►
Description
Join us for Kubernetes Forums Seoul, Sydney, Bengaluru and Delhi - learn more at kubecon.io
Don't miss KubeCon + CloudNativeCon 2020 events in Amsterdam March 30 - April 2, Shanghai July 28-30 and Boston November 17-20! Learn more at kubecon.io. The conference features presentations from developers and end users of Kubernetes, Prometheus, Envoy, and all of the other CNCF-hosted projects
K8s Conformance Working Session: Supporting Multiple Environments - Dan Kohn, CNCF
The K8s Conformance program and test suite will play an integral role in ensuring that customers have a consistent experience when porting their workloads between K8s providers. However, not all environments in which K8s will be hosted will be identical - either due to hardware differences, or the use of optional features. The question before the K8s Conformance WG is how to deal with these variants with respect to the Conformance Certification program. Several options have been put forward. This session will focus on trying to develop a generic solution, while using Windows as our first concrete use-case to address. This session is not introductory and will be more of a brainstorming session for interested parties.
To learn more: https://sched.co/J06P
B
A
This
one
in
because
the
there
is
just
a
substantial
area
to
focus
on
so
I'm,
going
to
ask
Patrick
to
kick
off
with
a
little
bit
of
background
and
then
ask
Doug
to
moderate
I.
Sorry
Brad
apologies
to
moderate
the
discussion,
which
hopefully
is
not
going
to
be
too
heated,
but
the
quick
context
that
I
want
to
give
is
that
the
certified
kubernetes
program
is
literally
one
of
the
most
successful
programs
in
the
history
of
software
certification.
A
We
just
got
started
a
year
ago
we
have
the
entire
industry
79
different,
certified
implementations
out
there
and
there's
some
real
strong
guarantees,
along
with
those
on
the
api
compatibility
and
they're,
actually
getting
better
as
the
conformance
Suites
gets
gets
better
and
and
we
have
all
the
mechanisms
to
avoid
people
falling
off
the
release
train.
So
there
are
services
out
there
like
being
able
to
support
windows
containers
that
we
know
that
a
lot
of
end-users
are
genuinely
interested
to
do
and
that
really
do
expand
kubernetes
into
new
markets.
A
C
D
But
for
those
who
don't
know
me,
I'm
Brian
grant
I've
worked
on
companies
from
the
beginning.
The
stick
architecture
co-chair
Sagarika
texture
owns
the
definition
of
the
conformance
suite
the
conformance
test.
Suites
we've
I
also
participate
in
many,
the
SIG's,
especially
those
that
are
building
plug-in
mechanisms
or
I'm
co-lead
at
of
kubernetes
at
Google.
So
many
of
the
sub
teams
that
we
have
working
across
the
entire
project.
We
have
people
involved
in
most
of
the
SIG's,
even
some
of
those
things
for
other
cloud
providers.
D
You
know
we
really
want
kubernetes
to
work
in
a
consistent
way.
We
think
in
all
the
places
that
it
runs,
we
think
that's
one
of
the
things
users
value
most
about
kubernetes
is
they
can
run
the
same
workloads
in
multiple
environments
that
they
can
have
a
consistent
operational
experience
in
multiple
environments,
and
so
all
the
great
tools
people
are
building
in
the
ecosystem.
Work
in
kubernetes
anywhere
and
all
the
environments
work
over
days
runs
so
we're,
starting
with
the
API
conformance,
which
will
help.
D
But
you
know
we
also
need
to
think
about
workload
portability
as
sort
of
the
next
step
and
we
haven't
really
got
there
yet
at
the
moment
we
started
with
the
tests
we
had
and
we
identified
tests
that
we
thought
would
be
portable,
that
we
thought
would
be
exercise
features
that
weren't
really
optional,
but
as
we
you
know
we're
looking
at
the
set
of
tests
that
we
have
and
whether
those
attributes
are
really
the
case
and
what
tests
are
missing.
That
would
need
to
get
added.
D
So
we're
really
focusing
on
some
of
the
core
features
like
the
pod
tests.
Aaron
Creek
and
Berger
has
been
doing
a
lot
of
work
identifying
more
of
the
tests
that
we
already
have,
for
example,
that
we're
in
the
node
conformance
suite
and
making
sure
those
get
into
the
conformance
suite,
but
as
we
keep
the
goals
in
mind
of
compatibility
and
portability
and
that
consistent
experience,
I
think
that
really
needs
to
inform
where
we
go
with
the
conformance
and
certification
program,
a
high
level.
D
So
we
can
get
into
you
know,
profiles
that
might
be
one
way
to
solve
it.
There
might
be
other
ways
to
solve
it.
So
just
keep
in
mind
the
higher
level
objectives
of
what
we're
trying
to
do
and
we
can
hash
out
together.
What
is
the
best
way
to
achieve
that
for
users
that
will
keep
the
program
understandable.
E
All
right
thanks
so
I
started
looking
at
this,
so
I
guess:
I
should
introduced
myself,
I'm
Patrick
Lang
I'm,
one
of
the
co-chairs
for
cig
Windows,
but
where
I
got
started
going
down,
this
path
was
when
we
were
trying
to
define
what
it
would
mean
to
take
Windows
support
stable
within
the
kubernetes
community.
You
know
one
of
the
first
things
we
looked
at
as
we
said.
Well,
you
know
today,
we've
got
this.
You
know
performance
logo
program
and
all
the
services
that
we're
customers
are
going
to
want
to
deploy.
E
Windows
are
using
this
logo
today,
and
so
my
initial
assumption
was
that
well
we
needed
to
get
Windows
conforming
once
we
started
digging
into
this.
We
we
found
that
there
was
the
first
thing.
It
was
like
just
from
end-user
standpoint
discovering
what
the
conformance
tests
actually
mean
is
actually
kind
of
hard
right
now
and
so
what
it
came
down
to
is
today.
E
The
code
is
basically
the
source
of
truth,
and
so
we
went
down
the
path
of
trying
to
modify
conformance
tests
such
that
they
could
run
on
Windows
and
there's
a
couple
different
approaches
that
you
know
that
could
be
used
to
do
that.
But
one
of
the
things
we
have
to
be
really
careful
about
is
we
want
to
make
sure
we're
not
changing
the
intent.
So
that
way
it
you
know
allowing
kubernetes
to
become
either.
E
You
know
to
behave
differently
than
it
did
previously
and
so,
like
one
example
of
how
we
did
some
of
this
on
Windows
was
there's
test
cases
today.
That
will
do
things
like
LS
slash,
you
know,
secrets
and
then
see
if
I
see
if
I
could
burn
a
secret
exists,
I
think
the
intent
of
that
API
was
to
be
able
to
say:
okay,
if
you
define
a
cur
burn
a
secret,
that's
supposed
to
be
mounted
to
a
volume
in
that
container.
E
Is
it
there,
but
when
you
use
the
LS
command
and
parse
the
textual
output
of
that
now,
you
have
taken
a
dependency,
something
that
is
I
guess
Linux.
Pacific
is
not
really
the
right
word
because
that
exists
on
BSD
as
well,
but
it's
something
that
is
is
invariant
and
not
necessarily
something
that
kubernetes
is
trying
to
guarantee
that
yes,
your
LS
command
works
and
so
like.
In
this
particular
case,
we
am,
you
know:
we've
been
working
with
some
some
folks
at
cloud
base
on
this
they've
had
very
similar
experiences.
E
Even
if
we
go
down
the
path
of
trying
to
sort
of
emulate,
Linux
behavior,
if
the
operating
system
underneath
is
still
different,
that
can
lead
to
some
differences
in
how
the
tests
work
and
so
I
just
kind
of
wanted
to
call
that
out.
It's
like
one
of
the
key
differences
that
were
that
can
affect
this
testing
and
show
a
difference
between
what
the
intention
was.
But
what
the
test
case
actually
does
just.
F
F
So
as
we
write
those
comment
metadata
things,
we
were
supposed
to
figure
out
what
was
really
being
tested
versus
what
was
infrastructure
to
make
sure
the
tests
actually
worked,
and
so,
if
you
guys
continue
down
this
path,
if
you
could
help
clean
up
those
comments,
because
maybe
maybe
they,
the
original
author
of
the
comments-
didn't
understand,
LS
really
wasn't
the
point
of
the
thing
right,
so
you
can
help
clean
those
up,
that'd
be
really
helpful
to
as
you
go
through
the
process.
Oh
thank
you.
So.
G
H
H
E
F
I
was
just
saying:
if
you
notice
problems,
please
don't
feel
free
to
help
clean
up
the
comments
on
the
same
lines
with
Aaron
was
the
same.
There
need
not
be
sufficient
for
variety
of
reasons.
Getting
on
a
sidetrack
is
what
point
out
that,
as
you
come
across
those
please
help
clean
up
the
comments,
because
that's
a
very
important
aspect.
H
Want
to
underscore
just
something,
though,
because
I
think
it
oftentimes
a
lot
of
what
is
slows
us
down
when
it
comes
to
changing
the
tests
or
adding
new
tests
is
to
understand
the
intent,
and
the
idea
here
is
if
we
can
try
and
front-load
the
documentation
of
the
intent
before
we
go
down
the
tricky
path
of
all
the
details
of
writing
the
code.
I
think
that's
something
that
this
group
could
probably
do,
spend
some
time
doing
to
really
help
Fedder
eight
out
the
follow-up
work
of
flushing.
This
out.
J
E
So
so
that
was
like,
you
know
one
example
of
sort
of
some
of
the
differences
we
found
and
some
of
the
challenges
that
we're
gonna
have
to
overcome.
You
know,
as
we
were,
testing
windows
support
there,
and
so
you
know
in
this
in
the
cig
architecture
and
other
meetings,
what
we
realize
is
there's
really
sort
of
two
separate
questions
that
we
were
looking
at.
E
One
of
them
is,
is
the
windows,
quality,
good
and
conformance
doesn't
actually
have
a
way
to
say
that
today,
but
getting
tests
you
know
within
our
cig
is
one
way
that
we
could
do
that
and
so
we're
pursuing
that
in
parallel.
But
when
we
want
to
go
back
and
say
when
this
cluster
is
certified,
what
is
going
to
work
on
all
of
the
nodes
and
have
all
the
nodes
been
tested?
E
There
wasn't
a
clear
path
forward
on
that,
and,
and
so
one
of
the
proposals
that
I
think
remember
if
Brendon
originally
made
it
or
someone
was,
but
they
said
you
know,
we've
got
this
this
proposal
out
there
for
something
called
a
conformance
profile.
So
that
way,
if
there
is
something
that
is
a
feature
that
may
not
apply
to
all
clusters
but
may
apply
to
some,
because
it's
like
we're
not
saying
all
clusters
must
have
a
Windows
node
in
them.
E
We're
saying
that
if
we
want
to
add
a
Windows,
node
2
run
Windows
apps
into
an
existing
Cooper
days.
Cluster
users
can
do
that
and
nothing
breaks,
and
you
know
it's
even
better
is
if
we
can
do
that
and
say
here's
a
list
of
functionality
that
you're
going
to
get
on
Windows
that
is
consistent
with
kubernetes
and
is
going
to
work.
E
The
way
you
expect
it
and
so
I
think
that's
kind
of
the
goal
that
we're
getting
to
performance
profiles
were
a
way
that
was
proposed
to
to
do
that
and
I
think
that
there
may
be
some
other
cases
it's
being
considered
as
well.
So
I
wanted
to
see
if
lift
Lilian
wanted
to,
since
he
was
the
first
one
to
propose
conformance
profiles
to
see,
if
you
wanted
to
add
any
color
to
that.
C
Thanks
metric,
yet
I'll
give
you
a
bit
of
a
background
on
this
proposal
because
it's
kind
of
been
circulating
a
while,
as
you
can
see
from
the
date
every
year
ago.
So
when
we
first
created
the
program,
we
thought
that
we
thought
that
this
case
would
actually
arise.
We
didn't
actually
have
Windows
in
mind
in
particular,
but
we
kind
of
figured
that
there
be
some
collection
of
features,
though
that
wouldn't
fit
in
in
the
corporate
file.
C
So
we
thought
at
the
time
to
avoid
to
avoid
people
not
be
in
conforming
or
to
avoid
kind
of
having
a
whole
body
features
that
weren't
part
of
program.
We
could
have
a
profile
system
to
capture
that
and
bring
that
user
benefit
and
I.
Definitely
it
back
to
the
users
here.
So
the
goal
is
definitely
to
help
users
understand
like
where
their
things
are
going
to
be
able
to
run.
So
if
I
have
a
dotnet
application
say
you
know,
I've
bought
it
from
a
SAS
provider
and
runs
on
Windows.
C
I
would
like
to
have
a
very
easy
way
of
just
discovering
where
that
can
run,
and
so
in
theory,
and
with
the
example
of
the
Windows
potential
Windows
profile.
If
that
work,
Lloyd
had
been
certified
to
run
on
any
conformant
distributional
platform
of
kubernetes
that
had
Windows
profile,
then
it
should
run
on
any
such
platform
of
distribution
and
that's
kind
of
the
end
user
goal
here
now.
As
for
the
actual
progression
of
this
proposal,
so,
like
I
said
you
know,
we've
had
a
couple
of
versions
of
this
at
the
last
meeting,
I
think
in
Copenhagen.
C
We
basically
just
agreed
that
we
we
thought
it
was
we're,
probably
gonna
need
it,
but
we
just
kind
of
decided
to
wait
until
we
had
like
a
really
compelling
reason
to
create
the
first
one,
and
so
here
we
are
today.
I
think
that
there
is
a
very
compelling
reason
to
always
consider
that,
and
it's
gonna
be
interesting
for
me
to
see
how
we
go
with
these,
because
I
think
it's
probably
hottest
on
the
first
one,
and
then
you
know
then
the
next
one
is
we
had
a
folios
template
a
lot
easier.
Can.
J
I
say
a
few
words
yeah,
so
so,
just
if
you
don't
know
me,
I'm
Brad,
topal
and
I
lead
the
upstream
folks
who
contribute
from
IBM,
but
in
a
previous
life,
I
actually
led
the
OpenStack
interoperability
challenge
effort.
So
I
don't
know
who
you
work
with,
but
it
wasn't
me
but
I.
Let
it
I
was
a
guy
up
on
screen.
J
It
was
very
stressful,
and
so,
when
I
think
about
interoperability,
that's
kind
of
the
thing
if
somebody
is
going
to
have
a
workload
and
some
yeah
muls
and
what-have-you-
and
this
is
how
it
works
today-
the
current
state
of
the
art
you
can
take
the
yeah
Mille,
ideally
and
and
if
everybody's
running
and
they're
certified
kubernetes,
you
should
be
able
to
take
that
yeah
Mille
and
run
it
over
here,
and
it
should
work
it
should
be
portable.
Now
the
profiles
are
interesting
and
I.
Guess
the
the
thought
well,
I
thought
of
with
profiles.
J
Would
be
for
something
like
an
option
thing?
Maybe
it's
something
like
GPU
or
something
like
some
extra
feature
that
not
everybody
has
the
profile.
What
I
didn't
understand
from
what
you
described
on
the
windows
is,
let's
say,
I've
written
something
over
here
and
I
didn't
start
originally
on
on,
say
the
Windows
platform
through
the
certification.
Are
you
worried
about
that
running
on
Windows
because
it
passed
the
certification
test
and
now
can
run
on
Windows,
or
you
worried
about
having
separate
different
workloads
that
ok,
this
was
really
meant
for
Windows
and
so
it'll
run
anywhere.
J
They
have
Windows,
because
there
is
a
slight
difference
in
the
expectation
as
a
user,
as
a
user
I
wrote
my
amell's
and
I
think
oh
well,
then
I
should
be
able
to
move
it
to
a
Windows
machine.
What
you'll
find
is
on
Windows
there's
certain
features
not
supported.
You
can't
mount
single
files,
there's
some
problems
with
the
configure
it.
You
know,
config
set
cetera,
et
cetera.
So
me
as
a
user
I'm
going
to
experience
a
subtractive
thing.
I
had
some
features
in
my
game
all
originally,
when
I
moved
over
to
Windows.
J
Those
features
are
not
there.
My
thing
fails
and
I
say
Dan,
your
stuff
is
not
portable,
like
you
told
me
on
stage,
and
so
I
want
to
be
careful.
As
you
look
at
the
profiles,
the
profiles
can't
be
a
crutch
to
push
that
under
the
rug
right
either.
You're
saying
this
is
just
a
profile
for
things
that
are
Windows.
So
if
you
do
this,
just
only
expect
it
to
run
on
Windows.
Does
that
make
sense?
There's
a
subtle
difference.
There
yeah.
D
D
Should
we
should
windows
be
GA,
so
I
think
that's
more
of
a
is
windows
ready
to
be
considered
sufficiently
complete
that
it's
usable
by
users
and
have
we
adequately
documented
that
and
do
we
have
any
validation
tools
that
users
can
use
to
understand
whether
their
configuration
should
work?
If
we
view
the
conformance
program,
as
kubernetes
has
some
GA
features,
some
of
those
features
are
optional.
Some
are
not,
but
if
someone
is
using
a
kubernetes
feature
say
running
pods
on
windows
that
that
works
the
same
way
everywhere.
D
That
windows
is
supported
right,
so
I
think
you
know
that's
three
questions
that
have
come
up
for
me
that
I
think
users
would
care
about,
although
it
would
be
maybe
good
to
talk
to
some
users
if
there
are
any
here.
Is
that
question
of?
Does
it
have
the
same
behavior
everywhere
that
I
run
kubernetes,
so
I
wouldn't
expect
a
Linux
program
to
work
on
Windows
or
vice-versa,
but
if
I'm
running
windows
on
kubernetes
on
a
specific
release
of
kubernetes
I
expect
it
to
work
the
same.
D
The
number
of
features
that
I
have
to
figure
out
if
they're
present
or
not
that's
a
you
know,
I
don't
want
to
have
to
read
through
a
list
of
200
individual
features
and
understand
if
an
given
kubernetes
distribution
or
service
supports
that
set
of
features.
So
there's
kind
of
a
granularity
question
which
is
similar
to
the
whole
cloud
native
mission
right,
like
I,
don't
want
to
have
to
understand
if
all
of
these
2000
files
are
in
sync
across
all
my
nodes.
D
In
order
for
my
application
to
work,
I
want
a
smaller
surface
area
that
I
have
to
care
about
and
know
that
the
system
is
taking
care
of
everything
else.
I
think
we
have
a
similar
sort
of
issue.
So
if
someone
uses
our
back,
if
a
community
supports
our
back,
for
example,
that
it
supports
all
the
our
back
features
and
not
like
some
random
subset
of
them-
and
that
leads
to
the
third
thing,
which
is
the
cross-product
of
all
the
combination
of
features
that
might
or
might
not
be
supported
right.
D
So
it's
probably
pretty
easy
for
me
to
figure
out.
Does
this
kubernetes
support
Windows
at
all
right,
but
if
it
does,
then
I
expect
to
support
the
same
features
as
every
other
windows
at
the
same
kubernetes
release.
So
I
think
that
if
we
can
constrain
the
problem,
so
I
think
the
big
question
for
profiles
or,
however,
we
solve
this
problem
for
the
optional
features
for
the
non
ubiquitous
features.
D
That
on
portable
features
is
how
do
we
solve
those
problems
for
users
and
convey
it
in
a
comprehensible
way
that
ends
up
being
useful
so
that
you
can
say
yeah
I'm
running
the
same
thing
on
all
these
16
kubernetes
distributions
onstage
in
a
useful
way.
So
you
know,
we've
had
the
discussion
about
the
cross
product
problem
before
if
there
are
45
profiles
and
becomes
meaningless
but
but
yeah.
F
So
Brian,
you
said
something
interest
in
there.
You
said
application
intent
around
in
Windows
runs
with
the
same
on
all
Windows
environments.
Do
I
grew
that
part
of
it.
What
what
isn't
clear
to
me,
though,
is
whether
we
have
consensus
on
the
I
say,
maybe
the
harder
problem,
which
is,
should
an
application
that
can
run
in
both
places,
meaning
Linux
and
Windows,
because
either
you're
writing
in
a
language
that
doesn't
use
anything
operating
system
specific
or
through
multi
arc
images.
B
F
D
My
current
answer
is
no:
like
I
wouldn't
expect.
We
are
pushing
hard
to
make
kubernetes
work
on
arm-in-arm,
64
and
PowerPC
and
Windows,
but
so
you
I
haven't
seen
any
user
who
wants
that
degree
of
portability
or
who
expects
it
out
of
the
box.
It
really
depends
on
the
language
depends
on
how
to
build
that
application.
It
depends
on
how
they
build
the
container.
That
would
be
really
really
hard
to
enforce.
D
F
I'm
not
talking
about
how
you
create
the
image
itself.
That's
the
users
problem
I'm
talking
about
from
the
kubernetes
configuration
perspective.
Are
we
looking
for
conformance
across
platforms
at
that
point,
so
you
could
say
here's
the
Amla
file
where
the
image
itself
is
portable.
My
problem
is
a
user.
Should
the
yellow
file
be
consistent
between
the
two
worlds
or
not
a
goal
for
us
as
I'm,
trying
to
figure
out.
B
K
Hi,
my
name
is
Nick
young
I'm
I
work
at
Atlassian.
I
am
an
end-user
of
communities,
obviously,
but
of
Windows
nodes
as
well,
and
so
as
an
end
user.
The
things
I
would
want
to
see
out
of
these
things
are
number
one
you
know
like.
Can
windows
nodes
work?
Do
they
work?
Well,
that's
the
first
thing
I
care
about
the
second
one
is:
if
they
do
work
see
the
other
thing
is
we
don't
use
a
vendor
platform
at
all?
We
have
our
own
custom
cooked
cube
because
we've
been
doing
it
for
long
enough,
and
so.
K
Secondly,
the
conformance
tests
for
us
are
a
good
way
for
us
to
say,
although
we
don't,
because
it's
only
for
internal
use,
we
don't
care
about
passing
the
cert
for
be
able
to
put
a
badge
somewhere,
but
what
we
do
care
about
is
making
sure
that
our
one
can
run
stuff
that
we
want
to
run
no
matter
what
it
is.
So
the
conformance
test
or
a
good
test
for
us
to
say
hey.
K
We
can
run
the
conformance
test
against
our
platform
and
say
this
will
run
the
same
Windows
things
as
any
other
as
one
of
the
managed
ones,
and
you
know
so
for
us.
That's
that's.
The
useful
part
here
out
of
the
the
profile
is
to
be
able
to
say
for
the
a
the
supports
windows:
B,
there's
a
Windows
profile
and
to
do
the
windows
profile.
K
K
G
J
B
D
L
So
the
node
conformance,
so
there
are,
we
have
a
normal
conformance
suite
which
is
used
for
the
certification
program
and
we
also
have
a
node
conformance,
which
is
used
by
like
various
CRA
providers,
to
see
if
things
are
working
or
not
and
it's
a
validation
swing.
So
that
would
be
enough
for
you,
you
don't
need
it
to
be
part
of
the
conformance
test
program,
yeah,
the
other
one.
I
wanted
to
go
back
to
us.
This
is
not
just
a
window
specific
thing,
because
we
had
the
similar
discussion
for
the
virtual
cubelet
right.
L
J
L
J
So
that's
both
excellent
input,
and
that
gives
me
a
question
for
Dan
right.
So
so
what
what
he's
saying
is
how
do
we?
How
do
we
keep
me
as
an
end-user
from
getting
confused
between
the
two
views
of
the
world?
The
two
movies,
the
current
right?
Now
we
have
a
single
movie
to
watch,
which
is,
if
I
pass
the
certification
to
us.
If
I
run
here,
I
should
run
there.
J
I
should
not
confuse
that
to
say
well
if
I
pass
the
certification
test,
I
could
now
go
run
on
Windows
right
and
somehow
was
a
messaging
from
the
CNC
F.
We
have
to
be
very
clear
that
there's
sort
of
there's
two
sets
of
tests,
there's
something
that's
for,
say,
Windows
and
then
there
was
something
that
we
had.
That
was
first
say
certified,
get
your
sticker,
get
your
whatever
does
that
make
sense
Dan,
and
is
there
a
way
to
manage
that
from
your
perspective?
Actually.
I
I
actually
think
it's
a
little
bit
different
than
that
in
the
sense
that
what
I'm
hearing
from
the
end
users,
what
I
think
I've
heard
from
everybody,
really
is
this
idea
of
consistency
right,
and
so
it's
not
about.
What
does
the
profile
mean?
It
actually
is
about
hey
I'm,
pretty
certain
that
if
I,
if
I
look
at
a
provider
and
they
have
the
same
checkmarks
across
each
provider-
has
the
same
set
of
checkmarks.
My
workload
that
I
started
in
one
environment
will
run
in
this
other
environment.
J
I
No,
no
no
I
mean
from
cluster
to
cluster.
Actually
is
what
I'm
saying
right
so
like
I
think,
generally
speaking,
what
the
conformance
program
is
about
and
what
I'm
hearing
the
end
user
say
is
hey.
You
know
what
we
have
an
environment.
We
built
a
big
application
in
that
environment.
We
know
the
checkmarks
that
that
application-
or
we
know
that
the
checkmarks
of
that
in
conformance
then
environment
checks,
checks
check.
Now,
if
I
look
at
the
dock
and
I
say:
oh,
you
know
what
this
other
provider
checks.
B
I
B
J
I
Then
right,
but
it's,
but
it's
really
not
about
I'm,
not
gonna,
go
and
look
and
say:
oh
I
want
to
understand
what
the
conformance
test
is
actually
saying
right,
I'm
gonna,
actually
just
say
like
here
check,
marks.
They're
these
clusters
are
the
same
right.
Well,
it's
and
it's
not
even
a
thumbs-up,
some
downs
at
some
level.
It's
like
a
consistency
like
it's
just
a
good
guarantee
that
you
know.
Okay,
if
I'm
an
ISV
I
know
that
if
I
test
it
in
one
environment,
it's
gonna
run
in
everybody
else's
environment.
All.
J
J
I
But
yeah
I
would
say
I
guess,
I
think
you're
having
I
having
I,
think
you
have
it
reversed
and
I'll.
Let
Brian
talk
after
that.
I
think
you
have
it
reversed
in
the
sense
that,
like
I'm
gonna
build
an
app
I'm
gonna
make
sure
my
app
works
in
an
environment,
I'm
gonna,
look
at
the
check
marks
that
that
environment
checks
and
I'm
going
to
assume
that
my
working
app.
If
I
go
to
a
new
environment
that
checks
the
same
check
boxes,
my
working
app
will
work.
I
It
has
nothing
to
do
with
specific
behaviors
of
kubernetes
api
s
or
anything
else
like
that,
because
I
don't
actually
even
look
at
conformance
until
I
start
building
an
application
and
I
have
working
application
and
then
I'm
thinking
about
portability.
It's
not
like.
Oh
you
know
what
I
need
to
understand
a
priori,
what
each
environment
doesn't
and.
J
D
So
so
your
example,
so
you
know
again,
as
we
which
features
Windows
supports,
is
really
a
question
for
the
as
the
Windows
functionality
is
being
developed
within
the
kubernetes
project,
about
whether
we
think
it's
sufficient
and
whether
it's
sufficiently
well
validated
and
documented
for
users.
So
when
a
user
goes
to
kubernetes
and
say,
I
want
to
run
a
Windows
app,
what
should
I
expect
to
work?
That
is
really
a
concern
of
the
implementation
of
Windows
within
the
project
and
the
documentation
and
the
validation
of
that
to
end-users.
D
Now,
when
we
come
to
the
conformance
program,
Windows
does
something
maybe
lifecycle
hooks.
Maybe
they
don't
kind
of
doesn't
matter,
but
Windows
does
something
for
the
things
Windows
is
supposed
to
do.
We
have
tests
for
that
as
part
of
the
conformance
suite,
and
we
can
ensure
that
those
things
work
the
same
way
everywhere
right.
So
we
might
decide.
Yeah
windows
really
has
to
support
life
cycle
looks
to
be
GA,
and
then
we
should
make
sure
that
there
are
tests
for
that
independent
of
the
conformance
program
right
and
then
we
can
have.
D
The
question
of
well
windows
should
really
be
under
the
conformance
program,
so
we
can
ensure
portability.
So
this
feature
is
supposed
to
work
that
it
does
work
the
same
everywhere.
So
that's
really
what
I
want
to
see
so,
as
we've
been
discussing
well,
what
is
the
bar
for
Windows
at
out
whether
it's
alpha
beta
or
GA?
D
L
D
Yeah,
so
the
challenge
we
have
had
so
far
with
conditions
in
the
test
suite
is
conditions
sneaking
in
that
really
shouldn't
be
there
that
allow
people
to
bypass
parts
of
the
conformance
suite
in
different
environments,
and
then
you
don't
get
the
assurance
of
that
portability,
so
a
different
approach,
then
conditions
would
be.
You
run
all
the
tests
and
you
report
back
the
results,
and
then
you
get
a
scorecard
of
the
features
you
support
and
don't
support.
So
if
you
fail
any
test
in
a
given
suite,
you
fail,
like
you,
don't
support
that
feature.
H
Hi
introducing
myself
for
the
video
how
many
ranking
burger
I,
take,
testing
and
steering
committee
I
work
for
Google
I've
helped
drive
this
program
forward
for
a
little
bit,
maybe
to
bring
it
back
to
what
I've
heard.
Brad
say:
I'm
under
the
impression
that
one
of
our
goals
to
use
the
check
box
analogy,
or
the
scorecard
analogy
is
to
reduce
the
number
of
entries
on
that.
H
H
Up
to
you
know
a
5,000
node
cluster
is
conformance
supposed
to
certify.
All
of
that
is
it
supposed
to
certify
like
some
subset
of
that.
That
is
useful
for
the
80%
of
the
community
and
then
finally
like
how
can
we
sort
of
frame
or
draw
boxes
around
our
profiles
so
that
they're,
really
simple
and
easy
to
understand?
I?
H
Think
we've
talked
a
lot
about
the
idea
of
profiles
being
additive
only
as
much
as
possible,
and
so
when
it
comes
to
the
behaviors
that
are
defined
for
Windows,
you
know
do
we
need
to
take
a
look
at
the
behaviors
that
just
cannot
possibly
ever
work
on
Windows,
because
that's
just
the
the
facts
and
take
those
out
of
the
core
conformance
suite.
Or
do
we
have
to
say
that?
D
Yeah
I
really
want
to
keep
that
question
more
orthogonal
and
windows
itself
is
a
set
of
features,
and
we
need
to
decide
when
that
set
of
features
is
sufficiently
useful
when
so,
the
conformance
program
should
really
be
looking
at
Windows
as
a
whole.
I
wouldn't
expect
all
the
Linux
tests
to
pass
on
Windows
like
selinux
or
a
Palmer
or
whatever
is
not
gonna
work
on
windows,
so
I
do
think.
We
need
to
get
clarity
just
from
a
windows
ignoring
the
conformance
program
from
a
usability
of
windows
perspective.
D
Independent
of
the
conformance
program,
it
should
be
independent.
The
conformance
program
is
really
about
that
consistency
and
portability,
so
I'd
like
to
keep
the
focus
on
that,
and
if
we
can
look
at
how
to
structure
the
conformance
profile
conformance
program
to
reduce
the
complexity
of
the
set
of
supported
features
that
there
is
some
set
of
features
that
is
supposed
to
work.
D
That
set
has
to
work
everywhere
for
whatever
we
call
that
feature,
whether
we
call
those
profiles
or
whatever
we
can
figure
out,
but
there
will
be
a
base
profile
right
now,
almost
certainly
there
are
things
in
the
base
profile
that
we
are
going
to
say.
Well,
it's
actually
reasonable
for
a
provider
to
not
support
that
or
in
a
particular
configuration
of
their
service,
to
not
provide
that.
So
one
thing
we've
talked
about
in
Sagarika:
texture
is
privileged
operations.
D
D
I.
Think
one
bucket
is
probably
not
gonna
work.
You
know
not.
Everybody
it'll
be
a
while
before
everybody
sports
windows.
Probably
and
probably
some
people
will
never
support
Windows,
but
it
should
be
really
easy
that
it
says
look.
Windows
is
supported
or
Windows
Server
2019
is
supported.
I
know
that
on
kubernetes
1.15,
that's
the
same
at
any
provider
says
that
the
past
that
could
forbid
sweet.
K
D
H
I
do
feel
like
this
session.
Maybe
we
wanted
to
come
to
some
consensus,
maybe
on
the
windows
thing.
So
I
don't
want
to
take
us
too
far
astray
from
that,
but
I
just
had
two
other
quick
examples
of
where
I
feel
like
we
maybe
are
moving
the
bar
slightly
higher
in
terms
of
the
full
spectrum
of
clusters
that
are
supported.
One
I
think
this
group
recently
came
to
consensus
that
a
single
node
cluster
is
just
actually
a
single
node.
H
It's
not
a
cluster
that
is
capable
of
passing
conformance,
the
reason
being
that
it
doesn't
realistically
demonstrate
daemon
sets
or
the
concept
of
a
rolling
deployment
that
has
zero
downtime,
which
is
a
feature
that
an
end
user
could
reasonably
expect
out
of
their
kubernetes
cluster
I.
Think
it's
not
sufficiently
useful
right,
so
I
think
another
place
where
we
are
lacking
at
present
is
I.
H
Usually
this
means
you
need
to
either
set
up
some
sort
of
software
based
filesystem
on
your
cluster
or
you
need
some
sort
of
vendor
provided
device.
I
feel
like
we
need
to
find
some
way
to
to
quantify
or
decide
whether
or
not
we
want
the
base
conformance
profile
to
never
support
volumes
that
can
be
arbitrarily
removed
and
then
mounted
on
other
nodes
or
whether
we
think
that
is
the
minimum
bar
to
support
stateful
workloads
and
be
sufficiently
useful.
So.
H
Stateful
sets
keeping
a
user
centric
perspective,
maybe
not
supporting
all
failure
modes
and
all
error
recovery,
but
at
least
enough
to
do
the
the
basics
and
then
perhaps
there's
sort
of
an
advanced
storage
profile
or
set
of
features
above
and
beyond
that.
But
that
would
mean
that
you
can't
just
spin
up
a
cluster
out
of
the
box.
No
configuration
required.
You
would
need
to
have
some
kind
of
storage
class
that
implemented
enough
of
these
guarantees
to
be
sufficiently
useful.
C
Thanks,
just
a
couple
of
housekeeping
issues:
real
quick
in
the
middle
here
I'm
taking
some
notes
in
them
in
the
running
minutes
doc.
If
you
haven't
seen
that
I
just
emailed
it's
a
list,
so
please
add
your
name
there
to
the
attendee
list,
and
you
know
it
so
you
have
also
we
have
stickers,
I'm
gonna
put
them
by
the
projector.
I
have
like
250
and
I
want
to
see
them
all
gone.
So
please
take
a
bunch
for
your
teams.
That
was
all
I
had
just
him.
C
D
C
Channeling
next
point:
I,
don't
think
any
reasonable
user
would
expect
their
Windows
app
to
run
on
Linux
I
know
that
that
has
never
been
the
case
for
me
and
I.
Don't
he
has
change
with
kubernetes,
so
I
mean
unless
anyone
disagrees
I'm
putting
out
as
a
bolded
non-goal
that
we
all
agree
on.
So
thank
you.
I
Well,
at
least
the
volume
part,
at
least
of
all
the
volume
attached
attached
and
as
such
I
think
that's
one
of
the
ways
we
should
decide
where,
where
profiles
are
relevant,
it's
like
things
that
are
in
base
really
90
percent
of
the
user
should
be
using,
and
if
only
50
percent
of
the
users
are
using
them.
It
probably
should
be
a
profile
because
I
can
build
a
useful
cluster
for
a
broad
set
of
users.
I
D
B
A
I
do
want
to
say
from
a
marketing
or
messaging
standpoint
that
I
think
when,
like
nine
year
olds,
can
really
argue
in
detail
about
this.
They
do
understand
some
of
the
basic
concepts
of
compatibility
and
what's
available,
I
want
to
go
ahead
and
throw
out
a
proposal.
Brian
mentioned
earlier,
the
idea
of
our
back
and
restrictive
kinds
of
things:
removing
functionality
from
the
base
profile
and
the
proposal
is
that
we
don't
address
that.
Yet
we
defer
it
because
it's
a
much
more
challenging
what
I
would
call
subtractive
profiles.
A
D
A
Then
yes,
so
then
there
so
that
would
eliminate
subtractive
profiles
as
a
concept.
But
the
specific
proposal
that
I'd
like
us
to
discuss
is
the
idea
that
there
is
no
conformant
kubernetes
cluster
that
only
supports
Windows
workloads
that
it
has
to
support
the
base
profile,
which
essentially
means
Linux
workloads
and
then
that
there
is
an
additional
functionality
when
it's
requested
to
support
Windows
workloads.
A
D
M
D
And
so
far
that
has
been
my
reasoning
as
well,
at
least
as
a
starting
point,
because
right
now
every
provider
supports
Linux
and
supports
Linux.
Only
so
until
there's
broader
windows,
support,
like
you
mentioned
about
50%
number
or
something
it's
hard
to
make
it
required
and
we
could
go
make
a
Linux
optional.
But
if
Linux
and
windows
are
both
optional,
then
I'm
not
sure
what
it
means
that
it
was
a
complexity
to
every.
A
D
B
D
I
I
Have
you
know
a
runtime
profile
of
where
it
runs
and
I
think
we're
gonna
need
to
be
able
to
do
that
and
I
agree,
that's
sort
of
like
the
null
set
if
I'm
only
conform
it
to
the
base
is
sort
of
weird
but
I
think
reasonable
users
can
figure
out
that
that
is
kind
of
a
vacuous
conformant
cluster.
At
that
point,
Brendan.
L
Can
I
offer
one
suggestion,
so
can
we
start
with
adding
Windows
labels
to
the
set
of
tests
that
currently
have
the
conformance
if
they
run
on
Windows
right,
that's
step?
One
then
step
two
would
be
to
you
build
a
test
suite
that
is
similar
to
note
conformance,
but
it
is
which
can
be
used
for
validating
for
the
end-user
right.
So
that
would
be
step
two.
Then,
when
we
get
to
that
point,
then
we
can
figure
out
whether
it
needs
to
be
I.
L
J
Can
I
just
say
word
and
it
makes
a
lot
of
sense
right,
but
Dan.
Here's
where
the
challenge
is
gonna,
be
ok,
as
you
start
thinking
about
extra
little
profiles,
profile
for
stateful,
set
or
profile
for
volume
storage
or
what
have
you
the
challenge,
we're
gonna
have
and
again,
let's,
let's
go
back,
how
you
really
want
this
to
work
you
you
really
want
to
be
able
to
show
that
you
can
take
a
pretty
complex
workload.
You
could
get
16
vendors
on
stage
and
everybody
can
pretty
much
run
it
and
it's
a
huge
marketing
win
the.
J
Yeah
but
of
course
right,
but
here's
my
point
to
do
that
since,
since
you
know
you
have
folks
in
the
room
who
are
liking
to
put
things
in
the
equivalence
classes
and
put
them
in
profiles,
because
it
helps
to
keep
things
straight,
there's
gonna
be
a
burden
Dan
where
eventually
you're
gonna
have
to
decide.
Well
what
subset
of
those
things
are
the
subsets
of
those
profiles
that
I
want
to
call
my
essentially
conformance
suite
and
go
talk
to,
hopefully
lots
and
lots
of
those
people.
You
already
talked
to
seventy
folks.
J
You
have
so
many
folks
on
board
and
say
well.
Do
you
all
think
you
can
all
support
that
subset,
because
at
the
end
of
the
day,
while
it's
fun
for
us
as
engineers
to
get
everything
well
organized
and
DIMMs
likes
things?
Organized
training
likes
things.
Organized
you're,
gonna
have
to
come
back
to
that
easy
ease
of
use
level,
because
if
and
I've
done
this
before
and
I
want
to
do
it
again
because
I'm
on
too
much
blood
pressure
medicine.
J
But
whoever
is
that
person
who's
up
there,
who
has
to
have
all
16
of
those
links
running
and
they're,
going
to
all
those
vendors
they're
gonna
want
to
make
their
life
easy
and
saying
well.
Everybody's
supporting
set
of
profiles.
Does
that
make
sense
Stan,
so
whatever
that
is
I,
don't
know
what
it
is
yet,
but
but
it's
ideally
you're
gonna
want
that.
Otherwise,
it's
gonna
work
on
half
of
them
and
that
doesn't
make
for
the.
D
Demo,
so
one
thing
that
I
think
would
help
is
data.
If
we
just
had
all
the
tests
tagged
with
the
things
that
we
were
concerned
about
being
non
portable
or
optional
or
whatever
reasonably
from
the
technical
point
of
view
yeah
and
you
could
just
run
the
full
test
suite
and
get
a
scorecard
saying,
all
the
things
tagged,
cluster
Pawel,
all
the
things
or
some
of
the
things
tagged
are
backfield
or
some
of
the
things
tagged
windows
failed.
L
A
D
Practice
like
I
actually
went
through
and
looked
through
scanned
the
list
of.
However
many
there
were
certified
at
the
time
to
try
to
ascertain
whether
I
reasonably
thought
that
all
supported
multiple
nodes-
and
you
know
it
looked
like
they
all-
supported
multiple
nodes.
But
you
know
if
there
were
just
score
cards,
I
could
go,
look
at
and
say,
yeah
and
practice.
They
all
support
multiple
nodes.
That
would
remove
I.
Think
a
lot
of
this
speculation,
mmm-hmm.
B
D
This
and
we
could
have
if
we
had
a
communication
channel
where
service
providers
or
people
building
distributions
could
come
back
to
us
and
say
well,
I,
don't
think
this
really
makes
sense
to
be
part
of
the
base
profile
or
this
other
profile
or
whatever
I.
Have
this
reasonable
use
case,
then
we
can
evaluate
adjusting
those
how
we
bucket
the
feature
sets
because
I
agree.
We
want
to
try
to
make
it
simple
for
users
and
have
kind
of
as
few
buckets
as
possible
and
I.
D
J
And
then
you'll
be
able
to
draw
your
thing
around
those
ones
that
make
it
just
a
pragmatic
question
for
the
folks,
we're
gonna
say:
have
the
two
check
boxes
say:
have
a
Linux
and
a
windows.
How
would
you
implement
that
you'd
look
at
the
workload
and
then
you
would
decide
you
didn't
go.
Oh
this
one
runs
on
Windows.
Oh
this
one
runs
on
the
Linux
okay,
but
that
and
then
and
then
dance
was
will
they
always
have
to
support
Linux
or
they
could
just
say.
J
A
A
H
We
are
talking
about
subtracting,
potentially
subtracting
functionality
out
of
the
base
profile,
but
we
are
saying
that
you
cannot
certify
your
cluster
as
conformant
unless
you
support
Linux
containers
and
let
it
today,
okay
in
the
context
of
Mac
this,
it
hits
the
data
and
hits
the
most
users
like
what
are
the
workloads.
We
are
actually
running
on
all
of
our
clusters
today.
How
can
we
get
data
about
that?
H
You
know:
how
can
we
identify
the
set
of
resources
that
make
the
most
sense
to
cover
and
the
set
of
functionality
to
cover,
as
opposed
to
looking
at
it
from
what
is
the
functionality
that
the
majority
of
providers
certified
providers
implement
today?
So
we
can
move
the
bar
like?
Where
do
we
need
to
move
the
bar
to
to
support
the
workloads
that
most
people
want
to
actually
run
on?
Kubernetes
because
I
feel
like
we
haven't,
really
looked
at
it
from
that
angle.
I.
B
D
Okay,
yeah,
we
had,
in
addition
to
the
work
that
ap
actors
doing
we
have
there.
I
was
also
data
that
has
been
published
like
by
data
dog
and
other
folks
about
what
containers
people
are
running,
and
things
like
that.
So
we
do
have
some
information
about
at
a
coarse
granularity,
at
least
about
what
api's
are
in
use
and
I.
Think
coarse
granularity
is
kind
of
what
we
want
like.
N
D
But
yeah
really
in
terms
of
thinking
what
is
a
profile?
What
does
that
mean?
That's
where
mentally
I
really
think
about
you
know
we
tagged
these
tests.
We
can.
We
can
bucket
the
test
by
functionality,
to
course
out
the
the
feature
sets
and
bucket
them
and
then
really
understand
what
passes
and
what
doesn't.
H
Just
hang
on
a
sec
just
to
be
clear:
it's
not
it's
beyond
just
like
what
containers
are
running
because
I
need
to
understand.
How
is
it
that
we
have
decided
to
deploy
this
workload
so
as
to
make
that
particular
application
the
easiest
application
to
operate
on
kubernetes,
so
my
sequel,
great,
but
like
what
storage
features?
N
Maybe
that's
a
segue
back
into
my
previous
point
that
I
was
trying
to
make,
which
is
that
is
v's
are
have
their
own
set
of
goals
and
objectives
and
requiring
ISPs
to
implement
for
multiple
target
environments
is
kind
of
contrary
to
the
whole
kubernetes
conformance
program,
which
is
fundamentally
focused
on
the
portability
of
workloads,
and
so
this
kind
of
goes
back
to
what
brad
was
talking
about
fragmentation
and
making
it
more
confusing
is
contrary
to
the
goals.
So
we
should
focus
on
that.
M
So
I
I
heard
similar
stories
from
ISV
perspective,
so
I
think
we
at
this
point.
Since
we
are
still
being
storming,
we
should
not
exclude
anything
so
portability
like
we
said
like
we
want
to
exclude
that
I
think
we
should
not
exclude
it
at
the
moment
because,
as
we
hear
more
because
right
now,
we
are
just
the
providers
talking,
but
as
we
hear
more
from
ISPs
and
more
and
customers
has
but
Brian
said,
like
we'll
collect
more
data,
we
can
exclude
things
later
on
at
this
moment.
I,
don't
think
it's
early
enough
that
we
should.
M
D
So
if
you
have
a
container
that
is
built
with
multi
architecture,
support
or
multi
OS,
or
something
like
that,
that
is
a
the
support
of
that
is
a
feature
of
kubernetes
kubernetes.
Doesn't
build
containers
and
receives.
Containers
right
in
an
environment
has
particular
particular
containers
that
run
on
them.
Maiav
x86
known
as
my
ab
arm
nodes,
I'm
I,
have
went
Linux
no
Sam
I
have
window
windows
nodes,
so
we
think
about
what
does
community
support.
D
Then
we
can
think
about
which
of
those
features
are
ubiquitous
enough,
that
they're
worth
testing
in
the
conformance
suite
to
and
sure
that
they
work
the
same
everywhere.
So
when
I
say
portability
is
like
a
Linux
app
on
Windows
or
whatever
is
out
of
scope.
That's
because
we
can't
really
control
how
people
build
containers,
or
you
know,
and
the
question
of
should
Linux
be
required.
L
L
L
L
L
L
J
Just
like
to
comment
how
hard
DIMMs
works
and
how
fast
he
works
and
I've
known
him
for
a
long
time
and
it's
it's
very
impressive,
so
very
impressive
I
had
one
kind
of
side
question
the
people
in
this
room
understand
that
we
have
a
list
of.
If
you
go,
look,
there's
a
list
of
conformance
tests
and
we
know
that's
the
conformance
suite
and
then
some
are
labeled
note
conformance
and
maybe
now
some
are
going
to
be
labeled
windows
conformance
from
a
messaging
point
of
view.
J
L
So
right
now,
as
of
right
now,
the
set
of
tests
that
have
conformance
tack
do
not
have
note
conformance
correct
so,
but
what
we
are
doing
is
move
things
from
note,
conformance
to
conformance.
They
are
both
you
sure,
okay
I
thought
we
were
removing
the
note
conformance
and
making
it
conformance
when
we
moved.
Okay,
I
see,
okay,
so.
J
So
from
you
know,
just
a
to
be
able
to
communicate
to
the
world.
How
do
we
keep
the
world
sanity?
So
they
don't
do
the?
What
I'll
called
the
Brad
mistake,
where
I
get
all
tripped
up
and
say:
oh
well,
they're
building
all
these
windows
tests,
but
they're
not
automatically
putting
it
in
the
conformance
corset,
but
they
called
it
windows,
conformance
and
well.
First.
D
Of
all
we
shouldn't,
we
should
rename
that
we
shouldn't
call
them
conformance.
This
is
legacy
from
before
the
certification
program
existed.
We
should
call
them
validation,
Suites.
Thank
you.
Enza
genius
and
the
things
that
are
in
the
official
conformance
program
are
the
only
things
you
should
be
labeled
conformance
now.
In
terms
of
helping
the
user
understand,
we
may
need
to
process
the
results
of
the
test
to
give
a
clear
scorecard.
B
D
Is
we
probably
need
some
tools
to
help
users
understand
that
I?
Don't
know
it's
on
a
buoy
does
exactly
you
know.
I
do
think
it
would
be
helpful
from
that
people
work
for
the
people
working
on
the
project,
at
least
to
increase
the
number
of
categories
of
the
tests.
So,
as
we
evolve
the
set
of
tests,
we
understand
the
features
that
comprise
the
suite
and
what
the
intent
is.
So
we
can
have
reasonable
discussions
down
down.
D
D
Again,
like
a
PR
get
sent
to
add
something
in
conformance,
we
have
to
re,
explain
why
it's
not
in
the
conformance
wheat
or
you
know
something.
That's
in
the
conformance
Wheaton.
Someone
sends
a
PR
to
remove
it,
and
we
have
to
debate
about
why
that
is.
You
can
just
categorize
all
the
tests
and
then
have
a
separate
document
that
says
these
categories
are
included
and
these
are
excluded
and
why
I
think
that
would
be
beneficial
I,
don't
want
to
prevent
people
from
adding
useful
categorizations
to
the
tests.
I
think
that's
not
constructive.
So.
E
So
related
to
that
one
of
the
questions
I've
had
is
that
for
someone
who
hasn't
been
in
all
the
discussions
when
I
look
at
this
doc
today,
it's
not
clear
why
a
particular
test
was
added
in
there
and
so
I
think
that's
something
that
would
help
users
as
well
as
implementers.
A
lot
is
if
the
proposals
to
add
or
remove
test
cases
are
pros
and
not
code.
So
that
way
people
could
talk
about
the
intent
we
could
get
user
feedback
on.
D
E
L
B
H
What
is
it
like?
Wood
behind
the
arrow
or
something
like
that?
Like
I?
Don't
I,
don't
really
think
we
have
enough
resources
dedicated
to
doing
this
stuff
we're
talking
about.
It
would
be
great
if
this
work
happened.
The
work
actually
needs
to
be
staffed
by
people
and
I
have
concerns
that
perhaps
putting
that
on
somebody
who
is
going
to
be
the
one
14
release
lead.
Who
is
a
steering
committee
member
who
is
the
chair
of
sake,
testing
who
has
given
six
talks?
This
conference
may
not
guarantee
dedicated
consistent
effort
over
time.
H
I
think
one
thing
this
group
could
really
use
is
some
amount
of
project
management
with
a
backlog
that
we
can
all
agree
on.
Perhaps
some
of
us
thought
today
we
were
gonna,
go
through
that
backlog
and
groom
it
and
walk
away
with
a
consistent
agreed-upon
set
of
priorities.
I,
don't
think
that's
reasonable,
but
I
want
to
understand.
H
Who
will
do
that
and
how
we
can
rally
around
that
like
for
these
Docs,
for
example
like
how
who
are
the
appropriate
people
to
actually
go
through
and
rewrite
these
in
prose
and
then
what's
the
appropriate
effort
or
way
of
organizing
effort
to
make
this
prose
sufficiently
understood
by
more
than
just
a
select
few
people,
you
know:
how
can
we
grow
the
decision-making
authority
around
this
conformant
stuff,
conformance
stuff?
Don't
get
me
wrong?
H
L
P
Lucky
here
from
Microsoft
and
I
I,
my
interest
in
this
session
was
to
get
what
are
some
concrete
items:
I'm
working
with
Patrick,
that
we
can
at
least
agree
on
to
unblock
this
thing
and
make
some
movement.
So
there
any
motions
or
proposals.
We
can
agree
on
as
a
reasonable
set
of
next
steps.
So
Patrick
and
I
don't
leave
this
room,
not
knowing
how
Windows
can
actually
make
progress
in
the
next
release.
P
D
Conformance
is
not
blocking
Windows
anyway
that
we
do
we
decoupled
that
decision.
Well,
when
it
is
blocking
Windows,
is
a
clear
description
of
what
is
expected
to
work
and
test
results
that
are
reported
back
to
the
community
and
we're
sick
testing.
Folks
are
working
with
the
windows
folks,
it
does
need
program.
Management,
Windows
has
broad
impact
across
the
project
in
node,
in
network
and
storage
and
they've
been
making
a
valiant
effort
to
outreach
to
these
folks,
but
really
it's
a
complex,
very
broad,
cross-cutting
effort
and
impact
testing,
release,
etc.
H
Want
to
get
a
car,
we
can
do
a
PR
review.
It
live
because
maybe
you've
already
responded
to
the
PR
review.
I
did
we're
like
I
think
we
agreed.
We
were
gonna
de
couple
conformance
from
this
we're
just
going
GA
and
there
were
some
tests
that
were
like
copy
pasted
into
a
separate
directory
called
windows
and
then
some
tests
that
I
think
we're
it's
like
we
adjusted
or
refactor
to
support
new
api
and
I,
as
a
reviewer,
was
really
trying
to
get
a
handle
on
what
necessitated
this
test
to
be
forked
into
a
separate
directory.
H
N
D
H
D
Discussed
that
particular
approach
in
seg
architecture
and
I
said
yes
go.
Do
that
and
the
reason
I
said
that
is
because
I
would
rather
have
test
coverage
to
prevent
regression
and
to
at
least
validate
that
some
fraction
worked
than
to
completely
then
to
have
no
tests
on
that,
especially
if
we're
touching
conformance
test
I
want
to
be
really
careful
about
changing
the
behavior
of
conformance
tests,
specifically
with
one
of
the
networking
tests.
Some
changes
there
were
pros.
There
were
LG
TM
by
multiple
people
that
actually
changed
the
meaning
of
what
the
test
did,
but.
H
I
I'm
cool
with
all
right
I
see
a
hand
up
over
there.
I'll
head
off
I'm
cool
with
the
pragmatic
approach,
but
just
I
feel
like
often
with
a
lot
of
the
end
end
tests
there's
a
lot
of
copy
pasting
rather
than
effort
spent
into
refactoring
the
tests
for
readability.
So
it's
clear
to
people
down
the
line
who
are
just
maintaining
or
reading
the
tests,
the
intent
behind
the
test.
H
Instead,
massive
structs
are
copy,
pasted
or
massive
blocks
of
code,
a
copy
pasted
and
then
one
or
two
lines
are
removed
or
modified,
and
the
reasoning
for
that
is
lost.
So
I.
Just
wonder
if
we
want
to
be
prudent
in
this
approach
and
like
try
to
aim
to
have
to
copy
as
little
as
possible
so
that
the
rationale
for
this
is
a
little
more
clear
or
if
we're
ok,
with
copy
pasting
everything.
But
then,
if
we
are
what's
the
right
way
of
dropping
to
do's
or
issues
to
track
this
to
clean
it
up
later,.
N
So
I
was
just
gonna
jump
in
and
answer
your
question
who,
who
can
own
the
program
management
aspect
of
this,
and
we
can
put
my
name
down
to
find
someone
I
will
work
with
our
program
management
leadership
at
Google
and
find
someone
so
until
that
person
exists,
I
will
be
that
person
on
the
list.
I
would
love
to
have
participants
from
a
couple
of
other
organizations
too.
I
find
that
we
need
two
roles.
We
need
both
the
program
management
in
the
backlog,
but
we
also
need
actual
execution
against
what
we
know.
N
We
want
to
do
and
every
time
we've
sort
of
staffed
that
as
a
side
project
for
someone
at
Google
they
get
busy
and
don't
make
significant
progress
on
it,
and
this
seems
to
be
universally
true
from
others
as
well.
So
program
management
will
get
us
the
prioritized
backlog,
but
it
will
not
get
those
tests
written
and
I
think
our
contractors
are
useful
in
some
areas,
but
some
of
them
require
actual
thorough
understanding
of
the
entire
stack
and
require
some
expertise.
H
Made
some
progress
but
I
agree
that
I
think
the
approach
of
like
waiting
until
the
very
end
to
clarify
all
of
the
details
and
requirements
is
the
slowest
way
of
getting
this
done.
I
would
really
enjoy
it
if
we
found
a
way
to
front-load
as
much
of
that
as
possible
with
us.
I
feel
like
focusing
on
the
pros
angle,
would
maybe
be
a
good
way
of
starting
this
and
to
be
clear,
I'm
not
trying
to
step
back
wholly
I.
H
D
More
like
a
plea,
if
you
are
making
changes
to
conformance
tests
or
adding
or
removing
test
from
the
conformance
suite,
if
you
are
not
a
domain
expert
in
that
area,
please
go
to
the
sig
to
try
to
get
one
if
you're
changing
something
in
the
cubelet
or
you're.
Changing
something
networking
related,
because
the
tests
are
not
super
well
documented
and
don't
always
say
what
they
actually
do
in
the
intent
isn't
well
expressed.
It's
very
easy
to
make
subtle
changes
that
actually
change
the
meaning
of
what
the
test
does.
D
J
So,
for
what
I
was
hearing,
there's
a
couple
different
work
items,
there's
new
tests
that
need
to
be
added
and
then
there's
pros
and
existing
tests
that
somebody's
found
is
wrong.
Do
we
have
a
place
where
people
are
opening
up
issues
that
says
this
prose
is
wrong
because
I
know
trainee
and
myself
are
happy
to
help
Aaron
help
fix
prose
and
push
the
you
know.
Prs
we're
happy
we're
willing
to
do
that.
H
D
D
H
We
want
to
focus
on
action
items
for
Windows
before
I
hand.
The
mic
over,
like
I,
did
say
at
Shanghai
when
I
was
talking
about
this
I
will
kick
up
a
whole
bunch
of
issues.
I've
just
been
heads
down
prepping
for
this
conference
and
then
I'm
probably
going
to
pass
out
for
a
week,
but
I
will
kick
open
a
whole
bunch
of
issues
on
tests,
I
think
need
to
be
written
anti-patterns
and
all
that
stuff.
I
just
need
the
group
to
rally
around
how
we
coordinate
on
all
of
ya.
D
L
Start
adding
Windows
label
to
the
the
ones
that
have
conformance
right
now.
That
would
give
us
an
idea
of
how
much
works
and
how
much
doesn't
work
and,
and
then
that
would
be
the
first
step
then
the
second
step
would
be
to
start
adding
validation
suite
of
your
own,
where
you
can
also
use
the
Windows
label,
which
will
be
specific
to
the
Windows
and
not,
and
we
can
skip
them
on
linux
right.
L
P
I
just
want
to
validate
something
that
I'm
hearing
here.
Conformance
for
getting
Windows
to
GA
conformance
is
non-blocking.
Okay,
that's
that's!
Okay,
but
in
saying
that
there
was
also
something
you
mentioned:
we
need
to
breakout
validation
test
from
conformance
tests,
because
there
are
conformance
tests
that
we
are
using
now
that
should
be
validation
tests.
That's
separate
from
conformance
the
sum.
P
D
The
GA
question
is
I
windows.
Is
it
Windows
is
an
optional
feature
that
no
providers
support
today
right,
so
it
there's
no
reason
to
attic
and
for
to
make
it
special
and
try
to
add
it
to
conformance
right
now.
The
issue
for
windows
GA
is:
we
need
clarity
on
what
is
expected
to
work
now,
what
is
expected
to
work
ever?
D
Essentially,
we
need
validation.
That's
that's
an
open
topic
to
discuss
user
experience
is
an
open
topic
to
discuss
like
if
someone
has
a
hybrid
clusters.
How
should
that
be
done?
And,
and
there
needs
to
be
tests
of
the
functionality
that
is
supposed
to
work
today,
and
those
tests
need
to
be
green,
just
like
any
other
feature
for
any
other
release
right,
whether
it's
beta
or
GA
or
whatever,
and
and
user
and
user
documentation
has
the
114
I.
J
P
There
was
just
we
were
using
the
conformance
to
actually
deliver
those
things
like
that
was
a
mean,
a
vehicle
to
deliver
everything
that
you're
saying
documentation
tests.
So
we
were
using
that
as
our
North
Star
in
terms
of
what
we
wanted
to
get,
because
we
knew
we
would
have
to
build
it
so
I'm.
Just
just
trying
to
understand
unwrapped.
I
know
this
decision.
P
H
So
this
is
maybe
orthogonal
to
conformance,
but
as
the
114
release
lead,
I
will
happily
work
with
you
on
this
I
want
to
bird
dog
this
and
try,
and
do
this
correctly
in
one
of
the
ways
I'm
gonna
try
and
do.
This
is
to
get
documented
test
plan
and
expectations
on
documentation
written
down
as
a
list
of
requirements
not
just
for
Windows,
but
for
pretty
much
anything.
H
H
The
way
node
validation
now
is
broken
up
code
wises,
there's
a
node
directory,
there's
a
common
directory
and
then
we're
thinking
there
would
also
be
a
windows
directory,
so
the
common
directory
would
be
the
set
of
tests
that
are
expected
to
run
Linux
only,
but
also
running
the
end-to-end
tests.
Some
of
those
are
conformance.
The
note
directory
are
tests
that
are
Linux
only,
but
also
require
more
privileges
and
access
than
conformance
tests
will
allow
and
the
windows
tests
are
tests
that
are
windows
only
and
may
or
may
not
fit
conformance.
We
don't
really
care.
E
K
So
I
just
wanted
to
be
clear
on
what
I
wanted
to
get
out
of
coming
to
coming
here
today.
So,
firstly,
I
wanted
to
get
some
clarity
on
our
profiles.
A
good
idea,
I
didn't
see
much
sort
of
agreement
from
everybody
on
on
that
on
the
mailing
list,
but
it
seems
like
everyone
here
and
I
wanted
to
get
this
written
down
in
the
meeting
notes
to
everyone
here
seems
to
agree
that
having
some
sort
of
profile
mechanism
is
a
good
idea,
so
yeah
I
just
wanted
to
yeah
yeah,
but
I
think.
J
Well,
not
too
many,
but
the
the
the
thing
that
that
Brian
came
up
with
coming
up
with
the
name
of
the
validation.
Suite
makes
everybody's
life
simpler
because
everybody
gets
the
ability
to
go,
learn
and
say
well.
What
can
we
support
in
Windows?
What
can
we
do
and
you
know,
and
then
at
the
same
time
we
don't
confuse
the
world
right
so
that.
J
K
J
B
H
Am
a
simple
man
and
I
guess:
I,
don't
understand
the
difference
between
the
words
conformance
and
validation,
but
it
might
be
okay,
but
it
could
be.
It
sounds.
An
awful
lot
like
y'all
are
talking
about
a
suite
of
tests
that
could
eventually
test
that
a
CRI
implementation
meets
the
appropriate
needs
that
are
defined
by
CRI,
whereas
conformance
is
about
the
way
kubernetes
functions.
Given
that
a
CRI
thing
and
a
CNI
thing
and
a
CSI
thing
all
meet
that
validation.
Is
that
right
or
can.
B
J
N
A
D
Just
just
to
give
another
concrete
example:
the
storage
teams
and
networking
teams
have
parking
mechanisms
as
well.
Si
si
si
and
I.
You
can
have
targeted
validation
suites
for
those
plugin
points
that
don't
test
the
system
as
a
whole.
Like
we,
some
of
the
note
conformance
tests
weren't
even
into
in
tests,
so.
D
We
have
for
can
the
conformance
suite.
They
just
nest,
those
little
boxes
exactly
so
one
requirement
we
have
for
the
conformance
week
is
it
it
has
to
treat
the
cluster
as
a
black
box.
The
provider
provides
a
cluster,
and
then
you
can
test
that
that
cluster
conforms
and
the
specific
concern
for
conformance
again
is
consistency
and
portability,
and
not
that
every
specific
feature
of
some
little
interface
within
kubernetes
works.
J
So
the
reason
for
the
two
names
is
ones
like
a
shim
right.
We're
gonna
get
all
these
validation
tasks,
but
then,
as
they
get
more
mature
DIMMs
is
gonna.
Look
at
me
as
look
at
me
and
say
you
know
these
should
go
into
the
conformance
we
potentially-
and
this
gives
you
sort
of
that-
that
pattern
of
being
able
to
be
simple
to
the
outside
world
and
saying
here's
our
conformance
set
yet
being
able
to
get
the
real
work
done.
That
I'd,
like
dims,
beats
me
up
that
he
wants
to
get
done.
J
Hey
Brad
I
need
profiles.
I
need
this.
I
got
go
check
all
this
stuff
out.
I
can't
not
have
tests
for
these
things,
so
it's
kind
of
a
two-step
process,
you'll
get
your
validation,
sweets
and
then
over
time,
Brian's
gonna,
look
at
and
go.
These
are
really
good.
Let's,
let's
move
them
into
conformance.
Is
that
your
vision,
Brian
cuz,
that's
yeah,
and
that's
why
you've
got
the
two
names.
D
D
Conformance
tests
sweet
has
an
additional
requirement
that
the
cluster
has
to
be
treated
as
a
black
box.
I
don't
want
to
have
tests
of
every
single
feature
of
kubernetes
that
are
both
black
box
into
end
kind
of
tests
and
that
our
unit
tests
right,
that's
gonna,
be
super
complex,
so
we
actually
need
some
serious
engineering
into
creating
a
test
framework
that
will
allow
us
to
run
tests
in
multiple
modes
at
the
end-to-end
tests
are
super
expensive
trust
me.
We
pay
the
checks.
D
I
know
they
are
super
flaky
they're
slow,
provides
a
real
impediment
to
contributors
to
the
project
to
actually
develop
their
features.
To
have
to
run.
Everything
is
in
twin
tests,
so
it'd
be
awesome.
If
someone
could
put
some
some
thought
and
energy
into
that
area
where
it's
going
to
require
some
creativity
sounds
like
a
really
fun
interesting
projects,
so
I'll
just
throw
that
out
there
as
sort
of
a
long-term
kind
of
maintenance
challenge.
We're
gonna
have
all.
C
C
C
E
C
J
One
thing
is:
eventually:
you
got
to
pull
Dan
back
in
how
you
mark
it
the
profiles
and
explained
it
to
everybody.
So
what
do
we
want
to
call
it
call
it
validation
for
now,
and
eventually
it'll
be
maybe
called
you
know,
conforms
profiles.
What
have
you,
but,
but
just
let's
not
worry
about
that
until
we
can
figure
it
out,
we've
got
all
these
other
steps.
J
D
I
think
it's
worth
doing
that
in
parallel,
because
we
know
we're
gonna
need
it,
so
we
need
to
figure
out.
Do
we
want
to
call
profile?
Do
we're
gonna?
Call
it
something
else?
How
do
how
do
we
want
to
report
it
back
to
users?
Do
we
take
the
scorecard
approach,
badge
yeah,
exactly
so
I
think
we
should
start
sorting
that
out.
D
That's
a
problem
I
posit
anyway,
so
we
should
think
about
what
what
can
we
do
to
grow
the
bubble
of
the
things
that
are
supported
and
we're,
starting
with
the
most
core
things
like
pod
features?
You
know
your
commit
environment
variables
should
work
or
whatever,
but
once
we
get
past
that
initial
hump
we'll
want
to
expand
the
set
to
other
things.