►
From YouTube: Kubernetes sig-aws 20190503
Description
No description was provided for this meeting.
If this is YOUR meeting, an easy way to fix this is to add a description to your video, wherever mtngs.io found it (probably YouTube).
A
We
have
a
couple
of
items
on
the
agenda.
Please
do
add
things
to
the
agenda
if
you
would
like
to
get
to
them.
We
have
a
quick
session
of
caps
and
I.
Think
some
discussion
of
koukin
and
most
excitingly,
perhaps
just
start
us
up.
We
havin
a
dear
who
is
gonna,
give
us
a
demo
of
cluster
API,
our
native
master
engineering,
yeah.
A
B
So
William
literally
cut
a
release
in
the
last
hour
of
version.
Zero
point
two
point
one.
So
those
of
you
not
familiar
with
class
API
I'm
sure
you
heard
a
lot
about
it
from
us
so
last
couple
of
months,
so
so
what
class?
The
API
briefly
is
what
we
consider
to
be
a
sort
of
future
model
managing
communities
using
kubernetes.
We
love
giving
at
you
so
much,
but
given
a
teaser
new
communities,
so
a
declarative
way
to
build
clusters
and
then
do
the
ongoing
management
of
it.
B
B
Are
these
providers
for
each
type
of
cloud
or
on-premise
a
controller
or
operating
voice,
known
as
that
handles
things
like
machine
sets
which
ran
logs
for
replicas
sets
and
machine
deployments
which
analogs
for
deployments
and
see
vital
Steve
be
strap
clusters
from
from
this
from
the
ground
up,
which
is
what
I'm
going
to
be
showing
you
today
so
in
terms
of
so
we've
been
at
it
since
last
August?
Probably
so,
we've
done
couple
of
releases.
We
haven't
demo
antistick
a
doublets
with
by
a
while
one
things
is
we?
B
We
added
support
for
h8
control
planes,
so
we
you
can
build
with
a
multi
node
control,
plane
out
of
the
box
now
and
we
support
which
communities
113
and
month
on
point
14
and
in
the
last
release.
We've
done
quite
a
lot
of
work
on
reliability,
so
improving
the
reconciliation
and
that's
really
great
to
have
contributions
from
others
like
New,
Relic,
10
scale
Factory,
who
really
thought
improved
reliability
last
couple
of
months,
so
I
happen
personally
not
worked
on,
plus
they
probably
Douglas
for
wild.
B
B
But
I
will
just
show
you
what
we
got.
So
we
have
the
output
if
they're
Carfi
on
once
you
unzip
it,
you
have
a
bunch
of
templates,
so
cluster,
a
definition
of
what
a
cluster
is,
and
you
kind
of
just
give
it
a
cidr
blocks
for
pods
and
services,
which
region
and
what
SSH
key
that
you're
going
to
use
in
emergency
to
get
onto
a
box
definitions
of
what
machines
are,
and
this
provided
component
space
which
essentially
just
they
you
a
a
deployment
like
you
might
download
for
any
other
bit
of
software.
B
B
B
It
just
templates
out
the
the
yeah
more
templates
that
we
got
and
also
injects
my
credentials
as
a
secret.
So
we
need
this
because
normally
when
you
you
running
AWS
base
stuff
like
cops
you're
running
on
your
laptop
in
your
come
on,
come
online
with
that
environment.
So
it
note
it
can
resolve
your
a
SS
credentials.
But
what
we're
going
to
do
here
is
we
actually
run
a
kubernetes
cluster
and
we
use
coin
for
this,
which
is
the
way
running
communities
in
docker.
So
we
actually
need
to
take
the
secret.
B
Then
we
need
to
take
your
credentials
from
your
environment
and
put
them
as
a
community
secret
and
run
them
inside
this
bootstrap
cluster.
So
we
do
provide
a
warning
that
this
provider
components
that
yeah
model
does
have
some
a
SS
credentials
in
it,
so
not
going
to
show
you
that
come
show
you
those
bits.
B
So
we
have
our
cluster
just
template
it
out.
It's
just.
You
know
test
one
highly
imaginative.
It's
going
to
use
this
SSH
key
that
I've
uploaded
previously
so
I,
actually
nothing
that
we
employed
in
this
artifact
it
in
this
tar
file
is
a
getting
started.
Guide,
smart
down
file,
so
this
goes
through
the
bits
and
pieces
that
you
need
to
actually
get
up
and
running.
B
We
all
also
have
a
deployment,
so
we
can
actually
create
a
bunch
of
worker
nodes
and
have
them
managed
automatically
by
cluster
API,
and
this
Adams
thought
Yama
is
Adam
here,
it's
just
a
Cali
Cali
deployment.
So
one
of
the
things
we
want
to
try
and
do
is
like
the
agnostic
of
CNI.
You
can
choose
what
type
of
networking
you
want
to
bring
to
your
cluster.
We
like
it's,
not
the
job
of
cluster
April,
I,
guess
780
force
an
opinion
on
you,
as
so
we're
going
to
deploy
calico
as
part
of
this
deployment.
B
And
so
it's
one
really
quickly
because
I've
already,
where
it's
already
instantiated,
so
this
class,
a
SS,
a
VM,
looks
at
your
AWS
accounts
and
create
some
I
am
Paula.
Seeds
templates
our
CloudFormation
template
engine
just
runs
it.
So
it's
the
see
we
we
provide
policies
for
the
controllers
that
one
for
cluster
API
itself
and
we
also
do
ones
for
the
nodes
which
are
going
to
be
running
the
AWS
cloud
provider.
B
So
what
you
would
see
in
any
other
communities,
cluster
that's
running
on
AWS
and
we
create
some
IM
roles
and
users
that
you
can
that
I
used
to
bring
up
the
instances,
so
I've
pretty
much
ready
to
start
a
cluster.
So
what
I'm
going
to
do
here
is
so
this
is
all
in
the
getting
started
guy,
but
just
go
through
some
nice
smoke
and
mirrors
here.
So
we're
gonna
use
a
kind
as
our
big
trap
cluster,
so
it
will
default
to
mini
keith.
B
We're
gonna
use
kind
here,
I'm
also
going
to
say,
I'll
go
free
what
this
does
in
a
bit.
We
have
this
specify
this
provider
aw.
Yes,
for
now,
that's
just
quark
or
quest
console,
and
then
these
things
are
like
we're:
gonna
deploy
some
machines,
so
I'm
gonna
point
at
the
machines
yeah
mall,
someone
got
a
point
at
this
height
H
a
control
plane
across
the
definition.
I
would
show
you
earlier
the
provider
components
which
is
just
the
deployment
to
get
cluster
API
AWS
running
and
add-ons,
which
is
the
Calico.
B
B
Okay,
so
plus
the
cutter
has
some
logs,
but
what
one
of
the
UI
issues
that
we
have
been
thrust
the
API
is
is
using
cluster
cutter
on
its
own.
It
doesn't
tell
you
a
lot
about
what's
actually
happening.
If
you
really
want
the
details
of
what's
going
on,
you
actually
need
to
look
at
the
logs
on
on
the
coin
cluster.
So
I
just
pointed
the
cube
config
at
kind
and
I'm
literally
just
gonna
get
the
logs
for
that
sort
and
see
that
in
real
time
set
it
can't
its
oh
yeah.
A
This
is
really
nice
and
you
I
really
like
the
the
way
that
you
have
wisely
from
my
experience,
put
two
calico,
for
example,
into
that
add-ons
thing:
I
feel
like
it
sort
of
bounds.
Look
problem
quite
nicely
right,
so
this
is
really
just
about
the
cluster
and
then
we
can
plug
into
the
other
efforts
elsewhere
through
that
mechanism.
Yeah.
B
Ideally
I
mean
is,
there
are
still
some
calico
bits
encoded
in
the
cluster,
so,
for
instance,
you
need
certain
security
group
rules
to
get
calico
to
work.
You
need
certain
ports.
Open
said
it
should
work
for
other
CNAs,
but
it
wanted
to
sort
of
rough
edges.
Is
that
yeah
we
specifically
put
stop
to
get
calico
working
into
the
cluster.
We.
A
B
Add-Ons
doesn't
really
get
you
all
the
way
there,
so
so
one
of
the
things
so
one
of
the
things
that
is
happening
right
now
is
that
we're
actually
going
through
a
rethink
of
what
cross
the
API
is.
So
one
thing
knows
so
that
cluster
object
I
showed
earlier
is
actually
what
that's
actually
doing
is
creating
a
VP
seed,
subnet,
Internet
gateways,
nat
gateways,
a
load
of
sort
of
networking
concerns
that
you
might
not
know
about,
and
we
have.
B
How
do
you
do
upgrades?
For
instance,
we
have
one
on
a
control,
plane
life
science
cycle.
So
what
are
the
various
ways
in
which
you
can
build
a
control
plane
data
model
which
is
going
back
to
these
cluster
machine
objects
about
thinking
about
redesigning
noise?
So
each
of
these
is
a
work
stream
that
has
its
own
meetings.
So
you
come
to
this
page.
This
work
strings
meeting.
We've
got
a
minutes
for
all
of
these
areas,
so
if
you're
interested
in
working
on
across
the
API
now
is
a
really
good
time
to
get
involved.
B
Class
api's
falls
on
the
sick,
clustered
lifecycle.
So
it's
a
lot
what
wide-ranging
effort
it's
not
just
me
and
where
it's
everyone's
involved.
So
if
you,
if
there's
something
that
you
haven't
had
before
from
other
tooling-
and
you
want
to
see
made
available
now
like
come
and
give
us
your
opinions
right
now
so
given
up,
let's
see
where
we
have
gone
to
you,
so
all
right,
so
I'm
just
pull
pulled
up
the
locks
for
this
plus
there
so
and
see
it's
creating
this.
It's
got
free
to
do
in
this
acute
groups,
and
we
can
see.
B
There's
some
hard-coded,
calico
stuff
right
now.
I
PMP
can
occur
so
what's
happening
is
that
we
have
like
two
loops
that
go
around
one
tool
to
come
closer
another
one
for
the
machines,
the
machine
one
keeps
me
trying
and
failing
because
bits
of
the
network
aren't
now
which,
but
that
is
fine.
The
standard,
so
kubernetes
model
keep
retrying
until
until
you
get
to
the
state
that
you
want
to
end
up
in
so.
B
Yeah
yeah
so
right
now
the
class
actuators
just
creating
a
Bastian
incidents
so
from
on
it
get
onto
that
control,
plane
in
cases
and
emergency
or
something's
gone
wrong.
Then
I
can
break
through
you
on
to
one
small
instance
and
then
get
through
you
on
to
the
rest
of
cluster.
So
yeah
we
deploy
control
plane
into
private
subnets
as
well,
so
their
access
to
the
cluster
happens
through
load
balancer.
That's
also
graded
by
the
this
cluster
controller,
which
we
call
actuators
for
historical
reasons
and
cuz.
The
API.
A
B
C
B
Absolutely
so
yeah
the
clustered
cotton
workflow
is
what
we
call
a
pivoting
worked
very
so
you'd
start
with
across
that's
running
a
set
of
controllers.
That's
running
in
your
kind
instance.
Once
that
clusters
created
the
objects,
then
move
across
into
the
created
cluster
and
the
kind
class
they
switched
off,
but
actually
most
people
don't
want
to
run
it
like
that.
What
they
really
want
is
a
management
cluster.
B
So
there's
nothing
to
stop
you
taking
the
provider
components
as
is
and
running
out
on
a
cluster,
and
then
you
can
actually
use
coop
cuttle,
because
these
are
just
CRD
objects.
So
you
can
apply
these
machines
and
clusters
with
cube
castle
and
just
let
happen
that
way.
There's
I
think
that's
one
of
the
things
that's
confusing.
We
always
talk
about
the
classical
workflow,
because
that's
how
you
get
anywhere
with
class,
they
join
start
with
variety
majority
cases.
You're
gonna,
probably
run
in
management
plus
they're
cool.
B
A
B
B
B
A
Think
maybe
you
should
come
back
in
a
minute
yeah,
let's
come
back
in
a
little
bit
actually
because
you
I
think
you've
nicely
Thank
You
vanessi
Twp.
The
next
item
on
the
agenda,
which
is
another
might
put
on,
which
is
talking
about
yeah.
You
wanna,
stop
sharing
your
screen
at
your
yeah,
perfect
I.
Thank
you,
which
is
an
item
about
our
caps.
We
have
one
significant
cap
that
is
go
or
one
cap
that's
going
in
to
the
next
release,
which
is
from
moonfish
Adam.
A
If
you
prefer
to
go
to
buy
moonfish
or
not,
but
yeah
I'm
in
fish.
But
it's
around
you
want
tell
us
what
it's
about
and
then
we
can.
The
thing
I
really
wanted
to
discuss
was
the
I
think
the
promotion
is
great
and
thank
you
so
much
I
think
the
the
previous
limitations
of
vieille
of
NLB
and
whether
they
have
been
addressed
around
hair,
pinning,
for
example,
so
yeah.
D
I
have
proposed
to
graduate
be
from
our
back
to
pat
you
know
and
or
15,
because
the
fisher
has
already
been
stabilized
since
its
app'
and
has
been
staying
up
for
more
than
up
five
virgins
and
were
sold.
It
has
now
had
beautiful,
back
fixes
and
support.
Cortona
Benson
and
s.
Apology
is
terminating
at
front,
were
either
at
Becca.
It's
time
to
switch
and
I
sing
a
Chris.
We
are
going
to
add
in
the
ueb
support,
so
we
should
be
fine
for
the
weather
to
our
to
better.
That's.
C
A
Behavior
like
it
seemed
like
previously,
you
had
to
create
a
three
P's
one
in
each
zone
and
it
wasn't
clear
what
happened
to
one
of
those
three
zones
failed
and
then,
as
a
nadir
just
mentioned,
the
hair
pinning
behavior
I,
don't
know
if
you're
familiar
with
either
of
those
with
a
lack
of
hair,
pinning
I
don't
know
if
you're
familiar
with
out
of
those
limitations
and
whether
they
are
have
been
addressed.
I
I.
A
A
D
A
And
like
yeah,
well,
why
don't
I
want
to
open
an
issue
on
whether
the
like
we
can
open
issue,
discuss
those
two
issues
and
then
we
can
close
them
out
hopefully,
and
then
that
will
be
a
good
like
discussion
point
for
everyone.
So
we
want
to
keep
talking
about
it
as
it
were,
but
I
think
every
out
be
great
and
we
also
have
to
figure
out
how
we
the
right
mechanism
for
updating
people
that
have
existing
services
of
type
load
balancer.
Let
the.
D
A
C
C
Yeah
so
there's
a
this
came
out
this
morning,
so
Ilya
started
a
thread
on
discuss
about
trying
to
get
people
together.
Do
like
a
birds
of
feather
around
eks
cops,
anything
related
to
AWS
at
mesh,
so
on
and
so
forth.
I
dropped
a
link
in
the
invite.
If
you're
interested
put
a
note
on
there
as
well,
I
dropped
another
link
at
the
bottom.
C
A
A
A
B
A
E
A
I
mean
we
have
a
so
there
is
a
sub-project
sequencer
life
cycle,
looking
it
using
operators
or
other
technologies
like
it
to
better
manage
add-ons.
It
feels
like
if
you're
talking
about
the
that
work
is
still
in
early
days.
We
have.
Some
that
happens
are
in
the
sync
cluster
life
cycle,
which
is
Tuesday
at
9,
Pacific,
12,
Eastern
or
I.
Don't
you
look
times
about
nine
Pacific,
but
in
the
alternate
weeks
from
sync
cluster
lifecycle
itself
and
still
early,
but
I
would
imagine
that
the
way
we'd
integrate
with
cluster
API
is.
A
You
would
create
in
that
add-ons
that?
Well,
you
would
create
the
operator
and
you
would
create
a
CRD
to
describe
what
you
wanted
to
install.
So
in
this
case
you
would
install
the
Calico
operator
anywhere
install.
You
would
specify
a
calico
c.
Rd
presume
you'd
also
have
to
specify
the
the
definition
of
the
cid
itself
or
the
CR
itself
as
well
like
Alec,
and
then
the
operator
would
come
up
and
install
calico
and
let
you
manage
and
upgrade
that
I,
don't
think
it
would
address.
A
E
A
The
add-ons
shouldn't
be
a
separate
tool,
I
think
I.
Think
the
the
idea
of
add-ons
is
that
I,
don't
wanna
speak
for
everyone,
but
I
feel
like
we're
all
pretty
agreed
that
we
want
a
declarative
way
of
managing
or
a
declarative
way
of
managing
add-ons
in
in
kubernetes,
so
by
a
coop
cutter
or
whatever
you
want,
makes
sense,
I
think
or
makes
more
sense
than
the
current
approach,
which
is
well
in
coo
bop.
A
It's
a
bash
script,
which
reads
from
direct
in
cops
that
add-ons
are
sort
of
baked
into
the
cops
binary
I
like
what
cluster
cuddle
has
done
where
they
or
it's
a
sort
of
external
file.
It
sounds
like
it's
still
shipped
with
the
with
the
release
so
something
there,
but
yeah
I,
think
I.
Think
so
I
said
effect.
Anyone
want
to
comment,
or
anyone
else
have
any
views.
E
Yeah
cool
that
is
really
useful.
The
reason
I'm
bringing
up
is
because
I'm
many
more
kind
of
CSI
drivers,
because
we're
gonna
have
already
have
three
drivers
now
I
know
it
might
have
other
more
and
more
items
being
added
into
the
cluster
EPS.
Are
you
carrying
specific?
So
that's
when
mine
is
starting
up
like
SRO
add-on
management
story,
yeah.
C
A
Thank
you,
I
hope
self.
The
I
think
the
I
think
the
nice
thing
about
the
other
operator
is
that
fundamentally
we're
saying
that
you're
gonna
specify
what
you're
doing
in
a
in
a
llamo
file.
So
if
it
turns
out
that
you're
not
gonna
use
operators
at
all,
if
you
want
to
just
directly
install
like,
as
the
cluster
API
demo
directly
installed
Calico,
if
it
turns
out
you
want
directly
in
siliceous
our
drivers,
that's
fine!
It's
not
a
big
change
right!
A
You
just
directly
embed
the
manifest
instead
of
having
an
operator
which
installs
the
manifest
I
think
that
the
key
reason
to
introduce
the
operator
or
a
key
reason
to
deduce
the
operator
is
when
the
process
of
installation
is
a
little
bit
more
complicated
than
just
a
coop
that'll
apply,
and
that
might
be
the
case
for
CSI.
So
yeah,
please
do
join
the
the
meetings.
We'd
love
to
have
you
alright,
probably.
A
Otherwise
I
will
say
thank
you
very
much
to
nadir
and
to
everyone
else
for
joining
and
see.
Hopefully,
all
of
you
in
coop
Connie,
you
I
think
we
will
still
have
a
meeting
next
Friday,
because
I
presume
not
everyone
will
be
traveling
to
Barcelona
on
that
Friday
I
might
be
so
I'm,
hoping
that
she
can
do
it
I
think
Nishio
so
volunteered
to
do
a
demo
of
Pulu
me
next
time.