►
From YouTube: Kubernetes SIG Node 20220202
Description
Meeting Agenda:
https://docs.google.com/document/d/1j3vrG6BgE0hUDs2e-1ZUegKN4W4Adb1B6oJ6j-4kyPU
A
Okay,
hello,
everybody!
It's
two,
two
two,
two
second
of
february
22
welcome
to
six
nook
ci
meeting.
We
have
a
very
short
agenda
today,
just
mic
fixed.
I
looked
at
the
kicks
like
you,
you
shifted
the
delay
in
different
place.
B
B
B
B
A
A
I
know
I
even
updated
the
dashboard
for
tests.
I
don't
know,
what's
the
recognition
called
today,
no
he's
not
anyway,
so
he
told
me
that
nothing
particularly
new
here,
so
there
are
a
couple
failures
on
here
on
kinder.
A
C
Okay,
daniel
my
gain
is
very
low.
I
can
barely
hear
you.
D
D
D
D
The
main
problem
is
trying
to
figure
out
how
to
do
that
in
a
way
that
is
sensible,
like
the
disk
eviction
tests,
I'm
increasingly
tempted
to
have
cont
kubelet
configure
itself
on
those
ones
to
use
a
like
temper
fest
with
a
forced,
fixed
size,
rather
than
you
know
like
trying
to
fight
with
the
system
for
other
stuff.
D
That's
going
on
partially
because
they're,
the
flakiest
tests
we
have
because
other
stuff
happening
can
break
them,
but
also
that's
a
horrible
idea,
and
so,
if
anyone
has
better
ones,
I
would
love
to
hear
them.
A
D
Yes,
they've
been
the
hardest
to
ever
get
going
green
and
I'm
tired
of
spending
time
on
them.
E
E
Specifically,
I
I
looked
this
afternoon
and
for
the
eviction
for
the
disc,
and
I
don't
know
it
looks
like
there
is
like
disc
pressure
for
like
two
minutes
and
then
the
disc
pressure
disappears,
even
though
the
the
big
disc
hog
container
still
has
the
same
size.
So
I
I
don't
understand
it's
resizing
the
the
available
space.
I.
C
Don't
know
how
these
tests
work,
but
I
know
that
probably
an
easier
way
to
make
them
be
more
like
I'm
not.
I
suspect
that
they're.
Actually,
I
know
that
I've
seen
on
some
of
them.
They
actually
try
to
fill
up
the
whole
disk
and
I
think,
a
better
way
to
do
this
would
be
and
more
reliable
way
to
do.
This
would
be
to
like
make
a
small
partition,
maybe
a
temper
fest,
maybe
on
disk.
I
don't
think
it
really
matters
that
much
and.
D
Disk
pressure,
if
you
point
the
cube,
so
that
is
what
I
was
saying
like
partially
like
what
they
do
right
now
is,
rather
than
filling
up
the
whole
disk.
Is
they
figure
out
how
much
free
disk
space
there
is
at
the
start
of
the
test,
minus
four
gigabytes
from
that
and
then
try
and
fill
up
10
gigs
of
space.
D
Which
is
horrible
for
many
reasons
partially
because,
like
if
you're
running
them
on
your
local
desktop
or
whatever
other
stuff
on
your
system
is
gonna
purge
disk
or,
like
you
know,
stuff
touches
disk
all
the
time
and
like
stuff
changes
which
makes
them
kind
of
really
flaky
part
of
what
I've
noticed
with
that
is.
If
you're
running
the,
if
you're
only
running
the
storage
tests,
they
only
fail.
One
in
like
200
runs
when
you
run
all
of
the
eviction
tests.
D
They
will
sometimes
fail
and
we
get
very
unlucky
on
ci
and
I
am
not
quite
sure
what
makes
them
fail
more
frequently
in
kci.
Yet,
given
that
I'm
running
the
same
image
configs,
it's
a
fun
time.
D
But
then
it's
not
an
end-to-end
test
right
yeah,
so
the
problem
is
like
the
ets.
They
kind
of
have
to
you
know,
go
through
the
whole
thing,
which
is
why
I'm
probably
going
to
end
up
doing
a
horrible
temper
fast.
C
I
don't
think
that's
that
horrible.
I
think
that
makes
sense.
I
think,
like
I
think
it's
a
good
idea
and
kudos
to
you
for
coming
up
with
it.
Fundamentally,
I
think
these
tests
have
just
been
broken
by
design
since
they
landed
and
nobody's
come
back
to
spend
the
time
on
them,
because
you
know
the
tests
landed,
they
mostly
worked,
they
weren't
in
any
blocking
lane
and
everybody
forgot
about
them.
I
think
it's
probably
a
good
summary
of
what
happened
and
it
would
be
cool
if
they
worked
because
they
yeah
they.
G
C
D
So
yeah,
like
I'm,
probably
gonna,
end
up
rewriting
most
of
the
eviction
tests
right
now.
They
all
use
a
very
dry
like
common
thing,
to
actually
run
them
that
makes
ever
debugging
or
fixing
them
really
hard
because
it
means
to
change
one
test.
You
have
to
change
all
of
the
eviction
tests
and
it
also
means
that
to
understand
one
eviction
test.
You
have
to
understand
all
of
them,
which
is
not
particularly
productive.
C
D
Yeah
like
right
now,
it's
just
like
the
cognitive
load
of
trying
to
debug.
Any
of
them
is
horrible.
They're
also
a
way
way
too
lucky
with
things
that
become
noise,
as
opposed
to
useful.
I.
A
Yeah,
I
wonder
if
it's
the
same
increase
as
we
saw
before
when
it
goes
up
and
then
a
few
like
50
days
later
it
goes
down.
Let
me
take
a
screenshot
and
then
double
check
next
week.
A
G
A
This
is
done.
Okay,
let's
go
to
the
dashboard.
A
I
wonder
if
it's
yeah,
it's
a
code,
change
product
change,
the
small
test
addition.
So
let's
keep
it
on
the
main
board
and
remove
from
city
board.
A
I
think
these
are
current
tickets
yeah,
it's
current,
but
they
they're
mostly
product
change,
and
this
music.
We
try
to
look
at
metric
test,
intro,
okay,
yeah.
A
That
means
that
they're
not
important
to
just
this
meeting,
concentrate
on
tests.
E
Yeah,
actually,
because
what
I
was
trying
to
do
doesn't
work
because
I
explained
there
and
I
asked
david
to
look
at,
but
he
didn't
reply.
E
E
Basically,
the
the
version
of
bus
ctl
that
is
available,
there
is
too
old
and
it
cannot
emit
signals
in
the
bus,
isn't
fedora.
F
A
A
Minimal
checkpoint
in
support-
oh
checkpoint,
it's
cap,
related
yeah,
so
many
pr's
just
need
to
be
filtered
out
here.
A
C
A
Okay
gt
sends
this
pr,
and
this
is
for
gci
folder.
I
think
yeah.
I
already
picked
walter
and
dawn,
so
I
should
yeah.
Let's
wait
for
them
to
improve.
G
A
Okay,
this
eviction
that
was
just
discussed
daniel,
I
okay
mia
signing
to
you.
D
Sure
I'll
take
a
look
at
cryo
stuff
too.
It's
probably
the
same.
A
Okay,
this
is
interesting
box.
This
is
alpha
dashboard,
like
alpha
grid,
running
non-alpha
tests.
A
Should
be
easy
change
and
then,
like
I'm,
not
sure
what?
Why
is
this
failures
here,
but
first
thing
to
fix
is
run
proper
tests,
so
name
will
match
the
definition.
A
A
C
G
Yeah,
the
other
bug
is,
is
on
my
plate
and
I
have
been
hoping
to
find
some
time
to
go
through
my
ever
growing
queue
of
upstream
issues
with
the
test.
So
hopefully
I
can
get
some
time
this
year
next
week
to
find
the
issue
and
it'll
get
two
birds
with
one
stone.
A
Okay,
I
think
we're
done
here
and
I
I
went
through.
I
need
review
it's
all
assigned,
so
I
guess
not
much
time
we
have
upstream
this
week
for
specific
tests.
So
let's
keep
it,
as
is
okay
without
a
new
last
call
for
test
related
stuff,
and
then
we
go
to
back
trash.
A
Going
once,
okay,
let's
move
to
puck
trash.
A
A
A
Let's
see,
storage
and
accept
trash
because
it
looks
like
real
issue
and
we
already
have
a
pr
opened.
A
I
don't
think
clayton
addressing
this
specific
issue
in
this
pr,
so
I'm
not
assigned
to
him,
but
I
should
beat
russian.
A
A
Now
also
importance,
so
you
know
fix
it
in
this
release,
since
it's
regression.
C
That's
triage,
remove,
remove
triage.
A
H
H
You
can
leave
the
trash
need
information
like
it's
always
good,
to
have
free
producer
for
the
latest
versions
and
par
in
part.
I
will
just
check
it
anyway.
It's
not
too
big
threat
under
the
issues.
E
A
C
They
don't
have
their
system
reserved
anywhere;
maybe
their
system
reserved
is
small
or
sorry.
Maybe
their
system
reserves
large.
The
actual,
like
amount
available
for
cube,
is
small.
A
C
Yeah
well,
it
definitely
says
they're
claiming
there's
only
cube
system
stuff
on
there
and
that
says
yeah
eviction
threshold.
I
bet
that
there's
some
workload,
that's
like
spinning
in
the
middle
of
getting
evicted.
I
agree.
C
C
C
A
Oh
yeah,
it's
just
daniel
fiction.
There
is.
C
Oh
no,
if
it's
completed
it
will
never
come
back
to
life.
Restart
always
is
only
on
a
failure,
so
I
don't
think
this
person.
Well
I
mean
that
the
crash
loop
back
off
that
looks
like
a
failure.
It
shouldn't
go
from
completed
to
crash
loop
back
off,
but
I
don't
think
that's
what's
happening.
C
It
should
go
from
running
to
crash
loop
back
off
back
to
run,
I
mean
crash
loop
back
off,
isn't
actually
a
status
it's
in
running,
but,
like
cube,
cuddle
will
display
crash
loop
back
off
if
that's
what's
happening.
It
should
never,
however,
go
to
complete
it
back
to
running
so
you
can
assign
this
one
to
me.
I
don't
think
that
this
person
understands
the
core
life
cycle
of
pots,
which
is
fine.
It's
confusing.
Maybe
we
should
write
a
doc.
A
A
E
C
C
I
mean
it's
only
supposed
to
happen
on
like
initial
pods
start
up
so
like.
I
would
expect
that
that
would
happen
if,
like
the
pod,
restarted,
that
it
would
run
again
because
it
basically
I
mean
the
point
of
the
startup
probe-
is
that
it
suppresses
the
liveness
probe
when
it's
not
ready
for
a
liveness
probe
so
like
that
would
happen.
If
you
know
a
container
died
and
got
recreated,
I'm
not
really
sure
what
they
don't
understand
here.
C
119
is
not
and
120,
I
think
we
have.
The
last
patch
release
is
happening
like
this
week
is
the
cutoff
kind
of
thing
so.
C
C
I
mean
if
it's
broken
it
upstream
then
yeah
like
you're
here
is
fine,
but
you
know
if
they
want
support
on
an
older
version,
then
they
always
submit
a
bug
to
red
hat.
C
D
Please
don't
ask
me
for
help
here.
Please
ask
me
for
help
in
this
other
issue.
Tracker
is
certainly
some
kind
of
response.
C
To
address
danielle's
previous
comment,
I
mean
sometimes
the
people
overlap,
but
sometimes
they
don't
so
in
this
case
they
got
lucky.