►
From YouTube: SIG Cloud Provider 2023-09-27
Description
Meeting agenda: https://docs.google.com/document/d/1OZE-ub-v6B8y-GuaWejL-vU_f9jsjBbrim4LtTfxssw/edit#bookmark=id.1h3q0nfeyssb
A
All
right
welcome.
Everyone
today
is
Wednesday
September,
27
2023,
and
this
is
the
Sig
cloud
provider.
Meeting
Sig
cloud
provider
is
a
sub-project
of
kubernetes
sigs
and
as
such,
we
follow
their
meeting
guidance
which
basically
boils
down
to.
Please
treat
everyone
as
you
would
expect
to
be
treated
and
explicitly
so,
please
be
kind
to
each
other
and
also
please
raise
your
hand
if
you'd
like
to
be
called
on
and
I
will
call
on
you
so
yeah
with
that.
A
Let's
dive
into
our
agenda
here
so
I'll
go
through
the
sub
project
updates
and
the
agenda
and
then
we'll
come
back
to
the
triage
at
the
end,
because
I've
got
a
topic
that
might
take
some
time
here.
A
So
I
guess,
like
John
I,
see
you've
got
an
update
for
IBM
Cloud
here.
Why
don't
you
take
it
away.
B
Yeah
just
announced
that
our
GA
of
the
1.28
and
I
managed
offering
came
out
last
week
on
921.,
so
that
was
it.
A
Awesome
that's
great
to
hear.
Does
anybody
else
have
updates
that
they'd
like
to
provide
today.
D
Yeah
I
decided
yeah.
Add
it
to
the
my
agendas
but
yeah
I
think
so
as
part
of
the
extraction
migration,
like
necklace,
has
a
cap
which
she
started
to
provide
a
framework
on
the
cloud
codebase
to
start
of
a
book
server.
D
My
PR
is
mostly
extension
on
top
of
an
exchanges
to
rested
of
a
book
as
a
validating
web
configuration
and
the
episode
that
way
like
now.
With
this
like,
we
should
be
able
to
use
the
I
mean
the
cloud.
Progress
should
be
able
to
use
the
framework
to
automatically
just
write
the
webwork
logic
and
get
the
auto
registration
terms.
A
That's
great
to
hear
I
guess,
since
we're
kind
of
jumping
ahead
to
this,
should
we
do
we
add,
do
we
need
to
add,
like
triage,
accepted
to
this
and
put
a
priority
on
it
or
what's
the
process
here,
I
guess.
A
A
We'll
see
how
that
goes,
all
right,
okay,
cool,
so
we've
got
that
one
accepted
and
do
we
need
to
add
the
priority
to
it
then.
A
D
Yeah,
that's
correctly:
yes,
okay,
great
yeah,
maybe
once
if
this
is
done,
we
can
I
think
the
next
step
after
this
would
be
for
the
cap
to
migrate
the
Web
book
to
individual
Cloud
products.
A
Great
so
I
guess
for
anyone
attending
or
listening
to
this
meeting.
If
you
are
curious
about
this,
please
take
a
look
at
the
at
the
pr.
We've
got
Linked
In
the
agenda
and
then
it
links
through
to
this
to
this
Camp.
Basically,
okay,
great.
Does
anyone
have
questions
or
comments
about
that.
A
All
right,
I'm
not
seeing
any
hands
so
yeah,
Kirsten,
I,
guess
we'll
just
try
and
get
some
reviews
on
that
and
maybe
bring
it
up
again
at
the
next
meeting
or
something.
A
Okay,
I'm
not
seeing
any
hands
going
up.
So
let's
move
on
to
the
regular
agenda,
so
we
kind
of
we
kind
of
covered
Kirsten's
topic.
I
didn't
know.
If
there
was
anything
else
you
wanted
to
add
to
that
Kirsten
or
did
we
did
we
pretty
much
address.
D
A
Formatting
here,
okay,
so
you
know
I've
been
doing
a
lot
of
work.
Looking
at
this
looking
at
updating
the
cap
and
we
need
to
answer
a
few
of
the
Readiness
questions
and
then
a
couple
questions
have
come
up
on
review
that
I
wanted
to
bring
here
to
the
Sig
to
see
what
people
thought
so
I'll
just
I'll
start
with
these
questions
on
review,
because
they're
here
and
then
I'm
going
to
switch
to
another
terminal.
So
we
can
look
at
the
other
questions.
A
So
Mo
Khan
had
a
few
things
that
he
brought
up
here
and
some
of
these
I'm
still
working
on.
But
you
know
we
got
down
to
a
discussion
where
I
think
you
know.
Moa
has
kind
of
two
questions
here.
One
is
about
doing
like
the
forward
migration
and
then
the
backward
migration
you
know
do
we
have
any
tests
that
show
that
this,
wouldn't
that
this
wouldn't
break
anything
by
going
forward
and
backward
like
that
and
I
think
you
know
part
of
what
he's
this
comment.
A
That
I'm
highlighting
here
is
where
we
kind
of
got
into
this
discussion,
and
you
know
I
thought
that
he
was
talking
about
the
code
in
the
KCM,
but
I
think
what
MO
is
asking
is
like.
When
we
do
the
migration
to
the
external
Cloud
providers,
do
we
have
any
tests
that
are
showing
the
migration
to
them
and
then
the
migration
back,
that
nothing
is
getting
broken
in
these
fields
and
I?
Don't
think
we
do
have
any
tests
like
that
and
I'm
not
sure
what
we
should.
A
You
know
how
we
should
respond
to
that
and
then
the
other
question
he
asked
me
as
well
was
kind
of
getting
into
like
are
we
you
know?
Have
we
done
anything
to
confirm
that
the
new
code
is
not
creating
more
load
than
the
old
code
and
I'm
a
little
bit
I'm
a
little
bit
confused
by
this
one,
because,
like
I
mean
was
where
we
started,
was
the
the
old
code
became
the
new
code?
But
then
those
projects
have
become
kind
of
external
projects
managed
separately
by
each
of
the
providers.
A
So
it
feels
a
little
bit.
You
know
kind
of
out
of
scope
for
me,
and
you
know
where
did
I
talk
about
this?
You
know
like
it
feels
a
little
bit
out
of
scope
for
me.
You
know
his
original
question
was
you
know,
whatever's
been
taken
to
make
sure
the
new
code
doesn't
create
a
bunch
of
extra
load
that
the
old
code
didn't
and
since
we're
moving
to
external
CCMS
at
this
point,
doesn't
that
become
like
a
purview
of
those
CCMS
like?
A
A
So
those
were
two
of
the
big
questions
that
came
up
on
review
that
I'm
I'm
kind
of
struggling
with.
Does
anybody
have
have
thoughts
or
comments
about
these.
C
C
Gonna
add
load,
but.
B
C
Well,
what
we
can
say,
though,
is
that
all
of
the
controllers,
like
the
core
controllers,
like
they're,
actually
the
exact
same
implementation
that
we
have
in
tree
like
the
controllers
themselves
and
the
calls
they
make
to
the
API
server
are
all
the
same.
And
the
main
difference
really
is
the
change
in
the
actual
like
provider
implementation.
C
So
if
you
call
like
one
of
the
methods
and
the
interfaces,
it's
possible
that,
like
the
external
implementation,
may
do
like
extra
calls
compared
to
the
entry
implementation,
but
like
you're,
saying,
like
that's
kind
of
implementation,
details
per
provider,
so
I
don't
know
if
that
helps.
Just
saying
like
we're
not
actually
like
the
actual
core
controllers
are
actually
the
same.
We're
just
moving
the
controller
logic
from
one
component
to
another
right.
C
C
Kind
of
two
approaches
we
can
take
for
that
question.
The
the
first
one
is
like
we.
We
can
run
like
the
scalability
tests
that
we
have
in
the
external
configuration
and
just
kind
of
use
the
data
points
in
that
to
say,
like
you
know,
just
to
catch
regressions,.
C
The
other
point
also,
we
could
just
say
like
hey,
like
we
actually
have
a
lot
of
providers
like
using
already
using
this
in
production,
and
it's
quite
vetted
like
in
based
on
like
real
clusters
and
and
users.
But
it's
not
necessarily
like
a
strong
data
point.
It's
more
anecdotal,
but
in
the
past,
like
that's,
been
a
fine
justification
as
well.
So
yeah.
A
C
Some
testing
we
did
as
well,
but
we
can
try
to
dig
that
up.
D
Yeah,
like
we
had
done
like
I,
mean
this
credibility.
Testing
like
as
part
of
the
not
I
mean
not
just
for
this,
but
we
do
this
as
part
of
every
release
so
that
we
so
as
part
of
the
novel
process
we
had
done
it
even
been
going
to
the
external
protosses.
So
yeah
we
didn't
so.
This
has
been
tested
scale
from
AWS
set.
A
Okay,
this
is
good
and
thank
you
Andrew
and
Kirsten,
like
I'll,
try
to
form
some
of
this
into
a
response.
I'm
guessing
I'm
thinking,
Moe's,
probably
going
to
get
back
to
me
today.
So
I'll
just
keep
the
conversation
going,
but
if,
if
either
of
you
could
take
a
review
on
this
and
maybe
add
your
thoughts
into
this
discussion
chain
like
that,
would
that
would
really
be
helpful.
D
A
Thank
you
so
much
Kirsten
and
then
the
other
question
kind
of
the
first
one
I
started
with
was
kind
of
like
you
know,
Moe's
asking
like
what
tests
do
we
have
that?
Do
we
have
anything
to
tests
the
upgrade
and
then
downgrade
mechanism?
A
You
know
this
is
where
this
is
where
I
was
kind
of
having
a
little
bit
of
trouble
because,
like
you
know,
Moe
was
saying,
like
you
know,
say:
a
field
in
a
resource
was
always
set
to
five
with
the
Legacy
code,
and
then
it
gets
sent
to
six
by
the
new
code
on
a
downgrade.
The
resource
would
still
exist,
but
now
the
old
code
would
see
a
six
instead
of
a
five.
You
know,
which
you
may
not
understand
now
kind
of
my
response
here.
A
You
know
I
tried
to
restate
the
problem
because
I
wasn't
quite
understanding
it,
but
I
said,
like
you
know
the
former
entry
controllers.
You
know
that
became
the
external
CCMS
I,
don't
think
they're
actually
making
modifications
to
the
node
or
service
objects
in
a
way.
That
would
do
that,
but
I
don't
think
we
have
a
test
that
demonstrates
that
does
any.
Is
there
any
like
data?
We
could
back
this
up
with
or
I
mean
I
I
had
kind
of
assumed
that
everything
was
just
working.
A
The
same
we'd
already
migrated
like
the
labels
and
everything
like
all
that
work
had
been
done
before
things
got
switched,
but
I
don't
know
if
it's
possible
that
a
CCM
could
modify.
You
know
like
a
node
or
service
object
in
a
way
that
would
create
this
incompatibility
and
I'm,
not
sure
how
we
demonstrate
this.
C
A
He
didn't
have
a
concrete
field
here.
That's
why
I
was
trying
to
turn
it
into
a
concrete
example,
because
I'm
not
I'm
I'm,
not
sure
that
I'm
not
sure
how
much
context
MO
is
bringing
into
this
like
I,
I
thought
I.
Think
maybe
he's
looking
at
this
like,
like
yeah,
we're
upgrading
some
component
that
has
a
an
API
crd
associated
with
it
that
they're
tightly
coupled
or
something.
But
it's
not
really
the
case
here.
As
far
as
I
understand.
C
We're
just
importing
over
like
the
same
controllers,
between
components
and
the
the
logic
I
mean
the
backing
implementation
might
change
what
specific
values
that
the
cloud
provider
would
return
to.
Those
controllers
which
could
change
certain
field,
values,
I,
guess
right,
but
I,
don't
see.
It's
not
I,
can't
think
of
anything
that
would
be
considered
like
incompatible
or
breaking
or
like
unusual,
because
it's
it's
for
the.
A
C
Not
changing
in
any
way
and
as
far
as
I
know,
like
most
providers
like
their
current
implementations,
are
based
on
the
Legacy
implementations,
with
maybe
some
like
minor
iterations
right
from
there,
but
for
the
most
part
I,
don't
see
why
it
would
be
different.
Yeah.
A
That
wouldn't
be
aware
that,
like
that's
all
kind
of
coded
into
the
core
framework,
but
it's
possible
that
someone
could
create
a
CCM
that
didn't
you
know,
that's
the
only
thing
I
could
think
of.
But
again
this
puts
the
onus
back
on
like
okay,
we're
part
of
the
activity.
We're
doing
here
is
migrating
these
CCMS
to
external
repos.
That
will
be
managed
by
different
project
teams,
like
these
components
are
no
longer
being
shipped
as
part
of
kubernetes,
so
the
user
has
some
responsibility
to
find
an
external
CCM,
and
then
the
question
is
like
okay.
A
C
A
C
The
Azure
angle,
if
there's
a
specific
issue
or
concern
with
the
Azure
implementation,
that
would,
where
that
could
be
more
possible
I
think
it'd
be
good
to
get
more
examples.
C
I
guess
like
I
guess
he
does
have
a
good
point,
though,
that
maybe
we
should
oh
it's
possible
that
Jared
might
have
added
a
test
for
this
already.
Actually,
we
just
have
to
dig
it
up,
but
a
test
that
does
some
like
run
some
basic
edoe
tests
like
while
it's
entry
and
then
migrates
to
this,
like
the
CCM
leader,
election
migration
logic
to
auditory
right.
A
But
then
you'd
have
to
go
and
then
you'd
have
to
run
it
again
then
migrate
back
down.
You.
C
A
Currently,
you
know
we
only
deep
deeply
test
the
cloud
controllers
on
GCE
right
and
with
a
subset
of
of
those
tests
running
on
AWS
and
a
subset
of
those
tests
running
on
Azure
and
as
far
as
I
know,
like
openstack
and
vsphere
used
to
be
part
of
the
entry
and
I,
don't
think
those
were
ever
tested
using
the
core
tests,
so
I
mean
we
had
code
in
there
before
that
was
never
tested
and
I
totally
I
love,
adding
more
tests
and
stuff
like
that.
But
it's
starting
to
get
to
the
point
where
it's
like.
A
Okay,
there
is
no
default
cloud
provider,
that's
packaged
with
kubernetes,
so
a
choice
first
has
to
be
made
that
you're
going
to
run
an
external
CCM
and
then
now
you
have
to
run
this
upgrade
downgrade
test
and
it's
I
think
that's
just
pushing
our
testing
kind
of
Beyond
where
we've
been
in
the
past.
So,
like
I'm,
not
saying
it's
a
bad
idea
to
do
that
kind
of
testing,
but
it
is
a
big
effort.
A
Yeah
like
we.
We
do
this
kind
of
testing
on
openshift
internally,
a
red
hat,
but
we
have
all
sorts
of
operators
that
we've
put
in
place
to,
like.
You
know
to
manage
the
external
to
internal
CCM
like
Transit
and
whatnot
and
like
I,
don't
I,
don't
think
we
have
anything
equivalent
in
kubernetes,
so
someone's
gonna
have
to
write
that
test
and
manually
write
all
these
like
transactions
to
like
upgrade
the
cluster
then
downgrade
the
cluster.
You
know
that's.
B
C
Yeah
and
if
it
helps
I
need
to
check
with
jakub
but
I'm
fairly
sure
in
gke,
like
we,
we
have
tested
like
the
migrate
to
CCM
and
then
migrate
back
to
the
old
like
because.
B
C
A
C
A
So
I
I
appreciate
this
conversation.
Deeply
and
I
will
try
to
express
these
points
in
the
pr
I
have
a
feeling
that
I
have
a
feeling
that
I'm
not
expressing
myself
in
a
way
that
is
coming
across
very
well
to
Mo,
though,
because
I'm
I'm,
failing
to
make
some
of
these
points
so
again,
like
any
help,
would
be
greatly
appreciated.
C
Yeah
I
mean
we
could
we
could
always
if
it
helps
just
like
set
up
like
a
separate
meeting
with
Mo.
C
A
B
A
A
You
know,
I
think
I've,
I've,
gotten
most
of
I
think
I've
gotten
most
of
the
major
points
answered
or
updated
in
here,
but
then,
when
I
was
going
through,
this
I
noticed
you
know
in
the
production
Readiness
questionnaire,
and
let
me
just
can
everybody
read
that
is
that
like
I
want
to
make
sure
it's
not
too
small
yeah,
but
what
I
noticed
was
while
scrolling
through
the
production,
Readiness
review,
then
I
started
to
notice
all
these
TBD
for
beta
TBD
for
beta
so
and
I
started
to
look
at
him.
A
Like
you
know,
how
can
a
rollout
fail?
Can
it
impact
running
work,
I
mean
I,
guess
I
at
some
level
like
I,
can
try
to
go
through
and
answer
some
of
these,
but
like
other
ones
like
the
specific
metrics
to
inform
a
rollback
like
I
could
use
some
help.
Answering
some
of
these
I
don't
know
if
we
want
to
try
to
go
through
these
here
in
this
meeting,
I'd
be
happy
to
if,
if
folks
are
up
for
it,.
A
C
Are
like
misconfiguration
right
right
so
that
that
includes
like
yeah,
like
maybe
you
missed,
like
the
our
back
part
or
the.
C
A
A
A
A
Okay,
is
there
any
sort
of
metrics
that
people
should
be
looking
for
to
inform
like
what
to
say
they
should
do
a
rollback?
I,
don't
I
mean
I,
don't
think
we
have
any
metrics
that
are
exported.
C
I,
don't
know
about
metrics,
but
some
some
things
to
check
would
be
like
if
your
node,
like
nodes
are
failing
to
register
like
there
should
be
generic
like
cumulative
metrics.
Probably
around
that
register
load
balancer
like
yeah
load
balancer
calls
failing
or
not
happening.
A
C
A
Right
I'll
go
back
and
look
at
those
as
I'm
updating
this
so
I
mean
this
tab.
This
has
been
tested
independently,
right,
upgrade,
upgraded
rollback,
but
I
don't
think
we
have
anything
in
CI.
C
A
So
is
the
rollout
accompanied
by
any
deprecations
and
or
removal
of
features,
I
think
yes
for
deprecations,
but
and
we
removed
some
of
the
providers
right.
D
A
Okay
so
I
guess,
if
the
if
the
workloads
are
if
the
workloads
are
using
load,
balancer
type
services,
that
would
be
like
the
only
way.
C
A
But
like
would
a
workload,
I
saw
what
you're
saying
they
could
look
at
the
logs.
C
A
C
A
C
Successfully
initialized
so
like.
B
C
Around
like
pod
scheduling,
so
if
the
CCM
was
not
working
similar
to
the
last
Point
CCM
was
not
working
and
then
the.
C
Yeah
Road
balancer
again.
A
C
B
C
Like
we
as
maintainers
of
the
kubernetes
project,
can't
prescribed
like
specific
slos,
because
it's
dependent
on
the
cloud
provider,
their
infrastructure
and
the
implementation
of
the
cloud
provider
so
for
for
the
the
pr
questions
that
are
like
asking
like
like
examples
of
like
specific
slos,
like
we
can
say
like
these,
are
things
to
consider
but
exact
objectives.
C
A
A
I'll
check
the
current
metrics
and
then
maybe
we
can
come
back
to
that
one
does
this
feature
depend
on
any
specific
Services
running
in
the
cluster
I
mean:
did
the
CCMS
themselves
count
as
a
specific
server
specific
service.
A
B
B
A
A
Yeah
this
was
the
quest.
This
was
the
question
that
that
that
set
Mo
off
here,
like
he
said
we'll,
enabling
using
this
feature
result
in
any
new
API
calls.
A
I'd
highlight
that
because
it
was
I,
just
remember
that
one
okay,
let
me
see
if
we
got
any
more
here.
Okay,
how
does
this
feature
react
if
the
API
server
and
etcd
is
unavailable,
I
guess
it
won't
work
right.
A
C
A
A
What
are
the
other
known
failure?
Modes.
A
C
C
I
can
think
of
is
like
a
failure
mode
being
like
the
cloud
provider
back
end
is
down
or
like
you're
being
like
you.
A
B
C
Is
not
necessarily
to
have
answers
to
all
these
questions
like
or
like
a
good
answer
to
all
these
questions,
but
my
understanding
is
that
it's
just
to
ensure
that
we've
kind
of
asked
ourselves
some
questions
and
thought
about
it.
So
I
think
what's
important.
Is
that,
like
whatever
we
put
in
here,
it's
kind
of
just
proving
that
we've
kind
of
like
done
the
exercise
of
thinking
through
some
of
these.
C
So
so,
for
for
this,
one
I
would
say
like
what
steps
should
be
taken
like
the
obvious
ones,
are
like.
Okay,
well
check
the
system
logs
like
what's
what's
the
issue
right,
if
it's
authentication
issue
like
make
sure
your
R
back
is
correct.
If
it's
a
quota
issue
with
your
cloud
provider,
apis
like
ensure
you
have
like
ensure
you
increase
your
quota,
make
sure
you
have
enough
quota
if
it's.
A
C
If
it's
like,
like
a
full
like
there's
a
bug
or
like
it,
just
doesn't
work
as
a
last.
We
can
mention
here
like
Last.
Resort
is
just
roll
back
like
you,
can
flip
the
picture
gate
to
back
to
off
and
then
go
back
to
using
entry
and
then
try
the
rollout
again
once.
C
A
Cool
I
think
that's
good
and
thank
thank
you
so
much
Andrew
I'm,
just
going
to
scroll
I,
think
that's
it
now.
So
I
will
update
this
again.
You
know
one
of
the
things
I'm.
One
of
the
things
I'm
concerned
about
here.
Is
that,
like
we're,
you
know
next
Friday
is
the
is
the
freeze
date
for
this
and
I'm
hopeful.
A
We
can
get
this
updated
and
merged
by
then,
but
you
know
there
are
still
questions
coming
up
on
the
on
the
review
and
I'm,
not
necessarily
sure
that
I
have
like
good
answers
for
all
of
them.
So
I
will
continue
to
update
this
and
continue
to
follow
it
and,
of
course
any
reviews
will
be
helpful.
C
Yeah
I'll
try
to
do
another
pass
on
the
pr
this
week.
A
Awesome.
Thank
you.
Thank
you
for
that
Andrew.
Okay.
So
let
me
just
let's
check
our
I'm
gonna
go
back
to
the
let's
just
check
our
open
issues
to
see
if
we
have
anything
that
needs
to
be
triage,
I
think
we
got
everything,
but
it's
always
good
to
double
check
right.
A
And
a
Sig
cloud
provider-
gcp
reboof
like
I,
don't
know
Andrew
are
you
have
you
seen
this
yet
or.
C
B
C
C
Is
it
in
the
comments?
Does
it
say
what
the
failure
is
like
the
reason.
A
C
All
right
can
I
just
read
that
comment
above
this
history
for.
C
C
Again,
for
us,
like
yeah,.
B
A
I'm
sure
everyone's
entertained
by
me,
fumbling
around
GitHub,
okay,
Cloud,
node,
lifecycle,
controller
shutdown
in
Clyde
murder
fails
for
node
with
no
provider
id.
C
A
B
A
Okay,
I
think
those
are
the
only
new
issues
that
we
have
so
cool
and
that's
the
end
of
our
agenda
did
anybody
else
have
topics
they
wanted
to
bring
up
before
we
go
here.
A
Okay,
going
in
three
two
one:
all
right,
thanks
for
coming
out
everyone
and
keep
an
eye
on
the
reviews
going
on
and
I
guess,
we'll
maybe
see
you
all
next
week
for
the
extraction
meeting.