►
From YouTube: Kubernetes SIG Windows 20200922
Description
Kubernetes SIG Windows 20200922
A
A
A
All
right,
it's
four
after
by
my
clock,
so
we'll
start
hello,
everybody
and
welcome
to
the
september
22nd
2020
sig
windows
meeting.
As
always,
please
be
aware,
this
meeting
is
being
recorded
and
please
abide
by
all
of
the
cncf
code
of
conduct
rules.
A
A
I
believe
we
have
a
couple
of
enhancements
that
are
trying
to
progress
this
release
and
the
new
one
for
privilege,
containers
that
we're
trying
to
introduce
please
reach
out
to
anybody
if
you
are
having
issues
finding
reviewers
or
kind
of
need,
help
moving
that
along
all
right,
and
the
next
agenda
item
is
kind
of
a
follow-up
for
this
week
or
from
last
week
about
docker
shim.
A
A
Currently
it's
compiled
into
the
cubelet,
where
container
calls
talk
to
docker
over
the
docker
api
directly
and
that
there's
been
a
lot
of
workarounds
in
kind
of
the
docker
shim
code
path
and
the
cubelet
code
path
in
general.
Some
of
them
are
for
windows.
Some
of
them
are
actually
for
other
kind
of
architectures
of
projects
and
because
of
that,
there's
it's
been
kind
of
identified
as
hard
to
maintain.
A
So
the
current
proposal
is
to
take
that
docker
ship
code,
and
I
think
the
first
steps
were
just
to
have
build
flags
in
kubernetes
to
not
build
it
and,
I
believe,
there's
a
docker-less
environment
variable
you
can
set
when
you're
building
the
cubelet
and
that
will
cause
the
docker
shim
to
code
to
not
get
built
in
favor
of
using
cri
code.
Pass
cri
stands
for
the
container
runtime
interface
and
it
was
meant
to
be
a
more
general
and
extensible
way
for
different
container
runtimes
to
plug
into
kubernetes.
A
Some
of
the
issues
that
we
have
specifically
with
windows
and
the
cri
and
docker
shim
are
that
well
there's
a
couple
things.
One
is
that
the
docker
itself-
and
this
is
because
so
docker
uses
the
mobi
moby-
has
its
kind
of
container
runtime
platform
under
the
hood,
and
the
mobi
platform
only
currently
supports
running
talking
to
windows
over
the
hcs
v1
protocol,
which
is
kind
of
old
and
outdated.
A
I
was
looking
and
there
are
some
there.
There
was
an
experimental
pr
that
was
opened
in
the
moby
platform
to
use
container
d
as
the
kind
of
the
runtime
platform
in
moby,
but
that
was
probably
over.
A
I
think
that
was
probably
about
18
months
ago,
and
that
was
when
container
d
was
very,
very
kind
of
early
development
for
windows,
and
that
has
been
completely
kind
of
untested
and
unmaintained.
So
there's
still
some
kind
of
unknown
or
ambiguity
around.
If
that
will
work,
but
so,
more
importantly,
the
way
that
you
configure
the
cube,
what
to
use
either
docker,
shame
or
a
different
container
runtime
is
with
a
fl,
a
docker,
a
cubelet
flag
called
container
runtime,
and
that
has
two
values.
A
Those
two
values
are
docker
and
remote
and
if
you
specify
the
remote
and
point
you're
expected
to
give
it
a
grpc
address
for
or
just
an
address
for
how
docker
or
how
the
keyboard
can
talk
to
the
container
runtime
and
that's
and
that
kind
of
triggers
the
cri
code
pass
with
windows
in
particular,
there's
a
number
of
switches
in
the
cubelet
code
base,
both
in
docker
shim
and
outside
of
docker
shim.
That
check
to
see
if
the
container
run
time.
A
But
if
that
variable
is
set
to
docker
or
not,
and
if
it
is
such
a
docker,
it
triggers
some
docker
specific
behavior,
which
was
needed
to
kind
of
help
kind
of
bootstrap.
The
windows
containers
in
general,
in
addition
to
some
of
the
end
most
of
those
kind
of
tweaks,
are
that
I
was
able
to
see,
are
just
related
to
a
path
kind
of
path,
massaging
either
replacing
backslashes
with
forward
slashes,
vice
versa
or
appending
the
a
volume
letter
to
those
pads.
A
So
a
lot
of,
I
think
it's
expected
that
those
or
well
back
up
a
second
so
there's
a
lot
of
kind
of
path.
Massaging
that
happens
in
order
for
the
cubelet
to
quickly
communicate
with
docker
over
the
docker
api.
A
In
addition
to
some
of
that
path,
massaging
there's
some
of
the
work
that
we
had
to
do
to
fix
an
issue
with
the
stats.
Timeout
was
actually
just
to
short-circuit.
A
The
stats
call
in
in
docker
share
with
hcs
shim
calls
directly,
and
that
was
done
because
at
the
time-
and
I
believe
this
is
still
the
case-
the
docker
api
to
get
the
metrics
endpoints
had
a
one
second
timeout
per
container,
and
the
intention
was
that
it
would
return
two
sets
of
metrics
the
first
with
with
the
timestamp
in
between,
so
that
users
could
calculate
kind
of
the
differentials
in
that
timeout.
A
A
The
the
default
timeout
for
scraping
the
metric
send
points,
and
that
would
result
in
a
number
of
things
like
cube,
control
top
to
stop
reporting.
Metrics
for
windows,
notes
and
also
it
would
prevent
a
horizontal
pot,
auto
scaling
from
working,
which
is
a
kind
of
a
big
draw
for
kubernetes
and
you
in
the
use
cases
there.
A
So
the
in
addition
to
all
of
that.
So
those
are
some
reasons
why
kind
of
windows
in
particular,
is
interested
in
this
docker
shim
deprecation
discussion
is
that
currently
container
d
is
the
only
runtime
that
kind
of
is
supports,
being
run
over
the
cri
interface
for
and
has
windows
support,
and
this
is
still
kind
of
an
experimental
feature
for
kubernetes
we're
saying
it.
A
We've
been
working
in
118
and
119
to
stabilize
this
and
hope
to
go
stable
in
120,
but
there
are,
we
just
haven't
had
quite
the
number
of
like
scale
and
perf
testing
and
kind
of
enterprise
usage,
as
we
have
with
docker,
so
we're
being
a
little
bit
conservative
there.
A
A
So
that
made
sense
and
also
say,
cluster
life
cycle
life
cycle
expressed
some
concern
for
updates,
because
right
now,
the
sierra
it's
not
really
documented
how
to
uptake
an
existing
cluster,
that's
using
one
cri
or
or
docker,
and
upgrade
it
to
use
a
different
one.
It
was
always
kind
of
assumed
that
nodes
and
node
builds
would
just
be
created
with
a
single
container
runtime
and
and
that's
it.
A
So
that's
some
of
the
issues
that
say
cluster
lifecycle
on
other
sigs
outside
of
windows
kind
of
raised.
With
some
of
this
that's
kind
of
some
of
that
information
here
I
will
open
it
back
up
to
the
floor.
If
there's
any
other
questions
for
this
meeting.
A
Yes,
we
are
planning
on,
as
in
in
120,
we're
planning
on
greatly
kind
of
improving
the
documentation
around
configuring,
the
cri
or
configuring
container
d
on
windows,
nodes
and
marcus's
question
was
if
there
could
be
documentation
about
how
to
configure
an
environment
with
continuity
and
and
other
cni's
like.
A
A
And
nothing
from
me
yeah
all
right.
I
guess
that's
it,
for
this
week's
sick
windows
meeting
see
y'all
next
week
same
time.
Thank
you.
Thanks.