►
From YouTube: Kubernetes SIG Node 20220316
Description
Meeting Agenda:
https://docs.google.com/document/d/1j3vrG6BgE0hUDs2e-1ZUegKN4W4Adb1B6oJ6j-4kyPU
A
A
Yeah
it's
march
16th
2022.,
it's
a
weekly
signal,
crmsync
and
let's
get
into
the
agenda.
We
have
couple
topics
in
agenda
today.
Imran
you
go
first.
C
A
Oh
okay,
so
if
I
just
added
this
item
to
agenda,
I
think
the
plan
was
to
move
low
contention
flags
to
config
file
instead
of
command
line
parameters.
B
No,
no,
not
not
like
pre-submit
jobs,
the
the
test,
script,
jobs
for
it.
Are
they
passing
now.
A
I
think
they
got
fixed.
The
problem
was
that.
A
The
problem
was
that
the
test
was
verifying
that
kublat
will
restart
after
it
was.
A
Down
because
of
the
law
contention,
indeed
the
test
I
mean
the
problem
is
that
something
went
wrong
with
the
image
and
the
image
wouldn't
restart
the
couplet
automatically.
So
we
just
removed
the
coupled
check
for
cobalt
being
they
started
from
the
test.
I
think
it
will
need
to
be
back
when
we'll
return
everything
together.
A
So
especially
when
we
switch
to
config
file
and
we'll
start
changing
config
file
during
the
test
and
change
it
back
after
the
test,
then
you
will
need
to
make
sure
that
kubelet
will
restart,
but
for
now
it's
good
so
yeah.
This
is
green
low
contention
functionality
works
restart
of
kubelet,
which
is
a
system
like
even
ccg
or
something,
let's
see
something
like
system
settings
that
may
not
work.
A
B
Yeah,
I
can
take
a
look.
This
one
is
an
api
review,
so
feel
free
to
assign
it
to
me.
A
Okay,
this
is
done
and
I
think
once
this
will
be
merged.
The
next
step
will
be
to
change
the
log
contention
test
to
use
config
file
and
move
it
into
all
the
serial
lane.
So
we
don't
need
to
run
it
as
a
parallel
lane
for
just
for
this
test.
D
E
D
A
I
think
any
test
that
you
run
will
basically
any
end-to-end
test
will
be
okay.
One
question
would
be
if
this
image
will
match
the
image
that
container
d
uses
it's
not
a
big
problem
but
yeah
the
worst
case.
Scenarios
on
garbage
collector
logic
will
take
off
and
remove
the
wrong
image.
A
B
A
Yeah
is
there
a
pr
for
for
the
change.
D
A
Yeah
so
answering
your
question:
any
end-to-end
test
will
be
will
suffice,
most
of
them
testing
that
the
plot
will
be
started,
so
it
should
be
fine.
The
mismatch
is,
can
be
caught
on
pull
request
review,
so
we,
you
may
need
to
double
check
that
the
runtimes
are
configured
with
the
proper
image
as
well.
B
I
do
not
do
we
have
either
we've
got
ryan
on
the
call
ryan.
Do
you
know.
A
A
D
For
instance,
no,
because
the
test
image
don't
change,
this
is
basically
for
a
moment.
This
is
a
new
proxy
in
front
of
gcr.io.
So
for
the
moment,
it's
not
impacting
the
test
images.
This
is
only
for
the
image
promote
from
the
staging
repository
and
all
one
migrate
from
the
domain
flip.
We
did
two
years
ago.
A
B
I
just
wanted
to
say:
hi.
We
have
a
couple
of
new
names,
at
least
to
me
on
the
call
tyler
mina.
Do
you
want
to
say
hi
introduce
yourselves
to
the
group.
B
Hey
mina
hi
there
we
go.
E
Yeah
hi,
so
I'm
mina
I
work
for
a
costing
container
optimized
to
os
team.
So
I
I'm
just
joining
these
meetings
so
that
if
there
is
anything
that
any
feature
that
we
need
to
develop
for
kubernetes.
B
Cool,
as
you
can
see,
we're
mostly
discussing
the
state
of
test
and
continuous
integration
or
ci
infrastructure
for
sig
note
so
stuff
pertaining
to
the
cubelet
and
other
node
related
matters.
B
B
I'm
not
sure
do
you
want
to
contribute
to
helping
out
with
node
tests.
It's.
E
Sure
I'll
I'll
just
attend
couple
of
meetings
and
see
whether
if
it
is
you
know,
help
benefit
us
awesome
well.
Welcome.
F
A
And
I
also
wanted
to
say:
ask
everybody:
do
you
want
to
start
taking
attendance
and
put
our
names
in
the
meeting?
I
think
we
had
some.
We
had
this
blog
post
published
with
like
mentioning
names,
and
we
wasn't
sure
whether
we
mentioned
everybody
who
wanted
to
be
mentioned
and
who,
like
visiting
this
meeting
regularly.
B
Yeah
we
used
to
take
attendance
and
I
don't
know
why
we
stopped,
maybe
because
we
just
forgot,
so
I
certainly
have
no
objections.
A
Okay,
because
it's
not
as
we
as
attended
as
main
signal
meeting
where
it
doesn't
make
sense
to
have
so
many
names
listed,
so
yeah
just
open
the
document
with
agenda
and
put
your
name.
If
you
will.
A
Thank
you.
So
let's
look
at
table
first
and
then
so
target
health
is.
E
B
I
submitted
a
pr
to
remove
it.
There
were
like
two
jobs
that
were
both
on
other
tabs.
Okay,
it's
apparently.
G
A
Okay,
yeah.
We
need
to
double
check
why?
Why
it's
still
here
but
yeah
apart,
is
still
in
as
you
there
and
supreme
so
good.
It's
green!
So
eviction!
A
D
A
Okay,
let's
move
to
triage
and
discuss
it
when
you
get
into
triage,
but
yeah
and
eviction
is
still
not
being
looked
at.
I
know
it's.
It
got
rotten
and
refreshed
back
to
alive
so
and
it's
failing
for
very
long
time.
A
B
Yeah,
I
know
that
they
were
on
danielle's
backlog,
but
I
don't
know
if
they
are
completely
fixed.
Yes,
I
know
that
she
had
at
least
one
pr
up.
I
think
for
them.
A
You
know
danielle
want
to
like
redesign
some
of
it,
but
the
design
may
take
longer.
So
maybe
there
is
a
quick
fix
possible.
I
didn't
mean
to
find
somebody
like,
let's
maybe
discussed
like
this
issue.
A
A
Npd
we
had
we
removed
coupled
like
one
job
at
least
because
image
didn't
have
gimko
of
proper
version
or
something
like
that.
So
job
was
removed
and
for
push
image.
There
is
a
fix
in
the
works,
but
it
wasn't
merged
because
of
this
test
was
failing.
Hopefully
it
will
be
clean
next
time.
A
F
Isn't
so
I
I
just
had
one
question
when
I
was
going
through
and
doing
the
triage,
I
noticed
that
the
like
the
right
click
and
file
bug
button
on
our
test
grid
points
to
the
like
global
kubernetes
project
is
that
I
don't
know,
should
some
of
the
tabs
point
to
like
different
repos
or
is
the
kubernetes
project.
That's.
B
A
great
question
most
of
the
tests
that
we're
going
to
be
looking
at
for
the
purposes
of
this
meeting
are
all
based
on
like
code,
that's
being
tested
in
kubernetes
kubernetes,
so
kubernetes,
kubernetes
or
kk
is
the
correct
repo
to
file
bugs
against
yeah.
B
Like,
unfortunately,
this
view
doesn't
really
give
a
lot
of
information
for
filing
them,
I
prefer
to
use
the
automated
sergey.
Can
you
go
to
the
triage
thingy?
I
don't
know
if
it's
working
since
I
feel
like
it
frequently
falls
over.
But
if
you
go
to
that
triage
link,
you
can
use
this
to
triage
various
test
failures
and
use
that
to
sort
of
like
see
where
flake
started
and
that
kind
of
thing
and
if
you
click
on
the
file
bug
for
that
test
down
there
sergey
perhaps.
B
It
links
you
to
a
bunch
of
example,
failures
and
like
a
bunch
of
other
info
automatically,
which
is
really
handy.
A
Yeah,
this
is
a
great
integration.
I
know
that
work
group
reliability
wanted
to
do
something
about
it,
but
I
don't
think
they
actually
did,
and
I
don't
is
this
bug
b
will
be
associated
with
a
failure
going
forward.
B
I'm
not
sure
I.
B
There's
been
some
talk
about
like
raising
the
reliability
bar,
which
I
continue
to
see,
come
up
at
like
community
meetings
and
whatnot,
but
like
what
action
people
are
taking.
There
is
not
particularly
clear
so.
A
Yeah,
tooling,
is
definitely
a
big
part
of
reliability.
Raising
the
reliability
bar
like
field
of
this
spreadsheet.
It's
not
ideal,
but
at
least
it
gives
you
like
historical
view.
I
I
really
enjoy
this
going
back.
Functionality.
A
Okay:
okay,
thanks!
Thank
you
daniel
for
bringing
it
up.
A
Yeah,
I
think
we'll
clear
them
out
really
quickly,
but
I
think
we'll
get
those.
B
A
Or
just
feedback
on
the.
E
B
D
B
Yeah,
oh,
I
can
take
a
look
at
this
one.
A
Yeah,
I
remember
that
I
I
even
approved
it.
We
still
need
six
sig
release
to
approve,
and
now
it
has
a
conflict
so.
B
A
But
to
be
so
fixed.
A
So
there
is
a
possibility
to
merge
this
cap
into
this
release.
I
think
it's
a
little
bit
late
with
a
soft
freeze,
but
if
it
will
go
smooth,
I
don't
see
a
reason
to
block
it.
D
A
We
cannot
approve
it,
but
it's
a
little
bit
related
to
what
we're
doing
it's
a
promoting
some
of
windows
tests,
as
conformance
because
container
d
now
supported,
and
they
have
they
support
the
host
process
like
what
is
it
privilege,
ports?
A
A
A
Yeah,
but
we
just
reviewed
one
from
andrew,
I
think.
A
Okay,
this
is
quite
straightforward.
Any
takers.
A
You
have
to
remove
a
bunch
of
documents
admits
and
on
one
of
the
prisms
windows
team
was
was
not
happy
with
removing
it.
I,
oh
I'm
not
sure
why.
B
A
Okay,
this
is
whatever
hat
made
to
conformance.
If
anybody
wants
to
lgtm,
I
would
be
really
appreciated.
B
A
Yeah
there
is
a
first
step
is
to
create
a
test
and
run
it
for
two
weeks
and
then.
A
That
is
under
not
on
the
cluster,
so
no
test
cannot
be
made
conformance,
so
I
needed
to
create
a
cluster
test
that
will
test
whatever
hat
api,
and
so
this
first
one
is
to
create
it,
and
then
second
one
will
be
to
like
move
to
conformance
after
it's
running
successfully.
A
A
A
A
A
A
A
B
B
On
the
you
can
click
on
the
job
runs
and
each
run
is
it's
once
per
day,
so
every
dot
there
is
a
day.
So
that
gives
you
kind
of
an
idea
of
the
scale.
B
B
A
Is
that
for
sure
to
bring
it
back
right?
Okay,
I
think
we
need
to
file
a
bug
to
investigate.
G
A
A
A
Okay,
so
this
is
full
up
from
this
out
of
memory.
I
think
we
just
need.
A
Now
you
moved
it
to
bug,
but
I
think
it's
a
feature
request
to
limit.
I
nodes.
B
A
Okay,
I'm
not
sure
I
know
there
are
two
bugs
one.
One
block
is
related
to
advisor
taking
too
much
time
to
traverse
oldest
directors.
Another
is,
I
note,
noisy
neighbor,
so
you
can
exhaust
all
the
file
descriptors
and
like
I
knows,
and
nobody
can
stop
you
do.
You
know
which
one
is
that.
B
I
didn't
read
this
in
depth:
I
mean
a
deep
directory
tree,
isn't
necessarily
gonna
exhaust
inodes
or
cause
a
noise
neighbor
issue
I
mean
I
definitely
there
is
the
it
might
take
too
long,
but
I
don't
think
it's
just
merely
in
c
advisor.
I
think
it's
also
in
the
container
runtimes.
A
It
may
not
be
even
c
advisor
issue
just
yeah.
F
B
Like
I
would
say
that,
if
the
killer
is
triggered
because
of
the
behavior
of
cubelet,
when
trying
to
remove
a
deeply
nested
directory,
that
is
a
bug
that
shouldn't
happen.
B
Like
an
obvious,
easy
way
for
something
like
this
would
be
to
found
or
validate
like
the
depth
of
directories
that
are
possible
to
make
here,
so
you
can't,
like
denial
of
service
attack
the
cubelet
in
this
way,
because
I
think
that
this
could
be
like
I
I
don't
know,
I
I
think
of
it
more
as
a
bug
like.
Maybe
the
way
that
we
fix
it
is
like
needs
a
feature
proposal,
but
this
strikes
me
as
a
book.
A
Okay,
we're
taking
a
little
bit
too
long
time
here.
C
Okay,
let's.
A
B
I
feel
like
that.
Cubelet
having
a
hard
like
dependency
on
like.
B
Time
being
set
accurately
like
crony
d
or
whatever
ntp
being
set
correctly
like
as
far
as
I'm
aware,
that's
like
generally
agreed
upon
requirement
for
nodes.
F
G
G
B
F
B
They're
just
saying
that
the
node
won't
create
containers
after
it's
already
been
like.
I
don't
know
it's
unclear
in
this
case
if
a
reboot
was
involved
or
if
they
just
went
and
like
set
the
time
back.
B
If
a
reboot's
involved,
I
would
say,
take
the
node
down
for
maintenance,
make
sure
it
has
a
reasonable,
accurate
time
on
it.
If
they
are
asking
like.
Can
I
just
set
the
time
back
on
my
node?
No,
that's
not
supported,
but
in
any
case
I
don't
think
it
makes
sense
to
entertain
trying
to
support
this
use.
A
A
Yeah,
I
tend
to
agree
that
correct
time
should
be
requirement.
I
just
wonder
if
one
minute
skew
is
enough
to
trigger
this
issue
or
not
enough
to
trigger
this
issue.
So
if
one
minute
is
enough
to
trigger
this
issue,
then
it
probably
needs
to
be
addressed
in
couplet,
because
I
can
see
how
close
you
can
happen
and
coco
will
change
one
minute
back.
B
The
other
thing
that
concerns
me
a
little
bit
is
that,
like,
I,
don't
think,
they've
reproduced
this
on
a
recent
version
of
kubernetes,
the
pr
that
you
just
opened,
which
was
rebased
the
person
who
tested
it
said
they
tested
against
120,
which
is
which
isn't
a
supported
branch.
If
they
can't
like,
I
would
want
to
see
them
reproduce
this
after
the
pod
life
cycle
refactor
at
the
very
minimum,
because
maybe
things
changed
since
then,.
A
I
think
this
is
reasonable
and
yeah
the
test
that
appears
that
you
mentioned
didn't
have
any
test
confirms
that
it
was
failing
before
so
have
a
hard
time
accepting
it.