►
From YouTube: Kubernetes - AWS Provider - Meeting 20200807
Description
Recording of the AWS Provider subproject meeting held on 20200807
A
Hello,
everybody.
This
is
the
aws
cloud
provider
office
hours
for
kubernetes
or
by
weekly
meeting
for
kubernetes.
I
am
your
moderator,
facilitator,
justin,
santa
barbara.
I
work
at
google
a
reminder
this
meeting
is
being
recorded,
will
be
uploaded
to
youtube
shortly
and
so
therefore
be
mindful
of
our
code
of
conduct,
which
boils
down
to
being
a
good
person,
and
if
there
are
more
people,
it's
good
to
you
know,
raise
your
hand
using
the
zoom
features.
A
We
have
a
relatively
light
crowd
today,
so
it
should
be
not
a
problem,
but
please
do
feel
free
to
add
things
to
the
agenda.
I
faced
a
link
to
the
agenda
and
then
we
can
be
sure
to
well
first
to
cover
it,
but
also
to
make
sure
that
it's
in
the
notes
for
the
video.
A
And
making
you
a
cohost
nick
all
right,
we
have
on
that
light
agenda.
We
have,
I
think,
one
item,
so
we
will
go
through.
We
have
one
item
when
I
checked
a
minute
ago.
We
will
go
through
that
on
a
item
which
is
a
cops
pr
from
nick.
I
was
expected
to
be
from
peter
okay.
I
confused
me
all
right
next,
you
want
to
talk
about
the
pr.
B
Yeah,
I
just
so
my
my
goal
here
is
basically
to
just
kind
of
maybe
make
make
a
little
a
few
tweaks
to
how
the
external
cloud
provider
for
aws
runs
on
cops.
B
I'm
not
sure
if,
if
what
I'm
doing
is
a
good
idea
or
or
what
what
you
guys
would
recommend,
but
I
figured
I
would
you
know
just
experimenting,
so
I
throw
it
out
there
then
my
goal
after
this
is
to
basically
document
that
in
the
aws
cloud
provider
repo
and
then
get
the
the
kubernetes
et
test
running
on
cops
using
whatever
method
we
we
end
up
coming
up
with
so
so
yeah.
B
I
just
wanted
to
didn't
make
a
ton
of
progress
on
it
this
this
past
week,
because
I
was
on
call
but
should
should-
should
make
some
more
progress
on
it
next
week.
So
I
just
wanted
to
kind
of
touch
base
with
you
guys
and
see
if
the
approach
that
I'm
doing
is
correct.
Basically,
I
found
there's
kind
of
a
couple,
different
ways
that
it
seems
like
you
can
do
add-ons,
and
one
of
them
is
this
sort
of
baked
in
way.
B
That
seems
to
be
some
of
the
more
core
add-ons
inside
up
slash
whatever
it
is
slash
add-ons
something
along
those.
A
B
B
Let's
see
so
yeah,
I
don't
know,
maybe
that's
enough
information
for
you
to
give
me
feedback
on
whether
or
not
this
is
a
reasonable
approach
or
not.
A
It
scanning
it
looks
absolutely
correct
the.
A
The
I
will
say
it
doesn't
work
yet
so
the
question
is
going
to
be
like
how
we
trigger
it.
So
we'll
probably
need
some
sort
of
field
in
the
spec
well,
either
a
feature
flag
or
some
sort
of
field
in
the
sp,
some
sort
of
way
to
trigger
this,
so
that
we
have
a
way
to
turn
it
on
and
off,
because
it
looks
like
right
now.
It
will
basically
always
be
on
and
we
want
to
have
some
mechanism
and
then
we
want
to
install
that
add-on
and
turn
off
the
built-in
one.
B
Okay,
yeah,
that
sounds
that
sounds
right.
Yeah
it
yeah
it's
it's
it
should
be.
I
think
I
was
attempting
to
trigger
it
based
on.
If
you
look
at
the
the,
if
statement,
for
where
it's
like
adding
the
add-on,
it
should
be
under
the
external
block,
so
you
have
to
have
the.
B
Field
defined-
and
I
think
if
it
just
isn't
empty,
then
it
will.
It
will
use
that.
So
I
don't
know
if
that's
the
right
trigger,
but
yeah
go
ahead.
A
Yeah,
no,
okay,
sorry
I
didn't
mean
to
yeah.
I
was
just
like
yeah
that
sounds.
I
didn't
spot
that
that
could
be.
That
could
well
be
good
enough.
I'll
have
to
think
about
whether
we
actually
like
fill
that
in
like
whether
there's
any
chance
it
could
be
filled
in
by
mistake
as
it
were,
but
it
seems
pretty
likely
to
be
reasonable.
A
We
might
need
a
secondary
thing,
but
yes,
this
sounds
good,
and
so
yes,
we
all
I
can.
I
can
have
a
look
at
this
and
try
to
try
this
out.
I
presume
that
the
image
is
real.
The
image
works,
the.
B
Image
yeah,
the
image
is
real,
so
I
I
haven't
tested
this
this
with
a
brand
new
cluster,
but
I
was
having
issues
updating
like
the
most
the
the
most
recent
cluster
that
I
had
created
to
these
changes,
and
I
had
I
did
the
whole.
B
You
know
build
all
of
the
binaries,
create
a
cops,
my
own
cop
space,
url
s3
bucket
that's
public
and
then
put
everything
using
the
same
directory
structure
that
like
the
ci
test
and
stuff
use.
So
I
hope
it's
correct,
but
yeah
it's
still
what
something's
wrong,
I'm
not
sure
what
it
is
yet.
A
Okay,
I'm
happy
to
like
try
to
figure
that
out
with
you.
If
you
want
and
but
yeah
the
pr
looks,
looks
good
and
I
will.
I
will
look
at
like
how
to
turn
off
the
turn
off
the
other.
The
other
like
the
built-in
integrations
when
this
is
on
as
it
were,
got
it
got
it
yeah,
okay,
cool,
but
thank
you
for
doing
that
looks
awesome.
A
I
lost
the
agenda
because
I'm
on
a
different
machine,
okay,
we
have
nothing
else
on
our
formal
agenda,
so
we
discussed
that
we
were
going
to
basically
go
through
any
new
issues
that
may
have
come
in
and
see
whether
any
new
issues
have
come
in
on
the
I
guess,
on
the
main
repo,
primarily
before
we
do
that.
I'd
like
to
open
the
floor
for
any
other
topics.
A
A
A
The
the
it's
silly,
we
can't
run
zoom
on
our
work
machines
for
security
reasons,
and
my
linux
laptop
has
a
hdmi
out.
My
mac
laptop
has
a
display
port
out,
which
means
I
don't
have
to
fiddle
with
the
cables
to
get
it
into
my
monitor
that
which
always
takes
a
very
long
time.
But
yes,
it
is
slightly
backfired
at
this
point,
all
right,
let's
see
if
this
works,
I
think
I
can
probably
bravely
no,
it
did
not
work
all
right.
I
will
minimize
everything.
A
A
A
A
B
I
could
do
it,
I
just
have
to
get
ready.
Yeah.
Just
give
me
a
second,
because
I
have
to
quit
to
to
do
this.
A
A
Cool
so
nick
we
can
see
your
your
browser
with
the
provider
the
best
meeting
up
on
a
similar
number
of
tabs
to
the
ones
I
have.
Let
me
quickly
look
at
the
make
sure.
B
A
A
Okay,
this
looks
like
a
good
list
if
we
go
through
like
the
last
two
weeks.
I
guess
that
would
be.
B
Yeah,
that's
fair,
so
we
could
start
with
like
july
july.
I
guess:
okay,
a
couple
of
scheduled
et
jobs,
fail
that
doesn't
feel
like
us.
We
can
have.
D
B
D
A
I'm
not
yeah
so
obviously,
like
my
muscle
when
we
definitely
have
this
as
a
scenario
in
the
case
of
the
aws
vpc
cni
provider.
Is
that
still
the
name
but
the
sorry
the
this
would
be
using
the
default
route
controller,
like
the
old
thing
that
used
to
be
cubenet
or
even
before,
cubenet
right
right,
right.
A
Air
configuring
source
dest
check,
okay,
yeah,
so.
A
Okay,
interesting
so
yeah
source
test
check
needs
to
be
turned
off
for
this
trick
to
work
for
tube
net
to
work,
because,
basically,
we
we
map
a
slash
24
to
a
typically
to
it's
always
previously
been
to
an
instance.
But
I
guess
technically
it's
been
to
an
eni.
A
I
bet
that
there
will
be
other
challenges
like
I.
I
don't
think
this
is
gonna
if
I
don't
think
we
just
have
to
fix
this.
A
D
A
What
I
don't
know
is
whether
this
will
actually
work
like
is
this?
Is
this
sufficient,
or
does
it
just
get
past
this
first
one
I
can.
I
can
definitely
have
a
look
at
this
as
well.
I
suspect
we're
not
going
to
get
this
into
119.
This
would
be
a
120
thing,
but
this
feels
like
something
which
we
definitely
could
get
into
120,
especially
as
there's
a
pr
for
it.
A
D
B
A
The
cubenet
works
with
yeah
the
big
downside
of
of
this
mode
of
operation
used
to
be
that
you
could
only
have
50
50
routes,
and
so
this
limited
your
cluster
size
to
50
nodes
less
a
couple
for
overhead.
I
don't.
I
I
believe
that
limit
may
have
been
raised
on
the
aws
side
or
the
there
was
like
a
there
was
a
default
limit
and
there's
a
you
can
ask
it
for
it
to
be
raised,
and
I
think
the
you
can
ask
went
from
a
hundred.
A
B
Yeah
that
that
that
aligns
with
what
I
remember
hearing
internally,
so
I
I
think
you're
you're
right
about
that.
A
A
Yeah,
we'll
see
how
it's
not
clear
like
how
critical
it
is
to
support
this
or
whether
they
probably
should
be
using
things
like
the
awesome
cni.
If
they
want
like.
Presumably
they
want
higher
performance
or
something
like
that.
So
maybe
maybe
that
would
make
more
sense.
So
we
can.
We
can
try
to
figure
that
out,
but
it
seems
relatively
benign.
B
Yeah,
I
think
it's
I
think
it's
I
mean
unless
we
are
taking
steps
to
deprecate
this,
I
think
it's
worth
keeping
it
improving
it
continuing
to
improve
it.
B
B
A
A
Yeah,
that's,
that
is,
that
would
be
a
bug.
There
is
a.
How
does
this
work?
There's
a
so
obviously
there's
a
limited
number
of
instances
you
can
attach
to
a
number
of
volumes
you
can
attach
to
an
instance,
if
I
recall
correctly,
we
hard
code.
That
is
that
fair
is
that
right,
but
then
there
was
a
a
request
to
have.
A
A
B
Sorry
I
was
reading
all
right
issue.
Can
you
say
it
one
more
time.
A
I
feel
like
long
term
we
should.
We
should
build
this
properly.
We
should
get
this
working
as
it
was
because
it's
a
viable
workaround,
but
the
long-term
answer
is
probably
to
put
it
into
the
aws
csi
driver,
the
aws
ebs
csi
driver,
but
I
don't
know
if
anyone
is
using
that
in
a
I
know
that
cops
doesn't
have
an
enabled.
I
don't
know
if
eks
has
enabled.
B
Okay,
it's
it's
in
there
somewhere.
So
we,
if
I'm
not
mistaken,
we
haven't
enabled
no
yeah,
we
haven't
enabled
it
yet
we
have.
We
have
plans
to
enable
it.
We've
we've
gone.
I
think!
No,
I
can
do
it's
additive
right,
you
can,
you
can
use
casi,
it
requires
some
extra
configuration.
Is
that.
A
Is
that
yes,
actually
you're
right?
Amazon
eks
now
supports
the
ebscsi
driver
as
of
september
last.
B
A
Seems
like
a
good
answer:
okay,
but
yes,
I
think
yeah
that
that
is
a
whole
nother.
Wrinkle
right
is
that
now
we
have
like
now,
it
can't
just
be
in
the
evs
csi
driver,
but
I
guess,
if
you
were
using
only
the
ebscsi
driver,
we
could
have
better
behavior
in
the
evs
csi
driver
that
that
looked
at
this
limit
more
accurately.
I
guess
or
understood
this
limit
more
accurately,
but
we
can.
We
should
get
the
environment
variable
working
so
that
people
have
an
option.
B
A
I
always
hate
it
when
people
ask
me
about,
like
the
gke
plants
on
public
in
a
public
place,
because
it's
like
what
is
what
like
what
is
on
the
record
and
what
it's
not
it's
always
hard.
So
I
apologize
for
doing
that
too.
No.
A
B
Yeah,
I
guess
this
shouldn't
be
the
one
that
we
just
looked
at
shouldn't
be
tagged
provider
aws,
because
it's
more
generic
than
that.
Would
you
say.
A
Yeah,
I
guess
so.
Yes,
I
think
that's
fair.
It
certainly
affects
us
more
than
others,
so
we
should
make
sure
that
it
gets
fixed.
A
A
B
B
E
B
B
I
don't
know
if
that
is
actually
what
we're
working
on.
I
think
we
might
be.
I
know
that
there
is
some
nlb
ip
support
that
we're
working
on,
but
that's
probably
pod
ips,
but
I'm
not
100
sure.
C
Anyway,
can't
talk
about
that
yeah,
we
are
going
to
support
the
iphone
4.
A
B
So
I
will,
I
guess,
I'll:
just
leave,
sharing
up
on
the
agenda
go
ahead
yang.
Do
you
want
to
introduce
your
your
topic.
C
Yeah
so
the
first
thing
we
are
going
to
support
the
ip
mode
for
nlp.
So
currently
we
are
registering
the
instance
as
target
to
the
mv
which,
using
the
new
protocol
and
for
the
ip
mode,
we
are
going
to
add
in
the
port
directly
at
the
target
of
mlb,
so
it
avoids
cool
proxy.
So
traffic
work
is
going
from
clients
to
the
nlb
and
then
to
the
course
directly
and
to
do
that,
we
have
to
use
using
an
external
controller,
because
the
current
into
a
cloud,
private
interface
is
using
instance
ids.
C
We
plan
to
do
that
because
nrb
and
aob
actually
shares
the
same
eop
v2
api,
so
there
is
no
sense
to
share
and
we
plan
to
just
content
both
of
them
in
the
same
controller
and
in
the
future.
We
can
also
move
the
instance
more
support
into
the
this
external
controller
as
well
another
into
a
car
provider
or
the
auto
recovery.
I
guess
return
implemented
elsewhere,
so
people
can
just
install
this
aws
node
as
a
controller
and
have
support
for
lb
and
lv
books.
A
That
sounds
I
have.
I
have
two
that
sounds
awesome.
I
think
I
think
the
two
questions
I
would
have
are
and
very
supportive,
but
let's,
let's
figure
out
how
to
turn
off
the
entry.
A
I
think
we
talked
about
this
like
briefly
two
weeks
ago.
How
how
are
we
going
to
turn
off
the
or
tell
the
entry
controllers
not
to
not
to
touch
one
of
those
one
of
the
your
load
balancers
that
should
be
in
this
mode
and
then
the
other
question
is
about
where
this
code
is
going
to
end
up
living
and
is
it
in
the
right
place.
C
A
That
makes
a
ton
of
sense
the
I
don't
know
if
you
I
I
I
feel
like
we
talked
to
this
before
so.
Forgive
me
if
I'm
repeating
anything
we
talked
about,
I
think
the
idea
of
provider
on
ingress
as
being
a
a
template
for
this,
something
something
like
that,
and
maybe
this
should
be
a
a
generic
field.
As
you
say,
and
I
I
suggest
we
should
open
an
issue
or
something
or
at
least
bring
it
up
in
the
sig
cloud
provider
discussion.
A
Sorry
no
worries,
I
mean
yeah,
I
just
there
or
possibly
see
networking.
I
don't
know
nick
do
you
do
you
have
maybe
both
which
is.
B
Yeah
I
mean
sick,
so
yanked.
I
know
you
were
thinking
about
doing
like
a
a
cap
in
the
aws
directory
and
and
the
caps
repo
enhancements
repo.
Are
you
still
planning
on
doing
that
or
is.
B
Yeah,
I
think
that
would
be
really
good
just
to
kind
of
make
it
easier
to
share
the
proposal,
and
I
think
personally,
I
think
it
would
be
really
great
to
have
this
at
sig
cloud
provider.
B
You
know,
ideas
or
concerns
around
it.
So
there
was
there
was
a
sick
cloud
provider
meeting
this
week.
Unfortunately,
so
the
next
one
is
not
next
week,
it's
the
week
after,
but
I'm
guessing
that
our
overall
timeline
for
this
is
something
maybe
akin
to
the
120
timeline.
Is
that
is
that
true?
Or
is
it
sooner
than
that.
A
Okay,
it's
a
little
more
aggressive
than
I
thought,
not
an
annotation.
Then
I
mean,
I
think,
yeah.
I
think
you
could
so
an
annotation
might
be
a
way
to
get
it
to
happen
immediately,
although
if
you
want
the
cloud
provider
to
focus.
A
C
A
C
A
C
C
A
Cool
that
sounds
that
sounds
great.
I'm
trying
to
I'm
trying
to
find
the
the
way
gce
did
it.
I
guess,
but
I
will
I'll
also
just
track
down
your
pr.
I
guess
and
try
to
see
what's
going
on,
but
yeah,
that's
that's
really
exciting.
I
think
the
other
question
I
had
was
where
this,
where
the
code
lives
right
now,
is
it
is
it
in
kubernetes
or
is
it
in?
C
A
Yeah,
that
would
be
really
great
if
anyone
else
has
any
other
questions
or
thoughts.
B
C
Is
that
going
to
be
cncf
yeah,
it
means
or.
A
It
it
is
enclosed
right.
Okay,
that's
that's
wonderful!
For
the
for
your
information.
I
don't
know
whether
other
projects
have
yet
done.
Renames,
I'm
trying
to
think.
A
C
Yeah
yeah
I
mean
technically,
it
doesn't
have
to
be
released.
It's
just.
We
added
a
support
for
lbit
mode
for
kubernetes
service,
but
it's
it's
just.
We
plan
to
rename
it
because
there
is
ongoing
another
networking
api
called
service
api.
So
we
also
plan
to
support
in
the
sim
controller
as
well.
So
it
makes
sense
to
rename
to
innovation
depends
the
controller.
A
Wonderful
cool
any
other.
B
Topics
or
anything,
did
you
just
update
the
agenda
with
with
notes?
I
did
not
okay
I'll
put
some
in
now.
If
you
want,
I
think
I
think
it
would
be
good.
Also
yang,
could
you
add,
like
the
you
had
like
an
issue
in
a
pr,
I
think.
A
Yeah,
I
don't
think
we
need
to
record
the
the
video
and
we're
we've.
We
did.
Probably
enough
of
our
of
our.
We
wanted
to
do
a
shorter
review
of
issues.
I
don't
know.
If
there's
any
other
topics,
we
can
resume
the
issue
triage.
I
guess
in
in
two
weeks
at
this
point,
sounds
good
yeah
and
there's
the
poll
I
will
request.
I
will
copy
that
and
thank
you
yang
and
I
will
update
the
notes.
I
don't
know
if
there
are
any
other
final
topics.
A
Otherwise
I
will
stop
the
recording
and
seamlessly
splice
out
the
the
the
from
the
alarm
to
the
the
sharing
of
the
whiteboard.
A
All
right:
well,
if
there's
no
other
topics,
then
I
wish
everyone
a
very
happy
two
weeks
happy
weekend.