►
From YouTube: Kubernetes SIG Node 20220614
Description
SIG Node weekly meeting. Agenda and notes: https://docs.google.com/document/d/1Ne57gvidMEWXR70OxxnRkYquAoMpt56o75oZtg-OeBg/edit#heading=h.adoto8roitwq
GMT20220614-170502_Recording_640x360
A
Good
morning,
everyone-
and
today
is
the
June
14th
signaled
weekly
meeting,
and
we
have
a
lot
of
our
full
updates
here,
and
the
first
updates
actually
came
from
let's
just
since
the
circuit
is
here,
needs
to
follow
our
regular
routine.
So
secret.
Do
you
want
the
update
of
the
active
pull
requests?
First,
thanks.
B
Yeah
we've
been
stable
on
our
pull
requests
and
I
I
looked
at
closed
PRS.
One
of
the
PRS
that
deserves
to
be
mentioned
is
like
work
in
progress
PR
like
given
its
work
in
progress,
but
it
was
adding
the
test
and
as
many
PR's
heading
the
test,
it
was
neglected
for
reviews
and
I
think
this
deserved
to
be
looked
at,
and
maybe
somebody
can
pick
up
or
think
whoever
author
was
maybe
Parker
I.
Don't
remember?
Oh
no,
so
yeah.
B
If
you
want
to
pick
it
up,
please
do
other
PRS
that
were
closed
were
closed
for
reason
and
it's
good
and
mostly
honest.
We
have
medical
like
flow
PR's
been
looked
at
and
merged,
so
yeah.
Nothing
very
specific
to
talk
about.
Many
actually
picks
this
time.
A
Okay,
thanks
Sergey
and
the
next
topic
is
from
me
I
just
want.
This
is
for
your
information
and
last
couple
days,
I
just
update
our
couple:
the
1.25.
The
signal
needs
to
have
the
carpool
monitoring
tracking
shift.
So
I
thanks
for
everyone's
efforts.
Last
a
couple
of
weeks,
so
we
finalized
the
cap
for
the
1.25
file
for
the
signal
so
I
convert
all
those
we
already
covered
like
the
wheel.
Most
of
the
card
was
included
that
it
is.
We
already
agree,
the
community
agree.
A
We
are
going
to
track
for
1.25,
so
we
express
it
identify
the
owner
and
also
owner
A
promise
is
going
to
actively
working
on
and
doing
this
release
cycle
and
also
we
most
of
the
pr
as
you,
except
one
and
already
clearly
identified
the
reviewer
and
also
a
poor.
The
only
one
I
just
I
did
for
the
Disco
quarter
and
promoted
to
Beta.
A
So
the
just
inform
us
to
yesterday
I
believe
and
I
just
included
into
the
track,
and
we
don't
identify
the
reviewer
and
the
Brewer
for
that
one
year
also,
I
asked
for
because
paiko
working
on
that
one
and
I
if
I
remember
correctly.
Last
time
we
asked
for
either
use
and
to
and
the
test
and
The
Benchmark
everything,
but
we
have
to
see
the
data
shared
to
the
community
share
back
to
the
community.
Yes,
so
this
is
why
we
might
need
to
discuss.
A
Do
we
want
promoted
to
next
phase
here,
but
at
least
I
include
that
for
the
tracking
we
also
totally
have
the
19
enhancement
proposed
for
1.25,
and
we
are
not
required
right
so
like
what
we
are
talking
about,
the
next
That's
the
menu
feature.
We
still
end
the
discussing
and
pretty
big.
So
it's
not
everyone.
Every
all
cap
is
guaranteed
will
be
promoted
like,
for
example,
the
new,
maybe
Google,
update
like
the
sticker
version
2.
We
try
to
make
that
yeah,
but
we're
based
on
the
user
feedback.
A
So
so
this
is
just
quick
updates
from
from
the
from
our
planning
progress.
C
B
Something
or
he's
just
Club
I
was
just
clapping
thanks
for
following
this
together.
Okay
and
I
just
wanted
to
make
sure
that,
like
everybody
understands
that,
we
need
to
keep
working
on
this
PR's
to
merge
them
by
enhancement,
fees
and
enhancement.
Feast
was
postponed
for
one
week.
So
it's
not
this
week.
It's
next
week.
D
A
A
I
believe
we
have
at
least
the
more
than
30
percent
of
the
cup
is
not
merged
yet
so
we
need
to
work
out
also
yeah.
A
Okay,
next
one
I
just
mentioned
that
that
this
quarter
so
I
include
it
in
the
cup
and
also
we
asked
the
view
packed
report
back,
but
because
I'm,
not
the
previous
PR
review
for
this
feature
so
I'm
wondering
because
he
did
indicate
the
pack
called
did
indicate.
The
Benchmark
is
ideal.
The
Entity
and
the
test.
I
do
say
the
end
test,
but
I
haven't
seen
The
Benchmark.
Yet
anyone
in
the
community
saw
the
Benchmark
demo
or
anything
for
this
feature.
F
I
haven't
seen
anything
done,
but
I
can
I
can
take
a
look
at
that.
We
are.
A
Is
the
feature
will
pick
up?
That's
already
Alpha
feature
he's
the
new
people
new
owner,
and
so
this
is
where
from
Robert
he
pick
up.
This
feature
is
being
in
there
for
a
long
time.
So
so
this
is
why
we
asked
in
the
past
I
do
say
that
in
the
test,
but
I
didn't
check
the
Benchmark
and
many
other
things
and
the
monitoring
Matrix
other
things.
So
that's
why
I,
maybe
we
need
to
fall
out
anyway?
I
did
ask
Paco
to
report
back.
A
Give
us
the
demo,
maybe
I
make
make
that
more
explicit
and
they
give
us
a
demo
and
what
enhancement
and
the
monitoring,
and
also
the
also
The
Benchmark
he
can
come.
I
know
there's
the
time
difference,
but
hopefully
maybe
can
record
or
maybe,
if
we
couldn't
in
person,
join
the
meeting.
C
So
quick
question
about
the
Benchmark
is
that
to
ensure
there's
no
performance
degradation
around
monitoring
the
the
the
quotas
themselves,
around
storage
was
that
the
Benchmark
concern.
A
Or
The
Benchmark,
if
I
recall,
but
it's
been
a
long
time
back,
but
if
I
recall
Nicole,
because
there's
a
couple
things
one
is
like
the
immediately.
We
definitely
want
is
because
current
we
are
doing
the
track
of
the
disk
space
usage
is
really
High
rate
right,
so
using
a
lot
of
CPU.
So
we
wanted
to
see
The,
Benchmark
and
add
this
feature.
A
The
result
is
the
seminar
within
you,
because
the
disk
space
tracking
actress
level
is
100
accurate
address,
so
we
do
require,
but
the
result
is
a
similar,
but
at
the
same
time
and
there's
the
reviews
of
the
CPU
usage,
so
that's
kind
of
see.
This
is
a
huge
Improvement
and
it's
just
and
also
we
want
to
see
some
under
two
in
the
test.
A
I
do
say
end
to
end
the
test,
PR
flying
and
the
back
and
forth,
but
we
didn't
really
look
at
the
detail
about
the
test
cases,
the
bar
the
two
things.
Obviously
we
want
this
feature
and
this
is
definitely
efficiency
right
improved,
but
on
the
other
hand,
we
worry
about
improved
in
introduce
of
the
integers
of
the
regression.
So
this
is
why
we
want
to
cover
both
cases
and
then
we
can
promote
it
better.
A
G
Yeah,
so
the
feature
doesn't
in
the
alpha
form
cover
all
ephemeral
storage
usage
that
a
pod
can
make
so
I
just
quickly.
Looked
at
the
enhancement
LinkedIn,
the
meeting
notes
I,
don't
see
how
it's
made
changes
to
close.
G
Was
there
like
an
audit
done
that
identified
that
Gap
just
so
we're
on
the
same
page
like
the
stuff
that
a
female
storage
is
accounted
against
right
now,
which
was
the
writable
layer
of
the
container
empty
their
usage
and
not
even
myself,
I'm
getting
confused
right
now,
the
the
ephemeral
quota
tracking
didn't
cover
basically
all
four
or
five
of
those
entities
in
the
alpha
form,
and
only
did
one
of
those
and
so
I
don't
know
if
anything
was
done
to
close
that
Gap
or
not
am
I
am
I
missing
something
or
was
something
done.
A
So,
thank
you.
You
didn't
miss
anything.
This
is
why
I
feel
because
this
one
don't
have
much
of
the
visibility
to
communicate
here.
Yeah.
G
A
Then,
okay,
this
way
I
feel
like
this
one
in
the
past.
We
might
not
track
closely.
We
do
want
this
enhancement
in
the
long
run,
but
the
the
feature
itself
I,
don't
see
the
enhance
too
much
and
I
want
the
the
owner
new
owner
report
back
to
the
signaled
Community.
Then
we
can
help
to
make
sure
decided.
This
is
a
better
quality
or
not
we're
not
deciding.
So
you,
if
you
see
the
back
comment
where
we
need
to
discuss
as
a
signal
of
the
community
promotes
because
we
don't
see
what's
ready.
A
A
G
So
I
had
reviewed
the
pr
through
some
comments
on
there.
So
a
lot
of
the
the
prior
updates,
I
requested
from
today
on
this
feature
were
made.
So
that
was
good,
but
there
was
still
a
a
kind
of
fundamental
concern.
I
had,
and
maybe
we
could
talk
about
it
in
this
audience,
which
was
right
now.
G
So
one
of
the
things
on
the
pr
and
I
don't
know
if
we
got
all
the
answers
yet,
but
maybe
Renault
or
Peter
or
Mike
Brown
or
anybody
else
who
works
in
the
runtime
tier
could
help
with
those
I
strongly
suspect
right
now
that
if
you
change
the
memory
limit,
the
qubit
will
believe
the
change
happened
when
it
never
actually
happened,
because
the
error
is
going
to
be
silently
swallowed
and
so
that
that
was
one
thing.
G
I
F
I
just
tried
it
out
and
I
pasted
my
findings,
so
basically
anything
anytime.
We
try
to
lower.
It
will
always
get
an
error
and
whenever
we
try
to
raise
it,
oh
only
time
it
gets
silently
ignored.
It
feels
like
it's
ignored
because
it's
only
allowing
4K
boundaries.
So
if
you're
trying
to
write
a
value
which
is
within
a
4K
Bounty,
it
gets
rounded
down.
But
if
we
do
it
in
increments
of
4K
boundaries,
then
it
should
be
respected.
F
This
is
on
V1,
I
haven't
tested
on
V2,
yet,
but
I
think
directly
should
be
okay
on
V1.
As
long
as
our
updates
are
at
4K
boundaries.
C
F
So
the
kernel
is
doing
the
rounding
down
right
now
and
we
are
not
I
mean
we
are
just
trying
to
write
the
value
down.
We
can
change
there,
but
I'm
not
seeing
why
I
mean
at
the
cubelet
level,
I
think
it's
always
that
MBS
or
do
we
also
support
KB?
F
I
G
Yeah
and
then
so
that
was
the
one
thing
just
getting
understanding
on.
Do
we
actually
think
update
container
resources
does
what
we
think
it
does,
because
I
did
feel
it
were
edge
cases
and
then,
in
the
scenarios
where
memory-backed
empty
doors
were
being
used,
I
do
not
believe
the
cubelet,
and
that
was
another
thing
I
needed
to
check,
but
I
ran
out
of
time.
G
This
morning
actually
reads
back
the
value
that
it
wrote
to
ensure
that
that
write
happened
there,
which
would
need
to
be
done,
independent
of
the
CRI
tier
so
and
then
there's
an
implication
in
the
pr.
If
I
recall,
that
is
expecting
that
we
get
like
a
plug
notification
when
these
changes
get
requested,
and
so
I
was
a
little
confused
looking
at
it
again
today
and
anyway,
we
can
catch
up
with
vinay
separately,
if
he's
not
here
today,
but
if
he
catches
the
recording.
Maybe
we
can
sync
up.
F
Yeah
I'm
not
sure
of
that
one
either
like
if
we
get
an
error
or
if
we
can
align
to
4K
I'm,
not
sure
why
we
would
need
that.
But
but
I'll.
I
G
It
was
more
to
like
trigger
a
change
in
the
internal
cubelet
State
machine,
okay,
so
yeah
just
something
to
work
through
but
yeah.
So.
C
G
C
Hey
yeah,
this
is
deep
again,
so
this
is
just
a
quick
bump
on
the
cap.
I
was
wondering
just
kind
of
bumping
it
on
direct
radar,
I
guess
and
seems,
like
Bruno,
got
a
chance
to
take
a
quick
review.
So
if
there
are
any
major
comments
or
feedback,
please
let
me
know.
C
It
also
seemed
the
prr
reviewer
for
this
cap
God
reset,
because
Alina
did
that
review
last
time.
That
was
a
bit
of
a
surprise,
so
maybe
just
a
announcement
for
the
community
like
if
you
had
a
PR
review
from
Ilana
I
guess
in
124,
but
the
cap
didn't
make
it.
It
seems
like
it's
resetting,
based
on
the
approvers
list.
C
Getting
changed
so
just
keep
an
eye
on
that,
because
when
I
did
some
updates,
it
seemed
it
kind
of
said,
like
your
prr
is
no
longer
approved,
so
I
had
to
like
re-request
that
prr
approval.
So
just
a
heads
up
in
case
other
gaps
have
that
happening
as
well.
K
Yeah
I
just
trying
to
bump
up
this
cap
here
so
I
basically
updated
the
ga
criteria
for
the
club,
lift
credential
provider
caps.
This
is
the
only
thing
I
mean
promoting
the
ga.
It's
the
only
thing
left
for
this
cap
so
trying
to
get
some
well
approval
from
signal
to
approvers
I
already
got
it
approved
by
Sergey
and
sorry
death
2K.
He
is
the
prr
yeah
I
got
through
my
Pros
already
so.
L
A
Yeah
I
didn't
approve
on
this
one.
It
is
because
I
do
want
to
bring
this
to
the
community,
so
it
looks
like
the
the
ga,
the
only
gear
criteria.
It
is
just
under
two
and
the
test
right.
K
Yeah
so
I
bring
this
here,
also
just
trying
to
see.
If
there's
any
other
kind
of
criteria,
we
want
to
add
I,
don't
think
we
have
any
user
feedbacks
yet,
and
that
is
the
reason
why
I
didn't
add
that
to
the
ga
criteria.
K
C
K
Just
we
don't
have
to
block
on
this
on
this
meeting,
we
can
move
the
conversation
offline
to
the
cap
itself.
Just
please
take
a
look
and
if
there's
any
other
thing,
we
need
to
do
or
we
need
to
check
before
promoting
the
ga
just
put
a
comment
there.
G
F
K
K
B
Is
a
vlog
actually
disables
in
three
credential
providers,
completely
I
mean
if
it
disables
them
completely.
Is
that
it's
as
good
as
removing
them.
K
Yeah
this
one
is
I,
think
it's
behind
a
future
gate
or
flag.
G
I
was
trying
to
think
about
how
you
could
get
feedback
and
like
if
the
existing
things
not
being
removed,
then
I
don't
know
how
you
get
feedback
on
the
new
thing
and
I
I
don't
know
who's
likely
to
give
you
feedback.
I
actually
don't
know
who
the
consumers
were
of
this
feature.
Like
I,
don't
know
Donner
Sergey
was
this
used
to
help
you
integrate
with
GCR,
or
is
anyone
from
Azure
to
help
like
I.
G
On
we
weren't,
obviously
using
this
right
now.
B
If
you
do
use
credential
provider
right
now,
it's
in
memory
and
with
this
in
external,
so
we
will
be
adopting
that
so
from
use
perspective,
it
will
be
used,
but
I
think
the
biggest
consumers
are
everybody
who
will
not
carry
these
dependencies
so
I
mean
to
some
extent
it's
it's
already
used.
G
Yeah,
that's
what
I
was
trying
to
figure
out
as
if
anybody
would
miss
the
entry
thing
being
gone
or
what
would
what
would
drive
signal
that
the
new
thing
was
being
used.
B
Ken,
it's
a
question
not
only
for
vendors
who,
like
there
are
three
vendors
right
now:
have
entry
credential
providers,
but
also
for
self-hosted
customers
who
just
use
Intrigue
credential
provider
on
corresponding
clouds,
but
holds
them
themselves.
So
it's
also
might
also
be
affected.
K
D
K
Do
we
want
to
like
first
to
promote
this
feature
to
GA,
and
only
after
that
remove
the
entry
code,
because
I
might
I
have
some
concerns
that
some
users
might
still
need
this
entry
provider.
K
But
yeah
anyway,
like
I,
said
we
can
move
this
conversation
to
the
to
the
cab
yeah.
This
one
is
currently
assigned
to
director.
If
you
have
time
just
please
take
a
look.
G
A
You
can
be
resign
to
me,
I
guess,
that's
the
fact
that
yeah,
unfortunately
I
also
don't
have
the
clear
data,
but
at
least
from
our
product
will
be
the
for
our
offer.
We
need
to
concern
about
yeah,
that's
the
concept.
Yeah,
okay,.
I
I
Up
mute
first
find
the
window
so
the
when
you're
doing
an
exact
you
know
into
your
container,
currently
you're
rude
and
that's
it,
but
users
of
Docker
are
used
to
being
able
to
set
their
users
right
on,
or
at
least
default
to,
the
user
of
the
container
that
was
defined
when
they
they
created
it.
But
right
now
it
we
don't
really
work
that
way
and
I
think
you
know.
Traditionally
that's
been
because
you
know
kubernetes
doesn't
see
this,
as
you
know,
a
space
that
people
really
want
to
do.
I
But
there
are
some
users
of
you
know,
pods
that
where
they
want
they're
expecting
to
be
able
to
set
the
user
so
that
it's
not
always
root,
or
maybe
they
don't
even
want
root
to
be
able
to.
You
know
exactly
into
the
Container
without
a
password,
for
example,
so
I
think
there's
some
use
cases
we
need
to
consider
here
and
and
I
was
gonna
I
just
want
to
bring
this
up,
because
people
are
opening
up
issues
against
the
container
runtime
saying
why?
Where
did
my
my
exact
go
right?
Why
can't
I?
I
Why
am
I
getting
a
different
Behavior
than
I'm
expecting
from
the
the
prior
implementation?
When
you
had,
you
know,
Dr,
Shim
and
I
could
exact
from
Docker,
for
example.
So
I
don't
know
what
you
guys
think
about
this.
If
anybody's
interested
in
you
know
doing
some
work
on,
this
is
something
we
should
pursue.
It
was
talked
about.
You
know
many
years
back
and
and
sort
of
languished
there,
but
I.
Don't
think,
there's
anything
insurmountable
about
doing
this,
but.
F
F
I
think
from
the
runtime
side,
it's
pretty
straightforward
to
wire
up
to
run
CE
or
the
runtimes
I
think
the
the
UI
may
be
a
bit
contentious
like
okay.
Are
we
exposing
too
many
knobs
like
what
knobs
make
sense?
Like
you
said,
do
you
match
the
container
user,
or
do
we
allow
a
knob
to
specifically
set
a
user
or
a
uid
I?
Think
that's
that's
a
tricky
part
here,
rather
than
the
actual
Plumbing
there
you
go.
I
You
also
have
you
know:
SSH,
you
don't
really
need
to
use.
Exec
I
think
the
issue
is
more,
we
have
exec
and
it
doesn't
work.
You
know
expected.
There's.
G
No
there's
no
option.
Maybe
one
thing
would
be
interesting
to
pull
like
I
know
for
our
users,
some
of
our
more
security
conscious
users
do
disable
usage
of
exec
entirely
yeah
and
I'm,
not
sure.
If
giving
them
more,
flexibility
would
promote
more.
I
We're
going
to
use
their
namespaces
and
what
does
it
mean
to
be
able
to
support
exec
when
you're
isolated
to
username
space
anyway
right,
so
it
probably
needs
some
work
right
and
we
probably
need
some
work
in
this
area
just
to
consider
what
it
you
know
when
is
it
viable?
When
is
it
value
it
and
is
root
documented
a
little
bit
explain
why
there's
a
difference.
Behavior
perhaps
provide
an
option
but
I'm
not
sure
I'm
bringing
it
forward,
but
I'm,
not
sure
either
I
I
don't
ligit.
I
A
Sex
yeah
thanks
mate,
Mac
and
also
thanks
the
direct
since
you
also
bring
up
the
informal
container
and
I
prove
the
promote
the
that
feature
to
the
ga
for
1.25.
But
during
that
time,
I
realized
I,
hardly
received
some
user
feedbacks
I
even
reach
out
I
believe
I
reach
out
also
menu
to
see
the
match
of
the
user,
because
I
do
heard
at
the
earlier
State
there's
some
user
inputs
from
the
openshift.
But
now
after
that,
I
haven't
heard
any
new
feedback.
A
Yet
so
so
I
reached
that
in
the
tab
originally
and
also
asked
me
the
original
owner
and
he
agreed
to
promote
that
to
the
ga
and
we
are
actively
working,
but
he
also
don't
receive
much
of
the
user
inputs
so
far.
F
G
Yeah
catch
up.
I
A
B
L
Hi
everyone,
so
it's
a
request
for
a
view
of
dynamic
resource
allocation
cap.
The
cap
was
reviewed
once
and
we
were
like
working
on
it
and
trying
to
address
all
the
comments
and
at
the
moment
the
second
review
round
started
and
team
is
reviewing
API
changes
and
we
would
like
to
ask
Derek
to
to
look
at
the
cap
again.
If,
if
he
has
time,
that's
that's
it.
L
Well,
not
not
just
dag
but
like
anyone
who
is
interested
in
in
like
in
this
area,
but
Derek
was
actually
reviewing
it
like
first
time
so
that
that's
why
it's.
G
B
G
To
pick
between
this
and
the
the
other
kept
right
afterwards,
but
is
there
a
priority
like
I,
feel
like
there's
a
lot
of
issues
in
Flight,
Around,
the
resource
model
and
I'm,
not
getting
a
sense
of
like
what
we
collectively
do
as
a
as
a
priority
here.
L
J
I
I
would
say
it's
it's
on
the
same
level,
because
it's
completely
too
independent
features
and
to
in
independent
areas
what
those
statues
so
like
with
Dynamic
resource
allocation,
it's
a
big
one
and
like
it
touches
a
lot
of
the
things
class
based
resources,
it's
small
change
on
CRI
protocol
level.
So
it's
not
it's
not
yet
on
board
spec,
it's
not
anywhere
else.
It's
it's!
Just
like
the
communication
between
the
couplet
and
runtimes.
At
this
moment,
yeah.
C
G
G
Yeah,
so
that's
what
I'm
saying
like
I'm,
a
mere
mortal
who
sometimes
can
get
confused
myself
like
everybody
else,
and
they
seem
overlapping
and
I,
just
want
to
get
a
sense
of
like
those
on
the
call.
If
there
was
a
priority,
it's.
A
The
park,
you
just
repeat
exactly
what
I
asked
a
couple
weeks
ago,
I
asked
the
exactly
the
same
thing:
I
confused
the
during
the
planning,
I
said:
okay,
I
still
think
well,
there's
the
some
level
are
for
at
least
the
problem
which
had
to
solve
there's
some
level
of
the
overlap,
so
even
I
get
to
the
answer,
but
sorry
about
it.
I
haven't
looked
into
the
detail
yet,
but
I
I
understand
we
try
to
make
the
cluster
resource
Capital
card.
A
One
is
the
smaller
scope
and
but
still
I
feel
like
the
we
try
to
attack
the
same
set
of
problems.
H
J
I
I
I'll
I'll,
try
to
probably
explain
the
difference
between
rows.
Two
again
might
be
just
for
name
or
naming
of
things
is
not
properly
correct,
so
Dynamic
resource
allocation
cap,
it's
pretty
much
revisit
how
we
are
working
with
devices,
so
it's
countable,
it
might
be
local
might
be
remote,
it
has
all
internal
properties
and
so
on
and
so
forth.
So
that's
one
scope,
class-based
resources,
it's
something
what
we
cannot
count,
but
multiple
ports
can
share
the
same
resource.
J
So
it's
like
a
good
example
is
blocker,
so
every
node
in
the
cluster
we
have
different
priorities,
different
description
or
different
disks
and
so
on
and
so
forth.
But
you
can
just
say
this
is
my
like
gold
level,
silver
level,
bronze
level,
and
you
can
say
this
uses
gold
storage,
which
uses
for
silver.
H
G
D
J
So
yeah
yeah,
oh
yeah,
so
we
were
saying
the
same
goes
to
a
memory
bandage
so
where
Hardware
doesn't
allow
you
to
to
specify
for
each
process
like
how
much
throttling
you
can
do
like
like
in
network
process
for
each
interface,
you
can
do
trottling
in
memory.
You
cannot
do.
The
hardware
allows
you
to
Define,
just
like
three
four
buckets
and
then
you
can
just
assign
approaches
to
those
buckets.
So
the
class-based
resources
is
just
directing
continues
to
one
of
those
buckets.
H
G
And
have
we
I
guess
of
jumping
ahead?
Have
we
demonstrated
the
class-based
resources
being
exercised
using
annotations
on
a
pod
today
and
where
did
that
fall
down
that
made?
It
have
to
go
onto
the
core
pot
spec
versus
the
dynamic
resource
allocation?
I'm
hearing
needs
to
be
in
the
core
interaction
pattern,
because
it's
dealing
with
a
real
counted
entity
that
is
similar
to
a
device
that
has
health
and
all
these
other
Concepts
associated
with
it.
G
H
G
Map
my
own
priority,
skew
to
this
I
would
say,
like
you
probably
want
to
have
Dynamic
resource
allocation,
because
it
unblocks,
maybe
a
new
class
of
devices
that
couldn't
be
supported,
whereas
class-based
resources
like
I'm,
under
the
assumption
right
now,
maybe
incorrectly,
that
people
can
explore
the
space
just
using
pod
annotations.
And
since
it's
opaque
to
kubernetes
in
general,
yeah.
H
J
It
can
be
yes,
we
we
kind
of
support
through
annotations,
but
we
Skip
and
blocks
were
like
the
future
work
of
improving
run
times
and
Kublai
communication,
so
practically
what
it
unblocks
were
Discovery
and
when
scheduling
so
right
now,
the
problem
like
the
biggest
problems
with
annotation
is
what
our
connotation
can
be.
Anything
scheduling
doesn't
care
about,
annotations
and
and
well
it's
it's
practically
uncontrolled
data
and
we
want
to
solve
it
like
step
by
step.
J
G
G
J
One
simple
use
case:
what
we
can
utilize
right
now
is
what,
for
example,
inside
cryo,
we
have
a
function
which
tries
to
detect
the
Pod
qos
class
based
on
the
C
group.
Name
and
words
is
a
bit
High
cash
and
not
reliable
thing
with
proposed.
Cri
change
allows
us
from
a
couplet,
explicitly
communicate
what
we
spot
is
guaranteed
class
and
we
simplify
our
code,
both
in
kublet
and
in
cryo
foreign.
G
I
guess,
but
either
way
it
sounds
like
there's
not
an
unclear.
There's
not
a
shared
priority
between
these
two
and
I'll
just
have
to
read
on
them
both
and
do
the
best
I
can
to
you
tease
apart.
H
L
Extending
CRI
and
the
the
first
one
is
like
much
more
complex.
G
G
Guys
be
willing
to
get
together
and
see
if
you
can
come
up
with
names
that
you
think
don't
overlap
and
better
separate
the
two
concepts.
J
Well,
it's
one
of
the
things
why
we
need
reviewers
with
better
ideas
how
to
name
it.
We
come
up
with
best
what
we
can.
Okay,
based
on
our
understanding,
like
a
reason
why
we
are
using
the
class-based
resources
because
it
comes
from
the
hardware
so,
where
qos
and
cache
site
or
memory
bandwidth,
it's
called
classes
and
that's
why
we
are
using
the
cluster
here.
J
But
even
if
there
are
people
who
are
interested
to
discuss
variants
or
like
do
brainstorming
I'm
happy
to
to
host
for
sessions
like
some
some
ad
hoc
discussion
to
to
help
with
that,
and
of
course
we
can
clarify
or
all
the
details.
You
know
yeah
that.
A
Was
the
new
one
using
class
based
resources
from
now
on?
So
then,
let's
just
forget
about
previous
one
consolidation,
training,
yeah.
L
A
A
And
we
review
that
previous
design
long
time
back
many
years
ago,
so
maybe
I
I
can't
find
someone
to
for
the
second
one
here,
like
the
I
think
someone
have
the
CRI
background
and
someone
with
the
kernel
background.
B
Like
from
a
review
perspective,
it's
more
important
to
understand
where
to
stop,
rather
than
like.
What's
the
right
implementation,
because,
as
Derek
said,
that
we
want
to
understand
what
scenario
we
enable
with
the
first
step
and
for
me
first
step,
it
seems
a
little
bit
too
small.
I
I
would
think
that
we
want
to
do
a
couple
more
steps.
Alongside
with
this
one
but
yeah,
we
can
discuss
some
capital.
A
F
A
B
Think
Northstar
is
more
or
less
understood
and
like
if
it's
agreed
on
like
I
mean
it's
one
question,
then
we
need
to
understand
how
big
First
Step
should
be,
because
right
now,
cap
proposes
one
CRI
change
and
then
a
cubit
reaction
on
some
photo
annotations,
which
are
not,
which.
F
B
Maybe
we
can
even
include
it
into
pod
spec
or
something,
but
obviously
it
will
be
way.
Bigger
discussion
then
exactly.
H
L
I
I
I
A
I
remember
we
discussed
this
because
we
see
the
based
on
this.
We
want
to
Through
The
annotation
to
see
this
is.
It
is
really
solved
the
problem.
We
want
to
demonstrate
that
problem
and
then
we
can
come
back
to
discuss
what's
next
I
believe
that's
the
Waterway
or
Grandpa,
but
I
also
I
think
the
circus
concern
is
more
likely.
We
want
to
see
the
entire
of
the
big
picture
and
in
that
cup
instead
only
just
say:
oh
here's
The
annotation.
We
want
to
do
this
one,
it's
more
like
from
the
Cog
perspective.
A
Here's
the
problem,
here's
the
block!
I
o
Lois,
the
neighborhood
issues,
how
I'm
going
to
solve
this
problem
for
my
giving
workload
so
I
think
that
maybe
this
is
the
one
we
could
improve
in
the
current
curve
here
and
explain
the
use
cases,
here's
the
problem
and
see
how
we
are
going
to
be
using
annotation.
So
this
is
one
these
cases
I
do
hope
not
just
to
CRI
people.
We
want
to
some
have
the
kernel
expertise
also
check
in
here,
and
she
won't
think
how
we
are
going
to
long
term
to
move
forward
right.
A
So
that's
really
solve
problems,
so
we
need
some
other
reviewer
and
yeah
okay
anyway,
I
already
included
in
1.12
or
some
sorry
1.25
tracking,
but
do
like
what
do
I
say
either
earlier
we
have
the
19
type
enhancement
total
we
have
to
consolidate,
and
sometimes
the
we
do
have
limitations
on
the
reviewer
and
a
poor,
bandwidth
and
and
also
in
previous
also
another
feature
is
pretty
big
feature
like
a
username
space
and
also
Dynamic
resource
allocation.
Many
things
actually
speak
features,
so
we
need
to
figure
out
the
priority.
A
L
A
A
D
E
Yeah
hi
everyone
I'm
from
the
uks
nodes
team
I,
was
actually
asked
to
bump
this
pull
request
by
someone
on
our
control,
plane
team
and
it
doesn't
look
like
anyone
from
control
plane
is
on
the
call,
so
I'll
try
to
give
whatever
context
I
have
here.
The
way
I
understand
this
right.
Now,
the
cubelet
only
annotates
the
node
object.
With
this
node
provided
IP
win,
an
external
cloud
provider
is
being
used
and
I
think
in
eks
123,
which
we're
kind
of
preparing
to
launch.
E
There
is
essentially
like
a
competition
or
setting
the
node
IP
when
you're
running
cubelet's
entry
cloud
provider
as
well
as
Cloud
controller
manager.
These
two
kind
of
controllers
compete
on
this
annotation
again,
I
might
have
this
wrong,
so
this
is
a
pretty
simple
change.
It
basically
just
unconditionally
sets
the
sanitation
when
the
IP
changes.
This
is
particularly
helpful
during
upgrades
when
the
ipe
May
flop
kind
of
back
and
forth
during
the
upgrade.
So
this
ensures
that
The
annotation
always
reflects
like
the
correct
IP.
E
As
far
as
I
understand
it.
This
is
kind
of
like
a
correctness
issue
in
123,
so
control
plane
is
trying
to
get
this
merged
Upstream
before
our
launch
and
I.
Think
they've
just
been
having
some
trouble
getting
some
eyes
on
it,
so
yeah
I
think
that's
my
current
understanding
of
it
again.
It
may
not
be
completely
accurate,
but
we
just
wanted
to
get
some
attention
on
this
PR.
D
E
I'm
not
entirely
sure
I
believe
I
believe
it's
an
issue
for
us
in
123
because
we'll
be
running
CCM
in
the
control
plane,
but
I
think
the
cubelet
is
still
running
like
an
Intrigue
cloud
provider
or
controller
of
some
kind.
So
it's
it's
the
presence
of
those
two
controllers.
That's
the
issue
for
us
in
123.,.
E
Obviously,
we'd
like
to
see
a
Cherry
Picked
to
123,
but
we
on
certain
occasions.
Oh
sorry,
on
certain
occasions,
as
long
as
things
merge,
Upstream,
we'll
just
cherry
pick
them
internally
to
our
internal
builds.
So
if,
if
the
community
isn't
willing
to
cherry
pick
this
back
to
123
as
long
as
it
gets
merged
on
kind
of
a
newer
release
line,
then
then
we're
comfortable
cherry-picking.
It
internally.
A
Yeah
I
mark
this
is
1.25
Milestone
yeah
it's
for
now,
yeah,
okay,
thanks
everyone
and
that's
all
for
today,
and
also
which
is
perfect
and
to
wrap
up
time
and
I
will
get
back
on
the
cluster
across
the
based
resources
and
I
I.
Think
we
want
to
have
the
two
week
viewer.
Maybe
we
just
need
one
kernel.
I
know
Manu
already
on
many
many
things.
So
that's
why
I
didn't
propose
his
name
there
and
many
things
we
so
need
you,
and
so
I
will
be
the
approval
follow-up
rule
for
that
one
yeah.