►
From YouTube: Kubernetes SIG Node 20201019
Description
Meeting Agenda:
https://docs.google.com/document/d/1j3vrG6BgE0hUDs2e-1ZUegKN4W4Adb1B6oJ6j-4kyPU
A
I
think
it's
recording
okay,
so
hi,
it's
cygnus,
weekly
meeting,
it's
10
19
october
19th
and
welcome
it's
everybody
slow
monday.
Okay,
let's
get
started,
I
will
share
my
screen,
really
quick
and
we
will
get
going.
C
Yeah
so
joe's
bringing
up
something
that
was
discussed
last
week
for
anyone
who
wasn't
able
to
catch
up
on
the
meeting
on
the
meeting
notes
or
something.
I
brought
up
the
problem
statement
after
some
empirical
discovery
that
it
seems
that
we
we
could
the
kubernetes
community
don't
properly
surface
a
lot
of
processes,
procedures
how-to
guides
on
how
to
actually
take
care
of
a
how
to
actually
take
care
of
ci.
C
After
chatting
after
chatting
with
a
with
a
couple,
people
like,
I
think
he
become
a
he,
he
became
a
little
bit
obvious.
You
know.
If
we
we
have
templates
for
how
to
open
how
to
open
up
an
issue
in
kk
and
try
to
debug
it
try
to
debug
and
test
a
testing
there,
but
we
don't.
We
don't
properly
introduce
people
to
the
to
test
grid,
how
to
use
it
and
what
do
all
the
things
in
there
mean
we?
C
Don't
we
don't
have
a
lot
of
information
on
proud
jobs,
exactly
what
it
is
that
you
need
to
look
at
the
details
and
when
you're,
when
your
debugging
test
is
not
everything
all
the
all
the
things
are
not
and
not.
C
Everything
is
super
obvious,
and
so
I
propose
to
write
something
like
a
chromebook
to
add
at
least
make
it
a
little
a
little
bit
more
discomfortable
for
people
for
people
whenever
whenever
they
see
a
test
grid
alert
in
their
mailing
list,
or
they
see
an
issue
in
slack
or
way,
whatever
yeah
so
throwing
them,
throwing
the
link
out
there
throwing
the
link
up
there.
C
They
you
know
if
we
take
away
the
fact
that
note
hituit
tests
are
in
some
other
directory
and
they
work
some
different
way.
At
least
the
other
70
percent
of
the
work
is
still
applicable.
That's
great
works.
The
same
brow
works
the
same
this
the
information
is
going
to
be.
We
have
different
information,
but
at
least
we're
going
to
use
a
similar,
similar
tools
and
processes.
C
So
I,
through
I
draw
pr
I
draw
a
pr
for
I
guess
the
first
version
of
the
in
between
quotes
wrong
book
and
it's
by
no
means
perfect
it's
by
no.
By
no
means
I'll
say
it
doesn't
have
it
all,
but
I
figure
it
might
be
a
decent
place
to
start.
It
would
completely
open
to
any
other
thoughts,
comments,
questions,
suggestions
and
criticisms
that
anyone
has.
A
Cool-
and
I
know
there
is
somebody
on
country-
backs
working
on
kubernetes
dot,
df
side
and
they
like
right
now.
There
is
a
work
item
to
convert
jordan's
video
on
how
to
deal
with
flakes
into
the
documentation,
so
it
it's
also
very
like
very
generic.
I
think
it's
mostly
for
it's
less
about
end-to-end
tests.
I
think,
but
it's
still
a
lot
of
fun
to
interest,
so
it's
pretty
aligned
with
what
you're
doing
right.
C
B
C
You're,
even
if
you're
brand
new
to
this
whole
thing
and-
and
you
are
not
completely
sure
what's
in
there-
please
take
a
please
say-
please
review
I'm
hoping
that
this
whole
set
of
wrong
books.
They
should
be
good
enough
by
themselves
for
someone
that
was
brand
new
to
the
community
and
wanted
to
contribute
in
some
way.
A
Cool
okay,
any
comments.
A
I
think
you
only
have
nine
participants
in
this
call,
so
perhaps
we
can
start
doing
like
like
sending
agenda
to
entire
mailing
list,
so
people
will
be
at
least
aware
what
we've
discussed
and
what
pr's
needs
attention
so
maybe
to
have
bigger
parts
than
just
none.
People
on
the
call.
A
D
Hey
so
do
you
want
me
to
explain
a
few
words
from
this
agenda
item
yeah,
please,
okay!
So,
first
of
all
hi
everyone
nice
to
see
you
again
and
a
bit
of
context.
D
I
contributed
a
big
chunk
of
the
end-to-end
topology
manager
tests
and
because-
and
I
still
care
about
them,
so
we
I'm
working.
I
work
for
reddot
and
we
have
now
some
more
capacity
to
run
the
tests
and
we
are
running
them
internally,
but
I
want
to
contribute
them
to
the
community
and
so
expect
more
patches
in
this
area
and
from
me
in
the
in
the
coming
weeks
and
months.
D
So
why
I'm
raising
this
particular
pr
in
particular,
because
it's
very
simple
and
solves
an
issue
pretty
relatively
bad
issue,
which
you
can
have
if
you
run
the
tests.
Why
didn't
surface
before
because
most
likely,
because
last
information
I
have,
we
don't
have
a
multinuma
lane
on
on
upstream
on
our
ci
on
pro.
D
This
is
a
conversation
I
would
love
to
start
soon
enough,
but
for
the
moment
I'm
just
contributing
a
bunch
of
fix
and
as
soon
as
I
have
more
time,
I
will
really
love
to
have
this
conversation
and
try
to
see
if
we
can
enable
some
form
of
multinuma
in
improv.
So
again,
normal
times
have
just
asked
for
reviews
in
the
channel,
but
I
think
this
is
both
small
and
useful
enough
to
erase
the
year.
So
thanks.
Everyone.
A
Should
you
know
what
the
serial
device
to
review
it
or
just
general
cleanup.
D
No
and
well,
let
me
just
elaborate
a
bit
the
way
topology
manager
tests
they
do
expect.
They
do
want
to
run
within
a
multinomial
environment
and
with
any
survey
device,
and
this
is
the
reason
why
they
run
the
survey
device
plugin,
and
this
is
exactly
what
it
looks.
If
the
test
fail,
the
srv
device,
plugin
pod
leaks,
meaning
you
have
a
stray
pod
in
your
system,
so
this
is
in
the
the
sequent
runs.
It
may
cause
cause
issues.
The
tests
themselves,
don't
really
well
a
portion
of
that
do
not
require
survey.
A
D
A
A
I'm
sorry,
I
don't
know
you're
the
name.
B
So
my
main
question
is
related
to
the
also
to
the
franchesca
questions
regarding
how
to
test
like
no
malware
cpu
opinion
or
in
the
future
memory
opinion,
because,
yes,
we
have
hope
that
we
will
like
have
memory
manager
as
alpha
in
the
next
release,
so
doesn't
possible
in
any
way
to
enable
like
fake
numa
under
the
like
ci
high
notes.
Once
I
I'm
running
tests
like
it's
part
of
ken
arguments
and
from
what
I
saw,
it's
really
impossible
to
pass
to
pass
any
ken
arguments.
B
To
pass
to
have
a
fake
name,
it's
like
to
create
a
new
image
that
will
already
include
all
needed
kernel
arguments
under
the
like
proxy
and
d-line,
and
just
to
use
this
kind
of
image
under
rci.
So
I
just
wanted
to
have
some
kind
of
brainstorming
how
how
we
can
to
achieve
it
because,
like
you
know,
if
you
will
have
some
fake
name,
it's
it's
not
a
real
hardware,
but
at
least
it
will
have
some
emulation
of
real
use
cases
for
the
multiple
pneuma
environments.
A
A
Okay,
he
he's
not
replying.
I
remember
on
signature
some
time
ago,
alexander
from
intel,
I
believe
he
was
talking
about
emulating
numa
on
dorker
and
he
was
saying
like
exactly
the
same
thing
so
like
at
least
you
can
vary
the
basic
scenario,
even
though
it
wouldn't
give
you
a.
A
I
don't
remember
what
human,
but
my
impression
was
that
it
wouldn't
give
you
enough
confidence
that
it's
actually
working.
It's
just
like.
Basically,
you
can
validate
some
logical
aspect
of
it.
Not
what
the
result
is.
Is
it
the
same
kind
of
situation
here?
Are
you
talking
about
the
same
thing.
B
Yeah
yeah
exactly
like
again
like
cpu
manager
and
some
kind
of
fake
device
plugin,
for
example,
you
can
check
if
topology
manual
manager
like
generate
correct,
hints
for
for
single
luminal
policy
or
for
restricted
policy.
D
I
can
elaborate
a
bit
I'm
teammate
without
without
him,
the
basic
need
is
to
have
the
test
run
on
a
multinuma
machine,
either
fake
or
better
real
multi-machine,
because
in
this
in
this
way
you
can
actually
exercise
exercise
the
code
which,
for
example,
check
about
the
pneuma,
the
new
alignment
or
the
the
numerous
research
in
the
numerator
selection.
D
As
far
as
I
know,
I
didn't
I
I
must
say
I
didn't
check
recently,
but
like
a
few
we
up
until
few
weeks
ago,
we
just
don't
have
lanes
in
in
you
know
in
the
pro
ci,
which
run
on
multi
in
any
kind
of
multinuma
machines.
A
E
So
sorry,
I
I
think
I
just
speak,
but
I
didn't
notice.
I
was
muted
yeah,
so
this
multiple
numeration
is
on
the
bare
minimum
machine
right
is
I'm
not
flow
on
the
gce
or
the
cause?
This
like
what
is
the
weather
is
supported.
There's
also,
one
group
called
topper
main
main
membrane
group
right,
sub
saber.
Like
a
group,
they
do
you
also
new
machine
if
it's
a
fake
numa,
I'm
not
sure
whether
it's
implemented,
and
we
can
give
it
a
try
wrong
course.
B
Yeah
just
emulation,
for
example,
under
the
kvm
you
have
possibilities
like
also
pass
some
cpu
topologies.
That
will
include
like
separation
by
numerals.
You
know
so.
For
example,
you
will
have
like
two
buckets
with
the
memory,
for
example
zero
and
luma
node
one.
I
saw
that
you
see
also
like
based
on
the
kvm,
so
I
thought
maybe
it's
somehow
possible
also
to
emulate
this
kind
of
instances,
but
I
did
not
find
any
like
options
under
gce
to
pro
like
to
pass
some
complex,
cpu,
topology.
B
E
I
think,
in
that
case,
if
some
warranty
or
carrying
invoice
we
and
to
monitor
those
pests,
we
can
try
right
to
import
it
to
see
whether
or
whether
it's
flaky
always
like,
is
consistently
succeed
right
or
can
help
detect
bugs
because
the
thick,
the
mock,
that
means
some
weather
will
fit
it
with
some
yam
file
or
some
like
the
test.
Data
right
and
yeah.
B
E
B
Yeah
exactly
you
need
to
pass
additional
kernel
arguments
like
the
problems
that
e2e
nodes,
the
test
himself
runs
runs
on
the
instance.
So
if
I
will,
under
the
test
like
try
to
change,
cmd
proclaim
and
we'll
reboot,
the
node
it
like
will
stop
the
test
at
all.
So
I'm
from
the
beginning
should
have
all
these
parameters
on
the
under
the
like
kernel
proxy.
Indeed,
so.
E
E
B
I'm
sure
it's
like
will
be
this
issue
under
the
sick
testing
or
something
like
this,
because
it
doesn't
really
relate
it
to
the
signal.
I
don't
know
like
like
it's
not
the
issue
under
the
signaled
code,
so.
E
A
E
A
Yeah,
so
did
you.
A
From
martial,
like
one
above
with
the
parameters.
A
Great
yeah-
and
we
probably
may
want
to
include
alexander
from
intel
in
the
discussion,
so
maybe
he
may
have
some
experience.
B
Yeah
yeah,
it
can
be
great
like
again
the
main
problem:
how
how
to
achieve
it
under
like
kubernetes
side
but
yeah.
It
can
be.
A
Make
sense:
okay,
jorge?
Can
you
update
on
what
we've
been
discussing
last
meeting
about
mailing
lists
and
emails
because
is
not
here.
C
Yeah
so
from
the
last
meeting
recapitulating
in
a
lot
of
it
tests
for
c
for
signal,
we
use
attacks,
not
conformance
and
node
feature
the
tl.
The
tltr
is
a.
C
Matt
is
going
to
send
an
email
to
see
to
seek
note
and
a
to
a
signal
just
to
retake
a
just
to
retake
this
convert.
Just
to
retake
this,
the
conversation
there
is
a
decent
chance
that
we
get
that
that
we
could
do
a
way
away
with
attacks,
given
that
we
build
different
test
binaries
for
kubernetes
test
and
node
e3
test,
and
also
just
to
catch
up
on
all
the
things.
Actually,
I
I
guess
to
follow
up
on
the
super
old
document
from
two
years
ago
that
and
introduces
it.
I
introduced
the
text.
C
The
other
email
is
something
that
was
never
cleared
up.
It
was
the
relationship
of
the
node
conformance
e2
test
with
overall
cluster
conformance,
and
so
so
I
guess
the
ai
was
to
also
send
the
email
to
see
with
it
to
sega
architecture
and
see
what
note
conformance
it
could
actually
be,
and
I
can
and
I'm
gonna
read
the
rest
from
the
midi
from
the
from
the
midi
notes
with
matt.
It
has
been
busy
and.
C
A
Cool
okay,
so
there
are
action
item
on
matt
to
follow
up
on
that
perfect.
A
A
Today,
okay,
then,
okay,
let's
wrap
it
up.
Thank
you,
everybody
for
joining
and
have
a
wonderful
week,
bye.