►
From YouTube: Kubernetes SIG Node 20200303
Description
Meeting Agenda:
https://docs.google.com/document/d/1j3vrG6BgE0hUDs2e-1ZUegKN4W4Adb1B6oJ6j-4kyPU
A
B
There's
been,
you
know,
comments
and
feedback
on
all
these
PRS
and
I.
Think
they've
all
been
addressed,
the
top
two
or
feature
related,
the
first
one
is
about
you
know:
alignment
of
resources
with
multiple
containers
in
a
pod
and
the
others
are
the
bottom
three
or
more
of
the
test,
related
PRS
and
so
I.
Don't
necessarily
think
there's
any
need
to
have
any
specific
discussion
on
other
than
just
saying
we
need
in
order
to
make
beta
for
118.
These
need
to
be
merged
by
this
Thursday
at
5:00.
If
there
are
any
questions.
B
C
So
it's
outside
of
Kevin
scope,
so
that
one
actually
just
needs
your
approval
and
then
the
one
below
that
is
the
feature
gauge
enable
enabling
that
by
default.
So
obviously
we
do
want
to
do
that
until
we've
everything
has
taken
care
of,
but
once
the
that
eight
seven,
seven
five
nine
is
merged.
We
can
move
forward
with
the
rest.
A
Okay,
that
sounds
good.
I
was
out
of
the
office
yesterday,
so
I
apologize
for
the
lag,
but
I
will
either
conclude
the
review
for
the
things
that
we
were
already
reviewing
or
trying
to
get
through
these
items
as
best
as
possible
note.
The
second
items
are
just
related
to
Docs
as
these
manage
right.
So,
yes,.
C
A
C
C
A
B
A
A
D
Sure
so
Clayton
is
going
to
the
cubelet
code
to
figure
out
if
there
is
any
slowdowns
and
if
we
can
make
improvements
in
reacting.
So
he
has
a
PR
open
with
a
bunch
of
issues
and
fixes
that
he's
testing-
and
one
thing
he
pointed
out-
is
that
the
cubelet
would
benefit
from
knowing
that
a
pod
or
a
container
is
basically
not
found
from
the
CRI
and
I
was.
I
I
was
looking
into
it
and
I
think
that
the
best
option
to
do
it
would
be
for
the
CRI
runtimes
to
return
a
custom
error.
D
A
E
There
are
two
peers,
like
one
is
a
bug-fix,
so
I
asked
on
the
previous
meeting
last
week
meeting
about
reviewing
it
and
it
it
was
kind
of
slight
activity
there,
but
if
it
can
be
reviewed
and
merged
before
the
deadline,
that
would
be
really
good
and
the
second
one.
Sorry,
the
first
one
is
documentation,
update
the
decode
is
already
merged
and
it's
in
release
branch
so
and
that
that
is
just
update,
documentation
and
is
blocked,
because
so
it
waits
for
signals
review.
Basically,
okay,
thanks.
A
F
We've
got
time
looks
like
we
do
so.
In
the
past
we
found
that
an
it
containers
prevented
HPA
from
working,
because
we
weren't
reporting
stats
in
the
cubelet
stats
provider
and
so
in
the
Londa
metrics
pipeline.
If
there's
missing,
CPU
or
memory
for
a
container,
the
HPA
refuses
to
take
action.
So
we
made
a
fix
in
the
first
linked
PR
for
any
runtimes
that
use
the
CR,
the
CRI
stats
provider
and
the
fix
is
basically
to
just
populate
empty
metrics,
but
with
zeros
for
CPU
and
memory
I'm.
Just
wondering.
F
Does
anybody
see
any
reason
we
couldn't
do
the
same
thing
for
the
CSE
advisor
stats
provider
because
we
actually
have
the
same
problem
for
any
runtimes
that
use
the
C
advisor
stats
provider,
which
basically
is
cryo
and
docker.
If
there
ever
containers
that
don't
have
those
statistics,
populated
and
C
advisor,
then
HPA
won't
scale
them.