►
From YouTube: Kubernetes SIG Scheduling Weekly Meeting for 20210715
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,
all
right
hi.
Everyone
today
is
july,
15
2031,
and
welcome
to
this
week's
week
scheduling
meeting
and
this
building
is
being
recorded.
I
will
be
uploaded
to
youtube
so
looking
at
today's
agenda
first,
why
is
the
gsoc
proposal,
which
was
tried
to
create
a
web-based
simulator
for
the
kubernetes
scheduler
to
sort
of,
I
think
visualize
every
decision
every
scoring
filtering
result
of
how
schedule
it
does
under
the
hood.
A
So,
let's
give
me
the
control
to
ken
say,
I'm
not
sure
if
I
pronounce
it
you're
correct,
correct
me
wrong
yeah
I
can
say:
could
you
see
if
you
can
share?
If
not,
I
will
make
you
the
co-host.
B
B
Yes,
you
can
okay,
sure,
okay,
okay!
So,
let's
start
please.
B
Okay,
can
you
see
my
screen.
B
So
hello,
my
name
is,
can
say
it's
sampo,
siho
and
yeah.
Most
important
part
is
I'm
not
good
at
english
yeah.
So
please
speak
me
with
a
very,
very
slow
english
and
I
working
on
this
issue
as
the
google
summer
of
code
project
and
I'm
going
to
give
you
a
brief
overview
and
after
that
I
show
you
some
demo
demonstration.
B
First,
it
can
create
this
resource,
and
so
we
can
accumulate
scheduling
that
depends
on
right
people
or
pbc,
and
we
can
see
scheduling
result.
So
we
can
see
how
the
scheduling
is
done
yeah.
B
After
push
start
simulator
button,
it
create
name
space
for
this
user
and
we
can
create
some
node.
B
B
We
can
see
this
port
is
scheduled
on
this
node
and
then
we
can
see
how
it
how
it's
scheduling
we
can
see
the
featuring
dragon
result
and
the
score
and
normalized
score
yes,
and
so,
for
example,
we
can
create
a
drag
sc.
B
Like
this,
and
we
can
create
a
new
bot
with
the
volumes.
B
B
These
volumes
add
this:
this
volume
is
bad
and
this
spot
is
scheduled
on
the
yeah,
so
this
is
all
of
our
generator,
our
simulators
future.
Thank
you.
A
Yeah,
thank
you
for
the
demo,
it's
very
impressive
and
I
suppose
this
project
is
still
ongoing
right.
Does
this
mean.
A
The
product,
okay,
so
still
ongoing,
and
when
you
create
a
new
part
which
kind
of
spec
do
you
support?
For
example,
you
you
specify
the
cpu
and
memory
request.
That
is
the
basic
one.
So
do
you
also
support
some
like
advanced
scheduling,
features
like
productivity,
no
dfinity
or
so.
A
Yeah,
do
you
support
any
other
other
advancement
feature
like
part
affinity
or
topology
spread
constraints,
so
is
that
on
your
working
left
or
it's
already
implemented.
A
Okay,
cool
so
yeah.
If
you
can
demo
advanced
the
example
that
would
be
great
for
life.
B
E
Perhaps
perhaps
you
can
give
an
overview
of
how
is
this
working
under
the
curtains
like
in
particular
interested
to
know?
E
Well,
I
assume
you
will
actually
have
a
real
api
server
running
and
then
the
scheduler
yeah?
How
is
it
running.
E
B
So
when
the
port,
when
the
port
is
created,
this
background
requests
to
api
server
to
create
port
and
then
the
scheduler
start
to
schedule
the
port
with
some.
E
B
Background
is
one
single
binary
and
I
create
api
server
and
scheduler
within
single
binary
and
run
only
one
are
one
background.
Binary.
A
B
No,
we,
I
don't
run,
no,
I
don't
create.
B
Sorry,
I
create
only
api
server
and
scheduler
and
people
controller,
so
no
cluster
is
running
and
we
just
see
a
scheduling
result
from
api
server
and
not
real
cluster.
A
D
A
A
So
the
question
is:
I
mentioned
that
if
the
back-end
architecture
is
like
how
we
wrongly
going
to
test
you'll
say
yes
right,
but
you
know
in
integration
test
we
need
to
because
api
server
is
pretty
difficult
to
mock
to
simulate.
So
in
the
integration
test
we
create,
we
start
the
lcd
battery
in
the
back
end
and
the
api
server,
it's
sort
of
like
a
real
api
server.
So
how
do
you
do
that
in
your
project?
What
about
the
dependency
on
api
server
and
the
electricity?
C
Here
just
jump
in
yeah
he
does.
He
is
starting
a
xcd
instance
that
runs.
C
Very
similar
to
how
the
integration
work,
integration
tests
work,
except
that,
instead
of
using
instead
of
measuring
performance,
we
just
use
the
output
of
the
scheduler
to
display
that
in
the
web-based
interface,
there
are
also
some
minor
changes.
For
example,
if
you
schedule
a
part
to
a
node,
there
are
annotations
that
are
added
to
the
part
to
tell
you
what
the
score
for
each
plugin
was
what
the
filter
result
was
etcetera.
C
This
is
obviously
not
a
part
of
the
integration
test,
but
the
but
the
simulator
does
that,
so
that
it
retains
information
about
the
score
and
filter
results
and
so
on,
which
is
what
you
see
in
that
table
when
you
click
on
a
pod.
All.
C
E
So
the
scores
and
filters:
how
so
did
you
modify
the
scalar
code
to
do
this
or
how
no.
E
B
Yeah,
I
so
on
on
my
background.
It
creates
a
plugin
to
record
the
scheduling,
result
for
its
default
fragrance,
and
then
they
record
on
this
annotation
tag
and
front
to
end
show
the
result
from
this
information.
B
B
So
I
don't
touch
scheduler
code,
but
I
just
use
my
plugin
to
run
no
when.
E
Oh,
that's
cool
all
right,
yeah.
A
I
think
yeah,
I
think
it's
valuable
to
get
this
designed
on
your
prediction,
tips
and
tricks
in
a
document
so
yeah
the
readers
and
the
other
guys
who
are
working
in
the
same
similar
idea.
Can
you
can
leverage
or
that
can
benefit
from
what
you
have
done?
A
B
Okay,
so
where
should
I
write
document,
I.
A
C
Okay,
I
just
should
just
chat
a
little
bit
more,
so
kenzie
is
around
the
halfway
point
in
his
project,
there's
around
four
or
five
more
weeks
left
in
his
gsoc
industry
project
and
even
though
he's
like
well
ahead
of
schedule.
C
So
we
originally
didn't
have
the
plan
to
have
filter
and
score
result,
etc,
a
part
of
the
first
half
of
the
project,
but
he's
already
ahead,
so,
which
is
why
we
already
have
all
this
already
done
in
the
project,
but
but
yeah
definitely
documenting
how
it
works
and
how
to
run.
This
is
probably
very
important,
and
I
think
kensei
will
probably
do
that
sometime
before
the
merge
into
the
the
main
repo,
or
this
is
not
structured
as
a
command
line
binary
by
the
way.
C
So
it's
all
under
the
cmd
directory
in
his
own
fort
right
now,
but
eventually
it
will
probably
be
a
part
of
the
main
repo.
C
Yeah
definitely
yeah
and
it
doesn't
have
to
be
the
main
report.
Yeah.
E
Yeah
one
night
pick
here:
I
don't
think
we
need
the
normalize
and
score.
We
probably
just
need
the
final,
which
is
normalized.
I
think
that's
good
enough,
but
I
was
going
to
say
this
is
great
great
progress
and
I'm
really
impressed
with
the
with
the
demo.
Thank
you
yeah.
E
F
A
A
B
No,
no,
it
don't
work
well,
but
previously
I.
C
It
works,
but
I've
seen
it
before
there's
another
thing.
Yes,.
A
B
B
B
Something
goes
wrong,
but
when,
when
put
a
period
to
schedule
route,
it
will
show
on
new
column
schedule
report
right
like
on
schedule
and
we
can
see,
result
yeah.
We
can
see
why
job
put
cannot
be
scheduled
from
this
builder
crown.
E
I
have
one
final
ask:
it
would
be.
A
E
B
And
things
like
that,
so
yeah.
E
If
you,
if
you
have
enough
time
left,
I
think
that
would
be
cool
yeah.
C
Yes,
thank
you.
He
has
that
plan,
so
we
he
has
a
plan
to
make
it
arbitrarily
configurable.
We
can
change
the
scheduler
config
and
profiles
and
so
on,
but
that's
a
part
of
the
second
phase
of
this
project,
which
is
just
starting
sometime
this
week,
so
yeah
excellent.
A
B
A
A
To
show
you
mark
or
or
robbie
some
meeting,
but
there's
some,
I
think,
there's
some
changes
to
you.
E
Yeah
we
each
now
v,
one
beta
two
is
the
default
and
in
one
b
one
beta
beta
two.
We
don't
have
this
plugging
anymore.
A
E
Sounds
good
yeah
I'll
bring
to
them,
but
yeah
the
features.
Still
there
I
mean
they
can
still
re-enable
it,
but
ultimately
yeah
you.
You
have
one
affinity
and
sorry
no
affinity
and
whatnot
toleration.
So
this
yeah
we
should
get
rid
of
this
feature
for
sure.
A
Yeah
all
right
so
last
week
we
had
the
cook
breeze.
So
only
the
critical
fix
those
sort
of
beers
freaking
fixing
the
flaking
test
can
be
eligible
to
merge.
So
yeah
be
aware
of
that
yeah.
Any
some
trivial
peers
will
be
wait
for
the
cycle,
123.,
so
yeah
and
before
I
end
today's
meeting.
So
is
there
any
item
ideas
you
want
to
raise
up.
We
have
do
two
minutes.