►
From YouTube: Kubernetes SIG Node 20210714
Description
Meeting Agenda:
https://docs.google.com/document/d/1j3vrG6BgE0hUDs2e-1ZUegKN4W4Adb1B6oJ6j-4kyPU
A
Hello,
it's
july,
14th
2021,
it's
a
ci
subgroup
of
signaled
and
we
are
talking
about
node
conformance
tests,
so
conformance
tag
was
created
for
kubernetes
certification
program.
It's
supposed
to
be
agnostic
test
that
works
supposed
to
work
everywhere,
and
if
you
pass
them,
you
certify
it
to
be
called
kubernetes
or
something-
and
I
think
initial
idea
was
that
note
conformance
will
be
in
the
same
vein.
A
A
So,
as
time
passed,
we
like,
I
don't
think
there
were
any
attempts
to
make
it
part
of
the
certification
program,
nor
any
desire
so
far,
and
so
the
word
conformance
is
initially
intended
to
be
same
as
conformance
and
cluster
test,
but
it
doesn't
actually
represent
what
it
currently
is.
A
So
I
think
we
can
like.
Today
we
have.
A
Yeah
ordering
is
a
little
bit
weird.
So
if
you
look
at
this
conformance
tag
description,
there
is
a
document
saying
how
to
do
cri
testing
policy.
I
have
this
red
sign
on
the
link,
so
this
policy
explaining
how
to
contribute
runtime
test
results
into
into
common
dashboard,
but
I'm
not
sure
whether
anybody
using
that
so
there
is
a
faq
and.
A
Also,
all
sorts
of
documentation,
but
I
don't
know
with
anybody
using
this
program
and
to
certify
the
runtimes.
I
think
we
are
running
like
container
d
tests
right
now
and
the
cryotests
it's
part
of
our
test
grid
and
there
is
nobody
contributing
back
cri
test
results,
but
nevertheless
I
think
this
is
the
reason,
by
the
way,
elena,
that
initially
you've
been
pointed
to
cry
as
a
not
conformance
kind
of
is
something
that
started
not
conformance
because
of
this
document.
A
I
think-
and
but
I
mean
description
of
this
document
saying
that
that
should
be
agnostic
to
cloud
providers
less
distributions
and
run
times.
A
So
it's
basically
a
set
of
tests
that
doesn't
require
any
special
handling
so,
like
everything's
supposed
to
be
working
without
any
special
configuration-
and
I
think
it's
a
fine
definition-
we
can
use
it
and
I
suggest
that
we
keep
like
we
rename
conformance
tests
to
something
more
appropriate
and
then
for
everything
else,
for
like
everything
else
being
alpha
beta
features,
plus
some
features
that
have
special
requirements.
A
We
can
use
feature
tag
plus
some
extra
text
like
alpha
or
special,
to
indicate
why
white
makes
it
so
special
and
then,
when
feature
is
graduated,
we
can
rename
it
into
node
conformance
or
whatever
new
name
will
come
off
is,
and
this
will
indicate
that
this
feature
is
supported
by
every
distribution
and
it's
part
of
like
big,
I
mean
part
of
required
tests
for
node.
A
So
this
is
a
proposal,
and
I
think
I
just
combined
this
document
out
of
different
sources
and
I
think
that
maybe
how
we
can
structure
tests
and
if
you
publish
it
as
a
policy,
we
can
clean
up
all
the
tests
like
right
now.
Many
tests
have
conformance
plus
feature,
and
I
think
this
is
not
what
everything
like
what
we
intended
to
do.
B
Like
amen
to
all
that,
that
was
great
thanks
for
doing
all
this
research,
sergey
and
brazil.
I
didn't
even
know
about
this
cri
testing
policy.
This
is
like
totally
news
to
me,
so
this
is
awesome.
Thanks
for
doing
all
this
digging,
and
hopefully
the
digging
that
I
started
at
sig
arch
like
a
couple
weeks
ago,
was
also
helpful
as
a
starting
point.
B
So
I
I
think
that
we
should
probably
like
I
would
say
next
steps
are
probably
to
send
something
out
to
like
sig
node
mailing
list
like
share
this
publicly,
maybe
and
get
feedback.
I'm
sure
that
dims
will
have
opinions.
I'm
sure
that,
like
if
we
could
get
like
aaron
or
ben
elder
to
weigh
in
they
should
have
opinions.
I'm
sure
the
conformance
folks
will
have
opinions
so
yeah.
I
think
just
getting
eyes
on
this,
and
we
might
want
to
come
up
with
before
we
we
set
this
publicly.
B
A
Yeah,
I
was
thinking
like
not,
I
don't
know
like
all.
My
names
are
stupid,
so
and
names
are
hard.
B
Well,
so
I
guess,
I
think
that
we
should
totally
move
away.
Well,
I
guess
it
depends
like
there's
kind
of
two
options
right
so,
like
historically,
we've
had
this
thing
and,
like
the
node
end,
to
end
test
to
this
day
like
there's
this
idea
of
being
able
to
run
a
subset
of
all
of
the
tests
just
with
a
node
without
like
a
fully
functioning
cluster,
and
so
I
think
that
we
might
need
some
way
to
indicate
that-
and
I
don't
think
node
conformance
is
a
good
way
to
do
that.
B
And
then
I
think,
there's
maybe
a
separate
issue
of
like
how
do
we
make
sure
that
all
the
cris
are
like
functioning
the
way
that
the
cris
are
supposed
to
function?
And
maybe
for
that
we
could
replace
like
right.
Previously,
we
had
this,
like
conformance
tag
with
the
hope
that
this
will
be
kind
of
like
a
you
know,
cri
conformance
or
that
kind
of
thing.
But
the
conformance
game
has
asked
us.
B
Maybe
please
don't
use
conformance,
because
that's
confusing
people-
and
I
agree-
I
was
confused-
so
maybe
like
what
about
like
for
the
cri
specific
ones
like.
Maybe
we
could
try
the
name
like
I'm
thinking,
cri
validation
or
something
like
that.
A
Okay,
but
this
is
not
shirai
specific-
it's
basically
like
it
validates
all
the
components
that
present
in
the
node,
so.
B
Yeah,
so
I
I
I
know
that
I'm
saying
we
shouldn't
we
shouldn't
like
wholesale,
just
rename
all
of
the
node
conformance
tests,
I'm
saying
for
whatever
meets
this,
like
cri
validation
criteria.
B
We
should
re-label
those
as
cri
validation,
and
I
think
we
should
figure
out
a
different
story
for
what
we
do
for
the
like
test
that
just
you
can
run
locally
kind
of
thing.
You
don't
need
a
full
cluster.
B
A
Okay,
so
you
suggested
that
there
will
be
no
two
categories,
but,
like
third
category
will
be
well,
I
guess
conforms
will
be
south
of
sub-categories,
like
cri
validation,
configuration.
B
So
I
mean
there's
like
the
actual
anything:
that's
a
real
conformance
test
should
be
marked
with
conformance
and
anything.
That's
not
a
real
conformance
test
should
not
be
marked
as
conformance,
and
so
my
suggestion
is
like.
I
think
that
you
know
all
of
the
like
the
standard
feature
things
that
we
used
to
call
node
feature
whatever
that
we're
trying
to
migrate
to
just
use
the
standard
feature
tags.
Those
are
fine.
B
The
reason
I
think
that
we
were
tagging
them
specifically
like
node,
as
opposed
to
like
you
know,
just
a
normal
feature
was
because
we
could
run
them
just
on
like
a
single
node
in
a
more
minimal
node,
end-to-end
test
setup,
and
so
that's
the
thing
that
we're
kind
of
losing
and
that's
part
of
like
why.
There's
like
a
bunch
of
confusion
here,
is
that,
like
historically
node
conformance
was
trying
to
be
two
things
thing
number
one
was,
it
was
trying
to
be
a
reduced
set
of
the
end
to
end
test.
B
You
could
just
run
on
one
node
and
second,
it
was
also
trying
to
be
a
set
of
conformance
tests
for
nodes,
and
I
think,
if
we
like
try
to
split
that
into
two
things.
I
think
there's
also
overlap
right,
like
some
of
the
things
that
are
conformance
tests
could
just
be
run
on,
like
one
node,
without
like
a
fully
functioning
cluster,
depending
on
what
components
are
required,
but
maybe
not
all
like.
B
For
example,
I
think
that
there's
no
scheduler
in
the
node
e
to
e
setup,
so
there
would
be
some
like
there's
like
some
things
that
wouldn't
make
any
sense
to
put
there,
because
you
have
to
have
a
scheduler,
for
you
know
like
full
cluster
conformance.
B
A
Yeah
I'm
trying
to.
A
On
a
single
note,
so
you're
saying
there
is
a
separate
dimension
for
that
is
which
part
of
cubelet
we're
testing.
B
Now,
okay,
so
that
way,
it's
like
clear.
B
B
Also
for
people
who
are
not
staring
at
my
screen
or
the
test
text
is
too
small
or
whatever
I
said
so
for
conformance
historically
node
conforms
tried
to
be
two
things:
one.
It
tried
to
be
a
set
of
end
to
end
tests.
You
just
needed
one
node
for
two:
it
tried
to
be
conformance
like
tests
for
nodes
and
I'm
suggesting
we
get
rid
of
the
name,
node
conformance,
because
it's
confusing
everyone
and
then
for
the
conformance
like
tests
that
we
want
to
like
have
the
cri
run
against.
B
We
should
call
them
something
else,
probably
cri
validation
and
then
for
the
tests
that
you
just
need
one
node
to
run.
We
should
come
up
with
a
new
name
for
them
that
is
not
node
conformance
and
it's
possible.
There
might
be
overlap
between
the
two.
So
I'm
sure
that,
like
some
of
the
cri
validation
tests
will
not
need
a
full
clustered
run,
so
they
would
also
probably
be
in
whatever
we
want
to
call
this
like
that.
This
needs
a
name.
B
And
then
we
could
get
rid
of
the
special
weird
e
underscore
node
folder,
perhaps
like,
because
that's
been
confusing
everybody
as
well.
So
I
don't
know
I
mean
that's,
that's
one
way
we
could
go
forward.
I
think
we
need
to
get
feedback
on
this,
because
maybe
that's
gonna
just
be
too
much
refactoring
effort
or
something
like
that.
B
The
international
concern
was
so
you
know
how
this
will.
This
definitely
applies
to
everything
in
the
test.
Slash
e2e,
underscore
node.
We
also
have
a
bunch
of
tests,
labeled
sig
node
in
test,
slash
e,
slash
and
then
I
think,
there's
a
bunch
of
common
tests
as
well.
That
are
us,
so
we
do
have
some
that
require
a
scheduler.
I
think
currently.
B
C
B
A
B
B
I'm
just
learning
so
many
new
things:
misplaced
and
legacy
tests.
Okay,
so
apparently
we're
supposed
to
clean
these
up,
sergei.
A
Okay,
but
yeah
we're
on
the
same
page
that
this
couplet
local
can
be
applied
universally
in
sick
describe
function
right
because,
like
the
same
way,
we
apply
sig
note
to
all
the
tests.
You
can
apply:
kubelet
local
there.
A
Okay,
but
then,
returning
to
your
first
saturation
about
ci
validation,
I
think
I
was
suggesting
that
we
will
basically
mark
everything
that
doesn't
require
any
special
requirements
and
graduated
with
some
conformance
tag.
So
you
suggestion
to
split
it
into
features
rather
than
keep
it
in.
A
B
Conformance
is
very
specific
requirements
like
right
now,
I
think
they
are
only
so
conformance
does
have
to
be
reviewed
by
a
conformance
reviewer,
which
is
why
I
suggested
we
maybe
poke
aaron,
because
I
know
that
he's
one
of
the
conformance
reviewers
like
not
in
new
zealand,
so
maybe
a
little
bit
better
time
zone
overlap
anyways.
So
as
far
as
the
conformance
tests
go,
I
think
they
only
really
test
stuff.
B
That's
in
a
user-facing
api
and
node
has
a
lot
of
stuff
so,
for
example,
anything
that's
just
testing,
for
example
the
cubelet
talking
to
the
container
runtime
anything,
that's
just
like
exercising
that
and
doesn't
involve
like
a
user,
submitting
a
specification
like
that's,
usually
not
considered
for
a
conformance
test.
So
the
conformance
test
is
just
checking
like
the
surface
area
of
the
external
api
yeah.
A
That's
right
right:
we
need
to
rename
it
for
sure.
That's
why,
like,
if
you
switched
as
a
document,
I
suggested
that
we
like
in
this
category
will
be
all
the
tests
that
feeds
like
that
are
agnostic
and
don't
require
special
environment.
B
Yeah,
it
was
never
that
so
that's
what
I'm
saying
I'm
saying
I
I
agree
with
the
plan
for
like
for
all
of
the
stuff.
That's
like
you
know,
features
and
whatnot
like
those
should
go
there,
but
like
for
the
stuff,
that's
ga
that
is
not
eligible
for
like
becoming
a
conformance
test,
because
I
think
there's
gonna
be
a
bunch
of
stuff
that
isn't
I'm
saying
what
do
we
do
about
those?
B
Well,
maybe
we
make
like
a
cri
specific
suite,
that's
a
subset
of
what
this
is
in
order
to
make
sure,
because
when
I
said
like,
can
we
just
rename
node
conformance
to
like
you
know,
cri
conformance
or
something
like
that?
I
got
no.
No,
that's
not
what
node
conformance
is
it's
also
these
other
things
so.
E
A
A
B
We
call
them
one
of
these
things,
but
then,
if
like,
we
want
to
continue
doing
whatever
this,
like
cri
validation
thing
is
which
right
now
we
are
using
to
select
tests
for
the
different
cris
like
we're
running
node
conformances,
I
think
presubmits
and
a
bunch
of
other
stuff
like
for
those
things
we
might
want
to
have
a
special
tag
for
that
and
that
maybe
that
should
be
cri
validation
for
tests
that
conform
with
this
policy.
A
So
all
these
tests
can
run
independently
from
environment
and
we
mark
them
as
single
tag.
So
we
can
run
it
as
part
of
like
pr
validation
or
something
like
that.
That's
why
I
wanted
like
common
tag
for
everything
that
is
not
the
same
as
kublet
local,
because
kubel's
local
is
a
superset
of
that.
B
Well,
swap
can't
be
conformance,
but
I
think
a
bunch
of
the
cri
stuff
also
can't
be
conformance.
So
this
is
why
I'm
saying.
A
A
Yeah,
not
the
conformance,
we
need
a
new
name,
but
not
conformance,
not
yeah.
We
need
a
good
name,
not
sanity.
That
mike
suggests
archon
suggested.
F
B
B
Well,
you
you'd
emphasize
like
and
like
we
wanted
to
be
agnostic
of
environment,
so
maybe
like
node
agnostic.
B
I
think
I
think,
honestly,
you
know
the
brainstorming
is
good,
but
we've
spent
a
while
discussing
this
topic,
so
isis
would
probably
like
the
best
thing
to
do
is
to
like
think
about
this.
Asynchronously
send
out
a
doc
for
feedback.
So
sergey
can
I
action
you
to
send
out
a
doc,
the
node
conformance
doc,
soliciting
feedback.
B
And
so
I
guess
maybe
also
like.
B
B
Okay,
let's
see
I
put
a
thing
on
here
for
122
burned
down.
We
were
in
really
good
shape.
The
last
time
I
looked
so
like
not
to
give
us
too
much
of
a
pad
on
the
back,
but
let's
take
a
quick
look.
B
So
last
I
checked
we
were
doing
really
good
for
120
to
burn
down.
So
there's
a
bunch
of
stuff
here
and
not
all
of
it
is
us,
but
like
we
only
have
nine
things
and
we're
like
not
even
at
test
freeze,
so
that
that's
better
than
last
release,
so
great
work.
Everybody
stuff
that
are
specific
to
us
is,
I
think,
this
bug
plus
the
four
tests
that
are
being
tracked,
and
I
think
that.
B
Is
this
one
this
one
this
one?
Some
of
them
are
like
they're,
not
regressions,
they're,
not
new,
so
I
think
probably
none
of
them
have
to
land.
This
release,
like
I
think
all
of
these
are
old
flakes,
so
none
of
them
are
going
to
be
release
blockers,
but
we
can
continue
working
on
them.
B
I
will
maybe
take
a
for
this
one.
I
will
try
to
follow
up
with
that
one
online.
Let's
see,
let's
go
through
these
rtm.
What's
going
on
with
this
one.
B
Cool
okay,
so
no
updates
on
that
one.
Oh
our
favorite
test,
matthias!
I
owe
you
a
review
or
something
I
think.
B
A
I
think
timing
looks
good.
I
only
have
one
comment
on
proposed
change.
I
don't
know
whether
this
change
is
actually
needed.
I
mean
it's,
not
a
bad
idea
to
do
it
basically
idea
here
is
that
we
have
a
radio
state
and
we
assign
it
in
the
very
end,
like
we
calculate
ready,
state
and
then
assign
very
end,
but
there
is
like
a
go
routine
in
the
middle
that
may
execute
faster.
A
So
I
think
I
gave
a
comment
that
there
is
one
more
place
that
may
benefit
from
being
properly
assigned.
B
You
so
one
of
my
concerns
with
this
is
what
basically,
when
I
was
looking
into
the
code,
I
did
a
bit
of
a
deep
dive
for
this
and
I
sort
of
concluded.
B
I
thought
that
this
test
was
written
kind
of
weird,
but
then
I
also
went
and
looked
at
the
underlying
code
that
it
was
testing
and
that
code
was
also
potentially
buggy
so
like
there
was.
I
think
this
yeah
this
case
where
I
was
like,
I
think,
there's
a
race
condition,
because
we
go
and
we
trigger
this
thing
asynchronously.
B
But
then
we
immediately
go
here
and
we
set
the
status
to
ready
which
might
overrun
whatever
ends
up
getting
run.
So
there's
a
there's,
a
race.
G
B
B
I
think
it's
probably
tough,
like
fine
to
tweak
the
thresholds
and
see
if
that
at
least
reduces
the
flakiness
of
the
test,
and
then
we
can
take
it
as
sort
of
a
backlog
thing
to
try
to
fix
this,
because
I
think
this
is
a
real
bug
and
I
think
the
test
is
catching
a
real
bug,
but
having
it,
flake
constantly
is
like
unpleasant
for
everyone.
B
Okay-
and
I
think
this
this
isn't
enough
compared
to
like-
what's
we
need
to
also
increase
the
frequency
right?
Yes,
okay,.
B
That's:
okay!
That's
okay!
I
just
wanna.
I
wanna
make
sure
that
I
write
down
accurately
because
I'm
going
to
put
a
comment
on
here:
disgust
in
sig,
node,
ci
meeting.
I
guess
today
is
july
14th.
B
B
Awesome:
okay,
that
was
in
great
shape,
I'm
so
glad
we're
making
progress
on
that
because
I
know
that's
been
haunting
all
of
us.
Okay
pod
should
run
through
the
life
cycle
of
pods
and
pod
status.
What's
going
on
with
this
one.
B
Something
oh,
this
is
the.
This
is
the
weird
it's
just
failing
for
cubelet
orphans
and
it
looks
like
there
was.
G
There
was
like
fix,
and
and
but
it's
still
failing
in
the
node
orphans,
and
now
I
have
like
a
debug
output
which
doesn't
really
help,
but
since
I'm
on
vacation,
currently
I
didn't
take
extra
time
to
look
at
it,
so
I
will
but
soon
enough.
I
promise.
B
So
perhaps
does
is
anyone
from
the
the
memory
manager
side
of
the
house
interesting?
Yes,.
F
I'm
interested,
but
in
general,
like
we
have
a
separate
lane
for
the
memory
manager,
and
it
should
be
just
in
excluded
from
orphan
tests.
It's
funny
that
we
have
them.
B
F
B
Making
such
good
progress
amazing,
okay
last
one-
oh
this
one
is
pause-
should
support
pod
readiness
gates.
What's
going
on
with
this
one
francesco.
D
Is
yeah
I'm
still
waiting
but
still
struggling
to
reproduce
locally
because
in
ci
I
will
it's
hard
for
me
and
true
to
be
told
that
was
super
busy
with
the
code
freeze,
so
yeah
this
one
and
another
one
I'll
need
to
prioritize.
It's
on
me.
Sorry,
understandable.
B
No,
no,
I
mean
it's.
It's
fine,
like
I
guess,
since
whatever
happened
over
here
like
this
is
flaking
less
I
mean.
Maybe
it's
just
because
code
freeze
happens
since
the
eye
isn't
necessary.
Don't
know,
I
don't
know
if,
like
frequency
of
pr.
D
B
One
thing
I'm
a
little
bit
curious
about
is
like
who
owns
this
feature:
pod
readiness
gate-
I
don't
want
an
olive
github.
B
Well,
it
looks
like
it
went
ga
and
114.
So
that's
probably
why
there's
no
cap,
so
it's
been
around
for
a
while.
D
We
have
also
pr
to
just
increase
the
timeout.
Let
me
fetch
the
link,
but
I'm
not
sure
why
why
it
is
not
producing.
D
D
H
B
So
this
just
looks
like
it's
kind
of
at
least
from
this
refactor.
This
looks
like
it's
just
kind
of
a
poorly
written
test.
B
B
Great
well,
that's
progress,
awesome
work
and
then
I
guess
francesco
you
had
a
couple
of
things.
We
want
to
talk
about
cereal.
D
Yeah
I
want
to
to
syrup,
because
it's
quick
and
arthur
may
pointed
to
me
that
the
cpu
manager
test
is
failing
lately,
so
I
believe,
but
I
need
to
check
and
again,
code
freeze
frenzy
was
why
I
didn't
check
yet
that
this
bug
I
filed
some
time
ago.
It's
actually
broken
now.
The
reason
is
the
cpu
manager
entrance.
They
assume
a
very
specific
cpu
layout.
They
are
just
fragile,
so
they
can
break.
For
example,
the
cloud
provider
just
changes,
cpu
generation
they
can
break
so.
D
It
could
happen,
and
this
is
of
course
outside
our
control.
So
while
I
can
say
hey
they
break
because
of
this
fragility,
this
is
surely
something
we
need
to
address
and
it's
a
good
chance.
So
once
I
get
the
chance,
I
will
start
looking
into
that
and
possibly
fix
this
issue,
so
I
was
just
point
out
that
it
could
be
the
test
fragile.
So
that's
something
we
should
take
into
account.
B
I
F
B
Probably
can
speak
to
that
since
that's
his
thing.
A
Yeah
it's
at
least
one
release.
I
think
the
only
user
right
now
we
found
is
maybe
alibaba
and
I'm
not
sure
how
vocal
they
will
be
about
removing
it
but
yeah.
I
am
aware
of
that.
That's
relying
on
that.
I
thought
that
duplication
will
go
faster
and
I
will
have
time
to
start
migrating.
Some
tests.
C
D
E
So
I'll
say,
failing
test
sake,
node
cc,
francesco.
B
B
Cool,
I
guess
that's
kind
of
all
I
had
unless
anybody
has
anything
else
for
the
serial
things.
B
I
need
to
take
a
look
at
the
ci
board
and
see
if
there's
anything
there.
That's
waiting
on
me
the
as
far
as
like
features
go
we're
in
great
shape
for
code
freeze,
and
I
know
everybody's
been
working
really
hard.
So
thanks,
everybody
for
all
of
the
effort
that
you've
been
putting
in
things
are
looking
really
good.
B
It'll
be
we'll
be
like
less
exhausted
from
all
of
the
code
freeze
and
bug
squashing
and
who
knows
what
else.
A
No,
I
look
at
ci
board
and
all
the
issues
are
assigned.
So
if
you
have
any
issues,
just
take
a
look
well
like
not
issues
prs
assigned
for
reviewers
and
yeah
now,
maybe
to
approve
like
we
have
13.
That
means.
B
B
Cool-
maybe
let's
just
quickly
look
at
this
one,
because
I
didn't
see
this
wow.
This
was
this
looks
like
a
scalability
thing.
B
D
B
I
think
we'll
see
like,
depending
on
what
the
fix
looks
like
and
then
I
can.
I
am
a
milestone
maintainer,
so
I
can
add
things
to.
B
D
D
B
Thanks
everyone
yeah
no
problem
sergey,
hopefully
next
week,
you'll
have
your
setting
yeah
fixed.