►
From YouTube: SIG Network Gateway API meeting for 20230213
Description
SIG Network Gateway API meeting for 20230213
A
A
little
more
hello
everyone
and
welcome
to
the
February
13th
edition
of
Gateway
API.
A
We
have
an
agenda,
that's
mostly
Rob
and
me,
so
if
anybody
has
anything
that
they
want
to
bring
up,
please
do
go
ahead
and
jump
on
our
Google
doc
and
add
it
as
a
reminder.
This
meeting
is
governed
by
the
kubernetes
code
of
conduct,
which
effectively
boils
down
to
be.
B
C
D
Yeah
I
I
left
a
comment
at
the
bottom
here,
but
I
I'm
really
just
trying
to
figure
out
how
this
fits
into
the
broader
picture
here,
because
I
know
there's
kind
of
this
longer
term
bigger
picture
discussion
of
hey.
Could
we
have
like
a
a
Gateway
class
cluster,
IP
or
or
something
like
that?
That
was,
maybe
even
you
know,
implemented
by
data
planes
like
Coupe
proxy
in
the
future.
D
I,
don't
know
you
know
like
there's
there's
that
idea
of
Gateway
kind
of
taking
on
the
the
service,
routing
capabilities
that
currently
live
on
service,
and
then
there's
also
where
I
think
this
is
really
coming
from.
Is
the
idea
of
trying
to
map
there's
a
huge
number
of
implementations
today
that
by
default
will
take
a
Gateway
And
when
they
see
a
Gateway
they're
going
to
turn
around
and
create
a
service
type
load
balance,
usually
type
load
balancer
and
a
deployment
behind
it?
D
That
would
be
some
kind
of
Gateway
implementation,
and
this
feels
like
a
tiny
subset
of
that
right
like
it,
it
seems
like
we're
trying
to
find
a
generic
way
to
represent
I,
want
that
service
that
you're,
creating
to
be
cluster
IP
and
not
you
know
not
a
load.
Balancer
right
is,
is
that
am
I
understanding
this
correctly
before
I
go
too
much
further.
E
I,
can
you
hear
me
I
guess
I
would
describe
the
problem
first
before
we
talk
about
the
like
potential
solution,
so
the
problem
we
have
and
I
guess
for
those
that
don't
know
I'm
kind
of
like
representing
key
native
here.
E
People
use
create
Canadian
Services,
which
kind
of
like
abstract
a
lot
of
the
underlying
details,
and
one
of
the
things
people
want
to
create
is
like
a
service
that
is
only
accessible
on
a
cluster,
so
it's
kind
of
like
a
private,
so
it's
not
by
default,
exposed
to
public
like
Ingress
and
so
that's
sort
of
the
problem.
So
the
question
is
like
hey:
how
can
you
do
this
right
now
with
giveaway
API,
but
in
a
portable
way?
I
would
say
it
doesn't
currently
exist.
E
So
this
is
sort
of
the
Gip.
That's
trying
to
address
that
so
like
how
can
I
create
gateways
that
really
only
come
back
with
a
status
address.
That's
only
accessible
on
the
current
cluster.
E
Whether
that's
implemented
with
like
a
kubernetes
service,
which
is
kind
of
what
it
lends
itself
to
sure,
but
I
would
say
it
doesn't
necessarily
like
imply
that,
but
even
though
it
yeah
I.
D
Think
I
think
what
your
the
way
you
phrase
that
I
feel
like
it
is
almost
entirely
tied
to
a
service
of
type
cluster
IP
because,
like
you
know,
I
I'm
thinking
of
cloud
load,
balancers,
naturally,
and
our
internal
load
balancers
would
not
be
limited
to
a
single
cluster
necessarily
so
that.
But
maybe
there
are
other
thing,
other
potential
means
of
implementation
that
I'm
missing,
but.
A
Helpful
anecdote,
but
to
give
some
I
guess:
I
have
some
context.
A
very
early
version
of
Kong's
Ingress
controller
actually
did
this
with
Gateway
and
it
did
use
service
type,
but
it
specifically
added
sorry
service
type
load
balancer,
but
it
specifically
added
the
cluster
IP
for
administrative
traffic.
A
That
is,
we
would
have
the
control
planes
and
that
may
actually
be
multiple
talking
to
the
Gateway
using
the
cluster
IP,
but
all
other
traffic
would
be
through
the
load
balancer
for
the
actual,
like
application
traffic.
We
ultimately
threw
that
away
because
there
was
not
really
much
support
in
the
community
at
that
time.
It's
probably
like
a
year
or
more
ago
for
having
that
for
differentiating
traffic,
administrative
traffic
versus
application
traffic,
but
I
can
see
the
use
case.
A
We
just
got
off
of
it
because
we
didn't
want
to
try
to
push
through
it
at
the
time.
So.
A
A
E
Yeah
the
way
to
think
about
it
too
right
where
hey,
if
I'm
running
a
cluster
and
I'm
running
key
now
I
would
want
to
expose,
for
example,
my
front-end
services
to
the
public
internet.
But,
for
example,
my
micro
Services
would
be
private,
but
I
still
want
to
potentially
benefit
from
all
the
L7
features
to
do
like
Canary,
rollout
traffic
splitting
and
things
like
that.
G
Can
I
ask
a
simple
question
from
everybody
new
to
these,
of
course,
what
I'm
seeing
that
was
as
I'm
preparing
all
this
info,
so
it
depends
upon
what
kind
of
traffic
where
the
traffic
is
for
application
based
traffic
or
it's
admin
traffic.
So
if
we
have
in
that
case,
if
you
think
that
the
Gateway
is
a
load
balancer
for
the
traffic,
then
that
should
be.
G
That
means
one
particular
load
balancer
other
or
the
the
local
Gateway
you
are
talking
about.
It
should
be
for
a
particular
app
particular
app
or
app
or
or
delivery
different
local
Gateway
for
the
admin
kind
of
traffic.
So
we
have
to
have
one
kind
of
classification
classification
in
the
top
right
foreign.
E
Use
case
the
There's
no
distinction
between
like
traffic
directed
for
the
app
well
I,
guess,
that's
pretty
much
everything
that
the
load
balance
or
that
the
traffic
going
through
the
any
Gateway
would
be
directed
to
the
user
app.
So
I
would
probably
distinguish
this
as
if
I
refer
to
admin
traffic
as
I.
Don't
know.
If
this
generalizing,
like
observability
like
Prometheus
scraping,
then
you
wouldn't
use
the
Gateway
for
that.
G
E
I
I
apologize
but
I
don't
understand
the
question.
Like
my
use
cases,
there's
user
traffic
people
are
deploying
apps.
They
want
those
apps
not
to
be
exposed
to
the
internet,
but
within
the
cluster
you
might
want
to
do
ll7,
load,
balancing
meaning
using
features
like
topic
splitting
and
things
like
that.
E
G
A
G
C
G
Have
even
application
Level,
so
my
my
thinking
was:
are
we
trying
to
prove
them
together?
Okay,
we'll
be
doing
one
set
of
application
and
in
another
set
of
hospital
within
another
set
of
application?
In
that
case,
when
you
are
talking
about
load
balancing,
you
are,
you
have
to
know
what
kind
of
application
your
load
balancing
is,
but,
but
if
you
are
to
reduce
that
complexity
to
one
level,
that's
fine,
only
all
application
will
be
a
one
load.
Balance
then
also
is
fine.
A
Yeah
in
Dave's
case,
it's
simply
the
difference
between
whether
or
not
traffic
should
be
allowed
to
come
in
from
outside
the
cluster
or
not.
There
should
be
a
way
to
discern
that
to
to
configure
that
Within
Gateway
is
what
he's
requesting
it
seems
really
to
me.
I
Rob,
you've
been
pretty
quiet
to
me.
It
seems
pretty
like
a
solid
thing
to
do.
I
do
get
that
it
has
implications
potentially,
for
you
know
everyone
else
who
doesn't
want
to
do
that.
D
Yeah
I
know
it
to
be
clear:
I
I
get
the
purpose.
I
get
the
idea,
I
think
I
think
it
makes
sense
somewhere
I'm,
not
sure
where
that
somewhere
is
for,
for
example,
let's
you
know
you
can
easily
go
down
a
rabbit
hole
of
well
hey
if
you
want
a
generic
way
to
configure
same
cluster,
why
not
have
a
generic
way
to
configure
same
network?
D
You
know
the
traditional
ilb
use
case
as
well
right,
because
at
some
point
there
will
be
more
weight
more
than
one
way
to
represent
that
as
well,
and
then
you
very
quickly
start
to
say:
well,
hey
right
now,
that's
represented
in
Gateway
class.
Why
isn't
this
concept
represented
in
Gateway
class
because
it
really
feels
like
a
you
know,
like
class
of
a
Gateway
right
a
this
is
where
a
Gateway
is
exposing
traffic
so
that
that
may
Belong
One
one
level
higher
than
this
would
currently
put
it.
D
D
If
we
call
this
a
cluster
IP
address,
does
that
prevent
or
limit
future
support
of
you
know
a
Gateway,
a
cluster
IP
Gateway
class
or
something
that's
kind
of
built
into
kubernetes
for
cluster
IP
I
again,
I,
don't
know
that's
pretty
theoretical
at
this
point,
but
I
want
to
at
least
leave
some
room
for
that
I
and
I'm
not
sure
how
this
would
interact
yet
I
know
that's
more
questions
than
answers,
but.
E
E
Because
there
are
needs
and
a
lot
of
user
requests
to
not
have
a
mesh,
which
is
essentially
why
keenano
went
through
the
whole
hey.
We
use
this
Shield
at
the
beginning
and
we
spend
a
lot
of
time
abstracting
in
a
way
to
give
people
optionality
to
not
use
a
mesh,
and
this
is
sort
of
like
the
requirement
needed
to
get
that
because,
like
candidly,
like
istio,
supports
assigning
gateways
to
the
Ingress
pod
and
then
they
have
like
you
can
set
a
Gateway.
This
Gateway,
referring
to
the
istio
resource,
one.
E
The
same
deployment
as
what
I
was
kind
of
hitting
at
and
like
that,
so
that's
like
a
way
to
get
around
the
full
mesh
thing
and
we
have
ways
to
do
this
with
key
native
sort
of
like
Envoy
Gateway,
precursor
called
Courier,
which
we
kind
of
hand,
roll
ourselves
and
also
like
in
Contour
as
well.
We
have
a
setup
encounter
to
accomplish
this
being
gay
native,
this
sort
of
visibility
in
terms
of
the
mechanics
that
you're
describing
Rob
like
I
I,
don't
necessarily
know
what
the
right
thing
is.
E
E
I,
don't
know
if
that
means
that
I'm
like
in
terms
of
Gateway
class,
it
would
just
sort
of
be
where
you
would
specify.
This
is
just
like
a
top
of
my
head,
like
IP
ranges
as
a
as
a
thing
like
more.
It's
like
a
more
of
a
templated
thing
and
then
gateways
like
the
the
stamp
of
the
template
like
this
instantiation
of
the
class.
E
A
You
I'm
looking
at
the
cat,
but
I'm
just
like
have
you
run
into
anybody
else
that
basically
wants
the
same
thing.
A
I
meant
other
implementations,
oh
because
that's
going
to
be
one
of
the
things
that's
really
going
to
help
to
kind
of
push.
It
would
be
that
you
know
this
is
something
that
more
than
one
implementation
needs,
but
if
it
is
something
that
only
one
implementation
needs
or
knows
that
they
need
that's
another
way
to
put
it,
you
may
be
the
only
one
that
knows
that
you
need
it.
C
E
Been
I
guess
validated:
we've
implemented
this
in
key
native
using
other
ingresses,
but
not
it's.
Not.
A
first
party
support
like
Contour
like
Contour,
implements
this
using
two
different
installations
and
we
program
resources
against
those
different
installations.
A
A
I
hear
you
yeah
Philip
did.
C
H
So
so
there
are
actually
two
related
things
that
we're
doing
to
what
we
just
talked
about.
We
actually
again
in
the
service
provider
area.
We
have
a
lot
of
a
lot
of
what
we
what
I
call
hairpinning,
because
we
actually
present
it
as
external
IP
addresses
but
they're
being
hit
by
something
internal.
H
So
you
know
one,
you
know
one
service
inside
a
namespace
may
call
itself
via
our
product
via
our
Gateway,
and
you
know
we
do
it.
We
we
don't
restrict
the
the
access
to
it
by
saying
that
it's
only
internal
access
but
I
could
see
the
value
in
that
the
the
other
thing
you
guys
were
talking
about,
there's
some
mention
of
sort
of
restricting
the
IP
address
range
as
a
part
of
the
Gateway
class.
H
Like
again
one
of
the
use
cases
we
do
that's
really
key
to
us
is
we
have
we
connect
to
a
number
of
network
domains,
so
you
know
usually
there's
only
one
like
internally,
maybe
there's
you
know,
there's
a
pod,
Network
and
there's
a
node
Network.
Externally,
though,
there
can
be
a
radio
Access,
Network
and
operations,
Administration
and
maintenance,
Network
and
services
network,
a
DMZ
to
the
internet,
Network
and
they're
all
separate
and
sort
of
at
the
Gateway
class.
H
That's
where
I'm
expecting
you
would
configure
which
of
those
you
can
reach
and
how
and
I
think
that's
similar
here
where
you
could
configure.
You
know
that
you
just
are
listening
on
that
internal
Network
and
not
on
any
external
network,
and
that
would
be
the
equivalent
of
having
an-
and
you
know
this,
this
just.
B
I
I
One
service
of
type
load
balancer
that
could
accept
external
traffic
and
then
another
like
service
like
a
type
cluster
IP,
and
then
you
kind
of
bind.
You
have
like
a
listener
resources
which
defines
like
what
course
you're
listening
on
so
you
just
have
you
create
two
listener
resources
pointing
to
whatever
importance
that
the
services
are
going
to
be
exposed
on
and
then
that
that's
how
you
can
get
one
listener?
That's
is
accepting
external
traffic
and
another
one.
That's
only
really
for
internal
traffic.
A
For
popping
in
with
that,
so
there's
use
cases.
So
that's
established,
that's
something
that
maybe
we
should
kind
of
push
up
a
little
bit,
but
then
there's
Rob's,
open
question
and
Philip
talked
to
it
a
little
bit
about
whether
or
not
it
belongs
in
light
Gateway
class.
We
are
kind
of
we
have
kind
of
spent
a
lot
a
lot
of
time
on
this
all
and
we
do
have
a
reasonable
amount
of
agenda.
D
I
I
can
put
this
on
the
issue
itself,
but
I
think
I
I
have
two
questions
here:
I
I,
don't
doubt
and
I
think
we've
seen
clearly
that
there
are
very
clear
use
cases.
The
two
questions
that
I
am
very
interested
in
answering
is
where
this
belongs
specifically,
if
it
would
make
sense
on
Gateway
class
and
and
then
very
related
to
that.
To
that
do
we
want
to
differentiate
between
same
network
and
same
cluster
or
my
conscience.
There
are
going
to
be
implementations
that
can
support
one
or
the
other,
but
not
both.
D
There
are
going
to
be
implementations
where
internal
means
internal
to
the
cluster
and
other
implementations
where
internal
means
internal
to
the
network,
but
there's
probably
not
a
lot
of
overlap,
but
just
clearly
stating
what
we
actually
want
and
I'd
like
to
make
sure
we've
covered
both
of
those
with
this.
With
this
work.
A
D
A
D
I,
that
would
be
my
preference
yeah
just
try
and
answer
these
in
the
Gap.
One
of
anyone
can
add
these
questions.
There
may
be
others,
but
those
are
the
ones
that
are
top
of
mind.
For
me,.
A
C
A
E
And
I
guess,
if
you
have
thoughts
on
potentially
it
being
on
the
like
on
the
Gateway
class,
it'd,
be
good
to
know
because
I
I
don't
know
what
implications
that
has
if
I
don't
know,
if
it's
simple
for
you
to
do
like
hey,
if
it's
on
the
daily
classes
or
the
pros,
if
it's
not
on
it,
these
are
the
cons.
D
Yeah
I
I
can
try
and
take
the
the
AI
to
explain
what
it
would
look
like
on
Gateway
class
I
am
obviously
coming
from
a
place
of
bias
where
we
already
have
ilb
Gateway
classes,
and
so
that's
where
we've
already
chosen
to
represent
this,
but
and
by
that
I
mean
gke,
but
definitely
interested
in
alternatives.
A
Roger
that
so,
since
we're
coming
up
on
the
half
hour,
then-
and
we
have
a
good
action
and
good
follow-up
I'm
happy
to
do
it
by
the
way
rob
it
sounded
like,
but
go
ahead,
I
mean
either
way.
Yeah.
D
John
and
so
I
sorry
I
missed
this
and
from
chat
but
I
wanna
I.
Think
John
asked
another
very
important
question:
I
is
this
bigger
than
just
cluster
IP
right,
like
I?
Think
John
is
talking
about
a
general
way
of
configuring
in
cluster.
Anything
like
how
you're
mapping
config
to
services,
because
I
think
maybe
90
of
active
implementations
today
were
some
very
large
proportion
are
taking
a
Gateway
and
mapping
it
to
a
service
resource,
and
although
this
is
just
one
piece
of
that
config,
that's
probably
another
good
open
question.
D
E
So
I
guess
a
question
then
would
we
try
to
put
in
Spec
in
Gateway
class
to
accomplish
this,
or
would
this
imply
customized
like,
for
example,
in
a
cluster
Gateway
resource.
D
I,
don't
know
that
that's
a
fairly
good
question
yeah
part
of
me.
You
know
I,
think
many
of.
F
I
was
just
suggesting
that
every
implementation
today
that
is
in
cluster,
is
solving
the
same
problems,
but
they're
doing
them
in
different
ways
that
you
know
whether
it's
a
cluster
IP
service
or
load
balancer
service.
It's
just
a
small
subset
of
the
problem.
There's
plenty
of
other
fields,
really
every
field
in
service
or
deployment
or
whatever
fields
are
being
created
as
something
that's
potentially
in
scope.
We
don't
necessarily
need
to
solve
the
equal
1000
fields
of
API
spec.
F
If
we
want
to
be
something
more
specific,
but
certainly
some
room
for
standardization,
there
I,
don't
think
I've
seen
two
implementations.
Do
it
the
same
way
yet.
E
A
A
A
Okay,
thank
you,
John!
Thank
you!
Everybody!
We
kind
of
are
pushing
the
half
hour,
so
I
think
we
should.
We
have
enough
things
to
follow
up
on.
We
can
kind
of
keep
this
going.
D
Yeah,
hopefully,
one
fast
one
yeah
so
I
this
has
come
up
for
a
while
and
I
got
pinged
a
couple
times.
I
think
Shane
and
I
have
both
been
asked
a
few
times
about
hey.
D
Could
we
have
a
meeting
time
that
people
in
Europe
could
actually
make
it
to,
and
it
would
be
nice
if
we
could
actually
try
that
out
or
experiment
with
it
by
just
trying
to
understand
the
potential
impact
of
that
on
the
attendees
of
today's
call,
if
we
were
to
consider
rotating
similar
to
gamma
already
does
and
have
one
meeting
every
other
week,
that
was,
you
know
eight
or
nine
a.m.
Pacific
roundabouts.
D
D
A
D
Okay,
so
maybe
maybe
what
this
looks
like
long
term
is
we,
we
always
have
one
meeting
that
is
friendly
to
each
kind
of
group
of
time
zones,
and
we
alternate
back
and
forth
whether
that's
Gamma
or
this
main
meeting,
but
for
I
think
the
first
step
would
just
be
to
experiment
once
and
yeah
Philip
I
I
work
with
a
lot
of
teams
and
and
have
the
same,
the
very
same
problems
you're,
describing
the
the
way
too
many
early
calls
that
are
already
taken,
so
I
yeah
I
think
it
will
be
hard
to
find
a
perfect
fit
yeah.
A
Yeah
I'll
get
that
updated
on
the
Sig
Network
calendar,
then
sound
good.
D
D
Okay,
so
my
my
next
one,
maybe
will
be
a
little
shorter
I,
don't
know,
is
Keith
on
this
call.
It's
maybe
this.
This
is
a
discussion.
Point
that's
come
up
a
few
times
and
I
just
kind
of
wanted
to
come
back
around
and
talk
about
policy
attachment,
it's
complexity
and
potential
ways
to
work
around
it.
D
Keith
did
a
great
job.
I
think
this
was
a
follow-up
from
a
gamma
meeting
of
actually
trying
to
describe
the
current
pain
points.
If
anyone
has
others,
please
add
them
to
this
list.
This
is
really
just
intended
to
be
a
discussion.
Point
of
anything
you
don't
love
about
policy
attachment
today
would
be
good
to
understand
and
quantify
with
that
said.
I
I
have
some
ideas,
go
ahead,
Shane
I'm,
sorry,.
D
No
I
have
some
ideas:
I
posted
them
below
in
here
and
I.
Think
they're,
largely
representative
of
things
that
we
have
already
been
talking
about
as
a
community
one
there's
this
idea
of
building
more
into
the
API
itself.
So
in
the
case
of
timeouts,
maybe
there
is
some
if,
if
we
allow
our
spec
to
be
just
a
little
bit
more
vague,
maybe
there's
some
space
in
there
for
some
kind
of
timeout
configuration,
I
I,
don't
know
it
they're.
D
Just
there
are
so
many
different
timeouts
that
are
configurable
in
nginx,
a
fair
number
in
Envoy,
a
fair
number
and
AJ
proxy
trying
to
find
the
intersection
is
surprising
complicated,
but
maybe
we
could
it's
just
in
the
timeout
example,
that's
kind
of
where
I.
D
A
It
absolutely
does
the
other
thing
I
was
going
to
add
real
quick
was
just
if
anybody
here
is
using
policy
attachments.
This
might
be
a
decent
thread
to
come
in
and
say
if,
particularly
if
you
have
pain
points
or
even
if
you
don't
come
in
and
kind
of,
add
hey,
we
are
using
policy
attachment
and
here's
how-
and
this
is
how
it's
painful-
that's
helpful,
I
think
to
this
conversation.
D
I'll
say
4G
for
gke
we've,
we've
implemented,
I
think
three
different
policies
today,
but
despite
them
using
the
current
model,
they
could
probably
be
represented
just
as
well
as
meta
resources,
because
they're
really
in
at
least
one
or
two
cases,
they're
really
tied
to
a
specific
resource
and
hierarchy,
doesn't
really
work
for
them.
So
that
seems
to
that
seems
to
go
well
with
Nick's
idea
of
meta
resources.
D
A
In
a
not
quite
like
that,
but
at
Kong
what
we've
done
is
danced
around
policy
attachment
by
almost
doing
it
multiple
times.
So
we
actually
had
some
existing
resources
that
attached
to
resources
to
basically
add
additional
configuration
from
the
Before
Time,
and
we
are
just
continuing
to
use
those
at
the
moment.
But
they
don't
they
they're,
not
in
the
vein
of
policy
attachment.
A
They
don't
follow
the
sort
of
the
specification
laid
out,
and
that
is
because
we've
been
kind
of
holding
on
policy
attachment
to
see
how
it
shapes
and
see
how
people
end
up
using
it,
because
we've
had
some
hesitations
for
these
reasons
as
well.
So.
A
A
D
Not
record,
no,
it's
not
required
for
GK.
We've
tried
to
group
related
things
together,
so
you
would
have
get
on
in
one
example,
we
have
a
way
to
configure
custom
health
checks,
and
so
we
group
everything
related
to
health
checks
together
in
one
policy
and
others,
you
have
a
policy,
that's
related
to
you,
know
gcp
specific
settings
for
our
backend
yeah,
yeah
and
interesting.
That
Lewin
is
saying
that
you
can
attach
policy
to
either
Gateway
or
route,
but
I
guess
that's.
I
am
specifically
that's.
K
Correct
yeah,
we
do
not
have
a
native
kubernetes
experience,
so
you
use
you
know
Gateway
API,
to
create
the
Gateway
and
HTTP
route
and
then
outside
you
know,
then
you
can
use
the
console
or
AWS
UI.
Then
you
attach
the
IM
policy.
The
long-term
goal
is
try
to
do
everything
within
the
cluster.
K
D
C
D
C
D
The
the
problem
with
any
kind
of
block
like
that
that
is
kind
of
you
know
open-ended,
is
that
you,
you
have
recreated
annotations
just
in
a
different
in
a
different
place
and
there's
not
really
good
way
to
validate
or
really
provide.
Much
of
anything
with
that
said,
it
may
be
preferable
to
policy
in
some
cases.
C
D
Notice
that,
for
example,
on
TLS,
we
have
an
options
field
which
it
feels
a
lot
like
that
it's
just
a
match,
string
string.
Maybe
there
are
some
places
where
we
could
I,
don't
I,
don't
want
to
completely
rule
it
out
I.
Just
we.
We
have
to
be
careful
that
I
I
really
want
to
wherever
we
can
push
for
proper
solutions
that
really
solve
it
the
right
way,
and
if
we
provide
too
many
easy
ways
to
kind
of
you
know
everyone
does
their
own
thing.
Then
we
have
ended
up
with
Ingress
V2.
A
A
All
right,
what
do
we
have
40
minutes
after
moving
on
to
the
next
topic,
which
is
also
Rob
man.
B
D
So
I
was
talking
with
who
you're
this
this
morning.
Actually
he's
been
doing
a
lot
of
really
great
performance
tests
around
redirects
rewrites
and
the
next
one
on
his
list
is
request
mirroring
and
that
is
more
complicated
than
I,
initially
thought
about,
and
just
wanted
to
throw
it
out
as
anyone
have
ideas.
So
the
idea
with
request
mirroring,
is
obviously
that
you're
you're
sending
a
request
to
both
the
main
primary
back
end
and
sending
a
copy
of
that
request
to
a
mirrored
back
end.
What
do
we
do?
For
conformance?
F
D
A
Guess
it
depends
create
a
uuid
foreign
request
and
then
send
it
to
make
sure
we
get
that
exact
request
on
the
back
end.
D
That
works
all
right,
that
was
easy,
yeah,
I'll,
I'll
follow
up
but
yeah
and
then
arco's
idea
of
enhancing
Echo
server
yeah.
We
probably
need
some
changes
on
Echo
server,
because
what
it's
logging
today
is
not
really
sufficient
for
what
we'd
need
yeah.
A
A
Okay,
this
is
quick
Blix
our
project
for
a
reference
and
testing
implementation
layer
for
implementation
for
Gateway
API
has
had
its
first
pre-release.
It
has
a
very,
very
basic
level
of
UDP
route
support.
So
I
just
wanted
to
shout
that
out
here
and
let
people
know
the
next
steps
are
going
to
be.
Finally,
getting
this
license
thing
figured
out,
there's
still
some
kind
of
feed
Dragon
when
it
comes
to
the
GPL
V2
we're
starting
on
the
TCP.
D
Rob
awesome
yeah
that
no
that's
great
I
am
looking
forward
to
having
something
we
can
run
as
a
reference,
especially
for
L4
conformance,
so
yeah
cool
and
you've
still
been
doing
the
Gateway
code,
Jam
right
on
Fridays
yep.
A
So
for
anybody
who's
interested
in
the
project.
It
is
a
kind
of
because
it's
a
testing
implementation,
it's
kind
of
a
fun
project.
It
doesn't
have
a
lot
a
whole
lot
of
it
has
weight
on
it.
A
But
just
not
you
know,
like
most
other
things
of
this
sort,
so
we
do
a
code
Jam
every
Friday
right
now
that
is
called
the
gateway,
API
code
or
yeah
Gateway
API
code
gym
we're
mostly
focusing
on
books
for
that
right
now,
it's
usually
about
six
people
show
up
and
we
just
do
like
look
at
code.
Sometimes
we
do
live
coding,
and
sometimes
we
just
talk
about
the
issues
in
the
project
and
stuff
like
that.
A
So
if
you're
interested
in
the
project,
please
do
join
in
we'd
love
to
have
you
very
laid
back.
We
don't
even
record
it
right
now,
because
we
just
want
people
to
feel
really
relaxed
and
just
have
fun
and
talk
about
code.
D
Cool,
that's
awesome
all
right,
yeah,
my
my
next
one
up
is
just
I've,
been
thinking
a
lot
about
070
and
specifically
the
idea
of
request,
rewrites
and
redirects
path,
redirects,
making
it
to
Beta
graduating
to
the
standard
Channel,
and
our
current
graduation
requirements
are
really
there's
two
forms
of
them.
There's
the
alpha
beta.
We
have
implemented
by
several
implementations
and
several
generally
means
more
than
two.
So
it
has.
D
As
three
and
then
for
beta
G
to
GA,
we
say
we
need
multiple
conformant
implementations,
so
at
least
two
conformant
implementations,
so
I've
been
trying
to
work
backwards
from
that
and
as
we
are
considering
going
Alpha
to
GA
at
some
point
in
the
not
too
distant
future,
I've
been
trying
to
think
about
what
that
even
means.
D
C
D
Of
the
things
that
we're
running
into-
and
this
is
very
true
in
this
case-
is
that
a
lot
of
implementations
are
waiting
to
support
a
feature
until
it
gets
to
standard
Channel,
which
you
have
this
chicken
and
egg
problem
of
we'll
implement
it
once
it
gets
to
a
spot,
but
we're
not
going
to
implement,
implement
it,
while
it's
still
experimental
I,
don't
know
if
this
is
the
the
best
solution
to
that,
but
just
the
idea
of
yes,
we
need
some
implementations
to
have
past
conformance
for
our
feature,
and
we
also
need
other
implementations
that
have
signed
off,
that
they
will
implement
it,
and
it
makes
sense,
as
the
current
spec
States
I
don't
know
is.
E
Dave,
just
a
quick
question:
what's
is
there
a
reason
why
people
don't
want
to
implement
some
of
the
extended
features
like
you
mentioned
before
it
lens
I
guess
into
the
core.
I
might
have
the
terms
wrong,
but
yeah.
D
So
the
main
thing
is:
oh
God,
it's
it's
a
good
question.
The
you
know,
problem
with
experimental
features
is
by
definition
they
can
change
or
be
removed
at
any
point.
So
there
is
some
risk
that
if
you
invest
time
implementing
support
for
an
experimental
feature,
it
could
go
away
or
it
could
break.
A
Yeah,
okay,
Mikhail
asked
a
good
question
in
chat,
so
yeah,
even
if
two
implementations
support
a
feature
are
we
sure
any
users
actually
tried
it?
This
is
a
good
question
no
by
by
default.
We're
not
like
positive
of
that.
You
know.
Obviously,
if
you
just
implement
it
doesn't
mean
people
are
going
to
use
it
because
most
people
I
think,
will
stray
away
from
using
anything
in
production.
That
has
the
word
V1
Alpha
on
it,
you
know
or
whatever,
but
we
could.
We
could
ask
more
about
that.
A
I
know
at
Kong
we
have
some
usage
of
UDP
route,
particularly
to
your
point.
I
think
this
is
pretty
fair.
One
thing
I
noticed
today
actually
was
I
was
going
over
because
I'm,
you
know
I'm
trying
to
Champion
The
L4
stuff
and
get
that
into
beta
as
soon
as
possible.
A
There
are
four
implementations
I
found,
including
Kong's
Implement,
two
implementations
to
comics
of
technically
five
that
currently
implement
or
sorry
four
that
currently
Implement
UDP
route
like
you
can
use
it
and
one
which
is
nginx.
That
says
it
will
but
isn't
from
what
I
can
tell
quite
doing
that
yet,
which
is
great
by
the
way
for
The
L4
story.
We
have
that
many,
but
I
think
getting
UDP
route
over
the.
We
don't
want
it
to
be
an
alpha,
because
people
won't
will
be
worried
about
wasting
time
on
it.
A
Hump
under
these
terms
would
be
better
so,
and
it's
a
very
simple
API
right
now
so
I
mean
that's
the
that's.
The
the
best
thing
we
can
do
strategically
is
okay
yeah.
If
we
take
this
kind
of
approach,
we
just
have
to
be
extremely
is
judicious
the
right
word,
I.
Think
judicious
is
the
right
word
about
what
actually
ends
up
in
those
specs,
because
we're
gonna
commit
to
you
know
maintaining
it
longer
term
sooner
so,
yeah.
D
Yeah
completely
agree
and
there's
been
a
good
good
conversation
in
chat
as
well.
Brian's
comment
that
no
one
wants
to
play
crystal
ball
gazing
game
of
guessing
which
to
invest
in
yeah
I,
get
that
it's
a
a
tough
problem
and
and
I
others
are
saying.
Yeah
I
would
love
to
have
end
user
usage.
I
wish
we
could
find
a
reliable
way
to
gauge
that
I
know.
A
Don't
know
I'm
curious
I
am
curious,
who,
among
the
implementations
of
any
particular
API,
how
many
implementations
today
are
taking
Telemetry
of
the
usage
of
Gateway
API
resources.
C
B
Are
being
proposed,
in
fact,
we
are
proposing
some
of
the
features
that
that
we
need
so.
A
Understood,
and
do
you
like
this,
this
potential
change
in
our
graduation
requirements?
Does
it
does
it
benefit
you?
You
think.
B
H
H
A
A
On
this
one,
no
yeah,
we
appreciate
the
the
input
on
that
I
guess.
I
guess
I
still
would
ask
the
question,
though,
like
so
as
somebody
who
feels
like
they're
more
on
the
user
side
of
things,
do
you
feel
like
Alpha
to
GA
having
two
conformant
implementations,
two
implementations
committed,
you
know:
do
you
think
this
is
a
good
change
from
your
perspective.
B
A
A
D
C
A
D
A
Okay,
I
will
follow
up
with
Matia
on
that
one
and
just
kind
of
offline
or
asynchronously,
rather
kind
of
get
that
one
moving
again.
A
F
A
Right,
that's
a
trick
at
the
moment,
yes,
and
for
for
anybody
who
didn't
pick
up
on
this
I
personally
am
trying
to
start
up
the
what
we're
calling
conformance
profiles
now
thing
which
would
help
to
kind
of
get
us
to
the
point
where
we
are
reporting
on
these
things.
A
So
we
should
be
hopefully
this
year,
starting
to
do
better
at
that
all
right,
so
that
one
just
kind
of
needs,
some
call.
It
updates.
A
Oh
this
one
I
made
because
we
had
a
conversation.
Well,
we've
had
a
conversation
about
this
multiple
times,
there's
actually
some
prior
art,
which
is
linked
here.
A
I
haven't
personally
had
the
Mo,
the
the
correct
motivating
feature
like
the
things
in
my
work
and
stuff
like
that
to
motivate
me
to
complete
it,
but
we
did
make
this
issue.
We
know
now
that
we
have
at
least
four
implementations
that
do
something
like
this
static
and
unmanaged
Gateway.
For
those
who
don't
know
is
the
idea
that
you
could
have
a
Gateway
implementation,
where
you
don't
provision
the
Gateway
or
you're,
not
responsible
for
the
life
cycle
of
the
actual
Gateway
resource
that
is
posted,
but
instead
that
life
cycle
is
managed
elsewhere.
A
In
some
cases,
that
may
be
by
helm
by
hand
by
a
human,
but
you
still
be
able
to
use
the
Gateway
Resource
as
kind
of
a
reflection
of
what's
going
on
with
that
existing
Gateway.
Multiple
implementations
do
this
today
and
there
is
some
prior
art
as
to
like
what
it
kind
of
means,
but
it's
never
been
fully
documented,
and
so
this
is
becoming
kind
of
a
sticking
point
for
gamma,
because
there
is
some
need
among
I
think
it
was
specifically
Linker.
A
I,
don't
know
what
to
say
about
this.
One
necessarily
accept
that
it's
it
needs
somebody
to
grab
it
and
kind
of
go
with
it.
B
D
A
Apparently,
with
Linker
D
and
I'm
probably
going
to
be
wrong
here,
but
I'll
try
apparently
with
link
a
d
I
I
know
with
Kuma.
There
is
a
there's,
a
Gateway
API
implementation
that
actually
supports
a
gateway
and
so
I
think
Linker
D
is
just
trying
to
do
something
similar
where
you
can
get
Ingress
requests
into
the
mesh
network
using
Gateway
API,
that's
my
guess,
and
so,
but
they
don't
have.
A
If
I'm
understanding
the
problem
correctly,
they
don't
have
an
operator.
So
unfortunately,
Flynn
was
the
person
who
really
particularly
wanted
to
have
this
kind
of
conversation
he's
not
here
right
now,
because
this
is
a
at
a
prohibitive
time
of
day
for
him.
A
So
I
think
the
best
follow-up
actually
would
be
for
me
to
go
ahead
and
just
kind
of
double
check
with
Flynn
and
make
sure
that
he
posts
something
in
here
stating
we're
like
making
sure
we're
sure
about
where
he's
actually
coming
from
and
that
I'm
right
or
wrong
about
what
I
think
is
the
reason
that
he
wanted
this
in
the
first
place.
A
We
got
two
minutes,
so
I'll
probably
be
the
last
one
here.
If
we
even
have
time
to
go
over
it
session
persistence
is
this
one?
That's
like
we
just
need
to
do
some
kind
of
a
catch-up
or
does
this
require.
D
J
J
Yeah
I
kind
of
had
the
same
question.
I
wasn't
really
sure
how
to
proceed.
I
think
reviews
would
help
and
I
I'm
not
sure
what
else.
If
there's
something
I
can
do,
let
me
know,
but.
D
J
C
A
A
You
all
right
cool
all
right
and
we
are
right
at
time.
So
let's
go
ahead
and
close
it
up.
We
had
one
issue
left,
that's
the
first
time
we
got
to
triage,
we
got
through
everything
at
least
did
enough
to
bump
them.
Thank
you
very
much.
Everybody
for
your
time
feel
free
to.
If
you
had
anything
that
came
up
here
today,
make
sure
that
you
put
stuff
on
for
next
week's
agenda,
or
you
know,
catch
up
with
us
in
slack
until
next
time.