►
From YouTube: Kubernetes SIG Node CI 20230118
Description
SIG Node CI weekly meeting. Agenda and notes: https://docs.google.com/document/d/1fb-ugvgdSVIkkuJ388_nhp2pBTy_4HEVg5848Xy7n5U/edit#heading=h.2v8vzknys4nk
GMT20230118-180528_Recording_1510x1120
A
A
B
Yeah,
so
this
is
related
to
a
device
manager
bug
that
I'm
working
on
and
as
part
of
this
issue,
what
is
happening
is
that
we
have
no
control
over
how
device
plug-in
pods
or
the
application
pods
consuming
those
devices
are
recovered.
So
there
could
be
scenarios
where
an
application
part
comes
up
before
the
device
plugin
pod
and
it
it
returns
and
leads
to
unexpected
Behavior.
Essentially,
what
we
see
is
we
don't
see
any
kind
of
error
and
at
the
end
you
know
when
the
application
Port
tries
to
access
the
device.
B
That's
when
the
error
kind
of
manifests
itself
so
as
part
of
kind
of
debugging
this
issue,
what
I
did
was
that
we
kind
of
made
sure
that
before
an
application
fault
comes
up,
we
have
an
existing
device.
Plugin
board
a
healthy
device
exists
on
the
system,
but
in
order
to
reproduce
this
issue
and
to
write
up
an
end-to-end
test,
I
had
to
make
some
changes
to
the
sample
device
plugin,
which
is
the
device
plugin
implementation
that
is
used
for
testing
and
what
I
did,
as
part
of
that
was
introduce
an
environment
variable.
B
So
you
know
Auto
register
by
default.
We
keep
it
to
be
equal
to
True,
which
which
keeps
backward
compatibility,
and
we
get
the
exact
behavior
that
we
was
getting
viewed
that
we
were
getting
previously.
But
for
this
particular
case
we
assign
that
variable
to
false
and
the
device
plugin
essentially
doesn't
register
itself.
So
I
think
that
is
kind
of
the
background
of
this
issue
and
what
I've
done
as
part
of
the
end-to-end
test
work.
B
What
I'm
trying
to
get
get
to
is
I
need
some
guidance
on
how
we
can
publish
now
this
device
plugin
image,
given
that
we've
made
some
change
to
it.
We
need
to
consume
this
as
part
of
the
end-to-end
test
for
this
PR,
so
I
I
thought
that
would
be
best
that
we
split
the
sample
device.
Plugin
changes,
but
I
came
across
a
kubernetes
image
promotion
process
that
says
that
maintainers
are
responsible
for
pushing
images,
so
I
kind
of
need
some
help
there
or
maybe,
if
others
are
aware
of.
A
Scenario,
I
know
for
sure
it's
a
credential
provider
what
Andrew,
updated
and
I
think
he
just
submitted
a
PR,
and
then
somebody
from
sick
testing
picked
up
new
images.
I
constantly
see
this
PRS
that
affects
inventory
testing
from
saying
bump
of
test
image
version
so
that
maybe
that
might
be
happening
automatically
I'm
just
not
sure,
but.
A
Absolutely
so
so.
A
Maybe
I'm
mistaken,
maybe
he
put
it
in
in
a
bundle
like
a
so.
Let's
see.
A
It's
not
that
yeah,
maybe
I'm
mistaken
I
I
thought
this.
Is
it.
C
A
I,
try
to
remember
is
maybe
it's
going
into
zip
file
like
end
to
end
Z
file
and
I'm,
not
sure
if
it's
delivered
as
an
image
is
not
or
not.
Okay.
I
can
give
you.
C
A
D
Swati,
are
you
talking
about
this
image
that
that's
used
in
a
job
like
that.
B
One
sec,
no
actually
I'm,
referring
to
so
the
image
that
is
used
as
part
of
an
application
or
a
payload.
Actually,
maybe
this
could
be
something
similar
as
well,
so
you
know
like
say:
device
plugins
or
even
nginx,
for
example,
like
we
refer
to
that.
A
Then
I
have
another
example.
I
thought
this
image
that
builds
like
entire
end-to-end
machine.
A
D
Which
is
required
to
run
the
test
cases,
your
commit
gets
merged,
and
then
you
just
wait
for
this
image
to
change
periodically
for
all
jobs
and
your
changes
will
be
there.
B
Okay,
so
then
that
means
like
what
I
didn't
hear
about
splitting
the
PRS
into
two
isolating
the
sample
device.
Plugin
changes
into
a
separate
PR
makes
sense
because
we
want
that
to
get
merged,
and
once
that
goes
through,
maybe
the
infra
processes
we
get
the
image,
and
then
we
just
have
to
update.
Maybe
the
tag
or
something
like
that.
You.
D
A
Would
be
yes
find
it?
Okay,.
A
Well,
I
think
the
first
question
is
how
to
up
upload
a
new
version,
so
this
is
where.
D
B
Who
does
that
I
I
see
only
kubernetes
staging
test?
Infra
images
are
being
updated.
So
what
I'm
looking
for
is
the
like
the
registry.khs.io
e
to
e
test
images.
B
B
Yeah
yeah,
so
so
I
think
the
only
difference
I
think
here
is
that
I'm
not
looking
for
an
image
pertaining
to
a
job
per
se.
This
is
kind
of
like
an
application
image.
If
that
makes
sense,
this
is
an
image
that
is
referred
to
in
the
end-to-end
test.
You
know
like,
for
example,
if
you
have
a
test
port
and
you
have
nginx
that
you
use
as
an
image
for
running
the
test.
Part.
C
A
My
Android
test
like
is
this:
what
what
I
did
for
a
pair
of
tests?
So
there
was
an
image
here,
a
perf
test,
an
image
quota,
TF
white,
deep,
so
I
updated
it
in
KK,
first
and
then
I
found
PR
from
Brian
I
did
similar
one.
So
this
promotes
this
image
by
specifying
its
quit
here
in
this
file.
So
I
would
assume
that
this
sample
device
plugin,
is
also
here
somewhere.
A
A
D
A
You
do,
is
you
submit
the
pr
here
with
a
new
version
and
yusha
and
it
will
be
promoted
automatically.
B
Okay,
cool,
please
yeah,
if
you
can
put
in
those
examples,
that'd
be
great.
Thank
you.
A
Yeah,
those
are
not
much
an
example,
but
close
enough.
I
just
didn't
find
exactly
one
that
promotes
changes
from.
A
Okay,
I'm
glad
it
was
answered
and
yeah
as
you
expected
sample,
I
I,
don't
know
like
we
discussed
a
device
plugin
last
time.
Francesco
was
here
and
you're
here
as
well,
hello,
Francesca
and
I.
Doubt
we
would
be
able
to
implement
this
end-to-end
test
with
idea
device
plugin,
it
may
be
harder
right.
I
would
assume
your
change
is
somehow
artificially
make
a
device
plugin
not
start.
B
Exactly
so,
basically
I'm
kind
of
waiting
on
some
sort
of
an
event,
essentially
a
unique
socket
for
an
event
on
that,
and
only
after
we
get
that
we
proceed
towards
registration.
So,
like
you
said,
it's
an
artificially
created
scenario
where
you
kind
of
pause:
the
auto
registration
of
the
device,
plugin.
A
Thank
you
anything
else
for
today
as
a
agenda
item
or
you
go
to
triage.
A
A
This
is
something
you
want
to
be
reviewed,
or
it's
still
work
in
progress.
E
Hey
you
are,
you
should
actually
ignore
it,
because
I
need
to
talk
with
Sig
Windows,
first,
okay,
the
context
with
this
PR
is
that
there
is
a
fairly
convoluted
state
with
respect
to
products
with
disappear.
We
cannot
get
rid
of
the
feature
gate,
because
windows
need
the
fissure
gate
to
disable
the
feature.
E
Otherwise
it's
gonna
crash,
so
I
need
to
sort
out
this
before
we
can
proceed
further
with
the
for
the
success
API
graduation,
so
I
need
to
make
understand
how
what
is
the
best
scenario
for
Windows
and
then
I
will
ask
a
signaled
review
to
say:
okay,
has
it
any
impact
on
signaled?
Do
we
need
any
anything
else,
but
for
now
I
will
contact
Sig
windows.
Thank
you.
A
A
Okay,
I
will
assign
and.
A
Okay:
support
to
control
registration
process,
it's
so
what
is
it
what
you
were
talking
about.
B
That's
exactly
what
else
working
about,
maybe
if
you
can
assign
it
to
someone
that
would
be
great
help.
I
think
Francesco
is
well
versed
with
this
area,
so
maybe
yeah.
A
What's
your
new
audience
right,
yes,.
A
I'm
very
interested
to
review
this.
It
will
change
it's
mostly
like
log
improvements.
A
Oh,
thank
you
I
think.
Maybe
chick
testing
also
will
take
a
little
bit.
A
A
Does
anybody
know
what
the
component
SLA
SLI
is
service
level.
C
Yes,
ready
for
a
review.
C
B
A
Because
I'm,
looking
at
this
logic
right
now
for
sidecar
containers,
cap.
C
C
A
I
think
we
have
approval
permissions
for
this
folder.
So
what
did
you
use.
A
Yeah
I'll
put
it
in
a
reviewer
needed
News
Review,
but
maybe
basic
windows
will
I'll
just
jam.
It.
A
Oh
yeah,
maybe
that
one
was
fixed
with
changing
the
process
with
it.
A
Yeah
because
we've
been,
we
wasn't
waiting
for
Port
to
to
be
started
and
then
we've
been
looking
for
port
and
sometimes
it
was
existing
from
other
tests
and
that's
what
we
read
unscore
from.
A
A
Okay,
I.
A
A
B
A
C
I
also
have
to
drop
for
a
conflict
if
anything
comes
up,
just
assign
it
to
me.
Thank
you.
D
One
more
test
that
is
getting
that
got
added
recently,
it's
failing
with
not
being
able
to
find
the
GCB
project,
so
I'm
interested
in
this.
D
Yeah
something
similar,
so
we
are
trying
to
add
a
new
node
E3,
relatively
evented
plate
that
we
added
and
everything
seems
fine,
but
the
job
field.
This
thing
unable
to
find
the
gcp
project
so
I'm
going
to
see
whether
this
is
somehow
related.
A
Cool
okay:
let's
go
quickly
to
bug
triage,
see
how
many
bugs
you
have
18
is
one
too
many
foreign.
A
A
I
assigned
this
to
myself
and
I
was
looking
at
a
little
bit
different
incarnation
of
the
same
issue.
So
I
will
be
updating
this
bug
yeah.
There
is
something
weird
happening
with
volumes
that
was
assigned
and
then
another
scientist
and
trying
to
be
mounted
again.
So
maybe
it's
the
same
for
config
Maps
well,
I
will
be
looking
at
that.
A
D
A
Haven't
okay,
yeah
because
stomach
if
fish
is
happening
and
some
of
examples
I
saw
is
the
docker
shim
is
working
and
the
continuity
is
not
I,
don't
know
about
cry.
A
Yeah
this
one
similar
but
different
and
yeah
again
I
I,
don't
know,
was
its
real
bug.
I
will
just
keep
it
around
in
Andreas
section.
A
A
Secret
cache
because
ephemeral
containers
yeah
taking
it
back
so
for
in
female
containers,
though
there
is
a
real
issue
with
cash,
is
not
being
updated
with
secrets
for
ephemeral
containers.
But
this
one
okay.
A
I
won't
maybe
I
need
to
triage
them
as
need
information,
but
I
just
don't
know
what
information
to
request
I
just
need
to
try
it
out
myself
and
make
sure
it's
certain
infrastructure
problem.
When
you
sort.
A
D
Of
course
yeah
this
one,
this
one,
this
person,
initially
what
I
thought
was
volunteering
to
handle
this,
but,
to
my
surprise,
they
asked
me
my
social
handle
and
updates
I'm,
not
sure
should
I
should
I
move
this
back
to
me
or
keep
it
with
them.
D
I,
so
if
you
go
a
little
up
here,
you're
saying
that
they
would
like
to
work
together,
so
I
thought
so
I
gave
a
clue
to
them
that
you
can
probably
start
debugging
the
list
of
parts,
but
then
the
conversation
went
somewhere
else
all
together
where
I
don't
know,
I,
don't
know
what
social
handle
I
can
give.
And
what
can
we
do
here?.
C
D
B
D
D
D
C
A
This
is
known
that
it's
taking
too
much
resources
to
do
all
those
calculations.
A
A
A
And
remember:
Elana
was
working
on
this,
but
I
never
completed
it.
A
A
D
D
A
D
A
A
A
Put
the
stock
on
terminating
because
we
are
not
deleting
file
I.
Think,
okay,
give.
A
Yeah,
because
we
keep
an
eye
I,
know
I,
know
too
high,
like
whatever
I
notify
on
this
file.
D
A
D
A
A
D
D
C
A
D
A
When
we're
done
with
back
trash
yeah,
it's
surprisingly
a
lot
of
bugs
today,
I'm
glad
we
went
through
all
of
them.
Thank
you
and
thank
you
for
watching
to
stand
with
me
to
the
end.
It
was
really
nice.
Thank
you
very
much.