►
From YouTube: Kubernetes - AWS Provider - Meeting 20210820
Description
Recording of the AWS Provider subproject meeting held on 20201211
New release, plan to move to beta CCM, removing calls to IMDS from CCM, discuss https://github.com/kubernetes/cloud-provider-aws/pull/248
A
All
right
started
the
recording,
welcome
everybody
to
provider
aws
meeting.
It
is
august
20th,
please
be
mindful
of
the
cncf
code
of
conduct
and
let's
go
ahead
and
get
started.
I
will
share
my
screen.
A
Okay,
so
cyprian
had
some
agenda
items
and
he
might
not
be
able
to
make
it
so
we'll
just
go
through
them.
We
need
a
new
release
that
includes
ipv6
support.
Yes,
absolutely.
I
can
probably
do
that
today.
A
Or
at
least
get
it
started
today
and
then
finish
it
up
on
monday.
This
is
this
is
okay,
so
that's
yeah!
That's
fine.
Next
item
is
what
is
the
plan
to
move
to
beta
for
ccm
and
one
of
the
blockers?
This
is
a
good
one.
I
was
initially
before
vacation.
A
I
was
had
gone
back
to
working
on
etes
and
I
was
initially
kind
of
just
without
really
thinking
about
it
blocking
on
getting
those
up
and
running
to
get
to
beta,
but
I
don't
know
that
we
really
need.
I
mean,
since
most
of
the
code
you
know,
is
a
mirror
of
what
is
in
tree.
I
don't
know
if
we
need
the
edd
test
up
and
running
before
beta
or
if
we
should
really
just
block
on
ga.
A
For
those
to
me,
it
seems
like
the
initial
sort
of
ad
hoc
testing
that
we've
done
as
we've
been
using.
This
might
be
enough
to
just
call
the
releases
beta
that
we're
doing
and
and
then
we
really
get
the
full
e
e
suite
up
and
running
before.
We
call
it
ga
and
then
so
yeah
any
any
thoughts
on
that,
and
and
what
else
might
we
might
want
to
get
done
before?
We
call
our
releases
beta.
B
B
I
don't
know
whether
people
expect
that
a
beta
is
a
little
bit
more
ready,
but
I
think
it's
you
know
it's
it's
a
great.
It's
a
great
forcing
function
to
have
that
discussion
right
that
like
when
we
have
a
group
of
different
projects,
a
kubernetes
beta
is
not
going
to
be
the
same.
It's
not.
B
The
same
quality
as
a
k,
ops,
beta
or
a
cluster
api
beta,
like
those
things,
are
not
going
to
mean
the
same
things,
and
so
it's
going
to
be
difficult
to
assemble.
A
I
I
think
that,
like
the
what
actually
matters
is
so
I'll
be
doing
this
release
today
and
or
monday.
Should
I
call
and
then
the
next
task
that
I
have
related
to
this
is
finish
the
ee
suite.
Should
I
call
this
next
release
alpha
or
beta.
B
A
Okay,
that
no
that's
totally
fine.
I
I
that
was
my
original
plan.
Is
there
anything
else
that
we
want
to
block
beta
on.
B
I
think
in
the
ede
coverage
we
should
have
two
tests,
one
for
like
clean,
install
and
one
for
the
migration
install
and
when
those
two
are
mostly
green,
then
that
would
be
what
I
would
call
beta
just
because
I
think
that
those
are
the
two.
B
That
much
functionality
of
of
the
cloud
provider
across
an
upgrade
anyway
so
like
it's
not
like,
we
will
actually
have
pvcs
or
load
balancers
that
you
know
we
check
that
they
continue
to
exist,
but
hopefully
one
day
we
will
have
those
tests.
A
D
A
A
A
Okay,
next
one,
how
do
you
feel
about
removing
the
calls
to
imds
and
instead
just
describe
all
instances?
This
seems
an
optimization
and
that
doesn't
have
much
value
when
the
ccm
is
external.
I
so
I
think
cyprian
is
talking
about.
A
Like
the
you
know,
getting
the
node
information,
I'm.
A
I
I
know
that
the
external
ccm
has
a
different
code
path
than
what
would
be
called
in
each
cubelet
right
does.
So
I
think
the
question
is:
does
the
you
know
if
it's
just
a
cubelet
code
path,
then
we
definitely
can
remove
it
if
it's
actually
used
there
is.
There
is
also
when
the
the
ccm
initializes.
I
know
that
there's
some
kind
of
a
like
it
initializes
an
object
for
its
own
node.
A
So
I
think
we'd
want
to
look
at
that
and
see,
I'm
not
sure
exactly
what
cyprian
is
talking
about
in
terms
of
optimization
so
yeah
I
would.
I
would
be
okay
with
removing
it
if
it's
like
purely
a
public
code
path,.
B
Yeah,
I
think
that
makes
a
ton
of
sense.
I
think
you're
right.
I
think
they
used
to
be
separate
methods
and
they
were
combined
in
a
sort
of
awkward
fashion.
I
think,
but
anyway
they
they
should
be
separate
methods
and
the
the
describe
instance
describe
local
instance,
which
uses
the
mediator
service
should
be
a
different
method
from
describe
some
other
instance
in
ideal
code
land.
B
The
I
might
expect
there
might
still
be
some
calls
to
imps
to
retrieve
like
what
region
am
I
in,
for
example,
or
some
things
like
that,
but
I
think
we
can
probably
test
this
by
just
actually
sending
one.
Once
we
have
the
ede
tests,
we
can
just
send
a
pr
or
assuming
the
edtes
run
on
prs.
We
can
send
a
pr
that
just
panics
in
the
code
path
under
question
just
see
what
happens
because
it
shouldn't.
I
agree,
it
shouldn't
be
called
it'd,
be
interesting,
it'd
be
interesting.
B
It
was
called
to
know
why,
whether
it
was
just
like
listing
all
the
nodes-
and
we
happened
to
be
on
that
particular
node,
or
whether
it's
actually
like
finding
about
itself.
D
A
I'm
not
sure
I
mean,
I
know
it
gets
credentials
from
imds,
but
I
don't
know
if
it
actually
would
get
anything
else
can't
think
of
anything,
but
all
right.
D
D
Yeah,
I
hope
so
we're
we're
look
going
through
all
the
caps
manifests
that
it
manages
in
clusters
and
making
sure
that
they
are
all
compatible
with
kubernetes
122..
D
I
don't
think
there's
anything
specific
in
the
project
that
uses
apis
that
are
being
removed
in
122,
but
we're
just
doing
a
general
pass-through.
So
this
is
just
updating
those
dependencies.
D
I'm
hoping
we
can
get
a
new,
a
new
tag
cut
with
this,
so
that
we
can
update
that
in
chaos
and
start
getting
it
under
test.
A
Okay,
yeah,
I
will
take
an
action
item
to
cut
the
new
tag
as
soon
as
possible.
A
Okay,
cool
so
yeah,
the
last
one
justin.
I
was
just
looking
for
video
like
previous
uploads
for
this
meeting
and
I
wasn't
sure
like
it,
looks
like
some
of
them
are
uploaded
to
your
channel,
but
I'm
not
sure
so.
I
just
wanted
to.
A
Mention
this
and
I'm
happy
to
do
the
uploads.
I
just
don't
know
how
I
need
to
figure
out
like
permissions
and
and
stuff
like
that.
So
maybe
we
can
set
up
a
time
to
sync
on
this.
B
Yeah,
that
would
be
great.
I
I
I
didn't
realize
I
had
a
youtube
channel,
so
those
actually
should
those
should
also.
I
think
those
are
the
videos
that
I
uploaded,
but
they
should
also
be
in
the
the
communities
aws
channel.
I.
B
Well,
I
think
also
I
I
was
sort
of
hoping
that
they
would
also
get
uploaded
automatically
when
we
recorded
it.
This
is,
I
think,
but
I
don't
think
that
they
are.
This
is,
I
think,
the
playlist,
rather
than
my
personal
channel,.
B
Like
what's
on
that
channel,
but
I
guess
the
that
should
be
where
they
actually
go.
We
could,
let's,
let's
chat
later,
though,
and
figure
out
how
to
actually
get
them
uploading
in
general.
But
yes,
I
I've.
I
have
been
remiss
about
uploading
them
because
I
was
hoping
it
was
going
to
happen
automatically
and
it
did
not.
A
Cool
no
worries
all
right.
I
think
that
is
everything
for
the
agenda.
Is
there
anything
that
anybody
would
like
to
add.
A
Okay,
so
I
think
we
can
take
a
quick
look
at
the
backlog.
A
Correct
or
not
so
I
will
looks
like
I'm
already
assigned
cool.
I
will
come
back
to
this.
Just
picked
up
a
code
scanner.
Oh
that's!
Pretty
cool!
All
right.
D
A
Okay,
so
this
one
was
a
pull
request
that
I
had
been
working
on:
use,
load,
balancer
name
for
classic
elbs.
I
think
somebody
asked
if
they're
still
only
going
to
work
so
yeah
I
had
started.
I
was
just
kind
of
messing
around
to
see
how
it
would
work.
It's
pretty
simple
to
have
an
annotation
that
specifies
a
name.
A
Cleaning
up
the
way
that
it's
written
now,
I
think,
if
you
change
the
name
of
the
annotation,
it's
just
gonna
abandon
the
the
balancer
that
it
creates.
So
I
wasn't
sure
if,
like
you
know,
I
like,
I
don't
know
if
we
want
to
make
sure
that
we
don't
do
that,
I
guess.
C
Do
you
support
editing
as
well
or
like
it
can
only
be
specified
at
creation
time?
The
lord
balance,
your
name.
A
I
I
suppose
we
would,
I
mean,
ideally
we
would
support
editing.
How
does
it
work
on
the
load,
bouncer
controller.
C
So
we
use
tags.
So
even
if
you
change
the
name
later,
we
don't
honor
it
like,
because
the
tags
tags
are
what
we
use
to
associate
the
load
balancers
with
the
resources,
so
name
changes
doesn't
matter,
got
it.
Okay,.
A
Well,
I
don't
think
we
need
to
go
beyond
what
is
supported
in
the
load.
Bouncer
controller.
C
A
Yeah
I
mean
that's
exactly
what
the
problem
I
ran
into
was:
is
that
changing
it
causes
a
new
resource
to
be
created
so.
C
C
To
leak
the
old
one
right
because
we
edit
it
so
we
have
to
get
the
old
name,
and
so
it
will
be
up
to
the
controller.
So
up
to
now
that
wasn't
the
case,
because
he
was
always
guaranteed
to
have
like
a
unique
name
every
time.
But
that's
not
going
to
be
the
case
with
this
annotation.
So.
A
Right
and
as
written,
it
leaks
the
elb,
which
is
why
okay
so
like,
I
wasn't
sure
how
to
store
the
state
like
what
the
best
way
of
storing
the
state
of
like.
Oh,
I
created
a
load
bouncer
and
it
is
equivalent
to
this
new
one
that
I'm
about
to
create.
So
I
need
to
clean
up
the
old
one.
I
considered
things
like.
A
You
know
something
on
the
service,
but
I
think
I
ran
into
issues
with
just
how
the
the
existing
code
is
written.
That
makes
that
difficult.
I
don't
really
remember
it
was
it
was
a
little
while
ago,
but
so
I
think
really
just.
C
We
don't
need
to
support
this
right
if
it
becomes
complicated
like
should.
We
even
support
this
annotation
yeah.
A
A
Yeah,
I
think
it's
fair.
I
think
I
was
just
responding
to
an
issue
or
something
but
yeah.
I
think
we
can
just
point
people
to
the
to
the
new
load
bouncer
controller,
if
it's
too
complicated
so
I'll
just
comment
and
say
that
I'm
not
currently
working
on
it
because
of
the
difficulties.
A
A
C
A
A
A
A
B
I
think
I
I
think
we
actually
have
a
bug.
Well,
we
have
a
bug
that
I
have
no
idea
how
to
solve
here,
which
is
so
there's
I
mean,
there's
the
bug
that
we
we
are
trying
to
configure
some
different
subnets
on
the
on
the
load,
balancer
and
user.
Some
things
argues
that
we
should
not
change,
not
change
them
at
all,
which
I
think
is
one
issue.
I'm
not
sure
I
agree,
we
shouldn't
change
them,
but
that's
one
thing
the
other
one
is.
B
If
we
try
to
change
all
the
subnets
on
a
elb,
I
don't
think
we
can,
because
we
end
up
like
trying
to
remove
them
all
which
you
can't
do
you're
not
allowed
to
do
that.
So
you
can.
You
can't
go
to
zero
and
I
think
we
try
to
go
to
zero
and
I
don't
know
where
that
yeah.
I
think
I
think
it's
because
it's
in
one
it's
in
when
we're
only
in
one
zone,
you
can't
have
two
in
the
same
zone
and
you
can't
have
zero.
B
So
we
can't
we
can't
change.
As
far
as
I
know,
I
don't
know
if
there's
a
trick,
I'm
missing,
but
that's
that's
one
problem
so
that
those
are
the
sort
of
two
problems.
C
In
the
auto
discovery
like
we
actually
choose
one
subnet
from
each
easy
and
we
verify
that
there's
at
least
one
subnet
available.
So
that
should
not
be
an
issue
for
the
auto
discovery.
Part
and,
as
far
as
I
know
like
for
clv,
we
can
edit
the
subnet,
but
for
nlp
we
ignore
it.
After
the
load,
balancer
is
created
in
the
entry
code.
C
A
C
Yeah
because
we
group
by
the
availability
zone
right.
So
we
only
if
there
are.
C
Subnet
in
the
az,
if
we
ought
to
discover
we
sort
and
we
break
the
ties
like
using
lexicographic
order
or
whether
they're
internal
or
not,
so
we
have
those
things
in
place
to
choose
only
one
subnet,
so
only
one
subnet
will
be
chosen
for
easy.
I'm
not
sure
like
why
how
we
went
into
this
issue.
B
Yes-
and
I
think
the
question
is,
why
do
they
have
multiple?
So
I
think
the
user
is
saying
this
elb
is
already
connected
to
one
subnet,
so
we
should
leave
it.
Be.
B
That's
not
unreasonable
to
be
honest,
but
I
think
it's
not
how
it
works,
but
it's
not
unreasonable
and
then
I
think
I
think
we
should
also
find
out,
though
from
them
what
from
the
user,
why
they
are
creating
multiple
subnets
and
tagging
them
all,
because
normally
you
have
one
subnet
tagged
for
internal
and
and
one
subnet
tag
for
external,
and
that's
it.
If
I
recall
correctly,
but
you
don't
you
don't
have
them
tagged
the
same.
You.
C
C
At
this
issue
as
well,
I
will
try
to
understand
like
how
exactly
they're
running
to
this
for
the
load
balancer
controller,
like
in
the
upcoming
version.
We,
what
we'll
do
is
like
we'll
only
assign
subnet
during
load,
balancer
creation
and
will
not
edit
them
later,
and
we
do
some
capacity
based
choice
as
well
like
if
we
only
choose
subnets
which
have
enough
ip
addresses.
So
that's
the
enhancement.
A
A
A
Okay,
yeah.
I
was
also
going
to
fix
this.
We
currently
in
our
so
before
we
were
building,
or
we
were
tagging.
A
By
hand-
and
I
think
or
our
build
automation
was
pushing
the
tag
that
basically
looked
like
just
this
and
I
think,
based
on
some
some
changes
to
build
automation.
We
no
longer
push
that
tag.
We
have
always
have
this
kind
of
date
thing
in
the
front
of
it,
which
is
kind
of
ugly,
so
I
I
did
actually
have
a
pr
open
that
I
think
might
fix
it,
but
yeah.
I
need
to
to
double
check
that,
so
I
will
do
that.
A
A
B
Yes,
this
would
be.
This
would
be
great.
The
only
reason
we
didn't
do
this
from
the
start
is
because
the
apis
at
the
time
didn't
offer
it,
and
we
had
to
do
a
lot
of
stuff
to
work
around
it,
and
so
yes,
now
the
api
is
off
right.
We
should.
We
should
definitely
do
this.
A
Okay,
this,
I
think
we
looked
at
this
before
I'm
assigned,
but
I
haven't
looked
at
it.
D
B
C
C
This
is
weird
right:
it
shouldn't
be
the
case
unless,
like
the
load,
balancer
is
not
like.
The
controller
is
not
picking
it
up
at
all.
D
A
A
C
C
C
B
This
is
where
I
think
so,
if
you're
running
nginx,
I
agree
it's
confusing.
If
you're
running
engine
xl
you,
you
can
also
have
a
like
a
traditional
elb
or
nlb
like
a
layer,
four
load
balancer
and
you
use
nginx
as
your
seven
load,
balancer
and
good
equipment
that
that
can
work.
But
I
agree
with
you
that
isn't
what
they've
like.
B
C
Yes,
it's
a
load
balancer,
so
it's
definitely
ccm
or
the
controller
manager,
but.
C
A
C
A
A
A
A
Okay,
that's
a
totally
fair
feature
request.
A
C
Of
course,
I
remember
doing
something
in
the
entry
provider.
So,
okay,
let
me
see.
A
Oh,
is
this
the
one
that
we
didn't
cherry
pick.
A
We
did
okay
yeah.
I
will.
D
A
All
assign
to
just
we'll
check
if
this
needs
to
be
cherry
picked.
A
Upstream,
okay,
all
right,
I
have
plenty
of
stuff
to
do
all
right,
so
I
think
that's
probably
good
enough
for
this
week.
Does
anybody
have
anything
else.