►
From YouTube: Kubernetes - AWS Provider - Meeting 20220513
Description
Recording of the AWS Provider subproject meeting held on 20220513
Agenda - https://docs.google.com/document/d/1-i0xQidlXnFEP9fXHWkBxqySkXwJnrGJP9OGyP2_P14/
Subproject Updates:
CCM
- new releases
kOps
- 1.24 release
- issue with versions in ccm
- planning to add more testing to ccm
LBC
- planning v2.4.2 release
Karpenter
- released v0.10.0
* KIT demo next meeting
SIG k8s-infra has ongoing project to migrate costs away from google and towards AWS for hosting container images in S3 mirrors. If you're interested in participating here, let hay know.
A
Foreign,
hello,
everyone
welcome
to
the
provider
AWS
sub-project
meeting.
It
is
Friday
May
13th
today
and
we
will
go
ahead
and
get
started
at
least
with
sub
project
updates.
I,
don't
see
anything
else
on
the
agenda.
So
if
you
have
anything,
then
go
ahead
and
add
it
otherwise,
we'll
just
do
the
sub-project
updates
and
go
from
there.
A
All
right:
well,
we
fill
those
in
I
will
start
with
the
CCM
Cloud
controller
manager.
We
have
new
releases
for
all
versions,
120
to
124..
We
have
merged.
A
This
was
done
by
the
managed
nodes
team
on
eqs,
we've
merged
a
new
controller
called
the
tagging
controller
and
we've
back
ported
that
to
think
all
the
way
back
to
the
120
branch
and
that
will
basically
take
a
config
and
apply
tags
to
nodes.
The
primary
use
case
for
this
is
to
help
eks
customers
understand
their
cluster
cost,
so
there
is
a
yeah,
so
that's
I'll
come
out
of
that
approached
controller
and
yeah
other
than
that
nearly
says
for
120
to
124.
I.
Think
that's!
A
It
Justin.
B
Yes,
so
we
are
working
on
doing
the
124
caps
release,
we,
you
know
123
was
out,
but
we
we
are
just
like
doing
the
final
external
components.
I
guess
we
call
them
instead
of
add-ons
AWS,
Cloud
controller
manager
being
one
of
the
big
ones
and
I
think
we
hit
an
interesting
issue
with
the
versioning
scheme,
particularly
on
the
older
branches.
I.
Think
I
think
we
had
a
meta
question
about
to
what
extent
you
know
we
can
approve
those
or
or
do
you
like.
B
A
I
spoke
to
well,
actually,
I
haven't
responded
to
him,
but
all
I
asked
about
this
and
I
was
going
to
suggest
that
yeah
somebody
from
Cops
does
be
an
owner
on
the
project
so
that
they
can
participate
in
releases
and
all
that
in
terms
of
those
builds
technically
like
we
rebuild
those
components
when
we
publish
them,
but
it
is
still
absolutely
you
know
important
and
yeah.
The
the
the
versioning
thing
I
think
is
actually
fixed.
Now,
we've
just
been
backboarding
that
fix
to
the
older
branches.
A
I
think
we
haven't
gone.
It's
still
broken
in
120
right
now,
and
we're
working
through
that,
but
yeah.
That
was
thanks
to
cyprian,
who
had
that
fix,
I,
believe
so
yeah.
A
Yeah
I
think
if
I
I
know
Ole
expressed
interest
in
helping
out
there,
so
yeah
I'll
message
him.
One.
A
B
Yes,
I
would
I
think
that's
fair,
like
it's
Friday
evening
afternoon,
so
yeah,
it's
it's
not
ideal.
Yeah.
A
B
We
actually
made
this
time
slot
in
alternate
hours
and
the
like
the
European
contribution,
Is
Not
Great
in
this
one,
but
we
also
have
a
Thursday
at
9
00
a.m.
Eastern,
which
I
know,
makes
the
West
coasters
cry
time
slot,
but
the
West
coasters
tend
not
to
show
up
to
that
one,
but
the
Europe
Europe
people
can
typically
come
to
that
one.
B
We
haven't
yet
found
a
time
slot
that
works
for
Asia,
but
that's
Thursday
9
A.M
is,
is
we
call
it
like
informal
office
hours
or
something?
But
we,
you
know
finding
an
informal
slot.
A
Definitely,
okay.
I
will
make
a
note
of
that.
A
B
E-Testing
to
the
AWS
Cloud
controller
manager,
repo,
there
is
currently
testing
on
the
master
branch
and
we
wanted
to,
like
you
know,
add
some
tests
on
the
other
branches
as
well
to
I.
Don't
know
there
were
suggestions
that
they
weren't.
We
weren't
testing
previous
releases
so
like
just
getting
some
more
test
coverage
on
this.
A
Yeah
I
think
we've
added
the
ete
test
to
the
branches
that
didn't
have
it
back
to
I.
Think
120
after
we
merge
a
couple
PR's
today,
which
I
think
will
probably
be
enough,
but
there's
still
more
coverage
needed
for
sure
more
tests
at
least
awesome.
D
I,
don't
have
much
updates
I'm
trying
to
close
on
the
242
capacities.
I
I
have
some
internal
dependencies
that
I
am
working
through
right
now.
So
once
I
have
address
the
dependency
issues,
I'll
release
the
path.
A
Cool
anything
to
call
out
in
that
patch
release.
C
All
right,
it's
a
carpenter,
so
yeah
there
was
a
new
release
for
Carpenter
dot
0.10.
There
is
a
new
interesting
feature
in
this
one
where
they
have
added
support
for
awareness
for
in-flight
nodes.
This
was
not
there
earlier,
so
it
might
be
useful
for
some
folks.
So
earlier,
when
there
is
a
pending
part,
we
Carpenter
would
provision
a
node
while
the
node
is
becoming
ready.
If
there
is
another
part,
it
will
bring
up
another
node.
So
now
it
takes
into
consideration
that
there
is
already
an
in-flight
node,
so
it
it's
a
little
intelligent.
A
Awesome
cool
all
right,
Jay
go
ahead.
I
think
you
have
both
a
note.
E
Yeah,
it's
the
same
one
that
I've
been
hanging
on
for
the
last
few
months.
I
know
Logan
Davies
is
the
contributor
of
that.
Pr
has
been
very
patient
with
us.
It's
been
through
a
number
of
rounds
of
reviews,
I
think
Judy
and
myself
are
plus
one
on
it.
I
think
we
just
need
a
final
check
from
from
you,
nick
and
or
biker
I've
asked
Micah,
but
he's
been
doing
other
things.
A
Yeah
I
was
planning
on
getting
into
it
this
week
but,
as
you
know,
I'm
finishing
up
my
on-call
shift.
So
it's
been
a
little
bit
hectic
yeah.
A
All
right
cool
anything
else
in
terms
of
some
project
updates.
Otherwise
we
can
move
to
I
guess
is
this
critique
look.
A
C
Yeah,
that's
me,
yeah
I
did
that
so,
basically,
as
part
of
the
scalability
team,
we
have
been
working
on
this
project
for
about
a
year
now
and
I
guess
we're
in
the
state
where
we
can
demo
to
the
group.
If
there
is
interest
and
people
want
to
start
using
it
so
at
a
very
high
level,
it
is.
It
is
a
collection
of
some
tools
which
allows
you
to
run
tests
against
kubernetes
clusters.
C
It
can
be
funny
like
kubernetes
clusters,
that
kit
can
provision
or
eks
cluster,
and
then
it
comes
with
some
pre-built
grafana
dashboards
and
some
tekton
tasks.
So
basically,
with
the
CLI
and
using
the
tecton
dashboard,
you
can
run
some
tests
and
you
can
say
against
an
eks
cluster
I
want
to
create,
let's
say:
10,
000,
nodes
or
I
want
to
create
10,
000,
Secrets,
10
000
parts,
and
we
have
all
those
tasks
defined.
C
So
I
just
want
you
to
see
like
if
there
is
interest
and
I
can
demote,
maybe
in
the
next
in
the
next
meetings
and
see
from.
A
And
let's
advertise
that
so
that
we
can,
you
know,
get
folks
from
the
community
to
come.
C
E
This
was
just
a
shout
out,
so
I've
been
working
in
sick,
Kate's
infra
on
offloading
some
of
the
costs
that
are
related
to
I.
Think
I
may
have
mentioned
it
before,
but
all
of
the
kubernetes
container
registry
GCR
is
hosted
on
GCR.
Obviously,
in
Google
data
centers,
a
huge
portion
of
kubernetes
users
are
running
on
ec2,
either
eks
or
on
you
know,
by
cops
and
those
instances
are
pulling
images
from
GCR
and
the
bandwidth.
The
egress
bandwidth
out
of
Google
data
centers
is
roughly
75
000
a
month.
E
Just
for
that.
So
what
we've
done
is
we've
set
up
a
bunch
of
S3
buckets
as
mirrors
in
The
11,
most
popular
AWS
regions
and
we're
mirroring
the
image
layers,
for
you
know
the
whatever
it
is:
1.5
terabytes
of
images
in
the
registry.cakes.io
space.
E
So
we
put
together
this
oci
proxy,
which
runs
in
Cloud
run
on
Google,
which
basically
says
hey
is
the
source
IP
for
the
client
coming
from
an
AWS
data
center,
if
so,
which
one
or
which
region,
and
then
we
send
it
to
an
S3
bucket
so
that
we
don't
incur
the
transfer
costs
right,
because
S3
bucket
to
an
ec2
instance
in
the
same
region,
is
for
free.
So
anyway,
that
is
it's
a
huge
reduction
in
costs
and
if
you're
interested
in
playing
a
part
of
this
I
could
certainly
use
your
help.
E
And
let
me
know.
A
That's
that's
awesome,
I
I
would
say
you
know,
I,
don't
know
how
far
along
in
the
process
you
are,
but
I
would
say
that
when
this
gets
farther
I
guess
it
would
be
cool
to
just
maybe
like
have
a
quick
presentation
and
like
this
is
what
we've
done.
This
is
where
the
images
are
now
yeah.
E
Absolutely
I,
I,
I,
think
and
Ben
Elder
and
myself
and
Adolfo
would
would
be
happy
to
do
a
little
presentation
to
the
other
sigs
saying
hey.
This
is
what
what's
going
on.
You'll,
probably
see
some
some
requests
from
some
pull
requests
and
or
slack
messages
from
dims
or
Adolfo,
saying:
hey,
we're
changing
image
Uris,
you
know
away
from
kubernetes.gcr.io
or
whatever
it
is,
or
kate.tcr
dot
IO
to
registry.case.io.
E
So
you'll
see
like
a
bunch
of
changes
to
to
the
image
Uris
that
are
referencing.
Things
like
cops
right
so
anyway,
there's
lots
of
different
moving
Parts
there,
but
hopefully,
within
the
next
few
weeks
we
should
start
seeing
a
pretty
dramatic
reduction
in
costs
and,
as
Justin
mentioned
on
the
on
the
agenda,
doc
yeah.
These
are
costs
that
cncf
bears
right,
so
they're
consuming
Google
credits
in
good
gcp
and
they
will
be
consuming
AWS
credits
in
AWS.
That's
what
we're
trying
to
shift
the
costs
so.
B
Yeah,
thanks
for
doing
that,
taking
on
those
costs
and
and
reducing
them
as
the
wonderful
Jay
and
thank
you,
I
also
want
to
mention.
There's
there's
also
a
benefit
for
users
as
well,
which
is
that
like
they
should
have
faster.
E
B
Better
performance
and
like
you
know
they
don't
have
to
Traverse
the
we'll
call
it.
The
public
internet
I
assume
that
we
have
some
form
of
direct
connection,
but
who
knows
we'll
call
it
the
public
internet,
but
they
they
can
consume.
They
can
just
it
should
be
within
the
data
center
or
within
the
region
at
least
so
that'll
be
that'll,
be
awesome.
E
A
cloud
run
function
that
right
now
all
it
does
is
just
forward
on
to
to
kate's.gcr.io,
but
in
the
future,
in
the
very
near
future,
it
will
only
serve
metadata
requests
from
case.gcr.io
and
all
of
the
V2
images
layers
or
the
digest
blobs
for
the
image
layers,
we'll
get
a
302
redirect
back
from
oci
proxy,
which
is
running
in
registry.k.io
to
the
pulling
Docker
client
and
the
docker
client
will
then
you
know,
send
a
request
to
whatever
the
S3
object
is
in
their
local
region.
A
Cool
all.
C
A
If
there's
nothing
else
on
that,
I
was
just
going
to
suggest
we,
you
know
the
people
who
are
at
kubecon
do
some
kind
of
a
meet
up,
even
if
it's
just
like
a
lunch
or
something
and
just
say
hello
and
be
nice
to
to
see
you
know
the
the
cops
contributors
that
are
EU
based
and
and
others
I.
Maybe
we
can
post
a
message
in
the
provider
AWS
slack
Channel
or
something
like
that.
Just
to.
A
Definitely
should
say
hi
there,
but
we
might
want
to
do
something
later
in
the
week.
Just
in
case
you
know,
there's
interest
and
and
people
you
know
interest
from
people
who
aren't
going
to
be
there.
B
Yeah
I
know
some
people
aren't
writing
until
Tuesday,
because
I
have
a
completely
free
day
on
Tuesday
I,
don't
know
what's
I'm
sure
it
won't
stay
free
but
like
it
seems
like
there's
a
gap,
so
I
don't
I,
don't
quite
know.
What's
going
on
there,
it's
like
all
the
associated
events
or
whatever
co-located
events
well,
yeah.
E
Aws
container
today
turned
into
AWS
container
two
hours
on
on
Tuesday.
It's
been
going
on.
Actually,
all
all
this
week,
AWS
container
day
has
because
they
turned
it
into
like
a
five
day.
Six
day
thing
yeah.
E
Nobody
knows
about
this,
but
anyway,
there
is
a
an
event
on
the
zero
day,
the
the
Tuesday
for
AWS
container
day,
but
it's
like
a
it's
like
a
two
hour
presentation
from
Nate
Tabor
on
our
product,
roadmaps
and
stuff,
like
that,
so
it's
not
the
eight
hour
straight
Twitch
TV
streaming,
Extravaganza
that
it
has
been
in
the
last
a
few
virtual
events.
A
A
Cool
all
right:
if
nothing
else,
then
we
can
wrap
up
thanks
for
joining
everyone
and
happy
Friday.