►
From YouTube: Kubernetes SIG Node 20220302
Description
Meeting Agenda:
https://docs.google.com/document/d/1j3vrG6BgE0hUDs2e-1ZUegKN4W4Adb1B6oJ6j-4kyPU
A
Good
morning
it's
march,
2nd
2022
welcome
everybody.
Today
we
have
empty
agenda,
so
the
meeting
will
be
mostly
about
triage.
Let's
take
a
look.
A
A
Yeah,
I
don't
think
we
need
to
see
flaky
here,
I'm
not
sure.
I
think
it
shouldn't
be
that
maybe
we
need
to
have
some
rules
like
when
do
we
mark
this
but
flaky,
because
I
think
we
did
it,
but
in
the
past
couple
times
already.
A
Yeah,
I
don't
think
it
should
be
considered
flaky,
because
it's
only
one
failure
among
like
many
tests,
okay,
but
I
don't
know
like
what's
the
threshold
to
market
here,
it's
like
because
here
it
says
like
humans,.
B
One
failure
is
enough
to
mark
it
as
flaky
like
it
should
never
flake.
I
guess,
for
the
release
blocking
test,
at
least
but
yeah,
because
those
are
old
releases.
I
highly
doubt
we're
gonna
go
back
and
like
fix
any
of
those
unless
they're
like
flaking
much
more
frequently
than
that.
A
Okay,
cool
blood,
so
lock
contention
is
a
new
failure.
We
introduced
this
low
contention
job
and
there
was
a
problem
with
cni,
so
we've
seen
I
probably
fixed,
but
the
lock
attention
job
is
now
by
itself
is
not
working.
It's
a
tracking
issue
that
you
will
need
to
trash.
A
Yes,
performance
shop
is
also
free.
I
think
our
bronze
job
was
also
just
introduced
very
recently,
and
let
me
make
a
note
here:
we
don't,
we
may
have
an
issue.
I
think
I
saw
youtube
for
performance.
C
A
Yeah,
I
think
one
of
my
pr's
changing
the
image
for
from
the
question
to
continue,
so
it
should
maybe
better
next
week
yeah
here,
which
is
same
whether
some
pg,
but
we
have
one
two
three
pr's-
that
attempt
to
fix
different
like
aspects
of
it.
So
hopefully
we
will
get
some
progress
next
week,
but
flick.
Is
it
the
same.
B
A
A
A
Yeah
you're
making
progress
unquestionably
just
two
jobs,
and
they
made
this
two
things
not
green
again
and
eviction
is
still
still
a
problem
here
in
korea,
I
saw
somebody
working
on
eviction
adding
some
on
chronic
direction
edge
in
the
log
extra
log
statement
to
troubleshoot.
I
I
was
quit
like
asking
on
this
pr.
A
I
don't
remember
who
supported
this
job.
Somebody
was
working
on
that.
B
Yeah
I
haven't
looked
at
any
of
the
container
d
or
the
is
it
container
os
jobs?
Thank
you
adidi.
If
you're
watching
the
recording
for
recording
this
danielle
did
say
that
it
was
okay
to
assign
her
things
if
nobody
else
wanted
them,
but
I
don't
know
if
there's
anybody,
maybe
from
google,
if
they
want
to
look
at
the
container
os
tests.
B
A
A
C
A
Okay,
just
puts
clean,
we
don't
have
any
needs.
Reviewer
everything,
origin
is
approver.
So
it's
let's
try
to
think
people.
I
think
many
of
the
sprs
outside
of
our
control.
We
need
to
think
more
people
as
approvers
anybody
interested
in
not
taking
more
tasks
from
to-do.
D
B
A
pretty
big
bump,
I
know,
have
we
had
anything
land
yet
like
to
run
c-bump
or
anything
like
that.
I've
been
working
with
bartack
on
trying
to
improve
cpu
utilization
of
the
cubelit
by
improving
the
cpu
utilization
of
metric,
scraping
and
c
advisor,
and
one
of
the
things
that
we
saw
was
that
we
had
significant
significant
cpu
savings.
B
However,
memory
was
jumping
quite
a
bit
more
than
we
expected,
like
nearly
a
hundred
percent,
and
we
were
going
to
do
some
testing
to
see
if
that
was
like,
because
of
just
those
changes
or
because
of
like
other
changes
that
might
have
landed
at
c
advisor
and
I'm
a
little
worried
seeing
something
like
that
that
maybe
we
just
had
other
memory
regressions,
because
that
looks
like
a
pretty
big
jump.
B
Because
I
was
very
surprised
to
see,
like
you
know,
if
we're
getting
70
cpu
savings,
I
wouldn't
expect
that
the
size
of
the
cash
would
be
so
large
that
we're
getting
like
100
memory
utilization
increase.
But
that's
what
we're
seeing
at
some
initial
tests
so.
B
A
That
maybe
because
yes,
there
is
clearly
a
jump,
so
let's
say
99
percentile
is
82
89.
I
mean
so
that
big.
A
Yeah,
it's
really
hard
to
say
it
looks
like
a
wave
right,
let's
see
same
in
the
beginning
and
then.
B
It
does
look
yeah
a
little
bit
like
I
would
say
it
looks
like
there's
an
increase
about
like
one
third
of
the
way
before
the
end
like.
If
you
go
back.
A
A
250
to
10
megabytes,
I
think
I
don't
think
it's
it's
a
two
breaker
or
anything.
A
Because,
later
like
this
will
disappear,
you
will
see
this
grow
grows
and
you
can.
You
won't
have
a
point
of
comparison
from
the
past.
B
C
B
Oh
and
there's
only
three,
so
that's
cool.
So
let's
definitely
look
at
those.
B
B
Great,
let's
get
these
four
secret,
v2
and
cubelets
seem
not
playing
well
together
on
non-systemd
os,
I
deploy
a
k3s.
Well,
we
are
not
k3s,
although
I
think
k3
is
used
as
a
kubernetes
cubelet
without
changes.
So.
B
B
B
D
Okay,
so
secret
v2
is.
D
B
A
So
it
doesn't
say
it
doesn't
give
any
information
like
what
exactly
means
that
it
joins
a
cluster.
A
So
the
issue,
as
I
remember,
is
they
saying
they
rejoin
the
cluster,
but
then
they
need
to
restart.
They
have
to
start
continuity
to
initialize
c9
plug-in
again.
I
have
no
idea
where,
like
what
it
means
to
join
the
cluster.
B
They're
also
using
a
very
old
version
like
almost
not
supported,
not
quite
not
supported,
they're
using
cube
adm
to
install
it.
Supposedly,
I
don't
know
what
k
nines.
A
B
Okay,
potter
could
get
lost
one
pod,
terminating
state.
This
is
quite
old.
D
D
D
I
know
that
this
one
has
undergone
a
bunch
of
work
since
we
stuck
the
needs
information
on
here,
so
I'm
gonna,
I
think,
remove
it.
I
don't
know
what
happened.
B
Okay,
we
want
to
quickly
look
over
things
that
are
in
need.
Recent
updates,
maybe.
D
Stale
answer:
we
added
a
comment:
survey
sale.
This
one
has.
B
B
D
B
B
D
A
Because
claims
that
after
he
started
everything
works.
B
B
Okay,
I
think
that's
everything
that's
been
updated,
so
I
think
that's
everything
for
this
unless
there's
any
other
bugs
folks
think
we
should
look
at.
B
Okay
looks
like
much
folks
are
up
for
another
meeting
and
I
think
that's
all
for
today.
So
thanks
for
joining
us
see
you
next
week,
bye.