►
From YouTube: Kubernetes SIG Network 20170309
Description
Kubernetes SIG Network meeting from March 9th, 2017.
A
C
B
E
D
C
A
B
E
B
So
the
testing
environment,
the
testing
runner-
can
be
run
independently
from
the
testing
system,
but
there
definitely
are
caveats
to
testing
system
that
are
very
hard
to
recreate.
As
we
environments
to
be
honest,
so
it
is
fairly
difficult
from
the
full
winter
wear.
It
brings
up
a
plunger
for
you
and
then
read
the
best
for
you.
E
B
B
A
And
I'm
so
that
next
to
me,
I
wanted
to
talk,
but
I
just
click
create
the
beginning
and
was
C
and
I.
So
people
hadn't
noticed
there
was
a
sea
and
I
release
about
50,
which
is
a
long
time
in
the
coming
and
came
out
yesterday.
Was
it
yep?
Yes,
so
I
think
they're
just
a
couple
things
related
to
this
one
yeah?
A
E
E
For
example,
multiple
interfaces
inside
a
pod,
multiple
edge
through
multiple
IP
addresses
inside
a
pod
of
some
stuff
around
a
provider
based
hypervisor
based
stuff
as
well
in
terms
of
think
of
some
yogurt
from,
but
all
in
all
I
think
it's
a
big
relief
for
C&I,
and
so
it
should
enable
most
of
the
plug-in
writers.
Another
part
of
this
big
on
improving
have
been
blocking
on
for
a
while.
The
other
thing
to
mention
with
the
release
is
that
attempts
to
be
backwards
compatible.
E
So
unless
you
actually
change
something
in
your
configuration,
even
though
you've
updated
the
code
to
ski
on
90.5,
if
everything
should
still
work,
you
have
to
expertly
on
in
do
this,
but
this
also
means
that,
on
the
crew
benetti's
side,
kuba
neji's
can
opt
into
these
features
and
its
crew
benetti's
responsibility
to
make
sure
that
it
doesn't
send
your
plug.
In
a
format
of
information
that
your
program
is
neither
yet,
though,
and
short
version,
plugin
writers
shouldn't
have
to
do
anything,
but
there
should
be
some
capabilities
and
then
your
future
writers
dish.
They
want
Dobson.
E
H
A
I
Auction
has
a
problem
is
proposed
that
general
purpose
will
I
use
that
for
current
to
currently
kubinashi
at
Java
deployment,
21
l
70
the
balancer
eunuch
last
episode
was
I.
Can
grasp
the
connection
about
the
a
single
11
know
the
balance
may
have
from
unwise
or
connection
number
and
patient.
So
then
them
maybe
some
neat
and
needed
after
do
the
wrong
to
create
a
motel
LOL.
I
know
the
balance
of
and
occupy
which
load
balancer
who
so
we
can
graph
and
the
in
talent
case.
I
J
Read
it
fine
here,
I
guess
I
mean
it
seemed
plausible
I
made
I
made
a
few
just
minor
typo
comments
on
the
PR.
The
concept
seemed
plausible,
but
I
have
no
real
Philip
familiarity
with
working
with
those
kind
of
things.
So
I
couldn't
really
tell
you
at
that
level
and
Tim.
Tim
Hawking
had
some
interesting
comments
in
the
in
the
issue
about
what's
a
class
and
what's
a
claim
seemed
that
was
valid.
A
K
Read
through
it
and
I
I
think
it's
a
good
idea,
I
think
the
idea
of
the
claims
it
could
make
sense
for
the
scenarios
that
could
you
describe
it
also
part
of
me
believed
that
a
load
balancer
should
be
a
infinite
resource
that
you
know,
if
you
think
about
a
multi-tenant
environment,
why
would
it
can
ever
be
denied
hello
down,
sir,
so
that
sort
of
conflict
with
the
idea
of
reserving
something
or
laying
a
claim
on
it?
And
there
could
be
other.
K
You
know
maybe
types
of
load
balancers
that
could
provide
the
isolation
or
characteristics
that
you're
that
you're
describing
not
not
saying
I'm,
not
criticizing
critical
of
your
proposal,
but
it
occurs
to
me
that
there's
so
many
different
scenarios
for
load
balancer,
those
bouncy
deployments,
inside-outside
regress.
It's
just
mind-boggling,.
A
I
K
Was
a
question
for
you,
so
it
getting
back
to
that
point
about
sort
of
inexhaustible
supply
of
load
balances.
I
mean
I,
know
that
netscaler
an
f5
devices
can
easily
easily
handle
hundreds
a
bit
and
that
circus
I
mean
on
a
massive
scale.
That's
not
a
big
number,
but
if
you're
going
to
deploy
this,
you
know
100
or
100
per
box.
So
you
seem
like
a
small
number
to
me.
So
is
that
a
scenario
that
or
is
that
a
method
that
you
funny
now
that's
the
way
you're
doing
it
today
are
foreign?
K
I
K
Yeah
I
understand
that,
so
that's
a
cloud
provisioning
service
example
I.
I
can
see
that
easy.
K
I
Yeah
I,
don't
oh,
and
that
people
currently
the
lamento
in
grass
a
face
you
put
how
they
put
up
I,
feel
they
have
clawed
her
in
front
of
ng
ng
vessel
in
Brussels
controller.
So
a
calendar
request
the
first,
the
rich,
the
actually
after
class
and,
as
we
have
class
left
for
the
request
to
ng,
distinguish
controller
and
the
ingress
controller,
then
for
the
ricotta
to
the
back
end
up
order,
our
team
has
use
a
clarified
car
deployment,
deploy.
I
I
Master
branch
has
implemented
a
little
action
in
legal
action
Angeles
in
black
controller.
So
the
main
may
there
is
another
approaches,
as
does
not
use
karate,
but
the
engine
single
controller,
dr.,
binder,
elechi
police
interfaces,
buy
it
by
itself.
So
I
think
you
sum
them
whistle.
Whistle
is
another
solution
for
a
tray.
A
B
L
Just
funny
what
we
did
seriously
so
when
you
have
a
big
hole,
yet
all
standardized
the
api's
for
the
benefit
or
the
qfd.
We
just
want
to
feel
our
community
that
we
are
doing
this
and
we
want
to
make
a
way
for
people
to
express
interest
in
the
network
resource
because
we
already
have
the
pvn
PVC.
It
has
to
be
illuminated
and
oh,
we
think-
and
we
already
have
some
requirements
from
our
customers-
that
they
want
to
shares
the
or
developer
or
their
customers
so
lets.
Basically
the
goal
we
have
a
the
pony.
L
We
can
show
some
demos
to
the
community,
the
police.
We
think
this
could
be.
We
can
do
to
subdue
the
latest
and
we're
resource
this
inverse
or
a
lot
of
them
all
the
work
field,
refining
it
now
next
step
would
be
to
do
the
recycling
stuff
of
the
lot
better
so
resource,
and
then
we
want
to
find
what
kind
of
networks
or
little
resource
we
want
to
define
like
the
bandwidth
and
iops.
L
Then
we
will
do
something
like
scheduling
for
the
load
balancer,
so
people
can
create
interest
of
how
may
I
oh
PS,
or
how
many
better
ways
you
want
and
we
want
to
find
a
way
to
enforce
their
in
the
load
balancer
for
the
qfd,
tough,
its
best
hen.
We
want
to
feel
to
the
community
that
we
can
do
this
with
qld
and
for
just
that,
quick
mention
will
also
can
use
f5
or
some
hardware
or
better
to
do
this.
L
L
A
M
So
he
hasn't
moved.
Ok,
so
I
think
you
basically
have
like
two
high-level
issues.
I
think
that
needs
resolution
on
these
multiple
networks.
Side
of
things,
I
think
they
are
probably
related-
is
what
I'm
thinking
I
think
everything
is
boiled
boiling
down
to
how
we
define
I.
Think
you
hit
on
the
point
in
your
earlier
comment.
So
how
do
we
define
the
role
of
an
api
user?
So
it
sounds
like
we
are
only
thinking
about
the
developer,
but
when
we
create
a
network
object
I'll,
be
expecting
the
developers
to
create
those
natural
objects.
M
I
think
thats,
that's
the
first
thing
that
we
need
to
agree
on
or
are
we
expecting
like
more
like
a
more
granular
user?
I
mean
it's
probably
not
only
the
developers
who
is
going
to
be
creating
all
the
ATA
objects.
I
can
see,
you
know
parts
and
the
you
know
our
application
of
the
Tomas
he
created
by
the
developers,
but
things
like
network
probably
is
going
to
be
created
by
a
developer.
/,
ops,
prison,
I.
Think.
E
It
it
also
depends
on
what
kind
of
hardware
and
capability
the
cluster
has,
because
if
network
creation
is
cheap
with,
whatever
plugin
is
being
used,
then
maybe
you
do
want
to
allow
developers
to
set
up
their
own.
You
know
custom
networks
and
whatnot,
but
you
know
if
it
requires
special
Hardware
on
each
node
or
if
network
creation
is
expensive
and
consume
community
resources,
then
you
might
want
to
restrict
that
doing
it.
Administrative,
okay,.
M
A
M
I
have
I
have
summarized
the
youth
is,
I
think
if
you
look
at
them
at
the
beginning,
I
have
summarized
some
of
the
use
cases.
I
think
the
rest
of
the
use
cases,
kinds
of
kind
of
fits
into
that
youth
is
at
the
top
that
I
have
summarized
mainly
mainly
boils
down
two
disjoint
networks
that
are
use
cases
where
the
networks
are
considered
disjoint.
In
other
words,
you
can't
reach
both
the
networks
through
the
same
interface.
That's
the
whole
point
of
having
multiple
interfaces
exist
apart
if
all
of
them
were
reachable
through
the
same
interface.
E
It's
not
entirely
true,
though,
because
I
think
with
different
network
capabilities,
even
if
you
could
reach
a
particular
Network
feel
like
a
you
know,
one
gig
interface
and
you
also
had
a
10
gig
interface
connected
to
that
same
network.
You
might
want
to
dedicate
that
10
big
interface,
the
specific
application.
M
That
that's
up
to
the
networking
infrastructure
to
take
care
of,
though
I
mean
you
can
you
can
have
the
same
nature
plug-in
connected
to
two
different
app
links,
and
then
you
can
use
just
plain
forwarding,
rules
or
policy.
You
decide
which
way
you
wanted
to
go.
I,
don't
think.
That's
the
issue.
A
recursive
here,
I
think.
M
Why
why
you
would
want
something
like
this
in
the
deployment
in
the
use
cases
a
lot
of
times
it's
because
of
operational
security,
because
you
don't
want
this
particular
Network
accessible
through
the
data
or
main
data
path.
For
example,
you
want
to
keep
your
management
if
you,
if
you're,
deploying
a
multi
done
in
the
application
you
think
given
it
is,
you
might
want
to
keep
your
infrastructure
Network
separate
from
the
ten
little
you
might.
You
might
be
using
it
for
some
sort
of
provisioning
or
control
like
that.
There
are
applications
like
that,
especially
involving
nsp.
M
A
A
little
bit
fuzzy
on
those
last
things
he
said
and
how
you
see
me
being
all
slow.
So
when
you
say
fundament
like
this
destroyed
storage
from
internet
works,
for
example,
are
you
referring
to
like
house
networks?
So
the
host
has
multiple
interfaces
and
you
want
to
bind
upon
directly
to
one
of
those
that.
M
Would
be
one
you
skate,
I
mean
I
think
in
order
to
have
this
join
networks.
Yes,
the
host
needs
to
have
multiple
physical
interfaces,
I
think
if
the
host
ultimately
only
has
a
single
Nick
I,
don't
think
it's
possible
to
have
this
jar
networks,
at
least
at
the
virtual
level.
Basically,
it
may
be
a
single
neck,
but
at
least
virtually
huge
traveling
to
different
mix.
In
order
for
us
to
be
discharged.
A
M
A
E
G
Oh,
actually
I'm,
not
sure
it
does
is
if
these
are
on
the
host
and
I've
got
virtual
interfaces
on
a
container
that
are
connecting
to
those
two
different
physical
interfaces
are
doing
that
through
some
kind
of
shared
connectivity
within
the
underlying
host
OS
right
see,
they're
going
to
be
a
bridge,
or
it's
going
to
be
a
routed
interface.
It's
going
to
be
something
so
at
some
point.
I
am
mixing
this
traffic
anyway
in
underlying
Colonel,
not.
E
Necessarily
I
mean
you're
mixing
an
underlying
Colonel,
yes,
but
you
might
not
be
mixing
the
storage
traffic
in
the
general
pod
traffic.
You
could,
for
example,
have
a
V
switch
or
a
bridge
dedicated
to
each
of
the
physical
mix
in
the
card
and
attached
sorry
virtual
nic
to
those
specific
bridges.
So
you
wouldn't
be
mixing.
You
know
any
of
the
storage
in
the
general
network
traffic
for
a
pod.
You.
G
M
M
F
Happen
but
also
have
interfaces
like
SR,
iob
or
stuff
like
that
it
would
go
on
with
luggage
by
the
pier
is
another
interface.
So
imagine
I,
don't
know
who
has
some
in
SD
case.
You
would
want
some
fri
of
inside
the
pot
and
you
would
have
the
wrong
moment
working
for
the
control
and
the
other.
One
is
for
data,
so
you
would
have
legitimately
have
two
interfaces
there
and
who
says
I
services,
some
critical
cases,
and
so
on.
I.
F
F
M
M
F
Fun,
that's
fine,
so
the
proposal
actually
was
it
goes
with
it
on
along
those
lines,
because
what
it
does
is
it
keeps
the
network
object
in
the
API
not
having
knowledge
about
what
is
the
nature
of
the
network
because
it
like
you,
can
put
it
there
as
metadata,
but
it
doesn't
know
about
specific
right.
I
might
be
before
I
PVC.
It
would
be
a
description
of.
M
M
We
have,
we
all
have
like
different
talk.
You
know
different
ideas
on
that,
so
we
will
arrive
at
a
conclusion,
but
I
just
wanted
to
make
sure
that
we
are
all
on
the
same
page
from
a
word
so
perspective,
because
it
appears
that
we
are
not
so
let's,
let's
get
there
first
and
then
I
think
we
could
consolidate.
You
know
your
ideas
with
what
we
already
have.
There
is
already
like
a
more
than
a
dozen
people
already
contributed
to
this
proposal
to
working
with,
so
you
know
us
will
be.
M
Then
it'll
be
great
if
we
can,
at
least
you
know,
arrive
at
a
conclusion
on
the
workflow
side
of
things
and
how
configurable
things
needs,
because
it's
easy
enough
to
take
an
take,
an
approach
where
you
know
all
we
have
from
a
network
of
object.
Perspective
is
just
their
handles
and
everything
else
gets
configured
out
of
band.
So
that
will
go
back
to
my
before.
K
F
K
Know
I
wanted
to
set
that,
but
you
want
to
get
to
make
much
sense.
My
second
year
I
understand
when
you
say
four
separate
networks
and
not
mixing
traffic.
I
get
all
that
and
I
believe
I
Carol
are
these.
But
my
question
is
simply:
if
there
are
two
interfaces
and
a
pod
that
are
attached
to
be
different
networks,
how
do
you
apply
network
policy
to
that?
Okay,.
M
K
M
Ok,
so
now
getting
back
to
you
know
the
workflow.
So
can
we
can
we
get
some
ideas
on
that
friend?
So
do
people
think
that
you
know
the
network
needs
to
be
just
a
just
a
handle
and
everything
else
needs
to
be
configured
completely
outside
of
Cuban.
It
is
or
do
we
think
that
needs
to
be
some
configurability
from
a
Cuban
at
his
side
of
things.
E
I
mean
I,
I,
guess
I'm,
just
not
seeing
how
I
I
do
see
some
advantages
to
having
configuration
with
incur
brunette
ease,
but
if
we're
not
assuming
a
specific
network
plug-in
and
a
specific
plug-in
driver
juicy
and
I
or
doc
or
whatever
I,
don't
see
how
kuber
Nettie's
can
be
of
much
use
here
other
than
basically
an
envelope.
You
know
you,
like
some
plugins
specific
configuration.
E
You
wrap
that
up
in
some
kind
of
thing,
so
paid
kaku
Brunetti,
and
it
sends
it
true
because
you
know
yeah,
maybe
you
could
send
it
not
a
fusion
the
routes,
but
it's
not
the
case
that
all
plugins
are
going
to
be
able
to
work
with
that
or
be
able
to
provide
the
capabilities.
That
Cuba
days
would
expect
right.
M
C
M
So,
for
example,
one
of
the
things
that
came
up
early
in
the
discussion
is
devout.
Inning
I.
Think
that's
that's
where
this
lot
of
this
discussion
started
right,
so
the
person
really
is,
should
Cuban
Aries
have
any
knowledge
of
this
or
it
has
to
be
completely
in
the
network
plugging.
So
the
situation
is
when
you
have
disjoint
networks.
M
Like
we
discussed
it
is
not
possible
to
configure
the
routing
entirely
in
one
of
those
networks,
so
potentially
potentially
the
two
networks
they
could
be
in
the
same
implemented
by
the
same
plug-in,
but
it's
also
possible
that
they're
implemented
by
different
plumbing
set.
So
if
you're,
if
you're,
covering
the
most
general
scenario,
I
think
that
routing
actually
belongs
in
the
part.
M
F
M
M
It
yeah
default
route
is
more
like
a
specification.
I
think
that
really
a
question
is
which
is
about
right,
so
that
decision
has
to
be
made
by
the
part
and
I
think
that's
what
that's
what
is
I
was
referring
to
now.
The
question
is,
you
can
take
different
vehicles
to
get
it
there,
so
if
you
want
to
be
very,
very
generic,
like
you
know,
to
support
different
networks.
Supper
by
different
plugins,
I
think
the
only
way
to
coordinate
this
is
from
the
pod
or.
E
E
M
That
that's
right,
so
we
can
easily
take
an
approach
that
you
know.
The
primary
network
is
the
default
route
and
all
of
the
other
networks.
They
only
deal
with
directly
connected
n
points.
Okay,
so
if
it's
directly
connected,
then
you
don't
need
an
explicit
route,
so
you
have
directly
connected
out
and
you
have
this
default
gravity.
We
could
leave
it
at
that.
I
was
going
to
leave
it
at
that,
but
then
Tim
pointed
out
that-
and
we
don't
want
to
reach
to
this
in
the
near
future.
M
Somebody
will
come
up
with
a
use
case
where
you
know
this
needs
to
be
more
generic,
more
general,
and
that's
why
you
know
we
are
trying
to
solve
this
in
a
more
generic
way.
Otherwise,
it's
easy
enough
to
I
think
to
cover
out
of
the
use
cases.
It
may
be
sufficient.
If
you
just
say
we
deal
with
all
the
secondary
networks,
they
will
just
be
the
directly
connected
traffic
and
the
primary
network
will
be
the
default
route.
We
could
say
that,
and
we
could
hear
that
so.
F
E
Matter,
obviously,
if
you're
attempting
to
reach
a
resource
that
is
directly
on
that
IP
range,
that
that
has
some
time
the
colonel
knows
exactly
what
to
do,
and
why
you're
saying
that
if
you
want
to
reach
an
IP
block
that
is
not
locally
connected
or
not
we're
subnets
of
those
interfaces.
How
do
you
pick
which
interface
that
is
right
exactly.
M
E
D
E
E
E
Mean
I
guess:
maybe
one
way
to
move
forward
here
is
to
sort
of
define
a
priority.
I
mean
like.
If
you
tag
network,
they
should
tag
a
pod
with
various
networks.
You
know
the
first
one
in
the
current
proposal
is
the
primary
network,
and
so
obviously
that
one
has
precedence
and
we
could
just
say
the
feed
list
is
ordered
and
if
there
is
a
conflict,
somehow
I'm
not.
G
G
Think
your
queen
or
not
louding,
then
how
do
you
enforce
this
I
think
a
cleaner
way
of
doing
this
would
be
just
follow.
Standard
IP
procedures
can
do
longest
prefix
match.
If
you
really
want
something
to
go
down
a
specific
interface
make
sure
it's
got.
The
the
address
is
our
longest
prefix
match
and
then
the
current
will
always
do
the
writing.
M
M
E
M
M
A
A
M
Think
so
I
think
the
rest
of
the
things
are,
like
minor,
think
I
think
everything
finally
boils
down
to
this
issue.
I
think
I
mean
there
was
this
issue
about
whether
you
pass
an
argument.
Okay,
I
just
kept
it
as
a
generic
placeholder,
so
I
think
you
can
probably
so
so,
let's
say
I
attached
the
network
and
there
is
a
way
to
pass
an
argument
to
the
plug-in
from
the
pod
spec.
M
A
M
F
A
M
F
M
A
M
F
M
A
M
A
M
I
don't
if
I
don't
see
them
resolved
in
a
day
or
two
I
will
I
will
move
them
to
the
salt
and
then,
if
you
don't
agree,
you
can't
reopen
them
so
that
stuff.
This
is
general
things.
I
wait
for
two
days
and
if
I,
don't,
if
I
don't
see,
issues
is
all
I
might
I
might
map
in
the
soil.
And
if
you
don't
don't,
don't
feel
offended
that
I
have
to
solve
them.
Yes,
they
open
them
and
tell
me
you're
wrong.
Yes,
that's
fine!
Yes,.
A
E
What
other
technologies
are
people
looking
at
for
keep
proxy
as
you
implemented
your
own
kind
of
version
of
cube
proxy
or
maybe
a
different
way
to
phrase
it?
If
you
could
do
whatever
you
wanted
with
cube
proxy
and
rewrite
it
in
some
way
or
modify
it
in
some
way,
you
know,
would
you
use
something
else
like
and
ETF
or
supply
store?
Ipps?
E
Yeah,
my
current
thought
is
that
to
solve
some
of
the
problems
with
iptables
contention
and
performance,
that
NS
tables
is
probably
the
clearest
path
forward
because,
like
I
said
it's
the
closest
analog
and
it
does
provide
some
of
the
things
that
are
useful.
Like
notifications
and
rules
change,
it
also
provides
a
much
more
granular
way
to
change
rules
and
replacing
the
entire
world,
which
is
the
cause
of
some
of
the
contention
issues
that
we've
seen.
E
But
NF
tables
may
not
quite
be
ready
yet
because
it
may
not
implement
all
the
features
and
iptables
right
now.
But
again,
it
seems
like
most
things
that
q
proxy
currently
does
with
iptables
could
be
converted
over
to
NF
tickets
very
easily,
whereas
something
like
you
DTF
for
splice
or
ippf
would
probably
be
a
very
large
rewrite
of
cute
and
both
all
of
those
have
their
kind
of
upsides
and
downsides.
But
I
mean
personally
I
haven't
explored
too
much
down
at
yuki,
EDG
f4i
from
PS
or
splice
pass
I
know.
E
E
So
it's
not
necessarily
an
either/or
choice,
but
it
seems
like
there's,
probably
not
enough
manpower
behind
to
proxy
already
and
so
attempting
to
expand
the
set
of
choices
might
not
be
the
best
thing
unless
people
step
up
to
maintain
those
going
forward,
but
then
also
you
know,
there's
kind
of
lonely
right
now:
1q
proxy
and
you
know
I
mean
do
we
keep
it
as
is
and
add
something
else
and
kind
of
deprecated
in
proxy.
Or
do
we
you
know,
convert
it
over
to
something?
I
guess.
E
E
E
H
I
think
it
would
be
a
good
idea.
Maybe
we
could
sort
of
figure
out
kind
of
what
a
generic
q
overlay
requires
and
then
a
good
look
at
the
technologies.
For
example,
one
thing
as
implementing
services
there
is
maybe
taking
it
to
you,
know,
implementing
the
overlay
and
so
forth,
yeah
and
then
you
how
about
slots,
because
I
think
right
now,
it's
like
very
focused
on
polk,
you
proxy
plus
IV
table,
and
maybe
it's
hard
to
see
that
yeah.
E
Definitely
that's
something
I
think
Tim
brought
up
as
well
that
there's
no
really
good
document
that
says
this
is
what
is
expected
of
a
proxy
for
cuva
Nettie's,
including
cube
proxy
itself.
So
even
if
you
were
to
write
a
separate
proxy
that
was
not
to
proxy
it's
hard
to
know
all
the
intricacies
of
what
cube
proxy
currently
provides,
and
so
whether
or
not
your
alternate
proxy
even
meets
those
requirements,
so
I
think
that
it's
very
good
point
we
do
want
to
create
a
document
that
describes
expectations
of
a
proxy.
E
A
C
As
far
as
limiting
each
namespace
to
only
be
able
to
contact
things
in
their
space
and
not
information
leak,
and
there
was
a
plan
potentially
to
introduce
an
idea
of
like
an
organization
thing
which
wraps
together,
multiple
namespaces,
but
when
I
brought
that
up
to
fig
off
it
was
kind
of
like
a.
This
is
going
to
be
a
great
thing,
but
it's
probably
going
to
sit
in
proposal
phase
for
quite
a
bit.
So
I
was
wondering
if
anyone
had
any
faults.
Well,
maybe
an
interim
step
to
get
so
there
yeah.
A
A
D
D
A
A
J
A
A
A
A
From
forget
exactly
from
who,
but
I
sunder
out
an
email,
p
cig,
lead,
zinc,
leads,
saying
yeah,
they
booked
these
rooms
for
on-site
stuff
and
put
your
name
down
for
one.
If
you
like,
okay,.
J
J
A
Go
yeah,
it
sounds
like
there's
minor
preference
for
the
later
one
that
I'll
just
double
check
that
we
actually
have
that
one
before
we
cancel
yours
yeah.
Somebody,
though,
go
slack.
One
will
do
all
right,
that's
our
time
for
today.
So
thank
you.
Two
weeks,
what's
two
weeks
from
now,
not
that'll
probably
be
in
first
time
two.