►
From YouTube: Kubernetes SIG Cluster Lifecycle 20170808
Description
Meeting Notes: https://docs.google.com/document/d/17J496IR2tXKw7k97fxwz2KUWOf9rpBD3pIEsmDiJQSw/edit#heading=h.bf0asbysznm2
Highlights:
- Update on progress towards 1.8 features
- Discussed how we should be represented in SIG-PM
- Check out https://github.com/heptio/sonobuoy
- Demo of the kubeadm upgrade command
- How to get started contributing to the SIG
A
D
A
B
E
E
B
A
F
So
we've
gone
back
and
forth
in
a
couple
of
conversations
and
there's
a
meeting
tomorrow
and
that's
part
of
my
PSA
from
yuju
and
I
are
on
the
same
page
with
the
general
miss
of
the
proposal
that
exists
on
and
I'm
going
to
talk
to
sig
note
today
too,
as
well.
She
has
specific
locations
where
she
wants
to
have
the
code,
but
the
code
itself
is
really
not
that
much
I.
F
Think
more
of
this
has
to
do
with
socialization
and
understanding
the
roadmap
after
the
initial
implementation
and
at
least
listening
to
concerns
and
making
sure
they're
addressed
in
the
proposal
once
we
get
it
finalised
and
pushed
up
to
the
community
repo.
So
I
think
at
this
point.
If,
for
those
who
are
interested,
we
will
discuss
it
tomorrow
with
other
interested
parties
as
part
of
the
ongoing
meetings
that
we
are
having
and
I
think
that's
pretty
much
it
unless
Erin
or
Lucas
either.
How
much.
B
F
We're
going
to
leave
it
as
a
stretch
goal
I,
don't
you
know
I
want
to
get
the
proposal
done
in
the
code
in
place
and
signaled
happy
before
I
do
anything
else,
because
the
review
process
and
everything
all
this
other
processes
that
are
going
on
right
now
are
taking
far
longer
than
I
had
hoped
for
and
require
far
more
socialization
than
it
does
actually
code
changes.
So.
A
Those
are
checking
in
I
was
just
reviewing
sort
of
our
planning
doc
41.8
there
couple
things
we
haven't
talked
about
recently.
They
we
had
marked
as
p0
so
I
just
wanted
to
touch
base
on
those
real
quick.
The
first
is
which
we
were
hoping
to
close
at
least
10
issues,
and
so
basically
fixing
some
bugs
in
cube,
admin,
repository
and
I.
A
Think
Jamie
was
going
to
take
the
lead
on
this,
obviously
not
doing
it
all
by
himself,
but
I
don't
know
if
he
or
anyone
else
is
actually
then
sort
of
keeping
track
of
our
progress
on
that.
So
there
we
should
know
if
we
are
anywhere
close
to
our
goal
of
10,
which
was
about
10%
of
the
open
issues.
At
the
time
of
that,
we
created
legal
I.
B
I
think
we've
reached
that
goal
and
have
been
even
more
successful.
I
think,
of
course,
I
mean
I
just
spawned
like
5
new
issues.
So
it's
it's
back
to
the
same
level,
but
anyway,
I
mean
I,
think
we've
closed
at
least
12
or
something
like
that
old
old
ones
and,
of
course,
like
attitude,
new
musicians
upon
for
our
vitals.
So
that's
that's
now
the
largest
pot.
Oh,
we
have.
A
F
B
B
Somehow
I
don't
know,
but
I
think
it
fits
pretty
well
into
like
the
discussion
we're
going
to
have
with
Jacob
like
how
do
we
actually
write
code
for
this
this
bust
so
like
in
a
more
efficient
manner.
I
mean
right
now
we
have
something
hard-coded
to
here
and
there
and
assumption
if
we
could
do
this
in
a
more
generic
way
without
telling
so
like
build
everything
into
testing
front,
it
would
be
better
true.
A
And
then,
and
also
from
our
side,
Jessica
I
started
looking
at
the
test.
Emperor
interoperate
Esther
keep
I've
been,
which
is
sort
of
more
aligned
with
making
upgrades
work
then
improving
test
coverage,
but
it
also
helps
send
someone
else
up,
understand
the
test
infrastructure
seriously.
That
will
help
yeah.
B
We
now
have
butcher
token
IDI
coverage
for
cuba
impact
app
head
and
notarized
ice
coming
as
soon
as
like
the
grand
new
plates
of
EDTA
is
reordered,
so
it
right
now
is
here:
it's
blocked
on
refactoring
single
step,
but
after
that's
done,
if
the
code
is
written
already,
so
it's
going
to
be
much
hopefully
by
next
week.
So
in
that
sense
we
have
better
coverage
and
also
much
better
coverage
on
you
test.
B
Yeah
so
I
attended
pick
Dan
from
minutes
ago,
and
we
talked
about
p.m.
representative
from
hits
big.
It
turns
out
that
Dan
has
been
our
representative
sofa,
but
he
has
other
works
to
do
right
now,
so
we
should
at
least
like
look
if
there's
someone
else
on
course,
that
can
take
on
being
the
team
with
representative
from
district
or
if
we
should
I
have
honest
about
it.
A
Do
you
know
is
the
intention
here
to
have
PMS
be
the
PM
representatives
like
p.m.
or
like?
Yes,
there
were
sort
of
two
ways
this
was
discussed.
One
was
to
have
people
from
sig
p.m.
attended,
the
other
sig
meetings,
the
other
was
to
have
people
from
the
SIG's
attend
lesyk
p.m.
meeting,
which
was
I
think.
B
G
It's
just
going
to
say
that
so
that
definitely
having
a
representative
from
this
group
in
the
sig
p.m.
meetings,
if
possible,
will
help
create
create
that
steel
thread,
especially
when
the
cluster
lifecycle
group
wants
to
sponsor
a
feature
and
a
stabilization
release
and
there's
there's
a
lot
of
questions.
I
know,
Joe
Joe
has
had
a
lot
of
concerns
and
I
hear
his
concerns
about
the
relevancy
and
the
importance
of
sig
p.m.
and
what
it's
trying
to
do.
G
I
think
one
of
the
important
lines
of
communication
that
we
want
to
maintain
is
between
this
egg
and
that's
eggs.
So
it's
really
about
having
somebody
from
this
egg
present
in
those
meetings
so
that
we
can
talk
about
what
features
are
coming
up,
how
we
best
promote
them
and
ensure
that
the
feature
life
cycle
is
on
track
in
in
accordance
with
whatever
type
of
feature
where
our
release
we're
putting
out.
H
There's
also,
then
so
sig
architecture
is
also
starting
up
and
we're
trying
to
figure
out
sort
of
you
know
how
do
we
capture
and-
and
you
know,
make
obvious
to
folks
sort
of
what
the
long-term
architectural
decisions
and
trends
are
and
we're
still
figuring
out
the
relationship
between
being
that
effort,
which
is
very
much
sort
of
technical,
long-term,
focused
and
and
what
role
cig
p.m.
has
now
I
personally,
like
I,
can't
make
a
cig
p.m.
meetings
and
so
and
and
I,
don't
see
a
lot
of
it's
difficult
to
actually
see
what
sig
p.m.
is
doing.
H
If
you
don't
go
to
those
meetings
which
is
I,
think
unfortunate,
because
I
think
one
of
the
biggest
things
that
they
can
do
is
is,
is
you
know,
help
to
to
sort
of
get
the
message
out
there?
So
that's
why
I
think
there's
a
ton
of
overlap
between
what
sig
p.m.
does
in
our
community
meetings,
and
so,
if
you
were,
you
know
we're
going
to
we're
going
to
try
I
believe
this
time
to
do
a
little
bit
on
lap
and
mix
things
up
a
little
bit.
I
am
really
interested
to
see
how
that
works.
Yeah.
G
G
I
mean
I
think
that
it's
8
p.m.
the
sig
p.m.
group
has
a
really
important
function.
We
just
need
to
really
lock
down
what
that
is
and
define
it,
and
especially
in
light
of
Sagarika
texture,
but
just
for
me
personally,
since
since
sig
cluster
ops,
almost
more
than
any
other
group,
is
so
end-user
focused
and
cluster
operator
focused
that
this
work
is
arguably
I
hate
to
use
superlatives.
But
this
the
group
has
a
huge
impact
on
the
in
the
wave.
Urban
abuse
is
perceived
in
the
world,
so
we
need
to
keep
a
really
tight
loop.
G
A
H
G
We're
actually
wrote
a
document
this
morning
and
it
is
the
stabilization
versus
a
regular
release
and
it
actually
has
some
specific
goals
and
guidelines
and
things
that
are
for
SIG's
to
participate
in
the
process
and
also
what
are
the
controls
which
cig
PM
is
responsible
for
so
I
think
it's
going
to
start
doing
that
demarcation.
Yes,.
H
They're
an
interesting
release
also
and
then
I
know
as
we
get
down
to
get
down
to
the
wire
there's
I'm,
trying
to
remember
the
name
of
the
mailing
list
for
actually
approving
changes.
You
know
after
code
freeze
to
get
stuff
in
I
mean.
Is
this
really
trying
I
mean
it's
a
goal
here
to
actually
be
able
to
figure
out
sort
of?
What's
in
and
what's
out
for
a
release?
And-
and
you
know
just
that
tactical-
you
know
what's
happening
on
the
ground
and
for
how
does
it
relate
to
seek
release
then
I.
G
G
B
J
B
J
Had
someone
brightness
on
the
call
just
start
it
would
some
we
have
to
decide
whether
it
makes
sense
for
someone
from
coral
ice
to
take
up
Dan's
role.
Obviously
that
would
guess
if
no
one,
if
no
one
is
clamoring
for
MIT
I
guess
it
certainly
seems
to
be
our
responsibility
to
find
a
replacement.
Since
we
held
the
position
previously
so
I'm
going
to
follow
up
offline
and
see
if
we
can
find
someone
who
attends
this
meeting
to
be
2
p.m.
J
A
I
mean
I,
think
that's
sort
of
not
the
model
we
want
to
propagate
is
that
you
know
someone
else
from
your
company
has
to
replaces
it,
because
I
want
from
your
company
was
doing
before,
like
the
point
of
the
sig
is
that
we
are
a
community
and
that
we
sort
of
take
turns
in
some
of
these
roles
and
so
I'd
only
a
reasonable
for
someone
else
to
step
up
oh
yeah
for
a
while,
and
maybe
we
should
make
it
more
of
a
rotating
position.
Officially.
Instead
of
somebody
doesn't
really
get
tired
of
it
right.
J
A
Maybe
maybe
we
should
do
something
like
say
like
for
the
1/8
release
cycle,
it'll,
be
this
person
and
it'll
take
someone
for
the
one
line
cycle
and
sort
of
rotate
that
way,
so
everybody
gets
sort
of
a
three-month
stint.
I,
don't
think
that'd
be
too
painful
to
do
it.
For
you
know,
bi-weekly
every
three
months:
it's
not
that
many
meetings,
yeah.
F
F
Do
this
informally
own
six
scheduling
to
outline
who's
going
to
have
what
major
rule
as
part
of
the
cycle,
because
it's
impossible
for
us
to
cut?
We
don't
have
as
many
people
assisting
and
it's
kind
of
it's
constantly
an
issue
to
make
sure
that
we
rotate
people
through
to
get
enough.
You
know
and
that's
my
chair,
distributed
amongst
the
groups
so
that
way,
no
one
person
becomes
a
point
of
failure,
bulb
screen.
So.
A
F
F
We
had
open
sourced
a
tool
as
part
of
the
CNC,
f
conformance
effort
called
Sunnah
boy
on
the
reason
why
I'm
mentioning
it
here
is
because
there
were
some
issues
open
previously
with
regards
to
post
cluster,
stand-up
validation,
and
that's
part
of
the
area
that
so
nobly
aims
to
address
and
for
the
I
could
probably
do
a
demo
at
some
other
meeting
and
I
plan
to
do
it.
One
at
Sig
testing
today,
if
you're
interested-
and
it
basically
just
serves
to
validate
I'm
sure
your
clusters
can
form
it
before
after
you've
stood
it
up.
F
If
you
enable
full
conformance,
it
takes
a
while
right
now
the
current
default
for
the
exit
for
the
QuickStart
examples
is
set
to
only
run
one
test
just
to
validate
that
it's
grabbing
everything
it
runs
on
your
cluster.
There
is
a
separate
page
that
outlines
all
the
details,
so
it
could
take
20
to
30
minutes
to
run
at
the
end,
depending
upon
the
size.
Your
cluster
right.
A
I
guess
I'm
wondering
like
if
we
should
make
this
sort
of
part
of
the
normal
great
cluster
flow.
Something
like
this.
If
it
takes
30
minutes
to
run
that's
probably
not
a
good
user
experience
or
someone
like
get
start
with
kubernetes
and
wait
half
an
hour,
so
we
can
verify
that
you're
three
nodes,
you
provisioned
on
a
cloud,
that's
pretty
reliable
as
conformant
and
when
it
should
basically
always
be
right.
If
it's
like
a
quick
smoke
test,
if
it
runs
really
fast,
maybe
that
makes
sense,
visibly
visible.
Yeah.
F
We
could
we
could
always
trim
down
the
set
of
tests
and
have
like
a
light
and
heavy.
The
current
conformance
list
is
167
s,
or
something
like
that.
So
it
seems
completely
reasonable
to
me
to
basically
have
like
if
we
know
Q
ADM
follows
particular
flow.
We
could
have
something
that
is
a
trimmed-down
version
of
it
that
will
run
in
orders
of
magnitude
less
than
it
takes
to
run
all
the
other
tests.
Yeah.
A
I
mean
I
guess
in
my
view,
the
thing
that's
probably
most
likely
to
is
networking
right.
So
if
we
did
like
a
quick
networking
test
that
verified
pada,
pada
networking
was
working
like
you
know,
things
like
you
know.
The
daemon
tech
controller
working
properly
are
probably
okay,
if
you're
using
the
same
version,
f2
badminton
that
that
we
tested
in
at
least
but
their
networking
side
like
I,
maybe
funky.
B
A
It's
actually,
if
you're
using
like
you,
know
different
places,
it's
too
bad,
then
without
the
whole
thing
at
the
end
and
I
think
you
could
much
more
easily
get
into
a
state
where
your
clusters
sort
of
looks
right,
but
networking
is
broken.
Then
your
cluster
sort
of
looks
right
and
same.
It
says:
okay,.
B
So
this
is
one
thing
when
discussing
in
the
Cuban
adoption
working
group
of
all
like
how
do
we
actually
verify
my
cluster
deployments
level
that
this
cluster
is
conformance
whenever
everybody
has
like
their
own
reference
and
that's
kind
of
hard,
but
hopefully
it
will
get
a
little
bit
easier,
at
least
with
like
when
it
every
one
of
most
most
one
eventually
use
key
better.
That's
a
minimum
base.
B
B
Cool
so
first
I'm
doing
a
normal
cube
etiquette
and
here
I'm
specifying
to
use
kubernetes
version,
one
seven,
one
and
I'm
setting
a
custom
API
server
port.
The
default
is
six
four,
four
three
and
now
I'm
setting
it
to
four
for
free
and
I'm
doing
this
to
demonstrate
that
if
the
user
did
this,
when
when
they
were
upgrading,
it
should
break
right
because
it's
not
the
default,
but
it's
the
the
customization
should
be
preserved.
This
is
something
or
all
for
upgrades
right
now
doesn't
do
so.
B
Unfortunately,
the
this
thing
didn't
like
make
one
one:
seven
zero
uploading
the
config
map,
but
it
will,
it
will
be
later
on
so
there
we
have
a
custom
and
then
yeah
clarify
if
I
now
run
cube,
am
everything's
perfect
Oh,
like
with
Alba
gate
by
an
awful
flag,
but
it
will
obviously
see
removed
given
an
upgrade.
This
is
what
it
looks
like
and
we
have
quiet
and
plan.
We
can
allow
experimental
upgrades
if
you
want
to
upgrade
to
be
turned
over.
We
want
subversive
Aussies.
We
can
do
that
as
well.
This
is
one
thing.
B
B
F
There
sort
of
a
clutch
that
gets
placed
up
this
December
check
on
things
that
are
not
released
versions,
so
you
mean,
if
you
are
using
CI
version
or
no,
if
you're
using
a
release
candidate,
even
there's,
we
should
put
out
some
warning
clause
that
you
know
you
are.
You
are
using
a
release
candidate,
you
know
etc,
said.
F
B
I
can
have
that
a
big
banner
or
something
I
mean
I.
Think
I
have
a
small
warning
or
something
like
that,
but
it
could
definitely
be
the
elantra
anyway
and
we
have
subsided
compact
flag,
keep
config
wedlock,
I
thought
so
I'm
pretty
comfy.
So
if
I
do
give
a
number
of
upward
plans
now
it
will
thank
Chris.
B
I
was
checking
like
like
if
those
cluster
of
credible
and
like
the
ISO
is
healthy
and
we
have
one
unhealthy
node
and
what's
that
then
well,
I
am
deployed
a
demon
set
on
this
node,
so
it
obviously
doesn't
work
but
requiring
a
network
is
is
great
like
from
our
perspective
because
or
like
requiring.
The
use
of
the
upgrade.
A
healthy
cluster
is
pretty
reasonable,
but
it
also
like
parts
around
no,
not
releasing.
We
have
in
case
we
have
to
use
the
phone
with
some
these
things,
so
we're
not
limited
in
the
same
way.
B
We
are
occupied
a
minute
but
yeah.
That's
that's
a
side
note.
So
when
we
then
okay
now
we're
ready.
We
can
do
that
again,
and
this
is
the
output
I'm
actually
considering
to
use
colors
I,
don't
know
if
people
would
would
like
that,
but
I
mean
that
well
kind
of
bit
of
texture.
So
if
we
could
highlight
like
this
is
healthy,
all
notes,
I
alter
the
static
part
of
the
control.
Plane
is
healthy
and
these
things.
D
B
B
B
Like
that
so
see
to
upgrade
possibilities,
so
first
we
have
like
the
normal
one,
and
then
we
have
the
external
one
and
I
mean
it's
it's
possible
to
operate
the
dispersion,
but
still
it
seems
that
we
have
some
kind
of
big
banner
before.
But
if
you
do
that,
you
have
to
in
any
way
in
any
case,
have
this
flag
and
allow
exponential
upgrades
before
you
can
do
anything.
B
B
K
B
If
we
specify
right
here,
it
will
pre-fill
the
images,
the
control,
plane
images.
This
is
basically
a
demon
set
with
the
images
that
are
going
to
be
used
and
they're
just
leaving
an
hour,
and
this
is
because,
if
we
do
this
kind
of
pre
poll
with
Dino
testing,
we're
not
going
to
when
we're
upgrading
like
fail
on
bad
internet
connection
or
something
so
that's
that's
one
good
thing
that
if,
if
we
don't
have
internet
connection,
it
will
just
arouse
in
a
people
face
of
in
the
actual
upgrade
face.
B
B
Showtime
output
like
this
is
where
I'm
writing
things
now
and
here
with
the
backup
like
and
if,
if
this
sale
I'm
going
to
automatically
roll
back,
this
backup
manifest
to
the
dole
place
and
things
will
be
resolved
and
API
so
was
updated.
The
same
goes
for
controller
manager
here
the
scheduler,
and
then
we
just
applied
the
adults
good.
So
look.
B
C
So
it's
doing
a
sweep
as
like
the
command
for
that
payment
set,
which
implies
I
mean
it
seems
that
there
is
a
sleep
command
inside
that
container.
That
can
be
called
yeah
move
to
like
smaller
containers
that
just
have
the
static
binary
in
there.
That
won't
work,
but
hope
I
mean
really.
We
need
a
better
pre-poll
mechanism
through.
B
B
B
B
K
B
A
F
B
Yeah
and
we
could
we
should
definitely
start
the
conversation
there
for
real
I
mean,
as
we
said,
there's
been
feature
requests
for
being
for
a
long
time
there,
but
well
now,
no
movement
as
far
as
I
know.
Anyway.
That's
the
workaround
we're
doing
right
now
and
we
can
see
that
cube.
C
del
Worsham
put
one
two
or
three.
B
K
B
K
B
D
Your
editor
few
different
flags
for
opting
into
non-standard
releases
and
I
think
one
was
specifically
are
seeds
and
when
you
say
that,
does
that
exclude
alpha
beta,
you
actually
mean
like
RCR
seeds,
or
is
it
just
pre-release
releases
and
the
other
one
was
experimental
and
I
was
wondering
if
experimental,
yeah
just
completely
continued,
like
steai
build
or
if
they
also
men.
Just
like
anything
that
we
don't
parse
us
some
baird
is
it
like.
D
B
That's
one
area
that
definitely
could
be
improved
like
from
the
US
side,
experimental
versions
in
this
context,
what
I've
done
so
far
is
open,
beta
and
all
C's
is
like
well,
it
experimental
is
all
for
bid
on
our
piece.
It's
allowed
for
everything,
all
those,
but
it
still
has
to
be
a
semantic
question
of
allow
us.
It's
just
a
low
quality.
D
B
D
D
B
D
F
B
B
A
A
B
One
one
thing
to
notice
here
that
I
didn't
mention
is
that
if
you
are
small
one
seven,
it's
the
following:
a
static
port
control
plane.
If
you
upgrade
from
170
180,
it
will
prefer
the
static
pod
which
okay
for
you,
it
will
not
operate
House
hosted.
But
if
you
do
cube
a
them
in
it,
a
new
cluster
on
one
eight
it
will
make
it
will
use
self
hosting
by
default.
And
if
you,
when
you
upgrade
them
one
nine,
it
will
upgrade
or
require
yourself
cred
to
self
hosted
before
you
can
do
that.
B
A
J
A
I
mean
that's.
That
should
be
pretty
easy
thing
for
us
to
check
in
the
upgrade
command
is,
if
version
before
the
version
where
we
know
we're,
sewing
too
big
map,
or
if
we
can't
find
a
config
map
in
it
to
let
the
user
know
like
by
the
way
we're
going
to
blow
away
all
of
the
parameters
you
specified
during
running
in
it.
Unless
you
reach
specify
them.
Are
you
sure
you
wanna
continue?
A
B
A
A
So
it's
not
all
landing
on
one
or
even
two
people's
shoulders,
so
sort
of
just
a
call-out
that
if
there
is,
if
you
do
feel
like
you
want
to
contribute,
even
if
you
don't
feel
like
you
are
necessarily
a
great
code.
Reviewer
one
way
to
learn
is
to
practice
and
you
can
sort
of
pair
up
with
someone
else's
doing
a
code
review
and
maybe
take
some
burden
off
with
them
in
the
process.
I'm.
I
A
K
I
L
A
We
talked
last
week
other
than
Doc's,
like
a
great
way
to
get
started
is
to
help
with
testing,
because
the
you
know
writing
tests
is
probably
a
good
introduction
to
figure
out
how
the
code
base
works.
So
I
would
encourage
you
to
sign
up
on
Jacobs
spreadsheet,
for
his
sort
of
PD
test
infrastructure
running
session.
Maybe
try
to
like
write
a
couple
of
tests
and
sort
of
see
how
the
process
works
and
pivot
from
there
into
edit
features.