►
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
Deep Dive: Azure SIG – Khaled (Kal) Henidak , Microsoft & Stephen Augustus, Red Hat
In the SIG Azure Deep Dive, we’re going to tell you all about why SIG Azure exists and the team behind managing it. From there, we’ll talk about what’s happened over the last few releases, Kubernetes 1.11 and Kubernetes 1.12, as well as some of what’s planned for Kubernetes 1.13. Additionally, we’ll glimpse into Azure support of multiple extension points e.g., CSI, CNI, CRI. Afterwards, we’ll have a quick demo of provisioning Kubernetes on Azure as well as a secret demo from our mad scientist and SIG Azure Technical Lead, Kal! Finally, and most importantly, we’ll talk all about some of the best ways to get involved with SIG Azure and all of the forums that we’re communicating on!
To learn more: https://sched.co/Gret
A
Right,
hello,
hello,
everyone
thank
you
for
coming
to
the
signature
deep
dive.
If
you
were
in
the
if
you're
in
the
signature
intro
yesterday,
you
had
an
opportunity
to
ask
a
lot
of
questions.
We
did
skipped
straight
to
the
questions
section,
because
Azure
is
we're
one
of
those
things
that
we
were
actually
trying
to
figure
out
more
and
more
about
the
kind
of
the
community,
the
the
user
base
and
and
what
what
use
cases
we
need
to
kind
of
attract
or
I
kind
of
understand
a
little
better.
A
All
right
so
straight
into
it.
What
have
we
been
working
on
so
we've
been
making
incremental
improvements
to
the
azure
code
base
on
kubernetes
one
around
pulling
the
pulling
the
in
triage
errcode
out
of
tree
moving
that
to
cloud
provider
specific
repos,
so
that's
kind
of
what's
been
happening
across
the
board
for
all
of
the
cloud
providers.
Aws
has
been
working
on
this
IBM
cloud
has
been
working
on
this
GCP,
so
we're
starting
to
see
this.
A
This
breaking
down
of
the
monolith
for
across
Cooper
Nettie's,
starting
to
use
starting
to
use
the
kubernetes
cigs
repo
more
frequently
so
we're
we're
broken
up
into
kubernetes
cigs
cuber
Native
Client
kubernetes
security,
which
is
a
secret
repo
and
and
and
a
few
others
that
SIG's
can
can
start
placing
their
code
in.
So
we're
we're
seeing
the
shift
away
from
away
from
the
day-to-day.
How
can
we
shove
everything
into
core
kubernetes
and
more
towards?
A
So
availability
zones
again,
if
you're
familiar
with
running
running
your
systems
in
multiple
regions
across
across
again,
we
have
to
now
just
do
to
the
AWS
constructive
availability
zones
with
the
addition
of
of
having
fault
and
zone
domains.
So
there
support
for
that
in
kubernetes
as
well.
I've
been
working
on
that
since
111
and
I
think
112
112
and
we're
set
to
wrap
in
114.
A
So,
as
I
was
mentioning
cluster
autoscaler
so
now
that
we
have
essentially
shifted
away
from
the
VM
AAS
or
virtual
virtual
machine
availability
sets
for
Azure,
we
want
to
make
sure
that
we
can
use
the
same
contracts
that
people
use
across
different
cloud
providers
to
be
able
to
scale
just
to
scale
your
clusters.
So
that's
that's
baked
in
as
the
the
vmss
support
1ga
right.
So
you
want
to
talk
a
little
bit
about
KMS.
B
C
B
We
went
ahead
and
we
built
an
integration
for
this
with
keyboard.
What
that
means
to
you
is
your
your
data.
You
can
configure
whatever
data
you
want,
so
you
can
say
crypt
my
secrets
or
encrypt
my
deployments
or
encrypt
my
config
maps
or
encrypt
everything,
and
that
data
is
encrypted
with
a
key
that
saved
in
a
dirty
book.
For
those
who
are
working
was
enterprises
such
as
banks,
highly
regulated
stuff?
B
B
B
This
is
the
answer
for,
if
you
want
to
do
snowflakes
implementation,
one
of
the
folks
I
talked
to
is
they
do
machine
learning
break
and
you
want
to
do
right.
Keep
your
cluster
we're
gonna
run
something
in
your
cluster
and
then
what
happened
is
that
cluster
will
span
B
cereus
VMs
for
ml
workloads.
So
it's
a
very
nice
use
case
for
trying
small
step
by
small
step
to
do
it.
There
is
a
bunch
of
repos
random,
rebo's
somewhere.
B
A
I
think
I
think
it
opens
the
door
for
is
kind
of
in
reviewing
the
proposal
for
cross
resource
group
nodes.
They
I
love
that
the
the
labels
and
annotations
kind
of
way
that
we
structure
the
code
opens
the
door
for
being
able
to
do
one
resources
across
multiple
Azure
accounts
right
across
multiple
Azure
accounts
about
across
multiple
regions.
Heck.
Why
not
start
connecting?
Why
not
start
connecting
cloud
providers?
Why
not
run
things
on
Prem
right?
B
Came
and
talked
with
us
and
with
me
personally,
and
they
said
you
know
what
we
love
everything
we
just
have
a
bunch
of
machines
on
Prem
and
here
on
workload,
that's
arcane
or
archaic,
and
we
just
don't
want
to
move
it,
we're
fine
with
it.
We
just
want
to
control
it
with
a
cloud,
and
this
kind
of
approach
is
what
what
we're
trying
to
solve
for
right,
we're
midway.
B
A
So
what
are
we
I
mean?
What
are
we
planning
for
the
the
next
few
cycles?
From
the
kubernetes
perspective,
we've
got
again
that
that
kind
of
migration
of
the
the
in
tree
provider
code
to
out
of
out
of
core
kubernetes
out
of
kubernetes
kubernetes
repos.
So
that's
continuing
along
that's
been
going
pretty
well,
I
think
at
some
point
we
will
start
looking
at
a
cluster
API
for
for
a
sure,
if
you're
worth
that
project
we're
all
right
so
yeah,
and
so
these
are
strictly
kubernetes
specific
updates.
I'm
sure
Cal
has
some
secret
Microsoft
sauce.
A
But,
as
always,
doc,
sir
would
help
us
iterate
over
code
a
lot
faster.
So
we
want
to
make
sure
that
the
docs
are
easy
to
read
easier
to
collaborate
with
people.
We
want
to
make
sure
that
the
testing
Suites
that
we
were
using
for
Azure
one
they
they
meet
them.
They
meet
the
bar
that
we
use
across
the
board
in
kubernetes
kubernetes.
So
that
means
adding
more
testing
to
to
test
grid,
which
is
the
urban
eddies
overarching
testing
platform,
as
well
as
starting
to
explore
this
idea
of
what
does
kubernetes
conformance
look
like.
A
A
And
then,
finally,
so
in
the
you
know,
in
the
spirit
of
conformance
and
collaboration
and
community
and
making
sure
we
have
consistent
consistent
guidelines
for
the
way
that
we
structure
kubernetes
and
deliver
kubernetes
code
on
cloud
providers,
there
are
we
spun
up
a
sig
cloud
provider
and
sig
cloud
provider.
It
has
a
quorum
of
people
from
multiple
cloud
providers,
AWS
AWS,
Microsoft,
GCP,
IBM
cloud,
so
on
and
so
forth.
So
what
we
want
to
do
is
move
sig,
asier
and
kind
of
slot
that
under
sync
cloud
provider
as
a
sub-project
as
opposed
to
a
sig.
A
A
The
sub
projects
are
essentially
delegated
delegated
responsibility
over
code
ownership
right.
So
a
sub
project,
like
a
sure,
would
control
the
code
around
the
external
controller
manager,
the
cloud
provider
of
the
cluster
API
and
and
anything
associated
with
Azure.
So
you
want
to
make
sure
that
we're
following
that
spirit
of
community
and
that
spirit
of
conformance
by
by
kind
of
bending
the
knee
right,
so
any
G
ot
fans
out
there
so
now
Cal
is
going
to
take
it
away
with
a
demo.
So.
B
B
B
A
B
This
all
right,
so
it's
a
cluster,
it's
fully
functioning
and
everything
the
one
catch
here
is
this
cluster.
If
I
go
if
you're
following
what
we
do-
and
this
is
manifests
and
then
I
go
cube,
API
server!
This
is
really
the
manifest
that
runs
in
API
server.
The
one
catch
here
is
this
guy.
If
I
can
see
it,
I
will
point
it
out
because
I
moved
it
so
this
guy,
it
see
these
servers,
equal,
HTTP,
karna
cat
CD,
and
this
endpoint
is
actually
not
in
cluster
this.
This
is
a
master
server.
B
B
What
does
this
mean
to
you?
What
does
this
mean
to
people
who
break
clusters,
especially
a
large
sum
of
clusters,
as
you
have
zero
operational
data
store,
if
any
of
you
guys
did
the
stuff
like
try
to
scale
up
masters,
go
let's
say
to
masters
or
upgrade
masters.
It's
always
a
highly
choreographed
tons
of
stop
this,
which
was
the
state
to
migrate,
moves
to
the
second
thing,
and
it's
usually
a
process
that
has
a
lot
of
a
lot
of
code,
sweat
that
the
person.
B
Is
basically
zero
operation
just
point
to
the
data
set,
and
so
also
the
good
news
is,
if
you
go
to
today
to
aks
engine
which
is
the
new
home
for
ACS
engine.
Don't
ask
me
why
with
it,
because
I
don't
know
why
the
we
rename
it
and
then
you
go
to
examples
and
then
all
you
need
to
do.
Is
you
go
to
the
this
one?
And
if
you
go
to
this
example,
you
will
see
that
it
has
only
one
flag.
B
B
E
D
B
E
B
You
need
to
do.
Is
you
go
through
the
arm
templates?
Oh
yeah,
you
go
through
the
arm
templates
and
there
was
up
there
zapped
by
the
arm
resource
of
type
cosmos
HDD.
But
there
is
a
white
listing
process
because
it's
a
very
lockdown
private
preview
to
a
very
selected
number
of
customers,
because
we
want
to
progress
this
slow
and
steady
because
it's
data
and
the
person
who,
by
the
way,
build
a
CI
CD
pipeline.
That
makes
sure
that
this
end
point
runs.
Is
this
guy
this
guy
then
his
name
is
Ernest?
B
E
B
E
B
That
that
data,
the
data
source,
is
actually
net
CD
data
source.
So
what
you
need
to
change
is
the
change.
I
showed
you
on
the
manifest
file
and
that's
it
that
Cuba
API
server
manifest
so
the
way
we're
on
Cooper
netis
is
there
is
a
thing
called
masters
and
then
they're
saying
call
nodes
right,
the
master.
They
know
the
masters.
There
is
a
place
where,
as
featuring
side,
Cooper
nets
called
static
manifests.
B
Yes,
this
is
what
most
or,
if
not
all,
used
to
boot,
up
to
bootstrap
a
cluster
you're
on
a
cubelet
and
you
point
cubelet
to
a
location
that
has
sort
of
manifests.
Each
manifest
is
a
static
pot,
one
of
those
pots,
our
API
server,
the
Transvaal
masters.
So
the
only
change
is
literally
three
flags.
You
need
to
change
the
certs
and
the
endpoint
you
all
right.
That's
it
that
changes
by
the
way
emerged
in
a
KS
engine.
You
can
actually
see
the
change.
Oh,
don't
do
it
city,
local,
because
I
just
assume
that.
F
B
B
B
I
do
all
right,
so
I
can
actually
show
you
a
simple
thing.
This
is
a
TC
d,
CTL
command,
so
yeah.
So
let's
see
the
something
was
cert.
Something
was
key,
something
within
point
pointing
to
Karnak
I'm
Egyptian.
So
this
is
the
temple
in
Egypt.
So
sorry,
I
had
to
name
it
stuff,
I
know
and
just
a
command
to
get
stuff.
So
all
right,
so
that's
basically
means
you
don't
need
any
tooling
investment
in
client-side
tool.
You
gonna
say
an
uchi.
There
is
no
upstream
changing
copper,
Netta's
openshift,
core
OS
should
run
as
this.
G
B
G
B
Right
you
talk
about
agile,
kubernetes
storage,
the
reboot.
We
have
yeah
right,
so
we
started
with
the
entry
provider.
The
story
goes
like
this.
Two
and
a
half
years
ago
we
had
the
blob
disk,
which
was
the
oldest
dyskinesias,
the
performance
of
that
disk.
I.
Think
people
can
tell
stories
the
natural
became
the
managed
disk,
which
is
easier
to
create,
attach
detach
and
a
lot
faster
was
the
premium
disk,
and
then
now
we
have
ultra
disks
ultra
SSD
disk
coaches
on
the
sixty
thousand
IO
per
second,
which
is
a
crazy
number.
It's
crazy
prices.
B
B
Tile,
the
up
and
down,
but
the
top
is
down
to
sixty
thousand
dollars
per
second
assume
Mobley
preview
orgy
recently
yeah
yeah
yeah
and
then
because
of
the
storage.
The
cloud
moves
at
the
speed
of
light.
If
you
want
moves
really
fast,
then
side
by
side,
we
were
doing
stuff
like
the
key
vault
flex
volume
so
give
it
like
keyboard.
You
attack,
you,
attach
your
keyboard
objects
as
if
they
are
disks
file
on
disks
and
we've
done
the
investment
in
all
right.
G
A
Just
Jade
yeah,
just
like
two
weeks
ago,
yeah
so
they've,
been
working
across
I,
think
the
last
five
five
or
so
release
cycles
to
get
it
GA,
I
think
I
think
the
next
improvement
that
we're
gonna
see
and
it's
not
so
much
an
improvement
of
disk
types
within
a
or
specifically
but
it'll,
be
the
improvements
to
topology
aware
disk
provisioning.
So
the
idea
that
you
know,
if
you
spin
up
a
cluster
in
you
know
they
spin
up
a
cluster
and
region
bla
that
your
disks
will
come
up
in
region
blah
as
well
right.
B
A
So
it
depends
we're
so
I've
also
run
Zig
release
we're
working
on
on
firming
up
what
the
patch
release
schedule
should
look
like
right.
Now,
we're
doing
right
now,
one
we're
creating
a
team
around
patch
release
and
release
engineering,
and
so
right
now
it's
it's
one
person
right.
So
whoever
gets
selected
for
the
release
team
is
basically
on
the
hook
for
nine
months,
maybe
nine
months
plus
to
provide
patches
and
be
available
to
to
do
this
stuff.
So
that's
that's
not
really
fair
and-
and
you
can't
really
calculate
when
people
are
gonna
be
available.
A
B
Having
that
in
mind,
plans
to
a
question
we're
moving
to
a
system
model
where
everything
is
outside
outside
of
the
tree
and
I
can
assure
you
that
the
key
vault
specifically
equivocal.
That
decision
was
just
looked
like
down
half
an
hour
ago
that
next
version
of
it
is
gonna,
walk
away
from
flexible
right.
So,
as
you
do,
and
then
most
of
the
stuff
eventually
will
flex
fall,
will
stop
all
right
and
use
with
CSI.
For
now,
if
you're
trying
to
make
the
choice,
use
the
the
entry
and
as
we
move
to
the
story,
yeah.
A
B
A
So
we
can
absolutely
continue
questions,
but
I
want
to
just
make
sure
that
you
know
how
to
reach
out
to
us.
So
again
the
chairs
of
the
cig
or
myself
and
Dave's
trouble
over
here.
The
technical
aids
are
Cal
and
thanks
a
and
the
if
anyone
wants
to
start
taking
pictures-
and
here
we
go
here-
we
go
of
the
the
the
home
page.
So
that
is
the
the
kubernetes
community
website.
A
A
So
all
channels
on
kubernetes
at
slack,
are,
are
absolutely
free
and
open
in
public.
The
only
channels
that
we
I
mean
there's
it's
a
community
policy
to
make
sure
that
all
of
communication
around
SIG's
and
working
groups
and
committees
happen
out
in
the
open.
So
with
that
said,
yes,
our
channel
is
open.
Once
you
joined
kubernetes
slack,
you
can
join
any
of
the
channels
that
you
want
what's
really
great
about
joining
our
channel.
Specifically.
Is
that
a
lot
of
the
people
in
this
room
who
are
from
Microsoft
live
in
that
channel
alright?
A
B
B
So
one
two
three:
let
me
answer
this
question.
First,
I
about
few
months
back
I
found
out
that
time
get
a
lot
of
distracted
because
of
slack
I'm,
also
not
on
just
this
one.
There
is
an
internal
slack,
then
there
is
teams
and
then
there
is
Twitter
and
there
is
email
and
there
is
phone
and
text
and
Santa.
So
I
decided
to
just
let
down
all
my
connection
down
to
email
and
text.
He.
B
D
Just
an
opinion
question
a
lot
of
new
ways
for
deploying
in
kubernetes
apps,
as
well
as
as
your
native
systems
like
a
Postgres
database
or
a
sequel,
server.
I
know
like
the
cloud
native
app
bundles
was
pretty
new
and
then
also
the
Service.
Catalog
is
also
relatively
new
ish,
but
it
doesn't
contain
everything
and
just
kind
of
curious
thoughts
on
the
best
way
to
tie
those
together
or
just
you
know,.
A
So
I'll
my
semi
opinionated
the
thing
about
service
catalogs
and
and
see
nav
and
whatnot.
It's
I
think
that's
great
I!
Think
that
we're
having
improvements
to
the
the
community
and
the
way
that
you
natively
deploy
applications,
but
I
also
think
that
it's
important
to
understand
what
your
use
cases
are
and
not
pick
up
the
next
shiny
tool
and
and
consider
using
it
too
I
personally
I'll
be
using
scene
Avenue
anytime
soon,
I
think
it's
it's
great
to
try
out
and
provide
feedback
for
different
tools.
B
My
opinion
has
always
been
the
same
for
the
last
22
years.
Use
what
works
for
you
don't
worry
about
what
others
use
right
if
there
is
other,
as
others
are
using
something
you
can
learn
from
use
it.
If
it
doesn't,
work
for
you
is
gonna
have
to
use.
I
have
seen
so
many
shapes
and
colors
of
tooling
and
scripts
and
Emmas
eyes,
and
whatever
works
right.
You
have
to
take.
My
primary
requirement
is
always
you
as
users
of
my
stuff,
the
stuff
I
built
and
the
customers
that
use
yourself.
H
Can
I
ask
a
second
question,
yeah
sure
sure
Jeff
several
but
I'll
just
try
to
one
big
window
shop.
We
have
legacy
apps
right,
well,
run
them
in
kubernetes.
So
if
like
right
now,
I
wanted
to
spin
up
like
a
new
cluster,
to
support
like
Windows
application
right,
which
version
of
kubernetes
would
I
choose.
B
A
B
B
B
B
Ci,
prohibitive
and
trust
me
I'm,
one
of
those
guys
are
saying
we
need
to
give
the
people
more
flexibility
around
cost,
but
we'll
get
there
I'm,
not
the
person
to
talk
Costa
eleven,
that's
the
first
thing.
The
second
thing
is
the
good
news
for
you
is
everybody
just
wake
up,
I
think
four
months
five
months
ago
and
said,
windows
and
I
mean
everybody.
You
wouldn't
even
imagine
that
they
want
Windows
I'm
talking
Google
I
mean.
B
One
of
the
other
things
I
will
always
say
is
there
are
a
bunch
of
things
coming
in
Windows
pipeline
I,
just
don't
know
if
they
are
disclosed
or
not.
That's
like
end
of
2019.
That's
gonna
make
even
the
things
that
you're
talking
about
specifically
legacy,
workflows,
Kham,
Kham,
plus,
all
of
that
nice
stuff
that
we
used
to
do
is.
C
B
B
H
Production
lewis,
linux
cluster,
where
you
know,
because
again
we
were
microsoft.
Sharp
with
you
know
when
cycle
windows
is
okay,
so
that's
where
we
have
a
students
on
right
now,
but
we're
you
don't
want
to
move
them
to
companies.
So
we
have
a
new.
You
know,
architecture
that
we
have
now,
but
what
we
having
is
a
lotta.
We
have
intermittent
timeout
issues
to
both
as
your
sequel
and
to
as
your
at
us.
B
This
is
the
same
discussion.
We
had
right,
yeah,
so
out
of
like
out
on
top
of
my
head.
I
cannot
tell
you
what
the
solution
is
X,
because
I
don't
see
things
right
and
what
we've
discussed
for
those
who
have
networking
generally
speaking
issues.
The
first
thing:
is
you
walk
the
path
of
these?
Do
you
have
anything
in
the
middle
that
says
no
to
you
all
right?
What
firewalls
IP
tables
all
the
funkiness
that
happens
there?
You
need
to
look
at
it.
The
second
thing
is
the:
what
else
did
we
say?
B
No,
no!
What
else
yesterday
we
talked
about
we've
talked
about
this,
so
one
of
the
things,
one
of
the
advice
and
I,
truly
believe
that
this
is
the
valid
advice
is
generally
spear
respective
of
where
you
want,
if
you're
building
hyper
systems,
if
you
have
a
pot
or
a
container
pot
that
doesn't
need
to
go
to
cube
DNS
or
gudinas
just.
B
B
E
E
B
Don't
don't
take
that
tree
yeah?
The
other
thing
that
I've
seen
people
also
suffer
from
is
even
in
a
small
cluster.
If
you
have-
and
it
depends
really
on
the
kind
of
APR
using
the
kind
of
framework.
So
Java
is
not
DNS
friendly
right
that
Network
yeah,
not
net
code
event,
is
not
DNS
rather
easy.
The
way
they
do
DNS
queries
and
then
gilepsy
versus
the
the
other
thing
muscle
were
yeah.
It's
it's
crazy
stuff
over
over
there.
So
while
you.
B
B
F
B
Aware
of
any
effort
just
happy
thing,
I
am
what
I
can
assure
you
is.
Our
preference
is
not
going
to
be
building
a
thing.
Our
own,
that's
gonna,
be
using
something
that
the
community
came
to
a
consensus,
that
this
is
where
we
want
to
invest
and
I.
Think
we
proved
with
proven
that
was
policy.
We
were
just
moving
policy
controller
stuff
away
from
agile,
because
we
really
want
this
to
be
a.
A
Concern
yeah
exactly
so:
I
expect
to
see
a
lot
of
service
providers.
Community
service
providers
start
to
build
tools
around
metering
and
charge
back
across
multiple
cloud
providers.
I
think
well,
we'll
see
eventually
is
it
will
be
an
API
on
top
of
providers
and
not
a
provider
provided
service
itself.
Thank
you.
Thank
you
all
for
coming
and
thank
you
all
for
hanging
out
extra
long.