►
From YouTube: Gateway API GAMMA Meeting for 20230516
Description
Gateway API GAMMA Meeting for 20230516
A
Hello,
everybody
Welcome
to
the
May
16th
and
since
of
the
Gateway
API
gamma
meeting.
This
meeting
is
governed
by
the
kubernetes
code
of
conducts,
which,
in
short,
boils
down
to
be
nice
to
be
respectable
to
everybody.
The
we
have
an
open
Agenda
here
for
the
API
government
meeting.
The
displays
already
beat
you
to
it
put
in
the
meeting
notes
in
the
description.
A
So
if
you
have
something
you'd
like
to
to
talk
about
to
the
community,
then
feel
free
to
do
so,
it
should
be
open
for
you
to
make
edits.
A
Oh
also,
as
part
of
on
this
agenda,
please,
you
know
to
admit
your
attendance,
so
we
can
make
sure
our
meeting
times
are
accommodative.
So
if
you
would
put
your
name
in
organization
and
affiliation
there
in
the
attendees
list,
all
right,
we've
got
a
couple
of
backup
topics,
but
I
think
we're
waiting
for
certain
people
to
take
care
of
to
premising
those
topics
so
we'll
skip
those
for
today.
A
The
first
topic
we
have
on
your
own
original
comes
from
Rob,
giving
doing
some
administrative
stuff
as
far
as
gamma
timing.
Take
it
away.
Yeah.
B
Yeah,
thank
you
so
I
know
this
is
actually
the
first
time
we've
had
two
Gateway
related
meetings
in
the
same
day,
depending
on.
B
Kudos
to
the
to
everyone,
who's,
making
it
to
their
second
one
for
those
that
are
yeah.
We
we
tried
out
a
morning
meeting
at
least
Pacific
Time
morning
meeting
for
Gateway
API
today,.
B
Yeah
it
did,
there
was
a
lot
of
attendance
there.
I
I
think
it
was
at
least
equal
to
the
kind
of
attendance
we
get
in
the
afternoon
in
a
later
session.
So
that's
great.
Maybe
we'll
need
to
do
that.
More
often,
it
also
seems
like
I,
don't
want
to
go
too
far
down
rabbit
hole,
but
it
does
seem
like
maybe
we're
seeing
the
same
pattern
with
gamma
that
the
earlier
meetings
are
at
least
slightly
more
popular,
so
yeah
interesting
with
that
said
my
main
goal.
B
The
actual
purpose
of
the
agenda
was
that
we're
targeting
Gateway
API
v0.8,
soon
like
in
six
or
so
weeks
soon,
and
the
entire
purpose
of
this
release
is
to
have
a
release
that
says,
gamma
is
experimental
if
other
things
happen
to
sneak
in
cool,
but
this
is
this
is
the
release
for
gamma.
So
that's
why
it
feels
like
it's
worth
bringing
up
and
discussing
here
that
six
week,
Target
is
based
on
rough
estimation,
discussion
whatever.
B
C
So
one
of
the
things
blocking
considering
gamma
experimental
is
this
Gateway
and
mesh
interaction
thing
we're
in
an
interesting
situation.
There
I'm
going
to
try
to
summarize
this
extremely
quickly
because
again,
I
have
a
hard
stop
in
22
minutes.
So
the
summary
goes
like
this.
As
Mike
and
I
kick
things
around.
C
But
they
also
have
the
interesting
property
that
it
feels
wrong
to
say
that
if
you're
using
a
layer
set
and
load
balancer
from
a
cloud
provider,
then
your
mesh
is
just
kind
of
out
of
commission
for
any
traffic
coming
through
the
English
controller.
So
happily
we
have
Rob
on
the
call
who
can
hopefully
answer
some
of
these
questions,
at
least
with
respect
to
gke,
and
maybe
a
you
know,
a
broader
point
there
is
we
aren't
100
going
to
need
to
talk
to
more
people
than
just
Rob
I.
C
B
C
B
Yeah
well
I'll
I'll
say
that
we
also
have
Lewin
on
the
call
which
I
guess
you're
working
on
something
slightly
different
than
this.
But
still
maybe
you
have
some
some
context
here,
but
I'll
just
throw.
In
my
perspective,
maybe
I'll
take
one
step
back
if
we
had
an
L4
load
balancer
and
that
L4
load
balancer
was
targeting
node
ports
on
nodes
on
their
node
port.
Is
that
close
enough
that
it's
something
like
that
mesh
can
intercept
so.
C
C
Let
me
describe
an
existing
implementation
pattern
that
I
have
seen
work
really
really
well
in
that
existing
implementation
pattern.
You
have
an
L4
load
balancer
from
your
cloud
provider
that
targets
an
Ingress
controller
in
front
of
the
mesh
and
all
the
problems
go
away,
and
this
works
really
well.
C
C
B
C
C
E
This
two-phase
routing
would
be
fine,
and
that
would
be
a
design
decision
we
make
you
always
do
two
tiers,
but
in
a
world
where
not
everyone
has
mesh.
That
means
sometimes
you're
in
one
tier,
sometimes
you're
in
two
tier
and
I,
think
that
is
where
all
the
problems
arise.
From
a
user
experience
perspective
now,
I
also
have
a
lot
of
implementation
concerns,
but
from
a
user
experience
perspective
that
that's
my
main
concern
so
like
adopting
a
mesh
would
be
it's.
A
I'll
also
add
some
user
for
experience
perspective
the
the
benefit
of
having
that
Ingress
that
two-tiered
approach.
That's
not
something
you
have
to
manage
and
scale
alongside
with
your
Cloud
load
balancer
and
that's
not
super
fun,
especially
for
those
customers
who
are
running
at
sale,
because
the
load
balancer
is
typically
symmetric
like
transparent
to
you.
You
pay
to
manage
that,
but
if
I
require
an
invest,
controller
now
then
set
the
that's
delivered.
B
Maybe
I
should
take
a
step
back
and
ask
what
are
we
trying
to?
What
are
we
hoping
that
mesh
would
add
in
this,
and
maybe
that's
a
dumb
question,
but
you
know
if,
if
the
options
are
managed,
load,
balancer
sends
direct
depod
or
managed
load
balancer
sends
to
intermediate
thing
that
mesh
intercepts
and
then
sends
to
pod.
B
What
are
we
hoping
that
mesh
adds
in
that
path,
and
is
it
similar
in
scope
to
what
we're
configuring
with
Gateway
API,
like
I,
I
I
I,
feel
like
there's
some
obvious
answers
here,
but
I
just
want
to
make
sure
we're
all
on
the
same
page.
Here.
C
From
my
perspective,
the
biggest
the
biggest
concern
that
we
run
into
you
know
talking
to
people
using
liquidy
who
are
also
trying
to
go
through
and
use
layers
up
and
load
balancers
that
we
can't
put
sidecars.
Next
to
this
thing,
you
run
into
things
like
they
cannot
participate
in
mtls.
They
lose
a
fair
amount
of
metrics
that
Linker
D
would
otherwise
collect.
They
lose
a
certain
amount
of
resilience,
because
we
there
are
things
that
we
can
no
longer
do
with
the
mesh
that
we
end
up
not
being
able
to
do
with
the
load
balancer
either.
C
E
Why
don't
you
just
add
mtls
and
let
the
load
balancer
do
the
load,
balancing.
E
C
E
E
E
C
Like
yeah
I
mean
it's
a
it's
a
thing
that
we
have
to
figure
out
how
to
do
it's
a
thing
where
to
get
it's
an
area
where
yeah.
If
we
want
the
mesh
to
be
able
to
help
you
with
that
aspect
of
security,
then
we
have
to
have
the
load
balancer
able
to
support
interoperability
with
the
mesh
on
the
you
know,
as
mtls.
C
That's
I
feel
like
that's
a
little
outside
the
scope
of
gamma
right
now,
just
because
we
have
always
kind
of
considered
mtls
to
be
sort
of
table
Stakes
for
the
mesh,
and
we
assume
it's
configured.
So
that's
a
an
interesting
thing
to
consider
and
by
the
way,
don't
get
me
wrong.
I
would
love
to
see
the
cloud
load
balancers
interoperating
with
linkertys
mtls.
That
would
be
a
nice
thing
from
my
perspective,
I
would
also
love
to
be
able
to
just
stick
an
MTL
or
stick
out
Linker
design
car
next
to
it.
B
Yeah
I,
you
know
the
the
I
I'm,
the
more
I
think
about
this.
The
more
I
think
that
this
is
something
where
we
can
encourage
interoperability
with
Gateway
API,
there's
absolutely
no
way
we
can
enforce
it.
I
think
you
know
the
vast
majority
of
no
I
shouldn't
say
the
vast
majority.
There
are
going
to
be
a
lot
of
people
that
just
want
Cloud
managed
load
balancers
from
cloud
providers
to
load
balance
directly
to
pods.
But
you
can
do
these
kinds
of
graceful
upgrades
potentially
including
Gateway
API.
B
You
know
with
standardized
configuration
for
a
concept
like
mtls
that
may
be
coming
like
the
mtls
configuration
may
be
coming
from
mesh
I'm,
not
sure
yet
I'm,
not
sure
what
that
looks
like,
but
maybe
we
can
provide
some
patterns
for
how
these
two
can
interoperate
and
I.
Guess
that's
the
goal,
but
I
I
think
that
it
seems
like
a
reasonable.
B
I
think
that
it's
unlikely
that
people
are
going
to
want
manage
load
balancers
to
you,
know
forward
to
a
node
Port
when
they
have
an
option
to
forward
directly
to
a
pod.
I
I
recognize
you
know
greater
observability,
that's
a
huge
one
that
you
just
can't
get.
You
know.
Mtls
feels
like
that's
a
solvable
problem
with
some
kind
of
standard
way
to
configure
that
not
sure.
C
C
Yeah
it
feels
like
it
should
be
solvable
at
the
same
time.
Key
distribution
is
really
interesting,
and
so
that's
you
know
going
to
be.
There
will
be
challenges
there,
but.
C
I
kind
of
feel
like
and
Keith
may
be
about
to
jump
on
this
one
as
well
I
kind
of
feel
like
I'm
hunting,
maybe
a
better
way
to
get
feedback
to
find
out
yeah,
you
know,
is
this
really
do
most
of
the
people
out
there
looking
at
Gamma
think
that
the
collateral
B
should
just
go
straight
to
a
pod,
even
at
the
expense
of
a
mesh,
or
do
we
want
to
try
to
support
some
of
this
stuff?
Do
we
want
to
be
able
to
make
that
two-layer
thing
the
common
case
or
not
Keith
go
ahead.
A
There
are
so
many
permutations
of
so
many
dimensions
of
this
problem
is
the
L7
load
balancer
case
cl4
load
balancer
case
The.
L4
is
kind
of
outside
significant
API,
but
L7
volunteer
case.
There
is
the
there's,
the
internal
Ingress
controller
to
to
a
mesh.
A
Acl
console
Kong
I
think
counts
as
well
that
class
of
Ingress
perspective,
but
then
there's
also
the
there's
no
match
that's
Cloud
to
Cloud
direct
deposit,
there's
cloud
service.
A
A
There
are
a
lot
of
different
permutations
is,
is
my
point
and
different
meshes
solve
this
different
ways
so
as
to
to
use
ntls
and
not
to
try
to
even
build
up,
because
you
just
entail
us
as
an
example
with
LSM.
What
we
did
is
we
had
a
you,
could
bring
your
own
Ingress
controller
and
the
Methodist
create
a
secret
for
you
to
put
into
your
English
controller.
That
was
just
going
to
be
ready
at
any
identity
for
it
that
you
configured
Etc,
et
cetera,
I.
A
Think
with
just
say
you
can
use
oidc
to
get
a
identity
and
some
certain
material
there
is
yeah.
There
are
different
ways
you
could
try
to
do
it
in
like
to
John's
point
in
the
chat.
I,
don't
know
that
it's
fundamental,
because
you
know
back
when
we're
doing
osm
stuff.
We
actually
had
a
design
where
the
an
Azure,
a
new
azimo
balancer
that
folks
are
working
on,
would
be
able
to
read
the
secret
from
the
cluster
to
be
able
to
do
ntls
from
the
cloud
load
balancer
to
the
mesh.
A
So
the
solutions
exists
are
the
solution.
Solutions
could
exist.
I
know
that
you
want
to
assume
they
couldn't.
My
point
is
I,
just
I
wonder
if
this
is
one
of
the
are
we
at
the
point
where
this
is
required
for
people
to
try
this
I.
C
C
That
makes
me
kind
of
sad
because
there
are
ways
that
I
would
like
something
much
more
standard
to
improve
interoperability
and
portability,
but
yeah
you
know
it.
It
feels
like
there
is
enough
Divergence
here
that
the
right
thing
to
do
is
to
move
on
without
it
Mike.
D
C
Well,
for
adoption.
D
And
implementation,
and
conformance
so
yeah
I,
that
might
be
the
best
possible
path
forward,
is.
C
C
Keith
I
think
you
and
John
are
talking
about
the
same
thing
in
the
chat.
Rob
sorry
go
ahead.
B
E
B
I
think
we're
on
the
same
page,
actually,
okay,
all
right,
yeah
yeah,
so
I
mean
that
that
was
the
one
thing.
I
guess
just
kind
of
going
further
on
that
it
seems
like
there's,
probably
something
there
about,
maybe
making
that
little
connection
easier,
like
hey,
mtls
from
Cloud
load,
balancer
to
mesh
sidecar
or
equivalent,
maybe
there's
a
pattern
that
can
be
universally
applicable,
but.
B
C
The
thing
the
thing
that
kind
of
bugs
me
about
leaving
it
this
way
recognizing
that
and
I
don't
mean
to
override
what
I
was
saying
earlier
about.
Yeah
I
think
we
need
to
not
make
this
a
gate
to
go
forward,
but
the
thing
that
bugs
me
about
it
is
that
in
a
lot
of
ways,
what
we're
saying
here
is
that,
if
you're
going
to
use
a
cloud
load
balancer,
then
the
cloud
load
balancer
dictates
the
terms
of
routing
within
your
cluster
and
there's
nothing.
C
You
can
do
about
it,
and
that
feels
a
little
weird,
because
what
we're
basically
saying
is
yeah
it's
going
to
Route
directly
to
not
a
cluster
IP,
it's
going
to
route
to
something
that
meshes
are
often
going
to
have
a
difficult
time.
Picking
up
and
doing
any
sort
of
advanced
routing
or
Advanced
features
with
that
feels
problematic.
John
probably
wants
to
correct
me:
go
ahead,
no.
E
I
think
you're
right
I,
just
think
that
I
would
frame
it
a
different
way
and
say
that
it's
actually
the
cloud
load
balancer.
That's
doing
it
the
correct
and
expected
way,
and
if
a
mesh
was
to
do
a
second
layer
of
routing,
then
that
is
stretching
the
definitions
of
the
API
and
is
actually
doing
the
wrong
thing.
C
C
E
B
B
Directly
yeah,
instead
of
sending
directly
to
individual
endpoints
they're
sending
to
a
cluster
ipe
but
there's
no
equivalent
for
or
yeah.
C
A
C
E
C
A
A
So
not
to
not
to
try
to
go
too
deep.
Well,
while
police-
not
here,
but
all
we've
got
folks
assembled
I
feel
like
I
feel
like
where
there's
agreement
is
that
this
is
a
very
complex
problem
and
it's
probably
not
required
for
the
answers
to
this
problem
are
not
required
for
people
to
be
able
to
play
with
it.
A
However,
I
do
think
that
there
is
I
wonder
if
we
can't
push
this
down
over
your
abstraction
and
say
that
if
you're
a
mess,
if
you're
a
gaming
compliant
mess,
you
should
have
documentation
about
how
you
interpolo
balancers,
granted.
Most
nurses
probably
already
have
that,
but
maybe
that's
some
something
we
can
do
to
make
this
I'm
just
looking
for
ideas,
but
maybe.
E
C
D
C
D
D
I
think
that
it
makes
sense
to
kind
of
remove
this
as
a
gate
from
saying
that,
like
gamma
is
a
thing
that
implementation
should
start
should
be
able
to
like
start
putting
together,
and
we
should
like
give
ourselves
the
space
to
figure
out
what
we
can
do
here
and
hopefully
that
leads
to
patterns
emerging
in
terms
of
just
seeing
where,
where
multiple
implementations
are
able
to
find
consistency
between
them
and
yeah,
I
think
the
documentation
thing
is
like
sure,
but
it's
still
like
a
it,
doesn't
get
us
all.
D
The
way
there
I
think
kind
of
like
the
risk
is
that,
whether
it's
documented
or
not,
the
risk
is
that
you
end
up
with
configuration
for
East-West
Gateway
API
that
looks
similar,
but
then
is
ends
up
in
different
data
plane
configuration
when
applied
to
different
meshes,
and
that
makes
it
difficult
to
have
any
kind
of,
like
conformance
between
implementations,
so
so
I
think
I.
Think
that's
the
risk
and
I
would
like
it.
D
If,
eventually,
we
can
find
some
way
to
at
least
like
Express
how
these
things
are
done
like
in
a
way
that
is
not
just
implicit
or
like
documented,
like
actually
make
it
part
of
the
configuration
but
I.
Think
for
now
like
we
should
just
accept
that
there
will
be
some
variance
and
try
to
get
try
to
unblock
the
ability
for
implementations
to
like
start
building.
This.
B
Yeah
I
agree
with
what's
been
said.
It
feels
like
there's,
been
a
lot
of
valuable
discussion
today.
That
will
be
very
easy
to
lose
context
on
and
then
repeat
in
a
few
months
and
so
I
don't
know
if
if
anyone
has
time,
but
if,
if
we
can
just
it
sounds
like
there's
already
a
dock
circulating
or
something
or
maybe
just
comment
on
this
issue
or
or
something
just
to
try
to
capture
this
discussion
and.
D
I
have
a
yeah
I
have
I,
have
a
10-page
Google
doc
that
my
goal
for
basically
at
the
end
of
this
week,
is
to
get
it
into
a
gap
that
may
or
may
not
merge,
but
is
at
least
like
a
place
to
document
this
conversation
and
a
lot
of
the
different
issues
that
have
come
up
and
been
discussed
as
part
of
this
yeah.
That
sounds
great
trying
to
get
it
off
my
plate
and
yeah
I'll
get
things
unlocked.
D
A
Yeah,
so
not
to
get
too
existential,
but
he's
had
several
conversations
about
what
is
gamma
is
gamma,
a
spec
that
describes
the
state
of
how
things
should
be,
or
is
it
something
that
codifies
the
patterns
that
exist
in
meshes
today?
A
This
feels
like
a
significant
step
or
a
milestone
in
deciding
deciding
that,
or
maybe
understanding
that,
because
I
think
you
know
if
it
was
just
a
question
of
getting
together
and
picking
up
the
way
we
want
gateways
and
measures
to
interact,
we
could
be
able
to
select
people
and
hearing
games
together
and
do
that,
but
the
reality
of
adoption
in
Cloud
load,
balancers
and
Ingress
controls
the
white
swap
of
things
that
are
out
there
rendered
that
conversation
well
physically
useful,
relatively
Limited
in
potential
impact.
B
It
yeah
I
think
that's
accurate,
I
agree
with
that
conclusion.
I
would
say
that
we
have
accepted
gifts
before
that.
Just
describe
the
current
state
of
the
world
without
proposing
a
solution
that
may
be
where
this
lands
I
don't
know,
because
I
I
think
there's
there's
at
least
some
interest
in
some
small
path
forward
right.
You
know
whether
it's
just
mtls,
maybe
it's
broader
than
that,
but
at
least
there's
something
we
can
do
here
so
getting
a
gap
just
to
explore.
That
seems
useful
to
me.
A
When
it
comes
to
Ingress
controllers
and
load
balancers
standardized
more,
there
are
lots
of
different
ways
to
go
about
it.
But
I
think
that
this
is
a
good
group
to
to
solve
that
problem,
because
I
think
it
makes
a
lot
of
the
high
level
questions
easier.
If
we
at
least
add
that
that
Bedrock
done,
but
so
yeah
I,
think
capturing,
it
is
important,
but
recognizing
our
limitations
today
is
a
it's
a
you
know,
kind
of
a
reality,
any
other
thoughts
or
comments.
A
Okay,
then,
in
that
case,
I
believe
that
is
all
of
our
agenda
any
last
minute
topics.
People
want
to
bring
up.
A
All
right,
in
that
case,
I'll
give
everybody
about
20
minutes
back,
which.