►
From YouTube: 20200922 SIG Arch Conformance
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
Good
morning
and
good
afternoon,
everybody,
this
is
hippie
hacker
and
I'll,
be
your
host
today
for
the
sig
architecture,
conformance
subproject
meeting
this
meeting
is
governed
by
the
cncf
and
kubernetes
code
of
conduct
so
be
kind
and
use
kind.
A
A
They
are
to
build
the
audit
logger
and
the
snoop
db
images,
and
we
also
have
an
app,
but
we
need
the
snoop
db
images
in
order
to
create
our
job,
and
so
I
can
drop
this
into
our
chat
area
as
well.
A
I
want
to
follow
along
at
home,
the
hook
seems
to
be
missing
from
the
cncf
api
snoop
board
into
kate's,
dot,
crowd,
dot,
io
and
I'm
working
with
the
sig
testing
on
call
to
try
to
make
sure
that
the
people
with
administrative
privileges
on
the
pro
cluster
are
able
to
have
temporarily
administrative
privileges
on
the
cncf
aka
snoop.
Repo
and
I
granted
that
yesterday
so
that
looks
like
we're
getting
really
close
for
this
to
merge,
also
drop
this
into
our
chat.
This
is
a
goal.
We've
been
looking
forward
to
for
a
really
long
time.
A
This
conformance
gate
for
this.
This
proud
job
is
intended
to
be
a
release
blocking
job
so
rather
than
informing
on
a
pr
when
it
adds
new
unconformity
apis
we're
going
to
go
ahead
and
let
stuff
merge,
but
we're
going
to
notify
sig
release
and
say
it's
part
of
our
policy
to
not
allow
this
to
be
part
of
the
release.
Please
coordinate
and
ensure
that
we
don't
use
any,
don't
allow
any
ga
endpoints
that
don't
have
conformance
tests
and
we're
hoping
that
this
will
merge
soon.
A
I
think
if
we
go
down
to
the
bottom,
we'll
see
that
our
tests
are
clean,
we're
just
waiting
for
the
approved
and
lgtm
labels
and
I'll
open
up
the
floor
at
this
point,
because
this
is
a
really
important
issue
and
I
want
to
make
sure
that
we
are
asking
the
right
people
for
the
lgtm
and
the
and
and
also
that,
the
next
step
being
to
work
with
sig
release
to
make
sure
the
processes
are
filed.
I've
dropped
some
notices
in
there,
but
not
a
super
large
amount
of
feedback.
A
Yet
all
right,
I
will
move
on
forward
from
there.
That's
our
release
blocking
job.
The
next
part
is
we
have
some
new
endpoints
introduced
in
120..
There
is
a
cap
out
for
storage
api
for
aha
api
servers,
and
part
of
that
includes
an
internal.
So
I
can
find
the
exactly
internal
api
group,
and
this
is
an
alpha
endpoint.
A
So
as
these
new
alpha
endpoints
are
added,
we'll
end
up
getting
new
alpha
endpoint
showing
up
in
api
snoop,
and
this
is
the
alpha
area.
But
if
we
change
that
from
alpha,
we
just
kind
of
back
out
from
alpha
and
change
this
to
to
stable.
We'll
see
that
whenever
you
have
an
introduction
of
an
of
a
new
api
group,
there
will
be
a
new
get
name
of
the
api
api
group,
so
the
internal
api
service.
A
So
this
is
going
to
be
inside
of
120
a
ga
endpoint
because
of
how
the
api
server
as
far
as
the
way
that
we're
looking
at
it
generates.
So
we're
we'll
need
to
add
this
likely
to
our
list
of
unconforming
endpoints,
but
I
wanted
to
try
to
follow
along
with
the
cap
so
that
it
gets
so.
I
think
this
is
the
kept
that
we
linked
to
earlier,
so
we'll
have
this
list
of
ineligible
endpoints
so
that
it
doesn't
confuse.
So
it
was
this
one
already
here:
rion
internal
yeah
internal.
So
it's.
A
One
of
our
I
know
this,
and
one
of
the
things
we've
been
trying
to
figure
out
over
time
is
how
to
know,
from
a
from
a
static
perspective,
an
analysis
of
code.
What
is
optional
or
not,
and
in
this
case
this
is
part
of
why
we
have
this
list
of
ineligible
endpoints
is
trying
to
document
why
it's
ineligible,
whether
it's
optional
or
not.
A
So
we
could
update
it
to
be
optional,
but
it
looks
like
it
would
actually
be
on
track
longer
term
2b
conform
it
when
that
kept
gets
all
the
way
through
to
ga
because
then
get
well
in
you
know,
api
group
would
be
there,
so
this
is
part
of
tracking
that
and
keeping
that
connection
with
that,
sig,
so
that
we
can
ensure
that
our
our
protocols
are
followed,
go
back
to
the
beginning,
so
it's
I
think
our
feedback
on
not
no
not
eligible
for
conformance
yet
could
also
be
it's
an
optional
endpoint
for
alpha
apis
to
be
super
clear
so
that
it's
better
understood
does
anybody
have
any
thoughts
on
a
better
wording
for.
A
C
A
Zach,
do
you
have
that
query
handy
that
we
can
link
to?
I
I
think
it
will
be
in
the
api,
smooth
repo
itself.
A
Have
some
other
logic
around?
If
we
see
a
get
some
type
of
api
group
and
the
only
there
is
no
content
in
ga,
then
we
could
automatically
remove
it
from
the
list
of
eligible
endpoints
for
now
we're
picking
out
specific
endpoints,
because
we
have
a
link
as
to
a
reason
why
that
doesn't
exist.
Yet,
whether
it's
optional
or.
E
A
So
right
now,
if
you
look
at
our
load,
open
api
functions,
we're
pulling
that
from
master,
and
so
we
would
need
to
find
a
way
to
temporarily
spin
up,
which
I
think,
if
I
look
in
the
kubernetes
source
code
on
how
open
api
spec
swagger.json
is
generated.
It
spins
up
an
api
server
temporarily,
does
a
curl
and
dumps
it
to
disk.
A
A
A
We
open
an
action
item
here
for
research
possibilities
for,
depending
on
a
swagger
json
generated
from
generated
from
a
conformance
run
instead
of
a
master.
A
Master
branch,
I
think,
we'll
run
into
some
issues
on.
We
depend
on
that
for
some
of
the
the
top
level
things,
but
it
might
work
good
for
conformance.
Let
me
swing
back
around.
A
I'll
go
ahead
and
move
forward,
then
we
we're
trying
to
get
rid
of
our
backlog
of
broken
things,
and
so,
if
we
go
back
to
what's
available
on
114
that
is
currently
not
tested,
so
I've
dropped
that
link
there.
If
we
sort
by
tested
release,
we
can
see
that
a
lot
of
these
endpoints
were
picked
up
in
117
and
eventually
119,
but
we
have
list
coordinated
lease
for
all
namespaces
and
also
a
lot
of
priority
class
related
endpoints,
and
I
think
rian.
You
had
some
information
on
leased
and
priority
class.
B
Yes
from
lease
I
actually
found
it
belongs
to
witches
team
and
they
picked
it
up
yesterday
and
created
the
test
where
we
can
discuss
that
in
the
second
part,
where
we
look
at
the
specifics
of
this
that
needs
approval,
then
about
the
other
one
as
priority
class.
I
was
already
researching.
Actually,
there
is
two
end
points
the
reading
and
create,
I
think
or
create,
and
one
other.
I
can't
remember
off
the
top
of
my
head,
I'm
just
all
researching.
B
So
there
is
some
endpoints
in
this
group
that's
already
hit,
so
I
was
thinking
if
there's
anybody
in
the
call
that
want
to
give
us
some
feedback
on
that
or
want
to
pick
it
up
and
say
it
belongs
to
us.
We
want
to
have
a
look
at
it,
so
if
not,
I
will
reach
out
to
the
sick,
whoever
it
belongs
to
that
created
the
first
endpoint
test
for
this
family,
because.
A
To
be
clear,
rian
you're
asking
for
feedback
on
priority
class
and
what
to
connect
to
on
that.
B
Yeah,
if
anybody
have
any
comments
on
that,
so
that
is
one
of
the
priorities
for
us
to
to
get
those
tested
in
this
release
and
if
there's
anyone
in
the
call
that
can
give
us
feedback
on
that.
B
Any
suggestions
about
the
way
to
test
or
not
when
it's
not
conformant,
so
we're
just
throwing
it
out
there
to
see.
If
anybody
wants
to
take
a
stab
at
this
force,
probably
twice.
F
F
But
could
be
enough,
I
think
it's
probably
take
scheduling,
but.
B
No
problem,
you
thanks
john
yeah,
as
as
I
get
more
information,
I
will
start
asking
questions
like
I
just
before
we
really
put
in
any
work
in
it.
I
thought,
if
there's
somebody
that
want
to
say
hold
on,
let's
not
conformant,
there
might
be
some
some
reasons
behind
that.
I'm
not
aware
of
so.
If
there's
nothing,
it's
not
about
not
a
bad
thing.
I'll
do
more
research.
A
Thanks,
ryan
and
john,
we
hold
our
test
writing
repository
where
we
do
all
of
our
documenting
creating
of
tickets.
So
this
folder
on
the
api's
new
repo
is
just
for
our
the
way
that
we
write
test
and
cluster
create
the
database
and
output,
the
markdowns
as
the
content
of
our
tickets.
A
It
was
a
bit
confusing
having
the
monorepo,
where
part
of
it
is
coordinating
and
getting
the
work
approved.
While
we,
you
know,
have
multiple
people
working
on
the
issues
together,
not
much
to
do
there
unless
you're,
actually
working
with
the
ii
team,
but
with.
I
think
at
some
point,
it'd
be
interesting
to
follow
up
and
have
it
getting
started.
A
B
Okay,
I
will
work
off
the
the
list
that
we
have
of
important
issues
and
then
also
tie
up
them
directly
to
that
to
the
price
report.
What's
not
on
the
project
at
the
moment
is
because
I
only
got
it
just
before
their
meeting.
The
ad
tests
were
listening
releases
for
namespaces.
B
As
I
said,
richard
picked
it
up
and
they
quickly
updated.
They
did
a
complete
life
cycle
test
for
all
these
endpoints
this
one.
Just
I've
missed
it.
I
picked
it
up
and
maybe
I
snoop
reach
out
to
them
and
they
quickly
sorted
it
out.
So
there
it
is,
it's
got
it
approved.
So
I
would
appreciate
it
again,
so
we
can
get
this
off
to
kill
some
technical
deck
518.
B
Test
that
I
want
to
talk
about
is
the
bot
proxy
and
the
node
proxy.
They
are
in
the
board
under
the
progress
problem
and
what
we've
done
in
the
august,
8th
and
september
almost
11
september,
8th
meetings.
We
discussed
these
and
we
because
the
proxy
actually
redirect
we
were
considering
to
push
it
out
to
ineligible,
for
conformance
and
in
the
meeting
it
was
decided
that
getting
a
301
redirect
would
actually
be
sufficient
to
show
that
the
endpoint
was
actually
totally
back.
B
So
stephen
went
ahead
and
created
the
test
for
pod
proxy
and
hitting
all
seven
endpoints,
and
then
we
in
a
similar
vein,
went
around
now
for
node
proxy
and
then
in
this
exactly
the
same.
E
B
We
have
under
the
triage
column,
we
have
service
proxy,
so
there's
been
a
lot
of
feedback
in
in
the
ticket.
B
G
B
From
the
way
we
wrote
it,
we
actually
really
run
the
test
for
what
proxy
and
known
proxy
two
or
three
times
already
through
all
bugs
sorted.
So
we
think
it
should
be
good
for
our
lg
team
and
approve.
If
there's
any
objections,
I
think
we
can
go
ahead
with
that.
B
Then
the
next
point
is
the
right
service
status,
life
cycle.
E
B
We
just
recently
it
was
an
old
test
that
was
approved
a
while
back
and
but
it
fell
off
the
back
burner.
So
we
re
wrote
it
and
we
would
like
to
have
some
feedback
on
the
way
we
wrote
the
test.
B
It
would
basically
hit
three
server
status
endpoints
if
there's
any
feedback
on
that.
If
we
can
go
ahead
and
create
the
test
for
for
the
pull
request.
A
These
are
the
four
issues
that
we're
we
would
love
to
get
some
feedback
on.
Would
it
be
useful
for
us
to
go
through
and
try
to
review
those
on
a
on
a
line-by-line
basis,
or
is
there
a.
F
I
B
F
Already
I
know
we
did
talk
about
those,
but
I
haven't
called
it.
F
B
Give
us
some
feedback
and
if
possible,
we
can
get
them
in.
Do
you
think
we
can
go
ahead
with
the
service
status
proxy
or
the
service
proxy
in
the.
A
Yeah,
we
were
looking
just
to
deprecate
it
because
of
possible
client
usage,
but
when
we
looked
at
the
actual
using
the
kubernetes
client,
it
actually
does
hit
that
301
redirect
and
follows
it
as
far
as
the
current
implementations
and
we
were
well,
how
do
we
do?
We
need
to
test
that
at
all?
A
C
Do
I,
if
we're
talking
about
the
get
request
specifically
or
get
and
head
request,
then
I
agree
that
verifying
the
behavior
makes
sense.
I
don't
agree
that
we
should
test
and
enforce
that
for
all
of
the
write
requests.
I
think
that
is
a
bug
that
should
be
fixed
in
the
server
since
it
will
not
actually
work.
C
So
if
you
look
back
at
where
the,
where
the
behavior
was
added
and
what
it
was
addressing,
it
was
to
support
browser-based,
get
requests
redirecting
and
ending
up
at
the
right
location,
a
post
request
that
the
api
server
intercepts
and
does
not
pass
the
post
on
to
the
backing
pod,
but
instead
sends
a
redirect
to
the
client,
which
turns
into
regret,
will
not
work.
That's
bad
correctly
and
the
the
fact
that
this
was
added.
The
only
reason
this
was
added
in
the
first
place
was
to
support
an
edge
case,
get
request
redirect.
A
To
remove
the
301
redirects
that
would
cause
bugs
in
kkk
that
would
reduce
our
number
of
possible
actions
against
that
endpoint
and
and
on
our
side,
reduce
the
number
of
stuff
we're
testing
to
get
and
head
on
those
three.
C
F
A
All
right
we'll
go
through
to
create
that
pr
or
that
that
issue
jordan
and
maybe
make
sure
that
what
we
articulated
is
what
we've
captured
on
this
call
and
go
from
there
and
then
we'll
we'll
pull
back
to
to
do
stuff
only
on
git
and
head
and
not
test
the
other
verbs.
I'm
not
sure
if
the
other
verbs
support.
C
A
Okay,
thanks
to
the
discussion
around
that
and
the
clarity
that
that
we're
able
to
provide
together
on
it
at
this
time,
I
want
to
make
sure
we
make
room
for
john's.
Thank.
G
F
G
I
So
if
you
recall,
we
started
the
idea
of
potentially
creating
some
profiles,
but
it
wasn't
as
clear
to
everybody
that
we
should
be
doing
it.
I
Wasn't
this
clear
to
everybody
if
we
do
do
it,
what
we
should
be
doing,
and
so
in
response
to
those
pr's,
I
wrote
out,
I
just
brain
dumped
the
things
that
sort
of
state
as
I
as
I
see
it,
and
there
was
a
follow-up
discussion
which
I
wasn't
present
for
which
brought
out
some
other
things,
one
with
respect
to
like
brad's,
asking
who
who's
who's.
Looking
for
this
right
now,
what
do
we
need
it?
Another
was
how
we
roll
it
out.
I
If
we
add
a
profile,
how
do
we
roll
it
out
in
a
way
that's
non-disruptive
to
vendors
and
users
and
everybody
this
particular
topic
we've
talked
about
since
well.
It's
been
talked
about
since.
I
The
performance,
effort
and
and
has
gone
into
circles,
in
my
my
view
of
things,
and
I
think
that
a
big
reason
it's
gone
in
circles-
is
that
we
haven't
really
clearly
stated
or
agreed
on
what
purpose
this
would
serve
and
that
essentially
there's
several
different
sets
of
stakeholders
that
have
an
interest
in
this
idea
of
profiles,
but
may
have
want
them
for
different
reasons
and,
in
fact,
more
to
the
point
there's
different
stakeholders,
in
conformance
in
general
and
so
in
this
document.
I
What
I
really
would
like
to
have
this
discussion
about
and
come
to
try
to
come
to
some
conclusion
if
we
can
is
as
a
sig
architecture
subproject
here,
you
know
what
is
what
is
the
real
reason
who
are
real
stakeholders
so
from
from
a
community?
Essentially
the
the
the
short
of
this
document
is
of
this
part
of
the
document
is
that
you
know
the
conformance
exists
for
the
community
for
cncf
from
that.
I
That's
those
stakeholders,
point
of
view
to
avoid
fragmentation
right
it's
to
avoid
there
being
a
million
different
kubernetes
that
don't
work
the
same
way
and
and
therefore
damage
the
whole
development
of
the
ecosystem
and
the
project
as
a
whole,
so
that
is
kind
of
that's
the
stakeholders.
Now,
if
I'm
a
user-
and
I
see
this
with
users,
this
is
what
users
think
users
think.
If
a
cluster
is
conformant,
they
can
run
whatever
they
want
and
it'll
work.
I
The
same
way
everywhere,
which
is,
as
we
know
in
this
group,
utterly
false,
because
we
have
an
extremely
concerningly
weak
set
of
tests.
I
Despite
all
the
hard
work
that
hippy
and
his
team
have
been
doing
for
years,
it's
just
really
hard
right
and
and
lots
of
progress
being
made,
but
even
if
we
did
all
of
the
tests
that
are
currently
allowed
to
be
in
conformance,
we
would
still
be
leaving
a
huge
number
of
things
on
the
table
that
that
could
cause
really
dramatic
differences
between
different
sets
of
kubernetes
or
different
distributions
for
kubernetes
and
how
they
work.
I
So
the
from
a
user
point
of
view.
The
program
doesn't
really
do
what
users
think
it
does
and
but
in
any
case,
from
their
point
of
view,
they
wanna.
They
want
workload
portability,
and
I
think
we
want
that
too.
In
the
community
and
as
vendors,
the
search
of
stakeholders
is
the
vendors
and
the
vendors
have
an
interest
in
workload
portability
as
well
and
in
lack
of
fragmentation
as
well.
I
But
you
know,
luckily
hasn't
been
a
problem,
but
there's
always
the
sort
of
risk,
in
my
mind,
of
bad
actors
in
the
vendor
community,
intentionally
trying
to
subvert
things,
there's
plenty
of
space
for
them
to
do
that
in
the
existing
conformance
program,
and
then
vendors
also
have
kind
of
a
goal
like
they
want
to
get
certified.
They
want
to
say
they're,
certified
kubernetes
and
be
able
to
put
that
badge
on
it
and
economically
they're,
not
incentivized
to
go
much
farther
than
that,
and
so
that's
another
issue.
I
One,
that's
why
we
have
trouble
getting
really
a
handful
of
vendor
a
couple
of
vendors
like
three
or
four
vendors
cops
who
actually
come
and
actually
contribute
in
this
forum.
You
know,
despite
the
120
vendors
out
there
or
so
so
I
think
that
we
haven't
sort
of
figured
out.
What
is
our
real
purpose
of
being?
Which
of
those
users
is
preeminent?
Which
of
those
stakeholders
is
preeminent?
I
I
would
argue
that
it
should
be
the
community
and
the
lack
of
fragmentation.
It
should
be
our
top
priority
and
our
top
goal,
and
I
think
that
if
others
agree
with
me,
I
think
that
changes,
sometimes
the
flavor
of
how
we
would
implement
profiles
so.
I
I'll
pause
there
for
comments
and
then
I
can
talk
a
little
bit
about.
J
J
Some
of
us
would
like
us
to
go
well,
no,
let's,
let's
just
go,
go,
ask
everybody
and
see
if
that
group
of
tests
could
could,
for
some
reason
go
in
the
base
or
how
many?
How
many
of
the
you
know?
This
is
what
we
did
in
march
2017,
how
many
of
the
interested
parties
can
implement
those
tests
today
or
how
many
of
the
interested
parties
think
with
a
little
bit
of
work.
They
could
get
there
and
that's
what
I
call
the
hard
path,
because
it
forces
us
to
make
a
discussion
about.
J
Should
we
be
raising
the
bar
up
and
just
getting
those
tests
and
increasing
that
in
in
the
baseline?
Or
do
we
really
have
a
split?
So
I'm
looking
for
the
the
sweet
spot
of
we
get
to
a
split
where
half
of
the
interested
parties
say
they
can
implement
something,
no
problem.
Maybe
it's
all
the
cloud
providers
or
something
and
half
say:
oh,
no,
that's
a
bridge
too
far.
That's
that's
you.
You
can't
ask
me
to
do
that.
J
Maybe
I've
got
the
first
group
of
tests
that
goes
in
something
that's
a
cloud
provider
profile
or
something
like
that,
because
if
those
hard
steps
aren't
done,
we
all
know
what's
going
to
happen.
Everybody
wants
to
be
an
engineer
and
they
all
want
to
have
the
joy
of
creating
their
own
profile
that
they
think
is
a
value
to
them.
J
Let's
go
look
about
having
the
timeline
to
get
these
into
the
base
if
we're
at
80.,
and
so
it's
kind
of
that
process,
step
that
that
john,
I
I
always
worry
that
you
know
everybody's
in
such
a
joy.
You
know
so
there's
that
and
then
there's
the
what
I'd
call
the
the
true
stakeholder
somebody
who's
showing
up.
That's
saying
I
am
so
frustrated
that
these
tests
aren't
getting
into
the
base.
J
J
I
So
so
my
concern
there
is,
is
I
guess
this
gets
back
to
like
how
how
broadly
we
want
to
make
people
conform
it
right.
So,
if
I'm,
if
I'm
selling
a
distribution
that's
for
edge,
and
therefore
I
want
to
assume
that
there
are
no
persistent
volumes,
everything's
ephemeral
right
now,
if
you're
saying
that
these
persistent
volume-
maybe
I
have
to,
I
must
provide
a
perfect
persistent
volume
implementation
in
order
to
be
conforming,
then
I
you
know.
I
I
There's
areas
we've
intentionally
left
out
of
conformance
because
they
require
features
that
we
don't
expect
will
necessarily
be
available
in
all
environments
or
or
even
that
there
are
features
that
are
not
of
from
their
inception.
We're
not
conceived
as
mandatory.
Like
our
back,
I
mean
you
can
run
a
cluster
without
our
back
and
today
that's
conforming
and
do
we
want
to
tell
people
they
must
run
our
back
now.
Here's
where
we
get
the
confusion
of
the
stakeholders
right,
an
individual
cluster
actually
isn't
conformant
distribution
is
conformant,
and
this
is
part
of
the
confusion.
I
A
distribution
has
to
be
capable
of
doing
it.
Every
single
cluster
that
the
distribution
creates
doesn't
need
to
be
incapable
of
doing
it.
It
needs
to
be
capable
of
producing
a
cluster
that
does
support
those
apis.
So
maybe
you
know,
I
don't
know
that
that
doesn't
mean
people
have
to
ship
an
implementation.
But
if
there's
a
you
know
an
edge
deployment
it
integrates
with
some
on-prem
csi
storage
provisioning
system,
then
you
would
need
it
so.
J
Well
so
so
the
the
the
step
of
asking
the
the
the
group
of
representative
vendors
is
an
important
one,
because
let's
say
it's
only
10
can't
do
the
persistent
storage
and
let's
say
it's
only
10
that
that
need
the
ephemeral
at
that
point,
you,
you
wonder,
and
you
go
okay.
J
J
I
mean
you
really
want
to
try
and
position
this
john,
in
which
case
you
end
up
with
probably
worst
case
scenario,
two
equivalence
classes
to
keep
the
world
simple,
where
one's
going
to
be
pretty
much
full
function,
kubernetes
with
the
storage,
my
guess
is
pretty
much.
Everyone
could
support
the
r
back
and
that's
why
you
want
to
ask
the
question.
I
could
see
two
different
and
these
were
the
two
classic
examples
we
always
talk
about
are
back
in
storage,
our
back.
Maybe
you
go,
ask
everybody
and
they
can
just
go
implement
it.
J
Fine
and
the
persistent
storage
is
the
one
where
we
finally
get
the
situation.
I
describe
a
true,
a
true
split
in
the
community
that
says:
80
percent
can
implement
20
can't
and
that's
20
says
they're
never
going
to
get
there
and
in
which
case
you're
going
to
just
say.
Well,
we
got
a
enterprise
profile
or
whatever
your
favorite
name.
Is
that
that
captures
the
fully
functional
thing
and
then
you've
got
all
the
other
folks
that
are
playing
around
with
the
less
thing
and
then
the
question
will
be
well.
J
Did
those
lesser
folks
really
need
the
certification
anyway?
Right,
if,
if
they
didn't,
if
they
did
fine,
but
that's
really
want
to
get
to-
which
is
a
model-
that's
trying
to
drive
towards
at
most
two
equivalence
classes,
because
if
we
really
really
need
this,
let's
apply
some
pressure
and
tension
to
the
system
that
really
causes
folks
to
fit
into
one
of
those
two
camps
yeah.
I
can
support
it
now
or
in
the
near
future
or
yeah.
I'm
I'm
never
going
to
be
able
to
support
that.
J
You
have
no
idea
what
my
thing
is
and
if
we
get
there
that's
fine
and
then
we
could
clearly
define
the
two
profiles
and
people
could
understand
those
worlds:
hey
because
you
know
for
edge
case.
You've
got
edge,
conformance,
that's
great,
but
98,
and
maybe
80
percent
of
us
aren't
going
to
care
about
it.
They
can
have
their
own
conformance,
sweet,
that's
fine!
The
sweet
spot
is
going
to
be
there.
I
J
But
in
one
of
the
previous
meetings-
and
I
think
jordan
was
here
for
one
of
the
previous
meetings-
we
did
come
to
the
conclusion
that
we've
been
talking
about
this
for
four
or
five
years
and
it's
pretty
much
those
only
two
that
have
come
up
in
four
or
five
years.
So
the
the
the
mysterious
hey
build
it
and.
I
You
have
different.
We've
also
talked
about
different
classes
of
users.
So
so
can
you
be
conforming
like
as
a
user
sort
of
user
level,
workloads
ordinary
users
that
don't
need
to
say,
modify
the
cluster
composition
either
in
nodes
or
or
web
hooks,
or
anything
like
that
right,
so
things
that
we
call
admin
level
things
where
you're
actually
modifying
the
configuration
of
the
cluster
or
adding
nodes
or
labeling
nodes
tainting
nodes,
that's
a
different
class
of
users
than
you
know,
those
administrators.
I
So
if
you,
this
is
where
I
get
back
to
it
kind
of
depends
on
the
stakeholder,
if,
if
I'm
a
user
right
from
a
user
perspective,
all
this
stuff,
what
what
users
hear
when
they
hear
certified
kubernetes
is
that
my
stuff
will
run
up
right
and
so
the
question
is,
it
kind
of
depends
on
what
that
stuff
is
and
if,
if
I'm
running
a
highly
managed
like,
I
think
openshift
you
get
a
name
space,
you
don't
get
a
cluster
right.
So
is
there
a
in
the
online
version
right?
I
Is
there
a
and
maybe
a
red
hat
person
correct
me?
If
I'm
wrong,
is
there
like
a
a
model
and
is
it
appropriate
for
this
group?
This
is
the
question:
is
there
a
model
for
defining
some
set
of
tests,
some
set
of
parameters
that
says
hey?
If
you
meet
this
level,
then
typical
user
workloads
will
work
and
therefore,
in
this
highly
managed
environment
it
can
actually
meet
that
level,
whereas
in
you
know
in.
J
Yeah-
and
I
agree
that
I
I
think
you
would
end
up
making
more
progress
is,
if
you
and
just
to
use
somebody
could
come
up
with
better
terms.
But
let's
say
we
could
just
drive
towards
two
equivalence
classes,
light
or
heavy,
and
we
look
at
the
features
and
we
figure
out:
can
it
go
into
the
light
class
or
the
heavy
class,
so
heavy
class
you're
talking
about
all
those
management
functions,
maybe
in
enterprise
features
and
persistent
storage,
and
you
start
trying
to
mark
them.
J
Okay,
well
we're
going
to
do
a
two
equivalence
classes
and
let's
go
see
which
ones
you
know.
Let's
try
and
push
as
much
as
we
can
into
and
to
heavy
and
make
it
you
know
as
as
good
as
we
can
and
or
or
I'm
sorry,
let's
push
you
know
what
what
goes
across
light
and
heavy,
but
let's
really
find
the
things
that
separate
the
wheat
from
the
chaff.
J
If
we
could
find
a
way
to
create
that
equivalence,
class
and
figure
out,
because
if
you
can
boil
this
down
into
two
things,
you're
going
to
be
successful
right
and
the
classic
model
is
I
I
don't
even
want
to
go
to
three.
So
the
classic
example
is
there
was
enterprise
java
right
j2ee
and
then
there
was
j2se,
and
then
there
was
j2
micro.
Whatever
the
name.
J
Wasn't
I
don't
remember,
but
my
guess
john
is
you
will
be
hugely
successful
if
you
push
towards
forcing
all
the
vendors
to
say,
hey
well,
if
you
need
to
stay
in
the
light
group,
well,
that's
pretty
close
to
what
we
already
have
and
so
don't
get
nervous
you'll
be
in
you
know,
you
know
kube
light
and
then
for
the
for
the
more
ambitious
folks
where
we're
really
trying
to
get
the
more
expressive.
Kubernetes
call
that
heavy.
J
I
would
call
it
enterprise,
but
I
I
think
that
that's
you
people
would
take
offense
of
that
term,
so
I'm
just
going
to
call
it
heavy
and
let's
go
figure
out
what
goes
into
heavy.
I
think
if
you
break
it
into
just
two,
I
think
people
can
then
start
making
a
decision
and
it
takes
all
the
abstraction
away.
John
and
then
you
know,
because
you
could
sell
too.
I
I
What
we've
said
to
people
is,
don't
build
stuff
in
tree,
build
stop
using
the
extension
points,
but
so
the
that
means
that
we
need
those
extension
points
to
really
be
part
of
the
base,
and
some
of
those
in
essence
require
some
level
of
privileges
that
we
wouldn't
necessarily
presume
an
ordinary
user
has.
So
anybody
else
have
any
opinions
on
on
this
or
any
thoughts
about
what
we're
talking
about.
K
Yeah,
thank
you
yeah.
I
mean,
I
think,
if,
if
I
totally
get
the
parallels
with
java-
and
I
see
why
that
makes
sense,
I
think
the
challenge
is
that
you'd
end
up
with
different
vendors
that
had
different
subsets
of
heavy.
That
would
all
be
categorized
in
the
light,
and
then
the
problem
you'd
have
is,
that
is
that
it
might
start
to
become
meaningless
for
users
right.
So
you
know
we
might
say
okay.
K
K
I
mean
I
can
give.
I
can
give
a
specific
example
from
from
our
use
case.
Is
you
know
we
have
we?
We
lock
down
specific
things
about
about
the
kubernetes
that
we
deploy
in
vsphere,
so
we
we
lock
down
host
network,
we
lock
down
node
port
and
there's
a
couple
other
things
that
we
that
we
don't
let
users
do,
because
you
know
we.
K
We
want
them
to
be
able
to
run
in
a
highly
secure
way
and
for
us
to
be
able
to
say
to
users.
Okay,
you
know
we
are
a
conformant
kubernetes
distribution,
but
we
can't
do
x,
y
or
z
or
we're
not
conformant,
but
you
can't
do
x,
y
and
z.
It's
going
to
have
meaning
it's
going
to
have
meaning
for
our
users
at
the
end
of
the
day,
you
know
they
they
and
this.
This
is
difficult
because
it
for
us,
that's
a
very
specialized
sort
of
use
case.
K
D
Yeah,
I
think
I
think
to
add
to
that
also
the
one
of
the
things
that
brad
brought
up
in
the
in
the
discussion
was,
you
know
how
do
we
work
with
environments
that
cannot
conform
at
all
right
and,
if
you
think
about
different
operating
systems
or
different
environments
in
which
kubernetes
runs
today,
for
example,
there's
lots
of
features
in
kubernetes
which
are
tightly
coupled
to
linux
right
like
host
ipc,
hostport
host
network?
D
I
I
I
That
seems
unlikely
to
me
honestly,
the
level
of
effort
involved
and,
if
microsoft's,
not
interested
in
doing
that,
I
don't
think
anybody
else
is
going
to
put
the
effort
in,
but
to
provide
some
sort
of
conformance
or
user
assurance
for
windows.
I
Nodes
would
be
looking
at
a
sort
of
runtime
conformance
type
of
program
which
isn't
this
program
today,
and
that
kind
of
gets
back
to
like
what
is
this
program
today
and
is
that
what
it
should
be
and
is
all
this
stuff
we're
talking
about,
really
makes
sense
in
the
context
of
that
and
I'm
not
still
not
getting.
I
sort
of
stayed
in
my
opinion,
which
is
that
our
primary
goal,
and
that
the
stakeholders
to
whom
we
are
beholden,
is
really
the
community
and
cncf,
and
this
the
health
of
kubernetes
as
something
that
does
not
fragment.
B
Can't
meet
it,
whether
users
can.
I
I
We
need
to
be
careful
to
not
make
a
water
down
like
the
light.
Just
exactly
what
ben
was
saying.
I
think
it
was
like
if
it's
too
watered
down,
then
everybody
needs
it,
but
everybody
does
other
things
on
top
of
it
that
don't
quite
meet
the
big
one.
So
this
is,
I
don't
know,
maybe
it's
just
a
subtle
judgment
call
and
maybe
like
some
of
the
stuff
brad
saying
about
how
how
we
make
those
judgments
is,
is
the
only
path.
I'm
not
not
hearing
a
lot
of
a
lot.
J
I
J
Like
storage,
okay,
so
so
what
do
you
think
about
this?
This
approach
and
I'm
going
to
call
this
the
cotton
gin
approach
and
if
you
know
how
an
old,
the
the
old
invention,
the
cotton
chin
was,
you
had
the
same
kind
of
problem.
You
had
all
this
cotton,
but
it's
all
bundled
together,
like
we
got
here
and
we're
never
able
to
make
progress
and
what
the
cotton
gin
does
is.
It
starts
pulling
the
cotton
off
whatever.
Is
there
right?
So
what?
If
hypothetically?
J
That's
more
of
a
fully
functional
kubernetes
and
the
first
thing
we're
going
to
go,
try
and
add
to
it
is
more
storage
apis
and
we
started
just
that
simply
and
and
went
around
and
said.
Okay,
let's
go
find
this
just
like
in
march
2017,
stakeholders
or
distros.
We
need
a
rep
because
we
really
need
to
know.
Can
you
support
these
apis
or
not?
I
J
Right
so
for
for
giggles,
I'm
going
to
say
enterprise,
because
maybe
maybe
I
can
get
away
with
that
and
we
go
shop
around
well,
who
can
add
the
storage
apis
and-
and
we
start
that
forcing
function
of
let's
start
go:
ask
the
distros
just
like
we
did
in
march
2017..
J
The
reason
we
made
progress
in
march
2017
is:
we
were
actively
going
to
the
vendors
and
saying:
can
you
run
these
tests
and
see
if
your
distro
passes
and
we
legitimately
made
decisions
based
on
that
eighty
percent
heuristic
of
well?
Eighty
percent
of
the
eighty
percent
of
the
distros
can
run
this
and
the
other
twenty
percent
think
they
can
get
there
that
that
that
was
the
process
many
years
ago,
john,
and
so,
if
you're,
you
know,
and
I
I
sent
your
frustration
that
hey
we're
not
making
any
progress.
J
I
see
that
as
the
the
first
legitimate
case
that
at
least
it
takes
away
a
lot
of
your
problems,
because,
if
you're
walking
up
with
storage,
everybody
agrees
that
that's
a
real
problem,
not
a
hypothetical
one.
So
you
have
all
that
built-in
legitimacy
and-
and
we
could
just
start
going
to
the
distros,
can
you
pass
these
these
10
tests?
J
So
you
just
like
we
did
in
the
way
in
the
early
days,
you
ran
sona
boy,
and
how
did
you
do
and-
and
I
remember
ibm-
we
didn't
pass
them
all
at
first,
but
we
talked
to
our
folks
we're
like
yeah.
We
can
get
there
on
those
next
day.
Those
last
eight
so
that
to
me
turns
us
from
a
hypothetical
conversation
of
of
designs
of
profiles
or
whatever,
to
being
a
more
process,
driven
approach
to
go
after
the
the
low
and
most
popular
fruit,
which
is
storage.
J
Because
then,
let's
say
you
get
success
with
storage.
Now
we
get
to
go
after
our
back
and
then
after
you
get
those
two
big
ones,
you'll
start
getting
to
a
point
where
you're
like
well,
maybe
it's
load
balancing
and
then
maybe
after
that,
when
you
get
to
well
there's
some
other
ones,
but
maybe
they're
not
so
important
right.
So
my
hypothesis
is
if
we
go
after
some
big
ones
and
show
some
progress.
I
I
One
thing
we
could
do
is
build
into
the
existing
conformance
tests,
a
provisional
set
of
tests
that
don't
count
towards
conformance.
But
when
people
submit
the
results,
then
we
get
to
see
right
there,
whether
they
pass
or
not.
So.
J
B
I
A
That
is
okay.
I
want
to
tell
you
both
that
we
can
support
that
effort
when
they
submit
the
results.
We
currently
do
check
that
they
have
run
these
explicit
set
of
tests,
and
if
we
want
to
enable
more,
we
can
just
be
sure
that
those
show
up
that
they
ran.
They
were
successful
in
the
pr
for
submission.
A
So,
okay,
thank
you
very
much.
Thank
you.
Everybody
for
attending
and
john,
if
you'll
upload,
that
sometime
this
week,
that
would
be
great
and
we'll
see
everybody
in
a
couple
of
weeks.