►
From YouTube: Kubernetes SIG Node CI 20230607
Description
SIG Node CI weekly meeting. Agenda and notes: https://docs.google.com/document/d/1fb-ugvgdSVIkkuJ388_nhp2pBTy_4HEVg5848Xy7n5U/edit#heading=h.2v8vzknys4nk
GMT20230607-170532_Recording_1840x1020.mp4
A
A
Okay,
first
item:
do
you
see
it.
A
You
so
first
I'm
Paco
wanted
to
join,
but
he
didn't.
Maybe
we
go
with
the
second
item
first
and
then
go
back
so
arm
and
multi-numa.
This
is
what
you
created:
a
pull
request.
B
A
B
I
did
that
today,
I
just
added
three
jobs
for
CPU
manager,
memory
manager
and
topology
manager,
and
it
basically
uses
multinomial
test
infrastructure
I
believe
we
had
added
that
a
few
months
ago
when
we
were
graduating,
topology
manager
to
ga.
So
now
we
should
have
signals
periodically.
Just
take
a
look.
I
wouldn't
call
myself
like
the
most
proficient
with
tests
and
try
jobs.
So
I
would
appreciate
if
people
can.
Let
me
know
if
everything
looks:
okay,.
A
Great,
we
just
went
through
same
exercise
with
Standalone
tests,
so
it
wouldn't
be
that
yeah.
B
It
was
yeah,
it
was
very
similar,
like
all
the
arguments,
and
everything
was
similar
other
than
a
few
fields
that
had
to
be
populated
differently.
A
A
And
I
checked
arm
tests,
they're
still
failing.
Let
me
get
the
link
so
yeah,
something
is
broken
with
a
startup
I
need
some
I
know
that
a
few
people
already
looking
at
me.
We
have
an
issue
with
this.
If
it
wouldn't
be
fixed
soon,
it
needs
to
be
I.
Will
I
will
try
to
find
people.
A
Okay,
so
let's
go
to
this
Uber
issue:
James
periodically
created
these
issues
on
permafiling
jobs,
and
this
time
it's
not
the
next
exception,
so
to
do
yeah.
A
A
C
A
C
A
A
Okay,
something
under
sick
testing,
I'm,
not
sure.
Okay,
let
me
put
an
action
item
on
my
agenda
to
follow
up
on
ownership
and
why
we
have
this.
C
A
A
D
A
D
A
Yeah
I
think
that's
that's
it.
We
only
need
to
charge
this
one
and
if
Ryan
cannot
work
on
this,
I
need
to
reassign.
Okay,
let's
go
to
the
TRS,
then.
B
Yeah
so
recently
we
came
across
an
issue
where,
after
cubeletree
start,
we
were
noticing
unexpected
Behavior,
so
that
led
me
to
use
sample
device
plugin
for
testing
cubeletal,
restart,
behavior
and
I
noticed
that
the
device
plugin
wasn't.
This
is
a
sample
device
plugin
that
we
have
in
tree
and
it
wasn't
re-registering
itself
after
cubed
restart,
and
that
was
because
we
were
only
allowing
it
to
register.
B
Once
I
made
a
couple
of
changes
just
to
make
sure
that
it
we
have
the
ability
of
re-registering
once
you
know
the
device
plugin
once
the
cubelet
was
restarted,
or
you
know
this
could
be
any
other
scenario
like
node
reboot
and
things
like
that.
So
that's
the
kind
of
rational
for
the
change
and
we
are
currently
looking
into
you
know
some
of
the
changes
that
we
made
as
part
of
device
device
manager
to
make
sure
that
from
tubulatory
start
perspective,
everything
is
in
order.
B
B
E
B
Is
when
there's
a
register
control
file
specified,
and
that
is
like
a
trigger
file
in
in
the
sample
device
plugin
that
triggers
registration,
and
this
is
used
for
cases
where
we
don't
want
the
device
plugin
to
register,
and
this
was
useful
in
some
of
the
end-to-end
tests
that
we
had
implemented
and
and
by
default
we
want
to
ensure
that
the
device
plugin
registers
itself
so
the
second,
the
else
part
of
it
is
corresponding
to
that.
A
C
A
A
So
what
is
happening
here
is
Google
Disturbed
and
then
some
actions
happening
like
some
registration
and
registration
of
pods
and
then
Google
starts
again.
So.
A
That
there
will
be
a
period
of
time
when
couplet
is
not
running
I.
Remember
Francesco
was
saying
that
there
are
some
cases
when
test
is
running
with
kublet
as
a
system
demon,
so
Cobalt
will
restart,
even
if
we
like
just
killed
him
with
a
stop.
A
If
not
I
can
just
comment
here.
B
A
E
C
A
C
A
B
Yeah
actually
I'm
not
sure
I
was
talking
in
the
context
of
device
plugins,
but
it
could
be
the
same
thing.
This
is.
This
is
Dr.
A
A
Oh,
this
one
is
flaking.
C
A
A
C
C
A
A
It
was
fish
to
cryo.
Okay,
that's
why
we
can
find
it.
A
Could
I
think
we've
done
with
a
test
side
of
it
things.
E
A
So
if
there
is
a
port
that
wasn't
admitted
in.
E
A
Race
for
germination,
Heats
Port
is
now
deleted
and
Demon
said,
can
create
a
new
pod
because
it
I
think
that
one
this
one
still
exists.
C
A
A
C
A
A
B
I,
if
I
remember
correctly,
at
least
in
device
manager,
when
you
have
a
checkpoint
file,
there's
no
update
at
the
time
of
deletion,
there's
only
update
at
the
time
of
creation
and
see
if
we
had
a
subsequent
part
that
was
requesting
CPUs.
The
checkpoint
file
would
be
updated
later
that
that's
my
understanding.
I
can
take
a
look
at
this.
B
A
C
C
A
Yeah
I
wonder
if
we
can
check
that
C
group
was
updated.