►
From YouTube: 20191022 SIG Arch Conformance Subproject
Description
Kubernetes SIG Architecture Conformance Subproject meeting for October 22, 2019
A
B
A
A
Run
of
the
tool
against
Pat's
back
to
get
a
list
of
behaviors.
What
I
need
to
do
is
I'm
going
to
go
through
and
curate
that
list
and
then
put
on
a
PR
for
that,
so
that
the
the
tooling
as
it
works
today
is
more
or
less
an
assistance
for
a
human.
That's
not,
and
it's
not
really
an
automation,
but
nonetheless
I
think
it's
useful
that
there's
a
PR
for
the
tooling.
B
A
We
do
have
around
this
broadening
pool
of
approvers.
We
do
have
a
session,
it's
not
on
the
schedule
yet,
but
I'll
be
giving
up
a
half
of
a
set
another
session
and
giving
it
to
conformance.
So
we'll
have
a
session,
probably
25
minutes
or
maybe
a
little
less
to
talk
about
conformance
at
the
summit
contributor
summit
and
to
try
to
recruit
people
for
the
effort
and
hippy
will
be
leaving
that.
B
B
B
They're
at
laser,
and
we
still
have
other
PRS
that
were
we're
looking
at
all
the
way
to
go
and
my
next
item
the
prowl
conformance
Bobcat,
thanks
Tim
for
looking
at
that.
We've
might
be
good
to
have
a
discussion
on
that
as
well,
just
to
make
sure
I'm
capturing
it
and
part
of
that
is.
That
is
what
I
put
on
the
agenda
for
today
around
conformance
numerators
and
denominators.
B
A
A
B
A
Can
be
reviewed
and
keep
moving
the
living
things
forward.
We
do
need
to
have
a
discussion
on
what
we
want
in
that
session
specifically
and
hippie
and
I
were
talking
about
that
sound
slack
earlier
and,
as
we
have
looks
like
we'll,
probably
have
time
in
the
agenda
today
we
can.
You
can
cover
that
here.
A
A
C
A
I
recall,
then
Tim,
do
you
do
you
or
I
guess
we
need
to
talk
about
what
help
we
need
right,
so
we
need
help
reviewing.
We
need
we
needed
to.
We
need
to
tell
people
that
the
the
policy
is
now
that,
when
you're
going
to
GA,
you
need
conformance
test.
That
means
we
need,
and
then
tests
are
already
running,
that.
C
A
C
A
The
sharing
time
with
this
other
sig
architecture
session
for
quota
division,
which
is
going
to
be
basically
what
we
don't
think
we
need
to
dato
anytime
soon
and
then
talking
about
this,
is
you
don't
need
to
put
this
in
there?
It's
sharing
time
with
that,
so
if
we
can
give
that
a
little
more
than
25
minutes,
as
my
main
point.
B
A
A
A
That's
an
awesome
idea,
there's
actually
at
the
summit
there's
a
dedicated
signatory,
so
you
say
that
the
Sigma
agreed
their
work.
There
will
be
representatives
from
every
single
sink
I.
Think
every
single
thing
has
signed
up
to
have
somebody
there,
and
so
we
should
make
an
effort
to
you
know
to
take
that
and
meet
each
of
them.
Take
5-10
minutes
face
to
face
with
each
sake
to
have
some.
You
know
more
human
contact
there,
which
I
think
will
be
a
lot
more
effective
than
just
how
many
people
over
email.
That's
what
also
my
dear.
A
A
A
Know
it
was
a
little
more
of
a
presentation
style
and
we
might
want
to
double
check
or
look
at
what,
what's
on
stead,
make
sure
that
we're
selling
it
the
right
way?
Because
if
it's,
if
it's,
if
it's
sold
as
a
presentation,
you
could
have
a
bunch
of
people
there
that
don't
necessarily
I
mean
just
in
in
doing
what
we're
talking
about
doing.
B
Know
I
would
have
a
room
or
a
boot
there's
something
where
we
have
big
screens
and
it's
intentionally
set
up
for
people
to
you
know
schedule
an
hour
with
their
with
someone
else
and
and
then
we
find
a
way
we
can
record
the
output
of
those
sessions
so
almost
like
twitch
for
pairing
on
writing
kubernetes
code.
We
don't
really
have
this,
especially
for
people
who
aren't
you
know
in
a
big
company
where
there's
people
near
them
all
the
time
to
do
this
pairing,
but
getting
that
culture.
How
do
you
code
on
kubernetes?
B
A
Are
all
been
requisitioned
and
all
that
stuff
like
we're,
probably
not
like
for
San
Diego,
it's
probably
a
little
late.
Try
and
you
organize
that,
but
the
summit
does
that
so
in
the
summit
there's
there's
there
sessions
and
workshops
sessions
are
more
just
blah,
blah
blah
blah
blah,
and
then
the
workshops
are
it's.
You
know
people
around
tables
with
their
laptops
and
somebody
leading
them
through
an
effort
to
actually
build
something
or
do
some
some
tasks.
A
We
could
potentially
look
at
not
not
at
San
Diego
it's
too
late,
but
we
can
potentially
look
at
doing
some
sort
of
workshop
sometime
in
the
future
around
something
like
that.
But
what
you're
talking
about
is
just
how
do
you
build
stuff
on
kubernetes?
How
you
develop
a
community
that
you
know
the
new
contributors
workshops
will
have
that.
Let
me
contribute
your
track
of
the
summit
will
have
some
of
that
too.
A
A
A
B
In
in
general,
the
I
think
the
contributing
to
writing
tests
stuff
that
we've
added
is
the
we
need
to
create
more
tickets
to
try
to
get
agreement
on
something
before
we
write
it
and
in
in
particular,
I'm
trying
to
make
sure
that
we
have
in
in
the
form
of
the
cap,
I
think
is
we'd,
get
it
formally
done,
but
also
as
we're
writing
it
to
get
feedback
on
these
different
priorities
that
communicated
parity
if
I
scroll
over
the
top
of
our
our
document
here
is
p1
p0.
This
pod,
spec
and
p1
is
watches.
B
A
A
And
a
watches,
watches
right
now,
that's
gonna,
be
we're.
Gonna
need
to
sit
down
and
really
just
speck
out
the
details
of
how
watch
behaves
in
certain
circumstances,
which
hasn't
been
really
explicitly
defined
up
until
now,
and
we
we
I
guess
I
mean
it
could
be
anybody
I
have
it
on
my
list
of
something
I
should
do,
but
the
best
person
to
do
it.
It
would
be
somebody
from
API
machinery,
Jordan
or
Daniel,
or
somebody
but
they're,
all
really
busy
so
anyway,
that
that's
a
great
question
is
who's
going
to
do
that.
A
B
We've
got
an
approach
for
both
of
these
I
kind
of
a
way
out
for
for
this
stuff
that
press,
let
you
know
what
we
found
and
make
sure
that
that
at
least
this
is
one
approach
and
if
we
don't,
then
we
can
go
down.
Okay,
let's
do
it
feedback
I,
don't
have
it
up
in
a
I
think
it
actually
is
in
a
ticket
off
the
cap.
B
And
so
the
and
I
think
Timothy
had
a
question
about
this.
Earlier
is
what
does
it
mean
to
be
hit
by
a
conformance
test,
and
so
one
of
the
definitions?
The
first
part,
is
it's
conformant
when
the
ete
binary
is
the
user
agent,
the
first
part
of
the
user
agent,
and
then
it
appears
in
an
audit
log.
So
we
have
to
go
all
the
way
down
to
seeing
it
in
an
audit
log
right
now
and
we
make
some
changes
early
on.
B
A
B
B
B
B
A
Yeah
I
I
I,
don't
have
an
answer
to
that
myself.
I
would
say
that
that's
approximately
what
we
want
and
for
now
that's
probably
what
he
said
it
doesn't
work.
Sometimes,
then
we'd
have
to
understand
why
it
doesn't
work
sometimes
yeah,
but
if
it's
more
in
most
cases
like
all
of
this,
is
it
at
this
point
here
is
that
great
and
we
need
to
improve
it
over
time.
A
B
B
A
This
doesn't
mean
conformant
endpoints
denominator,
total
stable,
endpoints,
sure
I
mean
yeah,
that
that
seems
reasonable.
The
the
we
know
that,
from
an
alpha
beta
perspective
that
that,
if
that,
if
the
kind
itself
is
stable
and
that's
fine,
it
may
still
have
alpha
beta
fields
in
it,
but
when
we're
just
County
endpoints,
this
was.
B
Reasonable
yeah
the
points,
the
the
endpoints
themselves,
don't
have
a
kind
associated
with
them
all
the
time,
and
there
are
a
lot
of
endpoints
that
we
hit,
since
they
don't
have
a
kind
associated
with
them.
They're
not.
We
have
to
look
at
the
path,
which
is
why
we've
chosen
to
look
at
the
path
versus
look
at
a
you
know.
The
group
version
kind.
B
Paw
spec
is
where
we've
needed
to
generate
a
full
coverage
map
of
all
of
our
fields
and
four
pots
of
POD
spec,
for
example,
I've
and
and
and
and
looking
at
the
every
every
possible
operation
that
uses
pots
back
where
it
goes
from
pod,
the
pots,
but
there's
only
687
texts
or
or
integer
fields,
and
we
they
group
together
quite
nicely
where
we
can
say.
Let's
knock
out
this
particular
subfield
mm-hmm.
A
B
I
thought
that
this
in
general,
if
we
want
to
focus
on
the
denominator,
and
so
the
definition
of
the
denominator
here,
is
initially
pod
to
any
upper
anything
that
does
pods
back
and
that's
around
687.
For
now
that
gives
us
something
that's
attainable
and
reasonable,
and
then,
when
we,
when
we
get
through
that
it'll
likely
fill
out
a
lot
of
other
stuff
too
yeah.
A
So
we
do
have
a
okay
so,
with
the
the
behave
I'm,
not
sure
what
this
687
is
here
in
in
the
scope
of
eventually
with
the
behavior
stop,
the
idea
is
that
the
coverage
number
we
get
is
the
denominator
is
the
behaviors,
and
then
the
numerator
is
the
number
of
behaviors
that
are
covered
by
some
test,
which
is
is
essentially
tagged
in
the
limit
between
tests
and
behaviors.
We
still
have
conceptually,
we
still
have.
How
do
we
come
up
with
that
list
of
behaviors
right?
A
A
We
have
to
assume
I
guess
that
the
behavior
list
is
comprehensive
in
what
we
consider
conformant
and
that's
the
sort
of
agreement
we
we
make
when
we
define
that
as
the
specification
for
conformance.
That's
all
I'm
trying
to
say
so
that
687,
what
I'm
trying
to
understand
is
that
at
687
you're
measuring
that
how
exactly
because
it's
gonna
be.
It
seems,
like
it's
gonna,
be
some
part
of
that
list
of
behaviors.
But
I
don't
know
if
it's
gonna
be
comprehensive.
B
A
Or
whatever
that
we
have?
Oh,
but
if
you
think
about
behaviors
aren't
one-to-one
with
fields
like
that's
the
fields,
I
expect
is
one
source
of
material,
one
source
of
law,
it's
one
set
of
raw
material
for
behaviors
and
other
raw
materials.
Talking
to
the
SIG's
talking
to
reading
that
there
may
be
things
I'm.
Just
being
you
know,
combinations
of
fields
like
how
does
this
particular
timer
and
that
particular
time
are
interact.
You
know
so
that
might
be
something
that
requires
multiple
descriptions
and
the
behaviors
that
need
to
be
potentially
tested.
A
So
my
point
is
that
I
guess
I,
don't
think
we're
gonna
ever
get
to
where
we
fully
automate
that
that's
that
denominator,
that
listing
of
behaviors
it's
it's
still
a
manual
process,
and
so
I,
don't
know
that
I
mean
I
get
your
687
here
is
a
place
to
start,
but
I'm,
not
sure
that
we
that
will
become
the
actual
denominator
and
eventually
in
the
list
of
well
or
rather
from
a
tooling
perspective.
The
list
of
behaviors
is
definitely
generally
what
is
conformant
and
therefore
it's
it's.
It's
just
a
static
number
of
behaviors.
B
Might
I'm
gonna
pull
this
out,
cuz
I,
cuz
pot.
Spec
is
one
thing
and
it
sounds
like
this
is
the
the
BDD
setup
and
yeah
having
it
be
a
separate,
a
separate
measurement,
and
it
may
be
that
bt
has
in
rap
and
has
a
focus
on
a
pod
spec
at
some
point.
But
it's
not
it's.
It's
loosely
defined
at
this
point
and
in
trying
to
get
us
to
a
point
where
we
have
a
PR
bot
that
is
clearly
defined
and
has
numerators
and
denominators
over
that
are.
B
A
B
We
have
a
measure
for
this
I'm
just
having
trouble
visualizing
it
right
now.
I
can
it's
probably
better
to
do
a
pairing
or
a
walkthrough
on
it
rather
than
trying
to
do
it
on
a
call
like
this,
but
we're
I
guess
I
was
trying
it
for
digit
in
general,
is
from
pod
the
pods
back
of
something
around
6
or
97
fields,
where
we
have
this,
where
we're
using
it.
Is
this
a
reasonable
start?
If
so,
I'll
write
that
up
more
precisely
because.
B
B
Of
objects
and
times
right
in
that
list,
is
around
130
for
the
top
of
my
head
and
those
are
the
denominator
are
130
for
operations.
That's
a
poor
watch
and
I.
Don't
have
the
report
handy,
but
somewhere
around
20
25,
depending
on
depending
on
what
we're
testing
today
you're
saying
yeah.
If
you
look
at
the
currently
released
blocking
jobs
where
we're
sitting
at
about
1/5
coverage
on
watch
and
that's
just
whether
we
hit
it
or
not,
not
whether
we
hit
it
in
a
meaningful
way,
yeah,
that's
the
same
thing
for
all
of
this.
B
A
B
A
B
B
A
There's
certain
I
think
this
is
important.
Nonetheless,
what
I
want
to
do
around
watch
isn't
so
much
just
basic
watch
functionality
of
every
resource,
although
that
that's
probably
helpful.
It's
around
defining
the
specific
semantics
of
watch
like
if
you,
if
you
read
the
documentation
of
watch.
It
specifies,
for
example,
that
that.
A
Sequencing
within
a
kind
is,
is
deterministic,
but
not
across
kind.
So,
like
you,
don't
have
a
global
single
global
version
now
in
practice,
because
we
back
with
that
CD,
we
have
a
single
global
version,
at
least
in
OSS,
but
there
are
vendors
who
want
to
swap
out
a
CD
I.
Think
Amazon
may
have
done
that
I
think
Microsoft.
Maybe
you
want
to
do
that.
There's
certainly
been
discussion
of
it
here
and
we
want
to
make
sure
that
and
then
the
kts,
for
example,
has
swapped
it
out.
A
We
want
to
make
sure
that
the
we
specify
the
minimum
semantics
needed
to
support
kubernetes
watch.
We
want.
We
don't
want,
watch
to
inherit
accidental
strong
constraints
from
that
CD.
That
would
be
impossible
or
difficult
to
implement
in
other
backends.
So
if
somebody
wants
to
shard
across
backends,
for
example,
the
single
global
global
sequence,
number
or
deterministic
ordering
is
kind
of
a
death
knell
or
makes
things
a
lot
harder.
So
those
are
the
kind
of
things
I
want
to
define
as
part
of
watch
that
haven't
really
been.
They
may
be
mentioned
in
passing
somewhere.
A
There
may
be
a
statement,
but
because
it
none
of
our,
we
don't
have
any
like.
In
my
ideal
world
we'd
actually
actively
check
a
provider
watch
functionality
with
potentially
with
clients
that
rely
on
looser
constraints,
the
Nazis,
but
what
we've
defined
as
conformant
and
watch
so
that
we
would
actually
those
resources
would
break
those
clients
would
break
because
we
don't
want
these
things
creeping
in
accidentally.
A
That's
what
I'm
talking
about
when
I
say
watch
is
a
high
priority.
That
said,
like
that,
that's
what
I'm
saying
like
also
the
failure
behaviors,
what
happens,
I'm,
disconnect
and
redelivery.
You
know,
there's
like
there's
lots
of
sort
of
pieces
like
that.
That
just
need
to
be
clearly
stated
if
they're
not
and
that's
what
I
was
saying.
B
A
Gonna
that
wouldn't
I
wouldn't
put
that
in
this
cap,
like
that,
would
be
something
we
put
into
a
Google,
Doc
or
or
into
I.
Think
there's
a
there's
actually
like
a
section
in
the
in
the
documentation
somewhere
that
talks
about
watch
or
in
maybe
in
the
developer,
documentation
they
can
protect.
You
know
there
if
I
could
find
it
I
think
there's
like
a
TBD
there
or
something
like
that.
Actually
in
the
document,
but
a
Google
Doc
is
sometimes
easier
to
comment
on
and
change
than
a
PR,
but
either
way
yar
has
better.
A
Anyway,
so
I
don't
want
to
derail
what
you're
doing
like
this
is
actually
useful.
Still
what
you're
talking
about
just
because
for
my
understanding
talking
to
the
API
machinery,
folks,
that
you
know
it
is
important
to
test
the
different,
the
different
resources,
because
somebody
could
do
something
wrong
and
then
have
a
resource
not
work
properly,
but
that's
not
necessarily
what
my
goal
was
in
that.
B
Okay
in
in
line
with
what
I
think
it
was
Timothy
went
through
last
week
with
us
around
my
our
API,
which
coverage
metrics.
It
was
164
if
you
go
back
to
our
our
Google
Doc
for
this
meeting
and
you
go
down
to
last.
Was
it
last
week,
yeah
October,
8th,
and
then
we
had
the
coverage
metrics
for
watch
and
just
underneath
that
Timothy
was
talking
about
the
steps
to
instantiate
an
object
and
iterate
through
them
and
rank
we've
taken
a
look
at
that,
but
we
want
to
make
sure
that
we
get
more
feedback
on
that.
B
B
And
I
just
want
to
make
sure
that
we're
doing
the
right
approach.
We
haven't
found
the
simple
way
to
list
all
resource
types.
Yet
we've
got
that
Emma
clone
going
and
then,
when
we're
looking
at
those
types
we're
going
to
create
them
it
would
it
be
easier
for
me
to
share
my
screen
for
this
one
or,
if.
B
B
Yeah
good
we're
having
trouble
with
number
three,
but
I
also
want
to
make
sure
that
this
seems
like
a
reasonable
approach
to
the
test
and
we've
been
trying
it
just
doing
it
on
our
own,
but
I
think
it
would
be
good
to
get
some
feedback.
I
think
we
saw
that
we
just
that
lava
lamp
was
a
sign
of
just
yesterday.
B
C
B
B
Comment
there
and
we
can
come
back
to
that,
that's
my
it's
my
I
think
one,
and
then
it
was
the
filter
for
those
right.
So
if
you
look
back
at
the
filter,
if
we're
gonna
go
through
and
actually
create
a
watch
and
create
them,
the
most
of
the
kinds
have
a
lot
of
requirements,
and
so
we
just
wanted.
The
initial
limit
is
just
say:
look
for
stuff
where
they
only
require
the
name
and
then
we'll
modify
the
name.
The.
A
A
You
know
some
of
these
things
end
up
creating
some
thing
in
the
cloud
provider.
If
there's
a
cloud
provider
with
clothes
alchemy,
that
would
be
a
few
set
load,
balancer
type.
For
example-
that's
not
really
a
big
deal,
but,
like
you
know,
we
would
know
if
we're
gonna
have
an
effect.
I
guess
here
we
automated
I'm,
not
sure
the
right
way
to
go
about
it.
Well,
he
did
but
Daniel
lava
lamp
and
he
had
this
same
same
key.
The
same
gods,
Tim
dude
of
like
K,
let's
go
through
and
we
can
automate
this
I'm.
A
For
certain
required
fields-
and
we
can
just
as
long
as
it
creates
some
even
if
they're
invalid,
like
you
know,
there's
no
such
image,
for
example,
on
a
container
image
they.
So
if
you
put
random
strings
in
there,
we
just
want
to
make
sure
it's
not
harmful.
That's
all
I
mean,
and
maybe
we
just
maybe
we
just
have
to
look
at
them.
You
know
one
by
one
at
that
point,
we'll.
A
A
To
Marty
for
now,
which
is
good,
we
can
use
that
as
a
template.
We
heard
so
we
need
to
I
have
an
action
item
to
update
the
summit
scheduled
to
include
this
conformant
session
and
I
can
take
Tim's
email
on
these
notes
here
to
put
together
a
description,
hippy
and
then
I'll
send
it
to
you.
If
you
watch
a
review
before
I,
let
the
summit
staff
now
well,
you
can
do
it
either
way,
I,
don't
care!
Oh.