►
From YouTube: Kubernetes SIG Cloud Provider 2019-06-12
Description
No description was provided for this meeting.
If this is YOUR meeting, an easy way to fix this is to add a description to your video, wherever mtngs.io found it (probably YouTube).
A
Okay-
and
let
me
share
my
screen-
alright
folks-
can
see
that
okay,
so
hi.
Everyone
today
is
June
12
2019.
This
is
the
bi-weekly
kubernetes
cloud
provider
meeting.
We
have
a
pretty
light
another
and
mostly
wanted
to
go
over
action
items
from
Q
Khan
and
also
the
go
through
our
115
backlog
and
make
sure
that
things
are
in
an
updated
State,
because
I
don't
think
we've
touched
it
in
a
while
so
yeah.
Let's,
let's
do
that.
Okay
actually
wanted
to
give
one
announcement
before
we
start.
A
Everyone
under
in
one
PR
on
July
12th
sound
reasonable
to
folks.
So
far,
all
right,
awesome,
okay,
so
there
were
two
big
action
items
from
coupon
I
mean
there
were
a
few
others,
but
I
ended
up
just
throwing
it
in
our
backlog
and
we
can
talk.
We
can
chat
about
those.
Then
the
two
have
been
formalizing
in
tree
removal,
timelines
and
user
group
adoption.
So
people
can
see
this
right
there
diagram
here.
So
this
is.
A
A
It's
gonna
be
a
lot
about
building
the
auditory
providers
getting
them
in
a
decent
state
where
we
can
get
early
adopters
and
all
that,
and
so
that
would
mean
if
we
called
its
timeline
q4
2020,
when
we
are
able
to
kind
of
delete
the
all
the
code
and
appendices
entry.
Does
that
seem
like
a
reasonable
timeline
to
folks?
Does
anyone
have
changes
they
want
to
suggest
your
I
mean
we
and
we
don't
have
to
like
finalize
anything
now,
but
why
don't
you
just
share
this
ahead
of
time.
A
Like
queue
for
2020
yeah,
I
guess,
it
would
just
feel
like
the
end
of
the
release.
So
whenever
alright
I
guess
what
what
would
be
more
idea
was
like,
we
would
ideally
get
all
the
code
deleted
like
at
the
beginning
of
the
release
of
q4
2020
and
give
time
to
kind
of
like
like
to
let
it
sit.
And
then
my
guess
is
that,
like
the
end
of
q4
2020
is
when
Yukon
North
America
would
be
so
it
would
be
a
done
deal
by
then.
C
Can
I
make
a
suggestion?
I
mean
the
timeline,
looks
right
to
me.
There's
only
one
thing:
I
think
this.
Maybe
you,
since
it's
a
it's,
a
pretty
big
change.
We
probably
warrants
like
a
fairly
warning:
I
see
that
q1
2020
there's
announced
removal
of
entry
club
bladder
I
think
we
should
announce
that
earlier
than
that,
potentially
you
know
q4
2019,
or
maybe
it
cube
cut
and
do
something
more.
You
know
some
fanfare
around
the
fact
that
they're
gonna
be
removed.
D
E
A
So
for
folks
who
want
to
be
looped
into
those
discussions
with
six
storage,
let
me
know,
and
their
club
they're
probably
gonna,
be
like
a
separate
meeting
for
that.
Cuz
I,
don't
want
to
call
everyone
in
when
I
don't
have
to.
So
let
me
know
if
you,
if
you
want
to
be
involved
in
those
discussions,
okay,
cool
Oh,
Steve
I,
noticed
you
had
an
announcement
you
want
to.
You
want
to
give
that
in
Austin
a
parking
point,
you're,
muted,
all
I'll
meet
you
for
you.
Actually,
I
don't
know
if
I
could
do
that.
F
The
thing
I
added
was
just
the
session
time
for
the
sig
session
in
Q
Khan
China,
and
that
might
be
a
change
from
the
original,
because
the
government
wanted
the
venue
back
on
the
last
day
resulting
in
the
CNC
F
shuffling
session
times
around.
So
that's
the
time
as
of
right
now
and
there's
a
link
to
that
session
and
Chris
and
I
composed
the
deck
that
think
is
already
visible
at
this
point,
I
think
the
deck
is
probably
a
draft.
A
Awesome
thanks.
So
next
action
item
was
user
groups,
so
I
know
Steve.
There
was
discussions
around
on
VMware
being
one
of
the
few.
That's
gonna
take
this
on
early.
Well,
there
are
any
other
providers
here
that
we're
interested
in
taking
up
the
user
group
thing
now
or
are
we
kind
of
just
waiting?
You
know
to
see
if,
when
user
groups
become
more
solidified.
F
I
haven't
eared
to
do
it,
but
I
don't
actually
know
who
I
need
to
approach
or
what
the
next
step
is
to
cause
that
to
become
official.
This
is
that
going
to
have
to
be
submitted
to
the
steering
committee
or
something
or
how
is
this?
Does
anybody
know
how
this
is
likely
to
work?
There
have
been
a
lot
of
conversations,
but
I
don't
think
that
there's
anything
that
I've
encountered,
that
I
would
interpret
as
some
official
decree
that
this
is
a
done
decision.
F
A
F
A
There's
a
charter,
and
then
there
is
you
get
a
folder
I
think
in
the
community,
repo,
where
you
can
kind
of
put
content
and
resources
for
the
user
group
and
I
think
you
need
two
chairs
and
in
total
for
users
who
who
are
interested
in
that
user
group
to
get
started.
Okay,.
F
Okay,
I
can
confirm
the
interest
in
doing
it,
but
realistically
I
don't
think
I'll
get
started
until
I
get
back
from
Shanghai.
Okay,.
A
F
Well,
Janine
hi
is
gonna
end
in
that
timeframe,
I
think
I'll
get
back
jet-lagged
and
then
it's
right
up
against
the
fourth
of
July
US
holiday.
So,
okay,
probably
look
more
like
second
week
of
July
to
get
started
so
I'm
not
sure
I'll
I'm
going
to
exactly
align
this
with
whatever
else
is
going
on
in
July,
but
it'll
be
close.
Okay.
A
A
D
D
A
We
should
follow
the
same
guidelines
I
like
API
changes,
but
I.
Just
we
just
needed
that
document.
It's
somewhere.
Did
you
get
around
to
this
love?
You
know
I
totally
forgot
about
this.
Okay,
so
we'll
just
no
worries,
yeah
and
I'll,
also
pee,
or
if
someone
from
Google
can
ping
Justin
about
this
that'd
be
great.
A
A
A
A
Nothing's
here
so
mike
is
still
working
on
the
kept
for
credential
provider.
This
is
probably
gonna,
be
alpha
for
116
as
well,
so
we'll
move
this
milestone,
cube
controller
to
control
manager
cap.
So
this
one
slipped
on
my
end:
I
couldn't
get
the
cat
approved
in
time,
and
so
this
one
this
one
couldn't
make
any
progress.
So
this
one's
gonna
be
116
as
well,
so
this
so
this
cap
is
probably
gonna
require
a
lot
of
like
design
thought
into
design.
A
So
if
folks
are
interested
in
how
to
do
this,
migration
more
seamlessly
or
like
if
you
work
on
a
cloud
platform
that
is
gonna
need
this
migration
mechanism,
then
feedback
and
help
in
this
area
would
be
great.
So
right
now,
as
part
of
the
extraction
migration
sub-project
we're
working
through
the
design
design
of
this
right
now,
okay
and
then
these
fear
Doc's,
so
I
wrote
some
Doc's
for
vSphere
and
pull
it
up.
C
H
A
I
D
D
It
good
question
at
this
point:
the
best
I
can
tell
you
is
depending
on
which
metrics
you
believe
it
seems
to
be
between
10
and
20
percent
of
all
cube.
Api
server
traffic
is
related
to
leader
election
okay,
gotcha,
it's
a
little
higher
than
I
think
it
should
be,
and
you
don't
worried
if
we're
gonna
bug
that
number
up.
Okay,.
A
D
Certainly
and
I'm
still
trying
to
drill
down
and
get
one
set
of
numbers
right
now,
depending
on
how
I
did
the
queries.
I
get
two
numbers
and
I'm
trying
to
understand
why
they're
different
and
then
I'll
start
breaking
into
where
exactly
the
load
is
and
why
it's
so
high.
So
that's
an
ongoing
I,
just
figured
I'd.
Give
you
guys
a
little
heads
up:
okay,
cool.
A
A
This
was
funded
because,
okay,
so
in
114
we
deleted
the
PD
controller,
the
out
of
true
PD
controller,
and
the
plan
was
to
provide
a
reference
implementation
for
like
out
of
tree
PV
labeling
that
uses
a
mutating
admission.
Web
book
I
feel
I.
Think
this.
This
was
delayed
mostly
just
cause
of
design
like
we
need
to
figure
out,
what's
the
best
design
for
dynamically,
applying
zone
and
region
labels
to
TVs,
without
embedding
that
into
the
API
server.
A
A
So
the
PR
essentially
is
supposed
to
migrate.
The
the
beta
labels
for
zone
region
and
insect
to
topology
dot.
Cooper
I
saw
IO
labels
so
there
so
there
J
this
one
slipped
on
the
cap
deadline.
So
hopefully
we
can
get
this
cap
in
early
in
the
cycle
and
get
it
started
so
I
know
Jeremy,
Morris
I,
don't
think
he's
on
a
call,
but
he's
he's
decided
to
take
up
this
work.
We
can
probably
divvy
up
the
work
like
one
person
does
each
label.
A
So
if
anyone
else
is
interested
in
this
one,
let
me
know
and
I
can
assign
it
to
but
looks
like
this.
One
is
also
gonna,
be
116,
yeah,
all
right
cool,
so
that
was
it
for
115
issues.
Anyone
want
to
discuss
anything
for
116
I.
Don't
want
to
do
that.
This
we
need.
We
can
do
the
116
backlog
grooming
for
for
next
time,
but
there's
any
specific
issues.
People
want
to
bring
up.
That
would
be
a
good
time
for
that.
A
B
Sure
so
the
issue
here
is
that
when
a
node
is
failing,
what
happens
is
that
the
node
switches
in
in
a
condition
not
ready
and
the
part
or
the
stateless
parts
get
rescheduled,
but
the
parts
that
are
by
backed
by
volumes
or
generally
speaking,
stateful
sets
do
not
get
rescheduled
because
the
control
plane
doesn't
know
if
the
nodes
gonna
come
back
or
not.
So
if
it,
since
it
cannot
know
if
the
nodes
going
back
up
or
not,
it
cannot
reschedule
the
sexual
parts
in
order
to
not
violate
the
Pulsar
safety
properties.
A
D
B
For
now
we
have
like
we
have
a.
We
have
a
method
on
the
cloud
interface
that
says
if
a
node
is
shut
down
or
not,
and
back
in
the
days.
What
we
were
doing
is
that
we
were
removing
the
some.
Some
providers
were
removing
the
node
object,
so
this
helped
on
this
issue,
since
when
we
delete
the
node
object
before
GC
controller
or
friends,
pods
the
that
were
affected
to
this
node.
B
A
Right,
so
it's
so
we
so
the
node
lifecycle
controller
handles
shutdown
and
it
applies
it
ain't,
but
because
the
nodes
no
longer
deleted
the
pods
on
the
node
stay
there,
which
means
we
don't
detach
volumes
properly.
Like
that's
the
main
issue
right
now.
Yes,
okay
and
it's
not-
and
it's
too
late
to
to
just
say
like
if
a
node
shut
down,
if
I
notice
shut
down,
we
should
just
delete
it.
Also
like
it's
that
that's
it's
too
late
to
go
there
right.
Yes,.
B
G
B
Of
the
yeah
one
of
the
big
issues
here
is
that
some
yeah
some
nuts
nodes
could
come
back
and
register
with
the
same
identity
and
list
all
the
paths
that
were
assigned
to
it
and
start
them
back.
And
so,
if
we
acted
on
this
change
that
informs
the
the
node
is
shut
down,
we
could
basically
violate
the
guarantees
of
safety.
B
What
the
I
think
this
is
divided
in
two
things.
First,
I
think
we
need
to
get
a
design
that
works
and
that
solve
this
issue
without
violating
any
properties
of
the
plug
safety.
And
then
there
is
the
part
where
we
need
a
indeed
implement
some
change.
The
change
would
likely
span
across
multiple
six,
at
least
cloud
provider,
storage
and
node,
and.
A
A
C
G
G
A
discussion
I
think
it
that's
why
the
timeline
for
the
user
group
discussions
are
important
so
that
by
the
CFP
4q
comes
in
Diego
we
have
I
was
actually
looking
for
the
user
group.
I
wasn't
able
to
dig
it
out
better.
My
email
there,
but
I
had
that
same
question
to
alert
those
that
get
here.
How
to
do
that.
So
maybe
next
next
of
these
meetings,
we
should
dive
into
that
more
and
just
to
make
sure
there's
clear
visibility
and
people
don't
miss.
C
G
A
F
You
mentioned
that
they
wanted
for
users
to
charter
a
group
and
one
of
the
things
I've
noticed
just
watching
the
VMware
say
attendance.
Is
we
don't
really
have
users
showing
up
but
they're,
not
consistently
showing
up
in
the
form
of
the
one?
Individual
never
misses
a
meeting
so
putting
these
users
in
kind
of
leadership
positions
running
the
thing
might
be
a
challenge.