►
From YouTube: Kubernetes SIG Node 20230606
Description
SIG Node weekly meeting. Agenda and notes: https://docs.google.com/document/d/1Ne57gvidMEWXR70OxxnRkYquAoMpt56o75oZtg-OeBg/edit#heading=h.adoto8roitwq
GMT20230606-170550_Recording_1868x1336.mp4
A
B
B
B
A
D
E
G
A
B
So
the
next
one
is
memory
Qs
try
for
a
beta,
any
updates,
dikshita
and
David
no
updates
yet.
But
it's
a
work
in
progress
where
I'm
trying
to
think
how
to
add
the
E2
tests.
Okay,
so
I
know
there's
one
person
from
Red
Hat
Andrew
Collins.
We
can
CC
him,
so
he
may
be
interested
in
helping
out
with
testing
this.
He
is
involved
in
performance
and
skill
testing,
so
I
think
that'll
be
good
to
have
sure.
Or
could
you
please
share
them.
B
B
I
think
next
one
is
a
repeat:
we
already
got
that
up
there,
so
we
can
remove
this
one.
So
the
support
username
space
we
got
Sasha
Granite,
Red
Hat.
He
did
a
cap
update,
it
needs
review,
I
and
I.
Think
Rodrigo
is
gonna
be
back
soon.
So
I
think
this
is
on
track.
Giuseppe
is
going
to
work.
There's
a
cap
update.
A
A
B
B
A
A
B
A
B
H
B
B
E
A
A
D
B
I
B
C
A
G
G
G
A
A
A
Yeah
I
think
one
discussion
here
and
I
posted
on
the
secret
gesture.
Here
is,
if
you
do
CRI
metrics,
if
they
go
in
beta,
is
it
the
right
timing
to
have
C
advisor
endpoint
be
marked
as
duplicated?
So
it's
kind
of
it
may
not
be
it's
a
good
user
experience.
If
something
is
deprecated
when
there
is
no
GA
replacement
yet
other
than
that,
I
think
it's
it's
good
to
go
and
I
need
to
mark
this
beta
by
the
way.
J
Yeah
I'm
open
either
way.
My
my
thought
on
marking
it
as
deprecated
is
like
deprecation,
doesn't
necessarily
mean
removal
like
it'll,
still
be
working
as
expected,
but
it'll
give
users
warning
that,
like
we're
beginning
the
process
of
like
removing
it
so
like
one
day,
it
will
be
removed
and
I
think
tying
a
deprecation
to
like
post
beta
will
like
possibly
not
give
them
enough
times
or
like
enough
warning
that
we're
going
through
with
that.
J
But
I
I
am
personally
open.
I'm
fine
dropping
that
Clause.
If
that's
a
blocker
for
you,
but
my
thought
is,
you
know,
marking
it
as
deprecation.
We're
not
like
we're
not
committing
to
a
timeline
and
we're
not
like
we're
not
dropping
support
until
the
and
then
the
dropping
of
support
and
the
marketing
of
GA
Could
Happen
atomically.
E
A
K
I
K
A
K
Manages
the
GC
already
for
the
images
and
the
the
only
thing
that
used
to
happen
on
the
container
runtime
side
is
to
delete
snapshots
things
like
that
when
the
containers
and
pods
are
removed.
We
don't
do
any
garbage
collection
of
these
images
in
the
container
run
itself,
at
least
not
in
container
D.
Unless
it's
manual.
K
A
K
A
E
Do
we
need
to
support
this
one
because
we
do
have
constant
I
haven't
I
haven't
seen
the
copy
it
so
I
also
not
really
comments
by
the
use
cases.
I
mean
there
are
still
have
used
cases,
but
they
also
have
the
use
associated
with
use
cases.
There
are
many
of
the
risks
long
so
I,
don't
I,
haven't
heard
any
reply
about
the
how
we're
going
to
manage
those
risks.
E
There's
the
earlier
have
the
one
feature
called
the
drop
in
kubernetes
configuration
directory
right.
So
so,
if
people
could
just
say,
oh
I
can
Dynamic
the
in
that
one.
Kubernator
can
pick
up
Reserve
like
the
system
reserve
and
and
Cuba
reserve,
and
the
change
of
the
unlockable
resource.
I
think
that's
the
good
start
point.
Then
we
can
think
about
okay,
even
the
whole
Dynamics,
the
node,
the
memory
resizing,
because
today's
basically
record
of
the
node
object
right.
E
B
A
B
D
A
F
F
A
F
A
B
M
K
I
M
A
B
A
A
M
J
I
can
take
this
one,
so
this
this
is.
We
haven't
talked
about
this
one
yet,
but
basically,
currently
cryo
has
some
special
behavior
to
do
to
clean
up
images
between
upgrades,
because
you
know
in
situations
where
there
are
automated
upgrades,
there
are
stale
images
that
are
left
over
and
even
though
the
cubelet
garbage
collection,
the
current
scheme
will
kick
in.
You
know
once
the
disk
hits
85
or
whatever
they
set
it
to
that,
isn't
necessarily
that
it's
like.
J
Sometimes
it
could
feel
kind
of
weird
to
have
all
the
like
the
images
laying
around
and
you
know,
populating
the
node.
So
so
in
some
installations
we
have
a
special
a
scheme
in
cryo,
where
we,
like,
you
know,
on
an
upgrade
we'll
remove
the
sale
images,
but
we
want
to
I
wanted
to
see
what
the
appetite
in
upstream
was
for
like
putting
this
Behavior
into
the
qubit
instead
of
into
the
CRI,
and
because
you
know
it's
currently
the
owner
of
the
garbage
collection.
J
If,
if
there
isn't
any
interest
in
that,
then,
like
you
know,
we
can
continue
to
do
it
in
crowd.
That's
fine,
but
I
wanted
to
see.
You
know
like
I'm,
not
really
thinking
that
it's
time
it
seems
like
we
have
a
lot
going
on
in
128,
so
I
don't
think
that
it
at
this
point
I'm
not
pushing
it
for
it
really
hard
in
128,
but,
like
you
know,
just
long
term
wanted
to
see
what
people's
thoughts
on
this
was.
E
I'm
really
interesting
to
know
more
about
the
letter
functionality
you
implement
in
the
quail,
because
today
is
the
kubernetes
image
garbage
collection.
Honestly,
we
we
just
the
first
step
for
the
1.0
kubernetes,
so
we
threw
something
there
not
to
really
wear
sort.
We
I
want
to
enhance
that
garbage
collector
so
many
times,
which
is
level
prioritized
so
I'm,
looking
forward
to
see
yeah.
B
To
give
us
a
summary
done
like
when
we
move
say
from
126
to
127,
we
know
that
all
our
images
are
changing
right
yeah.
So
this
is
the
optimization.
Why
waste
space
we
remove
all
the
all
images,
because
all
of
them
are
going
to
be
re-pooled.
So
that's
what
we
have
in
cryo,
so
yeah,
so
we're
taking
aggressive
on
GCE.
Since
we
know
nothing
is
going
to
be
used.
J
I
I
think
what
what
my
kind
of
Ideal
is.
Long
term,
though,
is
a
more
periodic.
Like
you
know,
my
idea
is
like
having
like
a
least
recently
used
cash
that
we
saved.
You
know
what
containers
are
being
used
and
then,
like
you
know,
every
I,
don't
know
what
period
of
time
that
we
want
to
run
it
on.
But,
like
you
know,
every
week,
every
two
weeks
or
something
cubelet
goes
through
and
cleans
up
any
containers
that
haven't
been
used
or
any
images
that
haven't
been
used
in
a
container
for
that
amount
of
time.
J
So
that
that's
another
idea
that
I've
kind
of
been
mulling
over
which
is
less
of
a
heavy
Hammer
than
just
like,
remove
everything
on
the
upgrade
and
would
require
the
Cuba
to
keep
less
State
between
reboots,
but
it
would
require
it
to
keep
State.
You
know,
while
it's
running
in
case
it's
a
really
long
running
cubelet,
so
that
that
was
one
idea
that
I
had.
J
But
you
know,
I
imagining
a
world
in
which
we
had
different
garbage
collection
schemes
and
like
supporting
that
I
think
you
know
it
sounds
like
that's
there.
There
has
been
an
appetite
for
that,
so
it
John
it
sounds
like.
Then
you
are
interested
in
this,
so
you
know
we
can
start
imagining
what
that
could
look
like
and
maybe
aim
for
a
129
at
this
point.
Unless
people
are
really
excited
about
it,
I
think
implementation
side.
E
This
is
definitely
it's
not
urgent,
so
because
our
current
garbage
collection
definitely
is
not
a
sophisticated
at
all.
We
we
understand
that
we
talked
about
this
before,
but
we
departures,
so
we
don't
need
to
rush
and
but
I
do
think
about
it.
I
also
not
tied
with
those
garbage
collect
based
on
upgrade
I,
totally
agree
with
you,
and
we
should
solve
this.
It
is
for
long
term
right
so
like
based
on
whatever
policy.
Currently
we
don't
have
those
clear
policy
defined.
E
So
a
lot
of
time
and
I'm
happy
for
the
performance
for
customer
like
the
startup
is
sometimes
it's
a
little
bit
difficult
to
predict
and
also
sometimes
we
prevent
or
remove
after
some
of
the
energy
to
earlier
and
hurt
their
performance.
So
that's
all
kind
of
problems
with
a
wireless
problems.
We
just
couldn't
fix
it,
but,
on
the
other
hand,
to
sound
there's
the
to
solve
something
like
this
other
app
issues.
So
a
lot
of
food
vendor
may
be
pre-loaded
some
email
to
solve
this
problem.
E
I
also
concerned,
like
the
garbage
collector,
maybe
make
those
systems
work.
So
that's
why
we
need
to
build
some
next
holistic
policy
and
yeah,
but
we
don't
need
to
rush
on
those
things.
Just
since
you
ask
is
that
interest
I
just
want
to
share
in
the
past.
Actually
we
have
a
lot
of
discussing
and
we
do
have
interest.
We
know,
there's
the
bigger
space
to
improve
yeah.
J
B
A
A
B
D
J
A
Next,
one
I
want
to
highlight
this
change
happening
in
this
PR,
so
PR
is
about
recursified
only,
and
the
interesting
part
here
is
Introduction
of
this
field.
All
like
API
for
runtime
render
features
I
wanted
to
highlight
it
here,
because
it's
kind
of
hidden
Beyond
like
hidden
in
this
enhancements.
It's
a
proposal
to
have
discoverability
of
new
features
in
runtime
I
would
love
more
people
to
comment
on
this
and
express
opinions
like
how?
A
If
there
is
something
not
working
today
in
how
we
discover
features
so
I
could
have
some
ideas.
What
else
needs
to
be
discovered
through
this
API?
Please
comment
here
so
I
mean
this
VR
like
I.
Don't
want
cap
to
silently
introduce
a
new
pattern
that
we
wouldn't
work
for
other
caps,
so
I
want
to
highlight
that.
H
B
F
About
yes,
yes,
so
it
I
think
it
will
be
quick.
It's
a
question,
it's
a
question
for
us
because
I
don't
have
any
knowledge.
So
basically
there
is
this
user
asking
about
an
API
and
what
API
should
return,
and
that
prompts
me
to
ask-
and
I
already
did
some
of
you
so
bear
with
me:
okay,
so
those
endpoints
those
cubital
points,
third-party
components
which
are
not
part
of
kubernetes.
They
can
query
them
and
which
kind
of
guarantees
apis
they
will
get
because
you
know
the
public
facing
API
has
guarantees
and
compatibility
and
everything.
F
But
you
know
in
order
to
either
to
answer
to
this
very
question.
I
would
like
to
clarify
first
myself
and
then
maybe
for
everyone,
which
are
those
guarantees.
I
mean
what
the
user
should
legitimately
expect
by
that,
and
this
is
why
I'm
asking
this
forum,
because
I
don't
know
what
how
to
find
this
information.
So
if
this
document
somewhere
I'm
happy
to
to
read
Vibe
and
do
my
homework
thanks.
C
E
K
I
K
E
L
E
Duplicate
that
API
and
how
to
register
also
authenticate
who
can
view
that
API
so
there's
the
security
concept.
So
we
have
a
really
enhance
those,
and
so
that's
kind
of
the.
So
we
ended
up
kind
of
a
little
bit:
Nimble
State,
it's
kind
of
internal
API
and
for
for
debug
being
it
yeah,
the
back
being
it
here
and
also
for
the
internal
component
to
consume.
We
find
another
way,
so
we
we
didn't
provide
any
guarantee.
F
E
F
Api
is
different
because
the
well
it's
also
a
bit
different
because
it's
stable
and
it
has
documented,
could
be
documented,
probably
better
and
but
okay,
let's,
let's
bring
that's
a
very
fair
question
and
I
think
we
in
that
case
we
have
and
we
want
to
have
stronger
guarantees,
and
we
totally
should
keep
this
conversation
going,
probably
offline.
But
yes,
we
should
keep
this
conversation.
Yes,
thanks.
Thanks
for
knocking
it
down.
F
A
Because
I
understand
that
it
may
be
affected
by
the
same
problem
when
Imports
are
about
to
be
gone
or
Google
just
restarted
and
I'm
not
sure
what
kind
of
guarantees
this
API
will
have
like.
Will
we
return
404
on
right
at
the
Kubota
started
or
he'll
return,
something
and
for
a
foreign
not
like
generally.
F
Yeah
right,
so
it's
like
not
forever
I
I,
I
I,
see
your
point.
I
agree
with
you
I
think
that
the
the
comment
I
think
I
can
make.
Now
is
that
the
fields
and
the
let's
say
the
syntax
and
the
the
contract
of
the
API?
That's
regulated,
that's
part
of
GAA
process.
The
semantics
could
could
use
some
clarification.
I'm,
not
sure
how
to
express
that.
But
you're
totally
have
a
point
here.
A
A
H
A
E
Patterns
I
had
to
look
at
those
CCS,
so
sometimes
it's
those
issues
caused
by
enemies,
infrastructure,
sometimes,
and
so,
but
there's
no
signal
for
people
that
I
don't
have
access
to
the
end
of
this
infrastructure
to
help
with
tragedy.
I
think
that
pack
holds
the
earlier
question
he
he
mentioned
that
he
even
don't
know
how
to
help
he
tried
to
help.
So
that's
might
be
the
things
that
maybe
we
can
see.
Okay,
I,
don't
know.
E
Honestly,
we
didn't
solve
that
problem
back
then,
and
a
lot
of
times
it's
just
people
have
access
to
the
Anthony's
infrastructure,
have
to
stay
on
top
and
give
the
signal,
and
so
that's
why
I
filter
some
issues
for
people
or
maybe
there's
the
automated
way
to
Signal
those
things
so
used
to
be
test
infrastructure
actually
give
some
signals.
So
then
we
can,
we
know
so
I
don't
know.
I
just
saw
some.
Some
idea
here
is
that
he
just
reminded
me
what
he
said.
This
reminded
me:
it's
just
harder
for
people
to
help.
I
A
D
A
H
A
Okay,
what's
next,
oh
I
forgot
this
topic.
So
the
reason
all
our
meetings
in
CI
group
was
like
Ci
group
at
this
music
and
sidecars
was
created
on
personal
calendars
and
I
wanted
to
get
the
shared
calendar
going
spare
instructions
for
all
the
six.
So
I
created
this
shared
calendar
and
I
suggest
we
move
this
meeting
in
the
shared
calendar.