►
From YouTube: Kubernetes SIG Node 20210818
Description
Meeting Agenda:
https://docs.google.com/document/d/1j3vrG6BgE0hUDs2e-1ZUegKN4W4Adb1B6oJ6j-4kyPU
A
B
Go
ahead
now,
first,
I'm
requesting
a
review
for
no
problem
detector.
Oh
there
was
an
issue
with
one
of
the
jobs,
and
this
basically
adds
a
file
in
order
for
the
job
to
properly
run.
A
Yeah,
okay,
we
generally
don't
track
npt
repository,
but
yeah.
I
will
add
it
to
the
board.
I
commented
already
any
anybody
familiar
with
cloud
builds.
B
A
A
So
this
will
unblock
which
test.
A
Cool
and
then
all
other
npt
tests
like
since
you're
looking
at
that
do
you
know
why
they
failing.
B
A
Yeah
last
time
I
saw
this
when
we
had
some
configuration
change.
C
B
Yeah
because
I
discovered
this
while
I
was
doing
some
some-
the
brigade
attack
migration,
and
then
I
noticed
this
this.
This
job
was
failing
in
this
constant
state.
A
Do
you
know
which
which
desktop
it's
it's
from
because
yeah
and
would
surprise
me
alpha,
doesn't
have
any
alpha
labels
on
on
test
name.
A
Oh
it
just
all
alpha
features
enabled
I
think,
okay.
C
C
A
I
think
this
tab
is
running
all
the
tests
with
all
the
alpha
feature
enabled
to
prove
that
you
can
enable
all
the
features
and
wouldn't
break
google.
It
like
main
scenarios.
B
A
So
I
mean
if
there
is
a
test
that
is
run
that
it's
likely
on
this
test
grid,
so
we
need
to
find
out.
Can
you
take
an
action
item
to
find
out
where
it's
run
like.
B
B
I
commented
the
new
changes
in
that
most
of
it
looks
pretty
good.
There
were
some
exceptions.
One
of
these
was
this
particular
test
and
the
other
ones.
So
let
me
just
share
the
pr
here.
B
So
the
last
comment:
basically,
I
ran
out
the
test
and
I
compared
all
the
tests
from
the
previous
run
to
the
to
the
run
on
this
pr,
and
you
can
find
that
there's
a
document
tracking,
all
of
them
there's
a
link
on
the
last
comment.
B
Basically,
the
only
exception
where
tests
don't
match
are
those
four
cases:
the
enzo
and
alpha.
It
has
failures,
cube
material
also
times
out.
So
I
don't
really
have
anything
to
compare.
Ubuntu
has
no
previous
runs
and
fedora
has
some
differences.
B
B
Yes,
you
can
go
on
the
link
there
on
the
comment.
It
should
be
a
shared
document.
That's
publicly
available.
A
A
B
Yeah,
the
only
difference
is
the
naming
changes.
Some
of
these
name
changes
are
expected
because
of
the
attack
removal,
but
this
one
in
particular
the
swap
fedora,
has
some
nodes
that
have
the
node
conformance
stack
in
previous
versions,
which
don't
have
data
on
the
new
version.
A
And
yeah,
okay,
next
quarter
thing
we
discussed
last
week
and
I
figured
out
what
this
chart
means.
A
So
apparently,
this
chart
well,
this
chart
is
indicating
number
of
projects,
so
this
18
is
actually
a
number
of
projects,
not
number,
of
course,
or
anything
else
like
we
discussed
before,
and
each
project
has
a
reasonably
high
quarter.
So
for
every
job
execution
we
rent
a
project
and
run
all
the
tests
in
this
project
and
then
clean
up
all
the
resources.
A
This
is
what
this
chart
represents
and
yeah.
The
only
way
to
optimize
this
chart
and
like
avoid
running
out
of
quota
of
projects
is
to
I
mean
first,
we
can
ask
to
for
more
projects,
but
you
can
see
that
generally,
it's
not
a
problem
or
we
can
get
and
minimize
number
of
jobs
that
we
run
so
less
jobs
means
less
projects
to
rent
and
that
will
improve
the
numbers.
A
So
yeah
also
discussed
before
with
combining
jobs
is
or
combining
tests
in
a
single
job
will
really
help
with
that.
That's
update
here
any
questions
on
that.
A
Okay,
martial
your
item.
C
High
folks,
so
regarding
the
offer
jobs
I
rejected,
and
in
general,
like
also
failures
that
currently
we
have
under
the
orphan
jobs.
One
of
the
failure
is
passing
under
the
syrian
job
because
it
runs
several
jobs
as
well
and
the
second
one
that
failing
is
part
of
the
conformance
test
suit,
like
not
under
the
signal,
but
under
the
conformance.
C
A
So
what
is
your
ideas?
I
forgot.
D
Oh,
my
good
one,
you
can
one,
let
me
send
it
to
you.
Sorry
all
right
was
full.
Oh
yeah
yeah.
Now
I
remember.
A
Cool,
thank
you
yeah.
I
think
it
like
it's
just
great
first
issues
and
you
at
least
like
understand
syntax
of
what
we,
how
we
skip,
that,
how
we
add
tests.
D
A
So
we
are
out
of
that.
I
also
wanted
to
show
what
I
checked
weekly
before
the
meeting,
but
we
can
probably
do
it
as
part
of
the
meeting,
so
the
performance
dashboard.
A
I
look
at
the
home's
dashboard
and
try
to
see
if
you
have
any
degradations
over
the
last
vehicle
cell.
So
I
don't
see
any
so.
You
can
see
that
yeah
we're
keeping
maester
quite
stable.
A
It
looks
a
little
bit
growing,
but
I
mean
I
I
try
to
look
very
hard
and,
like
I
I
don't
see
like
a
clear
trend
here,
so
I
think
we
are
good
right
now
so
yeah,
let's
keep
checking
performance
of
couplet
and
make
sure
we're
not
degrading.
Last
time
we
didn't
catch
it
in
time
and
it
cost
us
like
some
extra
time
being
in
the
release
blocking
mode
and
trying
to
fix
it
quite
urgently.
C
A
Yeah
there
is
a
what
is
it
legend?
I
think
it's
a
legend
here,
oh
yeah,
when
you
hover
over
it
will
show
the
legend.
So
red
is
50.
50
percentile.
I
think
it's
a
number
of
number.
A
Yeah
notes,
I
think
it's
a
hundred
nodes
test
and
50
percentile
is
zero
point
eight
and
then
all
the
way
to
hundredths
percentile,
like
hundred
percentile,
like
worst
case
so
worst
case,
is
zero
of
16.
I
don't
know
what
the
units
are
probably
course
yeah
016
course
for
a
node,
so
yeah
it's
percentiles.
C
A
I
think
I
mean
it's
the
best
like
we
used
to
have
a
better
dashboard
that
has
more
metrics
like
how
fast
it
takes
to
schedule
a
port
how
fast
it
takes
to
bootstrap
couplet.
So
we
used
to
have
the
dashboard,
but
then
it
degraded
to
the
extent
that
it
was
usable
so
like
this
is
the
best
metric.
We
have
right
now
all
right.
B
A
There
is
another
metric
that
shows
how
fast
100
ports
will
be
scheduled
on
a
node
or
something
like
that,
but
yeah.
Maybe
I
can
bring
it
next
time
to
just
educate
and
like
share
knowledge.
A
Okay
and
josh.
A
And
by
the
way,
thanks
for
igt
and
matthias
who
currently
not
here,
our
kublet
tab
is
now
way
better.
A
So
you
see
like
way
less
like
there
is
no
pr's,
like
president
means,
like
all
the
presidents
moved
here,
like
you
see
how
many
pre-submits
they
were,
and
now
they
all
live
separately
and
ci
group
monitors.
So
I
mean
ideally,
we
want
to
look
at
them
from
time
to
time,
but
this
is
what
we
can
concentrate
on
improving
vls
tabs
here
and
I
think
it's
beautiful
and
there
are
a
couple
more
optimizations
on
the
way
like
container
d
tabs
will
merge
together
and
some
continuity
tests
will
move
here.
A
D
A
A
A
A
A
D
A
So
yeah
some
investigations
we
made
in
the
past
uncovered
issues
with
pod
scheduling
with
api
throttling.
So
if
there
is
api,
throttling
and
pod
needs
to
download
config
map,
for
instance,
it
may
delay.
So
if
you
can
take
a
look,
what
port
specifications
are
and
what
it
needs
to
download
to
start
working
or
maybe
mount
that
may
answer
the
question:
why
is
the
delay
is?
But
I
don't
know
like
pink
slack
if
you
need
more
collaboration
on
this.
A
A
Contention,
oh,
this
is
a
famous
pr
that
going
back
and
forth
and
we're
removing
the
test
job
for
it
as
well.
A
A
Okay,
we
are
out
of
agenda
items
and
what
we
can
do
is
to
skip
to
change
to
triage
of
main
dashboard.
If
anybody
interested
to
stay
for
that,
otherwise,
I
will
stop
recording
any
other
agenda
items.