►
From YouTube: Kubernetes SIG Network 2019-03-21
Description
SIG network meeting from March 21st, 2019
A
B
C
A
B
I
went
through
all
bugs
that
were
sick,
Network
and
I
attached.
The
triage
unresolved
label.
Okay
for
now
is
an
indicator
that
we
have
not
looked
at
it.
Sig
contrib
X
has
got
a
proposal
to
actually
auto
label
every
new
incoming
issue
with
a
with
a
new
label
so
that
we
don't
have
to
have
this
race
condition.
So
what
I
did
today
is
I.
B
Looked
at
all
SiC
network
tag,
issues
that
were
newer
than
the
ones
we
looked
at
last
time
and
and
then
anything
that
has
also
tagged
unresolved
so
and
then
I
loaded,
a
bunch
of
them
up
in
my
tabs
and
will
do
like
last
time.
I
guess
we'll
see
how
many
we
can
get
through
I
actually
took
a
20
or
30
minutes
today
to
pass
through
a
few
of
them
commented
on
a
couple
closed
a
couple.
The
good
news
is,
we
got
a
bunch
of
them
from
last
week
done
the
bad
news.
B
Is
we
have
net
more
this
week
than
we
did
last
week?
So
hopefully
we
can
sign
up
for
these.
Thank
you
for
all
the
people
who
volunteered
last
time,
and
hopefully
we
can
get
more
volunteers
this
time
too,
and
hopefully
the
volunteering
process.
The
triage
process
isn't
so
painful
that
we
can
get
a
lot
of
people
volunteering
for
this.
So
let's
jump
in
and
shoot
for
30
seconds
per
issue.
Last
time
we
hit
about
45
seconds
per
shoe
so
again,
I'm
looking
for
volunteers
to
do
the
triage.
B
D
B
B
I
have
no
idea
what
this
one's
about.
I
mean
I,
actually
to
be
fair,
I
looked
at
it,
it
looks
like
they
have
two
I
two
IP
addresses
on
their
e0
and
it's
available
by
one
and
not
available
by
another
I.
Don't
really
know
why,
if
it's
supposed
to
be
listening
to
zero,
so
I
guess
this
is
a
case
of
somebody
should
try
to
reproduce
it
or
get
more
information
to
understand
it.
Any
volunteers.
B
If
you
don't
have
access
to
and
try
to
add
the
slash
kind
and
slash
priority
labels
us
that
we
can
get
a
sense
of
whether
it's
a
bug
or
a
feature
request
and
whether
it's
like
on
fire
urgent
or
just
let's
do
it
eventually
and
if
anybody
doesn't
have
access
to
to
use
the
slash
commands
for
the
bot.
Just
let
me
know
and
I
will
do
it
for
you.
I,
don't
have
a
problem
with
that.
B
F
F
G
F
G
B
B
Flaky
test.
These
are
the
best
I
haven't
looked
at
these
tests.
It
may
simply
be
a
case
of
like
our
timeout
is
too
short.
Does
anybody
have?
This
is
a
bigger
one
to
ask
people
to
sign
up,
for
anybody
has
bandwidth
to
spend
time
looking
at
a
flaky
test,
master
upgrade
it's
the
load,
balancer
firewall
rules
for
load,
balancer.
H
B
In
my
opinion,
I
would
rather
that
we
just
get
rid
of
the
cleanup
support
entirely
and
just
say:
look
switching
modes
on
a
cube
proxy
from
IP
tables
to
ipbs
without
rebooting
the
node
is
just
a
mess
and
like
we
were
trying
to
do
the
right
thing
here
when
we
were
switching
from
user
space
to
IP
tables
and
back
but
honestly,
I
think
it's
gotten
out
of
hand,
and
we
should
just
abandon
ship
on
idea.
Yeah
that
make
sense
to
me.
B
B
B
B
B
B
B
Okay,
so
I
think
we
can
make
it
into
one
end
of
114
one
or
something
with
the
cleanup
does
work,
but
they
overlap
with
each
other,
so
actually
for
us
to
make
it
work
properly.
We
have
to
do
sort
of
the
disjunction
of
the
cleanup
operations
between
the
different
modes
right,
which
doesn't
seem
reasonable
to
do
right.
I
think
we
should
just
give
up
on
it,
so
I'm
willing
to
argue
that
this
is
actually
a
bug
and
that
fixing
the
bug
is
by
disabling
the
cleanup
functionality.
B
B
H
May
be
related
to
the
PBS
captures
the
good
traffic
or
on
the
known
as
well.
I
think
so
might
be
related
to
that
volunteers.
B
B
B
My
guess
is
this
is
probably
a
network
setup
problem
again,
there's
not
enough
information
here.
Can
somebody
volunteer
to
extract
a
little
bit
more
information
like
what
cloud
are
they
running
in
what
network
is
it?
How
did
they
set
up
the
network
etc
and
maybe
point
them
at
the
general
network,
debugging
Doc's,
that
we've
got.
I
B
B
H
B
F
B
K
Running
Fred,
basically
like
there's,
there's
this
one
repo
out
on
like
it's
referenced
there
Microsoft
this
DN.
It's
got
some
wineries
published
there.
Basically,
that
doesn't
have
like
it's
basically
like
they'd.
They
want.
They
don't
want
binaries
published
on
the
that
Sdn
repo,
but
they
want
the
fineries
published
on
a
container
networking,
plugins
repo,
it's
more
us!
You
and
I
plugins
discussion,
I,
would
say:
okay
bug
per
se,
so
we
should
discuss
this
and
like
see
and
I
plug
in
meetings.
K
K
K
B
B
B
J
There's
a
there's
a
PR
open
for
that.
Essentially
right
now,
like
route
controller,
does
a
full
resync
and
it
doesn't
actually
act
on
any
node
events,
but
then
the
the
route
sync
is
also
really
expensive,
and
so
there's
discussion
on
how
to
do
the
route
sync
on
a
node
event,
without
making
the
controller
do
too
many
API
calls
okay
you're
on
this
one
yep.
C
B
A
B
H
B
D
B
B
If
you
see
new
network
tickets
that
are
marked
unresolved
or
that
are
that
are
newer
than
75
502,
which
is
where
I
started
today,
then
you
can
apply
unresolved
or
I'll.
Just
do
it
again
in
getting
ready
for
this
in
two
weeks.
So
it's
not
super
urgent
because
at
least
I
have
them
a
milestone.
Marker
I've
been
running
a
pot
over
unsorted.
K
Right
so
there's
a
open
issue
right
now:
we're
just
trying
to
promote
five
networking
tests
to
conformance
post
that
I
wanted
to
discuss,
promoting
these
tests
to
conformance
under
consideration
of
Windows
support,
but
before
I
get
to
that
I
wanted
to
ask
is
like
Signet
birthday
party.
Besides
promoting
network
tests,
the
conformance
is
that
another
sake.
Is
it
a
joint
effort
or
how
does
that
work?.
G
Well,
that's
a
great
question,
so
I
think
that
there
is
a
conformist
working
group
and
compromise
as
part
of
sig
arch,
but
certainly
the
so
that's
the
approval
process
for
the
tests
at
the
end.
Somebody
is
from
from
that
group
has
to
approve
the
conformance
tests,
but
obviously
subject
matter
matter.
Experts
are
you
know,
the
advice
is
taking
very
clearly
very
seriously,
so
it
becomes
a
joint
thing
in
that
sense.
I.
G
K
Sure
so,
there's
like
two
categories
here
of
these,
like
two
categories
to
group,
these
five
million
tests
under
I'd
say
so
the
first
two
tests
that
issue
that
I
linked
are
like
testing
partially
qualified
domain
resolution
and
that
will
not
definitely
not
pass
on
Windows
but
like
changing
the
platform,
we've
discussed
this
previously
and
I.
Think.
G
K
That
in
here,
that's
like
our,
where
we
were
working
the
issue
were
we
were
working
on
equivalent
tests
for
windows
only.
That
being
said,
there
should
be
a
way
to
make
these
tests
pass
on
Windows,
as
they
are
with
cry
container
D
support,
but
for
right
now
at
least
that
runtime
is
still
really
new
to
Windows,
so
it
might
take
a
while
until
we're
there
so
yeah.
G
L
So,
basically,
the
goal
this
canvas
you
know
to
remove
as
a
tunnel
from
the
khubaib
chat
server
code
base
and
to
replace
it
with
the
Napa
Junction
to
connect
to
the
notes
and
pods
in
a
more
complex
network
environment.
So
we're
gonna,
replace
it
with
a
generic
proxy
service
is
defined
by
a
RPC
and
the
implementation
we
intentionally
to
open
because
it
could
be
just
a
simple
as
just
as
a
tunnel
right
right
now
or
it
could
be
nothing
when
you
have.
L
You
run
your
master
than
working
on
the
same
network
and
it's
really
up
to
the
invitation
account
provider
to
implement
it,
and
the
benefit
is
basically
one
to
to
make
it
a
chess
server
simpler.
It
really
doesn't
make
sense
to
whether
that's
just
hung
inside
of
a
Chester
and
it's
kind
of
helpful
to
address
some
security
concerns
around
to
cuddle
proxy,
and
this
also
helped
for
our
and
the
cab
is
kind
of
multi
across
thick.
L
So
I
think
water
presented
last
week
in
the
Stig
HR
machinery
and
we
already
get
a
lot
of
good
comments
and
feedbacks,
and
we
really
appreciated
that
and
we
hope
that
anyone
was
interested
in
this
instinct
networking
to
take
a
look,
especially
those
who
were
running
HS
server,
a
master
and
working
those
two
different
networks
or
beer,
our
reliable
untrusted
Network.
You
can
take
a
look
and
we
hopefully
we
can
get
more
feedback
and
will
this
for.
L
D
Always
we're.
A
E
H
B
B
H
B
B
I
think
we
needed
a
volunteer
to
put
together
like
an
outline,
and
then
we
can
review
so
if
Valerie
is
volunteering
well,
happily
help
if
we
want
to
do
able
to
do
a
separate
zoom
meeting
like
next
week
or
something
maybe
this
time
next
week
or
something
with
a
smaller
group
of
people
to
just
throw
together
no
life.
Does
that
sound,
fair
sure
Jon's
in
alright,
we've
got
three
when
I
do
one
more
person,
one
more
volunteer:
okay,.
B
J
A
Awesome
I
I
won't
be
there,
but
usually
we
do
kind
of
like
an
informal
get-together
so
but
that
can
be
kind
of
figured
out
as
we
a
little
bit
closer.
D
A
B
B
B
C
I
B
B
C
B
I
E
B
B
Yep
yep,
so
Lucas
and
some
others
are
working
on
this
component
base
thing,
which
is
a
wrapper
for
a
whole
bunch
of
the
libraries
and
all
of
our
command
lines
and
I'm,
using
including
flags
and
logs
and
other
stuff.
So
the
I
think
the
real
answer
is
to
adopt
a
component
base
when
component
based
is
ready.
So
maybe
the
real
answer
is
just
to
remove
the
to
do.
Pr
is
welcome
sure.