►
From YouTube: Antrea Community Meeting 12/04/2019
Description
Antrea Community Meeting, December 4th 2019
A
A
It
was
probably
either
login
with
the
easy
logging
in
Soma,
with
VMware
credentials
with
the
corporate
credentials
anyway.
So
what
are
the
goals
for
this
meeting?
So
it's
something
that
first,
we
need
to
define
together.
In
my
opinion,
it's
a
it's
a
sort
of
a
crossover
between
energy
and
develop
a
developer
meeting
and
a
release
management
meeting
which
basically,
we
need
to
take
care
of
two
things.
One
when
you
take
care
on
making
sure
that
release
ahora
releases
are
correctly
planned.
A
That
items
that
we
expect
deliver
and
release
are
on
track,
and
if
not,
we
either
push
them
out
of
the
release,
or
you
know
we
take
corrective
action
like
seeing
if
we
can
focus
the
community
attention
to
this
particular
particular
items.
The
second
part
is
a
sort
of
developer
meeting
where
you
know
we
use
this
space
for
discussing
the
issues
for
discussing
proposed
proposals,
existing
proposals
and
Febreeze
storming
in
new
ideas,
perhaps
not
now,
but
hopefully
in
the
very
near
future.
A
This
should
also
become
a
sort
of
user
forum
which
we
will
discuss
not
only
among
developers
but
also
with
users.
What
are
their
most
important
needs,
other
their
pain
points
in
using
this
solution
and
how
you
could
make
their
life
better
and
say
that
if
you
think
that
we
could
add
something
else
to
the
goal
of
the
of
this
community
meeting,
please
just
let
me
know
speak
up.
A
All
right,
so
the
next
topic
on
the
agenda
is
to
go
together
through
an
tree
architecture
and
and
basically
I
know
that
most
of
you
on
this
call
already
have
already
know
it
very
well,
but
it
could
be
a
good
refresher
for
other
people
on
the
call
like
myself,
for
instance,
which
I'm
still
a
little
bit
struggling
with
a
few
concepts.
So
you
know
it
might
be
good
if
we
can
spend
some
time
ten
minutes
per
up,
providing
an
introduction
that
to
a
discussion
of
the
architecture
that
will
steal
recorded
years.
A
B
B
B
B
B
I
assure
you
guys
already
know
about
open
usage.
It's
a
high-performance
programmable.
We
switched
the
suppose,
both
ITA's
and
windows,
and
actually
also
bsd,
but
the
forest.
Probably
only
dogs
and
the
windows
are
more
relevant
open,
which
is
very
flexible.
So
writing
labels
intuitive
increment,
Cobras,
light
working,
the
layer,
2
layer,
3
and
the
title
call
is:
they
are
in
a
high-performance
manner
since
it's
programmable,
it's
very
easy
to
add
new
features,
incremental
e
and
Cinco
data
plan.
B
B
B
It
also
program
IP
tables,
but
it's
only
for
the
child
room
from
from
one
local
quad
to
the
node
network
or
external
network.
So
we
do
have
IP
table
drawer,
actually
a
single
rule
to
do
n
Snak
for
traffic
from
poor
to
little
network,
so
the
traveler
will
go
to
a
low
guy
peein
and
then
only
send
out
to
the
node
network
from
the
node
in
those
activities.
B
Last
one
we
have
a
controller.
Now
it's
deployed
using
a
custom,
coolest
Department.
We
still
only
a
single
radical
for
now
and
it's
it's
major
jewelry
computer
night
policy
covers
data
policy,
IP
size
and
the
number
ports
and
the
push
computer
policy
are
taking
every
nook.
Then
agent
can
actually
just
use
a
computer.
The
results
who
are
providing
research
flows.
B
Yes,
say:
I
too,
can
be
a
skilled
at
remotely
from
any.
No
that
can
accessorize,
API
or
Antioch
controller.
It
can
also
be
as
good
as
a
public
servant
judicata
plugging.
That
means
you
Ron
Kuby
Carter
and
then
specify
a
country,
our
CI
and
cut
to
be
the
command,
and
so
in
that
way
you
can
use
the
cookie
cutter
that
were
most
developer,
familiar
with
Ron
our
CI,
and
you
can
also
use
cookie
cutter
config
file
for
the
entire
API
discovery
and
the
ossification.
B
And
then
we
also
have
this.
You
are
plugging
in
optin.
Basically,
octane
is
the
open
source.
Ui
dashboard
for
gratis
I'm
sure
has
a
clocking
in
octant.
It's
also
for
debugging
covers,
and
they
can
show
some
basic
a
long
time.
Information
of
the
country,
our
components,
okay,
how
status
the
washing
and
some
other
information
in
hydrocal
all
information
from
Christ's
API
service
country
are
CRD
entire
customer
resource
definition.
Suppose
a
consumer
and
Ageing
to
work
with
status
in
class
API
to
his
post,
the
runtime
information.
B
B
We
don't
have
much
time,
maybe
let's
just
go
to
them
a
polite
working
part,
and
then
we
can
talk
about
how
we
configure
I
think
this
for
every
port
and
we
can
enter
them
into
OS
bridge.
So
it's
quite
straightforward
and
I.
We
know
that
we
have
OS
bridge
code
behind
I,
think
it
stands
for
Inca
grid
integration
bridge
and
we
have
a
single
subnet
for
every
node
actually
allocated
by
a
cool-ass
node
node
I,
Pam
consumer
high
school.
As
our
dream
code
and
for
I
report.
B
We
have
an
interface,
the
entire
cake
in
the
work
with
interface
and
allocated
one
IP
address
from
the
novice
event,
and
then
we
have
a
wizard
how,
with
the
inner
in
the
ways
that
you
asked
to
be
the
East
earlier
interface
of
the
court
and
the
outer
we
study
was
to
be
canonical
OS
bridge
and
then
an
approach.
We
have
the
Gateway
theory
interface
record,
it's
the
Gateway
of
the
subnet.
B
B
We
use
the
device
and
then
to
the
east
ability
was
inside
the
bridge,
I'm,
sorry
inside
the
that's
the
important
namespace,
so
chef
Eli
believes
who
has
previously
and
then
folded
traffic
between
two
poles
antiphonal
again
into
the
pedal
into
the
the
West
bridge
and
a
source
node,
and
it
would
be
fallen
to
the
tunneling
fence.
Not
here
are
two
wheels
down
what
you
leave
encapsulation
based
on
the
transformation
of
the
Andrea,
and
then
the
traffic
will
be
sent
to
the
favorite
core
interface
over
the
node.
B
It
work
goes
to
the
Ellis
bridge
that
will
be
fall
to
the
K
two
entities
and
OS
bridge
and
here
are
based
on
the
hosts
routing
tables
I
to
record
from
one
of
the
IP
notice.
I
know
only
table
all
who
have
equitable
jurors
to
the
s
night.
So
the
the
path
Arabia
I
started
to
the
node
IP
address
and
I
go
out
to
the
Vedic
network.
Maybe
a
we
should
say
on
the
network,
since
the
node
can
be
a
mochi
mochi.
B
The
last
one
for
the
shopping
to
the
cross
type
II
service
country
are
now
I
leverage
coolest
I've
dreamed
could
be
projected
to
the
teachable
load,
balancing
for
classic
you
service
traffic.
So
what
were
hyping
at
the
traffic
from
the
Polaroid
go
to
the
k2
interface,
since
the
kid
will
even
sits
in
the
house
default
namespace.
B
So
no
matter
you
use
IP
tables
or
IQ.
Yes
or
you.
L'espace
know
that
we're
on
to
be
proxy
cool
because
they
have
the
chance
to
intercept
the
traffic
and
the
to
denied
internet
along
the
bank
and
the
post
to
be
the
destination
that
Enoch,
who
does
the
IP
on
the
court
from
the
service
class
IP?
And
then,
if
we
focus
on
the
same
and
when
it's
important,
we
go
to
the
we
use
the
device
that
court
order
to
the
polling
place
in
the
opponent's
face.
I.
B
B
A
Perfect
I
mean
I
believe
that
is
it's
great,
that
you
know.
As
a
matter
of
fact,
a
layer,
two
broadcasts
are
or
layer
to
traffic
pure
layer
to
traffic
is
terminated
at
the
node
and
we
never
send
broadcast
across
standards,
which
should
be
good
for
for
scalability
and
performance.
Alright,
thanks
for
your
answer
is,
is
there
any
other
question
I.
D
Just
wanted
to
say
on
the
related
note,
I
think
I
think
two
important
missing
pieces
of
documentation
right
now
or
like
especially
like
developer,
facing
documentation
or
a
description
of
the
OBS
pipeline
and
a
description
of
how
we
compute
and
disseminate
network
policies
and
implement
network
policies
and
I'm
working
on
the
first
part
right
now,
and
so
that
will
cover
things
such
as
well.
What
we
just
talked
about
how
our
traffic
never
leaves
and
so
understand
and.
B
D
Talking
about,
like
maybe
like
a
detailed
example
like
if
you
have
such
and
such
communities
network
policies,
your
what's,
get
transform
that
the
controller
and
you're
what
gets
sent
to
the
node
and
your
other
flows
that
you
end
up.
Quiz
I'm,
not
user,
facing
documentation,
but
really
like
developer,
facing
to
understand.
What's
going
on?
Yes,.
B
A
C
D
A
A
Yeah
they'll
be
greater
the
more
we
are
the
more
the
merrier
anyway,
the
now
we
should
get
like
into
the
release
management
part
of
the
meeting,
which
is
what
I
am
usually
terrible
at.
So
we
need
to
decide
what
is
the
timeline
for
our
next
release
and
I
mean
not
really
decide
what
the
list
tentatively
define
a
timeline
and.
A
D
Far
as
really
striking
yeah
I
think
we
should
have
used
like
ISA
milestones
in
the
past
I
think
there
are
also
more
advanced
tools
and
detailed
as
well
like
project
tracking
and
what
that
kind
of
stuff,
but
I
think
milestone
is
a
good
first
start,
since
we
have
like
a
feisty,
fast
release,
cadence
and
and
not
that
many
computers
for
now.
So
basically,
we
should
just
like
the
size
of
a
time
line,
triage
issues
for
each
release
by
assigning
them
to
a
milestone
and
basically
keep
track
of.
A
That
has
a
simple
workflow
is
perfect
for
me.
So
you
know,
I
was
thinking
of
using
JIRA,
it's
going
to
be
it's
perfect,
I
love,
it
everybody
loves
it.
We
should
use
it
and
unfortunately
there
is
no
free
plan
for
JIRA,
so
it's
not
open
source,
so
I
will
get
up,
but
for
the
time
being,
I
think
that
my
stalls
are
perfect
because
we
just
need
to
label
every
feature
with
a
candidate
milestone.
That's
all
that
we
need
to
do.
E
A
If
we
want
or
amended
the
next
release,
where
we
think
we
may
have
some
meaning
of
meaningful
feature
to
add
or
a
more
stable
platform
so
for
the
patch
release,
maybe
I
was
thinking
that
perhaps
we
can
look
at
a
few
but
open
bugs
decides
we
decide
if
there
is
any
bug
that
deserves
any
of
release
to
be
to
be
publish
it,
and
then
you
know
once
the
bugs
are
fixes
and
the
tests
for
those
cases
are
added
to
the
CI.
If
applicable,
we
can
do
the
patch
release.
A
The
minor
release
is
a
little
bit
more
complex
because
I
would
like
to
know
from
you.
What
is
what
are
you
in
your
opinion,
the
team
which
should
be
the
of
these
new
release.
Stabilization?
Do
we
have
some
very
urgent
feature
that
needs
to
be
pushed
or
do
you
need
to
southward
do?
Do
we
want
to
add
some
integration
with
some
other
piece
of
software,
so.
B
Of
course,
the
item
policy,
but
even
for
other
features,
like
can
I,
hear
it
here,
since
we
never
have
rabbit
on
a
good
enough
casting
moment,
probably
a
little
surprised
to
see
any
put
into
users,
and
also
we
saw
I
wish
you
to
some
scale.
Tests
for
not
voicing
probably
are
not
is
packing
a
very
good
scale,
but
at
least
something
to
chew
a
are
not
doing
water
bed
and
then
I
will
lose
the
thing
about,
say,
I
on.
B
A
Make
sense,
I
believe
I
believe
that
this
this
is
probably
quite
good
enough
for
a
release.
I
mean
if
we
ever
our
release
target
by
the
end
of
the
year
doing
another
release
that
should
be
a
good
payload.
I'm
curious
about
skill
testing.
I
have
seen
on
github
that
we
are
already
working
to
make
this
test
reproducible
so
maybe
meaning
that
everyone,
if
you
want,
they
can
run
scale
tests.
A
B
D
Know
I
think
basically
I
think
to
the
0.2
video,
like
our
first
three,
that
we
actually
like
qualify
in
a
way
by
intense
and
network
policy
test
passed,
so
I
think
we
should
basically
hold
off
on
the
release
unless
there
is
something
urgent
until
the
test
can
pass
and
we
can
say.
Okay,
all
the
tests
are
passing
now.
Two
techniques
elope
somewhere,
so
I
think
that,
with
your
milestone
for.
C
A
A
B
D
A
C
D
A
You
know
that
part
of
the
network
policy
tests
having
them
as
a
part
of
the
let's
say
the
ordinary
CI
it
could
be
good
and
also
not
adding
those
tests.
Technical
she'll,
not
and
a
long
time
to
the
end-to-end
test
execution
ask
for
the
conformance
test
is
the
last
time
data
went
through
them.
There
are
actually
not
a
lot
of
conformance
tests
that
can
be
impacted
by
the
CNI.
Let's
say
that
probably
less
than
5%
of
the
conformance
tests
are
something
that
has
an
impact
on
the
CNI
plugin.
A
D
Ryder
select
a
subset
of
network
policy
test,
a
representative
subset
to
run
I
can
run
all
of
them.
I,
don't
think
we
can
run
all
of
them
on
a
public.
See
is
that
for
sure,
but
maybe
on
the
internal
jenkins
bed.
Maybe
I
think
ambition,
Abishag
probably
knows,
but
it
takes
something
like
it
takes
more
than
an
hour,
I
think
to
run
all
of
them.
If
you
don't
realize
anything.
A
A
A
A
A
E
D
C
C
A
See
any
issue
that
is
critical
enough
to
grant
a
bug-fix
release,
and
maybe
maybe,
since
we
are
discussing
bugs,
we
can
move
to
the
bugs
bullet
of
the
agenda
and
then
discuss
feature
proposals
in
the
last
minutes.
Last
few
minutes
of
the
meetings,
so
in
terms
of
bugs
I'm
going
to
share
my
screen
and
one
of
these
tubs
as
the
issues.
Oh,
yes,
the
one
that
I
have
open.
A
A
For
instance,
the
only
one
that
bothers
me
a
little
is
this
one
about
the
photometric
policy
rules
that
are
not
enforced
correctly.
It's
not
a
security
problem,
because
you
know
it
would
be
a
security
problem
if
pods
that
are
suppose,
if
traffic
that's
supposed
to
be
denied
is
instead
allowed.
But
in
this
case
traffic
that
is
supposed
to
be
allowed
is
not
allowed.
So
if
we
want
to
do
a
bug-fix
release.
D
Let's
talk
about
119
after
but
I
think
for
131.
One
issue
potentially
is
I
think
that
the
fix
for
that
maybe
depends
on
a
lot
of
changes.
We've
made
okay
support
since
the
original
1.0.1
auto
release,
so
maybe
actually
just
cherry
picking.
That
change
may
require
like
putting
in
like
a
lot
of
stuff
or
may
not
be
super.
A
A
If
it
is
in
easy
cherry
pick,
then
okay,
we
can
do
it,
but
if
we
ever
now,
if
the
code
is
if
the
code
has
moved
and
the
patch
is
not
a
clear
cherry
pick,
which
means
that
we
have
to
apply
multiple
commits
or
redesigned
initial
commit
in
a
way
that
can
be
cherry
picked,
then
maybe
we'll
say
we
can
wait
for
the
next
minor
release.
You.
D
F
A
Yam
and
I
found
out
that
the
golang
is
not
really
doesn't
make
really
easy
to
handle
conflicts.
I,
don't
know
why
but
I
get
lost
when
I
have
magic
conflicts,
single
leg
problems
anyway.
So
let's
say
that
so
131
is
probably
critical
enough,
but
it's
not
bye-bye.
It
by
itself
is
not
granting
the
need
for
a
new
release
and
then
the
other
one
that
I
wanted
to
discuss
is
119,
I.
Think
Antron.
You
have
a
yeah.
D
So,
basically,
to
give
context
the
issue:
is
somebody
tried
to
deploy
entry
on
vke,
whereas
en
I
was
already
running,
and
so
that
just
applies
a
entry
manifest
basically
and
some
pods
were
already
like
present
in
the
cluster
and
they
had
like
IP
addresses
allocated
by
the
previous
yanari
and
so
and
I
just
started
reallocating
those
IP
addresses
but
I.
Think,
okay,
the
coral
dish,
I
think
the
action
item
for
us
is
more
to
document.
D
What
should
be
done
when
you
want
to
deploy
entry
on
a
cluster
that
already
has
a
CNI
I.
Don't
think
there
is
any
like
bug
really
to
fix,
because
that's
kind
of
like
the
expected
behavior
sure
we
can.
We
could
always
say
well,
not
we're
going
to
look
at
the
existing
pods
from
the
API
server
get
their
IP
addresses
not
allocate
those
but
like
it
just
makes
no
sense
to
have
like
old
pods
with
old
IP
addresses
managed
by
you
previously
and
I
still
like
pending.
So
I
think
we
should
just
say.
D
A
And
I'm
sorry
I
didn't
know
that
this
was
the
G
key,
but
with
it
we
discussed
a
couple
of
weeks
ago,
already
I
told
that
it
was
another
one,
so
yeah
I
think
that
we
can
providing
you
know
providing
a
sort
of
statement
elect
the
one
that
you
just
give
us.
We
should
probably
consider
closing
this
issue
because
it's
not
it's
not
a
scenario
that
we
can
in
the
future.
Maybe
in
the
medium
term
future
we
need
to
have
a
solution
for
feeling
out.
A
I
want
three
I
can
work
well
with
the
CNI
chaining,
that's
something
that
we
have
to
do.
For
instance,
you
know
at
the
moment
an
3r
doesn't
have
the
you
can't
have
really
your
own
pumps
year
on
I,
Pantheon
I
for
essence
and
feed
the
IP
address
has
been
provided
by
the
I
am
into
Andrea.
That's
something
that
you
cannot
do
at
the
moment,
and
in
this
case
it
was
completely
different.
You
had
an
address
here,
9f
does
wiring,
and
then
he
country
was
supposed
to
do
rewiring
for
that
network
interface,
which
of
course
it.
A
B
A
D
A
D
I
actually
started
that
bugs
and
I
moved
to
a
proposal
and
I
was
like
no.
This
is
a
bug.
Go
back
to
so
there's
a
lot
of
info
in
that
report
there,
but
I
think
what
this
is
about
is
basically,
when
I
do
a
cube,
C
cube
control,
D
each
entry
I
do
expect
that
everything
created
by
entry
should
be
removed,
and
that
includes
like
the
bridge,
the
obvious
bridge.
D
D
If
somebody
is
running
Andrea
and
God
forbid,
they
want
to
move
to
a
different
CNI
while
doing
the
delete,
Andrea
and
then
apply
my
next
C
and
I
dot
e
mo
basically
should
work,
and
I
think
if
we
don't
clean
up
all
the
entry,
our
resources
properties
and
there
may
be
issues
and
conflicts
with
the
next
CNI
right,
because
we're
gonna
have
that
gateway
address
there
with
an
IP
address
assigned
to
it
and
routes
assigned
to
it,
and
that
doesn't
sound
too
good.
I.
D
Difficulty
is
it's
it's
it's.
There
is
no
good
way
right
now
in
kubernetes
to
differentiate
between
I'm
deleting
all
my
entry,
our
resources
we
keep
control
to
eat
and,
for
my
pod,
just
crashed
and
or
or
my
pod
got
deleted
and
is
getting
recreated
as
part
of
a
rolling
update
of
my
demon
set
and
for
those
things
we
don't
want
to
clean
up
the
gateway
and
clean
up
the
bridge.
Only
when
you
delete
all
the
entry
and
stuff
you
want
to
clean
up
those
things
so.
B
B
A
B
D
Know
this
may
be
like
a
good
way
to
end
all,
especially
at
the
beginning,
when
we
do
releases
that
may
break
like
backward
compatibility.
If
you
want
to
apply
a
new
version
of
entry
and
something
is
broken,
it
would
be
great
to
be
able
to
delete
the
old
entry
or
in
a
way
and
make
sure
everything
is
being
cleaned
up
problem.
A
A
B
A
B
A
Good
make
sense,
and
so
let's
move
quickly
to
proposals,
even
because
we
are
almost
a
time,
so
we
currently
have
six
open
proposals
I'm
just
going
through
to
list
them.
One
is
about
ipv6
support,
including
dual
stack.
The
second
is
about
implementing
a
cookie
idea.
Locator.
Then
there
is
the
proposal
that
children
has
been
working
for
a
while
now
about
IPSec
tunnels.
Then
we
had
these
a
the
octant
integration,
which
has
been
also
been
Brian
Mandy
for
a
while
and
then
I
was
curious
about
support
kubernetes
network
policies.
E
A
B
D
D
B
A
D
D
So
if
one
fails,
none
of
the
changes
get
committed
as
kind
of
that
kind
of
makes
like
cleanup
I
mean
error,
handling
easier
in
the
open
flow
client,
because
you
don't
need
to
do
any
kind
of
rollback
if
one
of
the
operations
fail,
especially
in
the
network
policy
case
where
we
install
like
a
lot
of
flows
at
the
same
time,
kind
of
like
and
we'd
like
to
do
that.
You
know
that'll
make
fashion.
Okay,.
D
So,
okay
I
can
get
in
a
bit
of
details
here,
but
basically
to
run
entry
on
kind.
We
need
to
use
entry.
We
need
to
use
an
the
net
dev
data
path,
type
for
obvious.
That's
because
kind
means
like
basically
all
your
community's
nodes
or
container
running
on
the
server,
and
so
they
share
the
same
the
same
kernel
on
the
host
and
obviously
on
Mac
OS.
D
What
do
you
not
have
yes,
and
so,
when
you
run
obvious
space
mode,
it
doesn't
take
care
of
dating
the
checksum
for
you,
and
so
basically
that
means
that
you
need
to
disable
checksum
checks
off
load
on
every
V
interface,
not
the
one
attached
to
the
bridge,
but
the
other
side,
and
so
for
containers.
This
is
pretty
straightforward
because
for
pods
I
mean
because
we
are
the
one
setting
up
the
networking
4010
trigger.
D
So
basically,
what
I
do
is
if
the
data
passed,
ID,
snap,
labs
and
I
just
called
pistol
in
the
pod
and
I
disable
like
TX
off
load.
But
the
issue
is
that
you
also
need
to
do
that
on
the
V
interface
that
takes
care
of
the
node
networking.
So,
for
that
we
see
interface,
one
end
of
it
is
on
your
laptop
basically
and
the
other
one
is
in
the
node
namespace.
D
Otherwise,
all
the
packets
which
all
the
traffic
that
goes
out
of
the
node,
all
the
TCP
traffic
as
an
incorrect
checksum,
and
so
on.
Linux,
it's
pretty
straightforward
because
of
our
like
dr
container
networking
work.
You
basically
just
do.
The
same
thing
is
to
on
the
bees,
but
on
mac
OS.
There
is
no
such
mechanism
to
disable
like
to
extract
some
load
on
the
mac,
and
so
plus,
dr
dr
networking
works
completely
differently.
F
D
On
mac
OS,
basically,
we
don't
have
direct
access
to
the
to
the
container
network.
We
just
have
like
port
mapping.
Basically,
so
I
I
spent
a
bit
of
time
on
this,
but
a
mac,
OS
networking
expert
that
I
couldn't
find
a
good
way
of
resolving
kind
on
like
OS
with
entry
right
now,
a
lot
of
stuff
is
working
but
to
pay
traffic
node
to
node
traffic.
But
if
you
try
to
do
no
to
external
traffic,
basically
you
get
like
corrupt
corrupted,
TCP
checksums
and
everything
coming
through
the
laptop.
A
D
Used
to
feel
that
way
too,
but
I
realized
that
good
programming
of
Mac
OS
was
actually
pretty
good.
Okay,
especially
with
like
cross-compilation
support,
it's
pretty
much
like
a
first-class
citizen
Ingo.
So
that
makes
it
easier
and
I
mean
card
is
not
the
ideal
place,
I
think
to
like
debug
and
develop
like
a
CNI,
but
it's
kind
of,
like
gaining
momentum,
are
running
conformant
tests
for
kubernetes
on
kind,
so
a
lot
of
developers
like
a
pretty
fond
of
kind
in
the
kubernetes
year.
So
that's
why
we
started
like
supporting
it.
F
D
Yeah,
so
we
may,
and
basically
yeah
that's
a
good
point.
Korean
I
am
I,
didn't
I
mean
what
I
wanted
to
do.
Basically
is
I
wanted
to
make
SV
all
changes
to
the
entry.
I
could
buy
it
itself
to
enable
it
to
run
on
coins,
and
so
that's
why
we
were
using
the
OBS
user
space
data
path
and
that
itself
as
a
lot
of
complications.
Basically,
for
example,
for
turning
to
work
for
the
overlay
network
to
work.
D
You
need
to
create
a
second
bridge,
and
every
node
has
been
a
major
headache
just
to
make
it
work
on
Linux.
Now
it
works
pretty
well
Linux,
but
yeah.
If
we
want
to
make
it
work
on
Mac
OS,
there
is
a
big
issue
to
resolve
and
I
kind
of
put
it
on
its
stand
by
because
I
didn't
like
a
good
way
of
I
tried
a
few
things
in
my
quest
to
try
to
disable
tea
extracts
on
off
road,
but
there
is
no
connector
interface.