►
From YouTube: Gateway API GAMMA Meeting for 20230110
Description
Gateway API GAMMA Meeting for 20230110
A
Hello,
everybody
Welcome
to
the
January
10th
occurrence
of
the
Gateway
API
gamma
meeting.
As
a
reminder,
this
meeting
is
governed
by
the
kubernetes
code
of
conduct.
So
everybody
please
be
nice.
We
do
have
an
open,
Agenda
I
placed
it
I
paste
the
pasted.
The
link
in
the
in
the
chat
here
so
feel
free
to
add
any
agenda.
Any
items
that
you
want
to
discuss
to
the
agenda
and
we
will
we
will
get
to
those
I-
will
also
share
my
screen.
B
A
Great
okay,
so
to
start
off
we'll
start
off
with
the
with
a
recap
from
last
week's
meeting.
Oh
and
please
yes,
add
your
name
to
the
attendees
section
of
the
of
the
agenda
as
well.
A
Just
so
we
can
know
who's
interested
and
know
and
Mark
that
you
came
here
so
as
a
recap,
from
our
last
meeting,
I
mentioned
that
I'm
working
on
gamma
conformance
test
for
the
Gateway
API
boarding
over
the
istio
bit
the
SEO
framework
to
a
Gateway
API,
at
least
the
parts
that
make
sense-
that's
been
progressing
slowly,
but
surely
the
next
Mike
had
brought
up
the
idea
of
you
just
wanting
to
see
what
the
status
is
on
PR
1493
and
the
I
approved
it.
A
There
were
some
other
longer
standing
questions,
I
think
the
consensus
that
the
last
time
we
talked
about
the
PRS
that
we
want
to
move
forward
with
it.
There
were
some
questions
about
you
know:
does
it
still
need
this,
but
John
nicely
commented
here
and
such
that
it's
less
important
but
they'll
still
actually
use
use
the
consumer
routing
part
of
this
so
appreciate
that
context.
John
I'll
actually
add
that
to
the
notes
here.
A
So
we'll
still
use
it
just
less
important,
so
I
think
we
are
looking
for
a
review
from
a
couple
of
the
Gateway
API
maintainers
to
kind
of
give
us
the
the
go-ahead,
I
believe
John.
It
might
be
helpful
to
resolve
a
couple
of
these
open
items
just
for
maybe
for
ease
of
review,
since
some
of
the
context
has
been
lost.
So
if
you
wouldn't
mind
doing
that,
that
might
help
get
that
review
out
there.
We
can
get
this
merged
in
any
questions.
On
that
bullet
point.
C
D
D
We'll
we'll
get
to
adding
it,
but
we
don't
think
it's
as
important
for
ambient
in
the
short
term,
most
routes
actually
are
producer
routes
we
found
so
that's
kind
of
what
we're
focusing
on.
B
I,
don't
I
think
that
Linker
D
has
seen
probably
more
situations
where
people
are
interested
in
overriding
consumer
Behavior,
but
yeah,
but
I
mean.
C
Do
we
know
of
any
I
mean
that's
that's
both
good
to
hear.
Do
we
know
of
any
other
implementations
that
are
interested
in
taking
this
potential
part
of
this
back
on
or
I
guess
not
spec
whatever?
This
is
experiment.
This.
F
This
thing
yeah
so
I'm
I
mean
to
get
cilium
service,
my
stream
to
have
a
look
at
this.
Now
that
we've
started
the
new
year.
We
just
all
gotta
yeah
bandwidth,
as
you
know,
but
yeah
so
I
would
like
something
in
service
to
be
able
to
do
some
of
this
at
some
stage
in
the
not
two
years
to
future.
But
it
is,
it's
I
use
the
words
not
too
distant
on
purpose,
though.
C
Okay,
so
I
mean
it.
You
know
one
of
the
things
we
look
for
in
in
many
things
is
just
that.
It's
broadly
implementable.
It
sounds
like
that's
the
case
here
and
not
just
implementable,
but
there's
it
seems
like
there's
intent
to
implement
so
that
that
seems
like
a
good
sign.
I
I
know
it's
been
a
while,
since
I've
looked
at
this
and
I
need
to
revisit,
but
given
that
Keith
is
already
approved,
I
think
it.
It
should
just
be
a
formality
for
any
one
of
us
to
merge
this
in.
F
Yeah
I
think
so
I
see
your
custom
I
think
the
the
key
part
there
is
that
this
is
an
experiment.
So
if
someone
can't
implement
it,
they
try
to
implement
it
and
they
can't.
Then
that
is
also
a
valuable
data
point.
E
I
would
say
it
can
be
implemented
with
a
sidecar
model
and
it
is
very
difficult
or
cannot
be
implemented
without
a
sidecar
model,
which
is
kind
of
the
reason
why
ambient
will
not
implement
it
and
so
I
don't
know.
If
we
want
to
impose
a
you
know,
mesh
requires
sidecars,
or
we
want
to
specify
that
you
know.
Consumer
overrides
are
only
four
sidecar
modes
or
measures
using
sidecars
I.
Don't
think
that's
a
good
idea.
F
I
think
I
think
right
now.
This
is
literally
an
experiment.
We're
going
to
say
this
I
think
what
we're
saying
is
we're
going
to
say:
Hey,
you
know
do
this
and
then,
if
people
can't,
then
we
come
back
and
say
you
know:
if
I
can't
get
psyllium
to
do
this,
because
we
don't
have
a
sidecar
for
doing
that,
then
I'll
then
I'll
come
back
and
say:
hey
I
can't
do
this
I
think
we
need
to
talk
about
this
again,
like
that's
all
like
this
is
very
very
early
for
days.
F
In
my
mind,
sorry
John
yeah.
D
I
completely
agree,
but
also
a
long-term
path
could
be
that
this
is
something
that's.
What's
called
implementation
specific
or
something
right.
Some
implementations
may
not
support
it,
or
maybe
maybe
we
don't
do
that,
but
I'm
just
saying
that's
a
possible
future
outcome
too.
B
Last
week
last
week
we
were
particularly-
and
my
my
apologies
I,
don't
remember
who
was
talking
about
this
kind
of
from
the
sdo
side,
but
there
was
discussion
last
week
about
the
fact
that
it
would
work
fine
in
sidecar,
but
not
so
fine
and
ambient
and
generally
the
place
that
we
left.
It
was
that
everybody
believed
that,
for
the
purpose
of
advancing
the
experiment,
it
should
be
fine
to
allow
that
restriction
to
exist
and
to
allow
istio
in
ambient
mode
to
just
set
a
status.
Saying
hey.
B
This
consumer
route
is
not
being
taken
and
I
I
agree
that
that's
probably
not
ideal
in
the
long
term,
but
I
think
that
for
purposes
of
gathering
actual
use,
data
and
figuring
out
how
things
actually
work
when
we
try
to
go
forward
with
this,
it
sounds
fine
to
me.
E
Yeah
we
do
have
plenty
of
data,
I
mean
is
your:
does
support
consumer
overrides
and
sidecar
mode,
and-
and
there
is
a
lot
of
experience
and
pain
associated
with
that.
That's
a
different
story,
but
I
think
there
is
another
kind
of
path
forward
which
is
kind
of
a
middle
ground,
where
we
may
support
some
subset
of
customer
override
or
some
some
sort
of
variation
which
can
be
implemented
with
waypoints,
with
with
regular
gateways
where
we
take
into
account
the
client
and
we
apply
different
routes
routes.
E
E
F
Yeah
I
agree,
I
agree
with
what
you
said
and
I
think
you've
raised
some
good
points,
but
I
think
that
the
yeah
I
think
this
is
the
best
we
got
for
now,
and
you
know
it
I
think
at
this
point
it's
more
important
that
we
keep
there.
We
keep
some
momentum
when
we
keep
some
things
actually
happening
here
and
we
try
not
to
get
too
stuck
because
I
think
you
know
at
this
point.
F
A
All
right,
so
it
sounds
like
we've
got
a
path
forward
on
on
this
bullet
point.
So
moving
on
through
the
recap,
we
talked
to
good
bit
last
week
about
goals
from
goals
in
the
new
year.
I,
you
know
not
everybody
could
make
it.
It
was,
after
all,
kind
of
the
first
day
back
from
many
of
our
our
holiday
breaks,
but
kind
of
talk
about
our
our
wish
lists
for
this
year
for
gamma,
which
was
very
very
awesome.
A
If
I
mentioned
at
Linker
d,
he
obviously
increases
it
gets
to
a
point
where
they're
able
to
decide.
You
know
if
it's
going
to
be
feasible
for
their
use
cases
implementing
gamma,
that
is
I
mentioned.
That
policy
would
be
great
to
start
flushing
out
still
a
lot
of
open,
open
field
there
console
won't,
wants
to
have
gamma
support
by
yours
in
I.
Believe
Mike
said
that
Kum
wants
to
play
with
it
more.
A
We
had
a
whole
aside
on
that
whole
interesting
aside,
I'll
say
about
multi-cluster
and
some
of
the
barriers,
not
some
hesitations,
I'll
use
that
term
towards
implementation,
yeah
towards
implementing
the
MCS
API
specifically
and
patterns
that
many
of
us
have
observed
in
the
wild,
with
like
a
hub
and
spoke
kind
of
pattern
for
configuration
management
multi-cluster
where
how
the
MCS
API
kind
of
falls
short
in
some
of
those
areas,
questions
about
the
status
of
the
MCS
API.
A
It
is
still
currently
in
Alpha
according
to
the
GitHub
page-
and
you
know
I
I,
mentioned
kind
of
my
hope
for
the
for
this
year.
Even
if
we
don't
write
a
single
word
on
a
gap
towards
multi-cluster
I
want
to
at
least
try
to
bridge
the
gap
between
us
over
here
networking
and
the
Sig
multi-cluster.
It
feels
like
they're,
at
least
from
my
POV,
because
of
the
need
for
there
to
be
an
MCS
controller
and
many
projects
reluctance
reluctance
to
become
an
MCS
controller.
A
We've
got
this
long
feedback
loop
problem,
where
it's
taking
so
long
to
get
MCS
controllers
out
there
to
get
people
playing
with
the
MCS
API
that
we
haven't
really
gotten
clear
signals
here
and
so
I
love
to
see
what
we
can
do
from
an
organization
perspective,
organizational
perspective
to
maybe
shorten
that
feedback
loop
or
get
and
get
multi-clusters
some
some
insight
here.
C
Just
well.
C
I've
got
lots
of
things
to
say
about
monkey
cluster
I
agree,
it's
unfortunate
that
it
has.
It
has
stuck
in
Alpha
for
quite
a
while.
Now
we
I
know
at
at
least
from
gke
side.
We
are
very
interested
in
pushing
this
forward
and
trying
to
do
what
we
can
to
help
move
it
towards
ga
I
know.
For
me,
I
I'm,
very
much
Gateway
focused
and
I've
been
trying
to
formalize
what
multi-cluster
Gateway
even
looks
like
I
know.
C
There
is
at
least
one
implementation
available
and
others
on
the
way
and
I
would
love
to
make
sure
that
what
we're
all
working
on
is
actually
consistent,
because
it's
not
really
well
defined
right
now.
C
So
those
are
things
that
that
are
very
much
top
of
mind.
For
me,
I
I,
would
you
know,
as
always
help
is
welcome
with
these
things,
but
my
my
goal
is
in
the
next
month
or
two
to
have
some
kind
of
PR
out
that
defines
what
multi-cluster
Gateway
means.
C
You
know
what
the
interaction
between
Gateway
API
and
multi-cluster
services
and
then
finally
I
this
is
this-
is
not
something
that
I
personally
will
be
leading,
but
I
I
am
I'm,
definitely
talking
with
people
that
will
be
helping
with
the
multi-cluster
service
API
and
hopefully
again
if
anyone
is
interested.
This
is
an
area
where
we
really
need
different
implementations
of
the
API
different
interested
parties
to
to
be
involved.
C
You
know
with
Gateway,
we've
been
really
fortunate
and
that
we've
had
a
lot
of
implementations
that
can
help
us
meet
criteria
for
graduating
with
multi-cluster
service,
there's
not
quite
as
many
implementations
today,
so
it
would
be
helpful
to
get
whatever
perspective
we
can
from
as
many
different
perspectives
as
possible,
so
that
that's
my
my
view
of
multi-cluster
service,
but
very
much
top
of
mind
for
me.
Going
into
this
year.
G
It's
definitely
something
we're
interested
in
the
comfortable
plan
too
yeah.
It
feels
like
it's
just
it
stabilized
earlier
than
I.
Think
a
lot
of
our
users
and
customers
are
starting
we're
starting
to
get
to
like
the
actual
problem.
That's
intended
to
solve.
So
I
think
that
is
a
bit
of
an
explanation
for
why
it
seems
to
have
like
not
moved
as
actively
perhaps
but
we're
definitely
seeing
interest
in
that
subject.
G
B
There's
a
lot
of
stuff
there.
That
I
would
certainly
like
to
learn
more
about
as
well.
A
Yeah
likewise
so,
hopefully
we
can
make
some
progress
on
that
here
in
the
New
Year
I'm
excited
about
that
I
had
a
new
topic
here.
A
As
far
as
wish
lists
around
operations
I
was
talking
to
who
was
it
I
was
talking
to
I,
forget
well,
I
think
it
was
I
think
with
some
of
the
SEO
folks
about
you
know,
we've
had
conversations
hearing
gamma
about
the
you
know
the
idea
of
representing
certain
mesh
operations
within
the
API
things
like
enrolling
a
a
workload
into
the
mesh
doing
that
with
a
resource
or
so
or
should
it
all
just
be
transparent?
A
Should
there
be
a
standard
way,
I
think
John
you
and
I
I
was
around
for
a
conversation.
We
were
having
about
something
in
the
ambient
world
with
operations,
and
someone
asked
if
there
is
a
standard
way
of
doing
that.
Do
you
do
you
have
any
more
context
on
that.
D
I,
don't
know
exactly
what
you're
referring
to,
but
I
can
give
some
insight
on
how
ambient
works,
so
Easter
sidecar
is
like
I,
think
the
rest
of
measure
this
label
on
the
namespace
for
ambient
those
kind
of
two
parts.
One
is
whether
you're
capturing
The,
L4
traffic
and
the
other
is
actually
deploying
these
Waypoint
proxies.
D
So
whether
things
are
captured
ml4
perspective,
that's
a
label
on
the
namespace
today
for
waypoints
we're
actually
deploying
them
with
the
Gateway
API,
with
its
own
Gateway
class.
That's
called
like
Easter
at
Waypoint
or
something
does
that
answer
your
question.
A
D
Oh
yeah
I
mean
that's.
That
would
be
useful
for
us
for
one
just
normal
Ingress
gateways,
but
also
waypoints,
because
they're
also
represented
as
gateways
and
they're
also
deployments
in
the
cluster.
So
anything,
that's
you
know.
A
Gateway
creates
a
deployment.
We
have
potential
to
upgrade
those
resources
Etc,
but
that's
well,
I,
guess
even
for
master.
D
You
may
want
to
configure
the
sidecar
so
I
suppose
it's
similar,
but
for
us
we've
already
solved
those
problems
for
sidecars,
whereas
we
haven't
solved
the
problem
of
oh
now
we
have
a
Gateway
that
deploys
a
deployment
rather
than
like
a
Helm
threat
that
had
all
these
customizations
but
I
could
see.
You
know
there
could
be
a
word
where
there's
a
match
and
it
has
whatever
customizations
you
have
and
whatnot.
E
One
point
I
want
to
make
is
at
least
for
Ambi
and
and
also
I
suspect
for
other
kind
of
mission
implementations
in
general,
because
it's
operating
at
not
level
and
because
it's
capturing
all
traffic,
it's
not
very
friendly
to
multiple
meshes.
What
we
discussed
in
the
past,
because
really
it's
very
difficult
to
have
a
pair
node
component,
that
it's
interactive
with
CNN
capture,
all
the
traffic
and
then,
for
example.
E
It
is
spitting
it
becomes
extremely
complicated
and
it's
also
kind
of
difficult
to
say
that
OneNote
has
measure
underground,
doesn't
have
mesh
so
All
That
Remains
is
really
ability
to
opt
out
or
obtain
some
workloads,
which
again
is
highly
debatable,
because
in
Siena
you
cannot
opt
out
somewhere
close
to
use
wire
guard
or
ipseconds.
Some
workloads
do
not
use
it.
It's
kind
of
a
general
cni
option,
so
I
suspect
the
use
case
for
having
some
workload
select
a
measure
dimensionable
be
mesh
disabled
is
it
becomes.
E
To
to
justify
to
to
to
to
find
the
common,
because
I
suspect
other
measures
will
have
the
same
kind
of
some
issues
on
by
default,
to
make
it
easy.
So
we
don't
have
user
having
to
worry
about
the
mesh.
So
maybe
we
should
postpone
it
until
we
have
more
more
feedback
both
on
Ambience
and
other
measures
that
want
to
be
on
by
default.
A
Got
it
got
it
got
it
so
yeah.
This
is
something
that
just
based
on
slack
in
our
conversations
that
I
didn't
know.
Other
folks
had
thoughts
on
any
non-stio
meshes
Representatives
have
any
opinions
on
this
operations.
Direction.
F
I
think
yeah
I
mean
I.
Think
that
there's
definitely
going
to
be
you
mean
for
psyllium
right,
like
there's
no
real
way
to
opt
in
or
out
right.
Like
you,
you
opt
in
around
on
a
clock
for
level
because
it's
a
CNA
so
like
it's
less
meaningful
for
for
us
yeah,
so,
but
that
doesn't
mean
that
there
can't
be
ways
to
handle
it
for
for
people
who
can
opt
in
or
out,
and
so
the
you
know,
but
again,
I
I
agree
with
costing
that
I
think
we
really
just
need
more
implementation.
F
Experience
I
think
going
too
far
down
this
thing
is
is
a
little
is
just
runs
the
risk
of
that
we're
locking
in
things
that
we're
gonna
have
to
change
later,
so
yeah
I
think
I
I
think
it's
a
bit
risky
is
to
make
to
go
too
far
on
those
yeah,
yeah
I
think
maybe
yeah,
something
that
opt
out,
maybe
but
yeah
that
that
may
be
yeah
like
I
would
like
to
see
so
anytime.
F
F
B
B
I
think
if
you
I
think
I've
heard
you
say
before
Nick
something
like
root
of
all
evil
or
something
like
that.
I
looking
over
this
one
I
think
it
actually
might
be
more
important
to
have
a
way
to
opt
out
ports
than
a
way
to
opt
out
specific
pods.
B
The
there
are
definitely
cases
where,
for
example,
there
are
cases
in
Lincoln
Land
where
it
becomes
very
important
to
be
able
to
tell
linkerty
the
Ingress
is
meshed.
Please
do
not
even
look
at
incoming
traffic
coming
from
outside
of
the
cluster
into
the
Ingress,
because
if
Linker
D
does
look
at
it,
then
it
becomes
correspondingly
more
difficult
to
preserve
the
original
Source
IP,
and
so
that's
an
immediate
operational
concern
that
I
think
is
going
to
end
up
being
important
in
a
shorter
term
than
otherwise
I.
Don't
think
it's
something
we
need
right.
B
A
E
One
more
thing
here,
I
mean
I-
think
for
this
particular
thing
about
exclude
in
particular,
it
will
be
various
looking,
not
not
crd,
but
some
place
in
the
API
to
specify
postplex
good.
So
a
user
who
is
implementing
G
approximation,
for
example,
doesn't
have
to
figure
out
what
each
mesh
implementation,
what
they
need
to
do.
I
mean
history:
it
is
a
label.
E
I
know
you
don't
like
labels,
but
that's
what
we
have
create
an
annotation
but
I'm
sure
each
vendor
will
have
something
to
allow
opt
out
for
a
particular
work,
and
it
will
be
very
painful
for
users
to
to
at
least
for
proximately
grpc,
to
do
to
deploy
stuff
just
support
these
students.
Someone
else
and
the
rest
will
be
your
pain.
A
Believe
when
I
see
you
have
a
question
in
the
chat
about
metrics,
could
you
explain
some
more
about
what
you
mean
by
that.
H
Yeah
the
reason
you
know
we're
talking
about
operation
right
many
times
operation
is,
you
know
one
of
the
key
factor
you
know
thing
of
operation
is
metrics,
so
my
question
is:
are
we
going
to
standardize
our
metrics
regarding
the
Gateway
objects
or
we
just
rely
on
existing
data
plane?
You
know
like
sdo,
you
know
you
have
sdo
metrics,
you
know.
Every
data
plan
has
its
own
metrics
so
and
boy
metrics
or
we
commonize
them
into
a
Gateway
metrics.
H
C
Know
that
that's
sort
of
like
a
question
and
I
know
that
has
come
up
before
in
an
issue
and
I
think
the
response
to
the
issue
at
that
time
is
it's
a
great
idea,
but
we
just
don't
have
time
to
take
it
on
trying
to
you
know.
Gateway
API
already
has
a
massive
scope,
so
it
was
a
not
a
no,
but
not
yet
answer
at
the
time
we
can.
C
We
can
reevaluate
that,
but
right
now
we
have
so
much
going
on
I'm
tempted
to
say
not
yet
is
probably
still
the
right
answer.
The.
E
Is
a
semantic
conventions
which
defines
names
for
HTTP
metrics
and
attempt
to
standardize
across
Telemetry
vendors,
what
metric
names
and
dimensions
to
be
used
and
I
think
which
transfer
should
not
be
that
we
is
not
in
school
forming
which
clearly
should
not
be
our
program
here.
It's
enough
duplication
in
Telemetry
about
each
vendor
defining
its
own
metrics.
Let's
give
open
Telemetry
a
chance
to
succeed.
A
I
completely
agree
with
that:
I
I
think
there
is
a
lot
that
we
as
a
field
are
experimenting
with
when
it
comes
to.
What
is
a
metric
specification?
Look
like
open
Telemetry
does
come,
you
know,
is
the
best
I've
seen
in
this
space
at
least
so
far
and
I
am
very
investors,
probably
I'm,
very
I'm,
very
optimistic
about
its
ability
to
succeed
with
other
things
like
SMI
traffic
metrics
have
have
failed,
even
when
it
comes
when
it
comes
to
naming.
A
It
would
be
great
to
have
standard
names
and
shapes
how
you
use
shapes
as
a
loose
collection,
depending
on
what
your
metric
system
is,
but
I
mean
to
some
of
the
to
the
point
of
some
of
the
folks
in
chat,
I,
think
yeah
cost
and
support
and
chat
like
cardinality,
starts
becoming
an
issue
depending
on
the
shape
of
certain
things,
and
now
you've
got
to
start
designing
for
a
certain
cardinality
in
certain
use
cases,
and
just
to
take
a
very
an
example.
That's
close
to
the
folks
here.
A
A
Just
slightly
different-
and
these
are
proxies
doing
largely
the
same
job,
but
in
different
context
and
to
Rob's
point
you
know,
we've
got
a
lot
of
things
to
take
a
look
at,
but
I
don't
even
know
for
me
personally.
If
I
want
I
need
to
do
a
lot
more
research
on
Metric
storage
and
things
like
that.
First
I'd.
B
A
A
All
right
fantastic,
so
this
will
be
hopefully
crossing
fingers,
quick
agenda
item
here
as
far
as
gameless
affordability,
I
plan
to
bring
this
up
in
the
Gateway
API
meeting
yesterday,
but
had
a
last
second
conflict
and
couldn't
make
it
it's
on
the
schedule
for
next
week,
but
I
wanted
to
well.
Actually
next
week
did
we
just
did
they
decide
to
cancel
it?
Rob
and
Nick.
C
It
will
exist,
but
it
will
not
have
an
agenda
per
se.
We
we
expect
light
attendance,
but
anyone
is
always
welcome.
F
A
Got
it
so
hey
this?
Might
this
is
good
a
time
I
think
you
can
talk
about
Game
of
supportability,
so
you
know
with
our
goal
of
trying
to
get
the
first
gamma
release
out
in
Gateway
API
07l.
If
it
makes
sense
to
me
to
try
to
make
sure
we
get
our
our
naming
right
when
it
comes
to
the
supportability
and
conformance
levels
of
gamma
and
what
it
is
within
Gateway
API,
the
the
best
that
I've
come
up
with
this
here
in
the
agenda.
A
That
gamma
is
a
provisional,
get
like
a
series
of
at
the
moment,
provisional
gaps
in
the
experimental
Channel
that
have
some
parts
of
its
spec
that
have
core
conformance,
which
is
a
little
bit
of
a
weird
mental
model,
at
least
for
me,
but
I.
Think
of
things
like
you
know
our
HTTP
routes,
API.
A
If
you
want
to
implement
gamma,
do
you
I,
don't
see
the
way
you
implement
gamma
without
doing
supporting
HTTP
route
in
the
way
that
you
know
the
Gap
is
specified
so
in
a
sense
that
that's
part
of
core
conformance
for
mesh,
but
so
far,
they're
being
those
directions
are
being
powered
by
provisional
gaps
and
everything
that
gamma
is
is
within
the
experimental
Channel
I
wanted
to
see.
If
that
is
the
Direction,
you
want
to
go.
A
F
I
mean
I
think
in
terms
of
in
terms
of
like
what
we've
talked
about
so
far
you
know.
Route
binding
to
service
is
one
of
those
things
where
you,
the
spec,
does
not
stop
you
from
doing
that
right
now,
right,
like
there's,
nothing
in
the
Specter
says
you
can't
do
that.
F
You
know
there's
nothing
in
you
know
so
like
it.
It's
just
implementation,
specific
behavior
so
like
that
I
think
that
being
a
provisional
Gap
is
fine,
because
it
doesn't
actually
need
spec
changes
and
as
long
as
we
stuff
as
long
as
you
do
stuff,
where,
as
long
as
most
of
the
other
stuff,
we
do
doesn't
add,
Fields
or
sort
of
you
know
do
other
sort
of
API
changing
kind
of
things.
F
Then,
then,
all
then
all
gamma
uses
a
series
of
provisional
gaps
that
that
are
implementation,
specific
support,
and
so
that's
fine
right
like
it's
more.
If
we
want
to
start
adding
conformance
tests
that
becomes
a
little
bit
more
complicated
because
then
it's
like
you
we're
kind
of
adding
conformance
tests
for
implementation,
specific
behavior,
which
we
don't
generally
do,
because
that
makes
it
extended.
F
That
doesn't
mean
that
we
can't
do
it.
It
just
means
that
if
we
are
I
think
once
we
start
talking
about
adding
conformance
tests,
that's
when
we
sort
of
need
to
be
clear
about.
What's
the
story
about
these
conformance
tests-
and
you
know,
do
we
add
them
as
like
a
separate
set
of
conformance
like
a
separate
Suite?
F
Maybe
you
know
like
you
know,
because
we
have
the
sort
of
the
sweet
level
functionality
and
we
add
a
gamma
Suite
that
you
can
then
run
in
addition
to
the
to
the
to
the
sort
of
the
Ingress
Suite.
F
That
seems
like
a
reasonably
clear
way
to
do
this
and
say
hey:
this
is
the
game
the
gamma
Suite
is
is,
is
completely
experimental
and
you
you
is
not
required
for
any
level
of
performance.
It's
there
to
sort
of
for
us
to
coordinate
on
developing
conformance
tests
for
gamma
behaviors
or
something
like
that,
but
but
then
and
then
yeah
and
then
yeah
and
then
at
if
and
when
we
start
adding
fields
or
changing
the
behavior
Fields
or
any
values
to
fields
or
doing
any
of
that
stuff.
H
G
So
I
can
add
a
little
bit
of
context
to
this.
One
of
the
specific
cases
that
prompted
this
discussion
initially
was
John's
open,
PR
for
client-side
routing,
as
opposed
so
that's
like
service,
consumer
routing
it
like
or
like
egress
routing,
as
opposed
to
like
service
owner
producer
routing.
G
That's
something
that
felt
like
it
would
be
appropriate
to
a.
If
you
choose
to
implement
this.
It
should
behave
in
this
way
so,
basically
like
our
extended
spec
or
our
expense
performance
level
versus
the
producer-defined
routing
that
exists
currently
in
the
provisional
experimental
Channel,
which
felt
like
core
functionality
for
gamma.
G
So
it's
almost
like
we
kind
of
want
a
different
axis
for
gamma
to
have
its
own
conformance
levels,
because
we
know
that
well
like
or
the
same
level
because
in
in
the
gamma
column,
just
kind
of
standardize
that,
because
we
know
like
there's
going
to
be
Ingress
only
implementations
of
Gateway
API
that
won't
have
any
of
this
and
that's
where
it
feels
inappropriate
to
introduce
any
gamma
Concepts
in
an
extended
or
core,
maybe
not,
but
at
least
core
and
being
able.
F
To
differently
yeah
I
think
the
thing
that's
relevant
is
we
don't
talk
about
it?
A
lot
but,
like
you
know,
the
things
that
are
currently
experimental
in
the
Ingress
Gateway
API
are
have
extended
conformance
right.
So
you
know,
but
you
know
so
they
already
have
extendables.
So
we
have
experimental
things
that
have
extended
performance.
The
you
I
mean
one
of
the.
We
don't
promote
those
to
Beta
until
they
have
conformance
tests
right,
so
I
think
I.
F
Think
the
it
feels
to
me
like
the
sort
of
the
canonical
place
where
we,
where
we're,
really
recording
how
the
there's
two
canonical
places
where
we
record,
if
things
are
extended-
or
you
know
core
or
whatever
that's
on
the
field
itself,
but
for
a
behavior
that
sort
of
crosses
fields
or
does
isn't
isn't
specified
in
a
single
field,
then
it's
the
conformance
tests,
and
so
that's
where
I
think
us
saying
hey.
F
You
know
gamma
gets
its
own
tweet,
you
know
is
a
good
way
to
sort
of
keep
that
as
an
orthogonal
concern,
largely
and
because
we're
because
you
know
it's
a
bit
nuanced
as
to
like
as
long
as
we're
not
talking
about
specific
Fields
I.
Think
talking
about
specific
Fields
is
much
more
complicated
because
we
need
to
be
like
yo
hey.
F
You
know
this
field.
You
know
this
field
is
standard
in
in
Ingress,
but
is
experimental
in
in
camera.
All
right
like
that's,
that
is
a
complicated
conversation,
but
but
I
think
the
good
part
is
that
right
now
we
are,
we
are
all
the
experimental
stuff.
We're
doing
is
explicitly
trying
not
to
add
fields.
C
No
I
mean
I
agree
with
everything
you're
saying
I
just
wanted
to
bring
in
a
few
other
Concepts
we've
talked
about
before
one
I
think
Nick
I
think
you're
the
one
who
had
this
idea
previously,
that
of
conformance
profiles.
The
idea
that
you
know
there
we've
thought
a
lot
about.
C
There
are
going
to
be
L4
implementations
of
this
API
still
on
the
Ingress
level,
but
they're
not
going
to
support
HTTP
route
more
than
likely
right,
similar
for
L7,
only
implementations
of
the
API
and
that
doesn't
even
start
to
get
into
gamma.
But
the
idea
that
we
could
Define
common
conformance
profiles
or
I
again,
I'm,
not
I'm,
not
sure
exactly
what
this
looks
like.
C
But
this
all
comes
back
to
that
discussion
that
we
had
yesterday
for
for
anyone
on
the
main
meeting
yesterday
about
the
the
idea
that
we
want
Gateway
classes
to
be
able
to
describe
the
features
they
support
in
a
meaningful
way
and
and
the
schema
for
how
they
do.
That
could
be
very
helpful
and
defining
so
much
else
right,
defining
a
table
at
some
point
somewhere
that
says
this
Gateway
class.
This
implementation
supports
these
features
and
we
have
a
clearly
defined.
C
You
know
idea
of
what
that
feature
means
right.
I,
don't
know:
I
I
think
this
is
going
to
be
an
ongoing
discussion,
but
I
very
very
much
want
to
see
this
go
forward
because
I
know
we
are
rapidly
growing
out
of
the
system.
Our
our
conformance
tests
currently
support,
and
it's
part
of
I
think
is
it
John,
whoever
whoever's
leading
the
effort
towards
gamma
conformance
this?
This
feels
like
Beyond,
just
figuring
out
how
to
do
tests.
It's
also.
How
do
we
represent
conformance
levels?
H
A
That
makes
sense
the
the
thing
that
so
as
far
as
Gateway
API
game,
that
all
makes
sense
in
sounds
like
it's
largely
a
problem
for
future
us,
like
you,
said,
I
do
I'll
get
that
later.
I
want
to
zero
in
on
something,
though,
because
what
you
describe
being
able
to
have
Gateway
class
described
the
things
it's
able
to
do
is
a
very
useful
capability
that
all
kubernetes
could
use.
A
I
was
just
having
a
internal
chat
with
the
like
Cal
about
this.
The
other
day
in
that
like
kubernetes
as
a
as
a
orchestrator
can't
say.
Oh
no
I
can't
do
this
request,
because
the
things
in
my
question
don't
have
this
capability.
Is
there
any
I?
Think
I.
Think
Gateway
class
is
a
great
place
to
start
that
that
work,
it
would
be
fantastic
if
that
schema
could
be
General
enough
to
be
used
elsewhere.
A
I
know,
that's
a
that's
a
ton
of
scope,
that's
a
giant
amount
of
scope,
but
did
want
to
call
that
out
because
that
definitely
be
on
my
wish
list.
Don't
know
about
anybody
else.
C
Yeah
I
feel
like
there's
a
tiny
bit
of
overlap
here.
I,
don't
know
if
anyone
saw
the
dock
that
Tim
sent
out
yesterday
about
rethinking
Alpha,
Beta,
And,
ga
and
one
of
the
one
of
the
bits
of
that
was
trying
to
provide
a
better
ux
around
using
Alpha
features.
C
It
it
seems,
related
the
the
idea
of
providing
a
way
to
to
opt
into
things,
but
also
to
communicate
that
these
things
are
experimental
may
not
be
supported.
Maybe
there's
a
way
we
can
tie
in
there.
I
I'm,
not
sure
I,
definitely
agree
with
that.
Longer
term
goal.
I
I
know
you
know.
Cal
definitely
brought
that
up
earlier
as
well
in
the
context
of
Gateway
API.
C
When
we
describe
this
idea
to
him,
I
I,
don't
know
I
I'm
I'm
torn
between
that
I
want
to
get
something
out
and
I
want
to
solve
the
bigger
picture
item
as
well,
so
I
I,
my
best
guess
is
we
have
to
start
relatively
small,
but
as
long
as
we
have
a
path
to
a
better
solution
that
it's
kubernetes
wide
I
think
that's
good
directionally.
F
And
I
think
that
there's
some
value
in
US
having
a
crack
at
doing
something
to
provide
data
for
the
kubernetes
wide
efforts
as
to
what
works
and
what
doesn't
as
well.
Given.
F
That
actually
yeah,
given
that
it's
much
easier
for
us
to
do
things
given
the
worst
crd
rather
than
a
core
resource
that
needs
feature
Gates
and
a
bunch
of
blah
blah
blah.
You
know
it's
much
easier
for
us
to
sort
of
have
a
crack
at
doing
something
like
this.
I
mean
it.
It
has
been
my
dream.
F
I
think
I
spoke
about
this
yesterday,
but
it
has
been
my
dream
for
a
long
time
that
the
yeah
it
feels
to
me,
like
the
ideal
way
for
conformers
to
work,
is
that
all
of
our
performance
tests
currently
have
a
gay
work
require
Gateway
class
to
be
installed,
and
you
need
to
tell
tell
the
components
test
what
Gateway
class
it
is
in
in
my
dream
world
that
way,
Gateway
class
tells
you
all
the
features
that
you
need
to
test
and
all
of
the
things
that
you
need
to
do
and
then
your
you
know
the
conformance
tests
test.
F
I
think
we
talked
a
little
bit
about
how
it
would
be
really
nice
to
have
a
implementation
table
that
says
implementation,
here's
what
it
supports
and
the
way
to
make
that
people
neutral
and
not
you
know,
sort
of
you
know
a
not
require
manual.
Updating,
but
also
B,
be
like
fair
to
everybody.
Is
that
that
thing
reflects
you
know
if
you
get
a
checkbox,
that
means
you
pass
the
test
or
a
suite
of
tests
and
yeah.
That
makes
it
fair
for
everybody
that
you
know
hey.
F
If
your
thing
doesn't
work,
then
you
either
need
to
fix
your
thing
or
fix
the
test
to
make
to
make
it
work
right
like
and
that
way
everybody
it's
the
same
for
everybody.
It's
100
Fair,
it's
machine,
fair
and
you
know-
and
we
also
get
this
documentation
of
what
everything
can
do,
and
you
know
you
get
the
ux
of
a
bunch
of
stuff
and.
F
Been
my
dream
for
a
long
time,
I
think
I,
communicated
or
poorly
to
start
with,
but
yeah,
but
I
think
this.
This
problem
is
like
is
is
an
isomer
of
the
same
problem.
I
teach
some
chemical
terms
for
a
minute.
A
Yeah
that
all
sounds
sounds
great
to
me,
it'll,
be
regardless
on.
If
the
The
Wider
problem
is
solved.
First
for
Gateway
class
itself,
first
I
I
sounds
like
Gateway
classes.
Probably
the
quickest
way
to
get
some
things
out
there
for
people,
but
I'm,
looking
forward
to
seeing
how
getting
excused
your
feedback
from
that
one,
because
that
would
be
super
exciting
to
have
that
kind
of
capability
anywhere
in
kubernetes,
even
if
it
starts
here.
First,
okay,
that
is
all
I
had
on
the
Gammas
portability
topic.
B
F
Yes
and
that
performance
for
gamma
is
complicated
because,
because
of
the
interaction
with
the
API
resources,
but
for
now
a
separate
conformance
Suite
will
cover
most
of
the
stuff.
That
we
need
is
my
opinion.
A
A
Okay,
it
sounds
like,
but
I
take
that
silence
to
me.
No
so
I
can
give
everybody
about
10
minutes
back.
Thank
you,
as
always
for
for
companies
meetings
participating
next
week.
Our
meeting
will
be
on
Tuesday
at
8,
8
A.M,
Pacific
time
the
I'll
be
looking
for
some
other
topics,
but
Belen
you
mentioned
in
the
past
having
some
feedback
about
the
HTTP
route.
Gap.
Are
we
getting
close
to
having
to
add
a
discussion?
I
hope
we
can
discuss
about
it.
B
Maybe
in
a
couple
of
weeks
is
what
I'm
thinking
yeah,
so
you
know
fingers
crossed
there.
B
Engineering
are
busily
working
away
with
some
stuff
I
need
to
finish
some
training,
certification,
stuff
that
I'm
doing
and
then
I'll
have
a
chance
to
put
a
little
bit
more
bandwidth
on
that
as
well,
I
hope
so
yeah
in
another
couple
of
weeks,
I'm
hoping
we
can
have
I,
don't
know,
let's,
let's
just
say,
early
February
and
bake
a
little
bit
of
buffer
into
there
to
be
able
to
talk
a
little
bit
more
meaningfully
about
that.
A
All
right
that
sounds
good
in
the
meantime,
we'll
go
through
our
I,
always
have
an
hour
release
checklist
for
gamma
and
talk
through
any
remaining
action
items
cool
all
right.
Thank
you.
Everybody
take
care,
have
a
good
rest.