
controllers
that
are
running
control.
You
can
see
manager,
kind
of
things
like
capi
controller
manager
and
cadzi
controller
manager.
So
this
cabsie
controller
manager
has
a
bunch
of
controllers
that
are
registered
to
it.
They
reconcile
so
for
every
crd
that
is
specific
to
capsi
has
a
controller
that
is
running
a
reconciler
watching
these
objects
for
create,
update,
delete
whatever,
and
that
is
responsible
for
reconciliation.
F
G
A
G
H
A
So
management
cluster
is
all
together
a
different
cluster,
so
you
could
actually
choose
your
own.
You
know
way
of
deployment
and
workload.
Clusters
are
like
you
know.
Technically,
you
can
create
on
the
same.
You
know
reason
or
you
know,
but
it's
just
the
deployment
strategy
that
you
have
to
choose
so
that
you
know
your
fault,
tolerant
and
highly
available
stuff.
Like
that.
H
A
Wow
so
this
time
it
works.
So
if
you
see
the
commands
that
I'm
you
know
executing
is
cluster
ctl
describe,
this
is
a
very
useful
command
when
you,
when
you
are
like
operating
or
trying
to
do
operations
using,
you
know,
cluster
ctl.
It
gives
you
the
status
of
you
know
where
we
are
at.
For
example,
we
are
waiting
for
the
control
plane
machine
to
come
up,
and
you
know
stuff
like
that.
A
If
you
can
see
on
the
screen-
and
there
is
one
more
command
that
I
usually
prefer
to
do-
is
cube
ctl
get
azure
machines,
so
it
actually
gives
you.
The
actual
machines.
You
know,
gives
you
the
status
of
actual
machine
that
is
being
created
on
the
infrastructure,
so
it
is
like
in
waiting
for
cluster
infrastructure.
If
nothing
goes
wrong,
this
should
get
created,
and
you
know
also
one
good
thing.
While
we
do
is
that
while
we
wait
for
this
is
like
you
can
say,
cube
city
I'll
get
pod
hyphen
a
so.
A
A
A
So
for
folks,
who
don't
know
like
what
cube
builder
is
q
builder
is
a
tool
to
write
kubernetes
controllers
and
let
me
just
pull
up
okay.
So
now
this
kcd
demo
control
plane
is
being
created,
we'll
see
the
status
and
of
it,
whether
it's
ready
or
not,
and
what
state
it
is
in
in
some
time.
But
meanwhile,
if
you
see
this
qcdemo
md01,
this
is
the
you
know:
worker
machine
that
we're
trying
to
create.
A
A
B
A
So
as
of
now,
I
don't
think-
or
maybe
I'm
not
aware-
we
can
do
that,
but
you
know,
as
far
as
I
know,
for
a
single
workload
cluster
that
we
want
to
create.
We
are
having
tube
machine
template,
one
is
for
control,
plane
and
one
is
for
worker
nodes.
So
if
you
are
having
a
machine
template
that
is
the
template,
that's
going
to
be
used
throughout,
but
maybe
you
know
I
don't
know
if
we
can
do
like.
You
know,
multiple
notes
with
different
templates.
So
you
know
yeah.
B
A
Right,
like
cluster
class,
could
be
one
way
to
do
that,
but
I
just
didn't
want
to
bring
that
up
right
over
here
yeah
that
could
be
one
like.
You
know
this
question
that
is
asked.
There
are
certainly
ways
to
do
that
that
I'm
sure,
because
we
run
couple
of
p2e
tests
where
we
have
heterogeneous
nodes
like
there
are
some
windows
node
and
there
are
some
linux
nodes
right.
So
how
we
do
it
to
use
wired
templates
is
what
I'm
not
sure
of.
Maybe
you
know
I'll.
Okay
call
back
on
this
on
slack.
A
A
Fast,
so
you
know
just
to
reiterate
this
control
plane
is
succeeded,
so
you
know,
while
we
wait
for
worker
machine
to
come
up,
what
we
can
do
is
we
could
actually
pull
in
the
cube
config.
So
what
we
did
just
now
is
that
we
created
a
target
cluster
on
azure
cloud
and
now
let
me
let
me
pull
this
command
to
get
the
cube
config
so
using
cluster
ctl.
A
A
Node,
you
know
come
up,
the
state
is
not
ready
and
the
worker
machine
has
not
yet
come
up
because
it's
still
being
in
creating
state,
but
now
we
we
can
see
like
we
have
one
worker
machine
and
one
control
plane
machine,
but
the
status
is
not
ready
and
the
reason
is
that
we
do
not
have
a
cni
solution
yet
installed.
So
what
I'll
do
is
you
know
quickly?
I'll?
Just
install
a
cni
so
that
our
roads
are
our
nodes
are
ready,
and
this
should
actually
install
the
cni.
F
A
Okay,
so
this
is,
you
know
what
you
want.
What
I
want
all
of
us
to
realize
is
you
know
you
see
how
easy
it
is
to
actually
create
a
kubernetes
cluster
just
using
these
yamls
and
this
aml's.
This
declarative
configuration
gives
us
a
powerful
way
to
capture
this
into
ci
cd.
So
now
your
whole
cluster
can
be
described
in
a
declarative,
yaml
file,
and
you
can,
you
know,
do
crazy
things
with
that.
A
Yeah
done
so
awesome.
We
created
a
kubernetes
cluster
of
version
1.21.
Now
let
me
quickly
upgrade
the
kubernetes
version.
I
know
there
are
other
logistics
involved
when
you
really
upgrade
the
kubernetes
version,
because
in
production
grade
you
know
you
deploy
the
workloads
and
there
are
a
lot
of
metrics
match
that
goes
on,
but
I
just
want
to
you
know
realize
that
how
easy
it
is
to
upgrade
a
kubernetes
version,
so
what
I'm
doing
is
I'm
going
to
upgrade
from
1.21
to
1.22
and
the
upgrade
works?
A
A
A
A
A
And
you
know
I'll
I'll
show
you
like,
while
it
gets
upgraded
I'll
show
you
like
how.
So
if
you
see
like
this
cube,
ctl
get
notes
right,
yeah,
you
can
see
that
there
is
a
control
plane,
node
that
has
came
up
with
version
1.22,
and
this
is
in
not
ready
state.
So
this
is
transitioning
right
now
and
once
this
is
ready,
it
is
done,
it
will
be,
the
scheduling
will
be
disabled
and
this
node
will
be
deleted.
A
So
this
is
how
you
upgrade
it
like.
First,
you
upgrade
the
control
plane
and
now
you
can
upgrade
the
worker
node,
I'm
not
going
to
do
a
worker
not
upgrade
just
for
the
sake
of
time,
because
taking
like
a
while
but
I'll
show
you,
you
know
what
you
can
do
to
upgrade
a
worker
node
after
you
are
done
with
control
plane.
So
you
can
just
do
cube,
ctl
get
md
and
you
can
write
cube
ctl
edit
md,
and
this
is
machine
deployment.
A
You
know
I
know
like.
I
have
not
talked
a
lot
on
the
terms
of
machine
deployment
on
all,
but
this
is
one
of
the
constructs
that
handles
your
worker
nodes,
and
here
also,
you
can
see
a
version
right
if
you
can
see
I'll
select
that
part
and
you
just
simply
have
to
change
it
to
1.22.0
and
all
the
worker
machines
will
get
upgraded
to
the
next
version
that
we
have
selected
here
right.
So
this
is
for
the
upgrade
and
let's
say:
if
we
want
to
scale
the
you
know,
worker
notes.
A
It
is
again
simple.
You
just
do
cube
ctl,
edit
md
right
and
you
just
search
the
replicas,
the
it's
similar
like
you
scale
the
parts
in
a
deployment
right.
You
read
the
deployment
and
you
say
that
I
want
like
five
replicas
six
replicas
here.
The
replicas
are
one.
If
you
make
this
replica
to
two
or
three,
you
know
you
will
have
more
worker
nodes.
Let's
say
you
have
two
replicas
there.
One
more
worker
machine
will
appear.
A
Just
before
closing,
let's
see
where
we
are
at
so,
if
you
can
see
this
is
scheduling
disabled
and
this
is
ready
in
1.22
and
if
we
do
upgrade
the
worker
node,
why
the
steps
that
I
told
you
it's
already
written
on
instructions
also,
it
will
also
become
1.21
and
if
we
have
like,
if
you
want
to
scale
it,
we
change
the
replicas
to
2.
One
more
worker
node
will
appear
here,
so
that
was
it
for
the
demo.
Apologies.
A
We
I
couldn't
complete
the
whole
op
the
thing
and
couldn't
show
you
live,
but
I
think
I've
shown
the
most
part
of
it,
which
I
wanted
to
and
in
the
interest
of
time,
I'll
stop
here
and
hand
over
to
anusha,
and
you
know
just
10
second
pause.
If
anybody
has
any
burning,
question
feel
free
to
ask
me
or
we
could
do
it
on
slack
also.
B
A
A
A
So
if
you
just
it
takes
a
while.
So
if
you
do
like
cube,
ctl
get
node
after
a
while,
you
know
that
node
will
be
automatically
removed
and
the
reason
this
transition
is
like
this
is
that
you
want.
You
always
want
to
be
nha.
You
don't
want
to
interrupt
your
workloads
or
other
stuff
right.
So
if,
if
it's
a
management
cluster,
you
don't
want
to
interrupt
your
management
cluster
operations
right,
you
want
to
be
highly
available
at
any
time.
You
want
to
create
a
cluster
or
stuff
like
that.
You
can
do
that.
A
So
what
happens
is
when
you
upgrade
a
humanities
cluster.
A
new
machine
is
brought
up
a
new
control.
Plane
machine
is
brought
up
and
then
you
know
the
older
one
is
cordoned,
it's
kind
of
drained
so
that
you
know
all
the
newer
things
will
not
get
scheduled
onto
that
and
when
the
newer
is
ready,
you
know
all
of
the
control
plane.
Things
gets
scheduled
to
the
newer
machine
and
then
once
it
is
done,
the
older
one
is
decommissioned.
A
A
Like
you
know,
when
you
do
upgrade,
it
is
similar
to
how
you
do
upgrade
deployments
right.
You
you,
you
know.
Let's
say
you
have
like
five
worker
notes
and
then
you
you
upgraded.
You
can
specify
how
many
percentage
of
worker
notes
you
want
to
be
actually
live
and
serving
the
cluster
right.
So
you
could
specify
that
percentage,
and
according
to
that
you
know
the
upgrades
will
happen.
A
So,
for
example,
if
you
have
if
this,
if
you
have
specified
that
okay,
I'm
okay
with
having
less
number
of
you-
know
worker
nodes
and
then
what
will
happen?
Is
that
immediately
it
will
a
couple
new
machines
and
delete
the
older
machines.
For
example,
you
can
have
like
one,
you
know
worker
machine
radiant
serving
and
then
all
four
into
deleting
state
and
all
four
and
new
four
worker
machines
coming
up
or
you
can.
A
You
know
play
with
the
number
and
set
up
like
you,
can
have
two
worker
machines
serving
and
then
rest
two
being
in
deleted
and
the
other
new
to
into
you
know
spawning
state.
So
you
could
specify
the
percentage
sirs
into
the
yaml
yeah.
This
is
how
the
rolling
upgrade
strategy
works,
and
this
is
this
is
the
standard
practice
you
know
to
do
the
rolling
upgrades
in
production
environment.
A
Meanwhile
I'll
answer
one
question
that
cornell
has
posted.
While
you
wait
for
the
poll
like
unmanaged
human
disk
cluster,
yes
by
unmanaged,
let's
not
get
into
like
that.
It
is
not
managed
by
campy
unmanaged.
Current
is
cluster,
I
think
cornell
you
mean
where
you
have
like
control,
plane
and
worker
knows
and
then
aks
or
eks
in
aws
is
like
you
know,
you
don't
have
your
control
plane
machines
access,
so,
yes,
we
do,
have
you
know
feature
to
manage
aks.
A
You
could
potentially
manage
aks
via
capi.
The
thing
is
that
when
you
do
aks,
you
don't
have
like
control
on.
You
know
the
control
plane
machines
right,
that's
how
it's
attracted
abstracted
from
the
cloud
provider
itself,
but
yeah.
This
feature
is
there.
You
can
utilize
capsi
to
manage
your
aks
clusters
or
you
can
have
you
know
the
plain
vanilla
one
like
I
mentioned
like
you,
can
have
your
own
control,
plane
and
worker
nodes
management.
Does
that
answer
cornell
thanks.
C
So
one
thing
I
you
know
this
is
the
first
time
that
I'm
actually
attending
a
class
capi
and
like
a
cab
provider
workshop
and
for
me,
I've
used
like
ui
and
like
ansible
scripts
before
this
one
definitely
seems
a
much
more
cleaner.
We
are
like
kubernetes
native
way
of
doing
like
managing
the
clusters,
also
pretty
cool
thing
that
you
guys
are
demonstrating
today,
thanks.
B
B
So
I
have
two
readme
files
here,
one
for
mac
and
ubuntu,
but
I
just
want
to
give
a
quick
disclaimer
that
the
preferred
flow
for
follow
come
on.
It
should
be
flow.
The
preferred
flow
for
this
workshop
is
using
ubuntu
vm.
If
you
all
have
that,
because
when
we
install
kubernetes
and
run
cube
adm,
it
might
involve
some
changes
to
your
system
settings
in
terms
of
turning
off
swap
firewall,
enabling
net
filters
and
so
on,
and
this
is
untested.
So
that's
why
I
don't
want
to
dive
into
a
mac
workflow
today.
B
B
Cool
so
I'll
quick,
very
quickly
bring
back
this
diagram
for
those
of
you
who
joined
late.
This
is
what
we
are
going
to
be
doing
today,
like
asito
said
we
need
a
management.
Cluster
byh
currently
supports
only
provisioning
workload
clusters,
so
we
have
to
rely
on
some
other
infra
provider
to
be
our
management
cluster.
B
So
while
this
is
happening
so
with
this,
we
are
getting
our
management
cluster
and
we
need
cappy
provider
and
beyond
for
a
provider
on
this.
So
from
our
sources
demo,
if
you
remember,
we
are
going
to
initialize
our
management
cluster
with
our
infrastructure
provider.
So
you
know
earlier
we
did
infrastructure
as
aws,
sorry,
azure,
and
now
we
are
going
to
do
as
byoh.
B
B
Okay,
so
if
you
see
they're
not
just
infra
providers,
there
are
a
bunch
of
control,
plane
providers
and
bootstrap
providers
as
well,
so
cappy
being
the
core
provider
for
bootstrap.
We
have
options
like
aws,
eks,
cubedium
and
talos
control,
plane
provider
and
look
at
this
list
of
info
providers
right
and
this
this
like
keeps
growing
day
by
day.
B
B
This
is
happening,
okay,
so
with
this
our
basically
our
management
cluster
will
be
ready.
All
we
have
to
do
is
now
spin
up
new
workload
clusters,
so
there
was
a
key
difference
that
I
was
mentioning
between
vivace
provider
and
rest
of
the
infra
providers,
like
with
azure
that
we
saw
it
was
creating
machines
on
fly
so
when
I
suppose
issued
a
command
for
applying
the
cluster
template.
B
That's
where
new
vms
were
created,
depending
on
what
os
you
want
and
what
kubernetes
version
you
want,
but
with
byoh,
like
literally
the
name
says,
bring
your
own
host,
so
we
cannot
create
the
host
for
you.
You
have
to
bring
the
host
to
us
so
for
this
demo
we
will
be
using
docker
containers,
as
our
hosts
so
we'll
be
creating
a
couple
of
containers,
because
we
are
going
to
do
a
workload
cluster
with
one
control
play
node
and
one
worker
node.
And
earlier
someone
asked
what
are
the
limitations
of
pyoh?
B
I
wouldn't
call
it
a
limitation,
because
limitation,
in
my
mind,
is
a
problem
that
has
not
been
solved
yet
so
and
by
just
like
a
very,
very
nascent
provider.
So
so,
at
the
moment
device
supports
ubuntu,
2004,
host
x86,
64
architecture
and
there
it
supports
kubernetes,
1.22
and
1.23
series.
So
let's
get
to
that
before
that.
Okay,
let's
do
a
cube,
cd
and
get
points.
I
hope
you're
following
with
me
and
if
not,
I
can
go
like
even
slower.
B
If,
if
you
think
this
is
a
little
bit
fast,
just
let
me
know
in
the
chat.
So
when
you
do
qctl
get
ports,
you
see
we
have
byos
system
our
namespace,
and
this
is
the
bih
controller
manager
that
is
running
so
this
manager
has
different
controllers
registered
to
it.
We
have
controllers
for
bio
machine,
a
bio
host
and
a
bio
cluster,
so
all
these
three
reconcilers
will
be
running
and
we
have
the
capi
system
in
which
the
core
capi
controller
is
running
and
then
similarly
one
each
for
bootstrap
and
control
plane
controllers.
B
B
B
B
B
Now
we
have
to
prepare
this
host.
Like
I
said
this
is
bring
your
own
host
right.
We
need
to
have
certain
prerequisites
again
going
back
to
this
image.
We
need
to
have
host
agent
installed
on
all
of
these
hosts,
so
this
binary
is
available
for
download
on
the
releases
page
on
the
releases
page
and
also
earlier
somebody
asked:
how
is
it
going
to
talk
to
the
management
cluster,
so
we
will
provide
the
cube
config
of
the
management
cluster
so
that
the
host
can
discover
the
ip
and
reach
out.
B
B
A
B
Yes,
because
these
hosts,
they
are
bare,
minimal,
ubuntu
hosts.
We
have
to
install
container
d
and
things
and
for
that
the
privilege
is
required,
maybe
in
future,
if
we
allow
customers
or
users
themselves
to
bring
their
own
container
runtime
and
you
it
doesn't
have
to
be
container
d
as
well,
then
we
could
probably
not
then
run
them
as
privileged.
So
just
because
of
container
d,
we
have
to
run
them
as
privileged.
I
So
just
to
add
to
it
I
mean
yeah,
I
mean
it's
fine
as
you're
telling
that
to
run
they
must
privilege
to
run
the
I
think,
the
cris.
We
need
to
run
them
as
privileged,
but
don't
you
think
I
mean:
will
it
be
a
good
practice
with
respect
to
the
security
side
as
well,
because
normally
we
don't
recommend
containers
to
run
as
privileged
unless
it's
very
necessary
right.
B
So
again,
this
is
a
very
development
kind
of
environment
right
where
your
containers
are
hosts,
so
in
an
actual
production
system,
you'll
have
bare
metal
as
host
bare
metal
servers,
so
there
you'll
have
to
run
so
the
agent
that
I'm
showing
over
here.
This
agent
process
needs
to
be
run
as
sudo,
so
so
consider
this
host,
not
as
a
container
but
as
a
bare
metal
server,
but
the
agent
has
to
run
a
sudo
as
long
as
we
have
to
do
and
install
this
container
do
things.
A
Okay,
I
just
want
to
like
retreat
on
what
anusha
said
someone.
So
this
is
like
this
certain
small
kind
of
use
cases
where
you
know
the
admins
has
to
do
this.
I
think,
for
example,
even
in
let's
say
you
want
to
bring
up
a
kubernetes
cluster
and
you
want
to
have
certain
kernel
modules
or
you
know
something
be
loaded
on
your.
You
know
nodes
right.
We
need
admin,
access
or
something
like
this,
something
similar
to
that.
But
you
agree.
A
I
I
I
mean
I
mean
why
I
asked
was
more
from
the
context
of
let's
say
when
we
are
bringing
byoh
as
more
kind
of
a
ga
on
the
environment
for
the
customer
side.
So
probably
one
thing
that
would
come
always
is
some
some
kind
of
security
audit
right
that
would
be
run
on
this
workload.
So
possibly
that
could
be
one
potential
query
that
could
come
that
why
containers
are
running
as
privilege,
and
we
I
mean.
I
B
B
A
B
B
So
as
next
step
from
this
diagram,
we
need
to
have
the
host
agent
running
on
each
of
these
hosts
and
also
they
need
this
cube
config
file
to
reach
out
to
the
management
cluster.
So
this
for
loop
is
doing
just
that
copy
the
hostage
in
binary
and
then
copy
the
cube
config
file.
So
I'm
going
to
run
this.
B
B
B
Similarly,
the
host
agent
on
the
worker
side
or
on
the
host
runs
a
few
controllers
itself,
so
it
is
also
constantly
trying
to
reconcile
the
state
of
the
host
to
the
desired
state.
So
if
you
see
you
say
starting
controllers
and
then
it
received
a
reconcile
request
and
then
it
said
machine
left
not
set.
So
what
does
it
mean
by
machine
left
not
set?
It
means
it
has
registered
itself
as
available
capacity,
but
as
a
as
an
admin.
B
B
We
can
look
at
that
over
here.
So
if
I
do
cube,
cpl
get
bio
host,
so
we
can
see
two
hosts
that
adjusted
and
that's
these
two
hosts
now,
let's
see
what
to
do
next.
Okay,
so
we
are
going
to
create
our
cluster
yaml
file
and
before
that,
a
couple
of
n
variables
that
we
need
here.
So
one
is
the
bundle
lookup
tag.
So
what
is
a
bundle?
So
I'm
not
sure
how
many
of
you
are
aware
of
an
image
package
bundle.
B
So
image
package
is
one
of
the
part
of
the
cardboard
tooling.
I
think
there
was
a
kavel
workshop,
so
what
it
does
is
it
lets
you,
you
know
package
all
your
kubernetes
configuration
or
any
of
the
debian
files
that
you
want,
or
you
want
to
reference
more
oci
images.
So
it
lets
you
package
all
of
this
as
a
single
bundle
and
you
can
host
it
in
your.
You
know:
oci
registries
oci
conformant
registries,
so
this
is
one
of
the
core
pieces
of
byoh
as
to
how
we
install
these
kubernetes
components.
B
So,
okay,
let's
come
here.
So
what
does
an
image
package
bundle
in
terms
of
byh
consists
of
so
like?
I
said
we
need
to
install
container
d,
cubedium,
cubelet
cri
tools
and
some
cni
so
to
bring
up
a
basic,
kubernetes
node
using
cube
adm.
These
are
the
files
or
services
that
need
to
be
running
on
your
host.
B
So
these
are
the
debian
files
that
are
pulled
off
of
their
upstream
github
repos
or
the
apt
packages,
and
these
debian
and
tar
files
are
bundled
as
an
oci
image.
So
this
is
where
I
think
by
is
really
powerful.
So
in
one
of
the
I
think
earlier
polls,
you
all
said
you
use
sibo
k3s
or
you
may
even
be
using.
You
know:
proprietary
kubernetes
flavor
that
your
you
know,
organization
wants,
but
the
cluster
api,
the
upstream
cluster
api
project
that
we
have
supports.
You
know
upstream
kubernetes.
B
So
if
you
want
to
bring
up
azure
cluster
that
we
did,
it
actually
gave
you
kubernetes,
and
this
was
conformant
with
the
upstream
kk
kubernetes
that
we
have
now
sub
today
in
this
bias
demo.
Also,
we
are
going
to
do
that,
but
I
just
want
to
like
take
a
minute
here
and
say,
if
suppose
say
you
wanted
to
do
mini
cube
or
micro
kits,
you
just
bundle
the
corresponding
bits
that
are
relevant
to
micro
case.
B
It
could,
I
don't
know
what
micro
gates
has,
but
it
could
be
a
different
kind
of
cubed
debian
that
it
needs,
or
it
could
be
a
different
kind
of
maybe
an
additional
configuration
file.
It
means
so
all
you
have
to
do
is
bundle
that
configuration
file
into
this
image
and
when
byh
actually
pulls
this
image
and
unpacks
and
executes
those
things
you
will
have
a
you
know.
Mini
cube
that
is
running
in
your
bbs
notes,
still
managed
by
cappy.
You
can
have
say
micro
gates,
running
in
your
b
watch,
node
managed
by
cappy.
B
You
can
have
upstream
kubernetes.
You
can
have
your
own
organizations,
you
know
private
kubernetes
sort
of
thing
still
managed
by
cappy.
So
this
is
where
you
know
the
power
of
bih
lies
moving
forward.
So
that's
the
bundle
lookup
tab.
So
you
it's
just
an
oci
image.
This
tag
is
equivalent
to
say
a
docker
image
tag
in
vivac.
Our
convention
is
to
use
the
same
say
I
want
a
kubernetes
123.5
version.
Our
tag
is
also
tagged
in
a
similar
way.
This
is
just
a
convention
and
the
control
plane
endpoint.
B
So
for
the
control
plane
nodes,
you
might
need
h.
A
right
so
b
at
the
moment
supports
cube
whip.
So
for
those
of
you
are
new
to
cube
whip
cubic
provides.
You
know,
clusters
with
a
virtual
ip.
All
it
needs
is
a
virtual
ip
and
it
can
act
as
a
load
balancer
for
your
control,
plane,
nodes
and
also
for
your
workloads
there.
There
are
like
other
options
like
you
could
use
vmware's,
rv
or
you
you
could
use.
B
You
know
h.a
proxy,
but
at
the
moment
beyond
supports
cubic
so
like
it
says
it
needs
a
virtual
ip
right.
So
that
is
what
we
are
going
to
provide
here
and
rest
of
the
things
are
like
pretty
much.
What
we
used
for
azure
we
use
cluster
cpl
generate
our
infrastructure
is
byh.
This
is
the
kubernetes
version.
We
are
going
to
create
one
control,
plane,
node
and
one
worker,
node
and
flavor
is
docker
because
I'm
using
docker
containers
as
hosts.
B
B
So
your
virtual
ip
has
to
be
in
the
same
network
but
outside
of
the
dhcp
range
outside
of
the
dhcp
range
because
say
some
other
container
comes
up
or
you
decide
to
introduce
a
new
bare
metal
server
into
your
capacity
pool
your
control
plane.
Ip
should
not
interfere
with
the
newer
ip
that
is
getting
assigned.
That's
why
it
has
to
be
outside
of
your
dhcp
range.
B
Okay,
sorry
about
that,
okay,
so
this
is
the
cluster
yaml
that
it
was
supposed
to
create.
I
think
we
were
looking
at
when
azure
cluster
yaml
was
created
right.
We
have
a
cluster,
but
this
time
the
infrastructure
f
is
pointing
to
bio
cluster.
And
similarly,
if
we
can
go
through
what
a
pyo
cluster
is.
B
B
F
B
B
So
if
you
want
to
have
a
kubernetes
version
or
distribution
or
flavor
of
your
own
choice,
you'll
have
to
create
a
new
bundle
and
give
that
in
your
cluster
yaml
saying
this
is
what
I
want
to
install
there's
a
little
more
nuance
to
it
right
now
by
understands
how
to
install
only
upstream
kubernetes.
That
is
the
steps
it
has
to
follow
to
understand
this
bundle,
so
you'll
have
to
probably
write
your
own
set
of
steps
that
can
install
your
version
of
kubernetes,
so
it
installed
all
the
kubernetes
bits.
B
Next,
it
is
going
to
bootstrap
the
case
node,
so
this
bootstrapping
happens
via
cubedium.
Since
this
is
the
control
plane
node,
it
is
under
the
hood
executing
cubadium
in
it.
If
it
were
a
second
control,
plane,
node
or
any
other
worker
node,
it
would
execute
a
cubidium
join
in
the
meantime,
let's
look
at,
let's
have
a
watch
on
the
machines
themselves.
B
B
Okay,
this
one
says
successfully
bootstrapped
and
now
it
should
ideally
start
the
same
set
of
steps
over
here
in
the
other
host.
So
right
now
it
is
randomly
picking
any
of
the
available
hosts.
For
you
know,
this
is
going
to
be
my
control,
plane
node,
but
if
you
remember,
when
we
issued
this
particular
command
to
start
the
agent,
this
also
accepts
another
argument
called
labels.
B
So
if
you
label
your
host,
as
you
know,
this
should
be
my
control
plane,
node
or
this
should
be
my
worker
node,
because
worker
node
is
where
all
your
actual
workloads
will
be
deployed.
So
if
you
want
to
give
high
capacity
to
that
or
like
if
you
have
a
host
which
says
you
know
eight
cores
and
something
else
with
only
four
cores,
you
can
attach
those
labels
to
all
of
these
hosts
and
in
the
cluster
yaml
you
can
specify
for
each
label.
You
you
have
a
corresponding
selector.
B
So
again,
it's
doing
similar
things
installing
kids,
we
have
a
bunch
of
warnings
which
I
need
to
look
into
after
this
call,
and
then
finally
it
says:
okay,
it's
not
successfully
bootstrapped!
You
come
back
here,
okay,
so
now,
if
you
see
both
of
these
machines,
have
a
provider
id
all
right,
you
go
here.
Okay,
both
of
my
notes
are
bootstrapped.
Let's
come
back
here.
B
B
B
This
actually
doesn't
have
to
be
a
separate
step,
because
api
provides
constructs
such
as
cluster
resource
sets,
wherein
you
can
specify
for
a
bunch
of
your
work
on
for
a
bunch
of
notes
or
for
a
given
particular
cluster
whenever
a
new
node
is
joined,
apply
this
resource
set.
Now
this
resource
set
could
be
a
cni
csi.
Anything
that
kind
of
applies
to
the
entire
cluster,
so
we're
going
to
apply
cni.
Now
I
have
a
cni
hosted
locally.
B
B
B
So
I
hope
this
goes
into
steady
state.
Yes,
so
first
the
control,
plane
and
then
everything
else
now.
I
think
I
think,
for
the
demo.
This
is
all
I
have,
but
a
couple
of
notes
that
is
in
azure.
We
saw
how
easy
it
was
to
upgrade
a
kubernetes
cluster
right.
You
just
had
to
go
and
upgrade
edit
the
kubernetes
version
and
the
azure
infrastructure
provider
created
new
vms
with
the
updated
kubernetes
version
and
then
deleted
the
older
vms.
B
So
here
it's
going
to
be
slightly
different,
because
again,
I'm
just
repeating
myself
that
this
is
bring
your
own
host.
That
is,
the
host
already
has
to
be
registered
in
the
capacity
pool.
The
infrastructure
provider
cannot
create
these
hosts
on
the
fly.
So
if
you
say
you
want
to
upgrade
your
kubernetes
cluster,
you
have
to
make
sure
that
you
have
spare
capacity
in
your
capacity
pool
and
also
ashitosh
mentioned.
We
don't
have
in-place
upgrade,
so
it
means
we
cannot.
B
You
know,
remove
all
the
kubernetes
versions,
the
kubernetes
components
on
this
host
and
then
install
new
ones.
We
do
need
new
nodes.
So
now
we
have
two
hosts
here.
If
I
want
to
upgrade
this
cluster
I'll
I'll
need
to
have
two
more
hosts
in
my
capacity
pool,
so
I'm
going
to
install
new
kubernetes
versions
on
those
and
these
will
get
deprovision.
Ideally,
these
will
go
back
to
your
capacity
pool,
so
this
is
again
a
difference
with
vivac
and
remaining
providers.
A
C
Bring
one
like
towards
the
end
cool
that
we
don't
give
a
crap
on
that
one,
but
just
wanted
to
call
out,
like
you
know,
both
these
projects
that
we
just
saw
or
like
completely
under
kubernetes
under
the
capi
and
as
you
are
like
are
already
there.
People
which
it's
in
the
process
of
like
you,
know
getting
into
kubernetes
six,
or
has
it.
F
C
Yeah
cool,
so
you
know,
join
us
on
the
kubernetes
slack,
as
well
as
the
repositories
with
the
issues
and
all
that,
and
then
we
can
continue
those
discussions
there
all
right
any
last
minute
questions
before
we
take
like
a
quick
break
of
five
minutes
and
then
we
totally
switch
hands
towards
the
pc,
which
probably
gives
you
like
a
little
bit
of
higher
level
implementations.
On
top
of
this
right,
aman
and
karupaya
over
to
you.
D
So
let's
start
with
the
unmanaged
cluster
yeah
so,
as
I
have
gave
us
some
brief
about
the
unmanaged
cluster,
so
it's
a
standalone
cluster
and
you
can
use
both
kubernetes
like
cube,
ctl
and
tensors
alloy
inside
it
and
it
comes
with
or-
and
you
can
use
all
the
features
which
tc
is
providing
like
the
plugins,
the
packages
and
other
stuff.
So
you
can
use
that
to
there.
So
if
I
start
with
the
demo
or
installation,
so
the
prerequisites
are
like
docker
cube,
ctl
and
install
tc.
D
So
I
think
I'm
assuming
that
docker
and
keeps
you
killer
there
if
in
case
tc
is
not
there.
So
I
have
written
down
the
shortest
way
to
install
a
pc
so
just
have
to
curl
or
wget
up
to
your
os.
So
I
am
using
a
linux
environment,
ubuntu,
and
so
these
are
the
commands
for
there.
I
am
I
so
same.
Commands
can
work
on
mac
too.
So
that's
that
so,
let's
start
with
creating
start
with
creating
an
unmanaged
cluster.
So
I
have
so
for
that.
D
There
are
two
ways
to
create
a
unmanaged
cluster.
So
one
is
like
a
simple
which
is
running
command
in
which
it
will
fetch
a
kind
image,
then
creating
a
kind
cluster
above
which
it
it
is
creating
a
cap
controller
cni
like
calico
and
then
fetching
the
repositories
and
installing
them.
So
that's
the
one
way
of
installing
the
unmanaged
cluster
and
another
way
is
that
if
the
any
cluster
is
present
in
your
environment
already-
and
you
want
to
convert
that
cluster
into
one
minus
cluster,
so
there
is
also
that
way.
D
D
So
it
is
fetching
the
kind
classroom
is
a
link
I
shared
with
you
earlier,
okay.
So
meanwhile,
the
on
my
cluster
is
installing:
let's
explore
the
all
new
features
which
we
are,
which
are
added
in
a
managed
cluster,
so
here's
the
one
like
support
and
my
cluster
profile.
So
what
does
that
mean?
Is
that
you
have
installed
unmanned
cluster?
You
have
installed
some
packages
there
or
installed
any
application
there.
D
So
you
can
save
that
profile
and
you
can
use
or
rerun
it
again,
so
you
don't
have
to
create
any
application
or
anything
from
scratch.
So
that's
the
one
feature
which
is
added
into
our
unwanted
cluster
so
and
the
there
are
many
more
I
have
just
fetched
few
of
them.
So
the
other
feature
which
we
have
added
into
on
my
cluster
is
like
with
an
existing.
D
So
the
same
thing
which
I
spoke
earlier
like
you
have
a
cluster,
you
can
convert
it
into
a
unminus
cluster,
so
these
are
the
two
features
which
they
are
multi-modal,
more
features
which
we
are
going
to
add
and
we
are
in
process
to
add
so
yeah
cluster
is
up.
So,
let's
start
with
some
running
some
commands
there.
D
D
D
So
let's
see
the
repository
list,
so
this
list
means
that,
like
the
list
of
packages
with
which
it
have
so,
these
are
two
lists.
The
advanced
cluster
consists
of
like
one
have
the
core
packages,
other
one
have
user
manage
packets
like
valero
core
dns,
multi
cni,
so
these
are
the
packages
which
it
includes.
D
So
we
are
starting
with
installing
a
fluent
bit
on
it
and
see
like
how
so
fluent
with
is
a
application
which
is
also
which
process
the
logs
and
show
you
in
the
form
of
matrices
and
in
a
simpler
manner.
I
would
say
so:
let's
install,
let's
see
the
version
so
currently
for
fluent,
but
there's
a
only
one
version
so
for
the
other
packages,
you
can
also
see
with
the
same
command
the
version
thing.
D
H
D
H
D
H
D
D
So
these
are
the
all
the
packages,
so
every
package
have
their
readme
file
so
how
to
install
them,
how
to
use
them.
So
you
can
go,
install
them
on
pc
using
pc
on
unmanaged
cluster
and
avail.
Their
services
like
graph
and
now
prometheus
and
lots
of
them-
are
there
so
to
delete
them.
Let's
we
have
already
deleted
the
fluentplate
once
let's
delete
it
again,.
H
D
D
D
A
J
J
That
you,
you
have
enough,
say,
access
to
create
stuff.
I
mentioned
some
of
the
resources
that
we
create
as
part
of
the
cluster
and
also
you
have
quota
available
like
if
you're
already
using
some
resources,
you
also
have
to
ensure
you
have
put
or
request
more
quota
and
once
you're
done
with
all
that
credential
stuff,
the
other
prep
you
have
to
do
is
like
you
need
to
have
ssh
key
available
for
yourself.
So
this
is
touch
the
ssh
into
the
machines
that
you
deploy
for
your
kubernetes
cluster.
J
G
J
Once
you
get
done
with
all
of
that
prep,
you
just
go
ahead
and
like
do
this
thing
called
tanzu
management
cluster
create
ui,
and
you
actually
get
a
ui
like
this,
and
I'm
actually
already
running
it
over
here
and
you
have
both
dark
and
light
interface.
So
I've
been
using
the
dark
one
moment,
so
you
notice
that
actually
actually
have
my
aws
configured
in
my
local,
so
config.
So
I
have
a
profile
over
here.
The
default
profile
and
my
credentials
are
the
aws
credentials.
J
So
I
can
just
tell
tansu
that
hey
you
can
just
use
my
credential
profile
and
use
the
default
credential
profile
and
the
region
is
usc
one
and
once
you
do
that
and
say
connect,
it
will
say
connected
like
this
and
you're
good
to
go
and
everything
else.
You
just
can
click
next,
you
have
the
default
settings
which
are
good
enough.
J
So
this
is
a
classless
inter
domain
routing
range,
so
for
the
vpc,
which
is
short
for
a
virtual
private
cluster
and
you
can
go
ahead
and
like
mention
a
type
for
your
machine,
so
what
this
means
is
how
big
is
your
machine?
Is
it
like?
Having
less
resources
like
less
cpu
and
ram
or
more
so,
we
mentioned
over
here
a
bit
of
a
large
one
so
that
you
know
your
humanities.
J
And
finally,
you
get
this
sort
of
ui,
which
is
what
I'm
actually
seeing
so
actually
this
sort
of
takes
around.
So
I
ran
it
around
like
12
30
and
it
took
around
like
some
half
an
hour
in
my
mac
os.
That's
because
I've
been
running
zoom
and
whatnot.
But
if
you
have
say
linux,
it
might
be
probably
less
like,
say:
15-20
minutes.
J
And
you
should
be
able
to
see
something
like
this,
wherein
it
creates
a
cloud
formation
stack
and
then
like
it,
basically
uses
clustering,
cluster,
api
and
cluster
provider
for
aws.
It
does
all
of
this
stuff
that
ashutosh
spoke
about
initially
and
it
creates
your
kind
bootstrap
cluster
and
everything.
Does
everything
in
one
go?
Basically,
I'm
seeing
some
chat
stuff,
don't
say:
hi.
A
A
J
J
J
J
J
J
Yep,
the
nodes
are
all
good
and
you
can
actually
also
see
the
cluster
info,
where
you
can
actually
see
that
it
has
a
public
ip
and
a
url
that
the
amazon's
elastic
load.
Balancer
gives
you
and
you
can
also
see
all
the
pods
that
are
running,
and
you
can
also
do
the
same
commands
that
ahman
wrote
for
tanzu
packages.
J
So
you
can
actually
see
tons
of
package
related
stuff
here
also,
so
you
can
same
commands,
you
can
run
to
install
fluent
bid
and
any
packages,
and
then
you
can
create
your.
You
know
workload
cluster
using
this
management
cluster,
and
this
has
all
of
your.
You
know
cluster
api
components
and
it
can
create
your
workload
cluster
with
that.
So
this.
F
J
H
J
Actually,
the
culprit
could
be
many
things,
but
if
we
want
to
find
exactly
what
it
is,
we'll
need
evidence
for
ways
we'll
have
to
actually
get
more
data.
Usually
what
you've
noticed
is
like
two
major
things:
one
is
if
you're
using
an
older
version
of
tc.
I
think
there
wasn't
some
support
for
some
particular
version
of
c
group.
I
think
v1
or
v2,
and
then
the
other
one
is
about
resources.
So
in
my
current
case,
I'm
actually
running
a
lot
of
resources
for
my
docker
one.
Second.
J
So
since
you're
running
on
management
cluster,
your
bootstrap
cluster
is
running
on
docker
and
then
your
management
cluster
is
also
running
on
docker
and
tanzu
creates
like
three
docker
containers.
One
of
them
will
be
pretty
heavy,
which
is
the
kind
container
and
actually
two
of
them,
the
control,
plane
node
and
the
worker
node.
The
third
one
is
just
a
proxy
which
is
hi
proxy,
which
is
pretty
lightweight,
and
your
bootstrap
cluster
is
also
kind,
so
there'll
be
three
kind
and
that's
pretty
heavy,
and
so
I'm
using
like
eight
cpus
and
for
aws.
J
J
J
So
here
you
can
actually
see
the
ec2
instances
that
I
have
so
it
says
tc
aws
demo
and
like
the
machine
deployment
similar
to
the
stuff
that
ashutosh
mentioned
and
then
there's
a
bastion
hose
for
you
to
actually
ssh
and
stock,
and
then
the
control
plane,
node
and
there
are
actually
more
resources.
These
are
the
problems,
so
you
can
actually
see
the
vpc.
J
J
C
Well,
like
not
related
to
the
workshop,
but,
like
you
know,
one
thing
that
seems
to
come
out
here
is
istanzu
like
a
cli
that
wraps
around,
like
the
various
tools
that
we
typically
use.
For
example,
like
you
know,
if
I'm
running
like
my
own
bare
metal
cluster
right
like
I
have
to
use,
help
to
kind
of
install,
you
know
set
of
add-ons
and
all
that
kind
of
stuff
or
use
like
enabling
level
of
let's
say
like
microcharacters
or
something.
J
J
The
editions
of
tanzanite
is
grid,
so
different
editions
use
this
as
the
core
and
then
tc
brings
its
own
features,
and
so
tanjo
has
its
own
like
package,
plugin
and
whatnot
micro
case
also
has
its
own,
like
some
sort
of
like
plugins
for
installing
packages.
For
example,
if
you
go
check
out
moment,
oops.
J
Okay,
so
here
you
can
see
micro
case
mentioning
about
this
thing
called
enable
dashboard
dns.
So
this
is
basically
micro
gates
way
of
saying
hey.
We
have
packages
for
these
humanities
dashboard,
cube
dns,
you
know
image
registry
stu
and
you
can
actually
just
deploy
them
similar
to
that.
Tanzu
has
like
you
know,
packages
where
you
can
say
fans
of
package
available
list,
and
you
should
be
able
to
get
all
the
packages
that
you
have
and
then
install
them
also,
but
it's
not
a
wrapper
around
existing
things.
It's
it's!
J
It's
mostly
on
its
own
and
you
build
on
top
of
existing
open
source,
for
example,
carville,
which
is
like
totally
different.
Word
is
like
what
do
you
say?
Helm
is
like
probably
simpler
when
it
comes
to
say
cap,
controller
cap
and
whatnot,
but
this
has
more
complex
features
and
has
more
power
in
terms
of
like
idd
cap
and
all
that
stuff.
B
A
Yeah
thanks
kira
and
drive
zany
for
this
wonderful
workshop
and
thanks
our
audience.
Like
anusha
said
we
the
murphys
thing
but
yeah
it
was
all
good.
Hopefully
you
had
good
time.
You
enjoyed
it
and
just
to
say
again
like
we
all
are
there
in
upstream
channel.
So
no
it's
not
over.
Yet,
let's
keep
chatting,
you
know,
feel
free
to
say
your
feedback
or
you
know
any
use
cases
that
you
want
to
be
solved
and
you
know
we'll
be
there
and
collaborate
there.
Thanks.