►
From YouTube: Kubernetes SIG Service Catalog 2018-04-23
Description
Discussing:
- move to prow
- compatibility with kubernetes versions
- 1.0 milestone review
- CRD proposal
- demo minibroker
- demo health checker service
- use case brainstorm
- WE NEED A LOGO!
A
B
B
If
there's
something
you
want
to
talk
about
it,
add
it
there
and
slap
your
name
on
it
and
we'll
just
walk
on
through.
If
you
have
a
question,
anyone
comment
or
say
something:
sometimes
it's
hard
to
interrupt
people
so
just
type
plus
hand
and
I'll
help
move
things
along
and
call
them
people
yeah
cool.
So,
let's
start
with
Jeremy
talking
about
what
happened.
Last
Friday
all.
C
Right
good
times
so
like
right
now
we
have
this
situation
where
the
the
helm
chart
that
we
use
to
install
Service
Catalog,
and
then
we
tell
people
to
use
to
install
Service
Catalog
seems
to
be
updated
on
every
commit,
not
when
we
do
a
release.
So
there
is
this
big
PR
that
went
in
rebased
all
of
the
aggregated
API
service,
stuff
off
of
110,
which
introduced
some
breaking
changes,
but
then
that
got
merged.
C
We
didn't
do
a
release,
so
the
version
of
Service
Catalog
in
the
chart
was
still
like:
v01
14,
which
didn't
have
the
new
flags
and
like
trying
to
track
this
down.
It's
not
apparent
to
a
lot
of
us
where
that
hellish
for
how
people
sync
is
happening,
and
it
seems
like
in
general,
we
have
an
opportunity
to
revise
and
revisit
what
we're
doing
for
CI
to
make
it
more
transparent
so
that
everybody
can
understand
it.
You
maybe
have
everything
baselined
in
the
repository
and
that
drives
whatever
builds
Travis
or
Jenkins.
C
You
figure
out
why
we're
using
Travis
and
Jenkins
instead
of
just
one,
maybe
look
at
moving
to
prowl
now,
I
think
and
then
I
think,
like,
coincidentally,
with
that
figuring
out
what
the
release
process
should
look
like
what
we
document
in
terms
of
what's
the
minimum
version
of
kubernetes
that
we
support.
For
instance,
I
can't
install
service,
catalog
and
I
can
install
service
catalog
on
like
a
1/8
cluster,
but
I
can't
install
any
brokers
on
it.
C
So
I
I
did
a
brand
new
one:
810
mini
cube
cluster
small
service
catalog
and
then
I'm
not
able
to
install
OS
Bo
or
the
UPS
broker.
I
get
this
cluster
service
brokers
that
service
catalog
that
case
that
IO,
then,
whatever
the
resource
name,
is
forbidden
request
and
that
happens
with
binding
anything
you
try
to
create.
The
things
are
registered
there,
but
they're,
not
they're,
not
available
I
can
do
a
coop
to
D
I'll,
get
on
the
resource.
Name
and
it'll.
D
C
Seems
like
there
might
be
some
potentially
breaking
things
between
14
1,
8
and
anything
past
1
8,
so
I
just
think
like
in
general.
There's
an
opportunity
or
a
need
for
us
to
kind
of
figure
out
what
the
CI
stuff
is
going
to
look
like
what
releases
are
gonna.
Look
like
him,
roll
that
out
and
then
move
to
a
new
thing.
B
So
I
was
just
suggesting
that
maybe
we
start
a
working
group
that
for
like
the
next
couple
weeks,
just
for
helping
people
to
organize
around
bringing
everything
back
into
the
repo
related
to
our
CI
and
if
there's
missing
doc
on
how
stuff
works,
we'll
throw
that
in
or
what's
required.
It
wasn't
clear
yeah.
A
So
I
would
say
that,
like
this
alien,
probably
the
most
salient
part
of
like
what
we've
just
talked
through
is
specificity
around
which
versions
of
kubernetes
are
supported
and
I.
Think
that
depends
on
how
far
back
we
want
to
support
I
if
I
remember
correctly,
the
there
is
something
about
the
aggregator
that
changed
that
we
require
at
least
one
nine
I'm
personally,
not
certain
that
there
is
much
value
to
be
had
in
supporting
more
than
one
older
release
like
current
release
and
current
minus
one.
But
it
depends
on
what
the
group
wants
to
do.
C
C
C
To
the
alum
chart
tomorrow
and
it
gets
merged
in
it's
gonna,
we're
gonna
have
the
same
problem
where
it
could
be
a
breaking
change,
use
of
either
the
released
image.
So
I
think
that's
what
we
need
to
fix.
First,
it
seems
like
that's
some
sort
of
magical
internal
Google
thing
on
the
jenkins
instance,
The
Vic
may
be
the
only
one.
That
knows
how
to
do
so.
He's
willing
to
talk
to
us
tomorrow
figure
out
what
we
need
to
do,
or
what
the
right
way
to
fix
that,
like
immediate
problem
is,
but
still
it's
like.
C
C
C
I
think
that's
the
best
idea.
I
mean
anybody
I'll
forward
it
to
the
whole
whole
list,
and
then
anybody
that's
interested
can
join
and
then
we'll
kind
of
spend
the
working
group
off
of
that
I
think
Scott
created
a
milestone
and
then
added
a
couple
equated
to
issues.
I
made
us
make
CI
a
better
issue
he
made,
so
we
can
kind
of
track
things
and
build
out
of
that.
Yeah.
B
Cool
sounds
like
everyone's
in
agreement:
I'm
gonna
move
on
to
the
next
one,
which
is
the
road
to
GA,
so
I
volunteered
last
week
to
go
through
an
update,
everything
from
the
face
to
face
and
kind
of
put
that
on
our
100,
milestone
and
github,
and
then
I
kind
of
realized
that
maybe
that
wasn't
the
right
thing
to
do
so.
I
want
to
talk
to
you
about
it.
First,
if
I
could
can
I
share
my
screen
real,
quick,
so
I
can
show
you
what
I've
done
so
far.
B
So
I
updated
our
wiki
page,
which,
which
had
a
road
map
and
I
went
through
and
marked
off
with
a
little
key
here
for
like
what's
been
implemented.
What
I'm
not
quite
sure
for
ready
to
call
it
done
or
not,
and
then
things
that
I
don't
think
anymore
seem
to
be
on
our
100
milestone,
but
was
originally
over
a
year
ago.
B
So
I
went
through
it
and
updated
all
of
this,
and
if
we
agree
on
that,
then
I
can
go
through
and
start
updating
the
milestones
and
and
get
how
but
I
want
start
thrashing
stuff,
because
it
wasn't
quite
clear
for
for
everything
to
find
some
of
the
questions
like
so.
For
example,
I
think
the
only
one
I
was
still
confused
on
is
a
our
back
for
individual
service
visibility.
B
A
So
II
I
think
this
was
originally
around
sharing
service
instances.
Oh
I'm,
sorry!
No!
This
is
for
service
classes.
I
think
this
is
this
pertains
to
the
idea
that
once
upon
a
time
there
is
a
discussion
around
trying
to
do
a
cold
filtered
on
the
cluster
scoped
resources
which,
as
we
all
know,
is
impossible.
G
B
A
So
our
back
gives
you
the
next
to
a
in
a
sense
of
which
service
accounts
can
provision
and
which
can
bind
and
but
not
to
a
specific
instance.
That's
that's
really
not
what
was
trying
to
do
I'd
be
happy
to
continue
or
restart
thinking
about
that,
but
I.
Don't
think
that
we
should
gate
oda
or
one
dot
o
on
on
that
those.
B
To
be
clear,
I
don't
have
I,
don't
care
if
any
of
this
goes
into.
100
just
was
lighting
stuff
that
that,
at
one
point
we
said
we
wanted
and
before
I
rip.
All
this
often
and
kind
of
change
the
planet
record
I
want
to
make
sure
that
people
agree
the
other
one
was
we
had
all
code
coverage
and
doc
standards
met
and
has
decided
in
some
magic.
B
A
What
I
would
say
is
that
I
think
we
should
consider
releasing
ODOT
two
after
namespace
broker's
are
in
because
what
I
have
heard
from
customers
and
folks
in
the
community
is
that
they
think
the
concepts
really
neat
for
different
reasons.
I
think
a
lot
of
folks
are
hesitant
to
use
Service
Catalog
until
we
have
namespace
scope
broker's,
and
it
feels
weird
to
go
from
o
dot
1
to
1
dot.
Oh,
that's,
a
typical
I
think
I
think
a
good
milestone
for
ODOT
2
would
be
namespace
scope.
Brokers
are
in
what
do
folks
think
about
that.
B
B
Hands
because
I
was
talking
is
sick
service,
catalog,
Morgan.
B
C
A
I
A
So
those
those
are
orthogonal
in
terms
of
the
namespace
scope,
Brooker
stuff,
the
API
surface-
is
mostly
in
I
believe
that
Eric
Nelson
had
some
internal
things
that
he
had
to
work
on
last
week.
But
he
told
me
that
he
was
resuming
work
on
the
control
loops
for
the
names
base
scope
stuff
today.
So
I
would
imagine
that
in
the
next
two
impossibly
three
weeks,
we'd
be
looking
at
those
being
complete.
A
Into
to
answer
the
second
part
of
your
question
Michael,
we
do
not
need
to
bump
the
API
version,
2
v
1
beta
2.
We
would
really
only
need
to
do
that
if
we
made
incompatible
changes
to
the
V
1
beta
1
API,
since
we
haven't
done
that
and
I.
Thank
you
all
for
not
having
done
that.
We
do
not
need
to
bump
the
API
version.
B
I'll
update
the
notes
with
what
we
just
decided
to
push
out
and
shoot
for
for
two
and
one:
oh
and
then
I
will
copy
everything
over
from
the
one
Oh
roadmap
wiki
onto
this
milestone
and
update
existing
issues.
There's
a
couple
issues
that
need
to
be
created
as
well,
so
I
will
make
them
if
they
still
don't
exist
yet
for
stuff,
like
generic
actions
and
direction.
Arrow
Corrections,
yeah.
H
B
E
Howdy
howdy,
so
I
I
posted
a
CL
with
a
proposal
to
get
some
feedback
and
I,
don't
know
if
anyone
read
it
or
gave
feedback.
Yet
I
looked
at
the
original
I
think
model
which
produced
a
CR
D
as
a
backing
store,
CL
or
PR,
and
it
looked
like
it.
It
was
pretty
complicated,
so
mine
is
gonna,
be
a
little
simpler.
E
A
Thing
that
I
think
is
is
relevant,
which
I'm
not
certain.
If
you're
aware
of
is
that
CR
DS
can
be
cluster
scope
now
they
do
not
have
to
be
namespace
scoped.
So
for
the
cluster
scope
resources,
it's
probably
a
better
fit
to
put
them
into
cluster
scope,
CR
DS,
and
then
you
don't
have
to
answer
this
question
of
which
namespace
do
they
go
into.
E
E
B
Okay,
kibbles
your
hand
went
down.
Is
there
anything
else
on
that
before
we
move
on
to
a
demo
of
mini
broker?
B
B
The
idea
for
mini
broker
is
it's
a
mini
broker
for
your
mini
cube
and
it
lets
you
provision
stable
home
charts
from
upstream
using
Service
Catalog,
and
it's
just
a
it's
a
little
broker.
That
kind
of
lets
you
do
that
and
I
finally
got
it
to
the
point
where
it
actually
kind
of
works.
We
just
we
just
bought.
It
works
for
installing
WordPress,
my
sequal
Maria
DB
and
Postgres.
B
Oh
dear
I
can't
watch
chats
at
the
same
time,
people
put
their
hands
up,
someone's
gonna
have
to
help
doing
stuff,
and
so
what
it
does
is
it
actually
has
a
sidecar
of
tiller,
so
it
has
its
own
private
tiller.
No
one
can
access
it
or
use
it
or
anything
like
that,
and
then
it
talks
directly
to
tiller
and
provisions
stuff
for
you.
So
somebody
get
when
we
make
us
like
way.
Bigger.
B
B
B
7-14
and
it'll
make
it
super
fast
because
there's
no
real
cloud
in
a
container
next
to
you
on
your
cluster.
So
if
we
take
a
look
at
what
that
made,
there,
I'm
also
gonna
show
a
new
feature
in
SVG,
yet
that
I'm
trying
to
get
merged.
So
if
you
want
this
to
see
this
in
the
next
release,
give
me
another
looks
good.
B
B
B
B
There
we
go
yeah,
it
doesn't
show
the
secret
data
until
the
secret
isn't
injected.
So
sometimes
it's
a
little
slow
I
have
an
open
PR
to
have
like
a
magical
flag
with
lots
of
warnings
on
it
that,
instead
of
telling
you
ten
bytes,
will
actually
decode
it
and
just
show
it
to
you.
So
what
this
ends
up
looking
like
is
get
it
secret
demo.
B
B
So
what's
cool
now
is
that
I
made
a
I
modified
the
upstream
help
chart
for
WordPress,
instead
of,
like
assume
certain
things,
I
added
like
a
couple
values
so
that
I
could
put
in
like
inject
parameters
like
the
name
of
the
database
and
the
user,
because
there's
there's
a
little
bit
of
assumptions
going
on,
and
these
are
like
small
changes
at
the
upstreams.
Probably
would
be
okay
with
I.
Don't
know
we'll
see,
but
it
just
lets
me
then
pass
in
these
parameters
wholesale.
B
So
then,
when
it
can
do
something
like
I'm,
not
gonna,
run
it
because
it's
a
little
slow,
but
basically
you
can
run
helm,
install
I
call
it
mini
press
and
it
references
this
local
WordPress.
That's
been
plumbed
to
work
a
little
bit
better
with
mini
broker
so
like
it
doesn't
assume
which
broker
you're
hitting
stuff.
Like
that
I
see,
you
can
switch
it
out
to
like
something
else
like
adder
and
then
what
that
ends
up
looking
like
is.
B
B
It
filled
everything
out
that
it
needed
and
if
I
go
to
WordPress,
no,
no
nothin
in
it
I
actually
have
a
live.
Wordpress
that
hit
that
and
was
all
provisioned,
which
is
pretty
cool,
so
I'm
just
finishing
up
getting
it
working
for
the
charts
and
then
I
have
people
from
like
cig
apps
who
are
kind
of
looking
at
this
Matt
Farina
is
really
interested
in
in
trying
this
as
well,
and
he
suggested
this
may
be
something
that
instead
of
having
to
be
an
OS
B
kit,
then
it
may
be
a
kubernetes
SIG's
thing.
B
But
just
because
it
seems
like
if
we
want
this
to
be
something
that's
owned
by
the
community
and
not
me,
it
may
find
a
better
home
there
than
and
they
always
wanted.
I
don't
know.
A
B
Six
months
from
now
I'd,
like
everyone's
thing,
because
some
of
my
plans
for
mini
broker
is
getting
it
to
the
point
where
it's
it's
the
way
you
can
try
out
Service
Catalog
without
paying
anybody
money,
because
right
now,
you've
got
to
either
use
UPS,
which
is
too
simple
to
really
help.
You
understand,
Service,
Catalog
and
all
the
other
ones
require
you
to
have
a
real
clock
right
or
somewhere
and
you're
paying
the
money.
So
this
would
be
a
nice
way
for
people
to
try.
B
B
A
B
A
J
F
So,
let's
see
this
full
request:
1904
there
was
a
discussion
at
the
sink
meeting
two
weeks
ago,
the
face
to
face.
Basically
what
we
did
was
we
wanted
to
create
another
tool
to
help
monitor
catalog
in
a
deployment.
You
know
a
lot
of
people
using
Prometheus
to
pull
metrics
monitor,
generate
alerts,
that
kind
of
thing
of
kubernetes
components
and
Service.
Catalog
is
one
of
those.
F
We
have
some
metrics
in
service
catalog,
but
we
thought
that
by
creating
an
additional
standalone
component,
which
exercises
a
Service
Catalog
and
exercises
a
you
know,
basically
a
test
or
no
op
Broker
we'd
get
some
end-to-end
flows
going
and
throws
so
the
health
check
would
do
that
and
it
would
also
explore
some
metrics,
which
you
could,
in
turn,
pull
from
Prometheus
or
scrape
from
Prometheus
real,
basically
on
Prometheus.
The
way
that
works
is
if
your
application
has
some
metrics,
you
can
expose
them
through
the
Prometheus
client
API
or
you
can
do
it.
D
F
It
is
like
here
this
is
our
I'm.
Sorry,
it's
the
trying
to
get
yeah.
That's
those
are
the
few
metrics
that
we're
exposing
from
Service
Catalog
they're.
You
know
very
rudimentary.
You
can
see
just
a
couple
brokers
here.
They
were
exposing
says
that
we've
got
one
class
in
the
OSB
stub
11
and
the
template
service
broker
and
goes
on,
shows
plan
count,
and
then
we
list
the
OSB
requests
that
we've
made
and
the
responses
that
we've
had
back
on
those.
So.
F
To
be
clear,
these
are
metrics
from
the
catalog
itself.
Right,
that's
correct!
That's
sorry!
What
I'm
trying
to
do
here
is
to
show
you
what
the
extension
point
looks
like
for
Prometheus
metrics.
So
this
is
just
hitting
the
end
point
for
the
Service
Catalog
controller.
So
likewise,
with
the
health
check
metrics,
you
get
something
very
similar
again,
you
get
a
metric
name,
followed
by
you
know
count
or
whatever.
F
If
we
pop
over
to
this
terminal,
you
can
see
I've
got
the
health
check
running
here.
It's
a
negative
application.
You
can
run
it
either
as
a
pod
or
a
native
app,
maybe
do
some
debug
or
whatever.
If
you
want
natively,
this
is
running
verbose
mode.
Normally
it
doesn't
print
this
much.
But
again,
this
helps
for
determining
what's
going
on.
If
we
take
a
look
at
this
from
the
Prometheus
side,
so
this
is
the
Prometheus
which
I've
also
got
installed
in
Cube.
This
is
showing
the
targets
that
we're
currently
scraping.
F
If
we
come
down
here,
we
can
see
on
the
source.
Endpoints
here
is
the
catalog
controller
and
if
I
click
on
that
again,
you
can
see
the
metrics
that
Prometheus
is
going
to
scrape
down
at
the
very
bottom.
I've
got
the
catalog
health
check
added
and
again,
if
you
click
on
that
link,
you
go
and
see
those
metrics
if
I,
if
I
get
back
into
from
EPS
itself.
This
is
the
previous
query
language
up
here
and
what
I'm
showing
here
is
the.
F
Duration,
how
long
it
takes
for
the
health
check
tool
itself
to
run,
and
you
can
see
we
get
a
graph
here.
You
can
see
at
the
99th
percentile
for
the
health
check,
completed
we're
running
in
well
under
three
seconds,
all
the
time
generally
about
two
and
a
half
seconds,
and
you
can
zero
in
on
the
90th
percentile
quantile
or
the
50
percent
quantile.
Those
are
programmable,
so
you
could
set
those
to
whatever
you
want.
F
The
catalog,
along
with
the
OSB
request,
count
what
other
ones
that
have
helped
a
number,
obviously
coming
from
the
health
check
tool.
So
on
the
prometheus
side,
you
can
monitor
these.
You
can
alert
if
things
take
too
long
or
if
your
air
counts
get
too
long.
That
kind
of
thing
I've
got
I'll
update
the
pull
request,
I
think
at
the
sync
service
meeting
there
was
a
quest
to
take
it
out
of
the
test.
Folder
move
up
into
bin,
so
I've
done
that
and
make
a
couple
of
changes.
F
Look
at
that
push
right
after
the
meeting,
so
basically
looking
to
get
some
eyes
on
the
request
and
get
some
get
some
feedback.
I
know
if
there's
any
feedback
that
people
wanna
give
at
this
point,
I
guess,
since
I'm
blabbering,
oh
I
had
a
hard
time
running
this
under
cubed
with
local
cluster
up
if
I
start
hitting
it
hard
or
with
too
many
iterations.
My
controller
starts
restarting
I
fail
health
checks.
That
kind
of
thing
the
demo
I'm
doing
here.
F
B
F
G
I
thought
we
might
have
some
extra
time,
and
luckily
we
do
so
I
want
to
bring
up
two
different
use
cases
to
get
some
opinions
of
people
on
some
recommendations
on
how
to
solve
it.
The
first
one
is
hopefully
the
easier
of
the
two,
so
we
have
a
situation
where
we
have
a
resource
controller
that
is
watching
for
new
service
instances
to
be
created
and
upon
them
being
created.
G
We
want
to
take
some
action
and
in
order
to
perform
that
action
we
need
some
extra
parameters
be
passed
in
for
this
action
to
do
its
job
properly.
Does
anybody
have
any
recommendations
on
we're
inside
of
a
service
instance?
We
can
stick
this
extra
data
because,
right
now
we
were
originally
thinking
sticking
them
in
the
parameters
section
and
the
broker
could
just
ignore
the
thing
that
doesn't
understand,
because
hopefully
I
won't
recognize
these
things.
G
G
I
think
it
has
to
be
part
of
it,
because
even
if
we
did
a
config
map,
we
still
have
to
have
a
pointer
from
the
service
instance
to
the
config
map
right,
because
otherwise
you
got
to
search.
Every
single
config
map
to
see
was
a
pointer
to
the
server
to
some
random
service
instance.
I'm,
not
sure
how
you
get
that
cooling
mini.
B
Broker
doesn't
have
a
back
in
store
for
all
of
it
in
the
way
it's
per
its
persistence.
Is
that
say
when
you
do
a
provision
I
make
a
config
map,
that's
just
named
after
the
instance
that
you
UID
and
then
I
know
how
to
find
it,
because
I've
got
an
instance
and
I
just
do
a
get
on
config
map
for
the
name
of
the
instance
and
then
yes,
the
other
extra
metadata.
G
Know
do
I,
have
some
random
user
create
a
service
instance
with
these
extra
parameters
and
we
don't
want
to
have
to
have
them
go
out
and
create
a
separate,
can
take
map
someplace
else
and
make
named
them
the
same,
and
they
do
some
syntactic
relation
to
the
two
annotation
Shh.
Can
you
put
random
crap
the
internet'
imitations
like
entire
object,
or
is
that
you
string
it's.
H
G
G
The
maintenance
nightmare
of
trying
to
keep
thousands
of
many
little
service
catalogs
in
sync,
with
a
global
service
catalog
in
essence
or
catalog
of
services,
is
just
a
problem
that
is
just
astronomical
right
as
you
remove
and
add
services
from
the
global
catalog,
the
idea
of
managing
fine
9000
clusters
and
trying
to
manage
those,
you
know
all
and
keep
them
all
on
say
for
the
master.
One
is
just
sounds
like
it's
gonna
be
a
midnight
Meredith
in
the
ass
I'm,
sorry
I
painting
they
took
us.
G
We
were
kind
of
wondering
if
anybody
else
is
running
into
a
similar
problem,
because
one
solution
that
we've
been
toying
with
is
the
notion
of
not
storing
a
catalog
in
every
single
koreas
cluster,
but
rather
allowing
kubernetes
to
query
something
external
to
the
cluster,
to
get
the
list
of
services
and
plans
and
stuff
like
that.
But
before
we
went
too
far
as
wondering,
if
anybody
else
has
run
into
this
problem
and
wants
to
brainstorm
with
those
or
not.
G
B
G
G
A
diff
but
I've
also
heard
cases
where
people
say
you
know:
I'm
not
gonna
have
a
single
broker
for
all
nine.
All
thousands
of
services,
I
have
right.
I'm
gonna
actually
have
one
broker
per
service
because
they
want
to
allow
assert
a
self-service
model
where
a
service
broker,
in
essence,
registers
themselves
with
a
platform
and
then
that
that
broker
then
gets
registered
with
every
single.
G
G
The
part
that
needs
elf,
so
it's
a
little
bit
of
both
right.
We
don't
necessarily
want
to
have
all
9,000
clusters,
have
to
hit
a
single
catalog
and
then
have
to
do
this.
This
diff
every
single
time,
but
it's
also,
if
you
turn
the
problem
around
right,
well,
maybe
phrases
it
depends
on
how
you
choose
to
set
up
your
environment
right
if
you
choose
to
have
a
I'm,
not
thinking
this
right
hold
on
a
second
minute,
think
about
this
before
I
say
it.
G
E
So
I
have
some
experience
in
this
in
a
previous
life
managing
regional,
catalogs
or
video
game,
and
there
yeah
there's
no
good
answer.
It's
you
have
to
decide
if
you're
gonna
use
push
or
pull,
and
then
you
have
to
layer
in
your
caching
strategies
so
right
now,
as
far
as
catalog
pulls
every
some
odd
seconds
or
minutes
or
whatever
you
configure
it
to
right
and
if
your
brokers
getting
overloaded,
then
you've
got
to
add
some
caching
or,
like
I,
said
it's,
maybe
a
CDN
or
something
it's
actually
a
real
solute
idea.
G
Yeah,
so
much
for
caching,
as
I
said,
he's
gonna
help
us,
because
chances
are
when
the
Google
account
of
what
gets
updated
will
probably
go
through
and
poke
every
you
know
all
9,000
kubernetes
clusters,
mm-hmm
yep
tall
realists
and
then
they're
all
gonna
want
to
relist,
and
so
it's
not
a
matter
of
caching
that
point
it
really
has
9,000
brain.
Your
requests
have
to
get
answer.
I.
E
G
See
the
other
and
I
didn't
want
to
mention
this,
but
I
think
I
might
have
to
the
other
issue
that
beasts
Seether
aspect
to
this
is
we
have
kind
of
a
complicated
strategy
by
which
different
service
are
visible
to
different
platforms
for
lack
of
a
better
phrase.
Sometimes
the
white
list
against
a
black
list.
Sometimes
it's
some
other
rationale,
and
not
all
of
those
rules
are
necessarily
expressible
within
kubernetes.
I
know
that
we
I've
is
it
white
lists
or
black
lists?
Now
he
wrote
one
or
both.
G
And
I
think
we
I
think
we
have
more
complicated
scenarios
in
mind
and
and
in
particular
ways
to
also
start
thinking
about
how
not
all
platforms
even
support
all
those
types
of
mechanisms
like,
for
example,
Cloud
Foundry,
doesn't
support
blacklist
or
whitelist,
and
then
we
also
support
Service,
Catalog
types
of
stuff
for
things
like
straight
docker
right.
So
we
start
talking
about
all
these
various
platforms.
A
long
time
ago,
we
had
decided
to
not
try
to
inject
this
logic
into
the
platform
itself,
because
they'd
all
been
support
it.
G
E
E
G
Step
myself
back
earlier,
it
was
because
I
I
was
thinking
about
the
push
model
and
I.
Think
we'd
still
have
a
similar
issue
with.
You
still
now
have
to
keep
9000
things
in
sync
with
the
global
catalog,
and
it's
it's
as
it.
As
you
said,
it's
basically
push
versus
pull,
but
either
way
you
still
have
this,
this
hassle
of
keeping
two
things
in
sync:
global
on
one
side,
all
9000
on
the
other
side,
and
it's
just
a
maintenance
nightmare
to
keep
everything
happy
at
that
point.
G
It's
ask
you
for
things
to
get
out
of
sync
and
then
we're
trying
to
avoid
that
that
problem
and
that's
Soho
I
was
trying
to
wonder-
is
one
brainstorm
for
ideas
and
thank
you
Scott
for
brainstorm
a
little,
but
also
just
in
general,
to
other
people
foresee
having
this
particular
problem,
because
it
doesn't
seem
like
having
9,000
copies
of
the
exact
same
thing
in
one
environment
like
all
the
bluemix
or
all
Google
cloud
or
whatever
is
them
is
a
very
nice
solution.
I.
E
E
G
E
E
G
B
E
B
B
E
B
I
was
just
gonna
say:
if
you
want
to
look
more
like
a
paw
print.
I
would
have
the
the
upper
triangles
not
go
all
the
way
down
to
the
to
the
center
of
the
shape
and
then,
instead
of
having
straight
edges
along
the
top,
they
would
be
round
so
they'd
be
a
little
more
circular
like
realto
beans.
It
would
still
overall
make
the
right
shape,
but
it
would
look
a
little
bit
more
like
the
thing
you're
going
for,
because
right
now,
I
don't
see
the
paw
in
it.