►
From YouTube: Kubernetes SIG-Windows 20231010
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
B
All
right,
hello,
everybody
and
welcome
to
the
Tuesday
January
1st
2023
iteration
of
the
kubernetes
segundos
community
meeting.
As
always,
these
meetings
are
recorded
and
uploaded
to
YouTube
so
be
sure
to
adhere
to
the
cncf
code
of
conduct.
Let's
get
started
with
some
announcements.
First
announcement
is
127.
Release
is
officially
started.
Yesterday
was
the
first
day
of
the
release
cycle
and
the
enhancements
freeze
is
probably
the
next
major
Milestone
that
is
on
Thursday
February
9th,
so
we
can
work
on
getting
all
those
in
and
then
hopefully
I'll
code
prsn.
B
Nope,
okay
next
is,
as
always,
we'll
see
if
there's
any
new
contributors
who
want
to
give
either
an
introduction
or
just
say,
hi.
What
you're
working
on
ask
any
questions.
So
if
there's
anybody
new
here
feel
free
to
unmute,
raise
your
hand
and
we'll
I'll,
let
you
introduce
yourself.
B
All
right,
it
doesn't
look
like
it
there's
a
couple
things
on
the
agenda.
If
anybody
has
anything
else,
feel
free
to
add
it.
The
first
agenda
item
you
want
to
click
on
that
James
is
something
that
I
wanted
to
discuss
in
a
more
public
forum
or
bring
attention
to
it
rather
than
just
in
this
PR.
B
But
so
this
change
is
proposing
using
a
different
way
to
calculate
the
memory,
that's
being
reported
by
the
node
stats
and
I
wanted
to.
You
know
just
yeah
start
a
more
public
discussion
on
on
this.
B
My
initial
thoughts
are
that
I,
don't
think
we
want
to
change
how
the
cubelet's
reporting
memory
stats,
at
least
not
without
some
sort
of
feature
Gator
a
way
to
you
know
progress
it
over
just
because
these
types
of
changes
also
often
are
surprising
to
people,
no
matter
how
much
we
say
it's
changing
or
are
documenting.
B
Why
and
I'm
more
inclined
to
suggest
that
they
use
something
like
Windows
node
exporter
if
they
need
to
get
these
stats
unless
there's
a
very
good
reason
to
have
the
cubelet
report,
these
stats,
but
I'm,
definitely
wanting
to
you
know,
start
a
discussion
on
this
I
know.
James
you've
been
doing
some
work
with
the
cri-based
stats.
B
You
might
have
some
opinions,
yeah
anybody
feel
free
to
comment
and
I
also
asked
I
was
hoping
Brendan
and
some
somebody
more
from
the
Windows
container
platform
like
internals
team,
could
also
comment
too
just
to
make
sure
we're
doing
the
calculations
correctly
or
reporting
what
we're
actually
thinking.
B
C
Are
we
is
this
they're
just
expecting
a
different
value,
particularly
like
the
way
that
memory
is
and
we're
reporting
the
right
thing,
but
there's
also
another
metric
that
they're
particularly
interested
in
that
shows
a
slightly
different
viewpoint
on
on
all
right.
My.
C
Yeah
I
guess,
if
it's
that
second
one
where
it's
like
just
a
different
Viewpoint,
then
maybe
either
a
new
metric
or
using
a
tool
like
note
exporter,
to
get
something
that
makes
more
sense.
That's
kind
of
my
thought
because.
B
On
on
Windows,
it's
like
your
physical
memory
usage
does
not
actually
indicate
that
you're
like
you,
can
be
very
well
yeah.
It's
just
it's
differently.
It's
like
the
paging
happens
way
differently
and
everything,
so
you
can
be
using
a
lot
of
memory
for
your
your
process
or
for
maybe
like
system
memory
and
not
actually
hit
all
of
its
glimmery
or
the
opposite.
So,
let's
yeah
maybe
ask
this
individual
again.
B
B
D
B
B
B
Okay,
next
agenda
item
is
something
that
I
added.
This
is
about.
I
know,
there's
been
a
couple
of
people
have
asked
about
this
arvind,
most
recently
in
slack,
so
for
people
unfamiliar
with
this,
like
normally
when
you
start
a
Windows
container
you're,
the
the
volume
that
gets
created
for
the
container
is
20
Gigabytes
and
with
Docker
that
was
configurable,
you
could
use
the
docker
CLI
to
set
store
like
a
storage
Ops
instead
of
bigger
size,
or
you
could
set
it
globally
for
the
docker
in
the
docker
config.
B
So
you
could
have
your
nodes
create
Windows
containers
that
had
where
all
windows
containers
would
create
a
larger
node
or
nodes.
All
the
containers
on
a
node
would
be
created
with
a
larger
volume
size
with
container
D.
The
plumbing
was
done
in
container
D
to
be
able
to
set
that,
but
there's
no
way
to
wire
that
into
construct
kubernetes
to
tell
containerdy
what
size
to
make
it.
So
this
PR
was
an
attempt
to
override
one
of
the
values
like
the
the
resource,
limit
values
and
signal.
I
didn't
think
that
was
a
good
idea.
B
Most
of
the
customers
that
have
run
into
this
issue
were
able
to
work
around
this
by
just
switching
their
pods
to
use
an
empty
their
volume
inst.
Instead
of
relying
on
that
container,
like
the
containers
volume
but
I'm
wondering
if
we
should
just
make
a
uthor
author,
a
cap
and
figure
out
how
to
expose
this
in
the
the
kubernetes
like
the
deployment,
specs
I
think
we
just
need
to
figure
out
the
either
pick
a
new
AP
like
field
in
the
API
and
figure
out
where
to
put
it
and
plummet
through.
B
But
most
of
the
work
is
going
to
be
discussions
around
that.
So
does
anybody
have
a
strong
opinion
either
way,
if
not
I'm
inclined
to
try
and
do
this,
even
if
it
doesn't
happen
for
127,
but.
A
A
E
B
E
B
Yeah
somewhere
in
HCS
or
HCS,
gym,
there's
a
default
that
says,
give
me
when
you
create
the
volume
that
you
unpack
all
of
the
contents
of
the
container
layers
into
rate
of
20,
Gigabytes
and
I.
Think
that
was
just
what
people
thought
was
a
reasonable
size
for
a
Windows
Server
core
container
image,
which
is
not
wrong,
and
since
it
was
then
provisioned,
it
doesn't
necessarily.
A
Yeah
the
concerning
bit
is
right.
Things
like
say
someone
is
putting
I,
don't
know
200
Politics
on
a
note,
and
each
of
them
gets
this
20
gig
virtual
volume
right
as
long
as
the
10th
provision
you
might
get
away
with
it.
Nothing
happens,
but
actually,
when
the
when
the
parts
start
getting
used,
bad
things
start
to
happen
on
the
Node,
because
you
start
to
run
out
of
space,
especially
if
they
are
not
aware
of
this,
and
they
have
not
given
enough
disk
space
for
that.
Node.
A
A
F
Yeah,
a
meme
I
think
you
added
the
k2s
stuff
yesterday,
so
I
just
saw
you
pushed
it
last
night.
So
that
means
we.
The
structure
of
this
is
now
kind
of
up
and
I
see
you're
there.
So
if
you
just
click
on
Windows
service,
proxy
I'll,
just
you
know
and
I
mean,
if
you
want
to
jump
in
here,
go
for
it,
but
it's
pretty
simple:
there's
a
Windows
backend
directory
and
that
back
end
has
the
kernel
space
proxying
stuff
in
it.
F
That's
not
that
different
than
if
you
were
to
go
to
PKG
proxy
Windows
and
look
at
the
entry
backend
kernel
space
and
then,
if
you
pop
out
of
there
and
you
go
to
CMD
you'll,
see
the
thing
that
we
use
to
run
it.
And
then,
if
you
go
into
k2s,
what
we've
done
here
is
we
made
a
couple
of
PR's
into
kpng
so
that
the
original
logic
of
kpg
was
this
idea
of
there's
a
cube
to
store
functionality
and
the
idea
there
is
you've
got
Kube.
F
You've
got
a
thing
that
watches
the
Kube
API
server
and
then
that
sees
stuff
and
then
it
pushes
it
into
an
in-memory.
You
know
sort
of
data
model
right
and
then
that
can
get
pushed
down
to
a
store
the
store.
In
this
case
you
know
we're
pushing
it
to
the
windows
back
end
now
and
the
way
you
do
that
is
you
is
you
literally
say
you
know
two.
F
You
know
two
windows
kernel
space
back
end.
Basically,
it's
compiled
in
now.
What
we've
done
here
is
we've
lifted
that
code
out
of
kpng
that
defines
that
defines
the
idea
of
watching
the
kubernetes
API
and
then
pushing
it
to
a
specific
back-end
and
we've
sort
of
copied
it
in
here
and
the
this
is
a
little
bit
of
a
not
the
way
you're
meant
to
use
kpng.
But
from
our
perspective
we
were
thinking
well.
This
gives
Windows
complete
control
over
anything.
We
can
add
our
own
command
line
options.
F
We
can
add
our
own
things
that
we
want
to
push
data
to
we
can
we
can
basically
do
anything
we
want
with
this
mechanism.
The
trade-off
here
is
that
we
aren't
exactly
doing
exactly
the
same
thing
that
the
other
backends
might
do
most
of
the
back
ends.
People
that
were
trying
to
extend
kpng
with
the
new
proxy
would
do
something
more
like
run
the
grpc
mode,
and
then
they
would
just
write
their
own
back
end
and
the
only
thing
the
back
end
needed
to
know
about
was
the
kpng
grpc.
F
So
there's
no
importing
of
the
kubernetes
API
there's
no
importing
of
the
kubernetes
client
there's
no
dependence
on
kubernetes
whatsoever.
So
it's
something
we
can.
We
can
back
out
of
this,
but
the
decision
we
make
here
is
a
it's
a
one-way
door
right,
so
we
go
in
and
we
do
this
and
we
own
our
own
proxy
forever,
except
for
the
core
KPMG
logic
that
we
depend
on
that
watches
the
API
server
and
translates
the
data
model
for
us
not
a
lot
of
technical
debt.
F
G
F
Is
that
we
would
be
depending
on
grpc
and
grpc,
would
now
Force
us
to
run
two
processes
and
the
initial
thought
I
think
from
the
Sig
was:
let's
not
have
two
processes:
let's
run
everything
in
memory.
Let's
try
to
be
similar
to
the
existing
proxy.
So
that's
why
we've
gone
with
this
road,
but
if
you
all
decide
we'd
rather
not
depend
on
the
entire
kubernetes
API
and
have
a
simpler,
lighter
footprint
and
we're
okay
with
that
grpc
model,
then
we
could
back
out
of
this
and
we
could
go
back
to.
G
B
B
Either
run
everything
like
this
or
depend
on
a
grpc
interface
is
when
I
was
the
last
I
heard,
which
was
probably
like
November
December
was
that
there
was
a
desire
to
move
kpng
itself
away
from
that
grpc
interface
and
to
find
a
golang
interface,
and
one
of
the
implementations
of
that
may
be
a
grpc
thing.
Is
that
still
the
case?
Because
if
that's
the
case,
we
could
probably
just
stick
with
what
model
we
have
here
until
that
all
gets
fleshed
out
in
yeah
the
kpng
proper
yeah.
F
There
is
this
idea
that
you
could
have
a
normal
for
most
people
that
want
to
run
everything
in
memory.
You
could
just
have
a
go
Lang
interface
and
you
implement
that
interface
and
then
everything
else
is
done
for
you,
but
even
in
that
case
you
would
still
be
having
to
vendor
in
all
of
kubernetes,
because
at
the
end
of
the
day,
if
you're
running
everything
in
memory
and
you're
watching
the
kubernetes
API,
then
you
need
client
go
and
you
can't
get
away
from
that.
F
F
E
Go
ahead:
this
is
the
couple
the
couple
enough
to
later
use
the
copy
original
Germany.
G
E
Thing
I
I,
think
it's
fair
to
do
is
like
bring
the
flags
of
the
back
end.
I.
Think
we
have
a
few
hard-coded
fields
on
it,
and
I
will
work
to
bring
the
the
current
Flags.
We
have
on
the
cube
proxy.
C
On
it,
the
one
we
want
to
get
rid
of
General
generally
so
yeah.
F
Yeah,
so
that's
yeah,
so
I
think
you
you
all
can
I
mean
I
think
this
will
be
something
we'll
have
instructions
for
testing
and
playing
with
it
within
a
week
or
two,
but
the
overall
structure
is
basically
there.
So
folks
want
to
look
at
it.
You
can.
If
you
want
to
think
about
it,
you
can,
it's
probably
a
little
premature
to
start
filing
bugs,
but
you
could
ALS,
you
could
totally
file
issues
against
it
or
ideas
or
whatever
you
know.
B
I've
got
another
question
to
it:
it's
maybe
related
to
some
other
kept
work,
especially
if
is
Michael
tapping
here.
Oh
no
I'm,
not
so
I
I
noticed
that
the
kpng
or
the
the
Q
proxy
code
in
kubernetes
kubernetes
had
some
feature
Gates,
which
I've
brought
up
in
the
past,
which,
like
enable
or
Windows
DSR,
and
there
was
one
other
one
that
I
had
actually
assumed
were
either
beta
or
already
graduated,
because
I
hadn't
been
touched
since,
like
118.
B
F
B
I,
don't
know
the
full
history,
especially
because
there
wasn't
this
wasn't
captured
in
it
kept,
but
my
understanding
was
like
the
DSR.
The
direct
service
turn
was
mainly
there
because
that
was
also
being
released
with
like
developed
in
Windows,
and
they
wanted
to
just
be
able
to
disable
it
or
enable
it
based
on
the
like
the
windows,
OS
patch
level.
In
case
there
were
issues
being
introduced,
but
I
don't
have
the
full
history
of
that.
Does
anybody
not
want
to
use
like
DSR
mode,
for
example,.
B
F
C
So
all
that
what
you
just
talked
about
Jay
about,
like
the
different
trade-offs
between
having
it
in
memory
versus
grpc,
did
we
write
that
down
somewhere
in
here,
like
just
a
short
paragraph
that
says,
like
we
decided
to
do
this
because
of
XYZ
I.
D
F
This
in
that
document
right
so
this
is
a
this-
is
kind
of
a
new
thing
that
I'm
just
bringing
up
today.
After
sort
of
thinking
about
it
and
I
talked
to
some
with
some
people
about
it
talked
to
McHale
some
about
it
and
we're
thinking
about
it
and
we're
like
well.
When
I
told
him
I
was
you
know
going
to
do
this
thing
of
copying
the
k2s
over
and
me
and
amine
were
doing
this
or
a
meme?
Did
it
I
did
the
you
know
got
all
this?
F
We
we
never
I,
know
I,
don't
think
even
me
and
amim
talked
about
it,
though
we
did
this
all
asynchronously,
so
I
guess
it's
time
like
we
should
yeah
I
should
just
write
this
in
that
document.
So
yeah
we
have
I.
I
will
write
it
in
the
document.
Then
I'll
repost,
the
document
in
here.
Should
we
move
the
windows
yeah?
Should
we
move
the
Windows
proxy
out
of
tree
so.
F
Definitely,
okay,
so
so
that
was
old,
notes
and
then
I'm
going
to
add
new
notes,
so
I'm
going
to
add,
updates
so
here's
the
document
I'm
going
to
link
it
for
folks
in
the
Sig
Windows
thingy.
Where
is
the
sync
Windows
here?
It
is
yeah,
so
I'm
going
to
relink
it
and
then
I'm
going
to
add
it
in
there.
F
And
we'll
just
keep
this
as
like
a.
A
F
F
F
F
Yep
yeah,
so
this
is
a
new
project,
so
if
anybody
wants
to
own
a
piece
of
it
or
a
part
of
it,
for
example,
Ci
or
you
know
doing
code
reviews
or
anything
like
that,
documentation,
anything
like
we.
We
definitely
would
love
an
owner
for
for
something
here
to
help
us
along
the
way
at
some
point.
But
we
don't
need
anybody
urgently,
because
right
now
we're
just
copy
pasting
code
and
moving
stuff
around,
but
within
a
few
weeks
I'm
sure.
B
It
may
be
completely
off,
but
I'm
doing
that
too,
and
then,
after
that,
I
think
we'll
have
a
good
at
least
place
to
measure
where,
where
things
are,
what's
working,
what
what
isn't
working,
if
it's
not
working
and
then
that's
when
we
can
make
like
start
to
make
decisions
about.
When
do
we
stop?
B
F
Yeah
yeah,
so
you
so
we
could
so
that
doesn't
sound
hard
right,
because
I
guess
we
could
just
add
a
make
Target
for
that,
and
so,
unless
there's
anything
official
that
you're
thinking
needs
to
be
done
there.
If
it's
just
a
matter
of
having
a
make
Target
and
a
CI
job
that
calls
that
make
Target
compiles
the
thing
and
and
uses
it.
That
seems
like
something
somebody
could
do
with.
B
We
could
even
just
like
for
for
some
of
the
integration
tests.
I
think
one
of
the
things
that
I've
known
just
from
looking
into
this
but
didn't
mention
here
is
we
have
this
Cube
to
store,
there's
a
mode
for
earning
kpng
like
I,
don't
know
what
it's
called
a
file
to
store
where
you
can
just
give
it
like
a
yaml
representation
of
what
all
of
the
routes
are
supposed
to
be
that
it's
supposed
to
program.
Tell
cue
proxy
go
program.
This
and
it'll
use
the
same.
B
F
C
Be
able
to
use
the
GitHub
Windows
VMS.
B
C
Okay,
make
2022
and
2019
so
and
I'm
pretty
sure
we
can
program
the
proxy
there
I
mean
I
can
run
host
process,
containers
and
GitHub
actions.
So
okay
should
be
good.
Only
thing
we
can't
do
in
GitHub
actions
right
now
is
run
like
hyper-v,
nested
virtualization,
but
I.
Don't
think
we
need
that
for
this
project,
so
it
should.
E
C
Yeah,
so
that's
I
think
what
Mark
is
saying
is
if
we
could
just
like
If
all
we're
doing
is
programming
the
hns
rules
then,
and
we're
not
setting
up
all
the
containers
and
running
cubelet
and
all
those
things
we're
just
making
sure
that
the
rules
get
written
into
hns
properly.
We
should
be
able
to
validate
that
as
like
an
integration
test
prior
to
having
to
like
you
know,
run
the
whole
suite
and
make
sure
everything
works
and
end
type
of
thing.
Yeah.
F
C
C
Yeah
and
that
now
that
spits
out
Json-
and
you
could
do
some
like
a
very
detailed
Json
about
the
state
of
the
network,
and
so
you
could
then
validate
that
pretty
extensively
which
we
aren't
doing
at
all
and
in
the
current
Q
proxy,
so
I
think
could
could
be
a
very
beneficial
yeah.
F
Okay,
I
just
filed
that
as
the
first
issue
and
then
you
know,
yeah.
C
Cool
so
we're
at
time.
So
is
there
anything
else
anybody
want
to
discuss
or
you
throw
it
on
for
the
next
meeting.
Otherwise,
I
want
to
make
sure
we
end
up
I'm,
respecting
everybody's
time.
F
F
Actually,
I
see
pramita
and
David
are
here
so
I'll
stay
after
hours
and
just
introduce
them.
Pramita
and
David
are
both
working
on
Sig,
Windows
devtools
stuff,
and
do
you
all
want
to
say,
hi
or
anything.
G
Hi
myself
and
I
just
started
with
working
with
Jay
on
indicating
Kimu,
with
the
background
I
found
some
some
of
the
issues
new
to
the
community
as
well.
I've
worked
on
majorly
on
networking
storage,
also
but
yeah
in
the
community,
a
newbie.
Thank
you.
F
F
Okay,
you
can
stop
the
recording
and.