►
From YouTube: Kubernetes SIG Node CI 20220803
Description
SIG Node CI weekly meeting. Agenda and notes: https://docs.google.com/document/d/1fb-ugvgdSVIkkuJ388_nhp2pBTy_4HEVg5848Xy7n5U/edit#heading=h.2v8vzknys4nk
GMT20220803-170449_Recording_2864x2100
A
So
I
guess
I'm
not
having
video.
We
don't
actually
have
an
agenda
today,
but
there
are
hello.
A
Okay,
cool
all
the
zoom,
you're
I
just
disappeared
from
my
screen,
so
I
wasn't
sure
so
yeah.
We
don't
actually
have
an
agenda
for
today,
but
there
is
some
stuff
we
should
talk
through
about
failing
CI
and
then
take
a
quick
look
through
the
open,
PRS
I'm.
B
Hoping
to
discuss
my
problem
too
sometime
during
the
meeting.
A
Cool,
do
you
want
to
just
pop
it
on
the
agenda,
so
we
don't
forget
yeah,.
A
So
we've
had
a
few
fun
CI
issues
for
the
last
couple
of
days,
mostly
related
to
the
cryo
tests
in
the
none
of
them
were
running.
For
the
last
few
days,
seemingly
due
to
like
a
change
in
the
Fedora
image,
we
were
using.
A
D
I'm
afraid
not
yet,
but
it
is
on
my
list
but
I've
I'm
kind
of
coming
back
from
vacation,
so
I
have
a
bunch
to
catch
up
on.
Oh.
C
D
We
just
need
I,
remember
looking
at
it
yeah
a
while
ago,
and
it
just
it
wasn't
apparently
clear
what
was
going
on
and
it's
weird
because
the
cryo
tests
like
there
are
different
sets
of
tests
that
the
different
jobs
are
failing.
So
it's
very
confusing.
A
A
Looks
like
I
missed
the
hack
in
another
test,
Suite
Joker,
but
containerdy.
We
have
one
unexpected
failing
test
still,
which
is
the
container
the
e2e
Ubuntu
test,
which
runs
regular
case
etes
and
has
been
sort
of
like
randomly
failing
for
a
few
days
and
still
not
sure
why
I
started
failing
when
we
updated
the
cause
image,
because
the
Ubuntu
tests
still
use
cause
for
the
control
plane,
node,
and
that
seems
to
be
something
wrong
with
like
connectivity
between
workers
on
the
control
plane.
A
A
B
That
would
be
great
I,
couldn't
add
what
I
needed
to
the
agenda.
I'll.
Add
it
afterwards.
That's
fine.
B
B
All
right
I
hit
the
share
button.
You
should
see
it
scrolling
to
the
top,
so
you
can
see
the
title
and
number
it's
about
some
slow
job,
which
is
label
this
flaky,
but
actually
fails
every
time
nowadays
and
I
figured
out
why
the
questions
how
to
solve
it
went
through
a
PR
process
which
I
didn't
wasn't
too
comfortable
about,
and
the
rest
of
the
community
was
not
either
so
we
I
shut
that
down
myself
yesterday
because
it
involved
putting
a
test
artifact
in
which
was
large
and
40
supporting
files.
B
B
B
Right
and
then,
if
that
all
fails,
a
totally
simple
idea,
I,
don't
know
why
it's
not
obvious
if
there
must
be
something
flawed
about
it.
This
is
just
a
load
test,
we're
doing
something
fairly
complicated,
a
tensorflow
job.
B
B
A
Okay,
I
cannot
use
computers
today,
that's
okay,.
A
So
add
e-tests
for
node
log
viewer,
which
is
apparently
some
new
feature.
I
apparently
looked
at
this
already
and
it
seems
like
they
fixed
things.
D
E
D
A
Okay
did
the
thing
that
is
good
I'll,
take
finishing
off
this
review
when
I'm
done
with
this
call
test,
put
volume
unit
test
to
Windows.
A
A
A
D
B
A
C
A
D
A
Just
say
that,
but
Cube
proxy,
why
is
this
triage
to
note.
A
A
Shutdown
node
host
the
kublet.
It's
graceful
note,
shutdown
enabled.
A
E
A
Windows
host
process
containers
are
missing
resource
utilization
metrics
for
metric
servers.
Let's
let
Windows
take
a
look
at
that,
assuming
they
haven't
already.
C
A
A
Is
the
best
website
for
saying
plus
plus
make
my
life
easier.
A
Assign
myself
to
that,
because
I've
been
looking
at
the
fiction
stuff
a
lot,
the
Kubler
is
setting
room
score,
adjust
backwards.
A
I
was
debugging
a
new
failure
recently
and
noticed
some
unexpected
behavior
for
how
the
kublet
is
setting
score
adjust
from
our
docks.
A
Exciting
I
guess
potentially
Michael,
Clayton
or
Derek.
D
Yeah
I
find
this
interesting,
but
I
I
can't
really
say
definitively
that
I
have
time
so
yeah
I,
don't
know
I
think
we.
E
Can
assign
it
to
occur
on
my
team
and
maybe
he
can
look
into
it.
A
A
This
is
a
rabbit
hole
that
I
do
not
have
the
the
capacity
to
commit
to
Master
disruption
mode
in
our
cycle.
Controller
not
doesn't
work,
sometimes.
A
A
In
versions
pre
119,
liveness
probe
starts
first
and
writing.
That
starts
after,
whereas
the
probes
now
happen.
At
the
same
time,.
E
E
Necessarily
run
at
the
same
time,
it
just
depends
on
whenever
the
timer
hits.
A
Can
I
see
someone
on
this
too?
Do
a
nice
answer.
A
Sorry
see
I,
say
I'm
about
to
putting
voices
to.
E
C
A
A
That's
a
future
request
this
disguised
as
a
bug
at
a
couplet
configuration
option
to
disable
C
advisor
metrics.
A
Triage
kind
of
cool,
so
does
anyone
have
anything
else
they
want
to
go
over
before
we
call
it
done.