►
From YouTube: Kubernetes SIG Scheduling Meeting 2018-06-14
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
Hello,
everyone
welcome
to
see
be
scheduling.
Weekly
meeting
today
is
our
5
p.m.
Pt
aspect.
Time
version
of
our
meeting
I
don't
have
a
whole
lot
of
items
for
today's
meeting,
so
hopefully
we
can
cut
it
short
a
little
bit.
But
if
you
have
other
items
you
can
you're
more
than
welcome
to
talk
about
those.
So
the
items
that
I
have
under
on
the
agenda
is
one
follow
up
on
image:
locality
from
Harry
Harry.
A
B
B
There
will
be
like
forty
forty
eight
parts
being
scheduled
to
the
nodes
of
is
this
image
exists?
Well,
the
expected
parts
number
will
be
thirty,
so
I
think
this
should
be
a
reasonable
number
and
even
in
the
worst
case,
if
the
image
is
very
large
and
then
the
priority
score
is
is
a
maximum,
then
the
hotend
nodes
will
only
have
same
thing
same
thing
or
more
has
been
scheduled,
so
it's
still
not
very
bad.
So
I.
A
A
B
B
Okay,
so
that
is
the
first
experiment
I
did
and
the
second
experiment
I
did
is.
We
can
actually
give
a
discount
when
calculating
the
score
the
discounted,
calculated
by
considering
the
distribution
of
the
given
image.
So
basically,
the
final
score
will
be
the
real
score
and
multiply
this
discount,
and
if
we
will
use
this
option,
the
test
result
will
be
more
stable,
because
even
this
priority
has
a
default
rate
than
the
final
score
will
be
calculated.
B
B
Is
if
this
image
has
been
existed
being
foregone
first
written
note,
then
I
say:
if
there's
more
notes
has
this
image
then
we
can
give
is
relatively
higher
score
during
calculating
because
it
it
means
it's
distributing
it's.
This
distribution
is
more
balanced.
So
this
is
how
I
calculate
calculated
this
score.
It's
just
another
porch
I
see
I,
see
yeah.
B
A
B
A
B
A
A
C
A
C
A
A
Basically,
and
also
there
are
some
small
differences-
are
there
than
just
the
plugins,
and
there
is
this
new
kind
of
reservation
and
admission
state
in
the
scheduler
that
helps
us
enable
some
more
advanced
features
such
as
maybe
gang
scheduling
and
dynamic
binding
of
resources,
and
all
of
that,
so,
if
you're
interested
please
take
a
look.
I
have
already
sent
an
email
to
a
deciduous
scheduling
mailing
list,
with
the
link
and
I
also
mentioned
in
that
same
email.
A
In
order
to
access
the
document
you
need
to,
you
need
to
be
on
the
kubernetes
dev
mailing
list,
which
is
one
of
our
Google
Groups.
If
you're,
not,
you
won't
be
able
to
access
the
the
document,
I
already
received
a
bunch
of
other
requests
to
give
access
to
the
document.
I
cannot
grant
individual
access
to
the
document.
A
D
The
definition
is
not
finalized
will
continue
to
to
up
that
updates.
That's
API
API
definition
and
another
point
is
I.
As
we
sing
half
of
line
with
the
coop
arbitrator
will
also
leverage
a
scheduling
framework
choose
the
pulpit
refer
to
support
to
support
bachelor
bachelor
cloud.
I.
Think
the
lesson
excited
to
share
the
same
framework
to
a
wise,
the
code,
duplicate
but
I
think
some
some
concept,
or
maybe
some
term-
maybe
batch
special
ticket,
for
example,
cure
or
something
others.
D
This
will
be
included
in
the
kebab
trader,
but
the
common
part
will
be
to
scaling
framework.
So
I
think
this
will
be
our
direction
to
for
this
one
yeah
and
already
have
a
very
traveling
first
reimplantation
about
it
or
API
part
hour.
They
openly
use
you
in
the
kebab
Kreacher
a
fool.
Anyone
how
interesting
or
how
any
comments
to
the
to
live
apart.
A
C
A
If
the
two
scanners
can
share
the
same
framework,
it
will
be
great,
so
if
we
can
make
that
necessary
changes
to
the
current
a
proposal
and
make
it
possible
to
use
it
for
building
the
batch
scheduler
as
well,
it
will
be
great
because
it
will
reduce
the
amount
of
maintenance
that
we
are
gonna
have
in
the
future.
Yes,
please
take
a
look.
I
actually
replied
to
one
of
your
comments
on
the
document
now
take.
A
D
I
think
we
need
to
share
the
same
scheduler
framework
because
we
don't
want
to
you
know
it
was
someone
interesting
in
the
true
part
that
should
be
really
hard
to
to
learn
different
thought
of
different
of
in
worker.
That's
true!
Yes,
this
doesn't
Google
for,
for
the
community,
I
think
we
need
to
share
the
same
scene
and
for
the
different
workload
attack
with
there
may
be
some
customized
or
some
special
concept
alter
me
in
the
different
the
scheduling.
But
it's
our
scheduling,
also
or
a
lot
of
the
student
is
saying
yes,.
A
Yes,
yeah
with
that
should
be
our
direction.
In
my
opinion,
we
should
try
very
hard
to
make
this
design,
or
maybe
the
design
already
supported.
I
I
cannot
really
because
I
don't
have
the
natural
context
about
the
requirements
of
gang
scheduling
but
feel
like
it
might
be
already
there
or
might
be
close
to
what
we
I
wanted
to
guess.
Oh,
it's
it'll
be
great
to
use
that
framework
for
both
one
question
from
you
is
that
in
so
you
are
building
cube
arbitrator
as
a
as
an
incubator
project.
Right,
yes,
yeah
it's
in
a
separate
report.
A
D
D
We
will
have
more
results
for
this
part
of
to
do
the
contribution
for
the
kebab
traitor,
I.
Think
of
template
for
the
a
batch
system
to
the
to
the
kubernetes.
You
know
we
we
our
products,
our
SF
ego.
We
have.
They
were
experiments
about
batch
system
for
this
part
yeah
and
for
the
other
contribution
I
think
we
have
some
user.
D
We
do
have
some
contributor
away
near
the
forum
chosen
twosome
ho
another
startup
company
in
front
channel.
They
would
like
to
wrong
imax
night
on
the
kebab
feature
to
support
a
GPO
top-loading
for
this
part,
I.
Think
of
this
booby
NASA
set
up
for
this
part
now
I
know
I,
think
that
there
also
a
I
think
there
also
a
TNT
tea,
maybe
TNT
I
can't
remember
the
name
of
that
part.
D
They
also
give
some
contributor
to
this
to
to
the
kebab
feature
for
this
part
and
a
startup
which
I
called
wrong
China,
they
I
think
this
should
be
tripod.
They
do
the
lead
to
the
integration
from
the
Cooper
flow,
the
chaff
operator,
to
the
integration
with
the
crew
operator
to
use
it
to
support,
accounts
caddying
yeah,
so
we
have
a
pre
worst
version,
a
single.
This
is,
though,
the
user
from
Ross
and
yes
and
because
ATI,
maybe
maybe
change
already
I-
think
neither
how
some
updated
to
the
chief
operator
to
make
it
API
relying.
D
Api
today
is,
it
is
still
using
pdb.
Yes,
the
previous
version,
where
I
using
pdb
to
define
the
minimal
while
well
yeah
and
in
the
new
for
the
new
api
single
image,
introduce
another
CRD
to
replace
pdb
people.
He
calls
a
PB
using
for
the
other
purpose.
You
know
you
know
the
cobra
and
has
I
think
I
will
replace
this
one.
Yes,
I
think
important
the
chain
rule
about
the.
If
you
have
our
yeah,
I'm
a
full
for
our
customer
I
think
maybe
I
can't
deal
with
the
name
of.
D
A
Okay,
these
are
all
that
I
wanted
to
say
I,
just
since
Robbie
is
here,
I've
already
pink
Robbie
to
update
the
resource
code,
our
documentation,
because
we
need
that
new
information
about
how
to
limit
users
at
certain
priorities,
priorities
for
111,
and
we
need
to
link
that
to
our
release.
Notes
for
111
Robbie,
please
want
with
us
yeah.
C
C
Instead
of
having
like
two
different
frameworks,
there's
two
different
schedulers
as
we
were
thinking
earlier.
If
you
could
combine
that
into
single
thing,
it
would
be
helpful
for
us
to
and
we
could
get
started
on
contributing
as
well,
because
we
have
some
customers
who
are
interested
in
using
it.
But
as
of
note,
since
nothing
is
in
beta,
which
would
be
Petra
view
in
open,
GFP
I
have
not
encouraging
the
customers
at
this
point
of
them.
So
if
we
have
a
single
framework
or
a
scheduler
that
could
do
both,
it
could
be
really
helpful.
C
A
So
for
the
as
far
as
framework
is
concerned,
I
I
guess
it
will
take
at
least
a
couple
of
release
cycles
before
we
can
see
a
working
version
of
the
scheduling
framework.
So,
depending
on
your
timelines
and
your
requirements,
it
may
or
may
not
be
ready
for
your
customers
in
time,
but
in
terms
of
combining
it
to
schedulers.
A
Yes,
that
will
certainly
be
beneficial
to
customers
and
also
us
as
developers,
because
it's
gonna
make
things
easier
for
us
if
we
can
have
just
one
scheduler
as
opposed
to
two,
whether
we
can
really
pull
it
off
without
making
scheduler
too
complex
or
not
as
it
is
the
question
that
we
need
to
answer.
If
we
can
actually
find
a
solution
for
that,
I
agree
that
it'll
be
great.
C
A
Please
also
look
at
that
scheduling
framework
is
on
document.
I
know
it's
a
little,
maybe
it's
a
little
longer,
although
I
try
to
basically
short
make
it
short
as
much
as
possible.
It's
like
10
pages
from
now,
but
I
couldn't
really
make
it
much
further
than
this
I
try
to
basically
skip
all
the
implementation
details
in
a
document
to
focus
on
the
architecture.
A
If
you
don't
want
to
go
through
all
the
details,
I
still
encourage
you
to
go
and
read
some
of
the
some
of
the
design
made
main
design
areas
towards
like
how
the
reservation
and
admission
of
POD
works
and
maybe
look
at
the
use
cases.
Maybe
that
actually
not
it
may
not
be
a
bad
idea.
Look
at
the
use
cases,
part
of
the
design
doc,
and
then
you
will
see
some
information
and
try
to
look
those
up
in
the
document.
If
you
don't
have
time
to
go
through
the
whole
document.