►
From YouTube: Kubernetes SIG Scheduling Weekly Meeting for 20201022
Description
No description was provided for this meeting.
If this is YOUR meeting, an easy way to fix this is to add a description to your video, wherever mtngs.io found it (probably YouTube).
A
Okay,
hi,
everyone
welcome
to
today's
scheduling,
weekly
meeting
and
this
meeting
is
recorded,
so
be
aware
of
what
you're
saying
and
be
respectful
to
each
other.
So
today,
there's
this
week,
there's
not
too
much
items
in
the
agenda,
so
I
will
just
go
through
the
first
one.
First,
the
first
one
is
the
feature
we're
going
to
graduate
to
beta
in
1
20
release.
A
Once
when
scheduling
this
kind
of
parts,
we
try
to
spread
the
paths
evenly
across
the
cluster.
So
it's
a
scoring
plugin,
it's
not
a
filtering
plugin,
but
in
the
last
release
we
think
that
it's,
maybe
not
a
good
idea
to
put
it
there.
It's
a
standalone
plugin
because
most
of
the
logic
has
been
duplicated
with
the
topology
spread
plugin.
So
in
lasso
release
we
make
a
transition
to
sort
of
put
the
logic
into
part
body
spread
and.
A
Give
user
option
to
describe
this
kind
of
option
using
the
arguments
of
the
part
project
spread
plugins
and
in
this
release,
we're
going
to
make
it
better
so
that
when
you
configure
the
path
to
project
plugging
there
is
an
optional
argument
so
that
you
can
define
either.
You
can
specify
that
what
the
type
is,
if
you
don't
specify-
and
the
feature
is
beta
so
that
it
default
to
system
system
means
we
will
apply
a
system
level
spread
constraint
to
that.
It's
a
soft
constraint,
no
hard
constraint,
and
instead
you
can
specify
that.
A
Okay,
I
don't
want
the
system
level
one.
I
know
my
cluster,
I
know
the
workloads,
so
I
want
to
define
the
constraints
myself
so
that
the
cluster
enemy
can
define
this
in
this
field,
so
that
is
basically
the
idea
of
default
topology
spread.
We
want
to
merge
the
similar
things
into
one
single
plugin
as
much
as
possible.
So
this
is
the
background
and
it's
tracked
in
the
in
the
120
release.
I
think
that's
the
only
feature
we
promised
to
deliver
in
this
release
and
the
another.
B
A
The
pod
matrix
one
that
is
raised
by
by
clayton
and
I
think
that
that
cap
got
merged
this
week,
not
100
sure
that
clayton
will
make
it
implemented
as
an
offer
in
120
release.
So
we
maybe
touch
base
with
clayton.
Later
quick
question
can.
B
You
share
link
to
this.
Please,
okay,
this
expression
yeah
the
spreadsheet,
with
the
I
don't
know
how
to
get
to
that.
Thanks.
A
The
dollar
store.
You
want
the
link.
Okay,
suppose
I
would
suggest
you
to
go
to
this
link.
Oh,
this
is
a.
This
is
a
new
one
with
charms.
A
I
think
this
is
a
new
one
that
the
release
team
bring
that
lately.
I
don't
see
it
before,
and
you
can
see
that
enhancement
yeah
there's
a
lot
of
handful
document
here,
so
it
has
one
tracking
here
and
also
the
master
date
like
the
recent
date
is
november,
20,
sorry
november
12,
and
we
will
do
code
freeze,
because
this
release
is
a
short
release
due
to
the
holidays
and
yeah.
That's
basically
the
milestone.
We
should
care.
A
A
If
we,
for
example,
we
finish
50
we
can,
we
can
do,
we
can
just
go
with
50
recently,
that's
no
problem,
because
it's
not
an
external
feature.
So
take
your
time
because
it's
a
hard
time,
it's
a
holiday
season,
so.
B
Yeah,
so
just
to
give
you
a
a
bit
of
context,
I'll
be
working
on
priority
and
fairness
for
that'll
be
most
of
where
I'll
be
spending
my
time
this
quarter,
but
but
yeah
I'm
definitely
going
to
be
working
on
the
particular
framework.
A
And
I
don't
see
much
participants
joined
today.
So
if
you
have
any
topic
you'll
just
spring
up,
we
can
we
have
20
minutes
to
have
a
free
talk.
If
you.
A
A
And
if
you
don't,
I'm
not
sure
if
your
mind,
I
spent
five
minutes
to
give
you
a
latest
update
on
the
scheduler
plugin
sub
ripple.
A
So
basically,
this
repo
is
is
a
place
to
incubate
out
of
three
scheduled
plugins,
so
that
maybe
in
the
in
the
first
time,
it's
hard
to
decide
whether
to
accept
a
plugin
in
the
upstream
right.
So
this
is
a
good
place
for
them
to
incubate
and
exercise
to
involve
the
api,
etc.
So,
recently
sum
up,
some
update
is
that
we
accepted
cap
to
involve
gun
scheduling,
also
also
known
as
co-scheduling
implementation.
A
So
in
the
before
we,
we
accept
your
invitation
just
based
on
the
part
labels,
so
you
define
the
co-scheduling
attributes
in
the
paths
label.
But
lately
we
accept
the
cap
which
raised
the
standardized.
A
I
think
is
a
reload
aware,
scheduling
cap,
so
this
is
a
long
discussion
topic,
so
whether
we
look
into
the
part
the
request
or
we
look
into
other
metrics,
like
real-time
cluster
utilization
metrics.
So
this
cap
made
a
trade-off,
so
it's
still
honored
the
ksm
kubernetes
resource
model.
That
means
it
doesn't
touch
the
heart
requirements.
It
doesn't
touch
the
filtering
plugin
like
to
own
a
part
which
requests
one
gigabyte
memory
to
cpu
status,
so
that
part
is
still
on
it.
Instead,
what
it
tried
to
enhance
is
try
to
enhance
the
score
plugging.