►
From YouTube: Kubernetes SIG Node 20230912
Description
SIG Node weekly meeting. Agenda and notes: https://docs.google.com/document/d/1Ne57gvidMEWXR70OxxnRkYquAoMpt56o75oZtg-OeBg/edit#heading=h.adoto8roitwq
GMT20230912-170601_Recording_2560x1278.mp4
A
Hi
everyone
hi:
everyone
welcome
to
Sig
node
weekly
meeting
September
12
2023
today.
We're
gonna
mostly
cover
129
planning,
all
right.
Let's
get
started
so
we're
trying
to
work
on
this
new
table
and
we'll
copy
things
over
from
the
previous
table.
So
first
thing
on
the
list
is
sidecar
containers.
A
B
A
B
B
Yeah,
we
also
you
will
reach
in
and
trying
it
on
some
real
workloads,
and
we
will
report
back
when
we
have
more
clear
data.
Yep.
A
Great
setting
priority
to
medium
is
that
okay.
A
I
guess
don
it'll
be
maybe
what
do
we
want
to
focus
as
ignored
as
the
highest
items?
I
guess
like?
What
do
we
think
the
top?
Maybe
we
pick
two
or
three
is
a
high
priority
and
the
rest
can
fall
into
medium
and
then
things
we
that
are
not
that
important
can
be
low.
E
A
D
Yeah
definitely
a
lot
of
users
are
asking
for
it,
but
we
are
not
changing
the
stage
at
this
moment
like
we
it's
beta
and
we
want
to
spend
more
time
in
beta.
So
from
the
effort
point
of
view,
there's
no
change
in
the
stage
at
this
point.
So
I
don't
know
how
do
you
categorize
that
if.
E
High
nominee
in
the
company
like
if
a
high
I,
don't
I,
don't
care
how
much
effort,
but
the
like.
The
everyone
will
treat
it
as
the
top
priority
I
mean
so
that's
kind
of
the
like,
for
example,
if,
as
a
brewer
I
will
approval,
I
will
spend
the
time
to
read
the
type
or
make
sure
this
is
can
be
proved
before
timeline.
So
that's
kind
of
the
that's
the
whole
thing.
How
much
you
you?
When
you
have
the
math
postings
in
your
queue,
which
one
you
should
spend
time
anyways
first.
F
A
It
tight
so
the
next
one
is
fine
grade:
supplemental
groups,
control
I
think
we
did
not
get
an
update
on
the
enhancement,
so
we
may
need
to
check
if
the
original
author
is
still
available
or
will
need
to
find
a
new
owner
for
it.
A
Question
is:
do
we
have
anyone
on
the
call
or
in
the
community
willing
to
take
it
forward.
B
A
So
we
try
to
find
an
owner
and
then
see
if
we
can
make
progress
on
this
one.
Okay,
maybe
I
should
have
added
a
third
option
of
BR
beside
GS
or
no.
G
A
C
Right
and
priority.
G
Yeah
I
can
speak
to
that
one.
This
is
Kevin
I've
been
working
on
trying
to
get
that
I
got
that
promoted
in
1.28,
press,
private,
Alpha
and
I'm,
hoping
to
promote
it
to
Beta
1.29.
G
G
H
G
It
node
API
accounts.
A
G
And
the
priority
and
everything
is
fine.
A
A
All
right,
so
the
next
one
is
support.
I
think
maybe
don
Sergey
I
think
we
were
looking
for
someone
from
Google
too.
E
A
All
right,
the
next
one
is
cubelet
plugin
model
based
of
dra.
B
So
I
I
was
copying
from
a
table
and
I
skipped
graduates
at
Google,
resource
and
point
to
GA,
because
it's
graduated
yeah,
okay
Qs
with
C
group
C2.
B
We
cut
it
out
and
we
don't
have
clear
path
forward.
If
anybody
want
to
call
will
disagree,
please
let's
read
them.
I
A
I
B
I
J
A
H
A
Can
clean
up
the
other,
but
the
next
one
is
limit
on
Parallel
image
pull
ribbon
if
you're
on
the
call
yeah.
D
So
Paco
talked
to
me
about
this,
so
we
can
work
on
this
together
in.
A
So
the
next
one
is
support,
username
spaces,
so
I'm
not
sure
if
Rodrigo
is
on
the
call.
I
can
check
with
him,
but
the
status
is
we
got
the
alpha
support
for
volumes
merged
in
so
maybe
it
can
Target
beta
I,
and
the
second
thing
is
this:
Sasha
is
working
on
Sig
or
support
for
the
Pod
security
policy
to
update
it.
So
that
definitely
needs
to
happen.
The
beta
I'm,
less
certain
of
I
can
check
and
get
back
here.
A
So
maybe
medium
on
the
auth
work.
A
L
A
So
I
think
no
so
this
is
like
there's
runs,
so
Ramsay
is
all
said.
Siren
is
all
set.
So
this
is
the
security
policy,
so
there's
a
baseline
policy
right.
So
we
need
to
tweak
it
when
user
name
spaces
are
enabled.
You
can
actually
be
root.
You
can
get
all
capabilities
because
they
are
not
capabilities
on
the
actual
Road
username
space.
So
there's
a
pull
request
open.
We
had
a
discussion
with
cigarth,
but
it
needs
review
and
merge
on
the
outside.
A
Yeah
yeah
that
and
then
the
actual
support
from
signal
perspective,
whether
we
move
it
to
Beta
or
not.
A
M
M
A
M
E
M
A
E
Can
we
just
leave
Alpha,
but
the
Kevin
Kevin
can
change
this
one
right.
So,
okay,
all
right,
yeah,
yeah,
okay,.
E
A
Ignored
warning,
node
condition
based
on
PSI
David.
Do
you
remember
the
details
on
this
one.
I
Yes,
so
we're
working
on
on
this
one
to
come
up
with
a
cap
and
kind
of
like
a
whole
prototype
Dixie
doing
it
make
sure
this
one
as
well.
O
Sure
I
added
in
the
meeting
notes
as
well
so
I'm
working
on
the
cap.
For
this,
the
draft
is
almost
ready,
so
I'll
send
out
the
pr
by
Thursday
this
week,
okay,
so
this
was
basically
our
PSI
metric
support
in
gubernetes
and
the
respective
PRS
I
think
in
C
advisor
and
run
C
are
almost
on.
A
Okay,
so
the
state
is
still
kept.
We
can
move
it
to
Alpha
if
we
can
get
it
merged.
What
about
the
priority.
E
This
is
block
or
something
we
can
longer
time
things.
This
is
does.
E
Yeah
yeah
yeah,
yeah,
yeah,
yeah,
yeah,
sorry,
okay,
and
so
relative
will
be.
Can
we
put
that
Medium
because
we
do
want
to
make
a
problem
guys
on
this
one
at
least
right?
So
not
not
technically
immediate
customer
leaded,
but
we
don't
handle
this
a
while
like
the
kubernetes.
So
we
can
talk
about
it
for
years.
So
that's
yeah!
This
is.
E
I
A
And
I
think
the
other
one
is
like
PSI
based
evictions
right,
that'll
also
get
unblocked.
Yes,.
A
Okay,
all
right
thanks
folks,
the
next
one
is
ADD
Max,
restart
time
zone,
failure
to
pods
back
David,
you
listen
as
the
reviewer.
Can
you
talk
to
this
one
or.
I
Yeah,
so
this
one
this
one,
we
had
some
discussions
about.
Basically,
we
want
like
the
limit
how
long
the
crash
loop
back
off
period
is
I.
Believe
that's
this
cap
Yeah.
So,
basically
the
idea
is
like
we
have
a
pod,
that's
restarting
a
lot
or
a
container
is
restarting
a
lot
and
you
don't
want
it's
going
to
crash.
Look
back
off
because
it's
expected
that's
sort
of
the
idea
here.
F
A
Sounds
good
the
next
one
is
cubelet
drop
in
directory
Peter
Jerry
this
one.
P
Yeah
I'd
like
to
move
this
to
Beta,
I'd,
say
I
could
be
say
low
priority.
But
let's
say
it's
going
to
be.
C
A
M
A
Maybe
just
keep
just
pause
on
this
one
and
come
back
to
it
later
in.
F
A
Right
thanks
tellur
memory
manager,
GA,
graduation.
A
Right
thanks.
The
next
one
is
static
policy
option
to
spread
hyper
threads.
Oh
yes,
this
is
from
by
duns
folks,
so
I
think
they
were
gonna
open
an
enhancement.
So
this
is
still
targeting
a
cap
and
if
it
gets
approved,
we
can
go
for
an
alpha,
I'm,
not
sure,
but
we
want
to
set
us
a
priority.
A
J
A
So
we
keep
it
low,
I
mean
priority
wise
or
make
it
medium.
A
B
Yeah
I
don't
know
how
potential
it
is
like
we
didn't
even
fail
to
add
Windows
support.
Lastly,
we
still
need
to
add
windows
and
then
after
windows,
we
will
get
into
a.
B
E
Yeah
in
this
feature,
yeah,
we
I
think
it's
high
to
us
to
find
new
owner.
Okay,.
F
E
O
B
So
that
makes
sense
there
will
be
a
talk
on
kubecon
coming
up
when
they
will
be
talking
about
the
feature
how
it
can
be
used.
At
the
same
time,
nobody
is
actually
working
on
making
it
useful
and
complete
and
I
think
one
more
issue
with
that
is
it's
designed
to
be
in
place
vpa
as
original
name
States.
B
But
now
most
people
want
to
treat
it
as
a
non
like
guarantino
restart
of
container
resize,
so
people
want
to
have
new
API
like
updated
apis
that
will
guarantee
that
Port
will
not
restart
when
resize
will
be
applied
and
I.
Don't
think
it's
what
designs
right
now.
So
my
point
is
I
think
the
feature
will
be
in
Alpha
for
a
long
time,
because
we
may
need
to
change
the
API
or
add
more
apis,
and
we
still
have
a
lot
of
Department
to
do.
E
B
Okay,
so
we
designed
the
API
saying:
please
try
to
update
resources
and
please
try
not
to
restart
people,
want
apis,
saying,
update
resources
if
it's
possible
to
do
without
restart.
B
R
A
quick
question
for
the
better
criteria:
do
we
think,
like
the
we
have
to
finish
the
Windows
support
and
the
new
API
yeah
I
asked
this
question
because
our
team,
the
the
violence
we
we
did
a
lot
of
improvements
on
the
In-Place
features
and
we
we
stopping
a
few
like
bug,
fixes
and
performance
Improvement.
R
So
if
there's
no
owners
at
this
moment,
probably
we
can
temporary
say
to
us-
and
we
can
just
finish
those
like
reported
issues
and
probably
it
won't
graduate
to
Beta,
but
we
can
fix
as
many
issues
as
we
can.
E
M
E
A
So
now
we
have
an
owner
okay,
so
this
is
still
medium.
Okay,
it's
all
set
the
next
one
split
STD
out
STD
here
our
log
stream
I,
don't
think
we
got
traction
from
the
original
owner,
so
this
is
kind
of
stuck
in
limbo.
A
J
B
A
All
right
sub,
second
granular
probes,
Mike.
L
Yeah
we
just
need
to
get
through
the
you
know.
The
cap
process,
rebase.
L
Out
with
this
one
I
think
it's
important
from
a
you
know,
performance
perspective,
especially
the
granular
part.
H
L
A
A
Right
I
set
it
to
medium
folks.
Let
me
know
if
this
disagreement.
P
Yeah
I
would
like
to
make
some
movement
on
here
in
the
cryo
World.
We're
getting
close
to
having
support
for
the
metrics
I
would
love
to
move
to
Beta,
but
I.
Don't
know
if
folks
would
want
to
block
on
continuity,
having
the
same
support
for
that,
we
could
have
like
a
house
by
default
beta
or
something
like
that,
or
maybe
find
someone
from
continuity
you'd
be
willing
to
implement
their
metrics.
A
Parity
so
Mike,
do
you
know
if
anyone
like
Mike
or
David?
Do
you
know
if
anyone
is
working
on
continuity
for
this.
L
H
E
P
So
we
also
we
have
yet
to
also
have
the
crowd
support,
so
it's
kind
of
blocked
on
CRI
in
general,
but
we're
moving
forward
with
your
cryo
changes,
so
I'm
hoping
to
have
those
done
by
129.
E
K
S
B
I'm
sorry
yeah
I.
A
So
I
I
know
that
Sasha
from
Red
Hat
is
planning
to
work
on
this
year.
I
pulled
progress
along
with
the
torrent
right
now,
so
maybe
we
can
pull
that
in.
A
Okay,
so
back
to
this
one,
so
basically
we
need
the
container
runtime
support
for
CRI
stats
right
before
we
can
move
it
to
the
next
stage
in
kubernetes.
I
I
think
that
plus
the
the
ete
tests
in
in
kubernetes
itself,
but
have
the
metrics
and
so
forth
right
does
that
make
sense.
Computer.
P
Yeah
there
is
a
PR
that
I
have
up
to
add
that
and
it
didn't
make
it
did
not
make
it
for
last
release,
but
we
also
I
also
have
someone
working
on
kind
of
making
better
and
eat
the
e-tests.
So
I
would
hope
to
add
that
in
129.,
so
I
think
the
biggest
like
thing
that
we
need
to
focus
on
is
getting
an
owner
for
containerd
to
support
the
mattress.
E
F
B
And
I
don't
want
to
sound
as
a
broken
record
and
I
think
I
already
voiced
it
last
time.
We
need
clear
documentation
for
customers
how
to
transition
metrics,
especially
if
something
could
change
I
know
we
don't
have
any
guarantee
on
Metric
name
and
shape,
but
I
think
many
people
took
depends
on
that.
So
we
need
to
have
clear
migration
documents:
learning
how
to
get
certain
like
what
metrics
used
to
be
what
it
they
will
be
with
CRI.
P
Yeah,
so
actually
a
a
pretty
important
piece
of
all.
This
is
like
in
the
the
expectation
on
the
CRI
implementations
is
to
have
metric
parity
with
the
advisor,
at
least
for
the
Pod
and
container
stats
so,
and
part
of
the
end-to-end
tests
that
are
going
to
be
added
will
prove
that
you
know
we'll
be
testing
that
so,
ideally
they're
actually
I
mean
we
can
have
like
a
little
blurb
about
like
what
is
changing,
that,
like
the
source,
is
changing
a
little
bit
but
like
the
endpoint
is
going
to
stay
the
same.
P
The
content
for
the
metrics
should
be
largely
similar.
It's
just
like
the
source
is
going
to
change
from
these
people.
I
say
as
they
move.
B
Yeah,
if
shape,
wouldn't
change,
it's
fine,
but
if
anything
changes
we
need
to
make
sure
that
we
communicated
it
and
we
also
have
CRI
test
repository
I
mean
it's
a
series
together,
but
you
can
add
tests
there.
So
you
can
test
any
runtime
on
that.
P
L
P
A
I
H
A
So
I
think
that's
that's
been
added
as
a
requirement
on
the
cap.
Besides,
the
original
owner,
like
Sasha,
is
also
gonna
work
on
it,
so
it's
still
kept,
but
I
think
this
should
be
a
medium.
What
do
folks
think.
J
You
can
remove
uh's
name
from
owners,
the
second
one.
Yes,
okay,
thanks,
okay,.
A
Sasha,
okay,
ensure
secret
pooled
images,
Mike
and
I
know
Peter
from
Red
Hat
was
also
working
with.
L
Oh
yeah
yeah
started
doing
the
storage
work.
It
looks
pretty
good.
We
still
need
to
get
through
the
cap
on
it
and
but
you
know
the
code
works.
The
issue,
though,
is
that
we've
been
in
this
work
group
we've
been
identifying
some
possible
changes
that
we
might
want
to
make.
So
we
might
want.
Let's
I
mean
it's
high
priority,
but
requires
you
know
some
work
agreement
right,
Peter.
P
Yeah
I
think
we
can
kind
of
focus
on
finding
the
Direction.
You
know
in
this
yeah.
A
Last
time
Mike
we
cigarth
had
raised
some
concerns
or
we
passed
like.
Does
the
current
design
address
those.
L
Right
and
and
that
had
not
happened-
I
mean
it
was
they
wanted
to
switch
from
the
house
to
Shaw
I
mean
we
can
do
that
real,
quick,
okay,
I,
don't
think
we
were
actually
doing
you
know
storing
it
unencoded,
but
you
know,
which
is
the
way
we're
doing
current
secrets.
L
L
It's
it's
just
yeah,
it's
it's
a
heck.
If
you
don't
Heck,
if
you
do
right,
we
had
an
in-memory
implementation
of
this,
which
was
more
secure
right
because
it
didn't
store
anything.
It
is.
But
then
the
request
came
in
to
store
it,
so
they
would
live
over
a
reboot
instead
of
just
you
know,
have
to
re-pull
over
a
reboot.
L
But
then,
like
said
the
auth
guy
said
well,
it
needs
encryption
and
you
know,
but
the
real
answer
is
don't
put
creds.
You
know
across
the
cry
wire.
C
L
A
K
F
B
Yeah
they're
still
looking
for
scenario
to
cover
with
this
I
think
one
of
the
problem
would
be
that
we
still
owe
the
refactoring
and
in
place
pod
update.
That
is
in
addition,
dynamism
in
in
this
area.
B
B
F
A
Low
priority,
which
stage?
Is
it
alpha
or
beta.
B
F
G
Yeah,
this
is
part
of
the
like
sorry,
the
image
work
group
the
Peter
was
talking
about.
This
is
around
splitting
the
the
writable
and
the
read-only
layers
and
and
having
cubelet
be
aware
of
that,
because
the
container
runtimes
allow
that
no
on
these
cryo
allows
that
so
the
hope
is
to
at
least
yeah
get
it
in
Alpha
and
one
dot.
29.
A
E
O
F
Yeah
yeah
I
think
the
viewer
maybe
assigned
Ryan
because
he's
been
involved.
I
Yep,
can
you
add
me.
F
Q
Yeah,
hello,
so
I
reviewed
the
the
cap
and
it
was
planned
for
GA
in
next
cycle.
Not
this
cycle,
but
I'll
just
get
in
touch
with
the
author
and
check.
If
that's
the
case
and
update
here
is
needed,
so
I.
P
Not
yet
so
this
is
I
I
need
to
learn
a
cap
still
I'm
hoping
to
do
that
like
in
the
next
couple
of
days.
But
it's
you
know.
P
The
the
cap
would
be
just
for
like
separating
the
like,
currently
there's
a
time-based
image,
garbage
collection
and
the
disk
based
image,
garbage
collection
and
to
have
the
time-based
one
trigger.
You
need
to
have
the
disk
addition
to
the
like.
You
know
the
disc
needs
to
be
low
enough
capacity,
and
so,
and
the
Hope
here
would
be
to
make
the
time-based
one
be
able
to
act
independently
of
the
this
size
this
one,
so
it
should
be
fairly
targeted,
but
I
wouldn't
put
this
at
a
wildly.
A
A
B
Yeah
yeah
I
really
want
it
high,
but
I
don't
know
like
how
many
other
people
want
the
same.
A
All
right,
so
we
still
need
to
get
the
cap
merged
right.
So,
okay,
yep.
A
N
A
N
N
A
N
Usually
Patrick
reviews
it
and
Kevin
so.
L
I
know
it
requires
CDI
on
the
container
run
time
side
I
need
some
push
or
some
you
know
reason
to
cherry
pick:
CDI
support
back
to
1.6
LTS
for
containerdy.
B
L
L
E
If
you
didn't
want
to
help
you,
but
I,
don't
say
that
at
this
moment
that's
right,
okay,
yeah,
unfortunately,
is
the
minute.
Do
you
have
some
like
the
from
the
from
the
eye
perspective
on
your
side,
though
yeah.
A
So
I
think
we
have
folks
trying
to
make
this
available
on
our
side
in
production,
so
I
think
I
looked
them
of
any
feedback
here.
L
B
Yeah
I
think
it
needs
to
go
to
stable,
I'm,
not
sure.
K
Yes,
this
one
is
promoted
to
Beta
in
1.25
as
in
and
reverted
for
about
an
hour
after
that
we
have
take
some
time
to
fix
that
and
the
citizen
we
I
think
there
is
no
blocker
for
this
to
promote
to
Beta.
A
Well,
all
right,
so
we
try
to
go
for
beta
again,
so
I
think
we
need
a
reviewer
and
a
brewer.
A
Thanks
truers
Dawn,
if
anyone
Google
I.
E
B
We
will
need
to
bring
them
up,
but
I,
don't
think
we'll
do
it
in
the
next
three
minutes.
So
there
were
like
one
for
within
a
lot
of
variables
from
file
system.
There
were
like
one
that
empty
dear
a
limit
needs
to
be
promoted
to
stable.
So
there
are
a
few
that
needs
to
be
tracked.
Somehow,
maybe
next
week
we
can.
A
E
A
I
think
our
Parmer
was
that
one
I
think
and
then
Kevin
needed
to
check
for
the
for
the
CDI
in
GPU.