►
From YouTube: Kubernetes SIG-Windows 20230620
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
Hello,
everybody
and
welcome
to
the
June
20th
2023
iteration
of
the
kubernetes
windows
community
meeting.
As
always,
these
meetings
are
recorded
and
uploaded
to
YouTube
so
be
sure
to
adhere
to
the
CNF
code
of
conduct
for
anybody
who's
new.
Here,
let
me
basically
just
boils
down
to
be
nice
to
everybody.
It
started.
A
All
right
first
announcement
is
last
week
was
code,
freeze
or
enhancement
freeze
and
the
next
milestone
for
the
128
release
is
code
freeze
and
that
is
coming
up
on
July
19th,
that's
about
four
weeks
from
now,
just
an
FYI
for
everybody,
as
I
was
kind
of
expecting
the
enhancement
that
I
was
trying
to
push
through
for
the
volume
sized
and
a
lot
of
reviews.
A
I
need
to
push
on
that
again,
just
during
I
think
during
not
the
enhancement
period,
unfortunately
I
think
Jordan
was
he
was
doing
some
reviews,
but
he
was
out
for
a
while.
So
I
think
that
was
a
big
reason.
I
didn't
get
much
attraction,
oh
well,
we'll
keep
pushing
it
for
that
for
next
release,
I
think
and
I
think
that
was
one
of
the
only
ones
that
we
were
tracking
right
now.
A
Does
anybody
else?
Have
anything
they'd
like
to
announce.
A
Okay,
no,
the
next
is
usually
on
these
meetings.
We've
got
a
space
that
there's
anybody
who's
new
to
the
the
community.
If
they
want
to
introduce
themselves
or
say
hi
or
you
know,
ask
any
questions
so
I'll
give
I'll
give
a
little
bit
of
space
if
anybody
wants
to
introduce
themselves
just
feel
free
to
unmute
or
unpause.
A
A
D
A
I
think
okay
well
I'll,
do
that.
If
we're
getting
into
the
agenda,
anybody
else
I
think
recognize
pretty
much
everybody
else
on
the
call
so
move
forward.
David
you
had
a
agenda
item
for
the
host
process.
Space
container
image
issues:
do
you
want
to
bring
that
up.
E
Yeah,
it's
probably
not
really
an
issue,
but
just
my
my
lack
of
knowledge
on
how
to
use
this
correctly.
So
I
was
trying
to
apply.
You
know
basically
a
manifest
like
a
yaml
to
my
kubernetes
cluster,
with
the
the
container
image
field,
pointing
to
the
host
process
based
container
image.
For
some
reason
it
was
not
pulling
it
down,
so
it's
wasn't
sure
what
what
the
issue
was
there,
but
the
this
still
works.
There
are
no
issues
with
it
right.
I
just
wanted
to.
A
A
Of
container
D
there
was
I
know
at
one
point.
There
was
an
issue
with.
A
There
were
the
some
build
tools,
changed
and
started,
pushing
the
images
with
a
like
I
think,
as
always
as
a
manifest
list,
so
that
it
could
include
some
provenance
information
as
another
artifact
that
was
also
on
there
and
that
caused
some
issues
with
the
platform
matching
logic,
and
there
was
a
patch
in
the
container
D
I
think
it
went
into
one
six
around
version
like
one
six
eighteen
to
loosen
the
the
validation
Logic
for
the,
because
we
just
have
the
single,
oh,
like
the
single
image
that
runs
on
all
the
OS
stuff,
so
try
and
run
the
latest
version
of
one
six.
A
If
you're
on
that
or
any
other
one
1.7
versions
should
support
that
and
if
not
try
posting
inside
I
can
mention
me
and
I'll
help
you
get
to
the
bottom
of
that.
E
Okay,
so
that's
good
and
the
the
latest
image
for
image
version
is
the
v1.0
or
the
base
container
image
and
for
container
D
Ultra
using
you
know,
1.7
or
one
point
1.6
or
higher
should
work.
A
C
A
D
A
B
B
B
Essentially,
you
know
consumes
a
full
state
every
time
when
any
event
happens
either
a
node
is
deleted
or
added
or
endpoint
slice
is
added
or
deleted.
Kpng
brain
part
will
trigger
this
callback.
This
callback
will
have
is
Will
essentially
have
the
full
State
all
these
services,
along
with
their
endpoints,
which
are
existing
in
the
kubernetes,
and
we
will
we.
We
will
be
basically
programming
the
data
path
for
all
these
services.
B
So
to
start
with,
I
have
this
hcn
service
I
have
our
own
definition
of
host
compute
endpoint,
Saint,
Paul,
load,
balancer.
B
B
B
So
one
more
thing
like
kpng
also
comes
with
a
digital
div
store,
is
basically
kind
of
service
change.
Tracker,
if
you
guys
are
familiar
with
Q
proxy
Windows
also
they
normally
have
a
servicing
tracker.
So
this
store
is
kind
of
a
new
generation
service
tracker
and
it's
a
generics
tool
where
you
can
track
state,
add
items,
and
you
know
after
processing
everything
you.
This
store
has
a
API
for
getting
the
new
objects
which
are
created,
the
objects
which
are
updated
as
well.
The
objects
which
are
deleted.
D
Yeah,
so
so
real
quick.
What
what
what's
happening
is
that,
like
this
sort
of
allows
us
to
do
a
few
things
all
at
once,
one
is
it
gives
you
the
potential
to
have
a
operating
system
specific
back
end
like
the
Windows
proxy
that
does
not
have
to
vendor
in
all
of
kubernetes
right
that
you
could
do
that.
D
It
also
gives
you
the
ability
to
like
to
to
have
a
Windows
specific
command
line
options,
Windows
specific
release
releases
for
your
back
end,
because
this
would
this
is
not
entry
right,
so
that
that's
the
other
sort
of
broader
thing,
but
yeah
yeah,
and
then
the
full
State
thing
that
Damon
talked
about
gives
you
the
ability
to
say
your.
D
Your
Coupe
proxy
will
get
the
full
state
of
the
network
every
single
time
and
that
something
changes
which
means
it's
a
lot
easier
for
Windows
developer
to
debug
it
and
play
around
with
it
and
figure
out.
What's
going
on
because
the
way
the
current
Coupe
proxy
entry
works?
Is
it
has
this
data
model
of
what
services
and
endpoints
are
and
it
sort
of
Aggregates
them
into
this
big
map?
And
nobody
knows
what
the
hell
is
in
that
map
at
any
given
time
right
unless
you
go
change
the
code.
D
B
So
to
essentially
a
programmers
service
in
Windows
we
have
to
create
host
compute
load,
balancer
and
host
compute
endpoints.
So
we
have
two
stores,
one
for
load,
balancer
and
one
for
remote
endpoints.
We
don't
track
and
maintain
life
cycle
of
local
endpoints,
because
that
those
are
programmed
by
CNN
and
we
don't
want
to
interfere
with
that.
B
B
B
B
So
it's
in
load
balancer
will
have
the
load
balancer
IP
the
list
of
XC
and
endpoints
the
flags
and
Source
VIP
Port
protocol
and
Target
Port.
B
So
for
programming
Services,
we
in
full
State
bargain.
We
just
add
all
the
services
which
are
required
to
the
div
store
after
adding
to
the
digital.
You
know.
Digital
has
a
EPA
method
for
done,
which
basically
computes
all
the
Deltas
yeah.
D
The
TF
store
is
essentially
the
thing
that
replaces
the
in-memory
model
of
the
entry
proxy
right,
because
the
entry
proxy
has
services
and
endpoints
it
Aggregates
them
into
this
one
data
structure
and
then
what
kpng
comes
with,
because
it
doesn't
do
that
for
you,
it
comes
with
a
thing
called
the
diff
store,
but
the
diff
store
does.
Is
it
allows
you
to
add
stuff
to
it
and
then
replay
it?
D
So
you
can
just
see
the
diffs
afterwards,
so
you
can
get
a
completely
stateless
represent
like
there's
no
state
in
the
model
of
what's
going
on
in
the
coupe
proxy.
Just
reads
in
the
whole
network
State
and
then
every
once
in
a
while,
you
can
just
get
the
diffs
out
of
it
in
it
is
separate
the
separate
your
call
yeah
go,
go,
go
ahead,
yeah
upper.
B
B
Coming
back
to
coming
to
this
function,
only
program
host
compute
objects.
We
iterate,
we
ask
the
digital
about
items
which
would
which
were
deleted,
so
we
Delete
all
the
load
balances
which
are
not
required.
Let
me
do
the
same
for
remote
entries
which
are
no
longer
required.
Then
we
iterate
over
all
the
items
which
are
changed
and
then
the
items
which
were
updated
and
finally,
we
check
for
the
load
balances,
which
was
either
changed
or
deleted
and
program
the
data
about
accordingly.
D
B
So
this
is
a
basically
who
mi
Port,
which
Returns
the
hostname
and
host
IP.
So
let
me
just
apply
this
manifest
so
yeah
in
my
node,
like
these
two
deployments
are
ready
now,
and
these
corresponding
services
are
also
ready.
So
if
I
go
here
in
the
Powershell
and
if
I
search
for
remote
endpoint,
so
the
final,
the
last
remote
endpoint
will
be
programmed
by
kpng
proxy,
only
we
can
simply
just
curl
it.
So
we
have
who
am
I
Windows
all
this
popular
cluster
ID
part.
B
B
B
B
D
B
And
I
can
check
try
to
hit
on
the
Linux
IP
as
well
for
the
notebook
service.
This
is
the
Linux
machine.
B
I'm
new
to
Powershell,
so
just
giving
us
the
mode
board
would
be
three
one.
Six.
Oh
one
yeah
works
for
both
the
IPS
billions
and
windows.
A
E
A
B
A
A
B
Experimental
group
in
Windows
node
what
was
that
I
think
there's
already
one
job
for
the
personal
space
back
in
in
it
falls
under
the
experimental
section,
yeah.
D
That
way,
we
don't
have
to
write
any
extra
code
for
a
new
CI
job,
the
other
one's
broken
anyways.
So
you
can't
be
any
worse
than
our
the
one
that
we
already
have
right
like
it
makes
sense,
it's
already
better
in
the
sense
that
there's
somebody
working
on
it
right
like
yeah,
true
and
then
you
don't
have
to
do
anything,
just
delete
the
old
make
your
PR
delete
the
old
code
and
then
the
CI
will
run
automatically
cool.
D
B
And
also
one
question
like
there
are
some
hard
coding,
as
in
the
host
Network
identifier
needs
to
be
hard
pulled
in
the
environment,
yeah.
A
B
I
have
to
figure
out
a
way
for
that.
Also,
the
cluster
cider
cider
will
be
already
there,
the
network
name
and
the
host
ipis
the
endpoint
IEP.
B
So
that
will
be
different
for
different
vendors
like
I
guess,
the
current
testing
is
implemented
with
kapzi,
Calico
and
again
I'm,
not
sure
which
value
I
need
to
add
over
here.
So
I
might
ping
you
guys
for
that.
Also
yeah.
A
And
I
think
I
especially
mess
with
cardio
at
the
beginning
of
or
during
the
triage
session,
two,
if
I
think
we
need
to
find
a
more
generic
way
to
handle.
This,
like
the
the
issue
is,
is
that
when,
if
the
cluster
first
comes
up,
we
need
a
way
to
talk
to
the
API
server,
but
the
in
cluster
config
and
the
environment
variables
here
point
to
like
the
local,
the
internal
service
IP,
which
we
don't
have
a
route
for.
So
we
need
to
figure
out
how
to
be
able
to
contact
the
API
server
in
previous
deployments.
A
A
Yeah
so
they're
still
they're
still
that
issue.
We
can
solve
it
the
old
way
or
figure
out
a
new
issue
or
two
but
I,
don't
think
I
should
block
any
progress
on
getting
actual
implementation
and
things
working.
D
E
A
A
E
Yeah
sorry
I
just
mean
it
the
so
yeah.
This
looks
really
good
thanks
for
working
on
this
and
I
had
a
few
questions.
So
was
this,
which
network
mode
is
this
using
now.
E
Okay
so
Ultra
Bridge
and
it
looks
like
there's.
You
know
there
should
be
a
considerable
performance
improvements
as
well
with
you
know
the
the
div
store
that
you're
using
versus
you
know
the
current
in
3
Q
proxy.
If
you've
done
any
measurements
yet
on
that
or
I.
B
D
But
let
me
talk
to
Mark
about
that
and
the
big
performance
Improvement.
There
is
just
that
it
takes
less
resources,
so
you
could
use
those
resources
to
dedicate
them
to
other
things
or
you
could
just
you
know,
rededicate
them
to
the
food
proxy,
but
the.
D
But
the
thing
there
that
we
talked
to
Mark
about
is
that
that
would
require
sort
of
changing.
If
we
were
to
go
that
to
that
extreme,
we
would
change
the
deployment
of
change.
The
way
that
Windows
thought
about
running
its
Coupe
proxy
entirely,
because
then
you
would
have
Windows
node.
That
was
really
completely
disconnected
from
the
kubernetes
API
server
entirely
right
and
then.