►
From YouTube: Kubernetes SIG Node CI 20220525
Description
SIG Node CI weekly meeting. Agenda and notes: https://docs.google.com/document/d/1fb-ugvgdSVIkkuJ388_nhp2pBTy_4HEVg5848Xy7n5U/edit#heading=h.2v8vzknys4nk
GMT20220525-170631_Recording_1920x2100
A
A
25Th
of
one
month
doesn't
know
May,
and
this
is
the
signal
CI
subgroup.
It
looks
like
we
have
no
actual
agenda
today,
unless
anyone
has
anything
last
minute.
C
Yeah,
so
so,
like
this
little
query
regarding
the
image
efforts
and
the
image
you
see
so
like
in
addiction
heart,
we
have
image
FS
got
available
that
is
at
215
by
default,
so
it
means
that
it
will
trigger
when
85
percent
of
the
disk
spaces
that's
consumed
so
and
in
the
emergency
higher
threshold
also,
it
is
set
to
85
by
default.
Like
means
image,
GC
will
come
into
action
when
85
percent
of
this
space
is
consumed
so
like
I,
want
to
know
why
that
thresholds
are
Same
by
default.
A
D
Is
it
just
one
quick
question:
are
you
referring
to
in
the
in
the
test
or
in
like
where
is
it
set
the
same
default
value.
C
D
D
A
Yeah
I'm
not
sure,
like
I,
haven't
looked
at
those
tests
in
like
a
solid
months
and
therefore
they
are
no
longer
in
my
brain.
But
yeah
I
can
have
a
look
when
I
next
look
at
them,
presumably
next
week
because
I'm
off
the
rest
of
the
week.
B
A
I
got
through
the
triage
backlog
because
there
wasn't
too
much
new
stuff
that
I
could
see,
and
so
we
now
mostly
have
a
backlog
of
stuff
that
needs
review
or
is
waiting
on
people.
A
A
I,
oh,
had
like
half
a
plan
for
figuring
out
how
to
set
up
the
like
testing
to
start
doing.
Node
restarts
but
haven't
got
around.
C
A
Like
building
it
yep
because
there's
never
enough
time
in
a
day
to
start
new
things.
B
A
A
A
A
C
A
A
That's
exciting
foreign.
F
A
A
F
F
F
E
E
A
B
E
Yes,
Sasha
on
my
team
does
oh.
A
Cool
then
I
will
open
an
issue
Farm.
A
Oh,
this
is
fine.
I
will
open
an
issuing
test
him
for
it
together
to
rename
those
jobs
cool.
So
it
looks
like
nothing.
New
is
failing.
That
is
good,
for
you
know
the
node
test,
like
things
didn't
get
worse
in
the
last
two
weeks,
which
motivates
me
somewhat.
B
A
D
C
D
E
I,
have
we
have
to
think
in
David,
but
I
I
don't
know
if
I'll
have
time
this
release,
but
we
had
talked
about
trying
to
move
it
to
Beta
And
125,
but
I.
Don't
know.
E
It
almost
seems
like
something
that
should
be
taken
care
of
by
the
CRI
Dr
D
maintainers,
like
you
know,
have
their
containerdy
or
have
containerdy
in
general,
be
more
clear
about
situations
that
it's
trying
to
support
the
CRI
or
not
I.
E
A
In
local
up
cluster,
we
could
at
least
check
to
see
if
container
Adidas
stock
exists.
E
Right,
but
it
will,
if,
if
it's
using
serial
idocity,
because
in
CRI
doctor
Docker
talk
to
containerdy
through
the
continuity
sock.
A
Doesn't
see
our
the
CRI
docket,
he
talked
to
Docker,
which
then
has
its
own
private
containerdy
socket
in.
As
you
don't
know,.
E
Ude,
maybe
it's
worth
getting
more
information
to
see
if
it's.
E
Okay,
well,
so,
if
it
doesn't
exist,
then
yeah,
that's.
That
would
be
something
that
we
should
change.
If
it
does
exist,
then
I
think
that
that
would
be.
B
A
E
Also,
we
do.
A
Like
what's,
oh
wait?
No,
because
it's
an
issue
not
a
period,
therefore
it
doesn't
actually
matter.
Does
anyone
have
a
canned
response
for.
B
A
D
E
D
A
That's
an
issue
opening,
containerdy
and
I
know
a
dto
is
working
on
it
before
she
went
on
leave,
but
the
current
state
of
it
seems
to
be.
A
No,
it's
flapping,
not
ready,
ready
notes
are
flapping
with
error.
Flag
is
not
healthy,
they
have
two
nodes
where
the
plague
is
taking
longer
than
three
minutes
to
respond.
A
A
A
B
A
Kublic
creates
leases
without
owner
reference
set
if
node
doesn't
exist,
yeah.
So
if
something
deletes
the
node
object,
while
the
group
is
still
running
I
think
things
break.
D
A
D
A
A
B
A
A
Oh
cool,
oh,
what
was
it.
B
D
Yeah
this
one
being
sent
to
me:
there's
a
there's,
a
GPR
open
to
fix
this
awesome
at
the
bottom.
It's.
C
D
An
issue
note:
actually
it's
an
issue
in
Sig
Network
in
the
endpoint
controller.
A
A
I
think
I've
seen
this
in
customer
escalations
before.
A
A
A
Is
this
potentially
related
to
the
segment
work
thing.
D
A
A
Pause
containers
remain,
undeleted
is
gone,
which
means
this
is
oh
wait.
No,
we
technically
still
have
to
backpot
some
texts.
I
guess.
A
A
D
D
A
A
D
B
A
D
A
B
A
Kubler
cannot
finish
terminating
a
pod
that
uses
a
PVC
with
volume
mode
block
if
there
is
an
Alfie
on
the
underlying
block
device.
That
sounds
more
like
storage
than
us.
A
That's
the
story
channel
that
liveness
profailing
when
the
load
on
the
application
is
maxed,
the
that
makes
sense.
B
B
A
B
A
A
The
other
day,
so
basically
the
Kubla,
like
cloud
provider,
stuff
iterates
over
all
of
the
interfaces
and
if
there's
a
bunch
of
interfaces
on
the
Node,
you
end
up
hitting
the
like
management,
API
rate
limit
because
you're
calling
it
every
10
seconds.