►
From YouTube: Kubernetes SIG Node 20210512
Description
Meeting Agenda:
https://docs.google.com/document/d/1j3vrG6BgE0hUDs2e-1ZUegKN4W4Adb1B6oJ6j-4kyPU
A
Hello,
it's
may
12th
2021,
it's
the
ci
group
subgroup
of
sig
node
meeting
and
agenda
is
very
short
today.
That's
everybody
been
on
kubecon.
Did
you
listen
any
talks.
A
Cool
okay,
yeah,
and
for
you
it
was
a
proper
time
zone
right.
B
C
A
A
So
yeah,
let's
go
to
triage
like
since
there
is
no
big
agenda
items
already
cleaned
up
the
board
a
little
bit.
Now,
let's
go
to
prs
that
means
review
and
then
we'll
go
into.
There
is
an
issue
that
in
progress
but
has
no
assignee.
Okay,
let's
go
through
a
pr.
That
means.
A
Reviewer,
please
start
cook
with
flak
is
I'm
in
here.
He
was
just
replying
to
one
of
the
prs
yeah,
but
he
didn't.
A
A
A
Does
anybody
else
wants
to
look
at
this?
It's
I
think
restart
will
unblock
some
of
the
activities.
A
A
A
C
Yes,
it
does.
There
are
two
fixes,
the
first
I
fixed
the
the
test.
I
explained
actually
why
we
should
have
the
sleep
before
startup
that
is
bigger
than
the
readiness
for
period
seconds,
because
otherwise
we
are
triggering
a
readiness
probe,
while
it's
still
not
accepting
any
any
problem.
C
A
D
Yeah,
like
I
already
mentioned
before,
like
to
provide
some
better
testing
for
cpu
manager,
topology
manager
and
memory
manager.
We
need
some
additional
luminols
under
the
in
our
upstream
ci
instances.
D
D
A
Okay,
anybody
interested
to
review
it
seems
to
be
infrastructure,
mostly.
C
A
E
Yes,
so
we
have
those
tests
which
you
know
to
properly
exercise
topology
manager,
cpu
manager
is
not
enough,
so
you
want
to
have
devices
during
the
topology
manager.
Almost
one
year
ago,
derek
was
pretty
convincing
that
yeah
better
use
real
devices.
So
what's
the
most
common
real
device,
specific
devices-
and
this
is
what
led
sir
here
now
on
ci-
we
don't
have
yet
any
kind
of
srv
devices
or
any
kind
of
device
plugins.
E
But
if
I'm
not
mistaken,
maybe
gpus
on
some
special
lanes,
but
still
so
test
will
skip
because
they
will
run
against
the
cpu
manager,
but
on
some
organization,
namely
reddot
they
we
will
have
machines
with
those
devices
srv
devices
in
our
case.
So
in
our
case,
if
the
test
skip
instead
of
failing
it's
bad,
because
devices
are
supposed
to
be
there
in
ci
at
the
moment
are
not
supposed
to
be
there.
So
it's
better
to
skip
and
run
what
will
possibly
run,
but
on
environments
in
which
a
devices
are
there.
E
They
need
to
fail
because
it's
a
it's
an
environment
failure
is
a
test
failure,
it's
something
you
want
to
check,
and
so
there
we
have.
We
have
we
had
this
option
to
make
bot
words
happy.
So
you
you,
if
you
know
you
have
devices
like
in
custom
environments.
Unfortunately,
you
enable
otherwise
you
keep
things
as
they
are
and
ci
runs
like
today
is
doing,
and
if
you
have
it,
you
use
it.
If
you
don't
have
it,
you
skip
it
and
your
it
works.
So
this
is
the.
E
C
B
E
E
D
E
Device,
that's
correct.
That's
correct
that
yeah
there's
something
I
would
like
to
check
eventually
didn't
manage
to
do
it
yet.
C
A
A
D
D
A
A
D
The
know
whether
who
is
responsible
to
set
level,
for
example,
first
first
good
issue,
because
I
saw
someone
asked
some
asking
for
some
issues
to
contribute
under
the
signal
channel
and
like
we
don't
have
enough
like
first
good
issue
issues,
and
we
have
just
a
bunch
of
issues.
But
it's
problematic
to
find
the
easy
one
to
like
to
code.
Something.
A
Yeah,
so
there
are
a
bunch
of
commands
for
that.
A
good
first
issue
is
one:
when
you're
setting
good
first
issue
thingy,
you
need
to
make
sure
that
you're
setting
you
given
enough
context
what
needs
to
be
done.
Otherwise,
the
good
first
issue
becomes
impossible.
First
issue.
D
D
My
like
initial
questions,
like
do,
we
have
some
like
grooming
to
do
it
like,
because
we
have
a
lot
of
issues,
and
maybe
we
can
to
do
like
once.
I
don't
know
once
per
year
or
once
per
month
to
label
issues
that
we
are
thinking
is
good
for,
like
first
contribution.
A
Yeah,
it's
a
good
point.
I
just
went
through
this
exercise
mike
just
joined
the
team,
and
I
suggested
some
few
issues
for
mike
but
yeah.
It's
it's
very
tiresome
process
to
find
a
very
good
first
issues.
Instead
of
non-blocking
important
and
enough
scope,
did
you
manage
to
find
some
issues
or.
D
Again,
probably
like
to
end
test:
it's
not
effect,
like
you
know
the
production
again,
it's
like
not
easy.
Again,
you
should
understand
the
infrastructure
but
again,
but
I
think
in
the
beginning
of
this
document
we
have
some
links,
some
relevant
links
to
better
understand
and
I
believe
this
group
is
started
also.
This
is
a.
C
D
E
E
E
Everyone
provides
a
sublist
and
we
review-
or
we
just
add
the
at
the
the
label
ourselves
without
sinking,
but
I
think
the
we
will
need
to
agree
about
or
find
a
good
definition
of
a
good
first
issue,
look
like,
for
example,
has
to
be
clearly
defined.
It
has
to
be
small
or
expect
to
be
small
things
like
that,
and
when
we
have
that
we
we
can
basically
move
on
in
a
lot
of
ways.
D
Yeah,
I
think
it
can
work
and
probably
also
like
writing
code
coverage
to
find
places
with
that
doesn't
have
enough
like
testing
coverage
can
be
also
some
good
point,
because
I
saw
someone
contribute
some
topology
manager
test,
like
notopj
manager,
cpu
manager
for
the
topology
package
unit
tests,
so
also
it
like
it's
not
critical.
So
now,
no
one
will
like
build
you
and
we'll
ask
you
when
it
will
be
ready,
so
you
will
have
time
to
understand
and
to
work
on
it.
D
A
Makes
sense,
yeah
increasing
code
coverage
is
a
good
thing.
I
I
don't
know
like
when
we
started
the
group
like
we
wanted
to
first
identify
critical
issues
and
like
my
critical
tests
and
my
monitor
that
they're
not
breaking,
but
then
we
also
wanted
to
increase
code
coverage
and
we
never
get
to
the
point
when
we
identifying
areas
that
are
not
well
covered
by
end-to-end
tests.
A
A
No,
no
okay!
I
need
to
figure
out
why
it's
not
on
the
board
but
yeah.
So
everything
looks
fine,
so
we
can
start
identifying
more
areas
that
needs
to
be
covered.
D
We
still
have
some
we
here
we
still
have
serial
drop
just
like
after
we
increase
the
memory
just
started
to
fail
after
the
six
hours
of
rank,
so
it
did
not
did
not
succeed
to
finish
after
the
sixth
house,
so
probably
yeah.
We
also
can
take
a
look
on
the
serial
group
or
at
least
start
to
divide
it.
Maybe
I
don't
know
because,
like
we
drop
anything
that
can
be
paralyzed
under
the
serial
lane,
and
now
we
have
such
long.
D
We
increased
the
amount
of
memory
it
just
started
fail
on
the
timeline
like,
for
example,
if
I
will
check
now
the
test
grid
for
the
not
cubed
serial.
D
D
A
Yeah,
I
think
we
closed
the
issue
with
increasing
of
memory
right
and
we
never
opened.
D
A
A
A
Tracked
seriously,
I
need
to
check
like
what's
going
on
because
I
clearly
like
saw
the
issue
I'll
double
check:
oh
yeah,
this
one
okay,
so
I
will
I
think
of
this
I
mean
so
for
me.
A
D
Yeah
by
the
way,
orphans
looks
like
the
easier
one,
because
again,
it's
have
something
around
fintech
that
failed,
and
it
also
has
already
unlocked
so.
A
A
Tests,
okay,
does
anybody
know
which
areas
are
not
well
covered
with
end-to-end
tests
and
needs
to
be
covered?
Any
suggestions
right
now
or
we
like
following
francesca's
suggestion
and
like
not
it
for
ourselves
and
maybe
discuss
it
in
a
meeting
or
two.
D
A
C
I
I
don't
know
I
would
say
no,
but
on
the
other
hand
we
always
find
some
funny
side
effects.
So
probably
we
could
increase
the
try
to
see
if
we
can
increase
coverage,
but
the
thing
is:
if
you
only
look
at
code
coverage
it
does,
it
does
not
test
like
all
the
the
cases
right,
it's
always
difficult
to
to
see
edge
cases
in
the
test.
Unless
you,
you
really
know
the
the
system,
so
at
least
unit
test
is
easy
to
to
determine
the
the
coverage
but
for
end-to-end
tests.
I
don't
know.
A
Yeah
makes
sense
yeah.
I.
A
For
that's
why
unit
tests
and
code
coverage
are
like
less
important
for
me.
I
really
like
it's
easier.
Like
I
have
this
data.
I
can
just
get
data
and
create
issues
for
uncovered
modules.
It's
fine,
but
end-to-end
tests
are
very
important,
like
you
need
to
understand,
what's
covered
and
what
features
not
well
covered.
A
Yeah,
for
instance,
I'm
still
concerned
about
initial
timeout
seconds
for
probes
and
I'm
not
sure
how
we
calculate
this
initial
timeout
so
especially
for
regions
problems.
There
is
a
startup
prop
as
well
so
yeah,
I
don't
know,
I
don't.
I'm
not
sure
we
covered
this
this
unit
test
and
we
are
checking
that
we
respect
this
initial
timeout.
C
A
Topics
dwighty,
you
didn't
have
chance
to
talk.
Much.
Do
you
wanna
share
what
you're
working
on
like?
Do
you
want
some
other
tasks
like
what's
going
on.
B
Yeah,
so
I've
been
working
kind
of
with
francesco
on
this
new
feature
related
to
making
it
strict
to
have
smt
alignment
in
in
case
of
workload,
request
cpus.
So
that's
pretty
much.
What
I've
been
working
on
and
like
at
this
point
in
time
like
that
that
has
been
the
focus,
but
if
there's
things
that
are
looming
around
again
similar
to
artem
and
francesco,
my
experience
would
be
cpu
manager,
topology
manager,
maybe
a
little
bit
and
the
resource
manager
side
of
things.
B
So
if
there
are
tasks
on
that
or
reviews
that
I
can
do
there,
I
would
be
happy
to
help,
but
other
than
that,
I'm
not
sure
yeah
like
if
there's
anything
that
is
important
or
something
just
hand
it
over.
To
me
I'll.
Do
my
investigation
and
try
to
chip
in.
A
A
Do
you
wanna,
like
I
mean,
do
you
need
any
tasks
or
like
are
you?
Are
you
looking
for
some
tasks
right
now
or
you
have
something
to
do.
A
Yeah,
I
think
I
have
one
pr
need
that
charge
except
yeah.
Okay,
thank
you.
Okay.
This
concludes
our
test
part
like
and
now
we
need
to
go
into
product
backlog.
I
will
stop.