►
From YouTube: Kubernetes SIG Scheduling Weekly Meeting for 20230406
Description
Kubernetes SIG Scheduling Weekly Meeting 2023-04-06T17:00:00Z
A
Thank
you
all
right.
Welcome
everybody
to
another
sea
scaling
meeting
today
is
April
6th
2023.
My
name
is
Aldo
I'm
gonna
be
your
host
today.
As
you
know,
this
meeting
is
recorded.
So
please
adhere
to
the
cncf
code
of
conduct
all
right.
So
we
have
few
topics
in
the
agenda.
The
first
one
Abdullah
I,
believe.
Let
me
put
this
one,
so
we
I'm
doing
this
working
on
the
annual
report
for
the
for
the
project.
Scheduling.
Do
you
want
to
say
Do
you
wanna
talk
about
this
Abdullah.
B
B
Like
please
take
a
look.
I
will
respond
to
the
comments
today,
just
like
a
quick
announcement
that
we
have
the
report
up
for
review.
Please
take
a
look.
B
Yeah
again,
as
I
mentioned
just
please
take
a
look
at
the
report
once
it's
merged.
Maybe
we
can
talk
about
about
it
like
how
we
can
improve
diversity,
reviewers
like
having
more
reviewers
helping
us
and
and
whatnot.
C
Okay
and
the
next
one
is
to
redefine
the
ownership
of
the
non-life
cycle
controller.
So
basically
my
memory-
it
was
developed
by
class,
who
was
the
sixth
scheduling
co-chair
by
that
time
and
then,
but
in
terms
of
functionality,
actually
the
components
across
ccaps,
I
would
say
cross
signal
and
six
scheduling,
so
I
think
it
was
initially
belong
to
six
scheduling,
but
later
on
was
transfer
the
ownership
to
six
no
I
would
say,
and
then
now
it
seems,
there's
any
some
discussion
to
talking
about.
C
A
Yeah,
the
the
reason
why
I
opened
the
issue
is
because
somebody
was
trying
to
propose
a
fix
or
some
changes
in
a
gap
for
in
c-gaps.
A
A
So
that's
why
I
brought
it
up,
but
it
sounds
like
it
should
be
under
our
our
sync.
So
if
there
are
no
no
concerns,
we
can
send
a
PR
to
update
the
owner
file.
C
B
Just
please
give
us
advocate
here,
like
most
of
the
controller.
What
the
controller
is
doing
is
actually
trying
to
understand
the
node
status
and
then
just
applying
a
tent
or
a
or
like
yeah
flying
things,
basically,
so
why
it
was
developed
by
someone
from
sex
catering
in
the
past
I
don't
know
if
that
behavior
is
something
we
are
the
experts
to
maintain,
because
I
mean
applying
the
tense
is
not
the
biggest
thing
here
in
the
in
the
controller.
It's
basically
understanding
what
the
node
status
is.
B
So
just
take
that
into
consideration.
If
we
are
going
to
take
that
in,
we
need
to
make
sure
that
we
know
that
we're
expanding
our
you
know,
area
of
expertise
to
how
we
translate
the
node
status.
A
So
basically,
one
of
actually
the
cap
proposal
is
to
split
this
controller
into
two,
because
it's
doing
two
actions
or
two
is
doing
basically
to
related
but
separate
things.
One
is
yes
adding
the
things
to
the
nodes
based
on
the
status
and
second,
it
is
a
victim
thoughts
that
don't
don't
have
the
Toleration.
A
So
maybe
once
the
controller
is
split,
it
actually
makes
sense
to
actually
to
give
one
of
them
to
node
and
the
other
one
given
in
in
scheduling,
because
the
things
and
tolerations
is
under
the
domain
of
of
scheduling
right.
So
maybe
that
that
could
be
the
outcome
once
the
the
two
con.
The
two
functionalities
are
split
into
two
controllers.
A
Do
you
agree
with
that
future
setup?
Of
course
we
have
to
bring
it
to
signal,
but.
B
C
I
think
maybe
so,
basically
this
kind
of
physical
we
can
have
a
top
level
on
this
files
and
also
with
some
soft
folders.
If
that's
very
specific,
to
Signal
those
C
gaps,
we
can
follow
the
pattern
that
we
involved
Michelle
or
other
sorry.
C
A
That
is,
that
is
true.
Actually,
yes,
so
the
the
two
functionalities
are
currently
split
into
folders,
so
I
we
could
already
do
the
separation
using
the
the
same
pattern,
so
I'll
I'll,
I'll
craft,
a
PR
and
send
it
for
review
for
to
signal
and
and
the
rest
and
see,
scheduling,
I
and
see
gaps
as
well,
of
course,
because
they
are
there
overall
Gatekeepers
of
Cube
controller
manager.
A
So
and
then
we
we
can
continue
the
discussion
that
way
in
the
pr
Maybe.
A
No
I
guess
kante
wanted
us
to
discuss
because
it's
late
in
the
East
Asia
time
zone,
so
I
can
summarize
what's
going
on
here.
Oh
yes,
actually
he
left
this
comment
here.
So.
A
This
is
a
an
old
feature
that
never
graduated
from
alpha
into
other
stages.
Basically,
what
we
have
here
is
an
admission
plugin
that
can
understand
the
this
annotation
in
a
namespace
object
and
what
the
plugin
does.
What
the
admission
plugin
does
is
that,
based
on
The
annotation,
it
introduces
node
selector
into
the
nodes
that
much
that
are
created
in
this
namespace.
So
basically
it's
defaulting
no
selectors
for
annoying
space
and
again
this
never
graduated
from
Alpha.
A
So
there
are
two
possibilities:
one
is
to
well
graduate
to
Beta
or
remove
the
feature.
A
I
think
kante
is
proposing
to
proceed
with
with
a
Beta
release
and
we
are
kind
of
looking
for
feedback
because
feedback
from
the
community,
because
this
Behavior
can
be
achieved
through
webhooks
today
right,
so
they
it's
not
clear
whether
this
Behavior
should
be
part
of
Upstream.
A
A
But
even
if
we
proceed
into
a
beta
stage,
most
likely,
we
should
abandon
The
annotation
approach
because
it's
hard
to
validate
and
but
not
so
an
Alternatives
would
be
to
provide
a
let's
say,
an
official
web
cook
or
implement
this
as
still
as
an
admission
plugin,
but
based
on
a
on
a
proper
API
field
instead
of
an
annotation.
A
So
there
are
a
few
possibilities
there.
So
if
you're
interested
please
get
into
this
issue
and
I
believe
condition,
either
a
dog
or
a
PR,
maybe
not.
A
Okay,
any
any
last
minutes
last
minute,
questions
or
topics
from
the
others
from
anybody
in
the
meeting.
A
Okay,
well,
actually
I
I'll
actually
prompt
way
and
Abdul
life.
You
have
any
any.
Are
your
learning
in
any
direction
in
respect
to
this
topic
of
the
admission
plugin.
B
I
would
like,
if
you
are
up
to
me,
I,
would
deprecate
I
think
we
have
a
proper
solution
in
web
hooks,
but
I
would
ask
people
who
have
been
like
kubernetes
much
longer
like
Clayton
and
others,
whether
they
and
Jordan,
whether
they
consider
an
alpha
feature
that
has
been
there
for
ages
as
a
almost
like
a
de
facto
supported
thing,
and
that
we
shouldn't
deprecate
just
because
it's
been
there
for
a
very,
very,
very
long
time
and
people
have
some
people
have
used
it
like
started
to
think
that
it
is
just
you
know.
A
So
my
my
thinking
in
terms
of
deprecation
would
be
that
we
we
take
some
time.
We
we
announce
it
and
then
we
remove
it
in,
let's
say,
as
any
other
beta
feature,
let's
say
three
releases,
but
the
question
Still
Remains
whether
we
should
keep
the
behavior,
regardless
of
how
long
we
take
to
from
how
long
the
deprecation
process
takes.
A
And
we
we
have
I,
guess
the
closest
precedence
I
can
think
of
that.
That's
a
similar
thing
would
be
limit
ranges
which
actually
I
don't
remember
if
they
are
namespaced
or
not,
but
that's
kind
of
like
the
the
defaulting,
the
only
default
pod
defaulting
mechanism
I
can
think
of
that
kind
of
could
be
used
as
a
precedence
but
regardless,
regardless
of
back
back
again
to
deprecation
the
The
annotation
has
Alpha
in
the
name.
A
So
definitely
you
cannot
stay
like
that
like
that,
so
we
all
have
to
change
it,
we'll
have
to
find
a
better
API
and
we
will
have
to
deprecate
in
a
stage
manner
to
avoid
avoid
breaking
users.
Unexpectedly,.
A
There
is
a
particular
decision
here:
I'll
try
to
involve
Jordan
and
Clayton
on
this
topic.
A
Oh
last
minute,
Shameless
plug
the
queues,
the
cube
sub
project
within
six
scheduling
is
getting
a
release
soon,
I'm
hoping
by
the
end
of
today
so
I
expect
that
to
come
soon,
and
if
you
have
questions
about
Q,
you
can
ask
in
Sea
scaling
or
in
working
group
batch,
but
preferably
in
working
group
batch
channels.