►
From YouTube: SIG Cloud Provider 2023-08-16
Description
Meeting agenda: https://docs.google.com/document/d/1OZE-ub-v6B8y-GuaWejL-vU_f9jsjBbrim4LtTfxssw/edit#bookmark=id.xdp8r8e6ot6b
A
All
right
welcome
I'm
gonna
get
the
document
shared
here.
Let's
see
we
are
looking
for
Sig
Club
provider.
A
Yes,
welcome
to
kubernetes,
say
cloud
provider,
it
is
August
16
2023.,
please
go
ahead
and
add
yourself
to
the
list
of
attendees
as
possible,
and
I
will
also
clarify
that
we
are
of
course
governed
by
the
kubernetes
code
of
conduct,
be
kind
and
considerate
of
others.
A
Okay,
today
I
did
take
a
look
at
open
issues
and
just
had
one
that
I
did
triage
except
on
and
if
it
looks
like
a
wild
Joel
has
appeared.
So
I
did
put
that
issue
that
I
just
triage
accepted
into
the
topics
for
possible
discussion
today,
looks
like
we
don't
have
a
ton
of
sub-project
updates
in
textually.
Yet
I
will
note
that
for
ours
for
provider,
Azure
I
just
dropped
in
the
links.
A
For
a
couple
weeks
ago
we
did
put
out
new
versions
for
1.27
26
and
25
we're
not
doing
further
releases
for
1.24,
of
course,
per
the
EOL.
Do
we
have
anyone
from
Ali
cloud
or
Baidu?
B
Hi
Luke
so
I
mean
nothing
specific
update,
but
we're
planning
to
start
to
work
on
moving
the
the
storage
web
hook,
things
to
use
in
the
native
way
of
which
only
has
implemented.
A
Right
feel
free
to
put
more
detail
if
you
like,
yeah.
B
A
All
right
and
our
gcp
friends,
what
do
we
have
there.
C
I,
don't
have
much
to
add
I.
Think
I
was
I,
had
a
long
chat
with
Michael
last
during
last
week's
project
meeting
about
pushing
Antonio's
PR
forward
to
use
the
gcp
controller
manager
by
default
in
CI,
so
I
think
yeah,
like
that's
probably
the
biggest
thing
on
my
mind,.
C
D
No,
nothing
there
I
mean
I,
think
there's
some
stuff
on
the
extraction
migration.
But
let's
delay
for
that.
B
Yep
no
updates
for
me
at
this
time.
A
All
right
fantastic:
do
we
have
any
open
stack,
vsphere,
Equinox
I
feel
like
we
don't
have
you
in
alphabetical
order.
We
should
change
that
sometime.
What
do
you
got
for
us
Chris.
E
Nothing
just
my
question:
Below
in
the
agenda.
A
All
right,
yeah
I,
did
move
your
question
because
I
was
like:
oh,
that
is
probably
more
of
a
discussion
topic:
okay,
all
right!
So
without
further
Ado.
Let's
move
on
to
our
agenda
Chris,
you
were
bringing
up
a
question
about
image
hosting.
E
Yeah,
just
we're
kind
of
finishing
up
our
move
into
the
sigs
and
thanks
everybody
for
your
help.
With
that,
the
question
that
came
up
is
we
currently
have
our
own
Docker
and
key
repository,
and
we
were
using
ghcr
from
our
old
from
our
old
repo.
Now
that
we're
in
the
new
repo
is
their
ghcr
hosting
available
in
this
org.
Is
that
cool
to
use,
or
do
we
not
do
that
here
just
trying
to
check
in
and
see
what
the
pattern
is.
C
Yes,
yeah,
we
do
have
like
GCR
Registries
that
you
can
use
for,
say,
hosted
projects,
I,
think
there's
a
I'm
I'm
a
little
I,
don't
remember
like
the
details
for
this
process,
but
we
should
be
able
to
dig
it
up
somewhere.
There
might
be
like
a
staging
process
where,
like
you,
have
to
like
push
like
a
staging
version
of
your
image
and
have
it
soak
some
time
before
it.
You
mark
it
like
public.
But
aside
from
that,
it
should
be.
A
C
C
A
A
And
yeah,
that's
again
with
your,
where
you
have
your
images
and,
of
course,
if
you
use
actions,
I
mean
there's
all
sorts
of
things.
I
feel
like
there
is
sponsorship
available
for
some
of
those
like
higher
tiers
of
stuff,
but
you
probably
do
need
to
get
a
change
or
turned
on
for
a
repo.
A
E
A
E
A
Thank
you
for
the
comments
that
people
provided
to
the
draft
doc
that
El,
Nico
and
I
were
working
on.
I
expect
we'll
get
that
Ian
in
the
next
week
or
so
it's
just.
He
and
I
have
to
sync
up
our
schedules
to
do.
One
final
pass
on
it
make
sure
it
looks
right.
A
A
and
we'll
get
our
delinquent
report
in
and
then
the
other
thing
I
noted
when
doing
issue
triage
is
Joel
put
in
and
I
guess,
I
already
had
it
open
in
a
tab,
but
that's
fine!
You
have
this
draft
pull
request.
Joel.
B
A
F
Yeah
I
just
kind
of
opened
this
as
kind
of
a
mark
in
the
ground,
so
there's
a
linked
issue
with
the
detail.
In
the
background
on
this
leader
election
thing,
I've
spoken
to
the
team
who
are
working
on
this
for
the
cube
controller
manager
and
basically
they've
been
going
through
and
making
sure
that
every
controller
passes
through
a
context
and
uses
that
correctly
so
reviewing
the
the
context
is
passed
through
correctly
to
different
API
calls
and
that
kind
of
stuff
to
make
sure
that
if
it
does
get
canceled
things
shut
down
correctly.
F
So
my
plan
was
to
one
build
this
PR,
which
just
changes
a
single
line
in
the
lead,
Direction
config
into,
say,
AWS
cloud
provider
bring
that
up
start
testing.
It
review
their
code.
You
know
see
you
know.
First
of
the
core
libraries.
F
Are
we
passing
context
through
them
and
kind
of
create
a
list
of
work
to
do
to
get
towards
this?
I
just
haven't
got
very
far
with
it.
Yet.
A
Okay,
good
good,
thank
you
for
getting
that
started
and
Walter.
You
have
some
input.
D
I'm
not
sure
if
it's
input
or
right
question
I
haven't
looked
in
the
leader
election
code
in
probably
two
years,
my
memory
of
the
leader
election
code-
and
this
is
owned
by
APO
Machinery,
but
I
see
you
have
David
eats
on
this
already,
so
that
that
sounds
good
was
that
it
basically
not?
Basically,
it
deliberately
kills
the
process
it's
in
when
it
loses
the
election
to
add
to
make
absolutely
sure
there
are
no
long-running
processes,
like
controllers
that
are
continuing
to
function
when
someone
else
takes
over
I.
D
F
Yeah
kind
of
so
you
have
two
things:
there's
the
release
on
Council,
which
basically
says
if
I
lose
leader
election
I'm
gonna
tell
the
lease
that
I'm
no
longer
holding
it
right.
So
that
means
that
something
else
can
take
over.
Then
you
have
the
on-stop
leading
callback,
which
is
normally
where
someone
goes
in
and
wires,
kaylog.fatal,
F
or
something
equivalent
to
panic
and
exit.
F
So
the
idea
is
that
the
context
gets
shut
down.
All
of
your
processes.
Stop
the
release
on
cancel
happens,
and
then
you
do.
The
EK
log
got
fatal
in
theory,
I
think
once
you've
once
you're
sure
that
release
on
cancers
work.
Incredibly,
you
know
following
your
context,
you
could
avoid
okay,
look,
not
fatal,
but
I
don't
see
other
people
changing
that,
even
though
they've
gone
down
this
route
to
make
sure
all
the
contexts
are
observed,
I
can
double
check
with
David.
D
F
D
No
I
think
that's
great
one
interesting
thing
and
I
don't
know
if
Andrew
has
any
input,
it
would
be
interesting
to
look
when
we
make
this
change.
I
want
to
make
sure
we
wire
it
in
in
a
way,
maybe
I
mean
maybe
it's
documentation
or
read
me,
but
there
are
the
controllers
in
the
CCM
that
we,
the
Sig
own,
and
then
there
are
the
controllers
running
in
it
in
any
given
Cloud
providers.
F
Yeah,
that's
that's
on
the
to-do
list.
As
far
as
I'm
concerned
awesome.
D
F
Yeah
I'll
I'll
have
a
think,
and
maybe
we
can
start
some
like.
Is
there
like
a
central
place
for
like
a
implementer's
guide
for
CCMS
or
some
sort
of
central
documentation.
D
There
is
a
sample
that
lives
under
KK
of
you
know.
Here
is
the
sample
version
and
here's
the
bits
you
need
to
replace
that
was
originally
written
by
I.
Think
CC.
A
D
D
F
Yeah,
so
at
the
moment
the
leader
election
stuff
is
is
kind
of
hard-coded.
If
you're
using
as
a
cloud
provider
app
I've
just
got
it.
So
it
turns
it
on
because
quite
a
lot
of
little
stuff
is
kind
of
hard-coded
in
there.
If
Theory
can
make
it
a
flag,
we
can
make
it
end
user
Choice
if
we
really
wanted
to,
but
certainly
making
it
so
that
each
cloud
provider
can
make
that
decision.
I
think
it's
it's
worthwhile,
so
we
don't
just
force
them
all
in
a
single
release.
So.
D
D
Online,
it's
so
there's
a
regular
leader
election.
That
is
one
that
is
just
for
the
case
yeah
and
there's
one
that
is
just
for
the
CCM,
and
then
there
is
an
optional
leader
election
flag
that
is
used
to
coordinate
the
KCM
and
CCM
when
you
want
to
migrate
controllers
from
one
to
the
other,
and
so
the
one
thing
I
will
throw
out
throw
your
way
is
Joel
is
that
we
probably
need
to
think
through
what
happens
if
either
like.
D
D
Don't
believe
so,
no
damn
it.
Okay,
no
I,
I!
Think
it's
an
I
think
that
is
absolutely
the
Assumption
most
people
would
make.
However,
when
we
originally
wrote
the
feature,
the
idea
had
been
that
there
might
be,
it
was
written
more
generically.
It
wasn't
written
just
for
the
cloud
provider,
migration,
the
cloud
provider.
Migration
was
thought
to
be
the
first,
but
not
necessarily
the
last
time
we
wanted
to
migrate
a
controller.
D
So
as
an
example,
the
idea
had
been,
if
at
some
point
in
the
future
and
I'll
I'll,
do
it
not
a
less
less
vague
one
as
well
at
some
point
in
the
future.
If
some
other
controller
started
need
that
it
runs
in
the
KCM
needs
to
have
cloud
provider
in
a
new
release
that
one
controller
might
need
to
be
moved
over
as
part
of
this,
and
that
might
be
either
done
by
this
sig,
because
it
everyone
needs
to
move
that
to
a
cloud
provider
or-
and
here
I
I
mean
here's
a
concrete
example.
D
D
The
other
thing
is
and
I
don't
know
if
any
anyone
else
makes
this
mistake,
but
we
we
at
Google
actually
have
multiple
controller
managers,
not
just
two
and
at
some
point
I
would
like
to
see
the
us
get
down
to
two
because
we
actually
have
CCM,
and
then
we
have
a
security
controller
manager,
and
yet
another
controller
manager
and
I'd
like
to
see
those
all
Consolidated,
and
so
there
are
I
think
there
are
multiple
places
where
it
would
be
valuable
to
be
able
to
migrate
controllers.
As
you
begin
doing,
consolidation.
A
Right
that
makes
a
lot
of
sense
by
the
way
when
I
was
taking
a
look
before
this
meeting.
I
did
see
this
issue
over
in
club
provider,
Azure
and
I
thought.
Oh
I,
wonder
if
some
of
this
you
know
releasing
and
deleting
things
is
relevant
to
other
folks
as
well,
so
because
I
think
we're
at
the
end
yeah
we're
at
the
end
of
our
planned
agenda.
A
I'm
reading
the
what
what
what
they
expected,
they
expect
the
cloud
controller
manager
not
tried
to
delete
an
IP
of
a
service
that
was
never
provisioned,
allowing
us
to
reuse
the
same
sheriff's
name
when
fixing
the
original
I'm
just
wondering.
If
anyone
else
ever
runs
into
this.
D
A
Sorry
I
think
this
is
azure.
This
isn't
Amazon
and
I.
Don't
know
the
answer
to
that
off
the
off
the
top
of
my
head,
like
it's
certainly
possible
we're
doing
something
very
different
than
everyone
else,
but
that's
why
I
want
to
check
and
see
if
this
sounds
familiar
to
folks
from
any
other
clouds
like
oh
yeah,
we
had
this
problem,
we
solved
it
by.
You
know.
A
And
I'll
put
a
link
in
the
notes
as
well
to
that
but
yeah
and
if
it
doesn't
sound
familiar
then
that's
fine.
We
can
try
to
investigate
more,
but
I
thought
I
would
see
if
it
sounded
familiar
to
others.
F
F
Few
weeks
ago,
I
brought
up
the
topic
of
AWS
and
her
customer
reporting
that
if
they
changed
a
the
load
balancer
from
an
NLB
to
a
clb
and
vice
versa,
they
would
leak
resources
at
the
time
I
hadn't
done
much
investigation
was
just
looking.
Some
initial
kind
of
ideas,
so
I
went
okay,
did
some
further
research
and
have
a
much
better
understanding
of
it.
Now,
if
so,
the
core
service
controller,
when
it's
looking
through
it,
calls
ensure
load,
balancer
and
calls
update,
load
balancer.
F
It
doesn't
really
have
a
concept
of
the
infrastructure.
That's
underneath
it
doesn't
really
need
to.
It
doesn't
have
a
concept
of
the
infrastructure
changing,
but
if
you
do
so,
the
way
that
the
AWS
one
works,
for
example,
and
I
suspect
this
is
similar
for
others.
You
know
it
goes
on
and
it
says
which
load
balancer
am
I.
Looking
for
clb
great
I'll
follow
down
this
path.
Nlb
great
I'll
fall
down
this
path.
F
At
no
point,
does
it
ever
say:
do
I
have
any
existing
NLP
or
clb
when
I
shouldn't,
and
they
always
in
in
AWS
at
least
always
goes
back
to
being
short
load
balancer.
So
my
kind
of
thought
to
fix
this
would
be
that
we
need
to
add
additional
finalizers
right.
So
if
you
had
a
finalizer
that
said,
AWS
blah
blah
blah
I've
created
a
clb
and
then
one
that
said,
I've
created
an
NLB,
then
the
Azure,
the
AWS
code
could
track
the
resources
created
better
and
could
then
go
and
remove
those.
F
Should
it
no
longer
meet
them.
It
would
complicate
the
update
process
a
little
bit,
but
it
would
be
better
than
leaking
resources,
possibly
the
issue
with
that
is
that
I
was
reading
through
and
I.
Don't
know
if
anyone
has
the
history
on
this.
F
F
If
the
answer
is
no
I
might
go
and
POC
this,
but
yeah
I
kind
of
brought
up
again
because
of
say,
like
I've
seen
this
on
AWS
I
suspect,
there's
similar
issues
for
resource
leaks
on
others.
If
there's
this
possibility
to
change,
Cloud
providers
like
low
balance
back
ends
or
something
in
general,
it
looks
like
we
don't
really
ever
track
the
resources
that
we
we
build
so
tearing
them
down
again.
D
I
I
don't
have
an
idea,
but
have
you
tried
doing
a
blame
on
the
comment
and
seeing
if
the
PR
that,
where
that
was
merged
had
any
more
info.
F
The
so
I
did
a
blame.
I
got
down
to
whoever
added
it
and
I
can't
remember
who
that
is
now,
but
I,
don't
at
the
time.
There
were
no
comments
on
there.
That
gave
me
any
more
inspiration.
A
F
A
G
That'd
be
great,
my
baby's
upstairs
right
now,
so
next.
A
A
F
Yeah,
this
is
not
helpful
and,
of
course,
my
MacBook
is
now
giving
me
security
issues
can
I
send
some
links.
Sure.
F
I
would
love
to,
but
now
since
I
have
to
quit
and
reopen
slack
to
be
able
to
do
that.
So
what's
that
Zoom,
that
issue
should
be
public
enough
for
you
to
read,
but
the
link.
That's
interesting
is
this
second
one.
A
I'll
start
sharing
again
so
that
I
can
yeah
yeah
I,
don't
have
a
red
hat
account,
so
I
can't
open
the
Red
Hat
Thing.
F
And
I'm,
like
yeah
Someone,
set
the
issues
security
level.
My
comment
is
public.
That's
weird!
Okay,
anyway,
if
you'd
show
the
GitHub
tab.
F
I
came
up
with
the
comment
last
week
of
in
theory,
the
load
balancer
could
apply
finalizers
blah
blah
blah.
However,
there's
a
note
in
the
interface
that
Services
must
be
treated
as
a
mutable
by
the
cloud,
not
sure
on
the
exact
reasoning
for
that,
but
there's
a
seminal
comment
which
doesn't
provide
you
much
context.
So
this
is
where
it
was
added.
F
So
that's
as
far
as
I
got
on
why
that
comment
got
added,
but
it's
this
one
here
on
line
83
information
implementations
must
treat
the
API
service
parameter
as
read-only
and
not
modify
it.
B
D
So
the
one
thing
I
will
say
is
Daniel
and
sadly
Daniel's
not
around
anymore,
but
I
know
Daniel
well
enough
to
read.
Daniel
is,
is
pretty
exact
in
what
he
says,
and
so
I
would
I
would
take
his
comment
verbatim
and
so
Daniel
at
least
believed
at
the
time
that
the
problem
wasn't
what
was
going
to
happen.
If
you,
if
they,
if
you
made
a
change
to
the
service,
it's
the
fact
that
it
might
not
be
persistent
right,
so
I
I
knowing
Daniel
well
enough.
He
that
is
almost
certainly
you
know.
F
Okay,
in
which
case
I
will
have
a
go,
probably
with
the
university
provider,
because
that's
where
I've
seen
the
leak
at
adding
some
finalizers
when
we
create
load
balancers
persisting
modes
in
my
own
API
call
and
then
using
that
to
track
which
I
haven't,
haven't,
created
and
then
see
how
I
go
on.
A
Yeah
and
of
course,
we
we
just
exited
code,
freeze,
I
believe
which
is
very
exciting.
So
now
is
your
moment
to
get
a
whole
bunch
of
stuff
in
before
the
next
release.
Before
we
get
back
on
that
Merry-Go-Round
again,.
A
All
right,
let's
see,
I,
don't
think
we
have
anything
else
on
the
agenda
Nick.
Do
you
want
to
tell
us
now
that
you
are
back
what's
what's
exciting
for
you?
What's
on
your
mind,.
G
Yeah
well
I'll,
just
I'll
just
I
just
want
to
mention
that
I
don't
know
if
you've
I
guess
Kirsten
has
been
coming
a
little
bit,
but
your
thing's
going
to
be
that
kind
of
it'll
be
us
representative
and
cloud
provider
and
I
will
come
as
much
as
I
can
so
I'll
still
be
around
I'll
still
be
attending
meetings,
I'm,
not
sure
if
I'll
come
every
single
week,
but
yeah.
If
you
haven't,
met
Kirsten,
please
introduce
yourself
to
him
and
other
than
that.
G
Yeah
I
got
a
nice
break
from
from
All
Things
technology
related
for
a
few
months,
and
just
hang
out
with
my
my
my
bit.
My
child's
and
I
will
mention
that
he's
actually
10
months
old,
so
I
kept
pushing
back
my
paternity.
So
so
he's
almost
he's
coming
on
a
year,
but
yeah
he's
a
very
happy
little
child.
A
Okay,
well
great,
seeing
everyone
welcome
back,
Nick
and
before
I,
say
goodbye,
I'll,
just
wave
and
say
Marco
hi
Marco.
Why
do
I
never
remember
getting
a
chance
to
talk
to
you
before?
Are
you
also
new
joining
us,
or
are
you
just
here
infrequently
enough,
please
say:
hi.