►
From YouTube: Antrea Community Meeting 08/02/2021
Description
Antrea Community Meeting, August 2nd 2021
A
And
yeah:
please
go
ahead.
A
B
B
Yeah
hi
everyone
and
my
name
is
rocha,
and
today
we
and
I
will
give
you
a
presentation
with
our
updates
for
design
multi-car
support
on
entry.
B
So,
after
last
presentation
we
have
some
new
thoughts
and
more
detailed
discussion,
and
so
I
think
it's
necessary
to
have
a
new
one
and
we'll
talk
this
time.
We
talk
design
in
more
detail.
Okay,
let's
start
so
again,
we
want
to
add
a
new
feature:
gate
multicast,
2
and
default
will
be
for
us.
B
You
can
see
this
crd
in
the
screen
and
user
can
define
the
multicast
group
addresses
which
are
expected
to
be
supporting
a
cluster
and
they
use
a
part
selector
to
set
target
pods
which
are
expected
to
access
the
multicast
address
and
multicast
group
defining
the
cid
could
be
ipv4
ipv6
and
if
it's
ipv6
address,
the
video
gates
for
joe's
stack
should
be
named
for
the
cluster,
but
we
are
not
planning
to
support
ipv4
in
the
first
version.
Maybe
we'll
talk
this
later
and
here
I
want
to
explain.
B
External
interfaces
in
crd
is
used
to
tell
entry
with
which
interface,
the
external
multicast,
client
or
source
will
communicate
with
local
parts,
and
local
here
is
used
to
decide
if
s
naught
is
performed
on
the
power
to
external
multicast
traffic.
B
B
Both
and
and
entry
engine
and
ng
controller
will
watch
multicast
crd
and
do
some
configuration
and
we'll
expand
them
in
detail
later.
B
So
for
the
entry
controller
part,
it's
really
simple:
let's
watch
the
multicast
crd
in
the
computer.
Switch
parts
are
involving
each
multi-pass
group
entry,
controller
leverage,
pass
letter
to
step
parts
and
write
a
part
information
back
to
cit
and
organize
the
result
by
nodes.
So
the
green
part
I
highlight
are
the
information
that
is
written
by
energy
controller.
B
B
B
Entry
engine
as
open
flow
entry
entries
on
the
on
the
current
node,
as
you
can
see
in
the
green
lines,
so
we
will
use
register
0
bit
index
27
here
to
identify
multicast
traffic
and
which
is
set
in
the
air
screen
routing
table
and
it
is
matched
in
the
l2
calculating
table.
B
So
ttl
is
decreased
from
multicast
traffic
in
the
ttl
degree,
decreasing
table
number
172,
please
number
172
and
note.
We
want
to
add
a
new
table
that
is
number
111
for
multicast
traffic
forwarding
and.
B
The
matching
this
table
is
multicast
group
config.
In
crd.
The
flow
action
in
this
table
is
a
group
bucket
in
the
group.
A
bucket
in
the
group
includes
three
parts
and
you
can
see
the
last
line
and
the
first
is
up
to
package
to
end
and
three
gateway:
zero
and
the
support
for
forwarding
packets
to
make
some
external
and,
secondly,
you
can
see
at
a
transaction
normal
and
that
is
to
leverage
only
asthmatical
smoking,
snow
pin
capability
and
for
the
pocket
inside
ovs.
B
Bridge
third
is
set
kernel
and
output
to
internal
port
ip
and
for
income
mode
for
no
income
mode.
No
additional
pipeline
is
needed
and
then
you
can
see
the
congestion
production
action
in
the
table.
111,
and
you
can
see
three
eyepiece
here.
B
zero
is
all
hosts
and
you
can
see
dot.
Two
is
operators
and
22
is
igp
version.
3
multicast
group
they
are
allowed
by
default
in
the
packet
packets
are
forward
to
longworth.
It
is
or
not
other
conflict.
Multicast
group
traffic
will
be
dropped
in
and.
B
According
to
the
crd
configuration
and
the
following
is
the
example
for
multicast
routine
entry
consequent
node
is
that
in
this
example,
ens
192
is
configured
as
external
interface
in
the
crd,
and
this
is
the
output
of
ip
multicast
command
and.
B
Second
event:
one
we
want
to
take
care
of
is
the
window
when
new
nodes,
where
parts
besides
our
join
and
existing
multicast
group
and
entry
engine,
has
different
progress
for
the
in-cab
mode
and
annoying
cam
mode
and.
B
B
And
for
the
no
in
camp
mode
entry
engine
as
multicast
routing
entries
on
the
host
to
forward
the
traffic
between
entry
gateway,
0
and
the
transport
interface,
which
is
used
for
routing
traffic
across
nodes,
the
multicast
routing
entry
is
configured
once
when
entry
engine
find
a
different
node
joins
to
the
multicast
group.
The
first
time.
B
Entry
process,
the
multicast
floating
entry
for
ports
on
entry
process
and
multicultural
entry
for
ports
are
different.
Node.
I've
seen
that
it
does
under
external
addresses.
Here
you
can
see
that
ens-192
is
the
external
x-ray
interface
and
the
ens
224
is
the
node
transport
interface.
B
So
no
it's
not.
We
support
no,
it's
not
as
described
before
in
the
local
field
in
the
crd,
and
it
also
has
an
impact
on
the
poor
traffic
across
node
in
the
knowing
cam
mode
entry
and
remote
parts,
ip
as
multicast
source.
Instead
of
remote
nodes,
ip.
B
We
have
some
questions
we
had
for
pre
previous
meeting,
and
most
of
them
can
be
answered
as
the
topic
I
mentioned
before,
and
there
are
some
discussion
before
about
ignp
proxy
and
market
and
we
won't
use
them
and.
B
Because
ijmp
proxy
only
supports
igp
v1,
which
one
downstream
interfaces
and
the
rgmp
proxy
adds
multicast.
Not
only
if
igmp
membership
report
is
downstream
interfaces
and
ignore
the
upstream
interfaces,
so
this
might
cause
local
multicast
traffic,
my
able
to
center
external,
which
we
don't
want
to
also
with
we-
are
not
using
dynamic,
multicast
water
demon,
I'm
rotating
on
pimd,
and
we
want
to
try
to
config
static,
multicast
routing
table
using
c
go,
and
for
do
we
support
multicast
four
to
external
cancer.
B
B
And
the
intention,
with
the
configuration
we
mentioned
before,
and
the
possibility
to
use
obs
only-
and
I
don't
know
exactly
what
the
question,
but
basically
we
implement
multicast
some
entry
by
adding
open
flow
and
entries
and
the
on
the
lvs
bridge
and
the
config
multicast
routing
entries
using
unicode
can
the
multicast
source
ipb
port
ip
when
there
is
no?
B
No,
not
the
answer
is
simply
simply
yes
as
we,
as
we
mentioned
the
before
in
the
knowing
cambod,
for
the
s.
No,
for
the
no.
B
No
s
not
support,
so
here's
the
answers,
and
next
we
are
waiting
to
have
a
further
discussion
on
this
approach.
We
have
discussed.
C
Hello,
I
think
I
want
to
restore
some
limitations
and
other
issues
about
the
design
and
I
think
we
could
start
the
discussion
about
the
design
details.
C
C
C
We
have
enforce
the
multicast
traffic
to
go
through
the
network
policy
related
tables,
but
I
don't
think
we
have
enough
policy
rules
focusing
on
mud
catch
traffic,
so
I
just
mean
for
the
first
release:
mudcast
might
not
be
limited
by
network
policy,
but
if
or
not
so
we
want
to
support
network
policy
and
multicast
traffic.
I
think
it
is
an
open
issue
and
the
conclusion
might
have
some
impact
on
the
table
designed
for
multicast
and
the
third
is
for
the
3d
design.
C
We
want
it
to
use
the
crd
to
tell
and
share
what
medicare
group
where
we
are
wanting
to
support
and
what
pulse
are
joining
the
medical
group.
Some
limitations
is
that
the
part
in
the
mathcad
group
is
now.
It
is
not
auto-discovery
but
frequent
configurable.
C
So
I
think
we
might
introduce
the
pulse
in
multicast
group
also
discovering
in
future
versions,
but
no
for
the
first
version.
So
I
still
think
the
therapy
to
tell
the
pulse
lecture
is,
is
necessary
in
in
distribution
and
for
the
ipv6
multicast.
C
C
Maybe
the
second
world
second
release
format
cast,
but
but
absolutely
not
in
the
first
release,
and
these
are
the
limitations
the
way
we
know
we
already
know
about
the
design,
and
then
I
want
to
have
some
discussions
about
about
all
the
issues.
Firstly,
is
about
crd
way.
We
want
to
know
if
namespace
selector
is
required
to
use
to
to
be
used
to
select
the
poles
in
the
zld.
C
We
want
to
know
your
suggestions
on
our
thoughts
about
it.
Currently
we
didn't
add
it
in
the
study
design
but
modify
if
it
is
required.
A
Hey
waiting,
I
had
a
question
kind
of
regarding
pot
selection.
Have
we
considered
using
like
a
pod
annotation
instead
to
have
a
pod,
join
a
multicast
group?
I'm
asking
because
let's
say
I'm
someone
scheduling
pods
in
a
cluster.
Maybe
I
maybe
I
know
that
my
pods
need
to
join
a
specific
multicast
group
and
it's
easy
for
me
to
add
an
annotation
to
the
pod
to
let
them
join
the
multicast
group.
But
if
I
have
to
go,
find
the
crd
and
modify
it.
That
means
I
need
to
have
like
permissions
to
edit
the
crd.
C
But
if
we
use
the
annotation
to
support,
I
I
think
for
annotation.
It's
true
should
be
ready
to
support
the
policy
lecture,
but
the
visuals
possible,
annotation
and
necessity.
I
think
the
it
will
introduce
more
work
for
the
multicast
implementation,
so.
A
Anyway,
right
because
you
need
to
configure
other
things
like
local
and
and
the
external
interfaces,
and
so
on,
but
yeah
I
I
I'm,
I'm
not
totally
sure
I
feel
like
if
some
multicast
groups
have
been
pre-configured
and
we
want
to
let
people
scheduling,
pods
like
join
the
multicast
groups
easily,
then
an
annotation
may
be
easier
but
yeah,
but
I
can
keep
thinking
about
it.
C
So
the
past
lecture
should
is
possible
to
remove
from
the
30
and
let
the
user
to
use
amputation
to
allow
his
paws
to
join
medical
schools.
Right
am
I
right.
C
Okay,
I
think
we
could
have
some
thinking
about
it.
Maybe
you
could
support
it
in
the
in
the
version.
D
D
E
Was
also
thinking,
annotation
probably
sounds
better
to
me
for
the
reason.
E
E
Sorry
allow
source
code
to
discover
a
receiver,
because
I
I
I
remember
you
mentioned
that
ign
is
looping.
So
if
you
already
enabled
looting
yeah.
C
Yeah,
if
we
already
enable
the
multicast
snooping
now,
whereas
the
port
ip
should
be
added
to
the
os
multicast
db
cache,
but
the
difficulty
is
how
shall
they
read
the
db
cache
from
less
maybe
using
the
os
mount
or
something
like
that?
C
E
E
C
I
just
think
it's
quite
changes.
Much
maybe
only
on
the
positive
gallery.
A
Or
in
that
case,
annotation
makes
more
sense
right,
because
it's
not
as
big
of
a
change
in
a
way.
You
don't
need
to
change
the
crd
and
conceptually
it's
kind
of
like
more
similar
right,
because
it's
the
point
but
joining
the
group
either
through
an
annotation
or
by
sending
a
an
igmp
packet
that
you
can
capture
at
the
at
the
switch.
C
C
A
No,
I
mean
for
the
crd
status,
you
can
have
the
entry,
a
controller
watch,
pods,
right
and
yeah
and
based
if
a
pod
has
a
multicast
group
annotation,
you
can
just
update
the
crd
based
on
that.
Maybe.
C
Anyway,
if
agent
has
has
worked
to
discover
the
the
the
power
default
multicultural,
no
matter
representation
or
less
mud
has
to
be
catch
and
creation
should
run
back
to
30
or
maybe
then
controllers
don't
need
to
write,
write
anything
to
do
this
rv.
If
agent
has
taken
over
the
job.
E
Yeah,
how
about
you
think
a
little
more
about
it?
I
mean
why
the
it's
it's
it's!
It's
significant
piece
of
theory
work
to
do
allotation
or
crd.
First,
if
it's
really,
I
mean
lots
of
user
will
work.
I
have
to
say,
let's
consider
how
to
discovery
from
the
beginning,
but
I
would
like
to
you
you
wanna
another
one
that
I
put
anything.
We
should
differentiate
the
debugging
information
and
the
the
the
involving
information
required
by
functionality.
E
So
if
you
are
saying
the
port
number
only
for
debunkability-
and
I
possibly
would
ask
why
why
you
must
do
that
in
the
first
version,
could
we
start
from
node
member?
If
that
is
more
efficient,
it
can
scale
better.
E
C
Since
I
think,
if
we
want
to
support
no
last
night
on
dramatically,
traffic
total
drives
should
be
included
in
the
street
status.
Otherwise
we
can't
add
it
to
the
medical
surrounding
entries
from
remote,
poles.
C
C
D
E
I
I
know,
but
I'm
I'm
thinking,
should
we
even
do
snap
or
we
should
never
do
stand
for
markers.
E
I'm
not
very
certain,
I
I
just
never
know
people
do
as
type
4
as
molecular
traffic.
Maybe
I'm
wrong
how
about
let's,
let's
double
check,
we
can
also
share
with
you
to
see
if
we
see
any
standard
requirement.
C
Okay,
so
if
no,
if
the
local
calculation
is
not
required,
I
think
they
could
don't
include
the
part
held
details
in
the
crd
standards.
We
could
only
include
the
node
node
information.
E
Sure
again,
I'll,
probably
let's
share
with
you
some
if
you
know
any
information
that
any
other
solution
for
style
for
americans.
My
counselor
pick
up
also
check.
E
Okay,
I
just
don't
know,
and
I
think
the
the
more
important
you
use
case
is
low
income
mode
right.
That's
what
I
have.
C
E
Way,
it's
no
income
or
why
you
do
that's
that
yeah,
okay
and
then
you
know.
D
E
E
A
E
E
C
A
C
E
I
don't
know
I
mean
if
you
ask
me,
of
course,
you
you
shouldn't,
make
a
decision
since
you
are
into
business.
But
if
you
ask
me,
I
have
to
do
the
income
first,
since
it's
because
it's
simplest
and
don't
require
many
chains
in
forwarding
pipeline,
and
it's
also
the
only
required
use
case
winner
and
then
we
can
do
in
calculator
he
first
met.
Then
I
probably
were
thinking
about
how
to
discovery.
C
Than
all
also
my
second
question
about
enough
policy
and
the
medical
traffic,
I
want
to
know
your
thoughts
about.
Is
the
requirements
existing
right?
So
we
might
have
some
policy
rules
to
allow
our
jobs.
Medical
strikes
now
right.
So.
C
Okay,
I
mean
for
network
policy
and
multicast
perfect.
E
C
E
D
Correct:
okay,.
C
E
E
For
example,
we
don't
allow
port
one
to
put
two
port
two
traffic,
but
you
have
a
smaller
car
traffic
from
port
one
to
port
two
and
you
want
to
drop
the
traffic.
I
think.
C
Okay,
then,
anyway,
if
enough
policy
is
possible
to
applied
automatically
traffic,
so
I
think
we
still
require
the
medical
traffic
to
follow
our
current
pipeline.
I
mean
the
package
through
people
should
be
enforcing
to
enter
the
inquiry
egress
tabling
and
the
ingress
tables.
E
Okay
yeah,
I
think.
Ideally
we
should
call
molecules,
but
if
you're
saying
it's
challenging
for
the
first
question,
maybe
we
can
do
later
just
my
poisonous.
C
Yeah,
okay
and
then
we
will
put
the
priority
for
that
policy
with
p
then
or
p2.
I
mean
the
priority
for
the
policy
on
my
cat.
E
C
Oh
disgusting
and
I
think
I
don't
have
other
alternatives
to
discuss.
I
want
to
to
share
another
point
about
the
madquest
resulting
entry
configuration.
I
remember
that
in
the
original
stride
genji
has
asked
us
if
we
could
support
the
any
thoughts
in
the
multicast.country
in
that
email.
I
have.
I
have
written
backsides
for
the
zero
address
is
positive
by
smc
route,
but
when
looking
to
the
linux
relating
multicast
routing
implementations,
I
found
that
we
can't
use
the
zero
address
to
use
that
anything.
C
We
have
to
add
the
routing
entry
with
every
single
source
of
device.
C
So
for
the
intro
asian
implementation,
we
will
use
the
segal
language
to
signal
to
to
watch
kernel.
Upcom
message:
I
mean
the
agmp
no
catch
which
will
be
sent
out
from
the
link
kernel
when
the
multicast
packet
is
not.
He
is
not
hitting
any
existing
routing
entry.
C
The
message
will
be
sent
out
to
the
demon.
The
entire
agent
will
catch
the
message
and
read
the
source
and
the
destination
of
the
ipo
dresses
and
then
add
the
multicast
documentary
into
the
writing
table.
E
I
mean
sounds
like
leela's
colonel
want
to
drop
the
package
for
the
ultimate
either
entry
for
the
unknown.
C
I
don't
think
the
linux
one
last
turn
around
to
drop
it.
It
is
because
we
don't
have
any
added
any
unknown
external
address,
but
they
don't
know
the
external
addresses
from
the
configurations.
So
we
have
to
let
the
link
kernel
to
use
up
call
messages
for
using
upload
messages
to
us.
Two
letters
and
extra
add
additional
login
entries.
D
Yeah
I
mean
it's,
it's.
C
Yeah,
it
is
different
from
the
personal
dynamical
learning
it
is
from
the
kernel.
C
I
think
other
tools
is
using
the
practical
learning
and
they're
sure
to
show
learning
dramatic
identity
message,
team
message
of
some
other
massacre,
related
protocol
messages
and
learn
the
source
and
the
destination,
and
then
the
dotting
entry.
I
think
the
final
action
is
is
the
same.
The
difference
is
a
very
get
on
the
how
they
get
the
message.
E
Yeah,
I
think
the
the
problem
that
we
don't
move
the
federal
interface
to
the
os
bridge
because
for
ignps
bridge
I
don't
think
you
need
to
your
sound
source
right
breach,
don't
counsel
you
just
manage.
E
Okay,
I
don't
know
if
that
way
works,
maybe
so.
E
Okay
orange,
I
thought
in
the
worst
case
we
can
use
tc
to
redirect
traffic,
but
then.
A
C
We
originally
want
to
use
the
crd
to
help
the
multicast
group
address
and
the
external
addresses
to
communicate
with
the
external
multicast
all
clients,
and
maybe
no
no.
Maybe
local
is
now
on.
Only
the
two.
C
There's
a
group,
then
the
theory
is
only
for
the
medical
school.
E
Okay,
again,
I
don't
know
if
you
want
to
do
annotation,
I
still
I'm
still
thinking.
Probably
you
can
still
do
one
allotation
with
multiple
groups,
but
I
was
still
thinking
if
we
do
auto
discovery,
we
don't
need
any
of
these
right.
C
For
other
discovery,
actually
is
because
we
they
don't
have
the
update
interface
from
os
about
multicast
db
cache
showing
how
to
loop
and
query.
We
require
the
multicast
db
cache
in
your
loop
then.
C
Actually
it
is
what
I
don't
want
to
use.
So
I
want
to
will
eventually
leave
some
time
for
os
to
have
to
provide
some
more
generic
functions
to
get
the
implementation.
E
Okay,
I'm
not
really,
I
I
don't
know
how
it's
handled
much,
but
maybe
you
already
did
something
asking.
Maybe
let
me
also
share
with
other
guys
to
say:
okay,
father,
if
the
first
version
is
no
income
only
again,
you
don't
need
any
of
this.
C
E
A
Thanks
for
the
presentation
watching
winning
and
roshan
and
yeah
for
the
discussion,
please
send
me
the
pdf
of
the
slides,
so
I
can
put
it
on
the
after
andrea.io
meeting.
A
A
All
right
to
quote
salvatore
going
once
going
twice
going:
thrice
and
yeah
thanks
guys,
thanks
again
for
the
presentation,
winning
and
russian
and
I'll
see
everyone
in
two
weeks.