►
From YouTube: Kubernetes SIG-Windows 20231024
Description
No description was provided for this meeting.
If this is YOUR meeting, an easy way to fix this is to add a description to your video, wherever mtngs.io found it (probably YouTube).
A
A
So
in
terms
of
announcements,
I
don't
see
any
new
announcements.
As
always
the
enhancements
freeze
is
coming
up.
It's
the
next
major
milestone
for
this
release.
That's
Thursday,
February
9th!
So
that's
a
couple
more
weeks,
a
little
over
two
weeks
to
get
all
of
the
enhancements
and
things
tracked
I
think
we'll
be
talking
about
one
of
them
today.
A
Not
can.
A
Which
is
usually
we
give
some
space
if
there's
anybody
new
to
the
call
any
new
contributors
that
would
like
to
come
and
say,
hi
introduce
themselves
feel
free
to
do
so.
This
is
optional
but
feel
free
to
just
submit
yourself
and.
A
Okay,
next,
we
can
go
on
to
the
agenda
now.
A
Yes,
okay,
James,
you
wanna,
discuss
this
sure.
C
Yeah,
so
we
got
out
of
the
report
from
a
customer
that
they
were
seeing
elevated
CPU
on
their
nodes
and
when
they
dug
into
it,
they
found
that
the
proof
counters
that
we
launched
at
the
beginning
of
starting
a
cubelet
was
taking
up
30
of
the
CPU
on
the
on
the
Node,
and
they
suggested
that
we
increment
The
Collection
time
from
one
in
10
seconds
and
when
they
did
this,
if
you
scroll
down
to
the
bottom,
they
they
saw
like
basically
the
CPU
dropped.
C
30
that
30
that
was
being
consumed
is
gone,
and
so
they
have
a
pretty
huge
Improvement
here.
I
guess:
I,
don't
have
any
major
concern
with
in
with
dropping
or
moving
this
up
to
10
seconds.
C
I
guess
the
biggest
thing
I
guess
I
would
see
is,
like
you
know,
breaking
a
customer
was
relying
on
one
second
usage,
and
you
know
now
now
they're
not
getting
the
Fidelity
from
and
maybe
they're
doing,
alerts
or
something
on
that
and
we
change
the
Fidelity
on
them
and
cause
kind
of
a
issue
within
their
monitoring
system.
I'm
not
sure
how
much
of
a
concern
that
actually
is,
but
I
thought
I'd
bring
it
up
here
and
see
if
anybody
had
any
thoughts
and
concerns
or
questions.
D
D
The
the
time
frame
for
birth
counter
and
if
I
remember
correctly,
it's
it's
over
a
couple
of
seconds,
but
I'm
not
sure
over
how
many
seconds
so
maybe
some
less
loss
of
measurements
might
be
lost
if
it's
too
high.
That
might
be
wrong
on
this
yeah.
C
That
was
one
of
the
concerns
that
I
brought
it
up
with
one
of
our
perf
teams.
Internally.
That
was
one
of
his
concerns
as
well.
Yeah
I
mean
I
I.
Think
configuration
is
possible,
but
do
we
want
to
map
something
all
the
way
through
cubelet
like
startup
Flags
or
you
know,
I
like?
Is
it
worth
that
is
kind
of
what
it
was?
I
was
wondering
or
I
guess
we
could
maybe
have
an
environment
variable,
but
I
don't
know
feels
like
a
slippery
slope
as
well.
B
A
C
Yeah,
it's
it's
literally
for
the
metrics
endpoint
in
like
if
you
go
to
cubelet,
slash
metrics.
C
Where
this
data
is
showing
up,
and
so
if
anybody's
scraping
that
the
Fidelity
of
the
metrics
would
change
with
with
this
change,.
C
A
One
other
interesting
thing
is
I
wonder
if
CPU
usage
would
go
down
if,
depending
on
what
the
process
priority
is,
I
could
see
that
this
being
another
issue
with
contention
and
it
would
give
the
kidlet
more
resources
it
might
fix
it.
I
might
try
and
something
out
with
that,
but
I
wonder
if
we
should
propose
that
that
this
would
be
configurable
in
the
container
runtime
to
then
and
cry
API.
That
might
be
a
good
middle
ground.
E
E
A
Okay,
I
was
gonna,
see
how
much
load
was
on
there
and
it
looks
like
it's
a
default
node
do
we
know
the
size,
I'll
I'll
have
a
couple
follow-up
questions,
I'm
curious:
what
side,
what
node
size
they
have
yeah
and
things
like
that,
but
I
agree
and
does
that
eventually
settle
down
or
does
it
stay
Spike
the
whole
time.
C
F
C
A
Yeah,
it
might
fix
a
lot
of
other
issues
if,
where
you
not
like
spending
CPU
spending,
CPU
Cycles
there
so
yeah.
E
A
F
I
mean
at
least
in
this
case,
if
you
change
the
interval
with
which
these
are
reported,
it
mostly
just
affects
the
monitoring
systems
right,
whereas
you
know
with
the
30
CPU
over,
impacting
the
actual
business
application.
Mm-Hmm.
A
B
C
Not
aware
of
scaling
yeah
exactly
that
I
guess
like
it
is
just
affecting
the
metrics
endpoint,
but
if
someone's
using
it
to
scale
or
something
like
that,
then
all
of
a
sudden
we
may
they
may
not
be
scaling
in
time
or
something
like
that.
So
but
I
I
mean
I,
I'm,
I,
guess
I'm,
leaning
towards
breaking
change
and
and
ship.
It.
F
Well,
and
with
the
30
CPU
usage,
you
know
we
they
might
be
over
provisioning
as
well
or
just
like.
You
know
in
scaling
too
much,
because
you
know
you're
you're
kind
of
seeing
the
notice
too
overload
it
just
because
of
the
metrics.
A
Yeah
I
can
write
a
comment.
I
think
I
I
would
like
to
know
what
node
size
they're
using
and
it
it
doesn't
seem
like
they're
doing
any
over
provisioning
Brandon
because
they
say
it's
a
fresh
install,
but
it
would
be
good
to
confirm
if
they
only
see
this
under
load
or
not,
and
if
the
Note
size
is
reasonable
and
if
they're
not
over
provisioning.
The
node
I'm,
definitely
in
favor
of
just
bumping
it
to
five
or
ten
seconds.
C
A
My
fresh
installed:
that's
what
I
was
wondering
too,
like
we
do.
Have
the
perf
dashboard
like
that,
the
perf
test
running
and
we
didn't
see
anything
using
30
CPU.
E
Anybody
yeah
I
think
we
should
get
more
data
from
here.
I.
Also
wonder
if
it's
it
shouldn't
be,
but
is
it
workload
dependent?
Are
they
running
something
interesting
on
there?
That's
causing
this,
which
it
shouldn't
be
right.
It's
just
a
counter,
that's
getting
pulled
so,
but
yeah
I
think
we
should
ask
them
if
they
have
node
configuration
and
things
like
that.
A
Yeah,
thanks
for
bringing
that
up,
did
you
add
this
James
yeah
yeah?
Okay,
our
event:
do
you
wanna,
take
over
for
a
moment.
E
Sure
yeah,
so
the
whole
no
log
query
feature
that
saga
continues,
but
I
think
it
has
reached
almost
a
Breaking
Point.
In
my
opinion,
if
you
read
Jordan's
latest
comment,
he's
basically
saying
that
the
login
point
is
going
to
be
disabled
by
default
and
they
are
strongly
suggesting
not
to
enable
this
for
conformance
purposes,
I
think
they're
having
security
concerns
about
this
being.
You
know
open
in
the
first
place,
and
it's
been
since
I.
E
Don't
know,
I
I,
don't
know
when
they
added
this
login
point
in
in
the
cubelet,
but
it's
always
been
there
as
far
as
I
know,
but
it
looks
like
Jordan
is
saying
like
we
shouldn't
do
this
and
I
spoke
about
this
internally
within
Red
Hat
also,
and
this
and
the
partial
there
is
like
yeah.
This
looks
like
it's
eventually
going
to
go
away
is
is
what
they're
saying
so
yeah
thoughts?
E
Do
we
still
keep
moving
forward
trying
to
get
this
in,
because,
if
Jordan
is
not
going
to
approve
this
right,
if
it's
actually
a
like
a
hard
note
from
him,
even
though
him
being
he's
not
being
cleared
about
this
I,
could
I
could
ask
more
questions
and
get
and
get
clarity
on
that,
but
I'm
a
little
hesitant
about
continuing
work
on
this
and
and
then
Jordan
finally
saying
no.
E
We
should
really
not
do
this
right,
it's
better
to
get
that
sort
of
feedback
early
on
this
has
been
going
on
a
little
too
long.
I
feel
I
delete.
This
sort
of
feedback
should
have
been
given
at
the
cap
stage,
and
we
should
have.
Never
even
you
know,
started
on
down
the
path
of
the
implementation,
but
I
almost
think
we
need
to
reconsider
this.
Do
we
need
to
come
up
with,
like
Alternatives
like
having
a
a
static
pod,
that's
running
on
Windows
nodes
and
Linux
nodes?
E
If
needed
that
performs
this
sort
of
similar
capability?
You
know
and
then
having
a
a
separate
client
that
that
hits
this
endpoint
and
not
go
to
the
not
go
to
the
cube
cuddle
at
all.
So
opening
the
I.
B
E
Yeah
I
can
keep
fighting
it,
but
if
it's
not
going
to
go
in
I
mean
it's
just.
You
know
just
getting
bounced
from
from
what
you
call
release
to
release
yeah.
A
I
I
think
I
added
a
comment
on
here,
because
I
didn't
see
this,
and
my
comment
was
Jordan's
comments,
seemed
kind
of
specific
to
things
that
definitely
run
in
pods
like
the
cube
API
server,
and
they
don't
want
to
expose.
They
want
to
limit
the
surface
during
that,
at
least
on
Windows.
We've
got
the
cubelet
on
the
system
so
and
probably
a
lot
of
money
since
I
was
do.
A
A
Like,
like
I
I,
think
that
Jordan's
comments
might
make
sense
for
a
lot
of
the
core
components,
but
I
don't
know
if
that's
applicable
to
node
and
I
kind
of
asked
here.
If
somebody
some
of
the
node
leads,
if
somebody
could
comment
I
think
if
somebody
from
node
comes
and
says
oh,
yes,
we
absolutely
want
to
limit
the
ability
to
the
logs
and
point
four,
the
cubelet
2.
Then
that's
probably
a
good
sign
to
just
like
step
away
from
from
this.
A
If
somebody
from
node
is
really
interested
in
this,
like
I
could
see,
I
I
feel
like
they
should
say
like
this,
like
limiting
will
expose
to
the
long
ten
point
make
sense
for
some
components,
but
for
cubelet
it's
not
so
I,
don't
know
if
who
we
want,
who
we
need
to
discuss
or
who
we
need
to
have.
A
That
this
will
be
from
node
responded
yet,
but
also
I.
Think
it's
it's
up
to
you
too
I
know
like
like
what
Jason
you
have
been.
If
you
want
to
keep
doing
this,
it's
been
a
long
journey
and
even
though
nobody
said
no
explicitly,
we
haven't
got
the
guesses.
We
need
either
yeah
yeah
I.
B
E
E
B
This
going
either
way
like
well,
because
I
think
that
we
want
this
feature
as
a
Sig
nobody's.
It's
not
like
we're
all
changing
our
mind
after
reading
that
one
comment
right
yeah
for
us.
It's
the
same
thing.
We
want
to
do
Coop,
CTL,
get
locked
and
and
see
why
containerdy
like
not
doing
anything
like.
E
Yeah,
what
I
would
like
to
see
is
I
I
for
one
I.
Think
Mark
has
a
good
point.
We
should
talk
to
someone
from
the
note
D,
because
I
need
to
go
back
and
get
reviews
from
the
notebooks
also
right
and
if
they're
going
to
also
push
back,
because
they
are
not
they're
not
seeing
this
because
the
way
I
looked
at
it
is
okay.
E
Let
me
get
buy-in
from
Jordan
and
Tim
and
then
I'll
go
back
and
ask
the
note
folks,
but
we're
already
sort
of
not
getting
full
buying
from
Jordan
is
what
I'm
seeing.
B
So
you
feel
like
it's
going
to
snowball
like
yeah
I.
Think
that's
my
nobody's
gonna
like
but
Superman's
gonna,
not
gonna,
come
out
of
the
clouds
and
like
could
it
yeah.
E
B
Mean
with
with
kpng
it's
kind
of
was
in
the
same
situation,
nobody
was
saying
no
and
we
could
have
kept
pushing
it
forward,
but
it
seemed
like
the
best
thing
to
do
is
to
make
the
women
proxy
out
of
tree
and
use
kpng.
That
was
kind
of
where
we
converged
and
if,
if
you
think
that
a
better
use
of
your
time
is
to
do
an
external
thing,
I,
don't
think
you
regret
that
no
I
mean
when
you
look
back
on
your
life.
You
know
50
years
from
now,
I,
don't
think
you
think
to
yourself.
B
Man
I
should
have
put
that
in
tree.
Like
that's
not
going
to
be
the
thing
you
know,
I
mean
like
yeah,.
E
A
F
A
A
Sign,
but
if,
if
he
comes
back
and
I
like
and
says,
I'm
just
looking
out
for
what's
best
for
the
project
and
for
the
components
that
I'm
most
familiar
with
I,
don't
agree
that
this
is
the
right
thing
to
do.
But
if
node
does
agree,
it's
the
right
thing
to
do.
I'm!
Okay,
with
that,
if
that's
the
answer
that
we
get
then
I
think
it
makes
sense
to
go
to
node.
But
if
there's
still
concern
even.
B
A
B
E
That's
that's
good,
so
I
think
what
we
can
do
is
you
know
independent
with
even
asking
Jordan.
Let's
just
get
the
note
side
of
the
story,
because
if,
if
no
one
says
no,
it's
easy
for
Jordan
right,
that's
like
hey
he's
like
okay.
No,
it
is
already
saying.
No,
so
let's
get
the
feedback
from
the
node
team
and
see
where
to
go
with
that,
and
then
we
can
post
back
at
Jordan
and
ask
him
also.
E
So
what
we
can
do
is
I
I,
don't
know
what
we
want
to
do
with
the
pr
at
the
moment.
Mark
do
you
want
to
add?
Would
you
want
me
to
add
more
comments,
or
do
we
just
add
this
as
an
agenda
item
to
the
to
the
node
side.
A
D
I
have
a
question
all
right,
so
correct
me
if
I'm
wrong,
but
technically
right
now
it
is
possible
to
get
the
logs
from
cubelet
or
whatever
node.
If
you
just
spawn
a
pod
on
us
on
this
particular
node
and
you
create
the
host
path,
mount.
D
Then
you
can
basically
exactly
into
the
Pod
and
get
the
logs
on
whatever
you
need.
So
what's
what's
wrong
in
simply
formalizing
this
type
of
behavior
and
having
an
explicit,
explicit
API
for
this
to
simplify
the
user.
Experience
then
to
just
you
know,
have
this
capability
anyways,
but
they
have
to
jump
through
a
couple
of
Hoops
to
create
the
pods
and
so
on.
To
have
those
Vlogs
that
the
user
needs.
C
So
I
do
I.
Think
I
asked
that
question
like
way
back
when
we
first
doing
this
and
that
the
answer
that
I
remember
was:
if
cubelet
doesn't
boot
up
and
you
can't
run
pods,
then
you
can't
get
logs
and
this
happens
on
Windows
fairly
regularly
and
it
makes
it
challenging
to
get
into
the
node
and
figure
out
what's
going
on,
and
so
that
was
kind
of
the
the
use
case.
From
my
understanding,
yeah.
E
A
E
There
was
some
talk
about
like
trying
to
get
this
gun
using
a
static
board.
Is
our
static
pods
not
that
dependent
on
the
cube
lip.
E
Okay,
but
I
guess
you
would
just
need,
because
even
this
feature
needs
a
cubelet
right,
so
I
guess
the
static
pod
would
not
use
cni,
and
things
like
that
right.
So
you
could
still
say:
I
could
drop
a
static
pod
in
here
and
look
at
it.
But
this
is
exactly
what
cloud
would
say
if
you
can
do
that,
using
that,
why
not
do
it?
But
David
had
a
good
explanation
about
this
about.
E
B
B
Yeah,
like
like,
you
could
easily
make
a
counter
argument
against
Claudio's
thing.
If
somebody
really
just
didn't
want
this
feature
like
they
could,
they
could
make
a
security
in
general.
If
somebody
doesn't
want
a
feature,
they
can
always
make
a
security
argument,
and
that's
certainly
so
here
right
so
well.
E
Yeah,
but
you
know
feature
gate
they
say
is
temporary
right,
like
you'll,
eventually
remove
the
feature
gate,
so
this
would
not
even
be
a
behind
a
feature
gate
that
would
be
like
an
option
that
you
add
here.
That's
going
to
be
always
disabled,
I
think
that's.
What
at
a
minimum
Jordan
is
suggesting.
B
B
B
The
smartest
thing
to
do
is
to
just
have
a
couple
of
really
direct
conversations
with
people
and
say:
look
I,
don't
want
to
spend
any
more
time
on
this
unless
you
think
it's
useful,
because
it's
a
waste
of
my
time
right.
Like
you
know,
we
we
had
to
ask
Tim
that
once
for
the
kpg
stuff
like
at
some
point,
we're
like
okay,
look
sounds
like
y'all
like
this:
do
we
want
to
do
this
or
not?
B
He
was
like
yep
do
this
this
and
this
and
we're
like
okay
and
we
kind
of
like,
took
that
and
stride
and
we're
like
okay
at
some
point,
we
will
do
all
those
things,
but
we
still
decided
to
Fork
the
Windows
proxy
anyways
for
now
right.
So
just
so,
we
could
get
it
done
and
so
I
think
you
know
Jordan's
a
nice
guy,
Tim's,
a
nice
guy.
You
just
talk
to
him
and
be
like
look.
E
Yeah
I
think
so.
The
way
forward
is,
at
least
in
my
mind,
is
you
know,
talk
to
the
note
team
see
what
they
have
to
say
about
this.
If
that
answer
is
also
hesitancy
Slash,
no,
then
we
need
to
drop
this
and
figure
out.
What
is
the
other
way
around
because
no
definitely
being
hesitant
is,
is
just
going
to
give
Jordan
more
data
points
that
hey
you
know,
I,
don't
think
this
is
a
good
idea
and
no,
it
also
does
nothing.
A
A
B
Yeah
and
and
let
them
know
the
situation
of
just
how
much
time
do
we
invest
in
it
right,
because
you
know
we
don't
want
them
to
put
the
decision
making
off
and
then
we
spend
another
two
weeks
waiting
for
that
right,
like
you
know,
if
they're
willing
to
work
with
us
on
the
feature
you
know
we
could
keep
going.
You
know.
E
Yeah
so
I
think
the
outcome
of
this
is
you
know
we
put
this
on
the
Note
team
I.
You
know
I
can't
plan
to
attend
it
this
year
this
week,
but
we
can
add
it
to
the
agenda
next
week
and
try
and
attend
the
meeting
next
week.
F
B
I
have
to
I
have
to
drop
everybody,
but
I'm
gonna
be
a
pairing
with
David
on
some
stuff
this
week
on
the
Sig,
Windows,
devtools
stuff
and
building
some
CI
for
the
Kube
proxy
Windows
service
proxy,
with
Damon
who's,
a
new
new
person,
who's
gonna
be
getting
involved.
So
if
anybody
wants
to
join
us
in
any
of
that,
let
me
know
I'll
be
around.
A
B
A
It
yeah
I,
quick,
I,
guess
a
teaser
I
had
a
cap,
Z
cluster
that
was
running
Calico
and
the
new
Q
proxy
and
I
did
was
able
to
deploy
some
IAS
pods
and
put
it
behind
a
load
Bouncer
and
get
the
external
traffic
rounded
to
it.
So
it's
a
step
big
step
now,
I'm,
just
struggling
with
the
pro
job,
trying
to
figure
out
why
I
can't
access
the
build
kit
to
commands
from
the
docker
CLI,
even
though
they're
in
the
image
and
raising
the
same
image
that
we're.
D
A
To
build
a
pause
image
which
calls
build
kit,
but
let's
start
with
that
later
all
right,
yeah
I,
need
to
drop
two,
so
I
think
I'm
just
going
to
end
the
meeting
for
everybody.
Thank
you
for
attending
and.