►
From YouTube: SIG Cluster Lifecycle 2021-02-23
A
A
Maybe
we
should
start
doing
introduction
for
new
participants
for
the
meeting.
Do
we
have
somebody
new
on
the
call
who
wants
to
introduce
themselves
to
the
group.
B
I
was
new
last
meeting,
but
I
never
introduced
myself.
My
name
is
kirsten
schumi.
I
work
on
the
eksd
team
and
we're
trying
to
just
get
more
involved
in
this
community.
So
I'm
encouraging
people
my
team
to
come.
So
hopefully
they
join
too.
A
C
A
A
Basically,
what's
what's
happening
in
this
change
is
that
we
are
now
waiting
for
the
node
list,
informer
that
the
couplet
has
to
synchronize
before
we
return
the
node
object.
The
reason
for
that
to
matter
ceiling
is
that
they
are
trying
to
avoid
unwanted
descheduling
of
pots
from
a
node.
A
A
A
You
will
see
that
basically,
we
are
pulling
thousands
of
times
in
a
certain
function
that
that
is
responsible
for
getting
the
node
object
from
you
know:
either
the
local
initial
node
object
or
the
node
object
from
the
cluster,
and
I
have
not
from
what
I'm
seeing
this
is
not
raising
enough
attention
yet
to
some
of
the
maintainers.
A
From
my
perspective,
we
have
to
fix
this
polling
because
it's
it's
just
very
buggy
yeah,
so
we're
looking
at
that.
I
already
posted
in
a
few
channels
because
users
started
reporting
it.
So
hopefully
more
people
know
that
we're
working
on
a
fix,
yeah
does.
D
Anybody
have
comments,
so
I'd
have
to
look
at
the
code
but,
like
it
seems
weird
we
is
it.
Is
it
a
separate,
go
routine,
that's
like
being
kicked
off
continually,
calling
it
because
there's
already
a
bunch
of
code
in
the
client
that
basically
does
exponential
decay
with
jitter.
D
A
I'm
very
confused
and
confused,
because
I
was
hoping
that
this
po
here
will
block
the
particular
goro
team
right.
But
it's
not
for
some
reason:
it's
either.
This
function
is
being
called
in
a
go
in
a
separate,
go
routine
itself,
because
I'm
seeing
the
pole
being
executed,
like
thousands.
F
I
was
just
thinking
it's
also
interesting
because
it
looks
like
some
of
the
motivation
is
to
make
sure
that
we
have
the
correct
node
labels,
but
in
a
previous
version
we
said
the
cubelet
can't
apply
node
labels,
so
labels
are
now
asynchronous,
so
there
is
no
guarantee
that
you
have
the
correct
node
labels,
even
if
you've
synced
like
there
is
a
missing
condition.
If
you
want
to
start
making
this
guarantee.
So
it
feels
like
the
whole
thing
maybe
may
need
a
deeper
evaluation.
A
A
And
somebody
like
here
posted
that
a
simple
workaround
is
to
just
return.
The
initial
note
object
that
is
not
from
the
question
hello.
A
A
Yeah,
but
basically
I
just
wanted
to
give
this
as
a
psa
we
are
working
on.
I
see
that
somebody
from
scheduling
join
the
conversation,
hopefully
we're
going
to
have
a
fix
in
a
couple
of
days.
One
annoying
problem
is
that
this
original
pr
from
derrick
was
backported
to
the
whole
support
sku.
D
And
we
I
mean
a
part.
A
part
of
this,
though
too
is
like
you
know
if
you
wanted
to,
because
it's
causing
issues
other
places
in
the
stack
like
we
can
pull
trump
cards
and
say,
like
you,
caused
a
regression
across
every
single
life
cycle
tool.
We
are
going
to
revert
this
change
so,
and
I
think
that
that
is
a
fair
thing
to
do
like
it's
a
fair
until
we
can
get
a
proper
fix
in
place.
D
They're,
probably
not
subject
to
this
because
they're,
not
necessarily
they're
they're,
running
they're,
not
running
the
with
a
hosted,
cooperate,
hosted
environment,
especially
for
the
control
plane.
So
you
know,
but
that
doesn't
mean
that
doesn't
affect
thousands
of
other
people.
F
Just
I
I
don't
believe
I
I
think
we're
running
it
as
a
as
by
the
cubelet
under
the
keeper.
I
don't
know
that
for
sure
I
can
check,
but
I
think
I
think
I
think
his
running
is
make
a
very
similar
configuration
to
cuba.
A
Well,
in
any
case,
I
from
what
I'm
seeing
in
the
ecosystem,
there
is
increased
demand,
for
you
know
better
support
for
edge
cases
in
kubernetes
and
things
like
that.
It's
really
it's
an
anti-pattern.
D
Yeah,
you
could
use
that
too,
as
well.
It's
breaking
a
bunch
of
deployments,
just
you
know,
revert
until
you
can't
have
a
proper
fix
in
place
that
solves
this
problem.
That's
a
fair
way
to
approach
this
and
you
know
if
you
need
fly
cover
for
that.
You
know
just
poke
me
and
I
think
that
is
standard
operating
procedure.
To
be
honest,
like
there's
a
number
of
times
where
a
fix
has
caused
a
regression
throughout
the
stack
and
we've
had
to
do
exactly
that
until
we
can
get
a
proper
fix
in
place.
A
A
D
Yeah,
I
definitely
put
some
gates
in
place.
You
know,
maybe,
as
part
of
the
the
revert
it'd
be
like
a
revert
plus
test
like
you
know,
or
do
two
separate
pr's.
A
Yeah
yeah,
oh
yeah.
I
will
definitely
propose
this
as
release
blocking
for
121
and
request
that
we
fix
it.
A
Cool
moving
to
soup
project
updates,
I
only
added
for
cube
adm.
We
don't
have
anything
interesting.
Mostly
a
fairly
calm
release
is
going
to
be
121..
We
are
following
some
interesting
changes
in
question
api
that
will
affect
how
cube
adm
is
used
there.
A
These
two
are
for
machine
attestation
and
the
other
one
is
for
note
agents
or
machine
agents,
I'm
kind
of
getting
confused.
What
some
of
these
names
are.
It
was
first
called
machine,
adm,
node
adm.
Basically,
I
already
reviewed
these
proposals.
If
you're
interested
in
these
topics,
you
can
have
a.
D
A
D
Today
I
did
want
a
quick
touch
on,
like
you
know,
engagement.
I
do
know
that
certain
areas
people
have
said
they
wanted
more
engagement
and,
if
amazon's
interested
in
throwing
down,
let's
get
them
engaged
right.
So
if
you
have
on
ramping
people
or
you
want
to
get
them
working
on
certain
areas
like
if
there
are
maybe
folks
could
schedule
some
time
to
do
quick,
one-on-ones
to
identify
key
areas
of
interest
that
might
be
mutually
beneficial.
D
You
know
my
my
open
source
hat
always
says
when
you
can
match
greedy
self-interest
with
with
community
interests.
Those
are
the
best
engagements.
So
if
we
can
find
what
matters
most
to
amazon
folks
and
what
what
is
also
a
community
need,
that's
the
best.
A
Yeah,
basically,
what
I
do
is.
I
just
tell
people
to
send
me
a
message
on
slack.
If
they're
interested
working
in
different
areas,
I
can
then
potentially
delegate
either
to
one
of
the
projects
that
I
maintain
or
something
else.
A
I
have
seen
people
dm
me
like
hey.
How
can
I
help
and
I
do
respond
with
an
issue
like
I
always
give
some
sort
of
an
example,
first
good
ticket
that
you
can
work
on
so
yeah.
If
somebody
is
interested
just
send
me
a
message,
I
can
delegate.
A
A
It's
either
due
to
you
know
time
they
have
problems
with
their
arranging
their
time
or
just
the
complexity
of
the
issue
that
I
give
them
is
too
much.
I
don't
know.
B
I
know
people
on
our
team
were
really
interested
in
customer
api,
so
there's
a
be
some
people
following
up
on
that,
I
can
pass
along
the
note
about
trying
to
align
greedy
self-interest
with
open
source
interests.
I
think
that
would
that
sounds
like
a
message
I
can
sell
pretty
well
to
the
team.
Yeah
we're
our
team
itself
is
pretty
new,
so
we're
currently
trying
to
onboard
much
people
but
yeah
cluster
api
is
an
area
definitely
where
I've
been
passing
along
information
about
how
to
go
to
sydney.
B
So
I
think
that
there's
a
meeting
for
that
that
people
are
going
to
start
attending
yeah
is
there,
should
I
just
have
them
contact
this.
The
people
in
charge
of
that
meeting
is
that
the
best
way
for
them
to
get
involved
or
just
pick
up,
issues
that
are
open.
D
D
So
I
think
you
know
reaching
out
to
do
an
intro
on
the
slack
channels
is
always
helpful
as
well
as
you
know,
in
person,
because
you'll
find
that
some
people
who
can't
attend
these
meetings
were
are
very
active
contributors
over
time.
A
And
something
else
that
I
I
did
I
I
know
dims
knows
about
this-
is
that
we
signed
a
couple
of
projects
for
the
lfx,
which
is
a
cncf
mentoring
program.
A
We
I
basically
submitted
the
project
for
cuba
dm
there.
This
is
another
way
to
gather
new
contributors
to
this
group,
so
unfortunately
I
don't
think
we
submitted
for
anything
else,
but
I
know
that
the
costa
radons
and
cops
had
like
google
server
of
code
participation
just
another
drink.
F
We
did
it
last
year.
I
don't
think
we
put
in
anything
this
year.
To
my
knowledge,
I
could
be
wrong.
There.
A
Okay
yeah,
so
we
have
also
these
two
opportunities
to
google,
some
of
coding
aofx.
We
have
them
to
potentially
gather
new
contributors.
A
I
think
the
like
the
best
way
for
contributors
to
engage
is
just
invest
all
of
time
and
start
looking
at
the
code
base
and
the
open
issues
and-
and
the
mentorship
aspect
is
quite
involved
involved
from
the
maintainer
side,
but
we
can
definitely
occasionally
like
record
videos
and
provide
like
a
walkthrough
of
the
code
base.
I
know
the
question
appears,
I
have
done
it
cube
adam
has
done
it.
B
A
Gathering
more
people,
okay
in
terms
of
cost
repair,
if
you're
interested
just
start
joining
the
meeting
on
a
regular
basis,
comment
on
the
slack
channel.
If
you
have
any
questions
or
you
know,
comment
on
issues,
just
that's
the
you
know,
general.
A
B
Okay,
that
sounds
good
yeah
I'll
pass
that
information
on.
Did
you
even
stand
up
today
right.
C
If
there
isn't
anything
left,
I
could.
I
could
briefly
mention
something
which
could
be
of
interest
to
folks
here
once
one
story's
term.
C
So
one
of
the
one
of
the
challenges
that
I've
been
solving
at
I
surveillance
was
around
testing
and
particularly
testing
psyllium,
which
is
a
an
implementation
of
cni.
There's
demon
set,
there's
an
operator,
there's
all
sorts
of
stuff
and
we've
been
testing
that
for
quite
a
while
in
a
very
homegrown
sort
of
setup
that
hangs
together
mostly
with
bash
and
and
uses
some
virtualization
that
is
available
on
the
given
hosting
provider.
C
So
one
of
the
ideas
that
I
had
and
had
time
to
implement
was
to
create
an
operator
that
manages
ephemeral
clusters
specifically
for
testing
this
operator
was
made
open
source
just
just
before
end
of
last
year
or
early
in
this
year.
I
kind
of
forgot
I'll
post
a
link
here
and
if
I
can
share
a
screen,
let's
see,
do
you
think
share
screen.
A
Yeah
yeah,
just
a
second
I'm
going
to
have
to
stop
sharing
mine.
You
can
also
add
the
links
I
already
you
know,
creating
a
test
specific
operator.
C
Yeah,
so
it
is,
it
is
targeting
gke
at
the
moment,
but
we
are
kind
of
considering
implementations
with
other
providers.
It's
just
that
gke
was
the
easiest
thing
and
sharing
you
didn't
enable
yet,
okay,
I
can
oh
just
I
can
add
a
link
to
the
document.
C
Okay
cool,
so
I've
posted
the
link
to
the
repo
and
let
me
share
a
screen.
Okay,
it's
this
one.
I
just
just
wanted
to
show
you
the
the
part
of
the
readme
that
sort
of
describes
how
it
works
it
is.
It
is
very
much
cluster
api
esque,
except
that
it
doesn't
expose
all
the
underlying
gke
parameters
through
the
api.
C
The
idea
is
that
you
want
for
for
testing.
You
could
avoid
having
to
expose
all
different
parameters
and
have
a
a
small
set
of
standard
configurations,
so
you
could
actually
pull
pre-built
clusters
and
speed
up
your
tests.
That
way,
for
example,
right
so
the
the
api
at
present
looks
looks
like
this.
There
is
a
a
config
template
name
which
is
which
is
referred.
One
of
the
well-known
templates
that
is
built
built
into
the
operator
right
now
is
called
basic.
C
We
could
add
more
templates,
we
could
add
templates
at
runtime,
potentially
as
well,
and
then
you
specify
the
like
once
the
cluster
is
created,
the
job
is
launched
and
you
specify
that
just
just
like
this,
it's
it's.
Basically,
it's
not
the
full
pod
spec,
but
it's
just
an
image
and
command
to
run
inside
that
image,
and
then
there
is
the
node
type
and
number
of
nodes
and
the
gcp
project
name
and
location,
et
cetera,
yeah.
So
I've
posted
the
link
to
the
project
in
the
document.
C
If
anybody's
interested
hit
me
up
on
slack
or
raise
an
issue
in
the
project
or
or
contact
us,
otherwise
so
yeah.
So
there's
there's
a
fair
bit
of
text
here,
which
probably
could
be
turned
into
a
blog
post,
but
it
was
convenient
to
place
it
and
read
me
just
just
to
to
say
why
we
did
it
and
and
what
are
the
challenges
with
other
approaches
that
are
that
do
not
involve
an
operator
like
this
and
for
the.
C
I
guess
some
of
you
might
have
a
question.
Why
not
use
cluster
api
for
this?
I
think
it
could
possibly
be
used
for
this
quite
easily,
so
you
just
create
an
abstraction
on
top
of
cluster
api
effectively,
but
it
was
somewhat
easier
for
us
to
to
build
something
on
gke
and
we
are
using
config
connector
for
this,
which
is
which
is
kind
of
a
different
take
on
the
cluster
api
problem.
C
Nevertheless,
it
has
very
similar
goal
of
exposing
underlying
cloud
resources
to
kubernetes
apis
and
allowing
you
to
describe
new
clusters
through
the
kubernetes
api.
That's
kind
of
all
I
had
to
say
here
so
have
a
look
at
the
project
and
hit
me
up
if
you're
interested
it
was.
I
I
think
it's
it's
it
it'd
be.
It
may
be
useful
to
to
some
folks
testing
kubernetes
testing
operators
on
kubernetes
testing.
Kubernetes
add-ons
is
kind
of
tough.
C
There
are
solutions
in
the
kubernetes
project
upstream,
but
not
those
are
not
easily
adaptable
elsewhere.
Thanks.
F
Thanks
for
that
that
that
demo,
that
was
really
cool
or
overview,
I
just
wanted
to
quickly
address
config
connector.
I
don't
personally
see
config
connector
as
a
competitor
cluster
api
like
config
connector
is
providing
krm
or
kubernetes
abstractions
around
or
crds,
really
around
gcp
resources
and
something
like
cluster
api
could
be
built
on
top
of
it.
Obviously,
for
the
case
of
gke,
where
we
do
have
a
managed
cluster,
then
yes,
there
is
functionality.
F
That
looks
a
little
bit
like
cluster
api,
but
but
I
don't
see
that-
and
I
don't
see
them
as
like
comparable
in
in
the
general
sense
and
then
the
other
thing
is
like
I
think
other
clouds
have
similar
ones
like.
I
think
aws
has
one
which
used
to
be
called
the
service
operator,
and
maybe
someone
can
remind
me
what
it's
called
now,
but
you
know
that
sort
of
thing
so,
but
but
I
I
yeah
config
connector
is
slightly
different
from,
in
my
opinion,
from
cluster.
C
Yeah
there
is
a
microsoft
one
as
well
yeah
it
is,
it
is
quite
different,
but
it
sort
of
allows
you
to
like.
There
is
only
a
cluster
api
implementation
for
gke,
so
config
connect
is
closest.
That
comes
to
that.
D
Yeah,
you
know
if,
if
google
wants
to
participate
in
updating
the
gcp
provider
and
adding
on
things
very
similar
to
the
aws
provider,
you
know
we'd
be
happy
to
help
that
along
the
way
if
they
wanted
to,
because
the
you
know
the
the
kappa
provider
actually
supports
both
eks
and
native
deployments,
as
well.
As
you
know,
auto-scale
deployments.
So
it
has
all
incantation
support
with
aws
provider.
F
I
would
like
us
to
do
that
through
the
scalability
scale.
Scalability
was
talking
about
doing
something
like
this.
I
think
we
can
all
try
to
make
sure
that
they
don't
do
something
else.
How
about
that
like?
If
scalability
builds
something
we
should
encourage
them
to
not
build
a
brand
new
thing
like.
I
think
it
should
be
costa
rica
in
my
opinion,
and
then
it
would
probably
be
capg,
but
it
could
also
be
cafe
or
other
things
as
well.
F
Out
to
way
tech
directly,
you
can
encourage
voy
tech
to
do
this.
Yes,
but
I
don't
know
if
I
think
it's
sort
of
stalemated.
A
C
I
thought
I
thought
I
could
mention
it
yeah,
it's
just
you
know.
Those
of
you
who
are
interested
in
the
the
topic
may
wish
to
kind
of
just
at
least
look
at
the
readme
and
I'm
quite
keen
to
hear
what
what
others
think
these
are.
These
are
some
of
the
ideas
that
I
had
and
having
thought
of
the
problem
before
starting
to
work
on
that
problem
within
the
cilium
project.
C
A
Topic
cool
all
right.
We
can
end
earlier
today,
30
minutes
back.
Thank
you!
Everybody
and
see
you
again
in
couple
weeks.
Bye.