►
From YouTube: SIG Network Gateway API Bi-Weekly Meeting for 20220502
Description
SIG Network Gateway API Bi-Weekly Meeting for 20220502
A
All
right
it
is
may
2
and
we've
got
a
lot
on
the
agenda
today
to
get
through
for
anyone
new
to
this
meeting.
This
agenda
is
very
informal,
feel
free
to
modify,
add
things
as
you
see
fit,
and
please
ask
questions
as
we
go
or
give
feedback.
I
want
us
to
be
as
interactive
as
possible
and
really
appreciate
different
perspectives,
and
with
that
I
think
candace.
You
have
the
first
thing
on
the
list
and
that's
this
very
outdated
beta
release,
plan
doc.
So
any
specific
comments
or
questions
here.
A
B
B
Yes,
so
what
I'm?
What
I
said
was
that
I
I've
only
been
here
a
little
while
and-
and
I
was
here
when
you
first
presented
this
doc,
so
I
didn't
think
it
was
that
out
of
date,
but
I
was
interested
in
an
update
on
some
of
this
date.
Specifics
in
here,
I
think,
there's
a
march
date
for
something
march
21st
at
the
bottom.
There.
A
Sure,
where
did
I,
I
believe,
there's
a
march
21
date,
but
I
don't
know.
A
This
is
in
progress,
yeah,
okay,
so
I
think
what
I
think
I
really
should
we
should
just
take
that
and
instead
pull
out
this
milestone,
which
is
really
tracking
a
more
up-to-date
thing.
That
is
probably
a
better
representation
of
now.
I
should
make
sure
that
everything
we
have
here
is
actually
an
issue
in
the
timeline
in
the
milestone.
A
A
A
Yeah,
that's
a
good
question.
I
we
got
the
initial
round
of
feedback
from
tim
and
dan
left,
one
or
two
comments,
but
it's
a
pretty
small
difference.
I
really
want
to
follow
up
with
a
wrong
issue
with
dan
on
this
one
to
see
if
he
can
provide
any
additional
feedback
as
well.
It's
all
it's
all
just
in
this
pr
here
and
there's
not
that
much
feedback.
Yet
much
of
what
has
happened
here
is
you
know,
follow-ups
in
issue
pr
etc.
A
But
this
is
this
is
the
this
is
the
only
bit
of
api
review?
That's
happening
right
now,
really
looking
for
feedback
from
tim
and
dan
and
cal,
but
I
think
cal's
out
for
a
while.
So
it
may
just
be
tim
and
dan,
but
yeah
just
need
to.
I
know
they're
both
very
busy,
so
I
just
need
to
block
a
bit
more
time
on
their
calendars
and
hope
we
can
get
one
more
round
of
review.
A
Well,
if
you
know,
if
you
happen
to
work
with
dan,
anything
that
you
can
do
to
help
help
him
take
another
look
at
this
would
be
great.
A
A
A
Great
any
other
comments
or
questions
on
the
release.
Thank
you
for
bringing
this
up,
because
that
that
doc
had
kind
of
been
circulating,
and
I
think
it
was
clearly
not
as
up-to-date
as
it
could
have
been.
B
A
Yeah,
I
think
that's
correct.
I
I
have
not.
We
have
not
added
that
as
a
graduation
criteria.
Okay,
the
graduation
criteria
specifically,
is
that
we
have
some
conformance
tests
in
place
that
cover
the
resources
we're
trying
to
graduate
to
beta,
and
we
have
that
it
does
not.
The
graduation
criteria
does
not
require
complete
coverage
of
those
resources,
the
more
coverage
we
have,
the
better
it
is.
Obviously
you
know,
I
think
we
all
want
more
conformance
tests,
the
more
conformance
tests.
We
write
the
less
e
we
test.
A
We
all
have
to
write
individually,
so
it's
a
win
all
the
way
around,
but
yeah
it
it.
I
don't.
I
don't
consider
them
blocking
for
the
at
least.
C
Sort
of
related
to
that
do
we
have
any
conformance
test
coverage
for
tcp
route
today.
A
No,
that
is,
I
added
that
to
the
agenda
because
I
wasn't
it's
been
a
while,
since
I've
looked
at
tcp
route
and
got
a
feel
for
who's
actually
implementing
it
today,
okay
and
so
on
the
agenda
two.
When
I
was
down
so
good
good
question,
it
works
cool,
maybe
candace
did
you
want
to
move
on
to
policy
attachment.
B
You
know
we
can
table
this
for
another
time,
because
I
don't
know
enough
about
this
myself
to
really
ask
the
questions
about
it.
I
was
trying
to
ask.
I
was
going
to
ask
my
my
local
team.
You
know
why
they
were
so
interested
in
it,
so
that
I
could
report
it
back
here,
but
I
didn't,
but
people
were
out
today
because
it's
a
made
a
holiday
in
for
those
guys
that
were
that
are
that
are
working
with
it.
So
I
can
ask
this
next
time
we
meet
yeah.
D
D
So
the
fact
that
it's
not
going
into
this
first
round
of
beta
doesn't
mean
that
it's
going
to
be
like
months
and
months
and
months
until
because,
like
it
goes
again
practically
because
of
the
api
review
requirement,
it'll
be
a
little
bit
but,
like
the,
I
think,
the
biggest
the
biggest
block
of
the
policy
to
policy
attachment
moving
further.
Is
that
there's
no
implementation
examples
yet
right
like
and
we
don't
have.
A
sort
of
we've
talked
a
little
bit
before
about
having
like
a
sample
upstream
one
that
that
does
some
something
common.
D
I
think
something
like
what's
in
the
examples
of
like
I,
don't
know
what
the
example
is,
but
there's
one
that's
in
the
examples
that
most
data
plans
will
support
the
the
the
key.
D
The
big
use
case
that
we
sort
of
were
thinking
about
when
we
designed
policy
attachment
is
timeouts,
because
that
that
came
from
the
fact
that
when
we
do
timeouts
when
we
tried
to
make
upstream
coffee
for
timeouts,
we
couldn't
because
every
proxy
does
them
differently
and
there's
no
way
that
there's
no
sort
of
there's
very
little
commonality
in
the
way
that
timeouts
are
handled
and
so
like.
D
We
need
to
have
a
sort
of
implementation
specific
way
to
handle
timeouts
that
also
had
a
sort
of
very
specified
behavior
in
how
those
settings
flow
up
and
down
the
resource
hierarchy.
So
that's
where
policy
attachment
came
from,
but
I
think
that
in
my
mind,
the
stuff
that
I
would
want
to
see
to
sort
of
graduate
the
idea
of
policy
attachment
to
beta
is
to
have
like
a
you're
sort
of
a
one
that
we
need
to
go
api
supply.
D
D
D
We
have
a
standard
set
of
behaviors
for
the
mechanism
and
then
once
that's
done,
then
we
can
say:
okay
now
you
can
go
and
make
your
own
custom
policy
attachment
stuff
and
then
because
we
have
a
good
example
of
how
it
should
work.
I
think
right
now
the
sort
of
the
how
it
should
work
is
roughly
outlined
in
the
docs,
but
it's
not
very
subtle
sort
of
it's
a
very
subtle
idea.
That
is
hard
to
explain,
and
so
it's
the
docs,
don't
do
a
good
job
of
explaining
it.
E
B
Yep,
absolutely
let
me
see
what
I
can
get.
Maybe
our
team
has
something
like
that
already.
They
want
to
share.
D
Yeah,
absolutely
if
someone,
if,
if
your
team
has
implemented
some
policy
attachment
stuff
like
has
implemented
a
policy,
then
you
know,
maybe
we
can
just
be
like
here's
an
example
right
like
that'd,
be
that'd
be
great
too.
Okay.
I
think
that's
that
in
my
mind,
that's
the
blocker
for
that.
For
that
stuff.
A
Thank
you.
Thank
you
for
that
question
and
and
great
answer
nick
by
the
way,
I
think
this
was
it.
I
I
kind
of
just
tied
this
recent
pr
of
mine
in
because
I
think
it's
very
relevant.
I
was
working
on
a
pr
as
part
of
our
release
criteria
or
milestone
of
what
does
it
mean
to
graduate
from
beta
ga?
What
does
it
mean
to
graduate
from
alpha
beta
and
I
realized
after
I
wrote
this-
that
it
doesn't
really.
A
I
don't
know
how
we
do
policy
attachment,
because
right
now
we
have
policy
attachment
is
what
every
vendor
brings.
Every
implementation
brings
their
own
custom
policy
and
attaches
it
to
our
api.
How
do
you
graduate?
How
do
you
write
conformance
tests
for
a
pattern
that
is
so
yeah?
You
know
it
really
does
help
what
nick's
saying
if
we
can
bring
some
kind
use
that
pattern
on
a
core
resource
like
if
there
is
some
kind
of
tls
configuration.
A
I
I
don't
know
something
like
that
is
shared
enough,
that
we
could
use
that
pattern,
maybe
conformance
test
work
there,
I'm
not
sure,
but
that
is
a
very
good
question
and
something
like
I
had
not
thought
about
in
this
context.
Yet
so,
thank
you.
That's
I.
I
I'd
also
echo
what
you
said
about
lots
of
teams
being
interested
in
policy
attachment
I've
seen
the
same
thing
within
google.
It's
kind
of
we.
A
We
have
lots
of
different
products
that
want
to
be
configured
via
kubernetes,
and
this
is
kind
of
the
way
that
you
can
attach
different
config
different
concepts,
and
so
it
it
is
a
very
interesting
concept
to
a
lot
of
teams,
but
yeah.
You
know
like,
like
nick
said:
we
we
don't
have
any
great
implementations
publicly
available,
yet
that
I
know
of,
but
the
second
someone
releases,
something
like
using
this.
D
Yeah
definitely
100.
I
think
I
just
added
a
thing
to
the
agenda
that
I
just
wanted
to
mention
and
have
in
there.
We
had
talked
about
this
briefly
before
about
renaming
the
reference
policy
objects,
so
that
to
not
confuse
it
with
the
general
policy
mechanism.
I
think
we
have
an
issue
where
someone
who
I
don't
think
is
at
the
meeting,
has
logged
an
issue
sort
of
saying,
hey.
D
A
Yeah,
that's
thanks
for
calling
that
out.
I
I
agree.
I
I
had
been
very
interested
in
rename
there
too,
and
I
just
I
think,
there's
some
some
conversation
about
that
in
previous
meetings.
I
think
there's
an
issue
like
you
said
discussing
a
possible
rename
and
then
it
just
kind
of
got
dropped
and
also
you
know
it's
just
that's.
Why
we're
not
an
alpha
yeah?
Why
we're
still
in
alpha
with
that
one,
but
if
anyone's
interested
yeah.
D
The
more
that
we
can
hand
some
of
these
things
over
to
other
people,
the
better.
My
bandwidth
has
drastically
shrunk
in
the
last
couple
of
weeks
and
yeah
with
some
with
qcon
coming
up.
You
know,
I
think
I'll
get
some
back
after
q
con
but
yeah.
So
I
think
that,
right
now
we
are
having
problems,
because
two
people
does
not
scale.
F
A
Yeah
well
said
so
with
that.
Let's
talk
about
well,
actually
was
that
all
on
policy
attachment?
I
know
we
kind
of
went
in
a
few
different
directions
there,
but
did
we
actually
address
that
question?
Well,.
B
Yeah
I
have
something
to
bring
back,
okay
and
and
maybe
I'll
you.
B
We'll
be
able
to
provide
something
to
help
it
along
no
promises,
but
I
can
definitely
check
into
it.
A
Cool
great
thanks,
okay,
so
a
broader
question
of
tcp
route.
I'm
curious
among
implementers
here
who
is
either
currently
implementing
or
has
plans
to
implement
a
tcp
route.
C
A
Okay
related
to
that
I
wanted
to
call
out.
You
know
something
that
we
had
discussed
at
the
last
meeting
of
pulling
out
gaps
735,
which
was
the
address
matching
bit.
That
is
one
that
is
significant.
It's
the
address
matching
for
tcp
and
udp
routes.
A
We
had
a
number
of
questions
that
were
difficult
to
answer,
and
so
basically
the
proposal
was:
let's
pull
that
back,
move
the
gap
to
provisional
and
continue
with
our
release
just
without
that
concept.
Yet
with
that
said,
since
we
do
have
implementations
that
have
either
implemented
or
in
the
process
of
implementing
tcp
route,
is
that
something
that
you
were
really
looking
forward
to?
D
Okay,
so
out
of
everyone
on
the
call
like
no
one
would
no
one
has
any
significant
issues
with
us,
calling
that
get
back
to
you
know,
provisional
or
whatever
I
can't
remember
yet
provisional
and
then
us
retooling
that
password
tool
in
that
gap
and
going
over
the
use
cases
again
posts
post
beta
and
using
the
experimental
mechanism
to
bring
it
in
okay.
So
the
goal
is
to
bring
it
back
in
right.
Yes,
absolutely
so
yeah
we
should
be
clear.
D
The
goal
is
absolutely
to
bring
something
that
does
the
same
a
similar
function
back
in,
but
some
of
the
questions
that
we
had
when
once
we
all
started
thinking
about
it
a
bit
more
like
we
thought
well,
maybe
some
of
this
is
better.
It
actually
might
be
better
to
put
this
on
the
gateway
and
put
it
on
the
listener,
rather
than
on
the
tcp.
D
G
A
Because
yeah,
I
there's
there's
also
some
interest
in
destination:
address
matching
that
had
that
had
less
questions
around
it
other
than
how
would
this
be
used
in
an
ingress
gateway
scenario,
because,
like
the
use
case
I
see
for
destination,
address
matching
is
for
mesh
implementations
right:
they
they
want
to
handle
routing
for
requests
that
are
intended
to
go
to
destination
foo
and
you
know,
send
it
or
apply
these
filters
or
whatever
it
happens
to
be.
So
what
another
go
ahead?
No,
no.
I
was
curious
what
what
your
use
case
would
be.
G
Yeah
I
mean
it's
part
that
some
some
people
also
use.
You
know
ipa
addresses
as
a
way
of
routing.
So
imagine
you're,
you
know,
even
in
an
internal
cluster
right,
an
internal
infrastructure.
Multiple
services
are
exposed
by
on
multiple
ip
addresses,
but
traffic
for
all
of
these
are
centralized
back
into.
You
know
consider
these
as
virtual
ip
addresses
and
all
of
these
traffic
for
all
of
this
comes
to
something
like
ingress
controller,
and
then
it
is
routed
to
corresponding
back-ends.
A
D
D
And
you
only
want
to
bind
someone
else
to
some
of
them.
There's
one
use
case,
but
there's
the
other
other
use
cases.
Are
you
want
to
send
all
the
traffic
to
the
the
gateway
implementation,
but
then
have
the
gateway
implementation
pick
a
route
with
using
the
address
as
a
discriminator
is
really
what
it
comes
down
to
right
is
that
right
here.
G
Yeah
that's
about
right
now,
some
there
are
probably
a
very
small
group
of
people
who
actually
ask
for
this,
and
this
is
mostly
like
very
large
enterprise
organizations
who
are
trying
to
move
into
kubernetes.
So
we
we
could
question
like.
Could
the
solution
for
source
and
destination
be
different?
You
know,
sources
like
easy
path
and
destination
would
be
a
bit
more
complicated.
A
A
How
how
do
you-
and
you
can
kind
of
you-
can
make
that
that
specific
use
case,
but,
like
you
said
it's
kind
of
for
the
you
know,
10
of
customers
that
are
or
users
that
are
enterprise
and
have
specific
workloads
in
mind.
G
D
I
would
have
possibly
said
obtuse
or
complex,
so
yeah
thumbs
up
on
that
one.
D
Yes,
I
think
the
just
just
to
make
sure
that
we
sort
of
since
those
people
here
are
doing
the
tcp
route,
like
that,
the
other
objections
were
like
what
you
tcp
route
has
a
whole
bunch
of
like
address
matching
at
the
moment.
What
happens
if
you've
got
like
multiple
tcp
routes,
specifying
the
same
address
that
are
all
attached,
the
same
gateway
in
some
way
or
what
happens?
If
what
happens,
if
you
have
multiple
tsp
routes
specifying
different
addresses?
How
do
you,
how
do
you,
how
does
the
conflict
resolution
emerging
and
stuff
work
and
because.
D
You,
because
this
is
like
a
firewall-like
functionality
and
there's
no
ordering.
You
know
you
can't
you
you
this
is.
This
runs
into
a
similar,
some
similar
problems
to
what
network
policy
you
had
originally
like
that
you
can't
have
because
there's
eventually
consistent
system
unordered,
you
know
you
can't
basically
have
some
allows
some
denies
because
they
must
be
ordered
and
so
like
yeah.
The
the
thing
that
I
liked
about
the
idea
of
moving
that
to
listener
is
that
you
can
enforce
some
ordering
and
some
other
stuff
like
that.
H
D
H
Is
the
thing
that
harry
said
is
the
difference
between
routing
and
firewalling
the
use
cases
he
has?
I
think
I
may
have
some
similar
ones
for
routing
and
there's
nick
helping
me
said
about
healthy
routing
and
kubernetes
doesn't
doesn't,
but
using
the
gateway
for
at
least
static.
L3
routing
is
an
interesting
concept
for
you
know
end-to-end,
so,
but
it
is,
I'm
not
sure.
Mixing,
firewalling
and
routing
rules
in
the
same
thing
is
a
great
idea
because
of
different
actions.
D
A
That's
that's
it
right
the
if
this
was
just
destination
address
matching.
I
think
the
question
would
be
is:
are
the
use
cases
sufficient
but
with
the
source
address
matching?
It's
are
we
implementing
something
like
a
firewall
but
not
calling
it
a
firewall?
Are
we
you
know?
So
I
think
this
is
unspecified
enough
that
we
can
probably
pull
both
of
them
out
and
that
that
is
what
I
propose
right
now,
but
I
I
can
be
convinced
that
someone
wants
to
leave
part
of
this
in.
D
Okay,
so
I'll
put
in
the
minutes
here
that
please
review
this
yeah,
I'm
I'm
plus
one
on
it.
I
can
lg.
I
will
lgtm
that
pr,
but
we're
sort
of
holding
it
to
have
some
other
people
who
actually
are
actually
implementing
the
tcp
route,
have
a
look.
D
So
those
of
you
who
are,
if
you
could
go
to
that
pr
and
hit
us
with
hit
us
with
the
go
with
some
lgtms
or
you
know
like
yeah,
and
if
we,
if
we
need
to
become
to
make
it
more
clear
that
we're
going
to
come
back
to
this
later,
that's
cool
too,
but
yeah.
I
think
we
definitely
this
is.
I
think
we
should
do
this
absolutely
do
this
before
we
move
to
beta,
because
once
we
move
to
beta,
we
can't
we
can't
do
it
anymore,
okay,
so
yeah.
D
So
this
is
absolutely
a
beta
blocker,
so
yeah
we
need
to
do
that.
Sorry,.
A
Cool
yeah,
so
this
is
pr1139
with
it
looks
like
I
need
to
push
one
more
update
to
this,
but
I
appreciate
any
any
comments,
even
if
it's
just
lgtm-
or
this
is
fine
with
me-
anyone
can
lgtm
this.
It
just
helps
to
get
some
a
variety
of
perspectives
on
this
and,
like
nick
said,
we
can
add
a
note,
add
an
issue
to
track
adding
this
back
in
getting
this
gap
from
provisional
back
to
implemented.
D
A
Yeah
yeah-
that
sounds
fair,
I'd
love
to
get
this
in
this
week.
So,
let's
say
by
thursday,
if
we
don't
hear
anything
plan
on
getting
this
through,
because
this
is
this
is
one
of
the
key
things
in
the
milestone.
I'd
say
this
is
one
of
the
bigger
ones
and
I'd
love
to
just
push
it
through
great
okay.
Well,
speaking
of
the
milestone,
let's,
let's
take
a
look
at
where
we
stand
now
we're
we're
finally
making
some
some
progress
here.
A
lot
of
these
are
getting
through.
A
We
have
some
prs
that
are
open
to
fix
some
of
these
issues.
One
of
this
one
of
these
is
a
pr
just,
I
think,
we're
in
an
okay
place.
My
biggest
concerns
right
now
are
the
status
thing.
We
have
lots
of
conversation
around
what
we
could
change
with
status,
but
we
don't
have
anyone
right
now,
working
on
it
that
I'm
aware
of
is
that
is
anyone
actually
working
on
this
or
planning
to
work
on
it?.
A
I
may
be
able
to
find
some
time,
but
please,
if,
if
you're
interested
in
contributing
this
is
a
good
way
to
help
out
this.
I
think
one
of
these
is
pretty
clear
right
now
and
what
we
are
suggesting
should
happen.
I
think
yeah.
I
think.
D
About
it
so,
okay,
we
can
get
a
pr
on
pretty
quickly
great
great.
D
And
already
ready
actually
needs
to
wait
it.
It
requires
too
much
talking
about
what
really
means.
Really.
I
don't
think
we
can
do
ready
for
v1
beta1
because
of
you
know
it's
yeah
yeah.
Maybe
we
have
a
different
type
or
something
like
that,
but
I
think
not
implemented
or
not
supported,
condition
type
for,
like
a
generic
one
that
that
you
are
allowed
to
add
on
to
resources,
but
and
if
so,
here's
how
it
works
is
the
way
to
go,
not
not
a
one
that
is,
has
a
mandated.
D
I
was
working
on
it
a
while
ago,
but
yeah
I
haven't
updated.
It.
D
So
yeah
so
ready,
so
we
did
talk
in
here
so
about
saying:
hey,
we
should
add
a
ready
condition
you,
and
that
would
mean
that
once
we
have
already
something
like
a
ready
condition,
then
it
makes
some
of
the
performance
tests
a
lot
easier.
However,
when
we
talked
about
this,
we
we
sort
of
found
that
many
implementations
can't
guarantee
that
the
config
has
made
it
to
the
entire
data
plane.
It's
like
a
very
strict
ready
condition
is
very
difficult
to
do
because
of
eventual
consistency.
D
So
you
can
your
controller
when
it's
you
know
so
for
envoy,
based
ones,
your
controller
can
be
like
hey.
I've
made
the
config
available
to
envoy,
but
unless
you
have
a
full
flow
of
the
ack
messages
and
stuff
like
that,
back
back
to
your
controller,
which
very
few
people
do
you
can't
be
like
this
is
definitely
configured
right.
Then
that's
what
I
mean
by
strict
ready
condition
is
that
you
can
say
with
great
confidence.
D
You
know
when
ready
is
true,
then
then
the
data
plan
is
100
configured
and
everything
will
work
will
not
should
will
we
can
have
we
can't
mandate
like
a
ready
condition,
because
that's
what
people
will
think
it
will
mean.
You
know
like
if
you
say
ready
people
mean
that
like
ready
right
now,
I
can.
I
can
send
as
many
requests
as
I
want
at
this
whole
thing
and
they
are
guaranteed
to
work,
and
so
we
we
talked
about
having
like
a
different
condition
type
that
would
have
a
different
name.
D
C
D
D
The
you
we
also,
and
so
that
was
the
last
thing
we
agreed
that
we
should
say
what
a
ready
condition
implies
and
that
it
is
allowed
for
you
to
use
a
ready
condition.
But
if
so,
it
must
behave
like
this,
but
it
must
be
that
you
can't
say
you
can't
use
a
ready
condition
unless
that
means
that
you
are
a
hundred
percent.
Confident
the
data
plan
is
completely
configured.
F
Yeah,
let
me
let
me
take
a
look
at
this
issue
and
if
I
could
take
it
on
then
I'll
sign
myself.
F
D
If
you
do
want
to
pick
up
any
of
these,
please
feel
free
to
hit
me
up
on
slack
or
you
know
we
can
do
a
zoom
or
something
like
that,
and-
and
I
can
talk
you
through
the
background
and
stuff
like
that-
it's
just
that
I
just
don't-
have
the
bandwidth
to
actually
sort
it
out
myself.
I'm
sorry.
E
Thanks
raising
my
hands
here,
if,
if
there's
any
way
to
split
up
this
work-
and
you
like
some
collaboration
that
I
could
maybe
help
as
well.
A
D
So
because,
here
I'll
keep
you
posted
on
what
happens
with
my
house,
one
about
the
other
ones
and
I'll
I'll
see.
If
I
can,
if
there's
any
other
ones,
that
we
can
put
in
there
about
sort
of
updating
the
conditions
and
the
status
and
stuff
like
that.
E
Yeah
in
the
it
sounds
to
me
that
there
has
yet
to
be
an
agreement
about
what
this
pack
would
look
like
right.
D
Yeah
yeah,
so
it's
more
sort
of
we
need.
We
need
someone
to
sort
of
write
up
like
here's,
what
his
work!
Here's,
where
we're
at
like
he's
what
everyone
has
said
and
sort
of
take
all
the
different
bits
that
everyone
has
said
and
sort
of
bring
them
together
and
say:
okay,
here's
what
I
think
we
should
do
so
that
we
can
and
maybe
that's
a
pr
yeah.
I
think
that's
probably
the
next
step.
A
So
just
one
tiny
thing
on
on
this
issue:
1077,
it
seems
like
we
can
probably
split
this
into
two
now,
because
essentially,
this
issue
covers
two
things.
It
covers
an
unsupported
or
not
implemented
kind
of
condition
and
a
ready
condition,
and
what
we're
saying
is
that
ready
part
of
it
should
be
split
out
to
another
issue
that
is
not
blocking
v050
release,
but
this
one
should
should
just
focus
on
unsupport
not
implemented
whatever.
We
call
the
condition.
A
Okay,
okay,
cool
who's,
but
any
any
volunteers
to
split
this
or
create
a
new
issue
to
track
the
ready
condition.
A
D
Cool
okay,
I'll
move
on
to.
Actually
there
is
one
in
there
that
is
a
that,
is
another
documentation,
one
the
path
for
redirects
and
rewrites
that
one's
we've
implemented
the
get,
but
we
haven't
documented
it.
So
that
is
a
pretty
like.
That's
a
reasonably
straightforward
small
pr
for
somebody
to
do.
That's
you
know,
go
and
go
and
look
at
the
spec
figure
out
how
it
works
and
document
document
how
it
works.
That's
that's!
It.
A
There
is
a
gotcha
there
and
that's
this
pr
where
we're
kind
of
switching
everything
around
or
proposing
switching
everything
around
on
rewrites
and
redirects,
so
that
I
don't
have
a
good
diff
here,
but
anyways
the
the
structure
is
changing
just
a
little
bit
that
doesn't
actually
change
how
you
document
it.
So
you
could
write
the
documentation
and
then
just
change
your
example
ever
so
slightly
if
this
didn't
get
in.
But
I
I
agree
with
nick.
D
Okay,
I
will
go
just
now
and
put
some
links
to
those
things
in
that
issue
so
that
it's
a
bit
more
discoverable
and
I'll
put
a
wanted
as
well.
A
A
D
A
To
it,
thank
you.
Thank
you.
Yes,
all
right,
let's
get
into
prn
issue
triage.
I
I
added
a
few
of
these
that
I
wanted
to
be
sure
to
cover
in
this
meeting,
but
I
know
there's
a
lot
open
right
now.
First
off
I
wanted
to
highlight
this
discussion
around
graduation
criteria.
A
Thank
you.
There
are
some
good
comments
already.
Thank
you
damian
yeah.
I
I
would
love
feedback
on
this.
I
appreciate
the
feedback
that
I
just
got,
but
I
really
need
to
think
through
what
this
means
for
policy
attachment.
A
I'll
say
just
briefly
that
this
first
alpha
alpha
beta
section,
is
what
we
used
in
our
cap
to
we
had
a
cap
upstream
cap
to
describe
what
it
meant
for
this
api
to
graduate
to
beta,
but
we
never
documented
that
in
the
gateway
api
repo
itself.
A
I
would
agree
with
the
comment
that
initial
is
not
very
clear.
It's
just
some,
which
is
again
not
not
great,
so
getting
into
a
bit
more
detail
there,
but
if
anyone
else
has
thoughts
on
what
it,
what
we
want
to
consider
when
we're
thinking
about
graduating
from
beta
to
ga
or
alpha
to
beta.
I
think
this
is
very
important.
A
It
may
seem
like
a
trivial
small
pr
now,
but
I
imagine
whatever
we
write
here
is
something
we'll
be
referencing
back
to
over
and
over
again
when
it
comes
time
to
hey,
there's
a
new
resource.
Does
this
meet
the
criteria,
so
I'd
love
to
have
a
sound
foundation
here,
whatever
whatever
that
is,
so
it's
just
meant
to
be
a
starting
point.
A
A
A
A
C
A
No,
no,
no
that
maybe
it
was
which
implementation
was
this,
but
we
we
did
a
a
fair
amount
of
work.
We
so
we
have
those
base,
conformance
resources
that
include
primarily
gateways,
and
the
idea
was
hey
you
can
you
can
set
these
up
beforehand?
A
The
key
thing
with
this
pr
is
that
we're
changing
that
fundamentally
we're
saying
a
gateway
is
fair
game
for
an
individual
test
case,
so
you
can
deploy.
You
can
like
each
individual
test
case
may
involve
spinning
up
a
new
gateway
since,
since
this
conformance
test
is
coming
from
steve,
it's
probably
fine
for
contour.
D
That
so,
which
we
did
want
to
allow
some
people
to
do
to
start
with
because
it
lets
you
get
started,
but
I
think
there's
a
pretty
strong
expectation
that
you
can
dynamically
reconcile
gateways
yeah
and
deal
with
them
being
created
or
destroyed
as
part
of
the
performance
anyway.
So
it
kind
of
makes
sense
to
me,
but
we
just
wanted
to
make
sure
that
we're
not
messing
with
anybody
who
is
assuming
that
you
can
do
static
provisioning
at
the
moment
we
may
well.
Contour
may
well
have
been
the
last
one
to
do
to
need
that.
D
C
D
Gotta
be
maybe
you
could
slap
an
lgtm
on
this
one
yeah.
C
A
I'll
I'll
call
that
promising
I
was
just-
I
had
a
hold
on
this
one,
just
because
I
wanted
to
make
sure
that
this
was
a
safe
change.
It's
it's
been
fine
for
us
too,
for
indefinitely.
A
Okay,
that's
that's
fantastic
I'll
I'll
remove
a
hold,
and
I
I
think
this
one
is
pretty
close
to
go
yeah,
great
glad
that
that's
already
working-
and
I
didn't
realize
it
all
right,
one
more
that
I
really
wanted
to
highlight
here.
It's
a
changelog
for
o43.
A
We
discussed
this
at
the
last
meeting.
I
just
there
is.
There
are
very
minimal
changes
here.
I'll
actually
just
show
you
the
change
log,
one
probably
the
most
noticeable
one
is
if
you're
relying
on
an
implementation
of
o4o,
which
or
o4x
which
most
people
are
you
like
to
be
able
to
deploy
the
web
hook
of
the
same
version
and
that
doesn't
work
right
now,
because
it's
stuck
on
some
old
cool
web
webhook
searching
image
that
is
out
of
date.
A
This
fixes
that
this
also
fixes
some
bugs
and
validation.
I
think
it's
just
a
net
positive,
a
good
thing.
We
should
have
a
practice
of
bringing
these
patch
releases
back.
A
I
can't
there's
no
actual
api
changes
here,
so
I
I
feel
like
this
is
pretty
straightforward,
but
again
would
appreciate
some
lgtms,
and
so
we
can
get.
This
release
kicked
off
yeah.
A
A
I
do
I
do
have
one
more
that
I
I
forgot
to
add
and
I'll
cover
this
really
briefly,
but
otherwise
yeah.
A
This
is
different
than
that.
This
is
one
that,
as
I
understand
it
today
is
a
plug-in
for
traffic,
and
so
it's
here's,
how
you
configure
http
route
gateway
whatever
for
traffic
specifically
and
it
can
be
expanded
in
the
future
for
other
to
support
other
implementations
of
gateway
api
right
now,
but
as
of
today,
it
is
limited
to
a
single
implementation
of
the
api.
A
My
the
way,
I
thought
of
integrations
and
clearly
never
documented
this.
I
don't
think
any
of
us
documented
this
anywhere,
but
what
I
thought
of
as
a
as
an
integration
is
something
that
worked
with
multiple
implementations
of
the
api
that
that
seemed
like
the
the
bar
that
we
should
have,
but
we
don't
have
that
documented
anywhere,
and
I
I
didn't
really
know
how
to
respond.
I
responded
with
some
clarifying
questions,
but
I
wanted
to
bring
that
up
in
this
meeting
as
a
whole
to
see
if
others
had
thoughts
on
how
we
should
approach
this.
D
Yeah,
I
think
it
does
seem
fair
that
there's
a
difference
between
an
integration
that
works
with
lots
of
vendors
lots
of
implementations.
C
D
This
integration
that
works
with
lots
of
implementations
and
an
integration
that
only
works
with
one
implementation,
and
so
you
know
we
need
to
have
a
way
to
call
that
out.
It's
not
just
a
general
integration
yeah
and
we
process
it.
I
think
there's
probably
two
things
that
we
need
to
do
number
one.
We
have
a
thing
at
the
top
of
this
page
saying
implementation
means
and.
D
Means
and
then
you
and
then
we
have
our
something.
That's
like
you.
We
have
some
other
word.
Plugin
extension.
I
don't
know
something
like
that
that
that's
it.
That
means
an
integration
that
works
with
one
implementation.
D
D
We
should
be
trying
not
to
be
king
makers
or,
to
be
you,
know,
favoring
anybody's
implementation
and
so
yeah.
That
kind
of
feels
like
a
minus
but
like
this
is
also
the
place
where
people
are
going
to
go
to
find
out
what
works.
What
so
yeah,
I
don't
know.
A
A
I
can
imagine
integrations,
would
be
a
very
long
list
yeah.
So
that's
my
bias
feel
free
to
join
the
conversation
on
on
this
pr.
I
I
did
not
respond
with
that.
Yet
I
wanted
to
get
kind
of
a
sanity
check
here,
but
I
think
I'll
proceed
with
a
response
like
that:
okay,.
A
No,
it's.
I
think
this
is
ready
to
go.
I
added
there's
like
two
comments
unresolved.
I
pinged
mike
on
slack
last
week
just
to
see
the
status
of
it,
but
I
I
haven't
got
a
response,
but
I,
whenever
like
there's
just
the
tiny
little
tiniest
little
bit
left,
and
I
think
this
is
good
to
go
from
my
perspective.
D
Okay,
yeah,
I
think
I
think
we're
far
better
to
merge
this
and
you
and
say:
okay
is
conformances
for
reference
policy
and
then
we,
if
we
change
the
name,
we
change
the
name,
the
conformance
test
version
anyway,
that's
the
whole
point
of
the
bundle
version,
including
the
performance
tests,
so
that
we
can
do
exactly
that
sort
of
break
and
change
yeah.
That
would
mean
that,
whatever
version
we.
F
D
We
rename
the
reference
policy
that
becomes.
You
know
that
that
would
have
to
be
6-0,
probably
because
that's
a
version
0-6-0,
that's
a
breaking
enough
change
and
yeah.
That,
then,
would
be
the
one,
the
one
where
we
rename
reference
policy
to
use
the
friends
naming
scheme
for
a
moment,
yeah.
C
Yeah
I
have
a
special.
F
C
A
Ironically,
I
think
that
will
mean
reference
policy
will
be
the
first
one
that
covers
all
the
conformance
tests.
Thank
you
yeah,
that's
it!
I
I
wanted
to
highlight
this
and
this
pr
went
in
last
week
and
not
I
feel
like
not.
Everyone
may
have
noticed
it,
but
there's
some
pretty
big
changes
described
here.
A
Most
notable
is
that
in
o50
there
are
going
to
be
two
release
channels,
you're,
probably
familiar
with
release
channels.
By
now
there's
experimental
and
stable.
The
big
thing
is:
if
you
install
stable,
crds,
stable
gateway
api,
it's
just
going
to
include
our
beta
resources,
which
means
it's
just
going
to
include
gateway,
class
gateway
and
http
route.
A
Yeah,
which
I
really
want
to
get,
I
think
I
think
that
will
encourage
and
and
make
us
all
push
to
get
reference
policy
or
reference
grant
or
whatever
we
call
it
into
beta,
but
that
that
is
a
significant
change
as
part
of
this
release.
I
just
want
to
make
sure
it's
not
coming
as
a
surprise
or
if
that
makes
people
uncomfortable.
We
have
that
discussion
before
we
go
forward
with
this,
but
that
is
what
merged
and
what
we're
planning
on
for
next
week
well
for
five.
A
A
You
know
so
yeah,
okay,
I
think
that's
all
for
today,
then
any
last
any
last
things.
D
I
think
probably
the
the
thing
that
the
that
the
the
conformance
test
for
reference
policy
thing
does
raise,
though,
is
that
we
probably
need
to
have
a
way
for
the
conformance
test
to
say:
hey,
I'm
only
doing
only
do
the
stable
resources
only
do
the
alpha
risk
like
do
the
alpha
resources
as
well,
or
something
like
that
right
now,
if
you
install
the
stable,
if
you
install
the
o50
stable
channel,
and
then
we
add
this,
then
you
try
and
run
conformance
it'll
probably
fail
right,
because
it's
going
to
try
and
do
the
conformance
on
an
object
that
doesn't
exist.
A
D
Yeah
yeah,
so
I
I
would
definitely
say
that
there's
definitely
something
some
some
space
for
people
to
think
about
how
we
handle
that
we've
talked
in
the
past
about
ending
up
with
performance
profiles
where
there's
some
way
that
you
can
say.
I
might
only
want
to
be
conformed
for
these
resources
and
there's
some
set.
That's
like
now.
You've
got
to
you've
got
to
be
confirmed
for
the
core
ones,
which
is
probably
gateway,
class
gateway,
reference
policy
or
something
like
that
and
then
there's
a
way
to
say.
D
I
want
to
opt
in
for
the
http
route
and
the
tls
route
only,
which
is
what
contour
will
do,
because
we're
not
good
we're,
never
going
to
do
tcp
route
and
udp
rail
yeah,
and
so
there
needs
to
be
somewhere
for
us
to
do
that
so
yeah.
Just
if
anyone
has
been
with
to
be
thinking
about
commerce
test
more
than
that's
something
that
we
need
to
figure
out
how
to
do
as
well.
D
And
ideally,
there's
some
way
that
you
can
do
that
and
then
the
last
conformance
thing
we're
going
to
have
to
do
at
some
point
is
we
have
to
close
the
loop
and
have
it
so
that
your
implementation
can
run
the
conformance
test
in
some
ci
or
something
like
that
and
then
have
a
the
same?
The
same
as
you
do,
with
kubernetes
conformance.
D
The
last
point
is
the
closing
of
the
loop
and
the
sending
of
the
conformance
report
putting
a
conformance
report
into
a
into
a
repo
somewhere
that
then
you
used
to
have
like
the
you
get
the
conformance
badge
thing
like,
so
that
is
down
down
the
road,
but
like
yeah,
I
just
wanted
to
make
sure
that
that's
on
people's
radars,
because
at
some
point
we
have
to
have
a
thing
where
you
finish
the
conformance
process.
A
Yeah
for
sure
working
on
kubecon
slides,
I
noticed
that
we
have
several
that
are
already
several
implementations
that
are
already
conformant
and
I'd
love
a
way
to
highlight
that
somewhere
and
what
you're
saying
is
is
kind
of
the
way
to
do
that
so
yeah,
but
we
we
have
actually
gone
over
time
today.
So
thank
you.
Everyone
for
a
great
discussion
time
has
flown
by
and
we'll
talk
to
everyone
next
week.
D
Yep,
I
think
we
missed
a
thing
from
travis.
Actually,
oh
that
maybe
sorry
the
chat
in
the
chat
that
about
that
yeah
that
it
looks
like
they
only
create
that
that
cueva
only
creates
a
great
way
and
route
resources
based
on
something
in
their
tool,
which
makes
it
not
so
traffic
specific.
So
but
that's
for
you.
I
made
a
check
until
later.