►
From YouTube: Kubernetes SIG Windows 20190205
Description
Kubernetes SIG Windows 20190205
A
Alright,
alright,
hello,
everybody
and
welcome
to
another
attic
windows
Meetup.
Thank
you
all
for
attending.
So
there's
a
couple
of
things
from
an
organizational
standpoint.
So
now
we
have
a
weekly
engineering,
stand-up
meeting
that
that's
happening
every
Thursday,
if
you
guys,
saw
my
email
to
the
secundus
Elias
last
week,
I
was
asking
for
folks
who
are
willing
to
come
in
and
how
booked
tickets
well,
that's
testing
them,
creating
bug,
fixes
or
anything
else
as
necessary
for
us
to
get
to
GA.
A
The
second
item
is,
you
know,
you've
seen
that
we
change
the
zoom
link,
you
guys
are
wondering
why
we
had
to
do
it
so
that
from
the
government
stamp
only
the
kubernetes
project
so
that
we
can
better
control
the
meeting
and
some
of
the
moderation
controls.
These
type
of
meetings
are
better.
So
so
that's
why
we
did
all
right.
A
The
rancher
experimental
windows
support
demo
has
been
pushed
again,
so
Jason
is
not
available
this
week,
so
we
won't
do
that.
So
so,
let's,
let's
get
into
our
own
agenda,
so
the
first
one
is
deep
or
your
online
see
dibs
here,
ok,
deep!
You
want
to
talk
about
the
PR
for
the
GMs,
a
I
know
you
put
that
into
there.
A
B
Jon
submitted
the
first
version
of
the
TR
it's
out
there
for
review.
This
mainly
covers
the
entry
items
because
we
figured
like
that
would
be
the
first
thing
that
we
need
to
get
in.
So
this
is
basically
the
changes
in
the
cubelet
and
the
darker
shape
for
the
Alpha,
and
one
of
the
other
things
we
need
is.
We
were
thinking
of
creating
a
repo
under
kubernetes
eggs.
That'll
hold
us
here,
dear
the
web,
hook
that
we
had
planned
for
alpha
or
fall
overall
implementation.
B
C
I,
don't
think
sig
architecture
strictly
needs
to
be
involved.
They've
got
a
process,
you
can.
You
can
follow,
I.
Think
it's
under
the
kubernetes
SIG's
requests
repo,
but
I'll
find
I'll,
find
a
link
and
pass
it
on
to
you.
But
basically
you
can
file
an
issue.
Saying
hey.
You
know
we
want
to.
We
want
a
new,
a
new
repo,
here's,
an
owner's
file
and
then
they'll
create
it
for
you.
A
A
C
So
that's
an
infield
and
we
need
to
string
for
Windows,
and
so
since
you
can't
change
a
type
of
an
existing
field
without
breaking
api's
I'm,
proposing
that
we
just
go
ahead
and
add,
add
a
new
field
so
initially
I
that
I
opened
PR,
seven,
nine
nine,
just
you
know
more
more
or
less
as
a
place
to
capture
all
the
thoughts
around
this
and
you
know
Brian
was
saying
well.
This
should
probably
be
a
separate
cap
since
it's
us,
it's
an
API
change
and
then.
A
C
A
Right
so
it
has
to
be
a
local
account
in
the
container
and
that
local
account
has
no
access
outside
of
the
container.
Obviously,
because
it's
I
mean
that's,
why
GM
si
is
around
so
I
guess
from
a
from
a
functional
standpoint
in
Linux
understand
why
that's
important,
but
in
Windows
that
account
is
meaningless
beyond
inside
the
container
running
as
a
local
user
called
foo
that
that
you
know
that
you
could
configure
yourself
when
you
package
your
container
or
when
you
write
your
app
anyway.
Yeah.
C
And
so
the
only
you
know
real
use
case
I
see
here
is
that
if
I
had
an
application
and
I
wanted
to
create
a
separate
user
that
was
had
permissions
to,
for
example,
read
logs
but
not
write
logs
or
not
access
other
files,
I
could
potentially
try
to
create
a
user
that
was
locked
down
and
had
permissions
only
on
a
specific
folder
inside
there,
and
this
couldn't
be
a
volume.
It
would
have
to
be
a
folder
actually
stored
in
the
container.
Then
I
can
run
another
process
as
that
restricted
user.
A
C
A
So
when
you
instantiate
your
container,
you
will
basically
have
a
run
as
a
specific
user
name.
Let's
call
it
foo
and
then
your
application
will
detect
that
it's
running
as
I
use
a
fool
and
we'll
go
and
set
up
folder
permissions
for
foo,
but
that's
the
only
account
inside
that
container,
or
at
least
its
account
that
your
process
is
running
yes,
yeah.
C
A
Okay,
to
say
that
we
push
this
for
further
max
release.
If,
if
a
specific
customer
really
wanted
this,
they
could
achieve
it
a
little
bit
differently
right,
so
they
could
pre
create
the
accounts
in
the
container.
They
can
run
their
app
as
a
service
using
one
of
those
accounts
and
they
can
control
the
permissions.
Our
way
I
know
it's
funky
and
you
can't
control
it
from
outside
the
quintero
or
you
can
say,
I
want
to
instantiate
my
container
using
the
user
or
using
the
user
bar
that
you
pre
created.
C
B
One
scenario
that
comes
to
mind
that
somewhat
related
is
around
like
SMB
file,
shares
I,
think
the
mount
sometimes
configure
it's
like
a
UID
or
GID
and
I've
seen
some
instances
where
that
tie-up
kind
of
helps.
Like
you
know,
the
user
name
needs
to
have
a
specific
UID,
so
they
can
access
some
files
and
do
the
right
things
beyond
that.
I
haven't
seen
anything
okay.
C
Yeah,
the
SMB
case
is
not
going
to
be
a
problem
on
Windows,
because
the
permissions
that
are
the
that
are
used
are
basically
for
the
container
as
a
whole.
It
doesn't
matter
which
user
you
use
inside
you,
I
DG
ID
mask
for
that.
But
I
know
exactly
what
you're
talking
about
on
Linux,
because
when
you
do
it
SMB
mount
on
Linux.
You
have
to
make
up
a
UID
because
lesson
B.
A
C
E
A
A
I'll
talk
a
little
bit
about
me
based
online
as
well,
so
so
we
had
a
chat
with
C
cluster
lifecycle
about
you
know
what
do
we
have
to
do
to
enable
heterogeneous
clusters?
So
that
means
you
have
compute
nodes
from
both
Windows
and
Linux,
that
work
together
and
one
of
the
biggest
concerns
and
and
Tim
Sinclair
and
me
and
LeBron
we
have
talked-
is
around
the
concept
of
test
cases.
A
We
need
good
test
cases
that
will
validate
scenarios
around
heterogeneous
clusters
and
and
basically
loop
Amir
has
volunteered
to
come
on
in
into
our
sig
and
help
us
evaluate
existing
test
cases
and
identify
what
do
we
need
to
do
to
make
that
a
reality
I'll
over
me,
you
want
to
talk
a
little
bit
about
that
effort
that
you're
gonna
start
spearheading
on
that
area.
Yeah.
F
The
problem
is,
we
don't
have
also
any
tools
that
support
Windows
correctly,
so
I
see
you
guys
started
using
cube
up,
which
is
basically
so
to
say
that
cube
up
is
deprecated
this
another
statement,
but
Google
still
using
it
and
I
think
it's
sufficient
for
you
guys
to
use
Cuba
to
going
GA.
So
it's
fine,
but
in
the
long
run
we
might
decide
to
like.
F
With
help
from
me
and
other
people
sequester's
like
me,
we
might
want
to
get
cube,
ATM
more
kima
windows
again
and
then,
after
that
we
might
basically
claim
better
support
for
like
kubernetes
and
windows
in
general,
because
q-pop
is
really
a
it
might
go
away
like
next
cycle.
You
never
know
Google
might
decide
to
remove
it
and
for
test
signal.
We
we
definitely
want
something
else.
G
A
I
mean
I
personally
agree
with
the
effort
to
go
down
a
path
of
deployment
and
deploy,
or
that's
accepted
by
the
cluster
lifecycle
and
that's
gonna,
be
the
long-running
support.
Adoption
I,
don't
know,
I,
don't
have
any
details
about
work
you
Bob
may
or
may
not
be
in
the
future.
So
you
have
a
lot
more
context
from
this
than
us,
so
trust
your
judgment
there
Patrick
do
you
have
any
thoughts
on
that?
A
C
H
So
we
introduced
Windows
support
in
Cuba,
which
is
in
in
the
tree
under
the
cluster
directory,
because
it
was
basically
expedient
for
us,
so
we
have
went
support
there.
We've
been
testing
a
version
of
what
we
just
pushed
into
the
upstream
repository
for
a
couple
months
now,
and
this
week
I'll
probably
be
submitting
a
new
test
job
that
targets
exactly
what
is
there
in
the
tree
and
that's
what
we
will
be
using
by
testing
on
GCE
for
the
foreseeable
future,
but
that
continuous
testing
that's
running
on
Azure
does
not
use
this
at
all.
F
D
F
Icy
so
yeah
the
skeleton
provider
is
fine
here,
I
guess
so.
I'm
sorry
I
was
confused.
I
was
thinking
that
the
drip
test
is
also
used
for
the
Windows
test.
I
was
I,
guess
I
was
mixing
the
effort
between
Google
and
the
sig
windows
in
this
case,
but
yeah.
So
we
don't
have
any
demands.
That's
the
good
news
from
Seaquest
lifecycle.
F
As
long
as
you
have
good
signal,
which
I
guess
sufficient
for
GA
and
maybe
in
the
future
Matt
next
cycle,
we
might
combine
some
efforts
to
get
some
of
our
own
tools
working
on
Windows
I,
basically,
the
other
day
I
right,
running
Hubei
TM
on
Windows
Server,
2000
99
a
failed
at
the
couplet
stage.
I
mean
Pro
people
like
Patrick
and
solve
this
with
a
slap
on
a
finger,
but
I
I
didn't
have
the
time
to
dig
into
the
problem.
G
Looked
at
trying
to
get
cube
idiom
to
work
and
last
time
I
tried
I
got
further
than
before.
I
didn't
take
very
good
notes,
and
so,
like
I,
can't
remember
what
hung
me
up
in
the
end,
but
using
some
of
the
override
functionality.
That's
now
available,
I
was
able
to
get
around
some
of
it,
and
some
of
it
was
just
like
you
need
to
prepare
the
Windows
VM
in
the
right
way,
but
yeah
I'd
be
happy
to
look
at
it
again
in
the
future.
Yeah.
F
Definitely
not
for
this
cycle.
Well,
we
basically
make
up
a
bunch
of
assumption
in
qadian
that
you
are
running
on
linux.
Note
like
secret
drivers
and
stuff
which
are
completely
irrelevant
for
Windows
and
yeah.
We
basically
need
to
reevaluate
like
the
abstraction
between
the
operating
systems.
It's
pretty
much
doable,
yeah.
A
I
mean
going
back
a
year
ago,
we
actually
did
get
cube
Adium
to
work
for
Windows,
but
things
have
changed
and
obviously
we'll
have
not
maintained
it
as
I
think
so
we
will
kind
of
let
that
slide
and
obviously,
since
we'll
be
introduced,
as
as
part
of
that
board
regressions
were
not
being
cut
and
what,
in
the
state
that
we
are
in
right
now,
Lumi
ooh.
The
second
part
of
my
discussion
with
Tim
Tim,
and
you
was
around
testing
and
maybe
identifying
some
test
cases.
A
A
sick
class
like
psycho,
has
that
you
could
potentially
adapt
and
and
test
windows
in
a
heterogeneous
class.
The
way
where
you
have
compute
nodes
from
both
operating
systems,
as
well
as
apps
running
on
both
on
both
types
of
nodes,
is
a
something
that
you
might
be
able
to
come
in
and
hop
and
identify
some
of
those
test
cases
and
see
booking
with
emotional
windows.
F
So
I
was
thinking
that
you
guys
already
already
doing
that
because
you
saw
the
the
control
plane
also
limits.
The
worker
mouse
can
be
Linux
and
Windows
to
my
understanding
and
you
already
have
in
a
way
a
heterogeneous
coaster.
So
now
it's
sympathy's
coasters
are
passing
conformance
tests.
Then
we
pretty
much
good
to
go.
A
A
I'm
thinking
of
the
other
cluster
master
nodes
are
on
Linux.
You
have
compute
nodes
in
Linux
and
compute
nodes
in
Windows
and
your
laps
from
both
to
operating
systems
and
you're,
publishing
them,
and
you
have
a
to
cluster,
that's
being
used
by
that
all
of
their
nodes
are
being
used
by
both
windows,
as
well
as
little
jobs.
A
Notes,
electors
and
things
in
place.
Obviously,
but
for
example,
you
publish,
you
could
have
an
app
that
has
components
on
Linux
and
components
or
Windows
and
those
components
can
only
be
deployed
on
on
their
respective
OSS,
but
you
have
more
running
at
the
same
time.
On
the
same
cluster,
we
dress,
not
you
yeah,.
C
And
I
think
I
think
we
are
gonna,
see
that
some
and
I
mean
the
pattern,
but
I'm
expecting
you
said
you're
the
Linux
or
the
Windows
app
would
be
published
as
a
service,
and
then
you
may
have
other
ones
referencing
it.
So
it
would
be
more
case
of
you
know
if
you've
got
an
existing
app
that
started
on
Windows
and
you
had
you
know
a
data
access
layer
in
there,
but
you
had
a
new
project,
that's
someone,
you
know
decided
to
use,
Linux
and
I,
don't
know
note
or
something
like
that.
C
C
Another
case
would
be
possibly
you
know.
People
wanted
to
run
the
new
and
a
sequel
2019.
They
do
have
templates
out
there
now
in
an
operator
for
managing
that.
So
if
someone
wanted
to
consolidate
their
date
or
date
onto
linux
nodes-
and
that
might
be
an
option
as
well-
maybe
they
use
a
combination
of
MS,
sequel
plus
some
other
stuff
running
under
kubernetes.
But
that's
not
really
my
area
of
expertise.
A
D
B
D
D
Images
that
pods
are
spawned
on
the
correct
node,
for
example,
one
of
the
limitation
is
that
we
need
to
specify,
as
a
beginning,
assess
to
use
specific
registries
from
where
to
pull
the
windows
images
right.
So
if
we
have
a
hybrid
class,
sir
there's
no
way
of
telling
it
at
this
point,
there's
no
way
of
telling
you
hey.
You
put
this
images
here
and
up
with
this
pods
here,
because
they're
for
Linux
suppose
those
ones
under
the
window
zone
example
indicating
best
don't
work
on
Windows.
D
F
Okay,
I'm
just
trying
to
figure
out
if
you
have
a
framework
problem,
because
I
know
that
that
framework
is
not
ideal,
like
the
testing
framework
in
kubernetes,
I
mean
the
KK
repo,
it
has
problems,
and
so
I
basically
try
to
evaluate.
If
this
is
like,
it's
definitely
not
a
bokken
4g8,
because
if
we
have
problems
with
the
framework,
how
can
we
guarantee
a
hit
organist
question
in
the
first
place?
So
maybe
we
can,
if
somebody
asked
why,
why?
Why
don't
we
have
the
chiquita
or
give
us
a
question
for
testing?
F
You
can
just
say
we,
because
we
have
problems
in
the
testing
framework.
It
needs
extensions,
and
also
minded
thing
is
that
we
have
problem
with
the
profiles
for
like
the
labels
for
Windows
tests,
because
we
currently
enabled
only
looks
only,
but
you
cannot
use
that
to
test
Linux
nodes
in
a
way
right,
because
if
you,
if
you
enable
demux
only
this
is
going
to
run
for
the
whole
test,
and
you
cannot
distribute
the
test
between
Linux
and
Windows
loans
because
of
this
label.
I
think.
D
Basically,
you
cannot
distribute
a
node
between
Linux
and
Windows.
If
you
have
this,
both
nodes
in
the
cluster
there's
no
way
in
the
family
to
ensure
hey,
run
this
test
on
this
node
or
that
from
viscus
on
the
windows
node
and
that
from
the
limo.
So,
for
example,
you
can
have
situations,
I,
don't
know
I'm
just
getting
here,
but
you
can
have
situation
where
very
few
tests
are
actually
run
on
the
Windows
environment.
C
A
So
essentially
the
well
you
it
is
possibly
two
different
runs
right.
One
run
with
only
Linux
nodes
and
the
limits
only
obviously
wouldn't
really
be
affected.
Then
you
can
have
Iran
with
only
Windows
compute
nodes
only
and
you
could
avoid
all
tests
that
have
Linux
only
and
run
everything
else
and
the
new
Koran,
a
third
test,
warehouse,
Windows
and
Linux
computes
a
detained
on
the
windows
notes
and
then,
essentially
all
your
tests
should
be
passing
and
that's
the
scenario
that
Patrick's
I
write.
A
The
pensive
windows
does
not
affect
anything
else,
because
nothing
should
be
scheduled
with
them,
because
you
haven't
modified
in
pot
specs
to
our
toleration.
So
the
princes
of
the
windows
notes
should
then
fail.
Maybe
there
are
some
of
the
I
don't
know.
I,
don't
know
that
question
answer
to
this.
If
attendees
present
would
that
be
one
said
still
go
on
the
note?
A
F
A
A
F
E
A
And
and
we'll
chat
more
about
these
things
in
our
engineering
standard,
basically
uncover
more
issues.
So
in
the
interest
of
time,
let's
jump
on
the
last
issue
really
quickly
Patrick
kind
of
flag.
This
a
couple
of
times
with
some
of
the
AWS
networking
issues
and
breakage
that
exists
on
the
networking
side
for
sick
windows,
particularly
about
that.
C
Yeah
I,
don't
really
have
much
much
to
say
here
other
than
we've
seen
a
few
reports
from
AWS,
but
I
don't
have
any
developers
that
can
build
and
test
on
AWS,
and
so
if
we
can
get
some
customers
to
your
depayne
AWS
and
see
if
they
can
join
sig
windows
and
help
debug,
this
stuff
I
think
that
would
be
a
big
help.
But
but
otherwise
I
don't
have
anybody.
I
mean
I'm.
Hardly
yet,
as
your
team
and
I
can't
assign
people
to
go
fix
bugs
on
AWS.
No
absolutely.
A
A
I'll,
try
it
many
of
you.
Have
it
it's
not
alright,
we'll
take
it
up.
What
other
time
guys!
Thank
you
all
for
attending,
we'll
see
you
guys
next
week
and
now
that
you
have
two
to
a
week,
especially
with
our
engineering
stand
up
on
Thursdays,
we'll
be
able
to
cover
a
lot
more
ground
unless
someone
has
an
urgent
question
have
a
great
day
all
right.
Thank
you.