►
From YouTube: Kubernetes SIG Scheduling Meetings 20170522
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
All
right,
I
started
the
recording
and
I
send
email
to
the
mailing
list
with
the
playlist,
where
all
of
the
weekly
or
bi-weekly
meeting
updates
will
be
the
going
on
YouTube.
So
you
can
check
there
came
this
a
meeting
and
want
to
see
what
happened.
It
seems
that
the
recording
is
working,
and
so
that's
a
good
thing
anyway.
Back
to
you
agenda,
so
Tim
Sinclair,
you
had
one
item:
do
other
people
have
stuff
they
wanted
to
I.
Think.
A
Yeah
we
can
either
talk
about
that
now
or
put
it
on
the
list
of
the
agenda.
Yep
later
put
it
I'll
write
them.
Okay,
anything
else
want
to
talk
about
now:
okay,
Tim
I,
I
hated.
It
I
forgot
what
the
thing
there
was
all
some
of
what
the
thing
was
you
had
mentioned
last
week.
You
wanted
to
talk
about.
Oh
it's,
four
or
five,
seven,
one,
seven
yeah,
so
my
high-level
comments
on
this
is
at
the
node
condition
and
taint
stuff
is
a
complete
mess,
not
the
implementation
of
paints
itself.
That
works
completely.
A
Fine
there's
been
a
huge
amount
of
confusion
around
when
to
use
paints.
How
node
conditions
are
handled
and
scheduling
what
different
cluster
deployment
tools
are
used
are
doing
in
terms
of
setting
up
paints
or
whatever,
and
in
an
attempt
to
we
like,
like
document
what
was
going
on
mark,
sent
out
a
spreadsheet
listing
like
the
current
behavior
of
the
different
node
conditions
and
the
what
we
the
taints
will
do
once
we
finish
using
those
for
the
knows
scheduled
conditions,
but
like
I
just
want
to
preface
this
with.
A
Like
yes,
there's
a
huge
amount
of
confusion,
people
who
are
confused
they're,
not
missing
anything
part
of
the
reason
for
doing
paint
was
actually
to
bring
some
sanity
to
this.
To
stop
using
node
conditions
in
scheduling
so
that
everything
would
be
very
explicit
for
scheduling
and
eviction.
Then
everything
would
be
very
explicit.
A
You
look
at
the
taint,
it
says
what
is
going
on
on
the
node
and
then
it's
obvious
from
the
taint,
whether
it's
doing
eviction,
whether
it's
blocking
scheduling
and
it's
obvious
from
the
pot,
whether
the
pond
should
schedule,
because
it
has
the
coloration
or
it
shouldn't
schedule,
because
it
doesn't
have
a
caller
Asian
or
should
be
affected
or
whatever.
So
like.
That's,
that's
that's
the
high-level
point,
but
Tim
now
I'll.
Let
you
talk
I'm.
Sorry,
I
just
wanted
to
again
for
a
minute
there.
What?
What
was?
What
on
that
issue?
B
So
there's
supposed
to
have
the
problem
in
that,
if
the
network
is
not
ready
and
notice
is,
is
by
definition
than
not
ready
right.
So
if
the
network
is
not
ready,
even
though
you
want
to
be
able
to
schedule
to
that
know,
you
can
currently
that's
the
problem.
So
it's
something
broke.
I,
don't
know
exactly
the
story
behind
this,
but
it
happens
in
the
1:7
cycle
or
something
introduced
in
the
one.
B
Seventh
cycle
broke
the
behavior
of
self-hosted,
four
couvade
again
so
could
Ian
comedian
right
now
today
on
master
does
not
work
and
it
does
not
work
because
you
get
into
this
weird
chicken
egg
problem.
That's
some
reason
is
listed
as
priority
urgent.
You
get
this
weird
chicken,
a
problem
where
you
can't
you
need
to
be
able
to
schedule
things
like
calico
on
to
the
nodes,
so
you
can
stand
up
the
network
that
you
can't
get
there
yet
right.
So
you're
in
this
weird
conundrum,
I
see.
B
B
A
D
A
So
I
mean
I
think
you
know
that
would
be
a
operating
compatible
chain,
so
he
was
suggesting
we
could
enable
a
with
a
flag,
I
think
arm.
It
sounds
to
me
like
something
introduced.
A
Saying
you
said
it's
in
1.7:
it's
not
because
of
a
change
in
Q
of
a
DM
you're,
saying
it's
a
it's
because
of
something
else
that
was
new
in
1.7.
Yes,
that's
weird
because
the
network
unavailable
was
was
in
was
in
1.6
yeah.
A
Yeah
I,
don't
know
yeah
I,
don't
have
an
objection
to
changing
the
hater
if
we
dude
in
a
backward
compatible
way,
but
I
think
we
need
to
understand
better
what
what
exactly
we
want
to
change
the.
D
Docs
mentioned
that
there
was
a
Kaku
Baku,
baby,
em
and
1.6
was
using
our
back
now.
So
any
plugins
that
needed
to
be
changed
or
added
or
anything
needed
also
have
their
are
back
permissions
at
prior
to
scheduling
the
things.
But
I
didn't
note
that
there
was
anything
that
would
make
it
unschedulable
if
I
definitely.
D
Because
I
run
kuba
DM
on
like
an
openstack
cluster
and
I'll,
spend
an
open
side
cluster
than
Farrow
on
the
daemon
said
for
calico
or
flan
or
whatnot,
and
it
seemingly
works
I'm,
not
I'm.
At
the
end,
they
also
on
a
bug
there
that
I
wasn't
sure
what
what
was
wrong
as
far
as
taints
and
everything.
So
you.
B
B
D
A
A
Would
certainly
explain
what
was
going
on
here,
but
I
don't
know.
A
So
anyway,
maybe
we
should
push
this
on
the
stack
and
come
back
with
it.
Come
back,
so
I
come
back,
come
back
to
it
and
on
github,
or
what
do
you
want
to
eat
in
I'm.
B
A
And
I'll
try
to
find
out
I'll
look
into
when
network
unavailable
was
introduced
to
make
sure
it
wasn't
actually
added
in
1.7,
or
you
know,
after
one
points
it
and
yeah
okay,
it
sounds
good.
Okay,
with
the
two
other
topics
we
had
here
was
the
deliverable
for
priority
and
reschedule
or
repose
the
reschedule.
D
C
A
C
B
A
E
Given
the
recent
discussions
with
with
signal
guys,
it
doesn't
look
like
we
can
really
reach
to
a
conclusion
in
time
for
41.7
David
Asheville
has
written
a
document
and
every
document
he
shared
with
me
on
Friday.
What
looks
like
there
are
a
lot
of
things
that
we
need
to
need
to
address
and
I
guess
it's
important
for
us
to
have
that
eviction.
Logic,
sort
of
in
sync,
with
the
preemption
logic
in
the
scheduler.
Given
all
of
this
I
feel
when.
A
E
E
We
don't,
but
what
I
mean
is
that
whatever
we
have
in
mind
for
it
and
how
it
should
work
should
should
also
confirm
it
should
be
in
sync,
with
the
eviction
you
can
lie
on
the
cubelet.
The
problem
is
that,
if
they,
if
they
or
not,
it
could
end
up
having
like
a
ping-pong
effect
and
a
system
cubelet
thinks
that
it
should
pick
some
pod
schedule
things
that
those
pods
should
be
scheduled.
So
schedule
schedules,
cubelet
removes,
and
it
should
not
happen
so
I
think
it's
important,
not
just
a
matter
of
opinion.
E
I
think
it's
important
for
these
two
logics
to
work
in
harmony
with
one
another.
So
I
guess
given
all
of
this,
we
need
to.
We
need
to
discuss
this
further
and
we
need
to
probably
have
a
separate
meaning
on
how
it
based
on
how
we
should
design
the
eviction
logic
at
the
Hulett
and
for
that
I
believe
it's
not
going
to
make
1.7.
E
E
Basically,
we
we
want
to
have
the
admission
at
want
to
add
the
admission
controller
right
at
all
these
these
fields.
We
want
to
add
the
logic
in
the
scheduler
to
work
with
the
priority
when
deciding
about
scheduling
which
part
and
in
the
next
phase,
we
want
to
add
the
preemption
logics
in
the
schedule
as
well,
and
all
of
these
required
it.
If.
E
E
B
A
E
A
I
mean
I
think
we
were
hoping
originally
that
at
least
the
cubelet
eviction
behavior
part
could
be
finished
too.
Then.
It
would
actually,
even
though,
wouldn't
control
preemption
of
0.7.
It
would
at
least
have
some
some
meaning,
but
I
agree
if
it
doesn't
have
any
meaning,
it's
probably
doesn't
make
sense
to
reduce.
E
Yeah
yeah
David
will
call
we
share
more
information
very
soon,
but
there
is
somewhat
of
clash
of
logics
between
costs
and
priority,
because
I
guess
there
were
a
lot
of
assumption
and
made
based
on
the
fact
that
we
only
have
flaws
in
the
system
across
was
sort
of
like
a
guarantee
guarantee
and
was
kind
of
acting
like
a
priority.
In
our
you
know,
current
logic
and
now
changing.
That
seems
to
be
a
little
hard
yeah.
A
A
I
talked
to
her
briefly
that
she
didn't
seem
too
disturbed
by
the
idea
that,
like
guaranteed,
would
no
longer
have
exactly
the
same
meaning.
Her
argument
was
the
guaranteed
never
had
the
meaning
that
you
would
never
be
evicted,
which
was
in
one
way,
which
is
one
one
way
to
live.
One
way
to
look
at
it,
and,
and
so
yeah
I
don't
know,
but
that's.
E
A
A
A
Okay,
any
other
stuff
people
want
to
talk
about.
B
A
A
Well,
I,
guess
that's
your
answer.
No
one
else
seems
to
wanted
anything.
Anything
else
want
to
talk
about
alright.
Well,
I
guess
we
had
a
short
meeting
today
and
yeah.
The
video
I'll
try
to
put
up
the
videos
since
I
had
never
supposed
to
actually
I
had
never
posted
anything
to
YouTube
before
ever
in
my
life
and
I
had
never
used
the
playlist,
which
turned
into
be
a
fairly
complicated
process.