►
From YouTube: Kubernetes SIG Node 20220301
Description
Meeting Agenda:
https://docs.google.com/document/d/1j3vrG6BgE0hUDs2e-1ZUegKN4W4Adb1B6oJ6j-4kyPU
A
Hello,
everybody
and
welcome
to
today's
edition
of
seek
node.
It
is
march
1st
2022.
We
have
almost
nothing
on
the
agenda
today,
but
we
do
have
prs
and
some
quick
announcements
before
we
open
the
floor.
So
sergey,
do
you
want
to
start
us
off.
B
Yeah
we're
doing
relatively
good
on
prs.
We
grew
a
little
bit
but
generally
keeps
a
level
the
same,
so
good
job
everybody
please
keep
reviewing
and
approving
things.
A
Awesome
and
then
for
my
side
on
announcements.
I
just
sent
out
a
reminder
to
the
mailing
list
as
well,
but
as
far
as
the
deadline
goes
that's
coming
up.
That
is
this
friday
march
4th.
We
expect
that
anything
that
is
either
a
beta,
graduation
or
a
deprecation
will
be
done
by
that
time.
If
it
is
an
alpha
or
a
ga
feature,
we
expect
that
there
will
be
a
reviewable
pr,
so
there
has
to
be
like
a
pr-
that's
open.
A
That's,
hopefully
not
work
in
progress
that
people
can
take
a
look
at
and
give
feedback
on.
So
I
sent
out
a
message
to
the
mailing
list.
I
included
a
list
of
all
the
pr's
in
each
category
based
on
what
was
in
the
enhancement
spreadsheet,
and
so
that
will,
if
you
are
one
of
the
affected
authors
on
that
list-
and
you
have
say
like
something
coming
up
for
this
friday
and
it's
not
there
we'll
reevaluate
at
next
meeting
if
we
want
to
remove
stuff
from
the
milestone.
A
So
I
think
that's
all
on
that.
There
were
quite
a
lot
of
things
that
I
think
were
due
for
this
friday
and
I
don't
know
that
a
lot
of
them
are
necessarily
close.
So
did
anybody
have
anything
they
wanted
to
add
or
concerns
or
discussion
around
that.
A
Sounds
like
no,
so
it's
happening
this
friday,
we'll
reevaluate
next
week
just
wanted
to
make
sure
that
folks
are
aware
of
it
and
then
I
think
the
floor
is
open
unless
somebody
snuck
something
on
the
agenda.
C
D
I
just
want
to
switch.
We
can
do
the
easy
topics.
First,
I
guess
so
I
I
know
I
was
supposed
to
help
get
an
annual
report
out
last
week.
But,
as
I
said
earlier,
I
was
not
healthy,
but
I
would
like
to
work
on
it
if
not
this
week,
but
get
it
done
by
next.
Is
there
anyone
who
wants
to
help
maybe
start
authoring
that
I
feel
really
bad,
that
I
was
not
well,
but
I
would
love
help
if
anyone
wants
to
help
write
that
annual
report.
D
Okay,
thank
you,
daniana
I'll
ping,
you
on
slack-
and
I
know
I
said
I
was
gonna-
get
a
draft
up,
but
I
I
was
not
able
to
but
yeah.
I
know
we
need
to
get
that
done
and
then
the
other
thing
was
the
main
thing.
I
wanted
to
just
note
on
the
the
in-place
resizing
pr.
It
is
big
and
thank
you
for
all
those
who
looked
at
it
previously
as
well.
D
I
deferred
anything
when
I
was
looking
at
that
c-group's
feet
too
aware,
but
I
think
we
need
to
work
out
how
memory
clause
will
work
with
it.
So
I
don't
know
if
the
if-
and
I
was
here
if
I
forget
the
gentleman
who
authored
the
memory
clause
pr,
but
if
if
they
could
take
a
look
at
that
afterwards,
that
would
be
helpful
and
that
was
about
it.
C
D
There's
stuff
in
the
keyboard
code,
when
I
was
going
through
it
that
seem
like
it
could
be
off
like
it's
only
when
it's
polling
for
container
resources,
it's
only
getting
the
request
and
limit,
but
not
getting
min
or
high,
which
is
set
when
v2
is
on
and
it's
using
checkpoint
state
in
a
way
that
I
need
to
understand
a
little
bit
better,
because
I
was
I
ran
out
of
time,
but
there
were
some
other
adams.
D
C
Okay,
the
reason
I
reach
this
concept
is
because
it's
kind
of
like
the
been
going
on
for
the
almost
four
years.
Every
time.
D
No,
I
know,
but
that
doesn't
work
in
parts
done
so
like
the
eviction
code,
that
I
saw
a
lot
of
updates
on
that
like
that
needs
to
be
updated,
to
actually
look
at
allocated
and
not
requests,
and
so
that
I'm
trying
to
find
the
minimal
set
that
we
can
merge
it.
But
I
don't
think
it's
quite
there
just
yet.
I
don't
think
it's
an
error
on
venezuela.
He
just
needs
some
help
and
we're
going
to
try
to
help
them,
but
I'm
not
trying
to
delay
I'm
really
trying
to
make
an
active
effort
here.
Okay,.
C
Thanks,
I
just
want
to
say,
because
this
is
let
me
finish
this
one,
so
I
feel
like
the
because
it's
not
online
so
every
time
we
go
through
that
one
then
there's
the
new
comment.
So
then.
C
I
just
tried
to
wow
the
well
the
the
the
developers
hacked.
I
totally
understand
where
you
came
from,
because
I
also
give
the
new
requirement
other
things,
but
I
just
try
to
this
time
to
represent
the
community
contributor.
It's
like
because
it's
been
delayed
so
long.
So
every
time
the
people
change
and
there's
the
new
feature.
So
they
have
to
make
sure
there's
a
new
feature
and
align
all
those
kind
of
things.
D
D
C
Haven't
looked
at
the
native
one
for
a
while,
because
I've
been,
I
think
about
the
one
half
years
ago
after
I
provided
all
those
feedback
and
did
the
address,
so
I
just
lit
its
goal,
but
I
will
take
another
look,
but
I
just
want
to
take
this
chance
represent
the
community
contributor.
This
is
why
it's
so
hard
for
the
big
features.
C
C
I
know
I
know
yeah.
The
problem
is,
I
believe,
in
the
past,
also
split
into
multiple
pieces,
and
then
people
also
commented
they
couldn't
see
the
whole
picture,
so
they
ended
up
to
combine
together
it's
just
too
hard,
so
this
is
kind
of
like
the
I.
I
think
this
is
the
real
problem.
What
we
have
the
community
not
just
for
this
particular
pr
these
particular
plans
just
for
into
that
that
one
I
just
try
to
represent,
because
I
heard.
D
This,
I
think
it's
understood
don,
I
think.
Well,
the
key
thing
is:
I
looked
at
twice.
Thank
you
to
everyone.
Who's
looked
at
previously.
I
think,
because
it's
so
big,
it's
taking
time
for
everyone
to
find
issues.
I
do
think
there
are
a
few
issues.
I
found
that
don't
make
me
comfortable
to
merge
it
just
yet
because
it
could
have
unintended
side
effects,
but
overall,
this
is
a
big
effort
and
I'm
trying
as
fast
as
possible
to
to
help
unblock
the
night.
A
Awesome
anything
else
on
that
one.
A
Sounds
like
no,
let's
keep
things
moving,
because
we
got
a
few
more
things
added
to
the
agenda.
I
think
that
the
issue
linked
next
is
probably
gonna,
be
a
long
open
discussion.
So
maybe,
let's
jump
to
the
class
resources
kept
first
class-based
resources.
Marcus.
Do
you
wanna
go
ahead
and
present.
G
A
G
Yeah
so
the
it
all
started
with
intel
rdp
technology,
which
is
basically
a
kind
of
class-based
technology
for
for
doing
cache,
allocation
and
memory,
bandwidth
allocation
and
amd
and
arm
has
similar
technologies
that
are
controlled
through
the
same
kernel
interface
as
well,
and
it
started
with
that
and
then
rdt
is
supported
in
oci,
runtime,
spec
and
now
now
it's
also
supported
by
the
runtimes,
and
we
wanted
wanted
to
get
this
rdp
supported
in
in
kubernetes
as
well,
and
in
addition
to
the
rdt
we
have
als
already
also
implemented
a
similar
sort
of
class
based
method
for
controlling
block
ioc
groups
or
configuring
that
local
c
group
controller.
G
So,
for
example,
in
this
rdt
case
we
would,
we
would
have
like
sort
of-
let's
say:
gold,
silver,
bronze
classes
and
and
gold
gold
will
get
most
of
the
cash
lines
exclusively
and
then
silver
and
bronze
would
share
the
other
remaining
remaining
cash
lines,
sort
of
qos
qos
resources-
I
don't
know
if
the
name
name,
class
or
class-based
resource
is
the
best
one
to
describe
sort
of
yours.
U.S
resources.
A
Yeah,
so
just
a
quick
question
for
you.
Typically,
we
we
ask
that
people
have
like
a
discussion
link
or
have
brought
this
up
in
a
sig
or
that
kind
of
thing
before
we
consider
the
enhancement-
and
I
don't
see
a
like
discussion
link,
it
looks
like
we've
just
kind
of
gone
straight
to
a
draft
proposal
for
this
one.
So
thank
you
so
much
for
sort
of
bringing
this
to
the
stake's.
A
But
we
haven't
seen
like
you,
know,
a
presentation
or
something
like
that.
Are
there
other
folks
that
are
interested
in
this
or
interested
in
contributing
that
kind
of
thing,
because
we
typically
would
ask
that
you
know
we
show
that,
like
other
folks,
also
want
this
feature
before
we
can
go
and
consider
it
for
inclusion
in
the
upcoming
cycle.
G
Sure
I
can
open
a
select
discussion
on
that
or
email
thread
they're
all
like
email.
G
Yeah
sure,
okay
I'll
do
that
and
I
can
yeah
come
up
with
a
presentation
as
well.
If
that's
kind
of
desire
to
describe
it.
A
Awesome
yeah,
I
think,
just
making
sure
that
you
know
the
folks
who
would
be
interested
in
this
are
aware
that
this
is
being
proposed
and
getting
their
feedback
and
ensuring
that
you
know
we
have
lots
of
interest
in
this
before
we
can
consider
it
as
a
sig.
This
is
kind
of
a
smaller
group
than
we
would
normally
have
in
a
full
sig
meeting,
because
we
didn't
well
coming
into
this
meeting.
We
didn't
actually
have
an
agenda
so
yeah.
A
I
really
appreciate
you
bringing
your
ideas
to
the
sig
and
I
think
yeah
follow
up
with
like
a
mailing
list
thread
or
some
other
form
of
public
discussion
to
sort
of
gauge
interest
and
get
feedback
would
be
awesome.
A
G
A
A
H
I'm
a
new
member
to
the
community
so
still
learning,
but
I'm
very
interested
in
this
topic,
and
I
want
to
bring
to
our
groups
awareness
and
have
few
discussion
on
next
step
on
it.
So,
following
that,
I
learned
basically
in
1.22
when
we're
trying
to
fix
another
bug.
H
We
changed
a
little
bit
on
how
we
do
local
admission
before
that
or
even
today.
The
the
couplets
and
scheduler
are
basically
sinking
on
part
face
to
determine
if
a
pod
is
terminated
and
implicated
that
the
resource
on
that
pod
is
freed
up,
so
scheduler
can
use
that
freedom
resource
to
on
other
newer
parts
with
that
fixed
in
1.22.
H
It
seems
we
continue
to
use
the
same
logic
to
update
part
phase,
but
when
we
do
local
animation,
we
do
and
some
extra
check,
so
even
a
part
in
succeeded
or
failed
states
will
still
be
calculated
as
its
resource
is
still
used.
So
a
new
part
will
be
rejected
because
sort
of
a
resource
such
as
out
of
cpu.
This
is
basically
my
understanding.
Prep,
please
correct
me.
If
anyone
know
this
is
a
better
or
there
are
more
information.
H
So
with
this
I
think
in
the
community
in
last
few
days
I
see
david
bobby
page
and
clayton
and
jordan.
F
H
Discussing
that
in
on
slack
to
figure
out
what
is
the
proper
way
to
move
to
to
to
basically
reduce
the
impact
of
that
behavior
change
and
make
a
minimum
risk
fixes
back
to
the
kubit
as
a
hotfix,
it
looks
like
now:
clayton
have
put
up
a
fix
which
a
little
bit
large
and
also
depending
on
another
pr
that
has
been
reviewed
for
amounts.
H
Is
this
one
here,
that's
what
is
the
our
group's
suggestion
on
the
next
step
on
it.
I
A
Thanks
mernal,
I
was
going
to
ask:
we
don't
have
clayton,
but
we
do
have
david
porter
david.
Did
you
have
anything
to
add.
F
Yeah
so,
like
I've
been
working
clayton
on
this
yeah,
so
like
I
mean,
I
think
when
june
summarized
it
very
well.
The
the
main
thing
is
this
is
something
to
do
with
local
kind
of
local
pod
admission,
where
basically
kubla
today
it
sends
the
phase
update
to
the
status
manager
too
early
and
in
certain
cases.
Basically,
when
we
have
something
like
a
job
or
something
like
that,
where
there's
lots
of
pods
getting
scheduled
to
the
node
and
terminating
very
quickly.
Obviously,
there's
like
a
lot
of
churn
on
the
node.
F
What
can
happen
is
we
can
get
a
race
basically
where
a
pod
can
be
in
terminating
and
a
new
pod
can
be
scheduled
and
since
there's
a
race
right
now,
where
the
phase
to
the
status
manager
is
sent
before
the
pod
is
fully
terminated,
there
can
be
a
race
where
then,
during
a
local
pod
admission,
a
pod
can
be
rejected
because
it
doesn't
have
resources
and
so
yeah.
It
gets
rejected.
F
Ultimately,
so,
basically
like
there's
a
few
different
changes
that
we
were
thinking
about
and
I
think
the
lowest
risk
change
is
to
basically
delay
the
status
update
until
the
pod
is
actually
fully
terminal.
F
That's
kind
of
the
change
that
clayton
is
proposing,
and
I
think
that
makes
sense,
except
like
there's
a
few
complexities
with
that
around
having
to
refactor
some
of
the
the
sync
pod
stuff
and
there's
a
few
there's
a
few
other
changes
that
need
to
be
made
because
it's
not
as
simple
as
simply
delaying
the
status
update.
So
I
think
whatever
change
we
need
to
do
like.
If
that's
the,
if
that's
the
option,
we
want
to
go
with
actually
delaying
the
status
update.
F
It
will
require
like
a
few
refactoring
changes,
as
clayton
proposed.
F
It
put
up,
I
put
up
a
regression
et
test
for
this.
I
think
this
is
like
the
second
issue
we
have
where
we
have
an
issue
with
local
pod
admission,
and
I
don't
think
we
have
super
good
test
coverage
there.
So
I
think
that's.
The
other
thing
we
need
to
think
about
improving
is
how
we
catch
this
in
the
future.
So.
A
We
have
various
areas
where
the
test
coverage
could
definitely
be
better.
Yes,
danielle
is
nodding
yeah.
A
D
C
For
sure
I
typically
hold
this
one,
I
saw
the
somewhere
already
approved
and
it
looks
good
because
I
think
we
we
need
a
little
bit
the
more
review
this
one,
at
least
for
test
results.
Right,
so
do
we
think
about.
We
want
to
merge
this
one,
because
I
noticed
that
it
is
pretty
small
change,
but
actually
the
impact
is
critical
in
the
critical
path.
So
do
we
want
to
just
merge
and
need
to
test
the
result
to
tell
us-
or
it
is
just
we
want
you
more
discussing
and
offline,
get
more
test
results.
A
I
assume
that
I
think
it
will
be
reasonable,
like
offline
out
of
band,
to
increase
test
coverage
renault,
I
think
you've
looked
at
this
pr.
The
most
recently.
I
C
Okay,
so
menu.
I
will
need
you
to
to
management
those
organisms,
all
those
kind
of
things
I
just
temporarily
hold
this
one
because
I
think
we
may
want
to,
but
I
agree
so
maybe
you
want
to
leave
this
in
first.
So
then
we
can
test.
Add
the
test
and
the
more
test
test
things
in
the
separate
one,
which
is
much
bigger.
One
right
thanks.
E
A
quick
question
for
david
david
does
your
end
to
end
seed
on
clayton's
patch,
the
bigger
patch.
F
Yeah,
so
I
have
a
so
as
part
of
kind
of
debugging.
This
I
put
together
an
ete
to
demonstrate
this
issue,
so
I
think
it's
part
of
clayton's
pr,
just
the
second
pr
to
actually
fix
the
issue.
Maybe
we
can
cherry
pick
that
as
part
of
that
pr
maybe
have
it
separate,
but
it
does
demonstrate
the
issue
and
it
does.
It
is
able
to
repro
it.
F
So
we
can-
probably
you
know,
add
that
test
and
use
it
as
a
signal
that
the
test
will
you
know
the
fix
will
actually
fix
the
issue.
Okay,.
C
A
Also,
I
think
we
are
out
of
agenda
items,
I
think
that's
it,
but
before
we
return,
I
just
had
one
quick
question.
Somebody
who
was
participating
in
the
meeting
today
asked
me
to
turn
on
the
live
transcript
from
zoom,
so
those
are
auto-generated
captions.
I
just
wanted
to
get
sort
of
a
feeling
check.
Were
those
helpful
today.
Did
people
appreciate
those?
Were
they
useful,
plus
one
in
the
chat
or
minus?
One
is
useful
too.
A
Awesome
got
a
few
plus
ones:
okay!
Well,
we'll
consider
continuing
to
do
that
in
the
future.
Glad
that
those
were
helpful
folks.
A
Okay,
I
think
that's
all
that
we
have
for
today,
so
it's
great
seeing
you
all
and
I'll
see
you
again
next
week
cheers.