►
From YouTube: Service APIs Office Hours 20200415
Description
No description was provided for this meeting.
If this is YOUR meeting, an easy way to fix this is to add a description to your video, wherever mtngs.io found it (probably YouTube).
A
A
We
had
a
lot
of
activity
last
in
the
past
week
and
lots
of
different
opinions
on
things
so
I
think
there's
no
shortage
of
things
we
could
talk
about
today.
There
is
no
formal
agenda
for
office
hours,
so,
as
things
come
up,
I
think
it's
worth
talking
about
know
it's
like
always
having
some
trouble
getting
into
a
zoom
webclient.
A
Let's
keep
on
going
here,
I
think
Bowie
should
be
here
soon.
I
wanted
to
start
off
just
with
what
I
think
is
a
really
no
Lepus.
That's
like
what
we
needs
to
be
around
for
one
of
the
things.
I
think
it
would
be
good
to
get
some
consensus
on.
Is
this
this
idea
not
actually
this
PR,
but
the
idea
of
what
the
minimum
version
of
kubernetes
we
want
service
api's
to
support
I.
This
PR
would
mean
that
service
API
is
only
supported,
kubernetes,
1.16
or
newer.
A
B
Think
the
main
reason
to
to
move
to
us
to
do
that
is
not.
This
is
not
just
for
the
defaulting,
but
also
for
all
the
other
stuff
you
get
out
of
CID
v1r.
Most
importantly,
pruning
and
and
I
mean
explain
it's
nice,
but
open
API
support,
but
the
the
pruning
is
really
important,
because
if
you
what
we
found,
we
HTTP
proxies.
If
you
don't
have
pruning
enabled,
then
you
can,
you
can
say
miss
type
part
of
your
yomel
and
it
will
go
into
the
object
and
be
visible
when
you
retrieve
the
object.
B
But
it
won't
do
anything
because
it
doesn't
matter
it
doesn't
DC
realize
into
the
code
types
so
you'll
have
conflicts
sitting
there
that
looks
like
it
should
be
doing.
Something
and
I
won't
do
anything
at
all
because
it
gets
accepted,
and
so
until
you
have
pruning
and
ignore
online
fields,
then
you
don't
get
that
and
without
that
you
need
to
have
a
validation
web
book
or
something
like
that
to
actually
validate
your
that
you
that
all
the
feel
to
set.
C
A
E
G
B
I
mean
I
think
I
think
that
the
pruning
which
requires
it
doesn't
doesn't
require
CID
everyone,
but
it
makes
it
that's.
The
only
big
one
makes
it
mandatory
makes
it
makes
running
something
that
does
this
sort
of
controller
makes
the
user
experience
much
much
much
better
enough,
in
my
mind
that
it's
worth
saying
yes,
kubernetes,
1.16,
1/16,
plus
and
yeah
I
mean
I
do
feel
like
this
is
a
pretty
experimental
feature
already
and
requiring
a
newer
version
of
kubernetes
doesn't
seem
like
that
big,
a
stretch
when
you're
working
on
something
this
is
fundamental.
F
G
D
Do
you
mean
to
be
dependent
upon
like
what
features
would
be
available
because
I
think
what
you're
getting
at
is?
Should
we
make
it
117
118
is
what's
coming,
then
I'm
not
aware,
but
is
that
the
question
you're
asking
yeah.
B
Don't
think
so,
I
think
that
once
the
once
you
move
to
the
v1
I
see
ID
actual
resource,
then
then
I
mean
you
get
all
that
stuff
and
also
you
know,
then
you
have
a
ga
contract
as
well
so
yeah.
So
then
we
don't
need
to
worry
about
stuff
changing
too
much
and
we
should
just
be
able
to
use
the
use
the
CID
resource
without
having
to
worry
too
much
about
new
features.
Any
new
features
that
we
add
we
can
choose
to
opt
into
already.
G
A
A
B
I
think
I
think
it's
pretty
important
for
us
all
to
talk
a
little
bit
about
about
some
of
the
question.
Questions
accrues
Andrews,
raised.
I.
Think
that
yeah
brief
yeah,
so
is
that
there's
that
there's
that
one
but
I
also
think
the
the
larger
thing
that
are
the
the
more
actually
it's
actually
smaller
but
I'm.
The
proximal
thing
that
the
other
political
thing
that
into
a
depression
the
inter
opened
in
my
mind,
ease
this.
B
Don't
started
from
our
discussion
about
the
listener
ports
and
having
a
you
having
a
array
of
ports
of
slicing
ports
in
you
know,
under
the
listener
and
I
think
that
I
think
that,
in
my
mind,
a
lot
of
the
discussion
that
we've
had
over
the
last
week
has
has
helped
sort
of
solidify
a
bit
that
you
to
mean
I.
Think
that
the
the
way
that
yet
this
one,
that
my
current
feeling
is
that
the
gateways
should
have
one
or
more
listeners
and
each
other
and
each
listener
should
have
one
or
more
ports.
B
The
port,
through
the
objects
similar
to
port
objects
in
other
parts
of
kubernetes
and
like
I'd,
be
willing
to
argue
with
people
about
it.
But
my
current
feeling
is
that
the
that
the
protocol
field
in
the
in
the
port
object.
It
should
tell
you
something
about
how
the
controller
that
manages
the
Gateway
should
behave.
You
know.
So,
if
it's
TCP,
then
it
should
just
pass
through
TCP.
B
If
it's
you
uke
view,
and
then
with
the
idea
being
that
you
can,
you
can
use
just
TCP
to
build
up
a
full
TCP
stuck
below
the
Gateway
or
you
can
put
in
new
HTTP
or
HTTPS
to
make
assumptions
about
what
about
how
the
how
the
forwarding
and
the
or
the
terminated
here
less
termination
and
things
like
that
will
work.
The
corollaries
to
that
are
that,
if
you
do
that,
then
the
listener
needs
to
her.
Each
listener
needs
to
have
I.
Think
a
bounding
are
binding
directly
to
routes.
You
can't
have
the
I.
B
You
build
a
TCP
one.
First
that
has
a
certain
set
of
stuff,
then
there's
a
TLS
one
that
has
like
the
TCP,
1
plus
and
Taylor
stuff
and
there's
a
V
one
that
has
all
the
stuff
of
the
probe,
the
higher
the
lower
layers
and
then
the
being
able
to
drop
a
HTTP,
listen
or
a
HTTPS.
Listen
what
so
H
to
be
binding
to
route
108
a
listener
he's
in
school.
B
It
means
that
you
sort
of
making
a
whole
bunch
of
assumptions
without
that
I
literally
have
been
thinking
about
this
in
the
last
24
hours.
I,
haven't
written
anything
down
here
and
I'm,
sorry
about
that,
but
I
just
wanted
to
I
think
that
we
really
need
to
discuss
this
originally
now,
because,
like
we've
got
you
thank
you
to.
B
We've
now
we've
got
a
yeah.
We've
got
a
use
case
for
people
who
are
actually
building
real
TCP,
only
use
cases.
You
know
on
bare
metal
and
places
where
there's
not
arm
you,
where
it's
not
just
ingress,
and
the
thing
that
you're
talking
about
is
not
in
it
is
not
running
in
the
cluster
and
though
that
is
the
really
important
difference
and.
A
A
Just
thought
I
understand
it.
Real
quick,
you're
you're
proposing
here
to
have
every
listener,
contain
a
lot
more
information.
Each
listener
can
have
multiple
ports
and
each
listener
within
a
gateway
can
is
responsible
for
its
own
route
selection.
So
route
selection
moves
from
Gateway
down
to
gateway
listener.
Yes,.
G
B
That,
yes,
because
I,
think
that
the
the
the
idea,
the
thing
that
we've
been
talking
about
with
having
like
with
merging
gateways
and
things
like
that,
like
it's
I,
think
it's
really
confusing.
When
you're
talking
about
like,
because
because
there's
so
many
use
cases
that
we're
heating
it's
very
confusing
and
we
either
need
to
you
know.
We
either
need
to
sit
down
and
like
matrix
out
like
if
you're
going
to
use
a
gateway
in
emerging
style,
where
the
gateways
all
getting
merged
into
a
config
that
ends
up
on
a
log,
dance
or
somewhere.
C
This
I
mean
I,
guess
this
approach
and
sort
of
confused
a
bit
about
what
problems
they're
solving,
so
that
we
can
just
table
that
if
you
want
to
build
everything
on
some
strong
notion
of
the
proxying
tcp,
what
do
you?
What
do
you
want
to
do
about
supporting
proxies
that
don't
do
HTTP
proxies
that
don't
do
TCP
proxy
there's
a
bunch
of
things
in
that
class?
Are
we
going
to
say
that
that's,
not
supportive
infrastructure
for
that.
B
Where
was
that,
where
we
have
been
designing
this
to
be
e,
to
make
it
easy
for
a
cheaper
use
cases
just
fair,
because
the
majority
of
these
cases,
but
we
need
to
be
really
clear
by
designing-
needs
to
be
easy
to
HTTP.
We
don't
exclude
please
late
local
Oh,
TCP
use
cases
like
because
we
don't
that's.
That's
the
point
that
I'm
so.
D
Been
percolating
in
your
mind,
so
the
the
complete
flip
of
this,
which
is
actually
after
we
all
spoke
last
week,
I
went
back
and
said:
okay,
somewhat
James
said
I,
see
and
in
my
mental
model,
shifted
and
I
started
doing
the
Gateway
as
the
virtual
server,
and
so
our
entire
implementation
is
currently
using
that
if
I'm
trying
to
think
and
I
share,
I
can
share
this
I'm
pretty
sure.
Well,
this
is
yeah,
so
yeah
the
I
do
I
do
have
concrete
use
cases.
B
D
Sent
where
it
gets
interesting-
and
this
is
what
I
missed
last
week,
that
Nick
pointed
out
if
your
load
balancer
is
running
in
the
cluster,
the
cluster
is
the
load.
Balancer
control,
plane,
okay,
but
if
your
load
balancer
is
running
outside
of
the
cluster,
then
your
load,
balancer
control
plane
can
be
one
of
a
couple
of
things
for
Amazon,
it's
the
ELB
service
and
I
would
argue.
There
is
no
real
construct
of
a
load
balancer
in
Amazon.
What
they
call
load.
Balancer
is
really
in
the
grand
scheme
of
things,
a
virtual
server
for
NSX.
D
You
have
the
NSX
API
and
that's
your
service,
that's
the
control
plane,
but
you
still
have
a
load
balancer
construct
to
which
you
attach
virtual
servers.
The
control
plane
creates
load
balancers,
and
you
attach
virtual
servers
to
those.
The
load.
Balancer
construct
is
important
because
it
has
things
like
size
and
max
virtual
servers
and
when
you
have
a
multi-tenancy
use
case
when
you
might
want
to
have
multiple
load,
balancer
constructs
in
different
namespaces.
That
becomes
important
because
for
I
I
don't
want
to
say
it
boils
down
to
on
Prem
and
not
on
Prem.
D
G
D
I
get
it
and
so
see,
that's
the
first
case,
the
third
second
case
and
then
Oh.
Finally,
there's
last
case,
which
is
H
a
proxy
which
is
literally
what
I've
been
working
on
since
we've
all
spoke.
It's
control
plane
is
even
more
fun
because
H
a
proxy.
The
control
plane
is
what
it's
the
data
plane
API
server
and
that
exists
on
every
H,
a
proxy
instance
that
you're
running
and
so
where
NSX
has
a
single
control
plane
for
all
of
its
load.
D
This
is
gonna,
be
brought
up
on
this
call.
That
knowers
is
where
you
have
a
product
where
you
may
have
a
load,
balancer
construct
and
when
I
say,
load,
balancer
construct.
Let
me
how
did
I
define
it
in
my
document,
because
it's
literally
the
first
thing
in
my
document
that
I
defined
I
said
there
are
two
constructs,
a
construct
that
listens
on
an
IP
address
and
ports
and
forwards
traffic
to
back-end
servers
example,
an
AWS
elastic
load,
balancer
to
a
platform,
slash
control
plane
that
manages
virtual
servers.
D
D
So
in
that
definition,
when
you've
got
a
product,
that
or
project
that
says,
I'm
going
to
have
multiple
load,
balancers
spread
across
multiple
namespaces
for
reasons
of
access,
control
and
sizing
and
are
I
already
said,
access
control
and
I
need
to
attach
virtual
servers
to
those
it
becomes
really
cumbersome
for
those
to
live
at
the
cluster
scope,
which
is
why
I
filed
that
issue
this
week,
I'm
saying
oh
great
James,
said
gateways
of
virtual
hosts.
That
makes
sense.
The
one
caveat
is
gateway.
Class
needs
to
also
be
namespace
tour.
D
There
needs
to
be
a
namespace
version
of
it
and
what
Nick
just
said
was
well.
Maybe
if
we
push
everything
back
down
to
the
listener,
which
I'd
already
backed
away
from
that's
what
I
originally
thought
it
was
based
on
the
documentation,
I
backed
away
from
them
like
ok.
Well,
this
makes
sense
because
routes
are
on
the
gateway.
The
gateways,
the
virtual
server,
either
way
we're
kind
of
shifting
from
the
middle
left
or
right
and
what's
important
to
me,
is
that
the
that
load
balancer
construct
is
modeled,
and
today
it's
it's.
D
It
is,
if
you
look
at
it,
it's
the
gateway
class,
but
that's
cluster
scoped
and
therefore
doesn't
meet
the
use
case
of
access
control
and
having
the
multiple
ones
per
load.
Balancer
construct.
And
if
it's
the
listener,
then
I
would
actually
go
back
to
what
somebody
filed
earlier,
which
was
like
a
proper
virtual
host
ID
and
actually
make
these
actual
resources
either
way.
You
shift
it
though
I
think
so.
I
think
that's
what
it
boils
down
to
for
me.
D
What
makes
sense
would
be
to
have
gateway
class,
be
the
load,
balancer
and
namespaced,
or
introduce
a
new
type
called
virtual
server
or
virtual
host.
Have
it
live
on
the
gateways
and
replace
the
listeners
and
then
define
the
listeners
there,
but
but
the
use
case
is
really
multi-tenancy
where
you've
got
that
construct.
That
has
to
be
modeled.
Does
that
make
sense,
yeah.
G
D
D
User
can
choose
so.
The
proposal
is
to
support
a
managed
environment
where
it's
it's
getting
wired
up,
based
upon
intent
to
find
elsewhere,
but
they,
the
eventual
goal,
is
that
it's
more
of
a
framework
that
there
aren't.
We
aren't
defining
any
of
room
types
right.
This
is
simply
say:
hey.
There
is
the
service
API
there
and
as
I
described
it
to
Nick
I
start
I
was
handed
this
on
March
30th
and
said
I
had
to
deliver
something
by
April
30th.
Hence
my
it's.
D
My
like,
showing
up
with
such
a
sense
of
urgency
and
I,
was
looking
at
just
the
of
my
own
type
for
a
while
with
my
team
and
then
I
discovered,
y'all,
stipes
and
I
described
it
to
somebody
else
after
I
read,
read
everything
that
y'all
did
I
was
like
well
crap.
If
I
had
to
build
this
from
scratch,
it
would
look
exactly
what
like
exactly
what
you
have
with
a
few
tweaks
here
and
there,
and
so
it
just
it,
doesn't
make
any
sense
for
us
to
do
her
own
thing
and
so
but
the
event.
D
Well
great,
because
guess
what
there's
this
well-defined
pattern
that
says:
there's
a
gateway
class
for
your
load,
balancer
and
there's
ease,
yeah
gateway
class
and
a
gateway
for
all
of
your
virtual
servers
and
if
you've
got
some
mechanization
to
create
a
gateway
gateway
for
every
one
of
your
services.
Oh
wait,
that's
called
a
CPI
right,
that's
what
we're
gonna
build
into
the
vSphere
CPI
that
just
creates
these
gateways
for
every
service
that
is
requesting
a
load
balancer
then
more
virtual
server.
Rather,
that
will
be.
The
mechanization
point
would
be
the
vSphere
CPI
or
CCM
yeah.
G
Like
the
way
that
I've
been
looking
at
the
API
is
one
one
thought
that
we
put
some
I
guess
effort
into
is
sort
of
thinking
about
it.
So,
for
example,
if
you
have
in
this
current
API
some
reification
of
the
notion
of
like
a
physical
load
balancer,
that
may
is
that
the
does
the
user
particularly
care
about
that,
or
would
that
be
a
mechanism
that
is
sort
of
like
somewhat
inside
the
controller
control
plane
well
sitting
on
your
implementation?
So.
D
You
have
to
ask
who's
the
user,
but
ultimately
I,
don't
think
it's
the
user.
So
much
is
the
product
and
it
can't
be.
It
can't
be
part
of
the
controller,
because
I
mean
I
guess
it
could
but
I
mean
I.
Think
James
said
in
one
of
his
comments
like
there
could
be
this
invisible
proxy
James
I
apologize.
Did
you
use
the
word
and
didn't
use
the
word
invisible,
but
I
think
the
implication
would
be
like
it
would
just
be
some
construct
that
will
be
managed
by
the
controller
without
being
modeled.
D
D
My
recommendation
to
them
is
admission
web
hooks
and
we'll
have
you
know
something
with
the
names
and
the
gateway
classes
to
align
them
with
an
inch
spaces,
but
like
their
kubernetes,
provides
part
of
this
with
the
way
they
do
are
back
in
with
namespace
and
the
other
other
cases.
So
you've
got
namespaces
that
represent
part
of
your
organizational
hierarchy,
Inge
one-inch
to
production.
Whatever
right
you
can-
and
this
is
one
of
the
things
I
just
love
a
little.
What
y'all
do
it's
all
top-down?
D
It's
all
finding
who
belongs
to
me,
and
so
you
can
move
a
gateway
from
one
namespace
to
the
other
and
if,
for
example,
the
load
balancer
objects,
actually
we're
the
same
name
or
the
gateway
classes
were
the
same
name,
but
they
just
referenced
actual
back-end
load.
Balancer
constructs
all
of
a
sudden,
hey
your
gateway
just
gets
reconciled
in
that
new
namespace
and
nothing
else
changed
and
I.
Think
that's
amazing,
that's
great,
and
you
get
that
today,
with
gateway
classes
at
the
cluster
level,
but
again
it
now.
D
G
It
would
be
good
to
be
good
to
figure
out
because
I
feel
like
I'm
hearing
sort
of
two
user
populations
like
there's
a
user
population
that
you're
describing
which
is
sort
of
your
control
plane,
is
going
to
be
reacting
and
acting
on
certain
objects
like.
What
do
you
expect
your
user
to
be
for
do
in
terms
of
giving
like
what,
which
are
the
resources
that
are
going
to
be
acted
upon
from.
D
Sewers
perspective
sure
so
up
so
right
now
we're
working
in
two
cases.
Our
MVP
is
for
something
internal
and
then
our
the
goal
is
to
move
this
in
the
the
what
you
just
described
that
aspect
of
it,
both
the
pattern
and
the
implementation
into
the
VCR
CCM.
So
from
a
user's
perspective,
they
would
create
a
service
just
like
any
other
service
type
load.
D
They
all
use
their
own
types
and
they
all
forego
the
CPI,
and
this
is
saying
well,
the
CPI
will
just
say:
okay,
the
user
says
that
this
is
service
type
load
balancer.
Here
they
specified,
maybe
the
name
of
the
gateway
class
or
the
name
of
the
load
balancer
to
which
this
service
wants
to
be
attached.
They
wouldn't
call
it
a
gateway
class
that
we
just
say
in
the
name
of
the
load,
balancer
and
if
one
didn't
exist,
if
they
did.
G
D
D
Be
but
it
would
be
represented
by
the
Gateway
class
and
the
idea
would
be,
and
from
the
perspective
of
the
vSphere
CCM
when
it
starts
it,
either
specifies
a
default
gateway
class
to
attach
all
services
or
gateways
rather
or
that
there
is
some
annotation.
That
is
standard.
That
says,
a
user
can
also
specify
that
if
they
happen
to
know
the
name
of
the
Gateway
class
that
they
want.
So
that
way
you
can.
Yes,
you
go
ahead.
Yeah.
G
D
I
think
Rob
brought
that
up
where
I
mentioned.
That's
so
I
wasn't
trying
to
take
away
the
notion
of
a
clustered
gateway
class
I,
actually,
like
it
I
think
for
I'll,
be.
It
makes
a
lot
of
sense
when
you've
got
this
single
control
plane
and
they
have
no
notion
of
a
load
balancer
again,
it's
a
virtual
server,
despite
what
to
call
it
and
so
you've
got
Internet
and
internal,
and
you
just
have
a
config
and
it's
points
to
the
the
control
plane.
D
The
EOB
service,
so
I'm
saying
that
that
could
still
be
a
cluster
scope
but
I
think
in
a
comment.
Rob
said:
well:
if
we
moved
it
to
namespace,
then
then
the
name
of
the
class,
then
if
there
wasn't
a
namespace,
maybe
it
defaults
to
a
namespace
I,
don't
know,
like
you
know,
service
API
system
so
like
by
default.
Everything
goes
there.
If
you
very
much
by
the
name,
some
yeah
I
wasn't
trying
to
move
the
cluster,
but
that
would
be
the
way
you
pray
like.
G
Sort
of
how
the
model
works,
because
the
class
model
has
come
where
both
ingress
and
storage
ingress
is
newer,
so
maybe
it's
a
little
bit
less
proven,
but
at
least
storage
classes
how
they
handle
some
of
these
things,
because
there
are
certain,
for
example,
I
think
there
are
some
storage
providers
that
sort
of
have
these
concrete
physical.
Like
let's
say
you
have
a
file
server
with
like
yeah.
D
D
Know
exactly
where
you're
going
and
I
don't
know
off
the
top
of
my
head,
because
I
was
about
to
tell
you
know
the
storage
control
plane
is
no
crap;
no,
they
might
have
individual
filers
and
I
I
would
have
to
go
look
to
see
how
they
handle
it,
but
yeah
they
would
have
been
dealing
with
the
same
thing
and
unfortunately,
from
the
VMware
perspective
because
of
the
sand
being
around
for
so
long
I
guarantee
you
that's
our
class
or
rather
CSIS
now.
So
it's
not
I
can't
look
at
her
own
stuff,
yeah.
G
D
G
D
Fair
right,
so
I
I
can
make
some
inroads
there.
I
I
was
one
of
the
founders
of
CSI
container
storage
interface.
I
know
a
lot
of
people
that
work
that
still
work
on
it
and
I'm
sure
they
were
gonna.
Have
some
answers,
I
just
didn't
when
I
created
it
I
wasn't
coming
at
it
from
a
kubernetes
perspective,
so
I
didn't
care
about
the
kubernetes
of
it
all
back
then,
but
assad
who
still
works
on
it.
I
can
check
with
him.
Although
he's
at
Google
with
you
so
yeah.
D
B
Think
I'm,
so
in
my
mind,
the
thing
that
I
was
trying
to
get
to
by
raising
this
today,
I'm
sorry
just
bring
it
on
your
injury.
Was
that
I
was
that
I
think
my
feeling
is
that
the
questions
you've
raised
raised
some
pretty
fundamental
questions
about
the
way
that
we're
structuring
these
things
and
I
don't
think
that
spending
more
time
tinkering
with
the
the
lower-left
down
stuff.
Like
you
know,
traffic
split
and
stuff
like
that
is
actually
going
to
be
useful
until
we've
like,
because
we
might
need
to
throw
a
lot
of
that
away.
C
Us
would
have
to
sense
regards
to
what
I
think
is
Andrews
perspective,
so
gateway
and
gateway
class.
The
whole
services
API
is
is
abstract
right,
it's
not
in
a
the
way.
It's
designed
I,
don't
think
it's
an
API
for
provisioning
or
configuring
infrastructure.
It's
completely
agnostic
as
to
the
infrastructure
underneath
it
and
it's
just
so
just
like
ingress.
It
doesn't
say
anything
about.
You
know
how
it
should
be
implemented.
You
can
implement,
however,.
C
G
A
C
G
G
Yeah
so
luckily
I
ordered
a
laptop,
a
personal
laptop,
so
maybe
this
will
be
improved.
The
yeah
I
think
I
do
agree.
A
hundred
percent
with
James
is
like
the
notion
of
this
abstract
description.
I
think
that's
kind
of
in
our
target
in
terms
of
the
users.
So
that's
why
there
are
some
tensions
in
terms
of
to
actually
make
a
course
on
one
to
one
with
his
ago
infrastructure.
G
A
Also
interested
in
the
storage
class
background
as
well,
and
why
it
was
chosen
to
be
cluster
scope
when
I.
You
know,
when
I
looked
through
this
and
talked
with
Andrew
some
more
about
this,
it
was
fascinating
to
me
that
it
seems
like
we
could
solve
a
lot
of
this,
at
least
for
this
use
case
by
doing
something
like
we'd
already
discussed,
doing
separately,
which
was
the
namespace
cope,
gateway,
class
I.
It
which
comes
with
other
advantages
that
are
positive,
regardlessly
right.
It
allows
better
modeling
with
our
back.
A
It
allows
easier
ways
to
feel
restrict
who
can
create
gateway
classes
where
and
whether
they
can
be
used
and
it
it
helps
in
some
some
other
ways.
I'm
not
saying
that
has
to
be
the
only
solution
and
I
think
there
are
other
things
we
can
clean
up
in
gateway
like
tying
ports
and
protocols,
together
as
an
example
makes
a
lot
of
sense
to
me.
I
I
think:
there's
there's
a
way
to
approach
this.
That
does
not
throw
at
everything
what
does
still
solve
this
issue
and
I
am
interested
in.
A
G
D
Soon
enough,
for
me,
I
mean
I'll
I'll
ping,
some
people
because,
like
I
said,
oh
I'm
getting
so
many
people
are
gonna
hate
me
I,
said:
okay,
we're
moving
to
this
paradigm,
we're
shifting
this
paradigm
to
gateway
class,
being
a
load
balancer
from
the
Gateway
being
the
load,
balancer
and
James.
Your
comments
are
very
fair,
III
understand
but
like
we
also
have
to
be
realistic.
We
live
in
a
world
where
infrastructure
is
important
and
you
all
come
from
a
world
where
you
don't
have
to
worry
about
infrastructure
in
the
clouds.
That's.
C
C
D
B
D
B
Where
does
TLS
live,
and
things
like
that,
like
I,
think
that
the
reason
that
that
has
been
tweaking
for
us
at
these
points
to
the
fact
that
that
that
the
model
is
not
because
if
the
model
is
bleep
in
that
the
place
to
pick
up
stuff,
we're
like
yeah
and
so
the
that
I
feel
like
a
ski
coming
in
to
ask
these
questions
is
really
useful
to
us
as
a
sub-project.
Because,
because
answering
these
questions
one
way
or
another
will
help
provide
clarity
on
what
we
are
actually
trying
to
do
is
enjoy.
So.
G
The
one
thing
it's
kind
of
the
response
to
unions
proposal
last
week
is
like
in
terms
of
API
layering.
You
know
what
could
it
be
possible
to
kind
of
get
what
we
want
for
Andrew
to
get
what
he
wants
out
of
the
API,
with
the
addition
of
say,
like
one
resource
to
model
the
physical
infrastructure,
that
sort
of
independent,
like
that,
in
my
mind,
is,
is
how
how
much
more
is
it
going
to
be
a?
D
So
I
can
answer
that
question
Rob
and
I
forget
it
was
Robin
our
Nick
and
I
discuss
this.
The
gateway
class
today
has
parameters
Ref
right,
but
what
it
doesn't
really
have
is
a
list
of
them.
So,
regardless
of
what
I
do
here,
I
have
to
do
something
really
kind
of
hacky
and
use
config
maps
and
secrets,
or
or
maybe
like
a
join
table
and
a
CRD,
but
whatever
that
is,
if
I
have
to
introduce
a
new
custom
type.
That
is
a
dependency
that
we
now
own.
D
G
D
Brent,
that's
brand
new,
like
the
the.
If
you
have
to
model
that
that's
that's.
Why
I
haven't
been
concerned
with?
How
do
I
model
this
and
visibly
or
with
a
custom
type
I've
been
concerned
with
introducing
the
concept
that
this
this
type
is
important
to
be
modeled
and
mapping
it
onto
something
with
the
little
it's
the
smallest
amount
of
change.
G
E
B
E
Get
back
to
this
I
just
want
to
hit
the
pause
button
on
this
topic
for
a
second
is
tomorrow
it
Rob
on
the
agenda.
There
is
a
topic
I
put
there
and
it's
about
SMI
the
service
mesh
interface.
We're
gonna
have
one
of
the
owners
of
that
project
join
us
tomorrow.
I
would
like
to
make
sure
that
that
topic
stays
number
one
so
that
the
this
person
from
SMI
that
joins
us
when
we're
done
that
this
person
can
leave.
E
But
the
the
point
that
I
want
to
make
is
just
that:
I
think
we,
as
you
know
as
a
community,
need
to
come
together
and
and
make
sure
that
we're
prepared
for
having
this
conversation
tomorrow.
So
if
everyone
wants
to
independently
look
into
SMI
and
the
API
specs
there,
but
I
think
it
would
be
a
good
idea
for
us
to
do
some
of
that
homework
and
then
figure
out
as
a
community.
You
know:
where
do
we
go
from
here?
E
E
It
was
just
an
open
discussion
brainstorm.
You
know
we
come
to
the
table
with
our
expertise
of
service
api's
and
I
forget
the
the
person's
name.
That's
joining
us.
It
should
be
on
the
agenda
is
going
to
join
us
and
potentially
others
and
they're
gonna.
Just
you
know:
they're
gonna
bring
their
brain
power
to
the
table
and
let's
just
discuss
it
and
brainstorm
it,
but
it
would
be
nice
if
folks
here
could
just
spend
some
time
looking
at
SMI
and
the
api
specs
and
just
try
to
mentally
prepare
themselves
for
the
discussion
tomorrow.
E
D
E
A
I,
just
I
just
will
not
follow
up
on
that
I
thanks
for
organizing
that
and
getting
that
on
the
agenda
and
that
for
anyone
looking
this
concept
came
up
in
poll
57
and
Damian's,
linked
to
the
actual
SMI
spec
or
traffic
split
here
and
I.
Think
a
number
of
us
have
looked
over
it
already.
I
did
I
agree.
It
does
seem
very
close
to
what
we
did
originally
proposed,
so
yeah
yeah,
yeah.
E
And
and
that
link
is
specifically
to
the
traffic
split
speck,
but
if
you
actually
go
back,
there's
a
HTTP
route
group.
There
I
mean
they
have
a
few
api's
that
that
to
me
seem
very
similar
to
what
we're
wanting
to
do
so
scroll
down
a
little
bit
here
and
yeah
you'll
see
the
traffic
specs
traffic
specs
go
to
traffic
specs.
E
Not
you
know,
James
brought
this
to
my
attention
in
the
traffic,
splits,
PR
and
so
I
took
some
time
now
did
a
quick
review
and
then
the
alarm
started
sounding
in
my
head
and
then
I
looked
at
the
issues
and
they
actually
have
an
issue
or
they
call
out
service
API.
So
that's
kind
of
where
I
synced
up
with
their
community
and
said
hey.
You
know,
join
us
on
one
of
our
one
of
our
calls
and
let's
kind
of
hash
through
this.
E
E
And
I
believe
it's
at
grandpa
burg
that
offered
to
join
our
call
tomorrow.
So
my
point
just
being
is:
if
everyone
could
just
you
know,
take
some
time
do
the
homework
so
that
we
gonna
make
use
out
of
grandpa
burg
and
anyone
else.
That's
gonna
be
joining
us
and
have
a
thorough
discussion
on
the
topic
and
I'll
be
doing
the
same
yeah.
It
sounds,
but
you
isn't
boo.
Okay
and.
B
E
E
E
D
D
Just
wanted
to
take
a
note
if
it
shouldn't
be
there,
I
can
move
it.
Just
I
noted
that
enough
for
on
this
storage
play
page
all
of
the
NFS
or
filer,
where
you
might
have
individual
pieces
of
infrastructure,
there
were
no
entry
implementations
which
I
found
interesting
specifically
for
those
and
then
looking
for
one
for
a
persistent
volume
claim
it
does
map
directly
to
an
in
effect,
NFS
server
and
export.
So
if
you
look
at
it
like
that,
it
is
kind
of
infrastructure,
it's
pointing
directly
to
the
what
would
be
at
load
balancer.
D
D
C
D
Si
si
si
si
is
another
layer
of
abstraction
yeah,
but
I
was
looking
specifically
at
NFS.
Yes,
it's
unfair
to
just
talk
about
si
si,
because
that
would
be
a
separate
layer
of
abstraction
and
your
your
example.
James
would
be
somebody
building.
Another
layer
of
abstraction
on
top
of
service
API
is
to
manage
the
infrastructure
so
yeah.
That
would
be
where
you
would
have
I,
don't
know
what
it
would
be
called
lb,
I
load,
balancer
interface,
and
it
might
be
based
on
resources.
B
Yeah
so
I
think,
like
I,
said:
I
think
it's
just
I've
wanted
to
rise
it
today,
because
I
think
it's
important
that
you
you
got
time
time
frames
like
Andrew
and
I
know:
you're
gonna
be
something
basically
if
we
can't
sort
something
out
in
a
reasonable
time,
and
so
yeah
I
want
to
that's.
Why
I
wanted
to
raise
it
today,
so
that
we
could
at
least
start
talking
about
it.