►
From YouTube: SIG Network Gateway API meeting for 20230522
Description
SIG Network Gateway API meeting for 20230522
A
All
right
welcome
to
the
Gateway
API
meeting
today
is
May
22nd,
at
least
in
California,
happy
happy,
Monday,
everyone
and
yeah.
Just
we've
got
a
pretty
packed
agenda
today,
but
don't
hesitate
to
add
things.
We
might
have
a
little
a
little
bit
of
time
here.
So
this
agenda
dropped
it
in
chat,
don't
hesitate
to
add
things
throughout,
especially
triage.
If
you
have
things
that
PR's
issues
whatever
that
could
use
some
attention,
please
add
them
to
this
list.
A
I
would
just
want
to
make
sure
that
we're
not
losing
track
of
things
and
there's
been
a
lot
of
movement
over
the
past
week,
so
we
probably
have
missed
at
least
one
thing
so
yeah,
that's
all!
As
a
quick
reminder.
This
meeting
is
covered
by
the
kubernetes
code
of
conduct
in
general.
That
just
means
to
be
nice
to
each
other.
Please.
A
We're
all
here
just
to
try
and
make
this
a
better,
a
better
project,
and
you
know
we.
C
A
To
help
each
other
out,
everyone
is
coming
from
different
backgrounds,
so
be
friendly.
That's
all
with
that.
I'll
get
started
right
away
on
the
first
item
in
on
the
agenda,
which
is
Shane's
looking
at
different
meeting
time
that
we
could
try
and
Alternate
with
maybe
Shane
you
wanna
yeah.
D
Simba,
it
is
time
we've
had
really
good.
We've
had
really
good
numbers
on
the
meetings
that
we've
done.
The
alternative
times
for
I
did
just
finish.
Sending
out
the
update
for
next
week's,
so
I
mean
this
was
just
to
make
sure
we
do
more
of
them.
I
would
like
to
think
at
this
point.
We
could
probably
commit
to
doing
a
regular
alternative
time.
D
I
feel
like
we
could
at
least
try
it.
We
don't
necessarily
have
to
do
it
as
a
bi-weekly
kind
of
thing.
I
think
it
might
be
okay
to
just
for
a
couple
of
months.
Try
like
a
one
one
a
month
and
then,
if
it's
still
really
good
turnout,
we
could
up
it
to
two
see
how
that
goes.
A
Yeah
I
think
that's
a
reasonable
Direction.
It
is
unfortunate
because
when,
when
Nick
is
no
longer
in
Europe
at
that
time
will
be
really
hard
for
him
and
always
good
feedback
from
him.
So
yeah.
Maybe
we
can
start
with
like
what
you
said.
A
you
know
try
this
for
once
a
month
and
the
reason
that
we're
moving
it
next
week
is.
A
It
was
basically
between
not
having
a
meeting,
because,
if
you're
based
in
the
US,
you
probably
have
a
long
weekend
and
are
not
working
on
Monday
and
I
think
there
was
also
a
holiday
in
Germany
as
well
on
Monday,
so
anyway
moving
it
off
to
Tuesday
at
least
for
next
week,
and
then
maybe
we
can
aim
for
once
a
month
or
so.
A
Keith
asked
a
really
good
question
on
this
doc.
That
I
wanted
to
highlight
at
least
the
goal
of
using
Tuesdays
was
was
we
were
trying
to
alternate
with
the
time
that
gamma
has
so
that
gamma
can
have
a
meeting
at
approximately
that
approximately
this
time
slot
and
then
in
roughly
the
same
time,
slot
Gateway
would
have
one
so
definitely
trying
not
to
conflict,
but
it
does
mean,
as
we
experienced
last
week.
A
It
does
mean
that
whenever
we
have
these
morning
meeting
well
morning
for
me
meetings
these
earlier
time
slots,
it
means
that
we
do
have
the
potential
of
having
both
a
Gateway
API
meeting
and
a
gamma
meeting
in
the
same
day,
which
maybe
isn't
great
either.
So
we
may.
We
may
want
to
fine-tune
that
a
bit
still
open
to
ideas,
yeah
and
I-
don't
know
if
anyone
else
on
the
gamma
side
wants
to
chime
in
here,
but
I
think
from
gamma
side.
A
You've
also
been
getting
a
bit
more
attendance
at
your
earlier
time
slot.
It's
that
accurate.
A
That's
good
here,
yeah
and
I.
Think
last
week
we
we
topped
out
around
23
I,
want
to
say
in
our
earlier
time
slot
so
around
the
same,
maybe
a
little
bit
higher
as.
F
A
So
yeah,
okay
I'll,
keep
on
moving
then
the
the
next
one
is
an
issue
that
I
raised
a
little
bit
ago
and
maybe
it'd
be
useful
to
have
a
bit
of
a
poll.
I've
asked
around
in
slack
too
and
I'm,
realizing
that
my
my
question
is
not
as
clear
as
I
would
like
it
to
be
so
maybe
I'll
drop
it
back
to
here
for
a
second
and
what
I'm
really
trying
to
ask
is:
should
a
request
to
Slash
Foo
yeah
like
this
match
a
well
okay.
A
Let
me
should
an
exact
match
on
slash
Foo
match
I
request,
a
slash,
poo
request,
I
think
that
the
answers
I'm
getting
suggest
no.
A
But
it
is
not
clear
from
the
spec
what
that
is,
so
maybe
we
can
just
open
up
for
a
little
bit
of
conversation
here
of
implementations
today,
I'm
hearing
some
that
are
saying
No
this.
This
would
not
match
Shane
go
ahead.
G
H
H
I
mean
we
should
check
the
rfcs
and
whatnot,
but
if,
if
not,
it
seems
a
bit
a
bit
sketchy
to
do
that,
I
agree
at
the
very
least
like.
I
H
One
more
thing
at
the
very
least
that,
with
the
trailing
slash,
should
not
match
one
without
a
drone
Splash.
Maybe
we
could
say
if
if
there
is
no
trailing
slash
in
the
match
that
we
match
one
with
the
trailing
slash,
although
that's
still
a
little
suspicious,
but
that
seems
a
little
better
to
me.
D
I
can
still
see
this
being
a
problem
potentially
down
the
road
somebody's
like
well
I
put
in
Foo,
and
it
wasn't
no,
but
it
does
make
seem
to
make
sense
to
me,
like
exact
match
means
exact
match.
You
know.
J
A
Yeah
for
for
prefix
match
the
spec
is
actually
very
clear,
and
we
have
said
that
a
trailing
slash
is
ignored
for
exact
match.
Our
spec
is
just
you
know,
matches
the
URL
path,
exactly
which
I
guess,
if
you
take
that
very
literally
then
yeah,
the
the
trailing
slash
is
Meaningful,
and
that
seems
to
be
what
the
the
overwhelming
feedback
here
or
for
con
a
little
bit
more
context.
A
I
have
a
related
issue,
which
is,
it
seems
entirely
reasonable
that
some
people
are
going
to
want
to
be
able
to
match
requests
with
a
trailing
slash
and
redirect
to
requests
that
don't
have
a
trailing
slash
and
if
we
said
that
the
trailing
slash
was
not
meaningful
on
a
exact
match,
request.
C
Oh
also
Ingress
V1,
the
conformance
tests
said
that
TransLink
selections
were
significant
if
that
matters
at
all
yeah.
C
A
F
Okay,
can
I
ask
a
question
real,
quick
yeah
not
to
open
up
a
whole
new
can
of
worms,
but
I'm
gonna.
Do
it?
Why
are
slashes
trailing
slashes
ignored
on
prefix
matches.
A
Yeah,
that's
I'd
have
to
go
back
to
that
conversation.
That
was
a
very
good
reason
at
the
time.
Let's,
let's
ask
chat.
A
I
Yeah
yeah,
that
was
my
point
I've
seen
I,
seem
to
remember
it
being
something
to
do
with.
I
We
were
thinking
a
bit
a
bit
about
I
think
we
might
have
been
thinking
about
inclusion
and
delegation
at
the
time
because
it
really
matters
for
inclusion
and
delegation,
because
if
you
are,
if
you
were
including
a
route
that
with
the
prefix
of
ABC-
and
you
have
another
route
that
has,
if
you're,
including
a
route
with
the
prefix
of
Slash,
ABC,
slash
and
you're,
including
another
route
that
has
slash
daf,
then
the
then
the
path
would
be
slash.
I
Abc,
slash,
slash
def,
but
the
the
I
think
the
intent
here
is
to
make
it
so
that
prefix
means
kind
of
is
intended
to
mean
path
fragment,
not.
C
I
You
know
like
a
a
piece
of
a
path,
not
like
literally
a
prefix,
because
you
know
it's
it's
kind
of
more
intended
to
be
used
for
slash,
ABC,
slash,
something,
not
slash.
Abcdef
yeah,
not
a
string
prefix.
Exactly
thanks
been
doing.
A
I
A
But
the
path
prefix
match
that
exists
today
is
is
what
what
Nick
described
it
as
a
matching
a
segment
of
a
path
and
that's
it
so
that
the
trailing
slash
is
not
particularly
meaningful
when
you're
just
saying
I
want
to
match
the
segment
that
starts
with
or
segments
that
start
with
this
yeah
anyway,
but
we
we
have
left
room,
you
could
have
a
different
prefix
match
type.
A
J
Will
motivation
something
yeah,
so
in
all
of
these
cases,
right
from
a
usability
perspective,
when
you
work
with
people
trying
to
leverage
this
Technologies,
they
tend
to
come
from
some
something
in
the
past.
That
they've
worked
with
pcre,
for
example,
is
a
great
example
of
something
that
people
are
very
familiar
with
and
when
they
attempt
to
use
this,
if
we
standardize
with
something
in
the
past,
that's
reasonable.
J
For
us,
we
should
just
advertise
that
as
our
path
forward,
we
are
compatible
with
the
way
matches
happen,
as
it
does
with
this
standard
piece
of
equipment,
so
that
you
know
that
that
sort
of
smoothens
the
learning
curve
of
it
and
leads
to
fewer
mistakes.
It's
just
been
my
experience
in
the
past.
A
Yeah
I
think
that's
very
true
and
a
very
good
tip
for
any
of
our
documentation
to
try
to
reference
things
that
you
know
our
concepts
are
coming
from.
You
know
where
you
know
the
the
way
that
we're
doing
this
kind
of
matching
is
entirely
compatible
with
X
technology.
I
would
be
very
encouraging
and
appreciative
of
any
any
contributions
to
docs
anywhere
like
that,
because
I
think
that'd
be
very
helpful.
A
A
It
yeah
just
an
FYI.
This
looks
awesome,
I.
A
Yeah
anyway,
this
is
the
thing,
so
cable
API
controller
is
coming
along
on
on
various
fronts,
but
it's
cool
to
see
another
Community
implementation
and
Dan
does
all
kinds
of
of
cool
things.
So
I'm
excited
to
see
where
this
one
lands-
and
this
is
he
was
working
on.
What's
the
the
other
one
LinkedIn
slack
that
I'm
missing
out
on
right
now,
yeah
Cube,
VIP,
cable
API,
which
does
anyone
remember,
is
this?
Is
this
the
same
thing
like
I
Know
Dan
was
working
on
both
of
these
yeah.
I
A
Yeah
cool
and
hopefully
we'll
have
more
time
for
him
to
present
talk
about
this
in
our
next
meeting,
because
it
should
be
more
friendly
to
his
time
zone.
Yeah.
D
Yep,
just
the
Prototype
merged,
or
rather
a
usable
prototype,
is
now
merged.
It's
under
an
experimental
go
build
tag,
so
don't
get
caught
up
on
that,
but
it's
an
alternative
test
Suite
that
you
could
try
out
I
believe
at
this
point
we've
had
two
implementations:
try
it
out
so
just
wanted
to
kind
of
raise
that
to
people
that
it's
there
for
checking
it
out
making
some
updates
take
a
look
at
the
gap
which
is
1709
and
the
next
step
is
kind
of
we're
heading
towards
marking
it
implementable.
D
A
Yeah
that
that
makes
sense
and
and
I'm
so
excited
to
see
this
move
forward.
I
think
since
many
implementations
are
represented
here,
I
want
to
highlight
one
of
the
takeaways
and
goals
of
of
this
performance
profile
work,
and
that
is
that
we
want
to
highlight
the
results
on
our
implementations
page,
so
we
want
to
really
make
it
clear
to
anyone
who's
trying
to
understand.
You
know
this
is
an
API.
A
So
that's
something
that
will
come
at
some
point
right
now:
the
implementations
paid
is
pretty
wide,
open
and
I
think
it
will
become
more
structured
and
there
will
be
definitely
some
incentive
to
to
run
these.
So
please,
the
earlier
the
better
in
terms
of
running
this
through
trying
to
understand
if
it
works
with
your
implementation,
if
it
doesn't
what
we
can
do
to
help
it,
you
know
be
a
better
fit,
but
this
is
definitely
something
that
we're
trying
to
fit
in
between
now
and
ga
yeah
I
don't
know.
D
Yep
and
we
only
have
a
the
go
test-
Suite
right
now-
nobody
picked
up
the
CLI
one,
so
I
think
Matia
finally
decided
to
go
ahead
and
get
that
started
so
that
we
can
help
people
that
way.
If
you
need
the
CLI
version,
just
please
reach
out
to
Matia
I'm
sure
he
would
be
happy
to
kind
of
he'd
be
happy
to
know
that
you
need
it
and
then
he
might
be.
We
could
break
it
down
into
things
that
we
could
work
on
together.
D
So
that's
kind
of
the
state
of
things
it's
going
good,
just
hoping
for
more
people
to
jump
in,
so
that
we
don't
have
to
come
back
and
change
things,
but
rather
we
can
change
them
on
the
way
in
so,
and
the
next
thing
I
got
is
a
bit
of
interesting
discussion,
so
I
think
a
number
of
you
I've
kind
of
reached
out
to
I've
noticed
with
my
kind
of
Gateway
API
maintainer
hat
on
that
lots
of
people
are
interested
in
multi-cluster,
but
not
a
whole
lot
of
people
have
gas
in
the
tank
for
it.
D
So
I
was
wondering
if
there
was
things
we
could
do
to
help
facilitate.
Basically,
you
know
basically
giving
people
more
to
work
with
to
get
started
as
a
way
to
kind
of
make
it
easier
to
land
and,
like
start
working
on
multi-cluster
implementations,
so
I
just
floated
the
I
worked
with
someone
from
Quadrant
named
Dan
and
or
Dave,
and
we
both
kind
of
brought
the
conversation
to
the
last
multi-cluster.
D
Sync
that
you
know
maybe
there's
room
to
kind
of
build
more
Machinery
in
multi-cluster
and
that
Machinery
being
kind
of
like
generic
Machinery
on
top
of
say,
multi-cluster
services
that
would
be
in
service
of
potentially
everyone
and
doing
that.
Upstream
might
give
people
kind
of
the
I'm
not
going
all
in
on
this
yet.
But
I
do
want
to
kind
of
make
make
moves
towards
this
kind
of
scenario
where
we
could
all
kind
of
get
Solutions
here
and
start
seeing
things
working
for
those
of
us
who
don't
have
implementations.
D
So
that's
kind
of
what
we
floated.
The
general
response
from
multicluster
Sig
about
like
doing
more
machinery
in
upstream
was
a
little
guarded,
but
I
think
the
main
takeaway
was.
It
seems
like
they
felt
if
more
people
were
willing
to
show
up
to
help
with
that
Machinery
they
would
be
open
to
it.
So
that's
kind
of
just
general
call
out
that
I
wanted
to
point
out.
There's
a
the
meeting
was
last
week.
D
So
if
you
want
to
go
check
it
out
on
YouTube
and
get
a
little
bit
more
context,
if
you're
interested
in
multi-cluster,
you
might
find
that
one
interesting
and
then
in
general,
if
you
want
to
start
showing
up
to
the
multi-cluster
sinks
and
continue
the
conversation
about
whether
or
not
there
might
be
some
room
to
do
more
Upstream
open
source
Machinery
instead
of
everybody
doing
their
own
implementation,
I
think
we
could
all
end
up.
D
G
Yeah,
so
this
is
extremely
exciting
to
me
for
a
couple
of
reasons,
the
main
one
being
this
is
kind
of
just
not
personal
opinion,
but
I
get
the
sense
that
there
is
I,
don't
know
that
MCS
API
is
as
complete
as
we
might
want
it
to
be,
especially
coming
from
a
mesh
perspective.
G
G
So
you
know
it's
not.
It's
indicated
to
change
some
of
it,
but
I
do
think
part
of
the
reason
why
it
has
an
involved
in
the
way
that
I've
been
personally
like
looking
for
is,
is
because
you
know
there
is
the
start
of
cost.
You
anything
will
take
cluster.
You
have
to
create
a
controller
and
there's
some
not
lots
of
non-trivial
problems
in
creating
a
controller
and
everyone
in
each
implementation.
He
does
think
something
differently.
G
I
think
that
there's
a
lot
of
conversation
that
can
be
driven
through
like
a
reference
implementation
for
people
who
just
want
to
play
around
with
it
or
start
to
evaluate
it
as
an
API
I,
think
it'll
reveal
some
I
guess
chinks
in
the
armor
lack
of
a
better
term,
so
I
I
love
to
see
it
happen.
I'm
worried
that
the
that
that
they
need
it
for
something
like
this
Machinery
cluster
to
happen.
G
It
just
facing
the
same
problem
that
the
MCS
API
in
general
has
had,
which
is
just
bodies
and
people
who
have
the
bandwidth
to
be
able
to
to
invest,
and
that's
where
I
currently
am
because
I
would
love
to
do
this
kind
of
stuff.
But
just
don't
have
the
bang
with
that.
At
the.
D
E
Foreign
yeah
I
think
bandwidth.
Discussions
aside,
I
think
one
of
the
biggest
things
that
Sig
multi-cluster
needs
in
terms
of
like
intersections
with
Gateway
API,
is
for
implementers
to
start
playing
with
it
and
I
know.
There's
a
few
folks
that
have
like
had
some
critiques
of
it,
where,
like
some
things,
feel
more
aligned
to
a
service,
Discovery
use
case,
which
is
what
it
was
designed
for
to
be
fair
and
it
doesn't
always
fit
white,
as
you
might
expect,
with
mesh
sorry
about
the
dog
yeah.
E
So
though
it
seems
like
the
core
resources
are
pretty
solid
and
it's
more
of
just
like
the
I
guess,
implementation
details
and
getting
more
folks
involved
and
whatever
we
can
do
to
help
facilitate
implementations.
I
know
that
this
work
that
started
a
conformance
test
I
think
that's
really
encouraging
and
can
go
a
long
way
towards
being
part
of
that
Machinery
but
yeah.
It
was
like.
E
D
No
I,
it's
a
great
point
and
I
think
I
agree
with
you
entirely
that
if
we
just
did
more
stuff
in
this
space
together,
we
would
probably
come
up
with
some
better
understanding
if
nothing
else
right
so
I,
I'm
really
down
for
that
and
I
think
I
mean
for
those
of
you
who
think
that
the
bike
shed
needs
to
be
purple.
D
I,
don't
really
have
any
specific
ideas
at
the
moment,
except
maybe
vaguely
we
talked
about
potentially
just
like,
starting
with
the
first
obvious
stepping
stone
which
is
like
solving
efficient
service.
Import
like
that
might
be
a
place
for
a
non-test,
related
Machinery.
But
again
it's
all
open
questions.
D
I'd
like
you
to
if
you're
interested
in
those
questions
and
interested
in
maybe
I
can't
go
all
in
on
an
MCS
implementation
today,
but
I
might
be
able
to
go
in
if,
like
you
know,
five
or
six
of
us
are
going
in
and
we
could
do
it
together
feel
free
to
PM
me
and
also
just
go
to
the
next
meeting.
We
can
put
something
on
the
agenda
and
kind
of
continue
the
conversation
a
little
bit.
They
say
they
seem
very
open
to
that
last
week.
E
Yeah
I
think
one
of
the
things
that's
tricky
about
MTS.
In
particular.
It's
just
that
an
implementation
is
basically
best
done
in
a
way
that
is
not
using
a
single
cluster
as
a
source
of
Truth,
which
makes
it
a
little
bit
difficult.
E
G
I
just
wanted
to
add
something
slightly
tangent,
but
currently
talking
about
segment
in
class
to
make
me
think
about
it.
The
other
big
piece
of
feedback
that
I've
gotten
from
internal
teams
that
can
Implement
NCS
aside
from
logic
Machinery,
has
been
the
kept
process
in
the
iteration
Cycles.
G
You
know,
I've
got
people
saying
that
we
we
love
to
work
more
Upstream,
but
we
can't
spend
three
years
getting
the
spec
promoted
through
the
the
cat
process
to
like
stable
before
you
know,
we've
got
to
have
our
product
in
you
know
within
the
year,
and
things
like
that,
so
I
would
be
I
think
if
we
could
I'm
very
interested
in
working
with
glitchy
cluster.
G
This
would
be
action
you're,
paying
with
four
like
seeing
if
they,
if
they're
open
to
something
like
a
like
a
depth
process,
I,
don't
even
know
what
that
acronym
would
be
or
how
to
pronounce
it.
But
if
there
is
kind
of
a
more
that
more
quickly
iterating
process
for
folks
I
know,
I've
got
people
at
Microsoft
who
would
be
more
open
to
contributing
Upstream
and
sharing
products
with
community.
In
that
way,.
A
Yeah
I
think
that's
a
really
good
point.
I
I
feel
like
there
are
definitely
people
that
that
want
to
want
to
do
that.
So
maybe
it's
just
starting
the
conversation
our
meeting
time
next
week,
8
30
Pacific,
will
be
immediately
follow
I
think
it
will
be
immediately
followed
by
the
multi-cluster
meeting.
A
So
maybe
there's
some
some
room
there
for
a
little
bit
of
overlap.
Maybe
we
can
invite
some
multi-cluster
people
to
our
call.
Vice
versa,
there's
a
ton
of
overlap
in
the
things
we're
doing
and
you
know
I
know
it's
been
hard
to
I.
Think
my
understanding
from
the
multi-cluster
people
I've
talked
with
it's.
A
They
they've
had
trouble
getting
a
lot,
many
people
to
really
contribute
and
buy
in,
but
I
think
I
mean
I'm
really
biased
here,
but
I
think
that
Gateway
brings
a
lot
of
value
on
top
of
multi-cluster,
and
so
when
we
can
combine
these
two,
it
becomes
that
much
more
interesting
and
compelling
to
work
together
and
just
build
value
on
top.
D
Can
you
hear
me
this
time?
Okay,
yeah
I
had
to
rejoin
I
can
see
my
things
working
now,
I
had
to
rejoin
because
I'm
happy
with
myself,
but
yeah
I.
Think
Keith.
If
it
wasn't
already
said
while
I
was
rejoining,
I
would
just
go
and
put
that
right
on
their
agenda
for
the
next
meeting
and
just
bring
it
up.
I
think
that
they'd
be
I
mean
obviously
they're
the
ones
that
need
to
hear
that
right.
D
Yeah
John
asks
in
chat.
What
are
the
gaps
today?
Api
changes
or
machinery
I
think
that
is
a
good
question
and
actually
Rob
I
feel
like
you
might
be
able
to
answer
that
a
little
better
than
me.
A
It
depends
on
your
perspective,
I.
You
know,
obviously,
from
the
like
GK
networking
perspective.
We
have
an
implementation
and
it
mostly
kind
of
works.
So
I
can
say:
oh
yeah.
A
Well,
it
works,
but
I
I
also
recognize
that
that
there
are
issues
and
some
things
that
that
could
be
smoothed
out
and
and
I
also
understand
that
the
Machinery
to
support
this
is
complicated
and
as
we've
gotten
towards
multi-cluster
Gateway
and
as
we've
started
to
talk
about
the
intersection
of
this
API
with
gamma
or
realizing
that
there
are
some
complexities
that
feel
like
they
don't
necessarily
need
to
be
solved.
A
Like
you
know,
one
low
hanging
fruit,
that's
already
been
brought
up
in
Sig
multi-cluster-
is
that
currently
the
spec
or
maybe
previously
respect
required
that
you
publish
endpoint
slices
in
every
cluster
and
kind
of
propagate
them
across
and
for
somebody
that's
just
trying
to
implement
this
via
mesh
or
in
cluster
implementation.
You
probably
already
have
this
information,
but
you
don't
want
to
go
through
the
effort
of
you
know,
keeping
all
the
endpoint
building
your
own
endpoint
slice
controller.
A
That's
I
think
a
good
example
of
things
that
we
can
remove
or
simplify
and
I
know
that
specific
one
is
already
in
progress,
but
I
imagine
there
are
others
that
we're
going
to
run
into
and
the
best
way
to
run
into
them
is
by
by
trying
in
or
just
even
looking
at
the
spec
and
trying
to
gauge.
How
would
I
implement
this
and
how
painful
it
would
it
be
and
trying
to
remove
those
so
John,
you
may
have
had
some
some
things
in
mind
and
I.
A
Think
Mikey
also
commented
but
yeah
that's
my
perspective.
I
Sense
from
from
the
psyllium's
point
of
view,
yeah,
we
don't
because
we
already
have
a
cluster
mesh
thing
and
it
does
not
use
endpoints
losses.
Also,
it's
important
at
the
moment
yeah,
the
more
that
we
can
not
have
to
worry
about
the
endpoint
slices,
the
better
off
it
will
be
the
easier
it
will
be
for
us
to
do
to
support
multicluster
API.
E
I
saw
a
talk
on
cluster
messages.
It
looked
like
it
would
actually
be
a
good
fit
some
of
the
important
details
inside
it
could
adopt
them.
Yeah.
I
Look
I
mean
we
customers
effectively
already
Imports
services
from
the
other
clusters.
It
just
doesn't
it
Imports
them
as
services,
not
as
service
Imports?
I
So
it
should
be
a
reasonably
straightforward
change
to
make
that
the
you
know
that
you
also
see
a
service
import
like
or
or
you
don't
see
yourself
I'm,
not
sure,
exactly
that
again:
implementation,
blah
blah
blah,
but
like
the
mechanisms,
the
mechanisms
to
do
all
of
that
stuff
are
already
there
in
psyllium
and
so
like
just
making
it
so
that
the
spare
concilium
line
up
shouldn't
be
Dart
hard.
If
you
know,
if
the,
if
the
endpoints
don't
need
to
be
included,
that
would
be
a
big
deal.
D
Great
cool
well
I,
probably
took
up
a
little
bit
too
much
time
with
this
one,
but
I
think
everybody's
thoughts
on
this
was
great
and
please
do
join
next
week
at
12,
30
Eastern
I
want
to
say
that's
9,
30
Pacific
for.
A
The
city
yeah
it's
it's
immediately
after
our
Gateway
meeting,
so
whatever
time
zone
that
is
yeah
I
had
an
hour
yeah
cool,
great
discussion,
thanks
for
bringing
that
up
Shane
next,
we
have
method
match
priorities.
Today.
Are
you
here.
C
Yep
looks
like
this
is
sort
of
getting
handled
just
generally
in
GitHub,
but
it's
one
of
the
recent
conformance
tests,
PR's
added
a
test
that
added
a
priority
for
method
matching
to
the
one
of
the
tests.
C
So
the
matches
in
the
test
had
header
matching
more
significant
than
method
matching,
that's
not
actually
specified
in
the
spec
yet
which
this
issue
was
open
to
specify
that
so
just
wanted
to
yeah
I
think
people
have
replied
on
on
GitHub
and
whatnot,
but
looks
like
this
is
getting
handled,
but
I
just
maybe
opened
it
up
to
discussion
on.
C
If
we
want
to
handle
this
in
the
zero
seven
release,
what
people
think
about
the
relative
priority
of
method,
HTTP
method,
matching
versus
other
attributes
and
requests
existing
PR
to
change.
It
looks
like
it's
adding
method
matching
as
the
least
significant
attribute,
I,
don't
know,
that's
what
people
agree
with
or
not,
but
yep
we're
just
calling
it
out
in
case.
You
have
opinions
and
want
to
add
it
to
the
add
to
the
discussion.
I
Yeah
I,
probably
would
have
put
it
I,
probably
would
have
put
it
above
header
would
have
been
what
I
would
put
it
personally,
but
yes,
I'd
be
interested
to
talk
more
about
that.
Probably
just
talk
on
the.
A
You
know
it
probably
does
make
sense
somewhere
above
where
it
is
right
now,
but
it
just
it
just
doesn't
match
with
the
rest
of
these
in
in
sentence
structure,
so
it'll
be
require
some
wordsmithing
I
guess,
but
I
I
don't
have
a
strong
opinion
on
exactly
where
it
lives.
It
does
seem
like
a
few
people
are
suggesting
between
path
match
and
header.
Match
seems
okay
to
me.
C
Yeah
conceptually
right
is
it
does
it
make
sense
to
segmented
subset
of
requests
based
on
the
method
more
significantly
than
the
header,
a
header
I?
Yes,
I,
guess,
I,
don't
know
people
do
lots
of
things
so
yeah
I,
don't
know.
If
there's
any
particular
I
mean
that
seems
like
a
valid
use
case.
I
mean
it's
a
quite
a
significant.
It's
differentiating
feature
of
HTTP
request.
C
I.
Think
I,
agree
with
that
General
sentiment,
but
yeah.
A
Okay,
that
that
sounds
good
to
me,
I,
don't
know
if
karov
not
on
this
call,
but
okay,
it
does
seem
like
this
is
already
getting
handled
in
in
GitHub.
So
maybe
just
we
can
keep
on
adding
comments
here.
It
seems
like
a
pretty
tiny
change
to
move
this
up
a
bit
and
do
a
bit
of
word
sniffing
around
it.
A
I
added
a
comment
on
the
the
issue
here,
just
agreeing
with
with
what
others
have
said
that
this
really
should
belong
in
O7
I
think
it
should
be
an
071
milestone
and
I'd
love
to
see
it
get
resolved
this
week,
because
we're
aiming
to
release
071
early
next
week,
so
I
think
we're
on
track
for
that.
Just
please,
as
you're
reviewing
things
prioritize
reviewing
this
specific
PR
as
it
evolves
over
time.
A
A
Cool
all
right
next
up,
I
just
wanted
to
to
come
back
to
policy
again.
I
know:
we've
been
discussing
this
a
little
bit
and
unfortunately
Flynn
I,
don't
think
is
on
this
call.
A
I
know
others
who've
been
involved
in
the
conversation
are
on
this
call,
so
we
can
discuss
a
little
bit
but
I'm
just
trying
to
get
an
idea
of
where
we
can
go
next.
I
think,
Flynn
and
and
I
think
Shane
helped
with
this
proposal
as
well.
Have
you
know
good
thoughts
here
about
the
the
complexity
of
of
the
model
that
we
have
right
now,
I'm
biased
in
that
I?
A
Think
a
lot
of
these
can
be
solved
by
making
the
existing
model
more
discoverable,
but
Nick
I
think
you,
you
have
your
own
thoughts.
I
know,
there's
been
lots
of
good
useful
discussion
on
this
PR
so
definitely
encourage
others
to
comment
here.
If
you
have
your
own
thoughts,
I
think
there's
been
some
really
good,
solid
discussion
throughout,
but
yeah
maybe
I'll
hand
it
off
to
Nick
before
I
go
too
far.
I
Yeah
Flynn
and
I
had
to
catch
up
on
this
on
Friday
and
I.
Think
I've
got
some
I've
got
an
update,
sort
of
drafting
it's
long
already,
but
yeah.
So
just
a
broadly
summarize,
I
think
we
talked
a
little
bit
about
I.
I
Think
I
saw
Europe
by
today
Rob
that
I
think
the
the
exact
hymns
sort
of
saying
it's,
not
the
clarity,
is
sort
of
based
on
a
based
on
a
sort
of
very
specific,
like
on
the
experience
of
Jane
in
the
in
his
example,
and
you,
like
yeah
I,
think
we're
basically
saying
the
same
thing.
It's
just
a
slightly
different
way
of
saying
it.
So
I've
got
a
bit
to
clarify
that,
but
yeah
I
think
I.
Think.
I
The
key
part
here,
in
my
mind,
is
clarifying
the
discoverability
and
and
adding
some
ways
to
to
sort
of
address
the
discoverability,
the
yeah,
the
discoverability
problem.
I,
am
sort
of
writing
up
a
bit
of
a
it's,
probably
going
to
end
up
being
a
mini
game
of
its
own.
In
that
comment,
I
think
just
sort
of
running
through,
like
here's,
the
things
I
think
we
could
do
and
he's
you
know,
here's
why
we
haven't
just
done
them.
I
You
know,
and
some
of
that
probably
will
end
up
getting
folded
back
into
the
Gap
as
Alternatives
considered,
or
something
like
that.
I
think,
but
I'll
put
it
in
here
for
now,
so
that
we
can
have
the
discussion
all
in
one
place
and
yeah
I
think
that
to
broadly
broadly
summarize,
I
think
is
for
the
discoverability
of
this.
There
are
three
things
that
a
a
user
of
a
route
needs
to
know.
Personally,
they
need
to
know
that
a
policy
is
affecting
their
route.
I
Obviously
you,
knowing
what
policy
settings
are
being
changed
on
your
route,
will
generally
tell
you
the
other
stuff
as
well,
but
you
you
being
able
to
keep
the
what
and
specify
it
is
actually
surprisingly
difficult
and
has
scalability
problems,
so
we
might
have
to
step
back
to
one
of
the
other
ones
or
have
some
other
way
to
do
it.
I
That's
what
a
lot
of
the
other
things
are
for
so
I'm,
just
in
the
process
of
kind
of
explaining
that,
in
a
way
that
doesn't
sort
of
have
me
rambling
for
30
seconds
and
it's
a
bit
more
comprehensible
but
yeah.
So
I'll
put
another
thing
on
there.
Soon.
I
definitely
encourage
other
people
to
not
have
a
rating
have
rated
discussion
and
say
what
you
think
this
is
getting
it's
sort
of
starting
to
get
down
into.
You
know,
let's
argue
about
API
Machinery
territory
and
so
which
always
gets
a
little
hairy.
I
So
yeah
look
that's
one
of
the
reasons
why
this
is
complicated
is
because
we're
sort
of
doing
we're
doing
stuff
that
hasn't
been
done
a
lot
in
kubernetes
before
this
sort
of
having
a
reasonably
complicated
system
of
of
objects.
That
also
has
other
objects,
sort
of
modifying
it
is
very
complicated.
It's
all
complex,
I
think
is
a
better
way
to
say
it.
Ideally,
it
can
be
complex
without
being
complicated.
I
You
know,
I
think
that's
what
we're
trying
to
work
towards
it
will
be.
It
will
always
be
complex
because
it's
solving
a
very
complex
problem,
but
ideally
we
can
make
it
so
that
it's
not
too
complicated
so
that
it's
easier
to
understand
the
state
of
the
system.
At
any
point,
one
point
in
time.
A
Yeah,
thank
you
for
for
the
update,
I
I
know.
I've
also
spent
some
time
chatting
with
Flynn
about
this
as
well,
but
it
sounds
like
you've
got
a
really
thorough,
update
coming
so
I'm.
Looking
forward
to
that,
thank
you.
A
I
think
you
know
that
yeah,
there
are
a
variety
of
things
here.
From
my
perspective,
I
I
wish
I
wish
there
were
a
better
way
than
what
we
have,
but
I
don't
know
that,
there's
a
better
way
to
accomplish
all
the
goals
that
we
have.
That's
again,
my
own
bias
perspective
on
the
world
with
that
said,
I
do
think
it's
really
important
and
this
kind
of
comes
back
to
what
you
were
talking
about.
A
Nick
of
we
need
to
make
it
easier
to
understand,
fundamentally
that
there
is
a
policy
that
is
affecting
your
thing,
that
that
is.
You
know
that
is
key
to
everything
here
and
it's,
it's
I,
think
led
to
a
lot
of
the
frustration
that
we've
seen,
including
including
this
you
know
the
the
parabelle
the
parable
from
Flynn
in
this
Gap
was
very
well
written
and
I.
Think
a
lot
of
the
frustration
in
that
story
was
that
the
the
author
of
the
route
did
not
understand
that
policy
had
been
affecting
their
route.
A
It
was
not
visible
to
them,
so
I
I
personally
would
really
like
to
invest
in
at
least
a
couple
ways
to
improve
that
one
of
them
kind
of
but
but
I,
see,
there's
already
a
couple
hands,
so
maybe
I'll
stop
what
I'm,
saying
and
defer
I
think
Shane
you're
up.
First.
D
F
Yeah
sorry,
okay,
yeah
I'll,
go
I,
haven't
read
all
this
stuff,
so
maybe
this
is
more
complicated
and
I.
Just
don't
have
the
contacts,
but
as
I
was
listening
to
you
talk
about
it,
it
sounded
like
this
is
something
that's
just
go
on
the
route
status
like.
If
a
policy
is
being
applied
to
a
route,
it
should
be
on
the
status.
F
I
There
there
are
reasons
of
basic
it
comes
down
to
Fair.
Now
it's
very
possible
to
have
one
policy
affect
many
Logics,
and
so,
if
you're
updating
the
status
on
every
route,
because
a
policy
got
updated,
it's
a
really
good
way
to
produce
a
lot
of
extra
load
of
the
API
server
and
historically
doing
that
sort
of
one
object.
K
Yeah
I
mean
I
I,
always
think
initially,
I
was
thinking
about
what
he
said,
but
I
think
you
know
so.
Flynn
and
I
have
talked
about
this
like
a
little
bit
in
like
previous
discussions
and
in
my
mind,
like
one
of
the
biggest
things
that
I
think
would
actually
be,
or
what's
on
my
mind
when
I
think
about
policy
attachment
is
the
the
debug
story
so
like.
K
If
I'm,
you
know
debugging
around
whether
I'm
Jane
or
whatever
fun
it
calls
the
infrastructure
engineer
manager,
you
know,
one
of
the
things
I
want
is
super
useful
to
know
is
like
having
a
full
state
of
like
how
each
of
the
resources
are
sort
of
related
to
each
other
in
terms
so
like
not.
So
it's
not
just
with
the
policy
attachments,
but
it's
also
like
with
the
browse
and
like
gateways,
and
you
know
like
which
gateways
are
being
affected
by
okay
with
classes
and
I.
K
Think
there's
a
lot
of
that
could
also
just
be
solved
by
additional
tooling.
So,
for
example
like
having
like
I
think
this
was
brought
up
in
in
the
gap
or
in
one
of
the
discussions
about
having
like
some
sort
of
control,
Plug-In
or
actually
I
would
be
I
would
even
prefer
to
be
part
of
Patrol
itself,
so
sort
of
like
you
have
like
a
control
get
you
know,
dashes
graph
option
that
basically
allows
you
to
visualize,
like
all
of
the
you
know
situation.
K
K
That
and
you
can
also
sort
of
like
chain
it
Upstream
small,
to
succeed
like
okay,
it's
part
of
this
Gateway
and
then
the
Skateway
class.
You
just
like
managing
all
these
gateways
and
things
like
that.
But
and
the
content
gets
around
like
the
sort
of
status
part,
because
now
you
only
hit
an
APS
server
when
you're
calling
it
through
the
CLI
or
some
other
way,
and
not
every
time
you
update.
You
know
your
policy
attachments
for
sure,
presumably.
D
A
Yeah,
those
are
great
points.
I.
That's
that's,
definitely
been
directionally
areas
that
I've
been
interested
in
I.
Think
there's
more
in
the
discussion
kind
of
like
you
referenced
so
there
there
is
a
an
open
discussion
on
just
like.
How
can
we
make
this
better?
Please
submit
your
ideas,
there's
some
ideas,
plenty
of
ideas
already
out
there
that
have
some
comments
on
them,
but
I
definitely
want
to
get
more
more
movement
here
and
I
think
we
already
have
a
couple:
Front
Runners,
one
of
them
like
the
route
status.
A
If
we
can
solve
fan
out
and
the
other
like
Coupe
cuddle,
control,
Cube,
cuddle,
plug-in,
hopefully
more
integrated,
maybe
included
by
default
soon.
I
don't
know,
but
yeah
yeah.
Absolutely
we
I
think
we're
all
agreeing
on
the
the
fundamental
problems
and
we
all
want
to
make
the
situation
better.
I
think
that's
yeah
100.
I
I
think
there's
just
one
one
thing
that
I
wanted
to
quickly
just
said,
which
is
another
thing.
That's
going
to
be
in
my
response,
but
I
think
one
of
the
other
metrics
that
we
need
to
have
is
that
we
need
to
try
and
keep
the
I
call
it
the
troubleshooting
distance
for
for
a
route
as
close
to
one
as
possible.
The
troubleshooting
distance
is
like
how
far
you
have
to
go
to
figure
out.
I
What's
Pro
what
the
problem
is
you
can
in
kubernetes
you
can
use
that
a
good
proxy,
for
that
is
the
number
of
Q
Kettle
commands.
You
need
to
do
to
understand.
What's
going
on
right,
so
one
is
the
best
right.
Like
you
get
one
you
get
the
you
get,
the
the
object
you
created
and
you
understand.
I
What's
broken,
it's
not
always
possible,
but
the
the
more
the
closer
you
can
get
it
to
one,
the
easier,
the
easier
it's
going
to
be,
and
so
you
know
two
is
better
than
is
not
as
good
as
one
but
it's
better
than
yeah
undefined,
because
it's
not
possible
for
you
to
figure
it
out
right
like
so.
I
That's
that's
sort
of
one
of
the
things
I
think
we
should
be
aiming
for
is
we
may
not
be
able
to
get
it
because
the
scalability
concerns,
or
whatever
we
may
not
be
able
to
get
it
to
be
one,
but
we
should
always
be
aiming
to
to
get
that
number
as
low
as
possible.
A
Yeah
well
said:
I
do
want
to
time
box
this,
but
I
know
of
not
you
had
a
a
hand
up.
Maybe
we
can
get
to
you
real,
quick.
J
Yeah
I
was
going
to
talk
about
the
operational
aspect
of
testing
as
well,
so
if
it
were
intended
for
operational
things,
metrics
blogs
other
debug
tasks
as
well
as
attaching
compliance
like
policies.
Let's
say
you
had
to
some
industry
standard
compliance
that
you
wanted
to
flow
to
switch
on.
Maybe
there's
some
way
to
extend
it
for
that.
A
Oh,
that's
really
interesting,
so
basically
to
have
a
set
of
policies
and
settings
that
you
know
comply
to
HIPAA
or
whatever
compliance
standard
gear.
You're
targeting
is
that
the
direction.
B
A
Yeah
that
that's
really
interesting,
yeah
and
and
I,
and
also
your
your
point
there
about
again
kind
of
on
the
observability
front.
We've
we've
been
focused
a
lot
on
the
debugging
front
with
you
know
the
the
application
developer
or
somebody
who's
really
close
to
the
infrastructure,
but
what
you're
describing
is
kind
of
the
other
view
of
this
from
from
the
top
down?
Where
are
these
policies
getting
applied
and
are
they
really
being
effective
everywhere?
I
need
them
to
be
for
compliance,
and
that's
something
you
want
to
see
via
metrics
are
similar.
A
E
A
These
are
really
great
I.
Please
I
encourage
everyone
I
that
that
last
one
is
like
an
area
that
I
completely
had
neglected,
I,
don't
think
anyone's
brought
anything
up
about
the
org
admin
looking
down
and
trying
to
understand
where
their
policy
was
being
applied
and
I
think
that's
really
important.
So
please,
if
you've
got
time.
If
someone
can
capture
that
in
this
discussion
as
an
idea
suggestion
whatever
would
be
great
yeah.
A
Great,
but
we
we
are
click
quickly
running
out
of
time,
and
we
got
a
really
important
one
here
at
the
end.
Candace
you
wanted
to
talk
about
TLS
from
Gateway
to
back
end
can
I
see.
Are
you
still
on
the
call
yep.
B
I
am
I,
didn't
write
this
tier,
so
it
might
have
been
someone
else
who
wrote.
B
A
A
B
Can
go
ahead?
Oh
so
generally,
we
you
and
I
and
Shane
had
a
discussion
about
the
the
Gap
that
Nick
had
updated
a
while
back
regarding
using
policy
attachment
to
to
work
out
the
details
for
backend
TLS,
as
in
the
in
the
minimize
limited
form
in
in
my
previous
Gap,
which
is
very
scaled
down.
B
Tls
back
in
TLS
discussion,
and
there
are
a
couple
there
are
a
couple
of
drawbacks
like,
but
does
this
freeze
it
in
place
now
the
idea
that
it's
I
mean
I
wanted
to
see
if
there's
any
implementers
who
are
using
policy
attachment
and
experiencing
issues
with
this
I
called
it
reverse
lookup,
but
you're,
calling
it
discoverability,
basically
so
you're
pointing
to
a
bunch
of
Services,
say
you
need
to
route
to
these
services.
But
how
do
you?
B
I
Yeah
yeah
I
think
that
one's
that
one's
going
to
end
up
looking
a
lot
like
a
reference,
Grant
I,
think
where
you
have
to
be
watching
all
the
reference
grants
and
figuring
out
when
you
get
a
you
know,
when
you
get
a
service
you're
going
to
need
to
get,
you
know,
you're
going
to
need
to
say:
oh,
you
know
am
I
doing
running
to
this.
I
Do
it
I'll
just
do
a
check
to
see
if
I've
got
a
TLS
backend
policy
that
covers
it,
so
you'll
need
to
keep
like
and
you'll
need
to
watch
them
all
and
just
see
if
the
service
that
you've
got
is
relevant
like
you
have
to
watch
both
service
and
TLS
back
in
policy
and
watch
them
against
each
other
as
part
of
the
reconciliation
process
inside
the
controller
that
doesn't
solve
the
problem.
For
you
know,
I
own
a
service
and
I
want
to
see.
I
If
my,
if
my
TLS
backend
policy
is
actually
in
use
by
a
Gateway
implementation,
that
is
the
discoverability
problem.
B
A
No
I
I
think
I
think
the
issues
are
are
present,
I'll
say
like
GK
Gateway
implementation.
We
have
multiple
policies
that
we
have
implemented,
supported
whatever
yeah
Works
approximately,
but
it
sure
would
be
nice
to
clean
up
the
ux
around
it.
A
So
that's
that's
why
I've
been
trying
to
push
for
the
and
I'm,
not
the
only
one,
but
trying
to
push
to
improve
what
what
I'm
calling
discoverability,
but
really
just
understanding
that
policy
has
affected
you
and
and
how
it
has
and
and
to
fill
in
the
other
gap
here,
I'm,
the
one
who
added
this
and
I
was
actually
I
actually
intended
it
to
be
part
of
the
discussion
around
policy.
But
it's
a
very
natural
next
step.
A
I,
think
that
where
we
may
end
up
with
this
TLS
from
Gateway
to
backend
is
a
built-in
policy.
Maybe
what
we
know
for
sure
is
that
we
need
to
extend
service.
We
need
to
say
here's
how
you
validate
TLS
connections
from
Gateway
to
this
back
end,
which
is
probably
a
service
or
service
import,
and
we
don't
have
a
great
way
to
do
that
right.
Now
that
isn't
policy
attachment,
we
could
try
and
fit
it
into
back-end
rep,
but
that
means
every
single
route.
A
Every
kind
of
reference
to
that
you
need
to
configure
it
separately,
that's
kind
of
not
great,
so
it
feels
like
kind
of
as
a
natural
follow-up
to
policy.
This
discussion
around
TLS
from
Gateway
to
backend,
May
kind
of
just
force
our
hand
of
really
making
us
ensure
that
what
whatever
we
have
here
works
well,
so
that's
that's
kind
of
why
I
added
this,
where
I
did
but
I
yeah.
A
A
Just
a
heads
up
that
right
now
the
the
trajectory
seems
to
be
towards
policy
attachment,
but
we'll
see
and
definitely
feedback
and
other
ideas
are
very
welcome
here
with
that
have
a
great
week
and
reminder
that
we're
meeting
at
an
earlier
time
on
Tuesday
of
next
week
and
we'll
see
as
many
of
you
as
possible,
then
have
a
great
one,
see
ya.