►
From YouTube: SIG Cloud Provider 2022-03-16
Description
Agenda: https://docs.google.com/document/d/1OZE-ub-v6B8y-GuaWejL-vU_f9jsjBbrim4LtTfxssw/
* Leader migration update
* PRs to review before code freeze
A
B
C
A
A
C
A
A
D
Sure
yeah
no
major,
updates
or
changes.
I
do
have
a
couple
of
pr's
we're
trying
to
get
through
and
I
had
talked
to
andrew
sidekim
about
the
second
one
he
said:
maybe
take
it
to
sig
network.
So
I'm
gonna
talk
to
them
about
that.
Hopefully,
but
question
about
that.
First,
one
andrew.
You
added
some
comments.
D
Yesterday,
I'm
just
kind
of
wondering.
I
know
that
we
obviously
aren't
you
know
getting
this
in
right,
the
second
or
anything
but
like
basically
it
looks
like
you're
you're
saying
hey.
Maybe
we
should
take
a
totally
different
approach
and
like
if
we're,
if
we're
trying
to
kind
of
I
kind
of
want
to
yes
and
that
like,
if
we're
trying
to
solve
an
actual
bug,
do
we
wait
for
the
optimal?
B
B
D
B
A
C
Ibm
yeah,
we
just
finished
our
124,
bring
up
working
with
red
hat
on
the
4.11
hippie
work
install
with
the
cloud
provider
there
and
then
we're
doing
some
refactoring,
our
vpc
load,
balancer
code,
if
you
recall
a
while
ago,
talked
about
the
open
source
for
that
we
have
kind
of
is
broken
into
two
bits
and
we're
trying
to
refactor
that
to
make
it
a
little
bit
easier.
I
know
andrew
you
had
mentioned
that
it
would
be
nice
to
not
have
those
two
repos
and
that's
what
we're
working
on
cleaning
that
up.
C
E
Sorry
I
had
to
find
the
unmute
button.
Yeah
server
network
network
proxy
they've
been
a
bunch
of
fixes
in
the
last
few
weeks
and
we
did
put
out
a
dot,
30
release
and
that
that
includes
some
fixes
on
the
client
library
which
goes
into
the
cube
api
server.
So
I
would
recommend,
if
anyone's
using
the
api
server
network
proxy
to
make
sure
they
pick
up
the
dot
30
release.
E
C
A
E
D
A
B
C
B
I
mean
we
still
have
two
weeks,
so
maybe
so
like
the
pr
that
you
have
up
is
is
just
the
the
v1
types
right,
but
not
right,
so
maybe
like
when
there's
a
follow-up
pr
to
bump
the
feature
gate
we
can
just
huddle
again
and
just
make
sure
that
we're
all
okay
with
the
g
graduation
going
into
124
and
discuss
it.
Then.
C
C
A
A
B
So
sorry,
nick,
I
have
a
question
for
you,
so
you
you
might
like
you're
testing
it
and
you
guess,
but
you
you
mentioned,
the
external
cloud
provider
hasn't
enabled
the
configs
yet
so
like.
How
does
that?
How
does
that
work
like?
Don't
you
need
the
external
provider
to
have
the
migration
config
wired
in.
A
B
A
B
A
A
B
Where
and
service
controller
is
one
of
those
like
controllers
that,
like
the
ownership,
has
hasn't
always
been
clear
like
if
it's
network
or
cloud
provider,
because
it's
it's
a
controller
that
operates
on
service.
But
it's
only
for
cloud
providers
like
it's,
not
a
generic
kubernetes
controller
but
anyways.
There's
a
there's,
a
bug
here
where
we're
not
updating
the
internal
cache
and
when
I
say
cache,
I
mean
service
controller
has
its
own
like
really
simple,
cache
implementation
of
service
that
it
populates
from
the
actual
controller
cache,
and
the
reason
why
it
has.
B
You
need,
when
you
delete
a
service,
it's
immediately
removed
from
the
controller
cache,
so
service
controller
has
its
own
cache
to
preserve
that
service.
That
was
deleted.
So
we
can
then
do
the
delete
operation,
and
that
was
before
we
added
finalizers
for
server-side
load
balancers
in
like
1,
19
or
118,
like
that
and
there's.
C
B
So
this
is
fixing
like
the
cash
update,
and
I
think
my
point
was
like
we
added
the
finalized
support
in
119
or
118,
which
means
that
we
don't
have
to
worry
about
the
service
being
deleted
from
the
cache
and
not
being
able
to
do
the
delete
operation,
because
we
have
the
finalizer
to
block
that.
And
so
I
think
we
can
just
do
some
cleanup
here
and
just
delete
the
service
cache.
The
internal
service
cache
and
just
use
the
controller
cache
directly.
B
But
I
do
I
do
agree
that,
like
oh,
it
is
like
a
bit
of
extra
work
to
do
and
like
we
are
like
two
weeks
out
from
code
freeze.
So
maybe
we
just
fix
the
bug
and
then
do
the
cleanup
after
I
think
I
could
like
I
think
either
we
could
like.
I
think
I
can
be
convinced
like
either
way
but
wanted
to
hear
what
other
people
thought
about.
D
E
It
sounds
like
if
we
do
the
delete
that
the
original
problems
having
to
do
with
with
deleting
the
the
services
were
probably
somewhat
subtle
and
subtle.
Bugs
worry
me
to
when
we
get
close
to
code.
Freeze
do
we
know
if
we
have
any
tests,
especially
e
to
e
tests
that
are
specifically
around
why
we
have
that
additional
cash.
B
Yeah,
I'm
I
very
certain
I
remember
reviewing
a
pr
for
adding
an
ed
test
for
this,
for,
like
the
finalizer
like
deleting
a
low
bounce
or
finalizer,
but
I
don't.
I
don't
think
like
the
bug.
Is
that
subtle
or
like
the
behavior
is
not
that
subtle?
It's
just
like
we
didn't
have
finalizers
unbelievable.
We
just
needed
a
way
to
keep
that
in
memory.
E
No,
no,
I
I
grant
that,
but
I
mean
you
know
like
the
the
problems
that
I'm
just
trying
to
anticipate
the
problems
that
may
occur
if
something
goes,
if
there
is
still
a
value
to
having
this
cash
that
we
haven't.
Thought
of
you
know
things
like
this.
I
mean
the
obvious
one
is
okay:
the
service
didn't
get
cleaned
up
properly,
but
the
more
subtle
ones
may
be.
B
Yeah,
I
get
your
point
and
like
it's
gonna
require
a
larger
refactor
of
the
code
and
that
could
create
more
subtle,
bugs,
I
think,
also
just
thinking
about
it
more.
If
we
did
the
refactor,
it
would
also
make
it
much
harder
to
back
pork.
So
maybe
we
maybe
we
should
just
do
the
the
simple
but
ugly
fix
that's
more
back
portable
and
then
in
125.
E
B
A
A
A
E
A
E
B
Yeah,
I
think
this
one
might
be
a
bit
contentious
like.
I
think
I
can
see
definitely
arguments
for
why
we
wouldn't
want
this,
but
I
made
some
decent
progress
on
the
adding
some
node
e2e
tests
for
the
crunchwriter
plugin
and,
like
ideally,
I
started
with
trying
to
figure
out
a
way
to
like
incorporate
the
existing
gcp
plug-in,
like
the
external
gcp
plug-in
into
pre-submit.
Somehow
and
like
a
dt,
did
a
bunch
of
work
as
well
on,
like
configuring,
proud
jobs
and
I
think
for
for
ga.
B
But
for
beta
I
took
a
shortcut
where
I
basically
took
like
a
bare
bones:
implementation
of
the
gcp
plugin,
and
I
put
it
into
the
the
gce
cluster
directory
and
then
I
updated
like
the
remote
node
e3
runner
to
like
compile
that
binary
and
then
configure
the
cubic
credit
provider.
So
we
can
run
a
test
for
putting
a
private
image
with
the
cubelet
credential
providers.
B
Disabled
and
basically
like
the
the
bare
bones
plug-in
that
we're
gonna
keep
inside
the
clutch
directory
is
only
for
testing
and
also
what's
nice
is
that
you
can
actually
build
the
whole
plugin
only
using
like
the
standard,
http
client
so
like
in
the
future.
When
we
remove
the
cloud
providers,
it's
not
going
to
be
an
issue
because
it's
like
importing
some
like
third-party
sdk
or
something
like
that.
B
So
like
the
good
news,
is
like
the
tests
passed
and
it
works
with,
like
with
the
disabled
keyboard,
creative
fighters
feature
gate
enabled
as
well,
but
yeah
like
it's
questionable.
If
you
want
it,
it
kind
of
feels
like
two
steps:
four
one
step
back,
because
we're
copying
an
entry
plug-in
in
in
the
main
kk
repo.
A
B
B
And
yeah
right
there,
if
you
scroll
up
a
bit
where
the
cubelet
flags
are
plumbed
in
right,
so
yeah
like
yeah,
like
we're
explicitly
setting
and
like
this,
is
actually
for
all
the
node
tests,
which
maybe
we
need
to
get
that.
But
this
is
effectively
disabling,
like
the
aws
azure
gcp
plugins
in
the
cubelet,
to
really
make
sure
that
we're
not
accidentally
offing
through
the
other
mechanism.
E
E
But
my
general
feeling
is
this
is
not
making.
This
is
not
materially
moving
us
in
in
a
is
not
adding
to
the
problem
we
already
have.
We
are
going
to
have
to
solve
the
the
extraction
problem
for
testing
and
I
don't
think
this
is
really
noticeably
making
that
any
more
difficult
and,
in
the
meantime,
it's
giving
us
test
coverage.
We
need.
A
Yeah,
I
agree
with
that.
I
think
that
we're
testing
the
pieces
that
we
want
to
be
testing
with
this
change
sure
we
have
a
kind
of
ugly.
You
know
cloud
provider
more
additional
cloud
provider
code
but
but,
like
walter
said,
we're
testing
what
we
actually
want
to
be
testing
with
this.
So
you
know
whatever.
B
Okay
sounds
good,
so
yeah,
please
review
it
and
I
think,
with
this
gr
merged.
I
think
we're
also
good
to
go
beta
on
this
feature,
because
we
also,
I
think
the
other
requirement
was
like.
We
have
at
least
two
implementations
out
there,
which
I
think
we
do
and
then
yeah
we're
going
into
ga
we'll
have
a
requirement.
B
I
guess,
like
the
the
g
criteria,
would
basically
be
we
can
set
the
cubelet
credential
provider
or
disable
cubelet
feature
gate
in
basically
like
all
the
proud
jobs
which
would
necessitate
like
walter's
effort
around,
like
fixing
that
circular
dependency
around
how
we
test
with
the
gcp
provider
and
whatnot.
So
I
think
it
will
be
a
good
forcing
function
for
ga.