►
From YouTube: Kubernetes SIG Scheduling Meeting - 2019-05-02
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
You
say
remains
on
the
Internet
for
a
very
very
long
time,
but
that,
let's
start
our
meeting,
I
have
a
couple
of
updates
for
you.
Folks
I
have
put
some
items
on
the
agenda
for
this
meeting
in
our
regular
meeting
notes,
one
as
with
regards
to
the
framework,
the
scheduling
framework
that
we
were
working
on.
A
Finally,
that
a
PR
that
I
sent
for
adding
the
interfaces
for
the
framework
or
some
of
the
extension
points
of
the
framework
and
the
context
and
stuff
like
that
is
already
merged.
So
you
can
take
a
look
if
you
want
and
with
that,
we
are
now
at
the
point
that
we
can
start
adding
more
extension
points.
So
I
filed
at
least
a
couple
of
issues
for
adding
extension
points,
in
particular
unreserve
extension
point,
which
is
used
by
plugins.
A
That
register
had
reserved
and
later
stages
of
scheduling
in
the
case
that
apart,
goes
to
reserve
and
needs
to
be
unreserved.
This
plugins
are
called
so
that
if
some
of
the
plugins
in
the
past
save
some
state
with
the
assumption
that
this
part
is
going
to
be
reserved,
now
have
a
chance
to
clean
up.
Knowing
that
the
reservation
is
gonna
fail
and
reservation,
the
reservation
is
not
going
through
so
the
unreserve
plugin.
A
There
is
a
issue
for
it,
and
hopefully
someone
will
I
think
I.
If
somebody
has
already
picked
it
up,
the
unreserve
one
I
also
filed
an
issue
to
add
permit
plugin
permit
plugins
are
super
important
for
us.
These
are
the
plugins
that
can
enable
advanced
the
scheduling
features
such
as
gang
scheduling,
or
course
scheduling,
as
we
call
them.
So,
if
you're
interested-
and
if
you
have
the
cycles,
please
take
a
look
and
take
that
issue.
A
I
also
find
another
one
for
adding
configuration
for
the
framework
on
the
our
design
document
for
the
scheduling
framework
were
basically
are
kept
for
the
scheduling
framework.
It
goes
to
some
details
about
what
this
configuration
should
look
like
and
provide
some
detailed
information
about
the
configuration.
Of
course.
When
we
want
to
implement
there
could
be
changes.
We
don't
always
necessarily
have
to
follow
the
exact
things
that
are
mentioned
in
the
cap.
We
can,
at
that
point
discuss
whether
those
changes
are
acceptable
or
not.
So,
if
you're
interested
in
that,
you
can
also
take
that
issue.
C
A
Basically,
our
main
goal
is
to
is
to
have
some
of
these
possible
in
115,
so
we
essentially
have
one
month
before
our
115
code
freeze
to
implement
these
I
implement
and
measured
them.
You
know.
Sometimes
the
implementation
takes
a
few
days
and
merging
reviewing
process
takes
a
lot
more
time.
So
if
people
are
interested,
please
try
to
take
them
and
start
working
on
them
as
soon
as
possible.
A
C
A
Luckily,
you're
not
a
new
contributor,
an
idea
yeah,
that's
actually
a
pretty
important
extension
point.
I,
don't
know
if
you
have
enough
time
for
it,
even
that
you
were
also
working
even
participating,
which
is
also
a
relatively
big
project
as
well.
Yeah.
A
Yeah
the
kept
for
supporting
less-than
and
greater-than
operators
for
injured
part
affinity
is
also
merged,
meaning
that
we
can
track
that
in
115
and
hopefully
we
will
have
key
feature
mesh
in
115.
I
saw
that
Leon
I
believe
send
a
send
a
key
or
floor
for
that
and
yeah
Leon
has
sent
a
PR
for
that.
I
have
linked
that
PR
in
our
meeting
notes.
It's
still
working
progress,
but
we're
hoping
that
we
can
finish
it
in
115.
A
A
Yeah
I
want
to
do
it's
important
that
any
PR
that
we
submit
must
pass
all
the
CIC
be
and
and
must
be,
of
course,
buildable.
We
need
to
be
able
to
build,
but
another
thing
that
I
wanted
to
say
is
that
probably
API
machinery
wants
all
the
eight
insider,
the
the
project
to
be
to
be
done
in
a
single
PR,
but
they
don't
necessarily
want
all
the
no
logics
of
the
scheduler
to
be
built
in
the
same.
C
D
B
A
D
D
A
Okay,
these
are
all
the
updates
from
my
side,
if
you
have
any
questions
or
comments,
it's
mounted
good,
a
good
time
to
ask
them.
Oh
and
I
would
like
to
quickly
also
introduce
Abdul
I.
If
you
don't
mind,
I
would
like
to
introduce
him.
Abdullah
is
one
of
the
Googlers
who
is
located
in
Waterloo
and
is
he
has
recently
joined
us
to
work
on
the
scheduler.
I
am
very
happy
that
is
his
in
the
scheduling
now
and
we
will
hopefully
see
him
more
often
in
in
our
sec
meetings.
A
A
D
D
D
B
A
A
A
C
C
So
you
can
specify
this
as
hard
requirement,
which
is
internally
called
predicate
and
also
you
can
specify
it
as
priority
means
we
prefer
to
do
these
options.
Okay,
so
today,
I
will
show
basic
example
of
predicates,
which
is
a
table
requirement.
Example,
we
have
Crestor
five
nails
because
I
run
in
your
local
host,
so
I
didn't
change
the
name
of
127.0.0.1.
C
Level,
D,
no
ands
on
to
each
note,
so
the
basic
there
are
three
nose
belongs
to
someone
and
the
nose
three
I
know
before
we.
Don't
you
don't?
Okay,
all
right
now,
there's
no
pass.
Okay,
I'm
sure
looks
like
so.
The
demo
looks
like
this:
we
introduced
new
API
field,
two-part
spec,
which
is
called
topology
sprayer
constraints.
It's
a
sliced
means
you
can
either.
So
it's
a
one
or
multiple
constraint
and,
for
example
this
example
just
spicy
moring
example.
Sorry,
one
constraint
it
means
I
want
to
spread
the
pass
with
the
most
tolerance.
C
C
Sorry,
when
unsatisfiable
means
whether
it's
a
predicate
or
priority
do
not
schedule
means
if
it
doesn't
satisfy
this
constraint.
We
just
let
it
be
pending.
We
don't
try
to
schedule
it.
Okay
and
the
labels
lecture
is
the
conditions
we
specify
to
ask
by
incoming
power
to
match
to
which
kind
of
party
in
the
cluster
so
right
now
I'm
going
to
just
create
Y
and
C.
So
basically,
there
are
a
lot
of
available
sauce
right
in
each
two
zones.
So
basically
it
can
lay
down
any
no.
C
Ok,
so
basically
the
blends
down
note
for
histone
zone
2
right
and
you
can
see
here-
five
notes
evaluated
and
the
finals
are
feasible.
So
it's
correct
and
let
me
scare
it
to
two
so
in
this
case,
because
there's
two
audio
already
one
party
in
zone
2.
So
basically
this
incoming
part.
We
can
only
be
scheduled
to
zoom
one.
C
E
C
C
A
C
C
And
see
whether
we
have
bugs
here
yeah
nice,
each
part
gas
landed
on
each
node
and
the
basically
they
don't
virus.
Sorry,
they
don't
violate
it,
either
constraint
right.
So
let
me
show
you
another
interesting
thing
so
because
right
now
we
have
one
part
for
each
node
right
and
what,
if
I
scale,
is
to
ten
what
things
you
will
think
it
will
happen,
so
you
so
buddy
what
symptom
would
you
expect
to
happen
so.
B
A
B
A
C
Exactly
so,
let
me
give
you
so
what
does
Bobby
mention?
Is
this
kind
of
status
we
want
to
them?
The
both
satisfy
the
zone
constraint
and
the
nil
constraint
part
is
infeasible,
because
what
the
schedule
cost
comes
to
nine,
the
tenth
part
cannot
be
scheduled
on
to
any
node,
because
he
the
violates
box,
for
example,
right
now,
he'll
top.
We
have
this
kind
of
situation
here,
so
the
10
spot
can
ask
to
be
landed
on
or
any
node
if
it
lands
on
no
two.
It
violates
this.
A
A
So
that's
this
is
fine,
and
now
this
is
fine,
because
users
has
a
specified
exactly
this
endure
in
their
constraints
and
have
also
specified
that
they
don't
want
to
be
scheduled
if
any
of
those
constraints
is
violated.
So
this
is
working
as
concerned
that
yeah,
if
if
they
want
the
part
to
be
scheduled
in
this
case,
they
should
probably
go
with
the
soft
affinity
or
with.