►
From YouTube: Kubernetes SIG Scheduling Meeting 2018-05-24
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
and
welcome
to
scheduling
meeting
today.
I
have
a
couple
of
updates
more
than
a
couple
three
updates,
maybe
and
I.
Then
let
you
guys
ask
questions
or
bring
other
points,
and
if
we
don't
have
much
more,
we
can
cut
the
meeting
short
so
about
the
updates.
One
is
that
we
have
decided
not
to
move
yet
cash
for
equivalence
cash
to
beta
in
111.
There
are
several
reasons
for
this.
One
is
that
we
are
not
so
confident
about
the
false
coverage
in
in
cash.
A
The
second
reason
is
that
we
felt
it's
not
a
good
idea
to
move
to
major
major
features
of
the
scheduler
into
beta
in
a
singular
release.
The
other
one
that
we
are
trying
to
move
to
beta
is
priority
and
preemption,
and
it's
not
a
good
idea,
probably
to
move
from
both
a
cash
on
fire
and
preemption
at
the
same
time
to
beta,
and
there
is
a
third
reason.
The
third
reason
is
that
we
are
not
so
confident
about
the
performance
improvements
of
because
we
are
still
working
on
it.
A
A
This
is
about
scheduling
framework.
There
is
a
a
third
item
that
I
would
like
to
talk
about.
It's
actually
of
interest
of
Harry
as
well.
Harry
had
filed
and
an
issue
to
move
one
of
our
priority
functions
to
food
as
a
default
priority
function
and
that's
image
locality.
So
basically,
this
priority
function
tries
to
put
a
pod
on
a
node.
That
already
has
the
images
for
that
pot
or
has
the
largest
number
of
images
for
a
pot.
This
is
a
very
important,
very
important
priority
function.
A
I
would
really
like
to
see
this
to
to
move
to
the
default
priority
functions.
This
is
going
to
improve
performance
in
terms
of
pod
startup
time
and
also
it
could
could
decrease
in
your
cluster
traffic
depending
on
where
the
image
is
fit.
It
can
actually
reduce
the
traffic
from
the
outside
to
the
Questor
as
well.
So
we
have
seen
that
this
improves
performance
in
in
Google
clusters,
internal
Google
cluster.
This
is
one
of
the
one
of
the
optimizations
that
worked
out
all
the
times,
so
we
need
to
definitely
move
this
to
default.
A
C
A
Yeah
sorry
I
have
been
a
little
busier
than
usual
recently
with
a
lot
of
other
small
things
going
on,
but
I
will
try
to
review
that
document
soon.
Hopefully,
this
week
I
mean
it.
So
that's
one
thing
and
then
the
other
thing
that
I,
along
with
the
same
thing
I
want
to
mention,
is
I
recently
get
a
lot
of
requests
for
reviews.
This
is
way
beyond
what
I
can
actually
mean
you.
A
A
Yes,
we
have
discussed
young
scheduling
as
well.
I,
don't
think
Klaus
is
here,
so
there
is
not
a
whole
lot
of
progress
in
that
area.
We
would
like.
We
still
are
thinking
about
basically
considering
all
the
requirements,
finding
all
the
requirements
and
considering
all
the
possible
solutions
to
address
them.
There
are
various
things
that
we
will
consider.
You
know
going
with
the
default
scheduler,
creating
a
separate,
scheduler,
separate
controller
and
a
lot
of
these
ideas.
A
Some
people
believe
that
we
may
not
even
need
yank
scheduling,
since
we
about
a
scaler
and
it
can't
possibly
work
without
a
scaler.
So
we
are.
You
have
a
lot
of
different
things
that
you
would
like
to
narrow
down
and
get
to
it.
So
far.
We
think
we
still
need
gang
scheduling.
It's
not
like
autoscaler
is
the
solution
for
everything
you
would
like
to
have
it,
but
whether
we
are
gonna
have
have
it
in
the
scheduler
or
what
kind
of
requirements
we
have
as
an
example
of
requirements.
A
I
can
point
to
whether
we
should
support
heterogeneous
parts
to
be
in
a
gang
or
what
we
should
do
about
the
types
of
objects
that
we
are
going
to
support.
You
know
we
discussed
it
last
week
as
well
and
things
of
that
sort.
You
know
these
we
would
like
to,
of
course,
have
all
the
features
in
in
place,
but
it's
often
difficult
to
decide
which
ones
to
start
with.
We
cannot
have
all
the
features
in
the
first
version.
So
here
is
still
in
our
face
sure.
A
D
A
A
E
A
Jonathan
and
I
Jonathan
is
not
here
today,
but
Jonathan
and
I
were
thinking
about
going
through
some
of
the
issues
that
we
have.
We
have
a
lot
of
issues
in
scheduler.
You
are
not
necessarily
Martha's
Help
Wanted,
but
we
feel
some
of
them
can
be
Martha
Help
Wanted
we
want
to
go
through
some
of
those
and
foil
and
label
them
properly
would
help
on
it.