►
From YouTube: Kubernetes SIG Node CI 20230517
Description
SIG Node CI weekly meeting. Agenda and notes: https://docs.google.com/document/d/1fb-ugvgdSVIkkuJ388_nhp2pBTy_4HEVg5848Xy7n5U/edit#heading=h.2v8vzknys4nk
GMT20230517-170443_Recording_1822x1020.mp4
A
Well,
hello,
it's
a
Wednesday,
May
17
2023
signal.
Ci
meeting
welcome
everybody
today,
I
wanted
to
start
agenda
with
a
few
discussions
and
highlights
first
of
all,
I
want
to
point
to
this
stress
test
that
is
being
added.
It's
not
quite
test,
it's
not
a
like
I
mean
it
doesn't
stress
entire
machine.
It's
more
I
mean
it
does
stress
into
our
machine.
It
tries
to
make
machine
run
out
of
circuits.
A
Out
of
sockets,
because
we
had
this
Improvement
recently
done
on
probes
implementation,
so
this
will
not
kill
the
machine,
but
it
potentially
can
and
I
think
this
is
one
of
the
few
tests
that
has
this
effect
as
it
does.
We
have
eviction
tests
that
also
kind
of
flaky
and
depend
on
the
environment,
but
the
eviction
test
intentionally
make
machine
go
into
bad
State
and
that's
why
we
have
them
flaking
that
much
this
test
tested
machine
will
not
get
into
the
state
so
suppose
it
will
be
more
stable
and
better
behaved.
A
Any
questions
on
stress
test,
okay
and
I
also
wanted
to
start
collections
more
of
information
of
what
kind
of
areas
we
improve.
Testing
in
I
know
that
you're
working
on
arm
tests
that
are
currently
two
PR
for
an
intestine
frame,
one
in
to
make
a
little
Implement
arm
tests
on
in
open
source.
So
I
know
that
vendor
is
running
some
arm
tests,
but
I
also
want
to
make
sure
that
OSS,
like
Upstream,
has
good
coverage
and
since
Google
Cloud
now
supports
our
machines
and
we
also
have
AWS
right
so
AWS
has
graviton.
A
So
maybe
some
of
the
test
cases
from
this
cap
from
this
PR
can
be
run
on
AWS
in
future,
but
nevertheless,
like
I
think
it's
a
good
idea
to
run
it.
So,
hopefully
we'll
get
this
to
the
finish
line
and
I
will
have
Auto
coverage.
A
Also
I
wanted
to
reiterate
on
the
Standalone
coupon
test
that
was
added
last
release.
So
what
happened
is
there
are
a
few?
My
Standalone
mode
is
very
specific
in
Standalone
mode,
coupler
doesn't
have
client
initialized,
so
it's
now
on
new
in
golang
and
if
it's
new
it
can
cause
all
sorts
of
panics.
A
Like
people,
assuming
that
there
is
a
connection
to
API
servers,
so
they
assume
that
it's
not
new
and
try
to
call
it
periodically
from
one
place
on
another
and
especially
it
happens
under
feature
gates
in
Alpha
features,
because
in
Alpha
features
you
have
less
of
a
coverage
and
more
Standalone
mode.
A
Couplets
are
not
running
in
Alpha
with
Alpha
features,
so
these
two
are
I
added
two
tests
in
CI
to
catch
regressions
in
previous
cities,
but
I,
never
added
them
in
video
objects,
so
I
think
it's
time
to
add
them
in
periodics
and
random,
like
maybe
once
a
week
would
be
enough,
maybe
more
often,
but
you
need
to
see
a
test
coverage
and
try
to
catch
early.
Any
problems
with
new
caps.
A
As
I
said,
we
also
have
multi-numa
in
periodics.
Swati
I
think
you
added
some
right,
so
they
editing
CI,
but
I
think
we
also
need
to
move
them.
In
periodic
say
they
were
quite
clean,
but
now
they
never
run
for
a
lot
for
a
while
outfit
test
grid
and
zero
like
there
is
no
recent
executions.
B
A
A
Okay,
so
I
think
these
three
are
new
areas
of
tests
are
very
good
if
you
can
get
them
from
Finish
like
I
mean
I.
Just
this.
A
A
B
A
B
So
maybe
it's
worth
sharing
the
pr
and
I
kind
of
made
a
few
changes
related
to
that,
but
but
I
think
in
general.
This
is
useful,
because
if
we
have
some
jobs
on
Master
Branch,
we
don't
essentially
enable
them
directly
on
our
release
branches.
All
these
are
disabled
by
default,
so
it
doesn't.
Actually
we
don't
lose
anything
by
enabling
them,
but
we're
just
adding
them
as
release
per
4
gives
us
the
ability
to
enable
if
you
want
to
do
it.
So
here
is
an
example:
I
posted
on
char.
D
Know
I
have
a
comment
on
this
option
or
come
on
so
I
I.
Look
into
this
and
I
think
it's
good,
but
the
the
potential
issue
that
could
happen
is
that
where
we
will
be
testing
with
old
image
versions
for
for
certain
release,
branches
or.
D
With
with
all
release
branches,
for
example,
for
testing
the
new
I,
don't
know
128,
when
it's
got
eventually,
it
might
cost
it's
not
bad.
We
can
also
have
it
like
that.
B
I
think
for
yeah
I
think
for
the
older
versions.
We
have
to
explicitly
add
them,
but
for
the
future
release
branches
I
think
it
happens
automatically.
So
we
don't
have
to
essentially
go
ahead
and
add
more
jobs
for
new
branches.
B
Yeah,
so
if
you
kind
of
go
through.
B
I'm,
sorry
I'm
not
sure
I'm,
just
losing
connection,
maybe
yeah.
So
in
this
case,
what
I
was
trying
to
do
was
enable
and
add
a
few
branches
for
sorry,
a
few
jobs
for
certain
release,
branches
and
Marco
who's,
I
believe
the
release
team
member.
He
recommended
that,
rather
than
doing
that,
you
know
like
in
addition
to
doing
that,
I
should
say
you
can
add
this
option
and
it
ensures
that
for
upcoming
releases,
this
job
has
the
option
of
running
these
jobs.
D
Right,
I
think
it's
good
to
start
with
a
four
gallon
later
on.
If
we
decide
we
need
a
specific
configuration
for
the
job.
We
can
create
a
new
one
yeah,
but
we
can
use
the
default
fork
for
most
cases
and
then
it
probably
will
catch
most
of
the
cases,
but
yeah
I
think
that's
a
good,
a
good
point.
Let
me
review
that
again.
A
A
B
No
I'm
not
proposing
to
change
something
I
think
in
in
terms
of
processes,
for
us
to
have
ability
to
run
certain
jobs
for
release
branches.
This
option
comes
handy
for
future
releases
like
right
now,
I
was
essentially
going
and
adding
the
adding
branches
for
specific
release
branches.
But
you
know
a
release
later.
You
would
need
to
add
another
job
for,
say
128.,
so
this
option
allows
us
to
automatically
create
or
have
access
to
jobs
for
upcoming
releases
and
it's
more
future
proof.
B
A
Makes
sense
one
potential
problem,
though
it
it
will
use
newest
images
right.
So
whenever
we
update
image,
we
will
need
to
run
version
of
kubernetes
like
three
versions
ago,
with
a
image
that
is
very
fresh,
so
it
potentially
can
introduce
some
unexpected
behaviors.
B
D
A
Okay,
in
the
same
vein,
I
think
Mike,
you
added
the
previous
release.
Periodics
right.
D
D
E
A
Yeah,
so
to
finish
like,
can
you
convert
this
in
periodics
sure.
B
A
C
B
A
Yeah
I
think
it's
a
good
achievement
recently
improved
coverage
quite
significantly.
A
Okay,
with
that,
let's
go
to
Performance
I,
always
check
performance
before
this
meeting,
but
I
don't
I
want
to
do
it
together.
I
think
it's
more
fun
this
way
so
yeah,
so
this
is
Google
doing
good.
So
far,
it's
see
okay.
So
let's
check
on
memory
yeah
still
good
and
runtime.
A
A
Okay,
then
go
to.
A
They
are
pre-opened
to
the
Box.
This
is
Dray
still
work
in
progress,
so
I
will
move
it
into
waiting
on
author.
A
Okay,
so
expect
oh
expect
Mega
equal.
F
A
G
It
looks
like
it's
marked
as
deprecated
in
code.
E
E
C
G
C
G
A
Okay,
so
we
will
expect
the
flood
of
this
refactoring.
A
Expected,
okay,
all
right
citizen
I
didn't
know
about
it.
E
A
Again,
oh,
it's
complete
as
well.
A
A
A
A
A
A
I
had
a
sick
testing
process
as
well,
because
it
seems
to
be
some
problem,
is
test
definition
so
like.
According
to
this
look,
it's
it's
failing
to
initialize
some
project
or
something.
A
Okay,
let's
keep
it
active.
C
F
E
A
Oh,
this
is
coming
to
the
two
cents
Mighty,
so
yeah
we're
done
with
this.
This
needs
to
be
out
of
this
project
because
it's
not
signaled.
A
Let
me
look
at
the
waiting
on
author.
We
have
like
23
here.
An
easier
base
needs
a
base.
Rebase
work
in
progress.
A
A
A
A
E
A
Progress
yeah
this
we
just
looked
at
okay,
Windows
tests,
new
windows,
test,
wise
service,
monitor.
A
A
So,
let's
use
it,
the
person
made
it
work.
G
I
thought
so
yeah
I
mean
there's
some
internet,
that's
running
yeah.
The
point
of
this
one
was
to
allow
something
and
tests
run
on
AWS
as
well
and
use
some
of
those
credits
which
is
occurring
now
so
I'm,
not
sure.
There's
anything
remaining
here.
You
can
sign
me
and
I'll
have
a
look
at
it
and
I'll
think
Mohammed
ask,
but
maybe
I'll
go
just
be
opposed.
A
Yeah
we
may
need
to
review
which
test
to
switch
I'm
curious,
like
I,
hear
about
the
CWS
credits
use
for
a
while
and
I
I
wonder
if
it
is
a
status
update
or
that
maybe
you
can
tell
next
meeting
you
know
more
information.
Yeah
yeah
I'll
find
that.
Thank
you.
A
Okay,
do
you
know,
do
you
remember,
what's
the
status,
it
seems
that
there
are
some
tags
missing
right
on
a
job
definition.
B
I'll
take
a
look
at
it
today.
The
tags
are
in,
but
the
jobs
were
removed
that
need
to
be
added
back.
Okay,
so
I'll
take
a
look
at
it
today.
E
No
I
did
yeah
I.
Sorry
not
yet.
C
A
E
G
E
A
A
I
think
we
out
of
things
for
test
part
of
the
meeting
we
may
go
to
the
box
or
just
one.
This
is
new.
E
E
A
A
So
it's
126,
so
it's
up
to
date,
enough
I
remember:
we
had
some
bugs
with
a
secret
config
Maps
not
being
downloaded
and
remember
why,
but
it
doesn't
feel
the
same
something
different
and
yeah
I
agree
with
you
like.
Why
won't
you
stop
watching
Quest
secret
update
okay,
so
this
is
done
of
course
here.
A
Since
we
reached
everything,
let's
go
to
need
information
and
see
if
we
can
fish
out
something
out
of
it.
A
I
think
yes
and
two
Francesca.
A
A
Called
long-term
priority
seems
to
be
okay
matching
the
expectation
and,
let
me
see
if
Francesca
is
actually
assigned
to
this
yeah.
Here's
a
sign.
A
Okay
personally
reply
to
one
question,
but
not
the
other.
E
F
G
A
E
A
C
A
Yeah
I
think
this
is
a
issue
everybody
just
plus
one
on
that.
So
the
request
here
is
sometimes
clear:
restart
counter
and
I'm
still
waiting
for
a
good
explanation.
What
the
scenario
is
like
how
One
will
know
that
the
start
counter
needs
to
be
cleaned.
A
A
A
A
Okay,
students
information,
so
we
have
I
think
it's
time
to
write
an
article
like
how
you
can
get
into
out
of
CPU.
State
Jordan
has
like
good
day
I
wonder
if
you
can
put
this
into
chat,
GPT
and
say:
write
a
blog
post
like
documentation
listing
this
CCA
scenarios.
We
will
write
a
useful
blog
post.
A
But
yeah,
nevertheless,
we
have
cases
of
out
of
CPU
on
kumblet
those
are
listed
here
and
it
may
be
a
new
case,
but
I
don't
know
how
effectively
applicants
can
affect
it.
I
mean
scheduler
should
handle
it.
Well
anyway,
let's
move
on.
A
C
A
Yeah
but
feel
free
to
move
it
the
way,
but
on
this
information,
sure.
A
A
Ten
minutes
left,
let's
take
a
couple
more
and
we'll
be
done.
E
E
A
A
Thank
you.
So
we
have
this
see
advisor
thing.
So
this
not
super
performant
and
apparently
it's
leaking.
F
C
A
E
A
A
Okay
I
suggest
we
stop
on
that
finish,
a
meeting
anything
else
for
today.
Coin
once
for
advice,
thank
you.
Everybody
have
a
great
rest
of
the
week
bye,
bye
thanks
for
everyone.
Thank.