►
From YouTube: Kubernetes SIG Scheduling meeting - 2018-08-23
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
B
Yes,
you
have
seen
that
the
the
poor
request
can
be
sent
and
we
have
passed
first
run
of
you
but
animation
that
we
are
thinking
about
how
to
move
nickel
in
catch
to
another
another
package.
So
we
don't
need
to
leak
some
critic
as
an
object
to
cash
and
we
actually
discussed
about
how
to
whether
we
can
use
metadata
to
pass
sequel
and
catching
fur
to
the
predicate
face.
But
I
noticed
that
currently,
the
metadata
is
shared
by
all
notes:
she's,
a
blocker
because
Oracle
in
cash
right
now
is
per
note.
B
B
B
A
Thinking
along
the
lines
of
passing
the
whole
cache
and
each
predicate
notes,
what
note
it's
it's
basically
running
the
predicate
on
right,
so
the
node
is
also
passed
to
the
predicate
yeah.
So
if
we
have
the
logic
in
the
cache
itself
in
the
e
cache
itself,
that
provides
like
guaranteed
safety
thread,
safety,
good.
B
A
A
B
A
C
A
D
You
me
yeah,
so
a
scientist
with
using
like
accelerators,
for
example,
like
GPUs,
or
something
that,
while
submitting
machine
learning
jobs
on
top
of
kubernetes,
so
they
are
facing
fragmentation
issues
when
they
see
that
therefore
accelerators
present,
but
they
are
not
able
to
use
because
depends
on
two
different
nodes.
So
the
kind
of
built
and
an
extent
scheduled
extender
which
does
bin
packing
for
these
accelerators,
and
we
were
hoping
that
visit
like
reasonable
enough
to
move
this
as
a
default.
Practical,
optional
priority
function
inside
the
default
scheduler
a.
D
Right
now
we
are
just
using
it
as
an
extended
resource
like
if
you
specify
any
extended
resort,
it
will
calculate
how
much
every
node
has
the
amount
of
external
resources,
and
it
applies
the
best
bit
of
algorithm
on
top
of
it.
So
it
will
suggest,
given
notes,
call
to
those
nodes.
Higher
notes
go
to
those
nodes
which
have
like
a
better
fit
for
that
extended
business.
A
D
D
Yeah,
so
there
is
an
option
like
giving
the
weights
to
every
every
priorities,
so
they
can
define
like
if,
if
you
give
similar
weights
to
spreading
and
and
bin
packing,
and
if
you
have
like
a
1/1
like
a
1
GPU
job-
and
you
have
2
GPA
to
try
to
still
spread
and
because
they
both
have
equal
priorities.
But
if
you
get
higher
weights
to
power
bin
packing
I
will
try
to
bend
back
over
scattering.
So
ok.
A
E
A
B
A
D
D
Yeah,
so
this
is,
you
can
take
it
as
a
default
case,
like
you
have
for
resources
for
resources
into
resources,
and
you
have
a
job
spec,
which
is
like
first
job
spec
requires
two
resources.
Second,
one
requires
four
third,
only
twice
for
yeah
the
first
one
goes
to
the
four
node
and
then
it's
a
pending
state,
and
since
these
are
machine
learning
jobs
along
for
a
long
time
and
you're
facing
issues.
A
B
A
A
E
A
E
A
E
Another
thing
is
that
it
actually
breaks
one
of
the
tests.
It's
not
flaky,
it's
just
that
in
some
of
the
tests
it
doesn't
run
the
cube
light.
You
just
write
an
API
object
like
local
house
or
to
the
API
server,
so
that
we
suppose
there
is
unknown
about
that
case,
but
the
lifecycle
manager
still
works,
so
it
treated
the
node
as
kind
of
status
on
now,
so
that
the
behavior
is
that
we
try
to
apply
the
tend
to
there
to
the
fecit
node
in
the
test
cases.
A
E
F
Yes,
I
seen
before
unreachable
for
your
test
and
as
for
your
things,
I
think
that
so
they're
reasonable,
because
even
universe
gather
our
wish
will
now
to
schedule
the
sapota
to
a
fake,
known
right.
I.
Think
yes,
for
for
the
for
the
behavior,
this
is
a
cracked,
but
we
need
to
confirm
is
a
test.
I.
Think
I
just
see
the
comments.
So
this
morning
the
task
is
from
from
Cooper
City
our
cool
control
right
yeah,
the
owners
tested.
The
medina
is
60-hour.
F
F
Photographs
gathering
I
think
I
I
do
something
for
type
about
the
queue
and
I
see
which
give
all
comments
about
whether
we
should
consider
our
administer
to
some
work
in
the
admission
to
for
the
transaction.
I.
Think
I
will
put
some
time
for
this
for
this
option
to
see
if
less
is
a
doable
yeah
and
the
photo
appear
part
I
think
we
are
going
to
use
the
notation
for
the
photo
group
name
in
the
powerful
alpha
for
the
Alpha
feature
right
now,
so
I
sing
a
fool:
I
don't
get
more
common,
so
about
HJ
definition.
F
A
F
So
I
think
I,
just
I
need
to
confirm
with
radiator
and
wish
to
see
if
there
any
more
comments
about
me
except
are
and
I
will
do
some
work.
Maybe
some
words
about
some
researcher
about
wish
suggestion
about
the
transaction
between
for
the
council
gathering,
yeah
I
believe
you
some
happens
here.
I'm
also
thinking
on
whether
we
can
do
a
perk,
give
a
separate
crew,
class
or
separate
design
document
to
get
the
API,
maybe
murder
or
can
I
be
higher
ed
it
for
for
one
thought,
wow
that
sounds.
B
A
Thank
you,
yeah
all
right
image,
vocal,
is
a
priority.
So
I
know
that
image
locality
priority.
We
have.
We
have
some
issues
with
the
test,
because
the
test
images
are
not
uploaded
to
G
cRIO
I
tried
I
was
able
to
upload
some
of
the
images,
but
not
the
top-level
item
in
the
tcri.
Oh
I
I
tried
many
times
reaching
Jeff
with
the
the
idea
of
ixd
vine
and
github
to
upload
the
images
for
us,
but
Jeff
was
out
of
office
for
for
several
days
and
then
then
he
came
back.
A
He
I
tried
to
reach
him
again,
but
he
was
not
responsive
anyway.
I
will
probably
try
again,
but
please
feel
free
to
ping
him
on
github
as
well.
If
there
is
any
further
updates
from
Harry
or
other
folks,
please
let
us
know
is
there
anything
else
that
we
should
know
about
him.
I
responded
recently.
Oh.
E
B
A
Thanks
there
is
some
effort
on
finishing
or
finalizing
the
party
scheduling
policies,
I've
been
working
with
Yaseen
and
Tim
old
Pharaoh,
and
this
it's
getting
there
almost
so
we
have
some
things
to
finalize,
recording
the
matching
and
how
these
policies
are
matched
against
parts,
but
it's
almost
there.
If
you
are
interested,
you
can
take
a
look
at
the
PR,
it's
in
our
spreadsheet
as
well.
As
you
know,
our
No
meeting
notes.
A
E
E
They
don't
add
the
Toleration
on
to
the
demon
side
paths
when
doing
the
calculation
see
say
how
many
knows,
sorry,
how
many
paths
should
be
running
on
how
many
nodes,
so
it
kind
of
off
in
the
last
step,
I
think
one
of
the
classes
PR
applies
the
on
schedule,
storage
onto
the
path
but
internal
race.
So
the
path
was
it
created,
for
example,
the
ice
they
are
to
knows,
but
actually
only
only
one
path
was
wrong.
So
I
think
that
is
a
minor.
My
usually.
B
E
A
So
we
have
only
ten
days
basically
the
for
the
code
trees
and
please,
if
you
have
peers,
please
send
them
as
soon
as
possible,
and
please
pay
me
don't
be
shy.
Please
ping
me
on
slack
or
whatever,
to
take
a
look,
Oh
Klaus
as
well,
and
we
have
more
more
approvers
now.
Actually
Ravi
is
an
approver.
Now
Ravi
can
help
reviewing
and
approving
Jonathan
or
mystery
kid.
His
idea
and
github
is
also
a
reviewer.
A
E
A
C
E
C
Regular
I
spoke
to
Derek,
who
is
part
of
the
steering
committee,
so
the
idea
is
to
move
D
scheduler
out
of
incubator
repo
and
for
it
in
the
sick,
scheduling
the
six
sponsored
ripples.
I
have
sent
out
an
email
to
the
six
scheduling
group,
so
the
way
forward
now
is
everyone
needs
to
like
four
actively
working
in
this
loop
needs
to
be
such
last.
One
saying
that
it
is
good
to
go.
If
you
are
fine
with
it,
then
we
can
actually
get
started
on
so.
A
C
E
A
I
see
yeah,
so
this
is
basically
III.
I
was
kind
of
expecting
this
to
be
moved
out,
righted
standard
component,
which
is
maybe
not
enabled
sorta
like
an
for
component
of
kubernetes,
which
is
basically
got
a
ship.
It
gets
shipped
with
the
rest
of
kubernetes,
but
not
enabled
by
default,
looks
like
we
are
not
gonna
do
that
at
least
in
this
corner,
or
with
this
release
driving.
A
A
Yeah,
so
you
know,
given
that
we
are
very
close
to
the
code
freeze-
and
this
is
like
a
huge
I-
cannot
say
huge
feature,
but
it's
it's
a
it's
a
component,
it's
it's
more
than
just
so.
It's
it's
a
bigger
effort
and
it
may
impact
clusters
if
you
wanna,
move
it
as
part
of
community.
So
we
need
to
give
ourselves
a
little
bit
more
time
to
ensure
that
the
quality
is
there
before
we
can
move
it
to
a
as
a
standard
component,
his
component.
A
C
E
A
C
A
So
one
issue
it
like
you
know
getting
it
from
a
separate
location,
is
that
they
will
have
to
do
extra
steps
and
probably
have
to
go
and
read
some
more
documents
and
basically
to
find
out
where
to
get
it.
How
to
compile
it
and
stuff
like
that,
if
it's
released
as
a
part
of
kubernetes
images
and
someone.
E
C
A
C
Atom
like
we
can
move
to
our
six
sponsor
Drupal,
but
we
will
let
it
stay
there
for
some
more
time
and
say
we
are
able
to
release
the
sorry.
We
are
able
to
match
the
released
cadence
of
Cuban.
It
is
along
with
the
features
that
we
want
because,
as
of
known,
the
features
that
we
have
are
like,
like
there
are
only
three
or
four
features,
that
we
have
indie
scheduler
and
it's
mostly
one
strategy
or
one
feature
that
people
are
always
interested
in
using
it
much
more
feedback
from
the
users
when
it
comes
to.
C
A
Well,
maybe
we
are
running
out
of
time,
so
we
cannot
really
expand
this
discussion
to
basically
the
details
of
things
that
are
neither
there
but
definitely
I'm
interested
in
and
looking
up
things
and
discuss
this
further
with
you
guys
to
see
what
we
choose
you
are
thinking
about
and
what
timelines
you
think
we
can.
We
can
have
this
as
a
part
of
kubernetes
standard
components,
think
it's
useful
still
for
many
users,
and
especially
those
who
run
larger
clusters,
probably
would
benefit
from
something
like
a
dis,
scheduler
sure.