►
From YouTube: 2017-07-28 17.01.31 SIG-cluster-lifecycle 166836624
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,
this
is
so
cluster
lifecycle
on
Friday
the
28th
to
July,
and
this
is
the
Friday
meeting
where
we
have
a
more
in-depth
discussion
about
the
in
progress
work
on
some
posting
upgrades
and
H,
a
probably
in
that
order.
So
does
anyone
want
to
take
it
away?
I
haven't
actually
looked
at
the
agenda,
yet
Tim
I
am.
B
B
There's
something
to
be
said
actually
just
just
to
give
a
little
props
to
some
of
the
work
that
Lucas
has
done,
the
continual
refinement
that
has
gone
through
and
the
shuffling
of
phases
and
all
that
other
refactoring
work
actually
helps
a
lot
so
for
readability.
If
we
were
to
leave
it
the
way
it's
the
previous
incantation
was,
it
probably
would
have
you
know,
grown
up
to
be
the
ball
duct
tape.
That
is
some
other
aspects
of
the
cone.
B
C
C
C
A
D
A
D
E
A
D
A
C
D
Okay,
so
August
9th
is
our
first
once
a
meeting.
Next
week
we
will
still
be
on
Friday
yep,
all
right,
so
I
think
an
issue
here
that
Tim
created
eight
days
ago.
I
thought
we
did
not
have
a
chance
to
discuss
last
week,
which
is
policy
for
versioning
on
end-to-end
tests.
It
seemed
like
this
was
going
to
be
an
issue
related
to
cell,
forcing
afraid
that
we
should
be
discussing
this.
B
Well,
it
doesn't
really
affect
self-hosting,
it
upgrades
it
has
more
to
do
with
features
that
were
added
in,
like
the
one
7
timeframe
and
the
intent
tests,
not
taking
versions
into
account.
So
I
know
Lucas
tagged
other
people
on
it,
but
I
was
it
was
originally
for
testing
because,
as
we
do
upgrade
tests
and
whatnot
there's
always
this,
because
I
also
work
on
cig
testing.
B
There's
this
black
hole
around
the
great
tests
that
occurs
and
a
couple
of
iterations
ago
we
had
issues
where
things
were
promoted
from
alpha
to
beta
and
the
tests
were
not
upgraded.
So
a
lot
of
these
tests
failed
and
it
caused
a
big
consternation,
and
there
are
many
people
involved.
So
the
whole
purpose
of
that
is
just
when
you
add
new
intend
tests
to
properly
version
them
with
a
version
check.
So
that
way,
if
they
take
that
test
suite
and
run
it
on
a
previous
version,
it
doesn't
explode
and
vice
versa.
D
Yes,
we
actually
talked
about
the
skull
years
ago
and
Zach
Lisbon
from
Google
had
suggested
adding
basically
capability
bits
into
the
API
server,
so,
instead
of
strictly
doing
version
based
checks
to
know
when
you
should
or
should
not
be
running
tests,
API
server
we're
basically
advertised
like
I
support
this
type
of
functionality,
and
then
you
could
test
the
functionality
of
the
advertise
and
that
sort
of
proposal
never
really
went
anywhere.
I
think
there's
still
an
issue
open
for
it,
but
that
would
be
a
really
sort
of
clean
way
to
do
this
also
yeah.
B
H
D
Cesc
you
in
both
directions,
so
we
run
old
protests
against
current
version
and
current
upgrade
tests
against
old
version
to
test
for
backwards,
breakable
changes
and
also
commend
for
forward
compatibility
right.
D
Well,
the
point
is
that
if
the
test
start
failing,
then
you're
making
a
change,
it
is
user
impacting
and
maybe
is
a
breaking
change
that
you
shouldn't
be
making
right.
So
if
we
start
guarding
test
saying
like
oh,
let's
stop
running
this
test,
then
we're
sort
of
potentially
opening
ourselves
up
to
an
easier
out
for
people
to
slip
in
breaking
changes
that
are
unnoticed.
If
the
reviewer
is
not
on
the
ball,
that
is.
B
D
So
I,
don't
think
anyone
is
is
doing
necessarily
doing
a
great
job
holding
the
line,
and
that's
we
rely
a
little
bit
on
testing
for
that,
because
the
test
has
caught
some
places
where
people
have
tried
to
introduce
breaking
changes
and
they've
gotten
past
the
review
process.
So
another
thing
we
shouldn't
put
version
guards
in
I'm,
just
saying
it's
another
way
for
someone
to
sort
of
more
easily
skirt
the
tests
right
now
that
would
catch
those
sort
of
changes.
So
we're
going
to
be
a
bit
careful.
D
B
B
B
D
B
D
D
C
Yeah
so
I
mean
the
ETF
I'll
Fabrice
I
have
a
document
somewhere
about
this
I
think
about
but
means
the
easiest
way
we
could
get
around.
This
is
just
having
just
executing
the
upgrade
face
like
in
the
like
in
in
companies
anywhere
like
the
op
op
stage.
We
do
both
first
and
up
and
then
upgrade
and
like
hack
it
that
way,
like
that's
possible,
to
do
in
five
lines
of
code.
C
If
we
have
to
like
do
the
full
I,
don't
know,
what's
even
required
to
first
do
the
out
face
then
instead,
then
some
upgrade
and
then
test
again
how
that's
possible
with
Christopher,
but
I
mean
we
can
do
a
quick
and
dirty
thing
for
one
eye.
If
we
want
so
I,
look
pretty
much
a
full
coverage,
but
still
not
have
to
focus
like
put
two
weeks
on
it.
D
B
Should
we
add
this
as
one
of
the
items
for
at
the
end
of
the
cycle?
I
don't
know
for
keeping
track
right
now
we
have
a
canonical
source
of
the
p0
items,
but
we
should
like
burned
down
list,
but
we
should
bear
them
now.
We
should
probably
like
add
that
to
the
canonical
list,
as
we
as
we
roll
towards
this
one,
a
release
I
mean.
D
D
B
Just
for
clarification
on
the
going
back
to
the
diamond
set
issue
for
self
hosting.
We
currently
don't
have
a
path
for
words
other
than
that
the
gaps
wants
to
add
a
new
strategy,
but
they
haven't
committed
yet
to.
C
D
C
D
F
F
D
I
think
I
think
we're
pretty
good
check.
That
I
think
we
have
agreement
from
the
gaps
that
if
we
write
the
code,
they
will
help
us
get
it
in
as
long
as
it's
reasonable
and
doesn't
raise
their
risk
in
terms
of
migrating
to
their
new
API
groups,
which
is
far.
The
reasoning
with
a
new
strategy
is
because
the
code
should
be
more
separate.
D
C
C
D
C
D
C
C
So
basically
it
had,
he
had
a
question
sleep
on
me
to
where
we
discuss
or
document
cube
and
compatibility,
but
that
was
not
as
cute
I
was
I
knew.
We
supported
when
I
said
that
we
need
to
sport
like
deploying.
If
we
have
a
cube
and
in
client
one
seven.
We
need
to
support
one
six
clusters
right,
API,
service
and
cubelets,
and
this
is
stuck
on
the
side
like
on
the
duck
cabanas
back
and
also
in
that
quest
to
follow.
But
may
should
we
be
even
more
clear
or
have
this
next
canonical
place
somewhere.
D
C
D
I
had
another
note
about
that,
because
the
official
version
skew
policy
for
kubernetes
and
official
in
quotes
is
we
support
version
few
of
two
or
two
minor
versions
for
cubelets,
so
I'm,
not
even
if
it's
the
current
version
mq
bed
and
can't
deploy
a
slightly
older
cubed
version,
I'm,
not
sure
we
should
be
blocking
people
on
a
to
version.
Skew
I!
Think
more
than
two
versions
is
fine.
That's
saying
you
can
only
skew
one
version
when
we're
doing
upgrades
may
not
be
sufficient.
C
B
C
The
the
plan
is
what
we
could
do
it
Lea.
We
add
the
API
by
using
a
job,
and
it
would
basically
edit
the
manifest
file
and
just
sing
like
local
health
CD
that
make
sure
it
comes
up
I
we
might
wanna.
My
plan
at
least
was
to
get
this
behind
them.
Oppa,
like
cubed
and
all
fourth
grade
applied
by
a
CD
or
something
like
I,
don't
know,
but.
C
Yeah,
so
basically,
we
had
to
somehow
reason
with
it,
but
I'm
not
sure
whether
it
should
be
in
the
main
I
didn't
include
it
in
it
like
main
apply
flow,
because
it's
kind
of
risky
and
Joe
said
that
the
normal
user
probably
wants
to
upgrade.
First,
the
control
plane
make
sure
everything's
working
then
maybe
move
on
to
my
cube
lessons,
maybe
a
CD
after
some
time.
Let
me
know
that
what
fall
couplets.
B
Start
much
safer
than
anything
else
to
upgrade
because
you
can
in-place
upgrade
them,
but
the
SCD
I
think
having
at
least
like
a
you
know,
I'm
going
to
rewrite
the
manifest
file.
I,
don't
I,
don't
know
whether
or
not
you
want
to
be
a
DM
or
a
job
to
do
it.
I
don't
feel
very
strongly
against
having
cuvee
DM.
Do
it
because
the
the
manifest
because
I
feeling
to
be
the
only
thing
that
some
static
manifesto
and
the
couplet
will
detect
change
and
re
start
right
so
well.
C
B
There's
a
time
window
interval
for
it
well,
there's
a
time
when
no
eternally
needs
to
get
them
in
the
code.
Now,
there's
a
time
window
internally
for
it
to
actually
do
the
channel
update
for
it
that
could
actually
updating
so
it
does
detect.
It
does
notify
the
main
and
then
restarts
it.
So
I
think
just
dropping
the
manifest
as
a
gated.
You
know,
fire
be
layer.
Type
of
thing
seems
reasonable
to
me,
unless
other
folks
have
objections,
as
you
just
going
over
at
the
manifest
yeah.
C
D
B
B
C
H
C
B
B
What,
if
somebody,
here's
the
weird
thing
like
how
you'd
have
to
on
the
node
detect,
because
somebody
might
have
configuration
files
that
might
vary
from
master
to
master
if
we're
getting
into
the
weird
space
right.
So
then,
then
like
what,
if
the
manifest
location
is
different?
What,
if
you
know
you
have
to
live,
detect
on
the
system
that
you're
running
on?
In
order
for
you
to
do
it?
Well,.
C
So
qubit
I'm
explicitly
at
least
in
the
define
duck.
It
says
that
supporting
and
apart
manifest
part
is
not
supported
like
the
next
to
diamond
command
in
the
clinical
and
it's
hard
coded
in
federal,
cubed
M,
so
I
think
that
she
finds
to
assume
but
safe
yeah,
but
well
indeed,
there
are
like
a
lot
of
corner
cases.
B
I'm,
okay,
with
a
job
but
you're
going
to
have
to
have
a
new
build
artifact.
Now
then
it's
going
to
be
a
pod
that
gets
built
this
part
of
coop
ADM
or
something
else
right.
Yeah
sounds
good
yeah,
it's
gonna
be
a
job
within
Canadien,
so
good
Ian
can
have
like
you
know,
pork
itself,
you
don't
I'm,
saying
right,
because
you
have
to
be
either
inside
of
Trevanian,
so
that
the
if
we
create
a
container,
it's
a
separate
thing
or
you
create
a
new
job
container
that
gets
published.
C
I
was
thinking
it
could
be
as
simple
as
using
button
box
or
something
but
I
mean
we
could
do
it.
We
could
say
we
can
also
engage
the
esidisi
in
in
like
cuban
amor
fourth
grade
and
something
at
CD,
and
have
it
like
manifest
like
right
directly
to
disk
right
now,
and
then
it
is
still
we're.
Keeping
this
like
cubed
M
upgrade
that's
not
over
dated
it's
still
like
working
on
a
strictly
like
API,
driven
API
communications.
D
C
C
D
C
C
So
it's
more
like
all
these
in
the
end,
I
did
end
up
with
a
lot
of
code
like
more
than
I
had
expected
when
I
start
thinking
about
all
importing
all
these
policies
that
we
have
like.
You
must
have
read
liked
you
better
before
this
one
and
such
things,
but
I
think
it
plays
out
really
well
I
mean
it's
it's
a
face
and
it's
in
a
separate
package
and
I
mean
cops
or
whoever
chorus
or
anyone
can
use
it.
Like
you
know,
the
CLI
code
is
really
small.
C
A
Be
worth
putting
that
link
to
that
talk
in
the
notes
as
well.
So
if
I
can
all
can
see
it
and
there's
also
one
tiny
thing,
but
I
noticed
that
someone,
when
reviewing
your
pull
request,
has
suggested
putting
line
breaks
between
sentences
because
it
makes
it
easier
for
a
new
markdown.
So
would
be
great
yeah.
C
C
D
A
Yeah
I
mean
these:
these
weren't
major
bits
of
feedback
but
I
think
they're
important.
So
the
first
one
was
just
that,
like
I,
don't
think
users
will
understand
the
terminology,
1.7
branch.
This
is
early
and
that's
like
a
humanities
developer
term.
What
is
the
1.7
branch?
Oh,
it's
an
implementation
detail
of
how
you
release
your
software.
Okay,
so
I,
don't
know
what
would
be
better
like
series,
1.7
series
or-
and
then,
if
there's
a
official
word
for
like
this
series
of
releases
that
make
up
a
set
of
patch
releases
within
a
minor
release.
A
Whatever
it
is,
we
need
that
word
instead
of
the
word
branch
and
like
when
you
say,
components
that
must
be
upgraded
manually,
you
have
to
give
the
user
some
kind
of
indication
of
how
now
that
could
just
be
a
link
to
a
document
on
Kiva
Nettie's
what
io
or
it
could
be
used.
The
package
manager
that
you
used
to
originally
install
these
things
to
upgrade
it
or
something
like
that.
D
Look
do
thing
instead
of
saying
branches
if
we
talked
about
like
released
tracks
with
that
make
more
sense
sure
understand,
because
we
could
just
drop
that
all
together
and
say,
like
latest
stable
version,
1
7
2.
Here
it
is.
But
then
what
happens
if
you're
going
you're
at
one
seven
one
and
there's
one
seven
two
and
a
1/8
and
add
you
pick
being.
A
D
A
A
bit
yeah
I
mean
just
for
the
sake
of
everyone
else,
so
I'm
the
wherever
there's
an
error
whenever
there's
an
error,
condition
I
think
the
last
thing
that
needs
to
be
printed
needs
to
be
an
excellent
explanation
of
the
error
and
also
it
should
return
a
nonzero
exit
code
and
I
mean
I,
can't
tell
from
your
Google
Doc
whether
it
returns
a
nonzero
exit
code
already
that
in
case
it
doesn't
yeah.
D
C
So
do
you
think
this
should
happen
with
plan
as
well
I
mean,
should
plan
return,
an
error
I
mean
this
is
love
the
output
of
plan.
It
will
definitely
return
an
error
if
you
try
to
like
upgrade
to
one
seven
two
and
you
have
one
seven
one:
cubed
em
and
yeah
yeah
I
think
not
specifying
the
false
flag.
I
think.
A
D
The
bottom
of
plan,
it
says
you
can
now
apply
the
upgrade
by
executing
the
following
command.
If
plan
fails,
it
says
you
need
to
do
something
else
before
you
can
do
upgrading
it,
presumably
wouldn't
print
those
lines,
and
it
would
return
a
nonzero
error
right.
So
if
it
fails
make
a
plan
or
if,
while
making
a
plan,
it
determines
your
cluster
is
not
able
to
be
verified,
it's
not
healthy.
The
config
is
not
okay.
If
it
can't,
you
know,
determine
what
version
should
be
rated,
it
should
return.
That's
real
good.
C
So
good
to
me,
I
mean
first
just
check
the
health
left.
Api
server
protect
the
own,
don't
already,
which
then
implies
that
there
are
C
and
I.
There
is
a
Siena
I,
provided
it's
all,
and
then
we
have
checking
that
it
is
self
hosted
if
it's
not
supposed.
If
it
is
for
the
use
of
that
it
will
upgrade.
When
you
do
the
apply,
it
will
happen,
it
will
upgrade
your
static
phone,
hosted
control
planes,
the
self
of
the
or
should
we
like
to
read
that
separates
I
mean
to.
D
And
I
guess
it
comes
up
because
it's
cube
admin
is
going
to
be
a
more
generic
administration
to
him.
It
should
be
a
client-side
tool
more
like
cube
petal,
which
is
cross-compile
for
different
client
platforms.
If
we
think
cube,
admin
is
more
of
a.
You
only
run
it
on
machines
inside
of
your
cluster,
so
you
always
run
it
on
a
master.
You
always
run
node.
Then
maybe
we
need
to
change
the
scope
of
the
tool
of
potential
changeling
yeah.
A
C
Is
it
I
mean
there
I
think
there
will
be
a
case
in
the
future
where
I
cubed
and
will
run
just
fine
on
Windows?
Well
like
the
only
thing
is
that
it
done
once
a
cube,
cube,
config
file
or
create
a
cube
holding
file,
and
then
you
we
just
have
to
like,
provide
a
package
for
Windows
and
also,
of
course,
a
some
kind
of
C&I
provider.
That's
working
both
linux
and
windows,
but
it.
D
Machine
in
the
cluster
so
I
think
that's
something.
Maybe
we
can
discuss
a
separate
meeting,
but
what
the
scope
we
think
of
cube
ivan
is
sort
of
over
time
and
whether
it
really
is
a
tool
that
we
expect
to
run
just
buzzer
note
or
whether
it's
a
tool,
we
think,
will
eventually
also
become
a
client
web
tool.
C
D
D
Ctl,
which
is
already
client-side
tooling,
we
have
cubed
admin
which
until
now
has
always
been
on
host
tooling
inside
of
your
cluster,
and
we
need
to
figure
out
what
the
future
of
both
those
things
are.
We
building
are,
we
s
are
building
more
client-side
tooling.
That
will
run
outside
of
your
cluster
for
administrative
purposes.
Would
it
make
more
sense
to
build
that
into
cube?
Control
itself
may
be
behind
a
queue,
control
admin,
you
know,
subgroup
and
I
know.
Openshift
has
an
open
shift.
D
Admin
command,
it's
separate
client-side,
tooling,
but
it
is
not
meant
to
run
it
by
the
cluster
and
then
to
run
like
you,
control
outside
of
the
cluster,
so
just
kind
of
a
metaphor
that
we
should
start
thinking
about
as
we're
adding
features
to
keep
that
and
then
is
whether
that's
the
right
place
to
put
them
or
whether
we.
What
do
we
want
to
restrict
the
scope
of
the
tool
or
sort
of
expand?
The
scope
of
the
tool
ya
know.
C
C
If
you
can
see
my
screen,
yes,
yeah
cool,
so
cube
Adam
and
it's
oh
forget
it
we'll
just
move
that
command
later
in
the
cycle
and
have
some
consequence
like
flag.
Sometimes
a
lot
of
experimental
upgrades.
I
think
this
will
be
a
useful
case
for
like
this
is,
of
course,
for
more
dams.
Like
oh
I,
wanna
upgrade
I
made
a
clasp
at
some
time
and
now
I
want
to
see
what's
new
in
1/8
and
what's
exciting
here
with
my
test
plaster
anyway.
Is
that
also
a
way
to
get
release?
Candidates.
C
D
C
Well,
it
reads
a
file
from
disk
and
uses
that
config
and
that's
config
all
the
rides
like.
What's
in
the
cluster
already,
we
can
fetch
like
when
we
do
Q
beta
minutes,
hopefully
from
173
on
we
upload
the
configuration
used
by
Celtic
map
and
then
just
read
it
and
keep
config
which
clients
use
for
talking
to
custom
and
print
config,
which
we'll
see
in
a
minute.
So
now,
when
I,
execute
cubed
I'm
also
play
cubed
an
upgrade
plan.
C
It
will
do
this
kind
of
thing
check
that
everything's
healthy
reading
configuration
from
the
cluster
checking
the
cluster
version.
Cube
am
client
version.
It's
stable
and
touch
cube.
Dns
closures
then
telling
me
that
I
have
a
1,
7,
0,
qg
cubelets
and
you
I
could
credits
one
son
to
and
yeah
sounds
good
to
me
to
add
something
here:
I
get
for
like
use
your
package,
manager'
jar
or
whatever,
to
create
giblets.
D
One
one
small
net
this
makes
it
look
like
I,
should
upgrade
my
cubed
1
7,
before
upgrading
my
control
plane
and
in
general
we
recommend
people
doing
it
in
the
other
direction.
I
think
we
technically
so
like
sort
of
maybe
are
okay
with
version
skew
and
patch
versions
where
the
cubelet
goes
ahead
of
the
master
a
little
bit,
but
we
don't
really
test
it
very
well.
So
it
is
true
that
those
going
to
be
upgraded
manually,
but
the
way
it's
printed
out
makes
it
look
like
you
should
do
that
first,
which
I
think
is
incorrect.
D
D
Have
some
sort
of
just
variant
or
or
maybe
switch
like
the
put
that
above
that
you
can
now
upgrade
so
it
says
upper
delays
patch
and
shows
you
what's
an
upgrade,
then
tell
people
with
FF
grade
manually
and
then
say
you
can
apply
the
upgrade
the
automatic
upgrade
or
whatever
using
fall
in
command,
because
in
the
very
last
piece
of
output
you
see
is
like
this
is
what
I
should
do
next
right
now,
it
looks
like
what
you
should
do.
Next
is
a
bigger
keep.
A
The
current
output
is
I
feel
like
it's
factually
correct,
but
that
the,
if
you
think
about
the
tasks
that
the
users
trying
to
achieve
from
that
perspective,
it
might
not
be
the
most
optimal
output
and
I.
Don't
know
what
the
answer
is
yet
I,
don't
know
what
the
like
that
the
I
don't
have
a
concrete
proposals
for
how
to
improve
it.
Yet,
but
I
just
want
to
say
that
that's
how
I'd
make
you
feel
at
the
moment,
and
so
I'll
probably
think
about
it.
C
C
C
H
C
Not
as
I
expected
to
these,
not
yes
I'll,
take
that
so
we
have
dry
run
fourth
and
version
yes,
should
we
read
from
standard
in
like
so?
If,
if
yes
is
not
specified,
should
we
like
say
that
this
is
what
I'm
gonna
do
press
like
right,
yes
into
your
terminal
or
something
to
accept
this
thing,
or
should
we
just
proceed
I
mean,
should
we
do
the
app
gets
kind
of
thing,
or
should
we
just
wait.
I
D
Refika,
cheaper
mode
has
a
similar
thing
where
cheek
liable
like
do
you
want
to
create
instant?
You
know
yes
or
no,
and
there's
also
a
way
to
tell
it
like
skip
user
input
like
just
go
ahead
and
do
it
and
so
for
automation.
You
always
have
the
flagger
set
the
environment
to
skip
user
input,
but
if
someone's
running
interactively,
then
you
prompt
them
to
make
sure
that
they
are
charged
yeah.
A
C
D
D
C
C
Yeah
so
there
we
have
three
demon
set
pods
that
are
terminating
because
I
think
there's
a
crazy
period,
some
seconds
which
I'll
remove
in
the
next
iteration.
So
this
just
deployed
the
image.
Oh
well
now
the
API
surveys,
one
for
this
one
API
server,
but
in
the
future,
all
others
of
all
and
one
more
thing
we
could
do
plan
print
config
for
so
for
those
who
are
really
interested
in
what?
If
exactly
going
to
happen
in
the
cluster
which
option
did
I
use
to
like
three
months
ago,
when
I
did
cubed
a
minute?
C
C
C
C
A
C
A
C
C
My
question
was:
do
you
wanna,
should
we
start
mucking
with
verbosity
flags
and
levels,
and
all
this
with
like
one
simple
command,
that
shows
this
amount
of
output,
or
should
we
just
have
some
some
small
I
mean?
This
is
essentially
also
print
config.
My
that
probably,
would
be
a
better
naming
wording.
A
D
And
that'll
also
be
a
good
like
if
you
were
running
an
older
version,
key
lab
and
it
didn't
put
the
contrary
to
a
config
map,
and
you
want
you
guys
in
print
config
that
says
couldn't
find
it
that'll
also
be
really
helpful
for
people
debugging
they're,
trying
to
run
it
up
for
Aidan.
They
need
to
pass
in
their
own
config
because
it
Disney
system.
If
you
pick
map
that'll,
make
it
more
clear,
yeah
and
it'll
say
sorry
doesn't
exist.
Most
have
been
running
an
older
version
to
that.
C
C
Well,
you
did
cube,
am
in
it
and
one
in
Cu
flannel
and
you
specify
like
for
network
subnets
or
whatever
the
flag
is
named
in
order
to
use
flannel
or
calico
or
whatever
that's
requires
that's
like,
and
then
you
forgot
well
cube,
Adam
didn't
upload
the
version,
the
config,
and
you
can't
remember
you,
the
death
of
you
upgrade
and
then
just
back
with
a
misconfigured
cluster.
So
should
we
like?
G
C
A
B
A
big
fan
of
like
I
know
we
don't
actually
do
this
much
in
Covidien
during
kubernetes
in
general,
but
uhm
of
just
prompting
and
shelling
to
the
editor
for
when
we
do
configuration
type
of
things.
I
did
this
before
and
some
configuration
management
type
of
land
where
we
had
super
complicated
configurations
we
had
to
deploy
to
multiple
clusters.
Oh
my
god.
The
world
repeats
itself,
but.
B
We
literally
punched
up
the
you
something
config
and
it
would
punch
up
the
config
for
you
and
you
could
edit
it
live
and
it
would
either
pull
or
push
based
upon
whether
it
have
the
data
right
and
then
whatever
you
saved,
it
automatically
push,
and
there
was
a
nice
user
experience
so
be
like
kuba,
DM
config.
And
then
you
just
punch
up
a
default
and
then
it
would
save
you
know
or
push
the
back
end.
C
D
A
It's
interesting
I
would
just
say,
like
my
own
personal
opinion,
on
the
like
launch
medicine
for
the
user
to
change
some
config
thing.
I
have
mixed
feelings
about
that.
It
feels
like
kind
of
an
advanced
user
tool.
H
C
A
Cool,
we
are
by
the
way
about
ten
minutes
over
time,
I'm,
okay,
to
stay
from
it,
a
few
more
minutes,
and
but
if
people
need
to
drop
off
and
that's
okay.
C
A
One
other
idea
is:
maybe
this
is
obvious
and
you're
going
to
do
it
already,
but
when
the
user
actually
does
run
the
apply
command,
then
you
should
print
upgrade
the
couplet
as
the
next
thing
that
they
should
do,
because
I
did
one
thing
and
it
told
me
to
do
the
next
thing.