►
From YouTube: Webinar: What’s New in Kubernetes 1.12
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
Kubernetes 1.12 continues to focus on internal improvements and graduating features to stable in Kubernetes. These new features mean increased security, availability, resiliency, and ease of use to get production applications to market faster. The release also signifies the increasing maturation and sophistication of Kubernetes on the developer side.
Join members of the 1.12 release team to discuss the newest features in this release.
A
Hello,
hello,
everyone.
This
is
Steven
Agustin's,
kubernetes,
product
management
chair.
This
is
the
CN
CF.
What's
new
in
kubernetes
1.12
webinar
we're
joined
today
by
Katelyn
Bernard,
it
is
a
112
communications
lead
thanks
Annie,
who
is
the
cig,
asier
technical
lead
and
one
vallejo
who's,
part
of
fake
CLI.
A
You
can
go
ahead
and
hit
the
next
slide,
Katelyn,
so
quick
agenda,
we're
going
to
go
over
a
quick
overview
of
what's
new
and
in
112
and
then
we're
gonna
break
out
into
separate
sessions
for
either
vmss,
which
is
a
VMware
scale.
Virtual
machine
scale
sells
kubernetes,
cube,
CTL
plugins
and
then
we're
going
to
do
a
question
and
answer
section.
This
meeting
will
be
recorded
and
available
later
online.
A
A
A
Go
for
it
yeah
all
right,
so
a
quick
overview
of
TLS
say
to
us.
We
have
kind
of
been
on
this
journey
for
providing
security,
and
this
a
day
zero
day,
one
story
around
security
in
kubernetes,
since
around
1.4.
Now
there
were
in
112,
we
finally
see
kubernetes
bootstrapping,
moved
to
GA
and
trinity.
The
TLS
server
certificate
rotation
is
now
beta,
so
the
server
certificate
rotation
is
kind
of
the
second
part
of
the
story.
A
After
being
able
to
instantiate
a
cluster
certificate
authority
and
then
and
provide
provide
credentials
to
each
of
the
components
within
the
control
plane
for
kubernetes,
the
next
problem
is
being
able
to
actually
rotate
those
those
certificates
in
the
in
the
face
of
a
security
event.
So
in
this
release
and
the
ones
well
release,
we
see
that
functionality
moves
you
to
next
slide
right
to
scheduling,
so
so
just
a
heads
up
that
each
of
these
slides
are
kind
of
broken
out
by
the
the
kubernetes
special
interest
groups
or
six.
A
So
there
are
quick
overviews
of
each
of
what
each
of
the
sakes
have
been
working
on
for
this
release
cycle
so
quota
by
priority
in
paints
by
priority
both
move
to
beta
in
the
112
release
cycle.
So
essentially,
this
is
functionality
around
being
able
to
prioritize
the
types
of
workloads
that
land
on
notes
and
if
you
have
been
following
the
kind
of
112
111
and
112
cadence,
there
is
the
introduction
of
pod
priority
and
preemption.
A
Each
of
those
those
functionalities,
we
kind
of
see
an
iterative
improvement
and
being
able
to
actually
set
quotas,
set
quotas
based
on
the
priorities
that
you
can
configure
within
those
clusters,
and
it's
like
so
with
an
API
machinery.
This
this
kind
of
hinges
into
the
the
idea
of
being
able
to
configure
quotas
for
four
clusters.
One
of
the
one
of
the
larger
problems
around
deploying
workloads
on
kubernetes
is
that
you
have
to
be
cognizant
of
the
types
of
workloads
that
you
deploy
and
the
resources
that
they
consume
by
default.
A
Workloads
that
are
deployed
on
kubernetes
don't
have
a
default
quota,
so
in
with
the
introduction
of
this
resource
quota,
API
we're
also
bringing
in
the
the
ability
to
be
able
to
default.
What
that
quote
is
should
be
for
all
workloads
across
clusters.
That
way,
if
you
run
your,
you
know
your
Bitcoin
mining
scheme
or
something
like
that
on
the
cluster.
You
can
ensure
that
the
the
the
resources
that
it
consumes
will
not
will
not
cause
will
not
cause
issues
with
the
other
workloads
that
you're
running
across
that
cluster.
A
Secondarily,
we're
introducing
API
server,
try,
run
and
API
server
dry
run
is
a
pretty
pretty
important
feature
for
the
overall
development
process
around
kubernetes,
for
reasons
being
that,
if
you've,
if
you've
ever
had
the
opportunity
to
use
a
CLI
to
land
and
use
a
flag
like
dry,
run,
equals
true
or
or
you
know,
apply
equals
false
or
something
like
that
being
able
to
essentially
mock
mock,
workflows
around
around
different
deployment
schemes.
You
can
now
do
that
without
having
to
actually
consume
resources
right.
So
we
can,
we
can
kind
of
trace.
A
We
can
kind
of
trace
the
you
know
the
the
mechanisms
for
for
each
of
these
each
of
these
different
pieces
of
the
code
without
having
to
actually
do
those
actions
within
the
cluster.
So
we're
we're
gonna,
see
more
and
more
improvements
across
the
the
113
and
114
115
lifecycle
around
that
try
run
feature
next
slide,
so
networking
a
round
network.
We
are
bringing
Network
policies
to
2ga
all
right.
A
So
two
of
the
three
of
the
features
within
network
policy
or
one
being
able
to
define
egress
IP
is
egress
cider
box,
as
well
as
as
well
as,
instead
of
instead
of
two
specific
IP
addresses.
You
can
define
a
set
of
cider
blocks
for
for
the
rules
that
you
define
in
network
policies,
so
both
of
those
features
move
to
GA
and
in
the
1:12
cycle
next
slide.
A
Alright,
so
a
few
features
from
sick
node
would
being
the
runtime
runtime
class.
So
that's
the
ability
to
actually
have
have
resources
and
have
resources
and
that
are
deployed
that
are
scoped
to
specific
cluster
actions,
which
was
not
the
case
before
the.
In
addition
to
that,
there
is
also
functionality
around
being
able
to
share
share
processes
within
share
process
and
the
process
namespace
across
pods
within
across
containers
within
the
same
pod
and
then
finally,
the
cubelet
device
plug-in
registration.
A
So
storage,
a
few,
so
storage,
six
storage
always
has
iterative
improvements,
brown,
the
CSI
a
framework.
So
one
really
interesting
thing
that
we're
seeing
moved
to
beta
in
112
is
the
idea
of
topology
where
provisioning.
So
one
concern
that
we
have.
Is
we
start
to
take
the
idea
of
deploying
kubernetes
in
any
specific
data
center
or
availability
zone
or
region?
Or
what
have
you
is
the
idea
that
all
of
the
resources
that
you
deploy
for
that
cluster
need
to
be
close
to
the
things
that
they're
going
to
to
action
upon
right?
A
So
if
you
have,
if
you
have
a
note
that
you
deploy,
that
that
is
compute
in
say
for
AWS,
it's
us
deist,
1c
right.
You
want
to
make
sure
that
the
storage
that
you
provision
is
coming
up
in
u.s.
East
1c
as
well
right
so
starting
to
bring
in
this
idea
of
topology
aware
clusters,
and
it's
apology
were
components
for
the
clusters.
A
So,
across
release
cycles,
we've
also
seen
improvements
in
lots
of
lots
of
out
of
tree
functionality.
One
of
one
of
the
out
of
tree
features
being
auto
scaling.
So
there
are
two
means
of
their
two
based
means
of
auto
scaling
in
kubernetes,
one
of
which
being
horizontal
pot,
auto
scalars,
which
are
essentially
default
in
and
crouppen
edits.
You
can
define
those
by
default.
The
vp8
or
the
vertical
pod
autoscaler
is
an
add-on
to
the
kubernetes
core
code,
but
so
in
112
we
see
an
improvement
in
the
api.
A
Our
continued
improvements
in
the
and
the
the
HPA
v2
API,
it's
a
better
algorithm
to
ensure
that
scale
up
and
scale
down
operations
are
handled
in
a
safer
way.
That
being
say,
you
have
a
some
scaling
event.
You
have
a
major
sale
on
your
website
or
something
and
you
scale
up
to
100.
You
know
a
hundred
dollars
or
something-
and
you
know
that
event
is
no
longer
happening
and
you
want
to
scale
back
down.
But
let's
say
between
the
time
that
you
plan
your
scale
down
operation.
You
need
to
scale
back
up
right.
A
B
Hello,
Aaron
trainee
from
Microsoft,
Asia
and
I'm
also
coat
accolade
of
stick
area.
I,
do
you
mean
also
noticed
I'm
the
irreversible
process
and
the
cluttered
scalar
area
had
been
reached,
their
general
availability
in
this
release.
So
what
is
Bushnell
mashenka
said
and
what
we
are
way
benefit
from
the
now.
Let
me
make
a
new
suctioning
of
them,
so
the
first
part
is
about
a
bushel
machines,
cause
as
a
devotion.
Machines
go
says,
is
a
group
of
identical
virtual
machine
with
with,
for
showmanship
go
says,
those
I
did
thank
Co.
B
Virtual
machine
could
be
created
and
man
injured
from
a
single
central
conviction,
so
it
is
much
more
easier
to
crater
and
manager,
those
damn.
It
also
provides
higher
ability
and
application
real
talents.
A
virtual
machine
process
allows
your
application
to
automatically
skill
and
research
team
on
changes
such
as
host
metrics
and
the
other
youth
provided
metrics.
B
Also,
it
provides
a
larger
skill
of
via
mutants
such
as
1000
intense
within
a
single
skill
set.
So
this
enables
you
to
easily
managers
and
it's
to
give
multiple
virtual
machine
to
provider
by
a
high
availability
a
so.
It
is
ideal
for
large
scale
applications
like
continual
close.
That's
why
a
virtual
machine
coasters
are
important
for
kun
based
on
error.
So
here
is
the
list
of
benefits
compared
to
traditional
virtual
machine
and
virtual
machines,
I
recommended
to
typical
octopi
availability
size.
So
sometimes
it
is
called
for
some
training
ability
side
compared
to
Bushnell
machines
cassette.
B
B
The
third
scenario
is
about
2h,
a
and
redundancy,
so
for
virtual
machine
ugly
size
you
need
to
manually
create
a
able
to
be
aside
or
distributed
and
check
vamps
across
his
own,
while
virtual
machines
go
says
will
use
an
automatically
for
you.
It
will
also
automatically
distribution
of
their
maintains
across
a
bill
bit
zones
or
resize.
B
The
last
scenario
is
about
us
or
scanning,
so
for
virtual
machine
you're
in
you
need
to
do
menu,
monitor
room
and
areas,
automation,
while
virtual
machines
go
says
or
just
go
downhill
for
you
with
a
metric
such
as
host
metrics,
Jim,
cast
metrics
and
application
inside,
but
headway
on
new
virtual
machine
CSS
on
a
way
of
the
internal
state
of
community
laughter.
So
this
to
any
it's
actually
not
recommended
humanists,
so
class
starts
together
is
recommended
it
for
managing
the
number
of
nodes
on
area.
B
The
current
status
of
virtual
machines
go
said.
Currently
we
have
a
supportive
many
features
for
first.
A
minute
ago
said
the
first
one
is:
you
could
create
a
multi
notes,
work
notes
with
virtual
machines,
virtual
machines
go
says,
and
also
you
could
make
a
combination
or
bush
memory.
God
says
and
availability
said
then
hit
your
master
hold
with
the
available
size
and
as
the
workers,
many
did
by
virtual
machines
go
says
also
for
of
per
virtual
machine
disk
code
attachment
I
important,
because
although
those
virtual
machines
are
identical,
the
attack
disk
are
different.
B
Also.
Also,
each
not've
will
were
beside
away
the
ability
zone
and
dynamic
volume
provision
will
be
also
spotted.
So
then,
a
disco
where
p
provision
him
some
zone
as
dictated
by
knows
his
own
in
the
future
version
browsers,
will
also
be
added
to
a
low
community
service
than
his
a
cast.
So
now
a
cast
only
suppose
a
bushel
machine,
a
baby-sized
and
the
Bushman's
cassettes
will
be
edited
later.
B
This
is
the
second
part
of
feature
about
a
clutch
or
scalar
class
or
scalar
allows
you
to
adjust
the
size
of
community
collateral
based
on
the
load
condition.
Basically,
it
could
create
a
new
noted
area
when
they
are
post
field
achieve
scale
or
because
of
insufficient
resources,
and
when
the
notes
have
been
under
utilized
for
a
period
of
time,
then
those
posts
will
be
drained
and
scheduled.
She
either
knows
this.
Node
will
also
be
removed
if
from
online
area.
So
the
result
Coast,
who
could
be
reduced
to
in
this
way.
B
Even
now,
all
their
four
types
of
surveys
as
coordinated
now,
the
first
one-
is
a
do
community
service
and
the
last
one
area
container
service.
This
tree
man
ended
its
service
area
and
a
says
it's
actually
an
older
version
of
acres,
so
I
guess
is
just
heated
for
the
new
community
clatters
and
our
new
features
of
communities
will
be
edited
to
a
castle.
So
if
you
are
running
cuneta
clutter
down
area,
you
should
use
a
cast
the
virtual
machine
set
and
the
standard
virtual
machine,
the
for
the
non
manager
the
way
to
draw
on
community.
B
So
if
your
cluster,
it's
also
running
over
shamanistic,
assess
them,
no
matter
which
deployment
to
use
are
you
using?
You
could
use
cluttered
scaler,
we
started
with
with
them
type
Statue
of
them,
as
I
said,
and
for
the
tender
devotion
machine.
It
is
only
spotted
with
SS
ancient
deployment
tools
listed,
because
we
need
to
figure
out
a
way
to
provision
a
new
virtual
machine
which
included
what
are
the
images
of
it
Museum
how
to
set
up
the
node
after
the
virtual
machine
is
created
so
wait.
B
We
need
to
integrate
such
thing
with
deployment
tools,
so
now
I
mean
SS.
Engine
is
coated
and
also
it
says,
engine
spotted
all
the
features
I
mentioned
here
so
such
as
the
feature
that
I
mentioned
about
a
bushel
minute
ago
says
the
availability
zones
a
today
skirts
and
that
time
dynamic
bottom
provisioning.
C
Thanks
Steven
hi
I'm
on
belay
hook
from
6a
light
and
we'll
be
discussing
briefly
the
changes
to
plugins
that
occurred
in
the
one
told
release
and
how
we'll
move
forward
in
future
releases.
So
next
slide,
please.
So,
since
the
11.8
release,
plugins
have
a
lot
of
users
to
extend
the
core
functionality
of
a
cue
control.
C
Second,
that
they
should
be
used
to
install
and
last
that
plugins
should
be
extendable
and
we'll
cover
what
this
means
in
depth
in
the
upcoming
slides
next
one
please
so.
First
plugins
should
be
easy
to
write.
This
means
that
plugins
should
be
able
to
be
written
in
any
language.
No
freaking
work
or
libraries
should
be
required,
and
all
flags
and
arguments
as
given
by
a
user
should
be
passed
untouched
to
a
plug-in
process.
This
means
there
is
no
filtering
there's
no,
you
know
basically
whole
back
of
any
flags
or
arguments
by
the
Q
potential
process.
C
This
means
that
plugins
choose
how
to
handle
user
inputs
and
parameters
themselves.
We
have
made
a
new
repo
under
their
current
ease
or
github
name,
CLA
runtime,
which
we
hope
assists
plug-in
authors,
but
like
in
development.
This
is
essentially
the
same
set
of
helpers
printers
and
configuration
handling
utilities
that
we
are
so
suits
in
Cube
control
commands.
C
There
is,
additionally,
a
plug-in
template
ingo
available
at
samples
that
the
sample
ceiling
plugin
repo
under
the
coordinators
order,
which
should
allow
you
to
essentially
start
off
at
a
well-defined
point
when
writing
plug-ins
and
go
and
again
to
reiterate
the
plugins
do
not
have
to
written
and
go,
but
should
you
choose
to?
We
have
provided
some
tools
for
you
next
slide,
please.
C
So
second
plugins
should
be
easy
to
install
is
pretty
much
means
that
there
should
be
no
plugins
specific
path.
Environment
variable
and
there
should
not
be
a
fixed
location
of
the
user's
machine
or
plugins
should
live
instead.
Plugins
under
the
new
design,
plugins
live
anywhere
on
a
user
spot.
Also,
there's
no
configuration
or
pre.
D
C
In
coming,
slides
and
last
as
part
of
an
installation
process,
there
should
be
no
external
tool
or
program
required
in
order
to
use
a
plug-in
with
your
control.
Next,
please
so
30
plugins
should
be
extendable,
but
we
can
even
maybe
this
is
that
it
should
be
easy
to
build
additional
functionality
on
top
of
the
new
mechanism.
So.
B
D
C
C
So
how
are
plugins
named
under
the
new
design?
A
plugin
is
nothing
more
than
an
executable
file
whose
file
name
begins
with
Q
control
and
it
lives
anywhere
on
a
user
staff.
So
in
this
example,
we
have
a
neck
securable
file
name
to
keep
control.
Who
am
I
which
will
be
invoked
whenever
a
user
types
to
control,
Who
am
I.
Mother
can
I
mine
commencement
miss
can
opt
you
to
use
an
underscore
as
part
of
their
plug-in
element
in
order
to
provide
a
cute
control
command
that
has
a
in
its
name.
C
C
As
part
of
this
update,
we
provided
a
handy
new
command
called
queue,
control
plug-in
list
which
allows
users
to
discover
all
plugins
available
to
keep
control.
Plugins
are
listed
and
executed
in
the
exact
same
worker
as
they
are
found
in
a
user
spot
and
plug-in
authors
are
able
to
obtain
warnings
and
errors
that
could
prevent
the
plug-in
from
being
executed
by
using
this
command.
So
a
few
examples
of
warnings
and
errors
provided
are
such
when
a
plugin
is
overshadowed
by
another.
C
Latter
will
never
be
executed
if
a
plug-in
post
name
is
keen,
but
it's
been
executable.
You'll
also
be
worried
about
this
and
last.
If
a
plugin
attempts
to
overwrite
a
core
to
control
command
such
as
version,
then
you'll
receive
a
warning
about
that
as
well.
Since,
as
of
right
now,
we
need
to
announce,
support
over
writing
or
extending
core
keep
control
commands,
there's
also
a
plug-in
executable
preference,
which
is
implicit
as
part
of
the
new
plug-in
design.
C
What
this
means
is
that,
given
two
plugins
with
almost
similar
names
under
a
directory
in
the
path,
the
longest
named
plug-in,
will
always
take
precedence
such
that
we
can
allow
it
plugins
to
behave
us
of
commands
of
others.
So
in
this
scenario,
if
a
user
were
to
invoke
the
command
to
keep
control,
prove
our
best
and
their
command
mine,
then
the
plug-in
file
name
with
or
the
plugins
cutable
with
the
longest
file
name
will
be
chosen.
C
C
To
this
new
design,
so,
although
plugins
can
extend
other
plugins,
it
is
currently
not
possible,
as
we've
discussed,
to
extend
core
queue
control
commands
such
as
version
there's,
no
additional
information
passed
from
the
queue
control
process
to
the
plugin
process.
An
example
of
this
is,
unlike
the
previous
design.
We
will
not
provide
plug-in
specific
environment
variables,
such
as
the
current
namespace,
to
have
our
variable.
C
D
C
D
C
Most
parts
invalidate
old
flags
and
arguments
passed
to
them.
Solo
queue
control
will
not
validate
or
parse
flags
for
you.
We
do
provide
a
seal
at
runtime
repo
which
allows
you
to
do
this
with
these
any
plugins
written
for
any
versions
before
one
12
must
be
migrated
and
will
network
with
newer
versions
as
well.
Next
slide,
please
so.
C
Of
plugins,
although
plugins
cannot
currently
extend
control
commands,
we
are
having
some
discussions
to
potentially
add
a
limited
support
for
this
in
the
future.
Some
key
questions
we
have
in
mind
are
whether
or
not
users
just
want
visibility.
In
the
first
place,
which
command
types
we
should
offer
as
extendable
command
paths
to
begin
with,
and
also
if
plugins
should
be
allowed
to
completely
override
existing
commands
or
simply.
D
D
A
B
Okay,
I
cut
thin.
The
answer
to
the
chat
chat
your
own,
so
first
question:
what
is
the
time
line
for
a
1.12
I
make
s
and
a
triple
minutes
ago
said
both
of
those
features
on
our
little
map.
Bush
commercial
disgustoid
be
available
earlier
December,
but
we
don't
have
a
name
detailed,
eight-year
it
and
also
a
wait
on
driver
and
detail
date.
Yet
for
one
point:
top
thanks
for
the
I'm
thankful
question.
A
C
C
So
I
don't
think
I
need
any
in
particular
come
to
mind
at
the
moment,
but
yes,
I've
run
across
a
few
that
are
pretty
useful
out
there.
So
if
you
on
github
it's
you
search
for
any
repos
with
the
queue
control,
plugins
tag,
you'll
find
a
lot
of
them
actually
that
are
written
to
work
with
the
new
zone,
some
of
them
offer
you
know
context.
A
C
A
A
A
B
B
B
A
list
without
a
ribosome
will
not
be
able
to
draw
the
nodal
with
different
prism
or
without
a
village,
so
I
the
we
are
no
than
to
the
dispossession
volume
I
created
a
type
naturally
via
storage
class.
So
when
the
storage
class
decided
to
create
the
the
potential,
it
should
note
first,
which
song
should
be
chosen
to
to
create
the
in
disk.
B
So
so
for
a
powerful
disc,
a
gesture.
Children
are
just
choosing
the
zones
from
the
list
up
nose,
so
each
node
will
have
a
labor,
always
the
don't.
It
is
Johnny.
So
then,
then
the
controller
manager
could
check
which
resume
are
available.
At
the
day
it
could
chose
one
from
from
the
listserv
and
created
the
in
disk
and
then
scheduler
to
the
dedicated
node.
A
Yeah,
so
that's
the
fun
I'm
yeah,
okay
right
so
I'm
gonna
answer
your
first
question:
if
you
check
out
the
kubernetes,
so
we
have
a
crouppen,
a
tease,
CSI
org
and
the
kubernetes
I
work
will
kind
of
show
you
a
lot
of
the
different
implementations
as
well
as
the
documentation.
So
the
the
link
that
I'm
sending
you
right
now
is
the
documentation
for
kubernetes
CSI.