►
From YouTube: Antrea Community Meeting 01/15/2020
Description
Antrea Community Meeting, January 15th 2020
B
A
A
A
A
A
A
Some
of
them
are
brought
up
from
previous
releases,
so
let's
go
to
to
the
state
of
each
of
them.
Maybe
let's
start
with
a
CLI
I
mean
let's
process
them
in
order,
the
order
that
we
see
on
the
screen,
I,
don't
know.
What's
the
status
of
this
year
like
and
maybe
I
don't
know
if
internal
origins,
you
can
provide
an
update
on
this
topic.
B
So
yeah
I
think
the
PR
is
about
ready
to
be
merged.
I
know
that
Jen
approved
it
and
I
also
left
a
small
review
yesterday,
because
I
had
not
taken
a
look
at
it
for
a
while,
so
I
think
is
probably
gonna
be
merged,
like
hopefully
before
the
end
of
this
week.
So
definitely
in
time
for
awesome
cuz,
it
was
nothing
broken.
Just
minor
things.
C
C
Ipsec
the
call
is
merged
for
I
begin
to
so
far.
I
can
make
the
walk
with
GRE
follow
according
to
the
OS
folks,
it
should
work
with
any
tunnel
type
like
based
on
G
live
and
STP,
so
I'm
still
is
causing
with
the
answers
from
always
say,
he's
an
implementer
of
them
I,
just
like
a
ton
of
a
message
to
say
what's
issue
but
anyway,
I
think
if
we
can
just
make
you
worry
I'll,
just
talk
and
I
decided
only
about
young.
We
can
say
it's
supporting
the
other
side.
Okay,.
B
Yes,
so
I
open,
so
basic
in
that
issue,
III
describe
the
approach.
I
wanted
to
do,
and
I
also
open
the
pr2
to
start
implementing
that
approach.
It's
it's
not
complete
right
now,
so
as
a
basic
idea
is
that
we
have
a
new
llamo,
a
small
llamo,
manifest
file
for
cleanup
and
basically
you
apply
it
and
it
creates
a
demand
set
which
is
going
to
perform
the
necessary
clean
up
on
on.
A
A
B
Basically,
there
is
a
demand
side
where
there
is
also
like
kubernetes
job,
which
is
going
to
wait
for
the
different
agents
in
the
daemon
set
to
report
their
status,
and
so
it's
going
to
collect
errors,
and
it's
also
gonna
make
sure
that
everyone
has
a
chance
to
complete.
Ok,
job
is
gonna
return
and
then
yes,
you
have
to
delete
the
yamo,
because,
even
even
if
we
use
a
job
to
delete
the
demon
set,
which
is
possible,
you
still
have
to
delete
the
job
yeah.
A
A
D
A
D
A
And
I
think
why
I
mean
it
seems
like
the
problem
that
you
are
having
is
more
like
a
packaging.
Let's
see
problem
like
making
sure
that
it
works
in
the
container
became.
Perhaps
it
might
be
the
case
of
pushing
the
code
for
review
so
that
we
can
start
providing
feedback
if
you,
if
you
can
do
that,
yeah.
A
Okay,
so
let's
see
if
we
can
get,
maybe
if
we
can
get
the
code
up
for
review
in
a
day,
maybe
we
can
still
keep
it
targeted.
Otherwise,
I
think
that
nobody
will
die
if
we
move
this
to
the
next
release,
then
support
for
knowing
cup
mode
support
for
knowing
cup
mode
I
think
that
these
at
this
stage
is
still
just
a
pro,
is
still
just
a
proposal
right.
We
don't
even
have
POC
code.
Formal
no.30
are
open.
C
A
C
C
A
C
A
A
A
good
point
all
right,
okay,
so
maybe
we
want
to
followup
with
the
PR
Otto.
Regarding
these
are
this
work
item?
Aye-Aye
I
mean
it's
very
high
level.
It
doesn't
feel
like
it's
in
a
stage
so
that
we
can
think
about
including
it
internal
injuries,
even
with
an
alpha
state.
But
I
would
like
closer
to
understand
what
is
a
European
on
it.
So
shall
we
wait.
A
B
B
A
B
I
did
this
like
right
now,
because
I
realized
that
actually
what
we
do
for
Linux
support
also
works
for
Mac
support.
It
wasn't
gonna
work,
so
I
didn't
bother.
Trying
and
I
did
some
more
investigation
when
I
had
some
time
yesterday
and
I
realized
that
it
actually
works
the
same
because
on
Mac
OS
with
docker
desktop
every
container
actually
runs.
You
know
Linux
hyper
kefir,
so
basically
what
we
do
for
Linux.
B
A
A
C
A
B
A
D
B
A
A
C
Sorry
for
the
basically
what
we're
hoping
that,
if
some
chronic
says
so,
is
giving
no
the
port
with
our
continent
mutation
and
I'm
similar
with
our
economy.
If
the,
if
the
service
is,
if
the
packet
called
understand
note,
lots
of
policy
will
not
take
effect,
I
mean
so
we
were
always
thought
knowledge,
traffic.
A
C
Think
the
rest
at
first,
if
for
I
mean
a
could
be
particular
scenario.
First
before
I,
the
policy
follows
in
local
research,
so
we
just
said
el34
the
announcing
that
I
believe
this
behavior
is
not
really
clear
that
if
an
incubus,
so
yeah
I
think
there
are
different
implementations
is
something
we
are
kind
of
thinking
that
maybe
better
way
to
open
a
psychic
I
mean
unless
you
have
a
policy
to
allow
the.
A
Right
that
make
sense:
okay,
we'll
probably
wait
for
this
one,
we'll
wait
also
for
some
feedback
from
city
and
city
Network
to
understand
what
should
be
our
behavior
in
this
case
all
right.
So
then
we
have
a
really
quite
a
lot
of
wishes.
Sylvan,
honestly,
I,
don't
remember
of
any
issue.
That
is
a
critical
enough.
That
really
must
be
included
in
0.30,
but
in.
A
I
mean
if
you
ever,
if
we
have
open
PR,
said
that
they
are
very
good
chances
to
marry
me.
Yes,
then,
let's
include
them,
speaking
of
which
maybe
in
order
to
define
which
issues
we
might
include
instead
of
looking
at
the
issue
list,
it
might
be
better
look
at
the
PR
list
and
ever
look
at
the
open,
PRS
and
see
which
one
are
mature
enough
to
be
considered
before
for
marriage.
So
the
one
that
you
were
just
mentioning
by
Shuang
is
these
one
about
improving
your
business.
So
this
and
that
process
don't
think
so.
A
One
the
deletion
process
perfect,
so
we
can.
We
can
merge
this
one
and
fix
the
issue,
and
then
we
have
documentation.
This
one
is
better
documentation,
you
think
kind
in
the
CI
which
that
doesn't
have
to
be
tied
to
any
milestone,
using
a
consistent
mechanism
to
skip
tests
on
the
kinda
cluster.
The
same
thing
delete
still
to
offset
flows
on
agent
restart.
We
already
discussed
about
it.
A
A
B
Like
I
think
improve
robustness
of
the
open
flow
client
because,
especially
for
medical
policies
like
one
update
from
the
controller,
can
map
to
multiple
flow
updates.
I
think
currently
we'll
try
to
do
a
rollback
in
case
of
a
flow
at
failure,
which
is
a
bit
clunky
when
we
do
bundle
because
of
the
transactional
semantics.
It's
actually
much
easier
to
do.
Error.
D
A
A
Good
and
finally,
the
antis
ETL
framework,
we
already
discussed
this
issue
and
I
guess
that
now
we
can
now
this
completes
the
list
so
I
think
now
we
have
a
clear
picture
for
what
goes
into
0.33
or
3-0.
Let's
say
that
the
two
open
items,
the
open
items,
are
what
we
do
about
Prometheus
and
the
other
one
is
about
no
in
kappa
mode,
so
far,
I
believe
we
are
leaning
to
move
them
out
of
the
released
for
both
of
them.
Probably
he
would
have
liked
tests
in
place
for
doing
a
move
yeah
even
from
Prometheus.
A
We
may
want
to
have
some
sort
of
testing
integrity
in
the
CI
right,
even
though
Prometheus
Prometheus
is
something
which
is
kind
of
orthogonal
to
Andrea,
so
you
can
enable
it
or
not.
So
it's
not
like
a
part
of
the
core
components
you
can
you.
Can
we
I
think
that
the
agent
that
Cobb
is
implementing
and
correct
me
if
I'm,
wrong
Kobe
is
being
done
in
a
way
that
you
can
run
it
not
run
it?
It's
the
users
choice.
Is
that
correct,
yeah.
C
A
A
The
only
additional
proposal
that
I
wanted
to
bring
up
for
possibly
a
future
release
is
to
run
Andrea
as
a
networker
follow
my
shift
for.
In
order
to
do
so,
we
will
need
to
develop
some
additional
components
that
interact
with
an
overshift
API
server.
They
have
a
fairly
complex
mechanism
that
in
some
ways
it
resembles
a
bit
Cloud
Foundry.
A
We
want
basically
means
that
you
start
the
cluster
and
you
get
hundreds
of
pots
that
you
don't
know
what
they
are
doing
pretty
much
like
when
how
the
foundry,
where
you
have
all
those
VMs,
that
it
takes
quite
some
time
to
figure
out
what
they
are
doing.
What
are
the
challenges
there?
The
challenge?
The
main
challenge
is
that
the
image
the
busy
major
cannot
be
Ubuntu.
A
Basically,
if
we
want
to
certify
Andrea
with
reddit
to
the
base,
image
has
to
be
either
red
hat
enterprise
linux
or
what
they
call.
It's
basically
her
OS,
but
they
have
a
different
name
for
it.
Like
unbreakable
filesystem
modifiable
faces
immutable
filesystem,
something
like
that.
So
we
will
need
a
different
base
image.
A
We
can
proceed
with
the
deployment,
let's
say,
from
a
strictly
open
source
perspective.
I,
don't
see
these
as
a
very
useful
thing,
meaning
that
if
we
want
to
focus
extremely
exclusively
on
the
kubernetes
upstream
community,
then
openshift
is
a
little
bit
out
of
date.
Let's
say
it's:
it's
it's
a
sort
of
a
different
ecosystem.
If
you
want
in
terms
of
adoption,
however,
a
it
could
be
relevant
in
terms
of
user
adoption,
because
so
far
OpenShift
has
only
used
its
own
Sdn,
which
is
still
open
with
switch
based.
A
With
calico
openshift
as
two
flavours,
the
enterprise,
which
is
a
properly
called
open,
shift
compute
platform,
which
is
a
paid-for
product,
and
then
there
is
ok
D,
which
is
a
the
open
source
version
summarizing.
This
is
just
a
very
high
level
of
what
what
we
can
do.
I
am,
if
you
think
it's
something
viable
at
least
to
be
tracked.
I
can
create
an
issue
for
it
with
a
list
of
items
that
need
to
be
done,
and
then
we
can
target
it
to
a
future
release.
C
A
The
thing
is
that
if
we
want
to
go
in
also
allow
enterprise
customers
to
run
on
tria,
then
we
will
need
to
talk
to
read
that,
because
there
is
a
certification
process
that
we
need
to
fall
off.
But
if
our
target
is
the
open
source
version,
you
know
it
used
to
be
called
OpenShift
origin.
Now
it's
called
ok
d.
If
it's
just
that,
I
believe
that
we
can
develop
an
integration
without
having
to
without
having
to
do
a
process
with
Reddit.
C
A
A
If
you
have
a
network
type
that
it's
not
recognized
by
their
built-in
operator,
then
what
happens
is
that
it's
up
to
you
to
deploy
your
networking
system,
but
ovn
and
overshift
Sdn
are
Big
D
in
the
system
I've
in
theory,
since
this
code
is
open
source,
it's
not
like
private
code.
We
can
even
think
of
pushing
for
supporting
it
out
supporting
Andrea,
but
those
this
code
is
in
repositories
that
are
controlled
by
reddit
there
are
there
open
source
but
controlled
Meredith.
So
in
that
case
it's
a
conversation
that
will
take
more
time
and
we.
E
Need
to
also
to
figure
out
the
differentiator
of
antair
compared
to
okay,
nor
rapid
shift
Sdn
in
the
in
OpenShift
environment
right
so
I,
in
order
to
set
some
kind
of
priority
on
this
right,
so
otherwise
why
people
would
go
to
an
Trier.
So
we
need
to
figure
out
why
why
someone
would
replace
the
default
OpenShift
s
the
enemy
country?
Well,.
A
A
E
A
Basically,
that
that's!
Why
that's?
Why
I
don't
want
to
go
into
starting
a
conversation
with
Reddit,
because
I
know
that
that's
going
to
be
time-consuming
and
I
know
that
there
will
be
questions
that
we
will
need
to
answer
like
if
people
can
use
ovn
and
not
really
okay
shift
at
the
end,
because
overshift
Sdn
versus
ovn
is
pretty
much
clear,
but
why
they
will
need
that
they
would
like
to
put
Andrea
on
the
same
level
as
ovn.
A
It
is
a
discussion
I
probably
go
on
for
years,
so
I
was
thinking
of
doing
the
integration
like
as
a
third
party
network
operator.
Okay,
so
basically
not
integrated
the
in
the
open
deployment
process,
but
Donna
in
the
same
way
as
the
third
parties
are
done
at
the
moment,
like
a
calico
for
instance,
and
I
believe
that
there
is
a
way
that
I
believe
that
even
see
Liam
can
do
that.
So,
basically
you
deploy
you,
you
tell
openshift,
hey
I'm,
using
psyllium
as
a
networking,
so
this
do
be
shift.
A
Network
operator
does
nothing
and
see
Liam
itself
as
some
component
for
deploying
for
deploying
itself
in
Europe
a
shift
cluster
and
the
same
thing
for
calico
and
initially,
if
you
remember
in
open
in
the
earliest
releases
overall,
my
shift
for
ovn.
It
was
the
same
process
then
now
they
made
it
diminutive
built
in
innovation,.
A
Anyway,
I,
don't
think
that
this
is
something
that
we
want
to
do
with
a
high
priority.
It
is
something
that
we
might
need
to
consider
and
everything
they're
under
on
the
roadmap.
Just
in
case
when
we
start
getting
feedback
from
the
user
community
we'll
get
the
Audrey
can
say.
I
want
to
run
on
three
on
open
shift
for
what
what's
your
strategy
for
that
yeah.
E
B
A
Will
take
care
of
that?
Yes,
probably
tonight
and
then
regarding
existing
issues.
I
have
been
in
the
past
weeks.
I've
did
some
work
on
the
dual
stock
front.
So
basically
we
need
to
redo
large
chunks
of
our
item
code,
because
I
realized
that
our
item
code,
at
the
moment
it
assumes
a
single
ipv4
address,
so
I
was
starting
working
on
that.
A
But,
unfortunately,
life
is
tough
and
then
I
got
caught
with
other
issues,
but
as
soon
as
I
have
some
time
some
spare
time
I
will
I
will
try
and
resume
work
on
that
once
we
address
the
item,
part
I
think
it
would
be.
The
next
step
will
be
also
to
test
to
do
some
changes
in
the
cni
client
to
make
sure
that
it
can
return
both
the
v4
and
v6
address,
II
actually
I
believe
there
should
not
be
this
change.
A
Basically,
we
need
just
need
to
mean
to
test
things
and
to
end
and
ask
for
network
policies,
it's
very
likely
that
they
will
need
also
some
changes
as
well
to
create
the
flow
move
using
the
v4
addresses
and
the
v6
addresses,
and
we
will
need
changes.
Also
on
the
sorry
I
was
saying
we
will
need
changes
in
the
way
we
deploy
a
cube
proxy
because
it
has
to
be
an
IP
vs
mode.
A
Even
though,
probably
for
the
next
kubernetes
release,
we
will
be
probably
able
to
use
a
proxy
in
activist
mode
also
for
dual
stack,
but
this
has
not
yet
been
done
right.
So
this
is
a
my
update
on
the
ipv6
last
up
front
and
say
that
I
will
shut
up
and
not
talk
until
the
end
of
the
meeting,
or
at
least
our
try
to.
A
A
F
Recognize
me
ok,
this
is
we
talk
about
a
lot
about
the
right,
hash,
stuff,
OpenShift
and
also
in
my
bio.
We
are
also
in
serious
consideration
with
the
entry
into
our
product.
So
one
of
the
product
we
are
thinking
about
it
on
the
on
the
PKS
and
the
other
is
I,
guess,
class
or
probably
know
on
the
I'm
just
wondering
do
we
have
some
document
which
compares
different
singing
I
like
current
to
the
entry
or
what's
the
functionality
it
was
country
provide
compared
to
others
in
eyes
like
calico,
because
people
talk
about
calico.
F
C
C
B
A
reminder
that
this
is
a
community
meeting
and
that
we
record
this
and
we
publish
it
on
YouTube,
okay,
but
still
a
relevant
question.
I
think
we've
been
running
some
scale
tests
internally
in
VMware.
That's
probably
some
results.
We
could
share
with
the
community
at
a
later
date
and
yeah
sure
I,
don't.
A
Yeah,
that's
right!
If
you,
if
we
refer
to
tools
that
people
know,
then
won't
be
able
to
access,
but
you
know
these
numbers,
it
might
be
good
and
once
we
polish
them
and
we
validate
that,
we
computed
these
numbers
with
the
correct
procedure.
It
might
be
nice
to
to
share
them
also
upstream,
not
in
a
form
of
comparison.
I,
don't
like
doing
that
and
I
don't
think
they
are
extremely
meaningful,
but
at
least
in
terms
of
absolute
numbers
referring
to
Andrea.
That
would
be
yeah.
A
It's
just
that
I
don't
want
to
just
take
them
and
share
them,
because
you
know
when
it
comes
to
scale
testing
the
process,
the
design
of
experiment,
the
metrics
that
get
collected
are
also
extremely
important.
So
first
we
probably
want
to
review
how
we
gather
this
data
and
then
once
we
assess
that
the
data
that
we
are
going
to
share
can
be
trusted.
We'll
do
that.
C
B
B
Not
the
octants
report,
maybe
because
I
know
that
mainly
did
a
lot
of
work
on
this
and
we're
publishing
an
image
now,
a
darker
image,
which
includes
a
öktem
distribution
and
our
plugin.
So
I
don't
know
if
you
want
to
consider
promoting
this
to
better
status
or
if
you
want
to
wait
for
the
next
race
to
see.
If
there
is
an
issue
with
that.
B
A
All
right
sounds
good
10,
so
we
can
stop
recording
here
thanks
thanks
for
coming
to
this
meeting,
and
oh
just
before
we
close
the
meeting.
So
what
should
be
our
agenda
moving
forward?
I
guess.
The
topic
that
was
in
agenda
for
today
should
get
pushed
to
next
week.
So
we're
going
to
have
next
week's
meeting
right.