►
From YouTube: Introduction to Kubernetes Scheduler Wei Huang IBM SIG-Scheduling Cochair OpenShift Commons Briefing
Description
Introduction to Kubernetes Scheduler
Wei Huang IBM
SIG-Scheduling Cochair
OpenShift Commons Briefing
June 29 2020
A
Welcome
to
another
OpenShift
Commons
briefing
as
we
are
want
to
do
on
Mondays
is
create
a
space
for
some
of
the
upstream
projects
to
have
AMA
introduction
sessions,
and
today
the
coop
scheduler
is
the
topic
of
the
day
and
way
is
going
from
IBM.
That's
going
to
tell
us
a
little
bit
about
who
he
is
and
introduce
himself
and
then
give
us
an
introduction
to
the
kubernetes
scheduler
so
way,
take
it
away
and
we'll
have
live
Q&A
at
the
end.
A
B
What
skips
calorie
is
what
cubic
keep
schedule
does
and
how
it
works.
Also,
in
this
title
here
how
scheduler
works
here
and
what's
the
latest
development
on
the
scheduling
side
and
the
last
part
is
to
because
we
are
expert
experiencing
some-
can
component
config
API
change
in
the
later
releases,
so
I
want
to
highlight
some
configuration
changes.
You
want
to
be
aware
of
the
last
part
and
the
last
part
is
to
how
to
contribute
to
keeping
any
schedule.
Okay,
the
first
part
is
what
scheduler
is
and
what
schedule
it
does.
B
So,
basically,
I
can
just
use
two
more
sentence
to
explain
what
scheduler
does
so.
What
schedule
does
is
to
do
some
magic
to
find
personnel
for
incoming
part
to
be
placed
so,
for
example,
in
this
and
the
upper
images
there
is
part
which
has
the
field
now
now.
That
means
it
doesn't
have
a
proper
now
to
be
placed
in
yet
so
the
scheduler
comes
into
the
stage
and
do
some
magic
and
fine,
the
best
nil,
to
put
their
note,
sorry
put
the
pad.
B
B
B
B
B
B
So
what
you
can
expect,
if
you
probably
they,
be
our
object,
we
off
cause
we
created,
but
there's
no
path
back
in
that
department.
Why?
Because
back
to
our
slides,
what's
your
career
Department?
He
can't
because
ABS
over
it
be
some
sort
of
cost
to
some
basic
check
and
the
persistent
API
object
in
this
case.
B
The
game
object
here
is
a
Department
position
it
to
the
EDC
and
then
Control
Manager
comes
into
stage
and
try
to
get
the
department
object
from
the
API
so
about
here
we
intentionally
shut
down
the
controller
menu,
so
there's
no
path
creating
here
you
can
see
no
path
and
department
managed
replica
set,
so
no
because
I
either
okay.
So
how
to
pick
that,
let's
move
over,
get
your
manager
back.
B
Okay,
there
should
be
some
time
a
second
for
the
tournament,
either
good
back.
No
one
sees
gets
back.
He
needs
to
do
some
internally.
We
call
that
cash
and
sinker,
but
you
may
need
some
time
to
wait
for
you
to
get
back
while
we
are
waiting,
let's
continue
to
the
map,
so
the
continue
the
controller
manager
takes
over
and
get
the
Department
event
back
and
internally.
It
well
created
the
object.
It
needs
to
ensure
the
running
path
expect
its
desire.
B
The
replica
numbers,
for
example,
chrono
manager,
will
try
to
see
what's
the
ruckus
out
there
and
how
many
paths
are
managed
by
the
Erica
set
and
juice
Americans
can
we're
consuming
and
make
sure
it's
state
is
desired?
Okay,
let's
go
back
yeah.
You
can
see
that
right
now
there
is
a
person
there's
replica
cell
career
and
those
are
there's
Patrick.
B
B
B
Schedule
finds
the
kind
of
worker
the
fossil
worker
for
the
part
to
be
placed
and
tequila
can't
dig
a
speck,
so
the
most
important
failed,
as
I
mentioned.
Is
the
speck
got
no
name
so
my
schedule
recites
the
no
name
to
the
current
worker.
Then
its
job
is
now
and
next
is
couplet
coming
to
stage
and
every
every
kid
played
in
issue
Kinnaird
wash
from
the
past
and
if
the
past
specter
new
name,
because
to
his
own
no
name,
then
he
says
okay.
It's
me
to
take
care
of
this
path.
B
B
Let's
take
a
look
so
best:
okay,
there's
two
major
phases:
the
final
decision
is
made,
so
the
first
is
called
filter
in
the
before
we
call
the
predicate,
but
I
will
use
the
two
words
interchangeable
play
in
this
session
and
if
you
look
at
some
documents,
if
there's
some
a
bit
o
document,
maybe
it's
use.
The
word
predicate
bracket
and
future
are
the
same
ran
out
because
we
are
doing
some
more
spectrum.
We
are
more
pre-internet
to
use
the
filter
and
use
further
parting
to
describe
this
phase.
B
So
practical,
it's
hard
constraints
that
you
apply
to
your
workloads
back,
for
example,
you
know
you
may
say:
I
want
my
pal
to
have
a
two
gigabyte
memory
and
one
core
CPU.
If
there's
no,
no,
no
fits
then
just
put
my
pad
honey
there.
I
don't
do
anything
else.
So
this
is
a
hard
constraint,
it
must
be
satisfied
and
all
hard
constraints
are
ended.
B
There's
an
internal
Q
inside
scheduler
and
the
Q
pups
are
apart
one
by
one
and
once
a
part
comes
in
to
schedule
the
main
logic,
and
it
will
run
all
the
registered
predicates
functions
against
each
node
and,
of
course,
some
that
were
fair
and
some
never
pass.
For
example,
here
there
straighteners
will
past.
So
by
the
end
of
this
phase,
the
upper
is
that
we
get
a
three
node
list.
We
know
that
LS
three
now
can
fit
the
part.
B
B
B
If
we
apply
note
affinity
to
the
part
and
say
so,
it
gives
some
some
room
should
not
only
choose
kind
of
worker,
but
here
is
just
say:
I
don't
want
my
path
to
be
put
to
their
third
worker,
so
it
can
be
working
while
welcome
to
and
there's
also
a
hint
here.
Is
that
if
you
look
at
your
spec,
if
there
are
some
key
word
like
require
means
it's
the
most,
so
it
usually
mapped
to
the
Pratt
case
of
filter
face
of
scheduling.
That
means
it's
a
must.
So
let's
take
a
look.
B
B
B
So
as
I
mention
there
may
be
some
user-defined
ApS
back.
There
impacted
the
priority.
For
example,
I
was
to
do
a
little
bit.
There
is
a
preferred
blah
blah
blah
spec
in
past
I
usually
means
that
the
best
kind
of
priority,
and
also
there
is
some
implicit
default
priorities.
For
example,
if
you
create
a
part
and
the
part
belongs
to
some
replicas
set,
we
are
try
to
advance.
B
The
parts
belongs
to
the
same
replica
sell
evenly.
So
that
is
some
implicit
settings.
Do
you
have
to
change
that
by
the
disabling
that
kind
of
priority
in
the
schedule
complete
configuration
of
animals,
but
if
they
don't
have
control
further
for
the
configure
most?
Maybe
you
have
to
return
to
your
crust
admin
to
turn
off
some
of
the
settings
that
is
implicit
default
priorities,
so
the
whole
step
is
the
logic
works
like
this.
B
So
there's
another
part
that
can
be
put
here
so
again
it
doesn't
want
to
be
put
into
worker
3,
so
it's
required.
So
these
predicates
and
also
I
put
priority
here.
It's
preferred
during
stuttering,
ignored
during
execution,
so
because
it's
a
scoring,
so
they
are
you
sure
it
comes
with
the
weight,
if
not
here.
If
yes
really
is
specified
in
the
Canadiens
configurations,
animals
and
that
specify
our
soft,
the
requirement
to
be
so
I
prefer
my
part
to
be
being
welcomed,
or
instead,
workers
to.
B
B
And
then
maybe
you
are
wondering
what
if
doesn't
know
candidate?
That
means,
if
all
the
predicates
fair,
it's
very
possible
right.
If
your
cursor
is
explaining
a
hybrid
cloud
and
almost
all
the
existing
workload
has
you
set
up
all
the
resources?
So
how
come
you
resolve
this?
So
in
kinetics,
110
or
1/9,
we
introduce
the
concept
of
priority
class
or
call
week
I,
proudly
or
preemption.
They
are
all
the
same
thing.
Oh,
let's
take
a
look
so
this
area
right
now
we
have
a
three
node
right.
B
B
Okay,
so
this
part
is
soon
and
on
work
of
three.
So
next
to
continuum
demo
I
need
to
do
some
tricks,
which
is
the
tan,
the
other
two
parts,
so
that
because
I
will
show
you
a
demo
there,
the
preamp
the
park
am
singing
and
I
will
just
ensure
that
premium
to
all
may
learn
something
same
note
as
the
Lowepro,
so
that
the
preemption
can
happens.
B
Have
a
land
into
a
parameter
which
also
requires
Reece
appears
because
in
total
we
only
have
both
abuse
on
one
note
so
and
then-
and
those
are
this
part
has
higher
priority,
the
product
cost
name
can
be
should
be
matched
for
an
existing
priority
classes.
So
buddy
father,
there's
two
system
level
broadcast
the
p1
prior
to
cross
it
created
by
me
manually,
and
it
has
a
very
wise
larger
than
zero
right.
B
You
can
see
right
now,
it's
a
interim
state
that
is,
that
is
terminate,
which
is
called
the
pre-empting
the
old
part
which
missed
go
up
to
show
you
the
idea
here.
Tp
is
AP
X
the
deans.
If
he
acts
is
being
terminated
and
why
is
a
new
party
speak
really
because
consider
manager
notify
now
it's
God
multiply,
that
is
old
records
being
delayed
here.
So
it's
trying
to
spin
up
a
new
one,
but
no
virus.
B
This
one
doesn't
have
any
number
associated
to
run,
so
it
will
be
parents
out
there,
and
here
you
can
see,
there's
a
interim
status
for
the
pre-emptive,
which
is
the
nominating
note
of
kind
of
Orchestra.
It
is
because
we
don't
want.
We
don't
do
this
preemption
once
training
cycle,
once
we
found
that
we
have
a
low
priority,
part
to
PN
preempt.
We
just
set
the
spec
nominating
now
to
the
to
the
preemptory.
Then
we
continue
to
the
next
cycle.
So
that
is
a
bit
of
difference,
but
this
state
is
very
transient.
B
B
How
preemption
books
I
mentioned
a
is
power.
Whisperer
across
prayer
cards
should
be
created
manually
and
then
the
logic
goes
like
this.
The
preemption
happens
on
every
node,
because
there's
no
candidate
know
that
at
all-
and
there
may
be
some
knows-
a
good
candidate
for
preemption.
Some
existing
path
and
some
well
well
come
out
to
the
preemption
candidate,
and
also
we
have
some
internal
algorithm
to
compare.
The
preemption
candidate
should
find
a
best
personnel,
for
example,
to
create
the
lowest
priority
passed
as
much
as
possible.
Instead,
the
preempted
the
higher
priority.
B
Okay,
let's
go
to
the
next
slide,
so
here's
just
a
summary
of
the
key
features
of
schedule
like
if
you're,
what
close
spec
ask
some
specific
resources.
No
matter
is
before
the
resources
like
CPU
memory,
storage
or
some
extended
resources
like
the
media.com,
slash,
GPU
or
whatever
extended
was
identifying
the
resource.
Fair
request
is
coming
to
the
schedule
and
also,
if
your
part,
has
some
toleration
and
toleration
the
match
up
with
the
chance,
also
down
in
the
schedule,
as
well
as
the
notes
lecture
in
the
definitive
part
of
it
in
Poland
1.8.
B
B
No,
so
some
storage
related
features
are
also
related
over
here,
for
example,
if
they
have
a
storage
class
which
has
a
is
of
the
type
wait
for
the
first
consumer,
so
that,
if
you
create
a
PVC,
then
the
PV
won't
be
created
when
the
first
consumer
comes
into
so
that
face.
It
also
happens
in
the
schedule,
because
you
know
ones
that
delay
the
PV
to
PVC
binding
to
the
scattering
phase.
B
That
is
basically
the
key
features
of
schedule
and
what's
new,
the
first
new
features
we
engineers
in
is
the
pod
party
spread,
so
Top
Pot
party
spread
is
trying
to
resolve
some
limitation
of
existing
features
like
part
affinity
and
part
I'm.
Finally,
today
so
pod
affinities
like
I,
want
my
past
to
be
scheduled
to
a
topology
domain
as
much
as
possible.
So
basically,
you
can't
control
how
many
I
want
to
be
placing
just
a
neurologist
today
and
how
much
I
don't
want
and
part
anti
affinities
are
not
a
legal
case.
B
B
B
The
part
about
the
sprayer
at
a
new
API
spec
called
topologies
where
constraints
and
has
some
idea
fields
here.
The
when
I'm
Scott
was
when
unsatisfiable
there's
are
two
values
enduring
else
about
it.
The
first
one
is
do
not
schedule
it's
a
hard
work
on
that
means
if
it
doesn't
satisfy
these
two
common,
which
means,
if
one,
if
the
balance
of
degree
is
larger
than
one
max
will
describe
imbalance
degree.
Then
I
don't
want
to
schedule
now.
B
Of
course,
you
can
make
it
a
priority,
which
just
means
a
software
requirement,
which
you
can
use
the
keyword
schedule
anyway.
So
you
can
basically
combining
use
the
to
you
because
as
much
as
constraint
as
much
as
you
want
so
right
now,
I
just
demo,
just
there's
one
constraint
so
basically
I
want
to
strain
webcast.
I
want
them
to
know
that,
in
balance
to
Moses
trick
balance
mode
with
only
one
topology
can
larger
than
the
other
by
one.
B
It's
pending
why
it's
pending
you're
sure
they
will
use
genetic
potential
describe
to
see.
What's
the
error
message,
if
you
schedule
it,
that's,
oh,
you
says
three
nails
didn't
match
power
to
power,
to
spread,
constrain
and
they're
actually
for
nose
here.
Why
are
there
are
four
nails
here,
because
the
masternode
we
also
count
as
no
so
the
first
part
where
I
definitely
tried
placed
onto
the
masternode,
but
methanol,
as
you
know,
must
know,
has
special
can't
to
avoid
the
incoming
part
should
be
placed
right.
B
Okay,
you
can
see
it
now.
It
looks
perfectly.
There's
nine
paths
and
one
there
sweet,
sukkah,
two
or
three
parts
were
exchanged
a
pass.
Okay,
here's
the
basic
of
the
part,
apologies
prayer,
works
and
there's
a
plot,
explain
this
feature
in
details
and
there's
an
advance
usage
on
this
feature.
So
if
you're
interested
just
check
it
out,
another
feature
we
came
up
recently
is
called
multi
profile.
Scheduler
I
will
demo
disappeared
later
so
because
the
motivations
because
of
the
limitation
of
multi
schedule,
so
not
we
use
the
super
multi
scheduler
modest
schedule.
B
It
means,
in
addition
to
the
default
schedule,
you
can
have
as
many
as
many
scheduled
processes
running
outside
or
whatever
running
somewhere
and
so
thick
works
together
by
identifying
which
comes
apart.
Each
schedule
will
cover,
for
example,
there's
a
pulse
backup
called
spec
dot
schedule
name.
So
if
the
don't
specify
that
the
devoxx
schedule
will
take
over,
but
if
you
explicitly
specified
aspect,
the
corresponding
schedule
will
take
over
this
kind
of
path,
so
you
can
use
II
think
that
there
will
be
some
raising
issues.
B
They'll
run
not
running
in
memory
to
sharing
something
they're,
doing
different
boundaries
and
also
right
now.
The
suggestion
we
gave
to
you,
so
you
feel
wrong.
Marcus
schedule
is
that
this
written
case
can
be
mitigated
by
dividing
your
cluster
into
several
parts,
and
each
class
is
governed
by
each
schedule.
So
this
can
mitigate
the
issue
and
not
resolve
the
issue
totally.
For
example,
some
advanced
scheduling
features
like
pod
affinity,
parties.
Prayer.
This
feature
doesn't
just
use
a
single
node
to
two
decisions.
B
B
So
the
solution
we
we
propose
is
using
multi
profile
mechanisms
so
that
we
still
recommend
you
use
the
the
scheduling
boundary
and
just
one
scattering
boundary,
and
within
that
we
give
you
some
API
spec
to
define
the
profiles.
You
want
so
each
provides
alike,
a
scheduling
flavor,
for
example.
This
flavor
wants
to
be
the
path
to
be
mop,
impact
that
flavor
to
want
to
be
whatever
takes
more
priority
on
some
mother
features
whatever,
so
that
the
banner
is
compiled
there
and
the
profiles
are
defined
there.
B
They
are
just
walking
into
one
boundary,
so
there's
no
resi
issue
and
also
the
consumers
can
specify
their
workers
to
using
what
kind
of
schedule
profile
they
want
to
use.
So
there
works
like
the
same.
So
basically,
we
want
to
resolve
this
issue
that
this
this
multi
profile
scheduler,
can
now
be
done
without
a
major
code
refactoring
inside
schedule.
So
in
the
last
three
or
four
releases,
the
most
important
thing
is
co2
refraction
and
we
reform
each
phase
of
the
schedule.
So
in
the
before
we
just
have
two
phases:
wines
theater.
Why
is
score?
B
B
I
sir,
I
thinkt
girl,
routine
sort
of
so
that
is
why
why
the
profile
can
only
happens
when
we
race
that
the
scaling
framework
so
friendly
is
just
we
redefine
some
internal
logic
and
expose
the
extension
point
that
the
consumer
can
consume
and
write
their
own
logic,
and
then
this
kind
of
logic
can
be
put
on
top
of
our
scheduler
to
combine
so
compiled
to
a
new
custom
and
scheduling
binary,
which
has
some
capacity
that
you
want
nor
the
default
capacity.
But
one
hundred
one
hundred
percent
of
the
default
schedule
capacity
is
capped.
B
B
B
Command-Line
arguments
now
recently
not
only
is
scheduler
also
in
qpid
mq
control
manager.
We
just
use
llamo
to
describe
this.
Can
this
command
line
arguments
we
use
like
scheduled
user
scheduler
component
config,
which
is
llamo?
We
can
you
can
pass
in
using
the
perfect
and
all
most
of
the
command-line
arguments
are
also
defined
there.
So
right
now
the
two
arguments
coexist,
but
maybe
in
the
future
we
will
totally
remove
the
command-line
arguments
but
keep
scheduled
component
of
it.
So,
let's
move
this
out
and
I
have
a
new
configuration
file.
B
This
demo,
this
yellow,
is
using
the
whip
1
alpha
2
component
config.
You
can
see
the
cone
here
schedule
cube
schedule
configurations.
So
basically,
you
just
defined
some
common
command-line
arguments
here,
like
the
cube
config
leader
election
and
additionally,
we
define
the
profiles
here.
So
you
can
see
here,
I
kept
the
fall
schedule
and
also
I
can
give
some
some
other
flavors
like
image.
First
image
first
is
to
give
a
very
high
weight
on
the
image
locality,
image
localities,
prehab,
priority
function
that
say:
okay.
B
B
So,
by
default
the
note
resource
in
list
I
located
the
project
is
defined,
which
means
I,
try
to
spread
the
overhead
on
HTML
so
further
the
pass
as
even
as
possible,
but
it's
not
a
case
for
like
cluster
autoscale,
because
this
kind
of
scenario
wants
to
use
the
resources
as
much
as
as
possible
and,
if
not
possible,
they
will
spin
up.
You
know
in
this
case
they
want
a
part
of
B
to
be
more
impact,
so
I
use
the
opposite
of
priority,
called
notice
us
most
educated
and
give
it
a
higher
score.
B
And
of
course,
in
some
extreme
particular
case
you
just
want
to.
Maybe
don't
any
don't
want
any
score
plug
into
participant,
because
you
may
think
the
filter
is
always
a
really
good
enough.
So
you
just
want
to
depart
when
I
learn
down
any
candidates
that
may
be
possible.
I,
don't
know,
I
just
used
a
demo
here,
okay,
so
you
so
that
I'm
going
to.
B
B
B
Look
like
this
so,
for
example,
app
impact
I'm
going
to
demo
that
I
just
want
to
deploy
ten
path
and
the
pen
pots
will
land
on
the
same
there.
So
here
is
that
you
can
use
the
you
should
use
the
schedule
name
to
want
the
profile
name
so
define
and
of
course,
to
bundle
here
is
that
you
launched
one
image
and
there
damn
you
launched
a
second.
B
B
Anyway,
we
can
give
a
little
time,
so
one
thing
I
want
to
mention
bit
here
is
that
if
you're
still
using
the
lower
version
of
Cuban
alias
the
multi
provides
no
support,
it's
only
story
on
180.
If
you
still
use
the
old
version,
maybe
you
have
still
reduced.
We
want
half
of
an
ocean
and
in
that
case,
deprecated
field
called
algorithm
sauce.
That
policy
is
used
so
that
policy
general
you
can
define.
B
Of
course
you
cannot
define
multiple
profiles
and
also
the
predicates.
You
have
to
use
the
keyword
prod
case
and
priority
to
to
define
or
modify
some
the
entry
default
settings.
The
thing
you
must
be
open
and
the
women
alpha
2
is
just
what
I
tried
demo
and
yeah
multiple
file.
Just
look
like
this.
So
the
aspect
of
scheduling
no
name
right
now,
adapter
to
two
phases:
sorry
to
to
kind
of
mechanisms.
B
You
can
use
the
multiple
schedule,
just
like
what
you
used
before
and
also
you
can
use
the
multi
profile
in
one
schedule,
but
in
the
communities
long
term,
road
map.
We
recommend
that
you
use
just
one
schedule
binary
to
avoid
any
recent
issue
and
define
the
profile
and
inside
schedule
all
right.
Here's
the
last
section
how
to
contribute
to
scheduling
some
high
level.
B
Definition
of
this
is
different
here
in
Chartered
MD
and
our
we
have
several
project
that
the
prowl
priority
is
the
default
scatter,
which
is
shipped
with
occupied
areas
to
release,
and
also
we
have
some
sub
project.
Like
a
schedule.
The
structure
is
initialed
by
rehab,
so
it's
trying
to
resolve
the
problem
that,
along
with
the
running
of
your
cluster,
some
balanced
situation
may
be
broken
resemble
at
this
moment
at
the
point
of
scheduling.
B
Maybe
your
workers
Emily
balance,
but
maybe
some
nails
have
been
shut
down
unexpectedly,
so
that
passed
to
other
nails
and
maybe
I
sometimes
does
not
get
recovered
back.
But
at
this
point
schedule
doesn't
do
the
rescheduling
Floyd
for
you
and
we
don't
have
a
priority
on
supporting
that
key
for
scheduler.
So
you
have
to
do
some
external
component
like
this
schedule.
If
you
really
care
about
parenting
or
some
specific
strategy,
you
want
to
comply
with.
So
you
use
the
schedule
and
the
second
one
schedule
proggy.
B
A
B
Ok,
yes,
for
example,
Alibaba
contributed
the
co
scheduling
plugin,
which
he
tried
to
scheduling
crew
plot
path
together.
This
is
a
typical
requirement
from
The,
Bachelor
clothes,
ai
and
missionary
field,
and
also
they
raised
another
proposal
to
consider
elastic
resource
quarter
so
because,
right
now,
Gustav
SCADA
is
defining
the
mission
hook.
So
we
mention
hook
means
if
a
path
specify
I
won't
use.
This
vu.
I
won't
use
this
memory,
but
it
fails.
You
still
keep
that
you
keep
the
request,
no
other
incoming
Park
and
zone.
B
A
Lead
in
the
chat
is
mentioning.
The
port
works
also
comes
with
its
own
scheduler
called
stork.
Oh
that's
it.
There
are
a
few
other
examples
of
other
schedules.
There,
yeah
I
hadn't
heard
that
a
lot
of
the
content
that
you've
covered
I
didn't
really
have
a
good
background
in
well.
Thank
you
very
much.
So
far,
really
good
and.
C
B
B
So
I
prefer
Caroline,
I,
think
and
maybe
it's
a
superset
of
mini
cubes.
So
basically
a
mini
cube,
just
a
lunch
at
a
wine
node
cluster
and
the
inside.
There
is
a
virtual
machine.
If
I
understand
it
correctly,
so
there's
one
machine
and
you
can
operate
with
the
virtual
machine
that
has
one
with
one
out
there.
So
of
course
you
have
some
counts
there.
If
you
want
to
some
scheduling
development
or
just
want
to
try
some
scenario
that
must
have
multiple.
B
You
cannot,
for
example,
if
you
remember
that
if
you
want
to
debug
the
priority,
if
you
have
only
have
one
no
of
course
they're
proud,
it
won't
happen
because
there's
only
one
choice:
there's
no
need
to
do
the
scoring
ranking
and
I
do
like
can't,
because
it
can
be
useful
in
the
open
source
available
in
the
entry
department.
So
once
you're
changed
on
some
Sasuke,
you
can
build
your
current
code
base
into
current
no
image
and
launch
out
there
and
also
some
other
features.
B
B
That
is
this
page
mention
before
you
just
choose
schedule.
The
racing
issue
can
now
be
avoid,
even
if
you
manually
separate
the
cluster
into
two
parts
in
the
schedule
just
govern
that
part
yeah.
The
two
reasons
I
just
mentioned
here,
because
if
you
don't
divide
the
cluster,
it's
possible
they're,
the
new
try
to
schedule
the
to
schedule,
try
to
schedule
the
same
part
at
the
same
moment
to
compare
for
their
resource
because
there's
no
distributed
log
deciding
which
scheduler
can
can
win.
So
the
final
arbitrator
is
to
just
queue
blade.
B
A
A
B
B
A
D
A
A
A
Now,
I'm
making
the
connections
between
all
of
these
projects
and
that's
what
this
is
all
about-
is
really
trying
to
figure
out
how
all
the
pieces
and
that
wonderful,
CN
CF
landscape
fit
together.
So
if
you
could
put
your
last
slide
up
again,
the
one
that
had
how
to
get
a
hold
of
you,
that
would
be
great
and
I
know
the
scheduler
sig
is
meeting
on
a
regular
cadence.
So
if
folks
are
interested
in
this
or
interested
in
just
reaching
everybody,
the
slack
channel
is
probably
the
quickest
way.