►
Description
Plans for finishing the working group, PR Review, KEPs
A
Hi
welcome
to
the
january
13th
2022
sig
cloud
provider,
cloud
provider,
extraction
working
group
meeting.
As
always,
these
meetings
are
intended
to
be
inclusive,
polite
and
considerate
of
all
one's
fellow
contributors.
So
you
know,
please
do
so.
We
we
follow
all
of
the
standard
rules
of
the
cncf
kubernetes
as
regards
to
behavior.
So
with
that,
let's
go
ahead
and
get
started.
El
mico.
I
believe
you
have
two
prs
that
you
need
reviewed,
one
of
which
is
on
me.
So
I
apologize.
A
I
will
go
ahead
and
take
a
look
at
that.
The
other
one
is
on
bowie
in
these
days
of
working
remotely.
I
do
not
have
as
easy
and
access
to
just
wander
down
the
hall
and
ping
bowie,
as
I
usually
do
so
I
don't
have
a
particularly
easy
way,
I'll
take
a
look
at
that
and
see
if
it
really
requires
bowie.
B
Yeah
someone
else,
I
think,
maybe
jerry
like
assigned
it
to
him,
someone
someone
in
that
thing
assigned
it
to
bowie,
so
I
I
just
assumed
that
was
the
right
call.
It
probably.
B
So
let
me
see
what
I
can
do:
okay,
yeah
yeah,
I
was
wrong.
Jayla
was
on
the
first
pr
there.
It
looks
like
vachezlav
recommended
bowie
for
the
second
one.
A
Me
see
what
I
can
do
with
those
all
right
cool.
Thank
you
cool.
So
this
is
the
cloud
provider
extraction
meeting,
so
I
would
like
to
have
a
little
bit
of
a
planning
session
with
regards
to
how
we
can
actually
finish
this
working
group.
I
would
love
to
actually
shut
this
working
group
down.
A
A
So
the
first
long
poll
is
there
is
a
particular
admission
controller
that
runs
in
the
cube
api
server,
that
is
responsible
for
tagging,
a
certain
storage
systems,
and
that
is
a
thing
that
is
a
cloud
provider
specific.
A
So
we
had
a
long
discussion
about
six
months
ago
and
with
six
storage
and
came
to
the
conclusion
that
the
way
we
wanted
to
move
forward
on
this
was
to
find
an
easy
way
to
allow
cloud
providers
to
have
web
hook
a
standardized
web
hook
to
deal
with
this.
So
I
went
ahead
and
I
wrote
a
cap
that
basically
described
how
we
could
extend
the
ccm
to
be
able
to
run
web
hooks,
and
let
me
be
really
clear
because
there
was
a
little
bit
of
contention
on
this.
A
It's
not
even.
I
am
not
necessarily
a
fan
of
running
the
web.
Hooks
in
the
same
process,
I
think
for
some
distributions
that
can
be
fastest
and
easiest.
A
But
since
what
we
have
is
a
framework,
the
idea
was
not
so
much
that
you
had
to
run
them
in
the
same
process,
but
that
we
would
provide
a
framework
which
would
allow
for
you
to
either
set
up
a
controller
manager
set
up
a
web
hook
or
set
up
a
process
which
is
both
and
that
it
would
be
up
to
the
cloud
provider
to
decide
which
way
they
wanted
to
go,
and
that
seems
fairly
reasonable
and
easy.
A
So
I
know
we've
had
multiple
people
who
thought
they
were
going
to
be
able
to
but
life
intruded,
and
it
hasn't
happened
at
this
point.
It
is
by
far
one
of
our
two
longest
pulls,
and
so
I
thought
it
would
be
good
to
bring
it
up
and
see
if
anyone
had
particular
feelings
about
trying
to
get
this
to
happen.
C
A
C
A
C
A
A
C
D
Think
that
I
think
the
intention
was
like
we're
gonna
like
we're
gonna,
build
that
one
web
hook
in
controller
manager
but
like
to
build
that
web
hook.
Specifically,
we
have
to
add
the
initial
wiring
in
the
controller
manager
to
support
webhooks
at
all
so
like
yeah.
Maybe
this
is
the
best
for
it,
but.
C
Initially
yeah
I
mean
I
see
that
I'm
just
I'm
just
trying
to
point
out
that
running
the
web
hook
in
the
controller
doesn't
to
me
seem
super
essential.
So
if
this
is
really
blocking
progress,
are
we
willing
to
compromise
and
say
let's
just
build
a
simple
web
hook
that
people
can
use
and
you
have
to
forget
how
to
run
it
and
if
anybody
wants
to
step
up
and
make
that
easier
great.
But
let's
keep
moving
forward.
E
C
E
A
B
E
So
I'm
happy
to
take
some
of
this
work,
and
I
we
had
this
conversation
before
like
I
just
can't
take
the
entire,
like
you
know
like
a,
I
don't
think
I
can
own
the
entire
like
building
out
of
framework,
but
I
could,
if
we
can
split
this
up
into
you
know
like
smaller
pieces,
and
you
know
I'd
be
happy
to
take
like
proof
of
concept.
Web
hook
type
thing
that
seems
much
more
doable
with
less
time.
A
A
It's
as
odd
as
it,
I
think
the
tagging
is
all
the
same.
It's
a
it's
a
question
of
filter,
filtering
it
and
getting
the
tack
right.
So
I
believe
it's
a
question
of
looking
up
the
the
persistent
volume
determining
whether
it
is
a
legacy,
persistent
volume
and
then
getting
a
cloud
provider,
specific
value
which
is
going
to
be
put
in
the
tag,
so
that,
I
believe,
is
what
what
is
cloud
provider
specific.
I
think
the
impactful
of
writing
the
value
to
the
pv
is
the
same
for
everyone.
A
It
looks
like
andrew
was
nice
enough
to
send
a
link
to
the
relevant
pv
admission
controller.
D
So
walter
I'm
wondering
if
so
I
do
agree
like
this-
is
important
because,
like
we
would
be
breaking
some
storage
stuff.
If
we
didn't
have
this
right,
but
I'm
wondering
for
like
124,
because
the
deadline
is
coming
up.
Do
you
think
we
should
spend
some
time
now
and
like
actually
maybe
like
list
all
the
caps
in
flight,
including
this
one
and
then
stack
rank
them.
A
E
D
D
E
Should
definitely
try
to
get
that
to
beta,
in
my
opinion,
we're
we're
testing
our
implementation
now.
Is
there
anybody
else.
C
Yeah
we've
tested
it
so,
but
it's
not
it's
not.
C
A
Okay,
so
two
quick
questions
on
that.
I
think,
but
I'd
have
to
go
double
check.
No,
no,
that's
fair!
So
I
have
a
couple
of
quick
questions.
One
did
we
ever
get
to
the
point
where
the
windows
was.
A
C
Official
images,
no,
we
don't
yet
it's
the
the
idea
was
to
get
the
basal
change
in,
which
makes
it
a
lot
easier
to
do
that
and
then
get
the
image
support
fully
added.
A
No,
it's
fair.
It's
just
my
thought
had
been
that
if
we
wanted
to
get
what
you
were
mentioning,
which
is
not
an
official
automated
test,
we
could
get
something
into
tests.
The
to
run
run
the
the
standard
tests
with
our
credential
provider,
but
that's
a
lot
easier.
If
there's
a
a
official
ins
official
version
of
that
image,.
C
Yeah
we
can,
we
can
prototype
the
flow
because
we
have,
we
have
a
a
stopgap
image
published
or
being
published,
so
we
can
at
least
get
all
the
infrastructure
working
and
then,
when
we
get
repeatable
images,
we
can
like
automate.
A
A
Bridget's
not
here
today,
but
I'm
guessing
she'll,
be
on
wednesday
fall,
so
we
can
check
with
widget
on
wednesday.
F
All
right,
any
other
caps.
D
The
controller
manager
leader
migration-
one-
I
don't
know
if
we're
doing
anything
but
I'm
just
listing
the
caps.
A
Yeah,
no,
no,
absolutely!
I
believe
that
is
in
beta
and
I
don't
think
we're
going
to
be
in
the
position
to
take
that
one
to
ga.
G
Why
not
like
I'm,
I'm
already
working
on
testing
that
with
kelps?
You
know
this
one
is
close
to
down
so
as
long
as
we
have
a
like
as
long
as
I
can
so.
I
I
already
add,
like
official
external
system,
support
for
collaborative
and
indirect
is
already
there.
A
If
you
want
to
try
and
push
for
ga
on
the
cap,
we
can
do
that,
but
I
don't
think
that
all
of
the
upgrade
testing
across
all
the
platforms
is
going
to
be
in
the
position
to
be
ready
in
this.
F
D
F
A
F
All
right,
unless
someone
has
any
other
ones
they
want,
do
you
have
any
other
caps.
B
D
A
We
need
something
we
can
argue
whether
it's
a
cap
or
not,
I
have
talked
once,
is
stephen
augustus,
who
ran
away
like
in
fear
with
the
things
that
we
were
going
to
want
from
sig
release
as
part
of
this
effort,
but
yeah.
I
do
actually
think
this
is
I
I
don't
know
if
that
means
it
is
a
cap,
but
this
is
clearly
a
cross
sig
effort.
A
So,
okay,
where
were
you
last
with
this
joe.
A
Okay,
so
I
had
jd
working
on
it,
but
jd
has
since
left.
So
from
my
perspective,
just
there
are
a
few
action
items,
so
the
first
is,
I
would
very
much
like,
and
there
is
a
pr
I'll
pick
it
up
and
move
it
forwards,
but
there
is
a
pr
to
allow
testing
to
turn
off
cloud
provider
in
kk
for
in
the
cluster
up
script.
A
That
will
allow
us
to
create
a
test,
a
suite
or
whatever
automated
test
suite.
That
will
show
what
happens
when
you
turn
on
the
relevant
feature
flags,
and
we
should
get
a
list
of
everything
that
breaks.
A
So
I
think
one
item
one
is
enumerating
the
list
of
items
that
break
that
need
to
be
dealt
with
and
probably
engaging
with
the
other
cigs
that
are
going
to
need
to
do
to
to
at
least
tell
us
how
they
want
to
deal
with
each
of
those
tests.
A
Dealing
with
those
tests,
in
my
mind,
comes
down
to
one
of
three
items.
One
of
three
paths
forward
so
path,
one,
the
easiest
is
they
say
we
don't
need
this
test
anymore,
delete
the
test,
I'm
guessing,
that's
not
where
we
most
of
these
will
land,
but
it
is
certainly
an
option
option
two
is
the
the
belief
is
that
tests
shouldn't
require
anything
cloud
provider
specific
which
means
that
someone,
preferably
the
sig
who
owns
the
test,
needs
to
investigate
why
the
test
is
failing
and
work
out
how
to
make
it
not
fail.
A
Common
examples
of
this
that
come
to
my
mind,
would
be
things
like
if
it
as
an
example
on
leagues
were
owned,
if
the
if
it
used
an
image
that
only
exists
in
gcr.io
or
if
I
know
a
lot
of
the
storage
and
networking
tests,
make
certain
assumptions
about
the
vm
that
they're
going
to
run
on
and
as
a
result,
you
know
the
test
may
not
be
testing
anything
cloud
provider
specific,
but
the
methods
that
it's
using
to
do
the
tests
are
cloud
provider
specific.
A
So
an
example
I
ran
on
ran
across
from
me.
I
forgot
it
was
networking
or
node,
but
there
are
tests
that
assume
that
there
are
certain
networking
start
and
stop
scripts
that
live
in
a
particular
location
that
you
can
sudo
to
restart
the
networking
and
make
sure
that
recovery
works
properly.
A
Obviously
we
want
networking
to
restart
properly,
but
that
you
know,
assuming
that
the
the
vm
is
set
up.
A
particular
way
is
a
particularly
cloud-provided
specific
task
and
then
the
third
option-
and
this
is
where
lkg
clicks
in
is
we
say
look
this
test
is
cloud
is
or
is
believed
to
be
sufficiently
cloud
provider
specific
that
we
think
we
could
just
move
this
code,
this
testing
into
a
cloud
deployment,
or
at
least
you
know
the
the
reference
implementation
and
that
we
will-
and
this
is
where
lkg
gets
a
little
complicated.
A
But
the
idea
here
is,
I
make
a
change
that
change
gets
shipped
to
the
cloud
provider,
the
cloud
provider
rebuilds
everything
they
need
to
build,
deploys
and
then
runs
the
test
and
then
ships
the
resulting
test
results
back
to
kk
for
a
given
release,
and
I
think
that
is
a
large
portion
of
what
the
lkg
proposal
that
joe
has
is
about,
and
I
think
that
is
option
number
three-
that
we
can
look
at
with
tests
that
fail.
A
Awesome
all
right
so
yeah.
I
definitely
think
that,
by
the
way,
when
I
mentioned
earlier
that
it's
one
of
the
two
long
poles
in
my
mind,
that
is
the
other
of
the
two
long
poles
and
I
think
it's
actually
probably
probably
pretty
important
that
we
at
least
get
the
list
of
tests
that
we
believe
to
be
a
problem
generated.
D
Cool,
so
I
prioritized
the
four
things
I
did.
A
rough
prioritization
of
the
four
things
we
talked
about.
Does
that
order
make
sense
in
the
agenda.
D
Like
because,
like
it
seems
like
we
have
a
like,
it
seems
like
resourcing,
is
the
biggest
farm
right
and
like,
if
there's
anyone
available
to
do
work,
they
should.
D
A
Yeah
and
the
other
thing
is,
while
I
do
think,
the
controller
manager
leader
migration
is
important,
it's
already
at
beta
right
right,
and
so
theoretically,
you
know
if
it
doesn't
move
to.
If
we,
even
if
we
don't
make
ga
criteria
this
time
around,
it's
there
are
two
more
releases
before
it
becomes
a
blocker.