►
From YouTube: 2017-05-16 17.03.46 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
A
C
Yes,
so
there's
been
some
discussion
about
where
we
are
in
terms
of
the
STD
operate,
it
was
proposed
to
actually
like
implement
get
to
the
operator.
My
understanding
was
that,
as
a
group,
we
were
waiting
on
a
better
understanding
of
what
the
SUV
operator
was
and
what
it
whether
it
met
our
need,
but
I
thought
one
way.
We
could
actually
figure
that
out.
It's
how
the
quests
are
working
on
defining
what
the
operator
is
and
what
it
gives
us,
but
I
figure.
C
B
The
first
and
foremost
requirement
is
that
everything
that
is
done,
this
cycle
should
be
listed
as
experimental
right
like
a
parliamentary
procedure
or
some
type
of
process
as
a
sig
that
we
follow
that
adheres
to
the
broader
space
that
basically
says
like
we
have
a
way
of
vetting
ideas
such
that
you
know
we
can
turn
prototypes
and
you
know
if
it
doesn't
meet
our
needs.
We
can.
We
can
expunge
them
right
because
I
do
not.
I
did
not
envision
this
being
a
first
class
thing
in
a
near-term
release,
cycle
I
envisioned
it
being
is
hey.
B
We
want
to
prototype
and
evaluate
it
and
see
if
it
works
for
our
needs-
and
you
know,
I
was
not
actually
Jamie
came
up
and
asked
on
my
on
a
break
out
that
way
created
from
the
last
AJ
meeting
and
asked
if
you
want
to
could
work
on
the
side
and
I
said
sure,
go
ahead,
so
you
know
I
think
we
should
absolutely
encourage
the
idea
prototypes
and
plug
ability
and
accessibility,
but
we
definitely
do
need
to
have
a
systematic
review
of
the
coordinates
as
well.
So
yeah.
C
I
think
I
think
there's
a
there's,
a
meta
point
there,
which
is
like
you,
a
DM,
is
well
selector
bicycle
we're
trying
to
like
figure
out
the
awesome
way
to
run
kubernetes
right,
and
there
is
a
danger
that
people
will
confuse
our
experiments
with
recommendations.
Well,
it'll
have
to
be
clear.
Then,
like
anything,
that's
marked,
the
question
is
like:
can
we
put
it
into
case
case
or
do
we
have
to
do
somewhere
else
right?
If
it's
an
experiment
like
blue
cube,
is
a
great
experiment.
A
C
C
Mean
as
long
as
we
are
yeah
I
think
I
think
this
stays
a
few
vidiian
is
confusing
to
me
and
I
feel
like
I'm,
relatively
close
to
it.
So
I
certainly
I
certainly
think
that
we
compute
that
maybe
we
maybe
we
continue
with
defining
of
requirements,
and
then
it
may
be
may
be
that
we
have
no
requirements
for
the
sake
of
Estoril
man
and
that
we
or
maybe
the
edge
to
the
operator,
meet
our
requirements
perfectly
and
we
can
just
integrate
it
and
we're
done
right.
A
To
be
clear,
I,
don't
I,
don't
have
a
strong
opinion
on
the
like
is
alpha
data
etc
sufficient
for
demarcating
experiments.
I
just
put
out
those
my
dear:
it's
no
strong
opinion
that
I
hold
no
but
sure
yeah,
okay,
so
requirements
the
fcd
operator,
I.
Guess
first
thing:
do
we
have
anyone
in
the
room
here
who
knows
the
LCD
operator
very
well?
Well,.
B
C
I
would
totally
disagree
of
that.
I
think
the
worst
thing
we
can
do
when
we're
trying
to
come
up
with
ideas
is
to
fixate
on
a
particular
solution
right.
So
what
do
we
want
right?
We
want
throughout
FTD.
Let's
go
crazy
right.
Should
we
get
rid
of
that
today
right
so
we
implement
our
own
entity.
Shall
we
in
OHS?
Hey,
you
know,
that's
the
IDH
or
whatever
the
verb
is
right.
You
don't
wanna!
You
want
to
fixate
on
a
solution
before
you
understand
what
we,
what
we
really
want.
So
yes.
E
D
C
D
So
my
first
question
is:
is
it
the
role
of
the
operator
or
some
kind
of
solution?
So,
if
you
have
like
a
solution
which
manages
the
data
store,
is
that
the
role
of
that
data
store
to
recover
itself?
What
do
we
need
to
suffer
a
component
to
handle
more
advanced
recovery
scenarios?
This
is
like
what
Justin
I
were
talking
about
when
I'm
not
quite
sure
what
like
the
dedicated
role
of
just
a
net
CD
thing
should
be.
D
E
The
way
that
we're
structuring
it
right
now
is
that
the
SPD
operator
itself
manages
making
backups,
but
then
the
recovery
doesn't
have
to
be
here,
but
we
put
it
into
a
boot
coop
sub-command.
So
there's
good
recover
so
right
now
that
just
is
able
to
recover
from
the
live
at
CD
and
then
what's
currently
being
worked
on,
it's
being
able
to
recover
from
an
SD
deep
back
up
specifically
for
self
hosted
at
CD.
So
it's
not
so
close
that
we
don't
know
how
you're
running
it
CD.
E
E
So
right
now
the
assumption
is
that
you
are
bringing
the
backups
to
a
node
that
will
be
able
to
recover
and
by
that
I
mean
like
there's
I
could
with
running
on
it,
making
the
tool
smarter
and
knowing
how
to
source
directly
from
SD
or
s3
or
from
some
other
volume.
We
can
do
that
over
time,
but
the
assumption
just
starts
with
bring
a
backups
and
then
use
that
back.
I
think.
B
If
I
could
take
a
step
back
and
like
make
sure
that
we're
requirements
aren't
slightly
myopic,
we
still
want
to
support
justing
McD's
clusters
that
are
external
to
the
system
right,
because
people
will
have
people
will
come
with
or
megan's
right.
So
I
think.
That's
absolutely!
You
know
she'll
be
listed
as
the
ingest
a
requirement
right
for
what
it
means
to
support.
E
B
E
Those
are
two
high
priority
things
on
the
sed
operators,
slash
self
host
identity
roadmap,
one
is
which
is
recovering
directly
from
backup
and
the
other
is
supporting
TLS
in
self-assertive
seek
answers.
So
there
are
open
issues
about
that
includes
design
documents,
but
those
of
those
are
really
high
priority,
and
so.
E
C
E
That
works
that
works
now.
So,
if
you
using
the
SPD
operator,
if
you
bootstrap
a
self-hosted
FPD
cluster,
then
it's
just
a
matter
of
interacting
with
that
third
party
resource
to
scale
will
be
a
city
cluster
like
you
would
any
other
sed
cluster
that
you
launched
using
the
operator.
So
you
just
tell
like
you
want
three
copies
and
it
scales
them
up
and
then
right
now
what
we
do,
but
this
is
optional,
is
just
we
force
that
home.
They
end
up
on
master
nodes
using
paints
and
node
selectors
and
such.
C
F
B
B
F
G
John
5:3
on
yourself,
you
can't,
with
the
view
of
booms
upgrades
to
and
and
the
next
manual
version
go
for
patch
version
you
can
done
with.
So
if
you
want
to
come
with
another
man
aversion,
you
have
to
our
recovery,
just
limitation
itself,
so
I
think
the
operator
will
not
have
you
do
anything
about
it
and
but,
like
I,
think
algorithm
will
take
I'm
Charlton.
The
current
snapshot
is
version
aware,
so
actually
it
can
help
you
to
like
renew
a
cluster
and
you
will
get
the
actual
for
that
version.
F
The
whole
version
of
the
snapshots,
yeah
I,
think
that
Ron
is
okay,
but
snapshotting
really
only
helps
you
if
you're
upgrade
fails
immediately.
But
if
you're
upgrading
that
CD
is
totally
broken,
you
can
restore
to
a
snapshot
because
you
haven't
lost
any
data
if
you
upgrade
and
ICD
stops
working
ten
hours
or
two
weeks
later,
and
you
have
no
path
to
go
back
to
a
version
that
actually
does
work,
then
I
think
that's
a
problem.
F
I
know
that
Matt
Leggett
at
Google
worked
very
closely
with
the
core
OS
guys
when
we
did
the
sed
to
that's
d3
upgrade
path
to
make
sure
we
had
a
downgrade
procedure.
I
think
it
was
a
little
bit
hackish
and
somewhat
manual,
but
there
was
a
way
for
us
to
rollback
a
live
cluster
down
that
city
to,
and
maybe
that
was
just
snapshot
the
new
database
and
we're
sorted
an
old
version,
but
but
I
think
it
is
possible.
It's
not
a
limitation
of
ICD
well.
G
That's
I
rolled
that
seems
only
the
mutation
factory.
So
that's
why
there
is
a
hack
arm
so
for
downgrade.
There
are
several
issues:
first
ones.
If
you
only
want
the
data,
that's
easy.
If
you
want
to
like
do
a
live
downgrade,
that's
hard,
because
your
composite
cannot
communicate
with
each
other.
If
you
just
want
to
like
take
everything
off
and
dump
the
data
and
then
load
into
the
old
database
that
feed,
ok,
yeah,
that's
what
we're
doing
then.
You.
B
G
G
Yes,
sure
I
will
do
it
later,
because
the
problem
is
like
if
you
are
running
a
harvard
university
and
its
use
a
new
feature
and
how
I'm
going
to
roll
back
that
feature,
I
will
just
ignore
it
or
what
happens
to
another
new
face
%
all
when
you
introduce
a
new
like
data
into
F
idiot,
what
I'm
going
to
do
with
that
new
data?
Well,.
B
G
B
H
I
have
a
question:
if
we
can
support
step
shelf,
restore
right,
wouldn't
that
solve
the
problem
but
say
everything
is
bad
and
you
know
upgrading
the
word
and
the
schema
has
changed
dramatically
during
GCD
upgrade,
but
even
before
that
great
we
do
the
backup
and
we
provide
guidance,
another
story
with
that
two
previous
version
right
with
the
previous.
That's
all
the
problem
so.
G
I
think
the
argument
is
like:
maybe
you
have
a
sub
tell
file
it's
very,
like
manual
bug
and
you
hit
it
over
time
and
the
Eurocopter
cannot
run
anymore
and
you
don't
want
to
keep
that
mental
usual
anymore.
So
we
want
to
a
some
grapes
and
you
see
one
keep
keep
the
delta
from
the
link
old
word
into
the
new
version.
You.
A
B
F
Don't
want
to
rattle
on
this
too
much
I.
Just
think
that
if
we're
listening
out
requirements,
then
downgrades
I
think
would
be
a
great
requirement.
I'm,
not
sure
if
we
can
actually
meet
that
requirement,
but
that's
something
I
would
love
to
see
us
build
into
the
system.
It's
not
build
an
automated
solution.
It
may
not
be
possible
to
automate
right,
but
we
should
have
a
way
to
do.
Taverts,
yeah.
G
B
G
A
little
bit
different,
if
you
think,
really
think
about
some
hoses
a
CD
because
on
your
API
server,
depends
on
ICT
to
run
and
your
itd
depends
on
cookies
to
start
and
your
cookies
depends
on.
It
gets
over
two
stars
so,
where
you
do,
when
you
have
a
total
power
of
not
in
can
start
because
there's
a
security
tendency
there
I
think
that's
a
drag
requirement.
If
we
want
to
sell
hole
to
a
TV,
I
mean
that
what
the
check
for
interest
for
right.
A
A
B
G
Our
fuel
case
it's
even
for
power
off
right.
So
the
fourth
case
is
your
power
off
or
part
of
your
like
cluster
is
like
you
have
three
muscle
noted
power
off
one:
that's
not
even
like
desire,
and
you
do
me
like
you,
don't
even
need
to
point
to
make
it
work
right.
Another
case
is
pop
your
power
off
all
of
the
machines.
That's
where
cap
point
takes
into
place
and
even
for
like
recovery.
G
There
are
several
pace
so
the
first
case
you
lose
quorum,
that's
already,
but
if
you
distance,
one
I
think
I'm
going
to
have
three
dozen
not
like
in
terms
of
recovery,
so
I
say
the
operator
can
just
handle
it
by
like
giving
a
number
and
add
a
new
member
in
right,
so
like
II
live
will
recover.
There
are
several
cases
if
you
won't
be
accurate.
H
Basically,
when
you
upgrade
the
cluster
or
see
many
times
because
of
the
top
issues
like,
for
example,
some
containers
could
you
know
just
walk
in
for
various
reasons
and
basically
that
results
in
certain
amount
of
notes
not
being
upgraded
just
operating
system
issues
or
anything
like
that.
So
it
will
be
nice
to
design
somehow
provides
some
guidance
on
how
to
get
from
a
state
either
from
continuing
the
upgrade
anybody
thinking
it
doorbells
or
drink,
something,
that's
being
timed
quite.
G
A
lot,
so
that's
actually
it's
pretty
important
to
like
operate
actually
because
if
you
have
a
lockdown
has
any
members
and
your
closets
in
Radian
mode,
and
if
you
lose
one
number,
maybe
you
will
have
to
go
into
the
disaster
recovery
mode
because
you
lose
clone
so
for
I.
Take
the
operator
up
for
our
indentation.
It
does
periodically
house
checking.
So
it
builds
that
one
number
it's
not
healthy
for
like
10
seconds
like
you
can
set
the
threshold.
You
will
just
give
the
number
and
add
a
new
number.
You.
C
As
this
other
stuff
on
the
agenda
now,
the
discovery
like
is
that
going
to
be
something
which
we
think
is
part
of
this,
or
is
that
made
a
separate
system
which
this
will
depend
on?
Do
we
expect
like?
There
are
two
ways
to
discovery
right?
One
of
them
is
to
say:
oh,
we
have
that
TV.
It
did
discuss
somehow,
so
we
every
the
other
one,
which
is
to
say,
there's
a
discovery
system
and
this
operator
plugs
into
it,
and
other
people
have
a
view.
We
say,
like
discovery,
question
mark,
do.
C
G
B
B
What
we
can
do
is
we
can
put
this
into
an
issue
and
there's
already
the
self-hosting
issue
with
the
entity
operator
in
the
Covidien
repo,
and
we
can
refine
this
list
and
we
can
also
ducktail
the
document
as
we
change
over
time.
But
I
don't
think
we
shouldn't
make
sure
that
we
enumerate
any
changes
that
we
want
to
or
any
PRS
that
we
want
to
that.
It
covers
certain
items
and
we
also
list
on
what
are
hard
requirements
and
what
are
softer
nice
to
have
I'd.
I
A
Sounds
good
just,
do
you
get
a
12
to
take
step
back
and
ask
a
little
bit
of
context
around
this
discussion?
A
B
Does
tells
us
both
I
had
not
originally
intended
to
work
on
it.
This
cycle,
but
Jamie
had
cycles
and
was
interested
in.
You
know,
taking
a
stand
better
I
think
I
do
agree
that
perhaps
we
should
take
a
minor
setback,
but
I'll
also
allow
the
prototyping.
So
as
a
sake,
we
might
want
to
have
a
defiant
mechanism
by
which
we
want
to
add
the
ability
to
for
people
to
hack
on
things.
We
don't
need
to
incorporate
them
right
away
right.
E
A
I
mean
that
I
wouldn't
have
turns
up
anyone
doing
that.
Yeah
I'm
just
I'm,
just
curious.
That
way
where
we
see
this
landing
and
when's
the
rush.
Let's
go
wise
yeah.
I
I
So
it's
it's
important,
but
I,
don't
think
for
in
a
real
in
a
timeframe
like
having
something
stable,
but
we
could
maybe
have
a
fork
or
something
a
well-known
fork
somewhere
in
the
meantime,
maybe
Jamie's
or
whatever,
but
there
was
some
pushback
from
from
others
in
the
community
for
for
merging
things,
if
right
now,
with
with
the
dependencies
so
like
the
gold
at
once,
and
should
this
this
TPR
stuff
under
chorus,
the
chorus
repo?
Should
this
live
inside
core
and
so
forth.
This
will
be
easier
to
deal
with
after
we
moved
cue
Batum
of
people.
K
L
Note
on
the
fork
we
kind
of
did
that
with
cube
admin
in
our
initial
development
and
one
problem
that
I
kind
of
saw
with
this
was
we
ended
up
kind
of
pushing
in
a
bunch
of
code
all
at
once,
because
the
fork
developed-
and
there
was
about
you-
know
five
thousand
lines
of
code
that
all
entered
kubernetes
codebase
at
once.
So
I
would
just
request
that
once
we
finally
do
merge
the
fork,
we
do
it
incrementally
in
in
reviewable
chunks.
So.
A
I
I
A
Any
line
I
don't
think
we
need
an
explicit
sort
of
sake,
managed
place
for
these
experiments
to
happen.
I
think
it's
fine
for
it
to
just
happen.
In
the
completely
natural
sort
of
github
style
of
some
work
happens
in
someone's
fork.
Some
that
person
can
give
other
people
access
to
push
that
fork
if
they
want
to
collaborate
on
it.
That's
all
give.
I
A
A
H
Name:
oh
okay:
okay,
a
little
is
good.
There's
a
saucer
of
English
it
last
week,
I
started
digging
into
engine
testing
framework
and
adding
CI
plugin
support
and
basically
a
couple
of
pull
requests
and
flies.
I
know
the
Jenny.
It's
a
work.
I
also
use
kids
work
to
basically
test
the
plants
in
a
plugin
right,
so
I
just
try
to
understand.
H
L
H
L
H
F
H
H
H
H
F
I
F
B
There's
a
bot
I
mean
like
if
we
have
BOTS
labeling
in
place
on
this
repo
I
didn't
I
didn't
see
the
Box
like
Auto
doing
stuff
from
its
repository.
That's
another
issues.
We
had
that
auto
tooling.
We
might
not
have
to
give
broader
Commission's,
but
you
know
there's
plenty
of
folks
who
run
this
column,
we're
already
meat
eaters
across
number
of
other
areas,
that
that
know
the
jurisprudence
and.
I
I
I
I
think
it
would
be
beneficial
if
we
could
get
more
more
than
three
persons,
because
people
will
use
or
are
using
the
class,
the
lifecycle
bugs
tic,
toc
lifecycle,
feature
requests,
PR
reviews
proposal
etc,
and
they
think
they're
targeting
the
right
people
but
they're
going
to
print
it
3-1.
So
if
we
could
get.
I
K
I
L
I
D
I
I
We
have
the
certificate
faced
by
Fabrizio
Apple,
pull
request
there.
It's
a
really
good
one,
but
I
haven't
had
time
to
review
it
fully.
It's
a
couple
of
hundred
of
lines
so
have
taken
me.
Some
some
days
have
other
stuff
to
do
as
well,
but
I
looked
it
and
preview
as
soon
as
possible.
I
think
he
was
ISO
Fabrizio
here
somewhere,
yeah
cool.
A
I
I
Late
in
the
one
six
cycle,
as
we
know,
we
we
switch
to
using,
we
should
switch
the
CNI
provider
to
using
like
a
status
hook
in
the
cube
list
and
that
had
a
consequence
that
the
node
is
marked
not
ready
when
when
CNI
is
punished,
uninitialized
right
so
and
when
a
node
is
marked,
not
ready.
It's
unschedulable-
and
this
is
kind
of
well.
I
This,
as
we
know,
broke
1/6
the
16.6
old
release,
and
we
we
just
did
the
first
a
key
thing
that
to
not
check
if
the
node
is
ready,
just
see
that
it
traced
it.
But
still
when
we
buy
self-hosting,
we
want
to
first
start
up
the
new
excel
sorted
api
server
that
works
well,
because
demon
sets
aren't
using
the
scheduler.
I
But
then
we
want
to
deploy
the
scheduler
as
a
deployment
and
the
controller
manager.
And
there
we
hit
some
problems
because
the
scheduler
can't
schedule
popstar,
not
ready
node,
and
this.
The
only
know
that
exists
is
not
really
because
CNI
is
an
it
initialized.
So
there
this
is
a
problem
and
it's
pretty
well,
it's
breaking
self
hosting
at
the
moment,
and
there
is
a
possible
fix
and
it's
probably
going
to
be
merged
for
one
seven,
but
it
still
needs
work.
The
first
thing
to
do
is
make
the
node
controller.
I
A
I
J
A
A
A
A
B
There's
a
big,
long-standing
gripe
that
I've
had
for
I,
don't
know
how
long
the
organs
project
now,
here's
the
about
having
a
actual
written
down
state
machine
for
the
Goulet
right.
If
you
had
an
explicit
state
that
outlined
this,
which
is
kind
of
what
you're
proposing
that
would
alleviate
this
problem,
which
is
you
know
if
we
marked
every
possible
condition,
which
is
all
the
edges,
there's
an
active
state?
That's
that's
a
lot
of
edges
right,
yeah,.
I
So
the
proposal
is
that
the
the
node
controller
will
create
pain
from
the
conditions
and
the
scheduler.
Well,
any
workload
that
you
want
to
schedule
must
tolerate
those
paints.
If
you
want
to
schedule
on
them,
but
the
problem
is
we
have
to
keep
backwards,
compatibility
and
the
scheduler
will
in
1/8
I
think
it
was
proposed.
I
C
I
It's
problematic,
I,
don't
know,
I
still
have
to
talk
to
David
Oppenheimer
a
little
bit
more
I
was
hoping
that
we
could
get
well.
I
was
hoping
that
we
could
get
the
the
new
behavior
of
the
scapula
to
sports
scheduling
on
not
ready
nodes
into
one,
seven
behind
a
be
a
feature
flag
or
something
because
that's
coming
in
one
eight,
probably
anyway,.
H
I
A
A
I
F
L
A
I
L
A
I
I
Yeah
they
they
run,
we
can
make
them
run
on
the
whole
network.
Yeah
yeah,
that's
not
a
problem,
that's
what
we've
been
doing
so
far,
but
we
somehow
have
have
to
like
fight
off
with
scheduling
the
schedule
on
until
it
works.
The
way
it
should
work
with.
I
should
be
able
to
tolerate
the
node
network,
isn't
ready.
I
I
A
I
Currently,
we
so
first
initially
it's
a
static
pod
and
the
self-hosting
phase
in
cube
a
DM
basically
turns
it
into
a
deployment
for
for
the
scheduler
and
controller
manager.
And
that's
that's:
okay,
that's
what's
working,
but
this
specific
issue
was
that
the
self
or
the
static
port
hosted
scheduler
or
the
scheduler
in
general
doesn't
allow
scheduling
on
not
ready
nodes.
So
it's
it's
kind
of
a
problem,
but
the
general
static
port
self
hosted
migration
that
works,
so
they.
E
Were
like?
Oh
sorry,
I
was
going
to
say
I
linked
in
in
the
nodes.
Another
issue,
that's
related
to
the
flakiness
issue,
about
the
the
API
server
actually
turn
off
health
checks
right
now,
because
that
comes
problematic
when
we're
using
the
file
lock
and
coordinate
between
the
rio
api
server
and
the
static
api
server,
both
during
the
pivot
and
during
check
framing
so
I
kind
of
outlined
in
that
issue.
But
it
is
something
that,
when
you
introduce
the
the
checkpoint.
E
A
To
interrupt,
but
we're
about
to
run
out
of
time,
I
just
want
to
make
sure
that
other
items
on
the
agenda
that
people
think
are
important
to
get
a
30
seconds
can
I
also
suggest
that
I
think
the
self
hosting
issue
is
really
important.
People
who
are
interested
in
it.
Please
I
encourage
you
to
carry
on
chesil
and
slack
about
it
up
the
meeting
and
and
anyone
who's
interested
in
volunteering
to
work
on
it.
A
Please
do
make
sure
that
you
talk
to
the
right
people
to
get
involved,
so
let's
just
run
through
these
last
few
items
quickly
who
will
bring
bootstrap
tokens
to
the
beta
them,
because
you
said
you
had
a
proposal
up
soon.
Yeah.
I
A
B
Instead,
it's
a
secrets
but
there's
there's
a
there's,
broader
authorization
questions
that
folks
will
want
to
take
a
look
at
think
about
because
we're
pushing
everything
into
a
single
namespace
and
for
our
backers
that
kind
of
lead,
there's
multiple
owners
and
we
have
multiple
people
who
could
potentially
own
those
certificates.
So
it's
just
a
broader
question.
I
think
we
need
to
start
thinking
about.
As
we
start
to
push.
You
know,
search
into
actual
secrets,
yeah.