►
From YouTube: Antrea Community Meeting 02/01/2021
Description
Antrea Community Meeting, February 1st 2021
A
Doing
now,
so
welcome
to
this
instance
of
the
andrea
community
meeting
today
is
tuesday
february
the
2nd
or
monday
february
1st,
depending
on
your
location
and
for
today
on
the
agenda.
We
have
a
few
topics
that
we'll
have
a
quick
announcement
about
arm
from
antonin
and
then
ray
will
present
about
a
generic
operator
design.
B
Go
let
me
share
my
screen:
it's
not
going
to
take
long,
just
a
couple
minutes
all
right.
So
can
you
guys
see
my
brother?
B
Yes,
yeah,
perfect?
Okay,
so
we
opened
an
issue.
I
mean
someone
opened
an
issue
like
almost
a
year
ago
to
have
support
for
arm
platforms
in
andrea.
So
if
your
cluster
includes
arm
arm
nodes
and
so
for
a
while,
we've
been
postponing
adding
this
support
and
that's
mostly
because
it
was
a
bit
challenging
not
necessarily
to
build
images
for
arm
whether
it's
arm
64-bit
or
arm
32-bit,
because
docker
has
support
for
a
qmu
emulation
which
lets
you
build
images
for
those
architectures.
B
But
I
think
the
challenge
was
to
be
able
to
to
test
to
test
those
binaries,
especially
on
32-bit
arm
test,
those
docker
images,
sorry
and
so
a
while
back.
Someone
from
the
community
pointed
pointed
us
to
this
website,
which
is
the
open
source
lab
at
oregon
state
university
and
they
provide
the
arm
64
servers
provided
by
mpr
computing
for
open
source
projects,
and
so
I
applied
for
a
few
servers.
B
I
think
for
four
vms
we
have
and
it
was
granted,
and
so
now
we
are
using
those
vms
to
build
and
test
the
arm,
64
and
arm
32-bit
images
and
so
yeah,
basically
we're
working
on
updating
the
documentation
right
now.
But
if
you
go
to
our
entry
docker
up,
you
can
see
that
there's
a
new
image
available
called
entria
dash
ubuntu
dash
m
arc
for
a
multiple
architecture,
and
if
you
click
on
it,
you
see
that
it's
actually
a
docker
manifest
with
support
for
multiple
platforms
amd64.
B
So
that
would
be
the
same
image
as
the
standard
one
entria
slash,
entry,
ubuntu
and
yeah.
There
is
also
a
support
for
arm64
and
arm
v7,
which
is
a
32-bit
arm
platform
used
for
raspberry
pi's
and
things
like
this
yeah.
So
that's
about
all
I
had
oh.
I
just
wanted
to
point
out
that
building
and
pushing
that
image
is
not
done
through
our
main
github
repository.
B
It's
done
through
a
private
github
repository
and
that's
not
because
there's
anything
to
hide
it's
because,
because
we
use
self-hosted
runners
for
the
ci
jobs
that
are
going
to
build
and
push
those
images
and
because
there
is
no
insecurity
problems
when
using
self-self-hosted
runners
with
with
public
github
repositories.
Basically,
anyone
can
open
a
pr
and
and
submit
some
code
which
is
going
to
run
on
the
runners,
and
it
can
run
arbitrary
code
on
the
runners,
which
is
not
a
good
thing.
So
right
now
we're
using
a
private
repo
for
those
build
scripts.
B
You
can
see
that
the
images
are
built
every.
I
think
it's
every
about
every
six
hours
and
it
takes
like
an
hour
and
30
minutes
to
build
them
and
and
run
the
community.
Some
communities
conformance
tests
on
a
rancher
cluster
using
those
arm
images
yeah,
but
as
soon
as
github
addresses
the
security
limitations
of
using
self-hosted
runners
for
ci
with
public
repositories.
B
My
plan
is
to
merge
all
this
infrastructure.
I
mean
all
this
infrastructure,
it's
just
a
very
simple
script
to
the
upstream
github
repo
and
after
that,
the
plan
would
be
to
use
that
image
by
default.
In
the
entry
yaml
manifest
so
that
there
is
arm
support
out
of
the
box-
and
you
don't
need
to
point
your
cluster
to
a
different
image
by
editing
the
ml
manifest
manually.
A
Clusters,
sorry,
maybe
you
already
said
it
it's
for
arm.
There
is
no
feature
limitation
or
whatsoever
everything
that
works
on
amd,
64
works
or
arm
as
well.
B
Yes,
everything
that
works
and
md64
works
on
arm
as
well,
essentially
yeah.
Basically,
all
the
packages
we
use.
The
ubuntu
packages
are
available
on
arm
platforms,
no
problem
and
we
can
build
go
binaries
of
entria
for
both
of
those
architectures
without
issues
as
well,
and
they
run
fine
great.
C
So,
in
the
in
the
arm
case,
typically,
it's
like
the
water
nodes
are
arm,
or
you
can
have
hybrid
case.
Some
of
those
are,
for
example,
some
some.
Some
are
sorry
somehow.
B
B
Historically,
I
think
there
were
some
issues
with
running
the
communities
control
plane
component,
so
the
api
server
and
the
controller
manager
on
arm
platforms,
but
I
go
through
the
rancher
k3s
distribution
which
supports
like
arm
nodes
for
both
the
control
plane
and
the
worker
nodes,
and
I
haven't
had
any
issue
with
that,
but
I
think
in
a
lot
of
deployments
you
would
run
the
you
may
run
the
control
plane
on
x86.
A
D
D
D
So
currently
the
restriction
is
that
the
entry
operator
only
is
only
supported
to
deploy
or
manage
the
entry
components
on
the
openshift
platform,
so
to
make
the
operator
can
be
run
on
my
platforms
when
we
need
some
changes
based
on
the
current
current
design
yeah.
So
if
you
you
are
interested
for
about
the
current
design
details,
you
could
go
to
this
go
to
this
google
doc.
D
You
know
from
a
continental
confidence
I
will.
Firstly,
I
will
talk
about
the
two
two
challenges:
two
changes
in
the
operator.
It
includes
the
channel
for
the
entry
install
crd
and
the
change
for
the
to
refer
to
the
controllers,
the
current
controller
logic
and
besides,
we
will
need
some.
I
will.
I
will
want
to
do
some
enhancement
to
current
operator.
D
D
Entry,
install
crd
is
the
single
source
for
the
configurations.
We
when
we
deploy
or
deploy
the
entry
components,
the
operator
will
read
all
configurations.
It
needs
from
the
entry
install
custom
source.
D
There's
two
changes
we
want
to
introduce
the
first
one
is,
and
we
want
under
new
field
in
this
mic,
to
allow
users
to
specify
the
the
kind
of
the
of
the
deployment
or
running
platform
and
the
second
change
exams.
We
will.
We
will,
under
some
info
in
the
entry
install
status,
to
show
the
current
status
of
the
installation
of
the
configure
realization
status.
D
Currently
we
will
we
will
support
the
overshift
with
our
current
implementation
and,
besides,
besides,
we
want
to
add
a
new,
a
new
kind
which
nominates
the
cronies.
D
D
So
when
user,
when
user
defined
the
value
in
this
field,
the
operator
will
choose
the
to
the
different
of
the
right
behavior
according
to
the
kind
of
the
platform
specified
by
the
user.
D
Currently,
on
overshift,
we
also
show
the
status
of
the
installation,
but
we
use
a
different
crd.
The
name
is
the
class
operator
this
already.
This
is
it's
private,
because
it's
only
created
in
the
openshift
platform.
So
on
other
other
platform,
we
wouldn't
we
don't
have
this
crd
to
show
the
status
of
entry
operator.
D
So,
in
summary,
we
will
introduce
two
fields:
almost
two
field
in
the
entry
yeah
and
to
install
crd
one
is
for
for
user
to
specify
the
current
platform
and
another
is
to
show
the
status
of
the
operator.
D
Okay
and
the
next.
I
will
talk
talk
about
controller
first,
okay,
with
about
changes,
we
will
need
to
change
some
logics
in
a
controller
to
attempt
to
attempt
to
about
changes.
D
D
This
controller
will
be
response
from
the
to
apply
the
configurations
with
space
file
in
the
untrue,
install
signalr
and
another
is
the
port
controller.
D
It
will
monitor
the
monitor
if,
if
the
component
is
alive
and
if
one
component
is
deleted
for
some
reason
it
will
resume
resume
the
component
so
for
the
config
for
the
config
controller,
it
will
check
it
will
check
which
platform
is
running
on.
D
So
if
it's
not
a
open
shift,
it
will
don't
need
to
read
the
reader
network
configs
from
the
from
the
api
server,
because
currently
we
we
will
read
the
service
cider
and
defaults
mtu
from
the
openshift
crd
and
the
next
one
is.
Then
we
will
also
turn
to
update
the
class
operator
on
non-operating
platform.
D
On
an
adjust
side,
we
will
use
the
status
field
of
entry
install
to
show
the
status
yeah.
So
there
is
a
configure
realization.
The
operator
will
we
update
our
status,
we
update
and
reinstall
those
status
about
the
current
configure
reconciliation
states
and
for
the
port
controller.
D
Sorry
here
is
a
mistake
and
for
the
product
controller
it
it
will
also
update
the
status
field,
adjust
the
config
controller.
The
difference
is
that
it
will.
It
will
change
this
field
according
to
the
account
to
the
status
of
the
android
controller
and
the
agent.
The
status
of
these
two
components.
D
And
the
first
two
parts
I
just
mentioned
is
the
is
necessary
for
the
anterior
generic
operator,
and
here
is.
It
is
an
additional
additional
parts
I
want
to
introduce
and
it
may
not.
It
may
be
not
necessary
for
the
for
the
current
stage,
but
it
will
be
good
for
the
long-term
design.
D
Employment
and
it
posts
some
common
functions
in
the
android
people,
and
that
these
common
functions
can
be
used
to
be
other
components,
for
example
the
entry
operator
and
the
under
control
and
others.
D
Let's
take
the
latex
configuration
related
function
and
the
example,
for
example,
when
we
in
the
entry
android
agent
or
android
controller,
we
need
to
load
the
configure
from
a
yama
file
and
send
defaults
about
the
config
about
the
config
values
and
after
that
we
also
need
to
validate
manage
the
configure
if
the
configure
is
is
the
red
values.
D
So
I
think,
for
this
config.
These
functions
are
also
needed
by
the
operator,
because,
when
operators
load
the
configurations
from
the
entry
install
crd,
I'd
also
need
to
do
the
sim
check
before
before.
It's
created
and
created
entry
components,
so
I
think
for
this,
for
the
functions
of
the
configs
like
exam,
it
can
be
shared
and
a
post
to
the
entry
operator.
D
D
Besides
that,
as
soon
as
here,
here's
a
more
more
complicated
common
functions
which
can
show
that,
from
the
entry
level
to
other
rep
to
other
components,
just
like
we,
we
talked
about
the
config
config
functions,
for
example
the
logic
about
how
to
update
the
config
of
entry
and,
for
example,
how
to
upgrade
the
entry
android
components.
D
So
we
can
choose
the
config
companies
and,
as
an
example,
I
don't
have
a
write
page
reader
pr
about
how
to
how
to
roll
out
the
entry
config
changes
using
the
a
mutable
configure
configure
yammer
file.
D
So
in
this
in
this
pr,
I
only
write
the
logic
to
to
treat
the
config
map
as
immutable
and
when,
when
a
user
wants
to
set
a
new
update
or
configure,
it
will
generate
a
new
version
of
yammer
files
and
then
do
the
layout
about
the
ruling
out
on
the
entry
entry
components.
D
So
and
this
part
of
logic
can
be
imposed.
It
can
be
imposed
in
one
function.
So
if,
if
this
function
is
it
is
exposed,
then
we
can
leverage
this
function
from
from
outside,
for
example,
if
we
use
under
control
we
can.
We
can
link
the
uncontrolled
code
to
this
piece
of
code,
and
we
can.
We
can
build
the
command
like
this
and
control
config
update
and
when
we
execute
this,
this
piece
of
command
with
it
will
call
the
function
in
this
part.
D
So
just
like
that,
the
entry
operator
also
can
leverage
this
piece
of
piece
of
logic
and
do
not.
I
don't
need
to
implement
another
another
another
copy
on
an
operator
wiper.
D
Yeah,
okay:
this
is
the
third
part
I
want
to
mention
about
some
enhancements.
D
And
for
the
last
part,
is
about
the
images,
so
here
we
will
have
two
two
main
components
which
we
need
to
build
images.
The
first
one
is
entry
image
and
another
one
is
the
operator
image
for
the
entry
image.
Currently
we
we
have
the
image
many
based
on
the
sound
ubuntu,
but
on
the
but
on
the
openshift.
We
will
need
the
image
based
on
the
ruby,
so
this
this
is
to
do
and
how
and
for
the
operator
image.
D
D
So
here
is
a
flow
performance
formula,
formant,
combo
and-
and
here
another
I
need
to
mention-
is
the
photo
image.
I
think
here
we
don't
have
the
requirement
for
the
photon,
so
we
could
leave
it
as
to
be
determined.
A
Thanks
all
right,
this
has
been
a
very
informative
and
the
only
thing
that
I
want
to
ask
you.
Yes,
thanks
for
sharing,
that's
exactly
what
I
wanted
to
ask
you.
If
you
could
please
share
the
design
document.
I
I
think
it's.
I
think
it's
a
very
good
design.
It's
you
made.
You
presented
it
also
in
a
great
way.
A
You
know
the
only
thing
where
probably
there
could
be
some
questions
is.
I
don't
know
whether
it's
a
good
idea
to
add
the
entry
as
a
dependency
for
the
operator.
I
don't
think
it
should
be
a
problem
this
I
believe
you
know.
As
you
pointed
out,
there
is
a
an
advantage
in
doing
so
and
what
could
be
the
downside
here?
A
A
A
C
So
program
for
crashing
here,
I'm
not
sure
it
makes
sense
or
not,
but
I'm
thinking
I
could
use
this
operator
to
do
something
for
windows.
D
Yeah,
I
hope
so
I
hope
so
yes,
so
we
can.
I
think
we
can
leverage
the
operation
to
do
some
configuration.
You
know
we
are
hard
to
set
up
the
environment
needed
by
the
entry
windows,
so
we
could
leverage
the
operation
to
to
prepare
the
the
network
needed
by
the
os
yeah.
I
think
it's
a
is
a
good
idea.
C
Okay,
busy
still
have
the
same
restriction
over
what
what
the
windows
container
can
can
do
for
the
host
network.
D
Yeah
this
other
written
will
be
wrong,
will
still
be
wrong
running
the
hack
host
network.
It's
because.
C
F
C
I'm
thinking
could
we
do
something
to
simplify
the
node
set
load
confusion
but
just
as
we
said,
since
we
still
wrong
operator
in
container,
I
guess
we
still
have
these
same
restrictions.
We
have
in
our
agent
demon
side.
G
C
F
D
Yeah,
I
think
it's
also
a
potential
way
to
to
help
set
up,
for
example,
the
community
service
api
server
for
the
agent.
D
Currently,
the
cleanup
is
not
considered
in
the
kind
of
current
design
because
we
have
not
yet
applied,
provide
a
way
to
about
how
to
delete
the
launch
here.
For
example,
yeah.
You
know
we
have
notified
the
behavior.
B
Yeah,
I
think
this
is
also
something
we
could
do
with
some
common
functions
that
we
put
in
the
andrea
repo.
A
H
I
have
a
small
question
about
the
android
cni
config.
So
is
it
the
file
the
ci
file
you
wanted
to
install
on
all
the
nodes.
H
How
do
you
plan
to
I'm
thinking
about
implementation?
How
do
you
plan
to
to
do
that?
Use
ssh
like
how
do
you
go
to
the
is
it
like
a
going
through
agent,
android
agent.
D
Okay,
I
think
I
got
a
question
so
if
I
think,
if
we
want
to
set
up
the
environment
for
the
entry
on
node,
I
think
maybe
we
should
provide
the
like
the
dimmer
set
away
to
run
the
sentence
right
instance
on
every
node
to
help
install
the
signal.
Plugin.
C
Because
for
lila's,
pretty
simple
you
can,
you
can
just
run
a
demon
set
and
mount
the
host
file
system
into
the
port
and
then
from
there
you
can
copy
the
file
to
the
host
defenses.
D
Okay,
you
mean
you
mean,
what's
the
difference,
what's
the
dependency.
H
Yeah,
my
understanding
is
that
this
config,
this
crd,
install
android
agent
as
well,
so
there's
some
kind
of
yeah.
It's
like
a
you
install
this
intra
operator
which
will
install
enter
agent
and
cni
for
you
right.
H
Okay,
so
it's
kind
of
I
see
so
you
install
basic
financial
agent
first
then
depend
on
agent
to
install
cni.
D
F
H
B
H
See
I
think
one
thing
is
that
sometimes
the
cni
is
kind
of
different
different
solution
users
seeing
it
differently.
They
like
they
put
more
stuff
in
it.
It's
kind
of
a
cinema
can
be
changed.
For
example,
for
example,
flannel
can
be
chained.
Sometimes
he
still
also
has
something
in
the
cni.
H
So
it's
kind
of
you
know
it's
common.
It's
a
it's
a
component,
it's
a
file
that
everyone
is
changing,
so
I
I
don't
know
I
could
yeah.
I
think
it
be
careful
like
how
to
how
to
change
it.
D
Yeah,
that's
one
point
we
need
to
we.
We
need
to
be
careful
so
so
currently
we
only
consider
the
the
single
the
single
thing
line.
So
I
think
it's
a
good
question.
We
need
to
to
think
about
how
to
collaborate
with
other
thing
eyes.
We
install
the
configure
config
config
list
about
a
single
plugin.
F
Actually,
we
we,
we
put
the
content
of
the
zener
config
file
into
the
development
yam
file
right.
So
if
someone
wants
to
update
the
the
same
item
or
use
another
name,
he
or
she
can
update
the
deployment
yam,
which
has
the
hard
code.
The
thing
I
named
file,
I
think
it
has
to
be
turned
and
andrea.com,
but
the
content
can
be
customized.
D
Yeah
correct
so
as
on
the
side,
currently,
we
defined
the
file
values
the
content
of
the
file
in
yamo
and
configure
configmapiamo.
Your
user
could
edit
it
or
update
this.
D
A
A
Okay,
if
there
is
no
additional
follow-up
on
this
question,
perhaps
we
can
move
to
the
next
topic
in
the
agenda,
which
is
open
discussion
since
we
don't
have
any
any
other
topic
booked
for
today.
A
A
A
Well
then,
it
appears
that
we
can
have
a
maybe
a
shorter
one
today,
since
there
are
no
more
topic
to
discuss.
I
don't
know
if
the
maintainers
like
antonina
and
jon
jones,
when
you
have
anything
else
that
you
would
like
to
bring
up
for
today,.
B
No,
I
just
want
to
say
that
0.13
release
is
approaching
it's
going
to
be
in
about
10
days
and
so
yeah.
Please
review
open
prs
and
if
you
have
pr's
that
need
to
make
it
into
0.13,
please
address
review
comments
and
update
the
prs.
C
So,
what's
it's
fabulous.
A
G
Do
we
have
a
meeting
the
next
month
in
the
next
bi-weekly
meeting,
because
I
think
monday
is
a
u.s
holiday.
A
A
Yes,
so
maybe
maybe
we
want
to
move
that
meeting
to
tuesday,
the
16th
so
that
you
know
we
don't
lose
the
meeting
and
in
any
case
next
next
monday
there
will
be
sorry
next
tuesday
on
2
p.m.
Pst.
A
But
still
another
another
another
chance
for
you
know
for
the
community
to
come
and
ask
questions.
So
I
don't
know
I
don't
want.
I
don't
know
if
we
want
to
keep
keep
move
the
meeting
from
february
15th
to
february
16th
or
whether
considering
that
it
will
be
a
public
holiday
in
china,
we
might
want
instead
to
cancel
it
or
to
move
to
make
a
move
and
move
it
to
next
week.
A
B
A
No,
no!
No!
No!
No,
I'm
not!
Sorry,
I'm
not
talking
about
the
office
hours,
I'm
talking
about
the
next
community
meeting,
which
falls
on
president's
day.
So
we
have
three
options
here:
one
move
it
to
the
day
after
two
move
it
to
next
week
and
cancel
the
office
hours
next
week,
so
we
move
it
forward
so
that
you
know
we
also
we
don't
catch
the
chinese
new
year's
holiday
and
then
the
third
option
will
be
to
just
cancel
it.
A
So
I
don't
know
what
you
prefer.
I
mean
we
don't
have
to
decide
now.
We
can
also
decide
on
the
luck
channel
to
me.
I
will,
I
don't
think
it's
a
good
idea
to
lose
a
community
meeting.
It's
always
a
valuable.
You
know
a
valuable
moment
to
meet
together
and
discuss
about
features,
so
I
will
probably
move
it
to
choose
lady
16,
but
I
don't
know:
what's
your
feedback.
A
A
Yeah,
yes,
we
did
it
for
for
martin
luther
king's
yeah.
I
believe
okay,
so
I
mean
that
this
means.
I
will
ask
the
question
of
luck
to
everyone
to
see.
If
there
are,
you
know
there
are
voices
that
disagree
with
this.
Otherwise,
we'll
go
with
the
lazy
consensus
approach
and
the
next
community
meeting
will
be
moved
to
will
be
moved
to
tuesday,
the
16th,
and
I
will
update
the
andrea
community
calendar
accordingly.
B
Vikki
is
pointing
out
that
a
lot
of
people
are
going
to
be
off
in
china
from
11th.
A
A
That's
that's
a
good
point
here.
I
will
ask
you
know.
The
thing
is
that
I
from
what
I
remember.
People
in
china
tend
to
take
two
weeks
for
new
year's
holiday,
but
it
could
be
an
idea,
so
it
also
could
be
an
idea
to
move
it
to
thursday,
the
18th.
A
So
it
follows
on
friday
the
19th
for
for
people
in
china
anyway,
so
I
don't
I'm,
I
I
think
we
have
several
options
here.
We
probably
can
discuss
on
the
slack
channel
and
finalize
them
by
the
end
of
the
week.
E
So
I
think
that
maybe
offline,
I
can
double
check
between
colleagues
to
see
their
opinions.
If
we
can't
join
so
we
can
draw,
you
can
keep
it.
A
A
Waiting
a
few
more
seconds,
that's
it
well!
It
appears
that
all
the
topics
that
have
been
exhausted
for
today's
agenda,
so
thanks
very
much
thanks
a
lot
for
attending
and
thanks
a
lot
to
antonin
for
completing
work
on
army
integration
and
array
for
this
great
design.
Regarding
a
generic
operator
design,
this
has
been
a
very,
very
informative
session
and
thanks
again
to
everyone
for
attending
and
hope
to
meet
you
again
in
the
entry
office
hours
meeting
next
week.