►
From YouTube: Kubernetes SIG Scheduling Weekly Meeting for 20220602
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
Hi
everyone
welcome
to
today's
six
schedule.
Meeting
last
meeting
was
cancer
due
to
the
cut
call.
So
this
is
for
two
seconds
and
there's
not
much
items
in
the
agenda,
so
the
first
one
is
the
delete
only
proposal.
This
is
a
very
old
issue
back
in
2015.
So
the
background
is
that
in
the
very
beginning,
we
don't
have
a
clear
gbk
group
version
kind
concept.
So
a
lot
of
resources
were
defined
in
the
top
level
like
the
bonding
resources.
A
So
the
bonding
is
operation
to
ask
the
api
server
to
modify
the
pad
spec
dot
domain,
as
you
mentioned,
as
you
may
already
know,
but
it's
now
a
top
risk
top
api,
which
is
on
the
v1,
but
that
symmetrically
doesn't
make
sense.
So
the
the
proposal
is
to
make
it
on
the
path,
so
that
is
path
slash
by
binding,
so
that
can
distinguish
each
resource
bindings
requests
separately.
So
the
proposal
makes
sense
to
me
totally
makes
sense
to
me
and
yeah.
B
B
I
see
well,
it
sounds
like
something
that
will
require
a
cap,
probably.
A
B
B
But
who
is
requesting
this,
why
why
is
it
important.
A
A
Is
not
that
suitable?
So
if
you
look
at
some
other
patterns
like
this,
some
other
external
project
says
okay,
I
want
to
do
a
bonding
to
my
resources
like
so
they
need
to
implement
the
specific
resources,
slash
banning
and
instead
of
get
confused
by
the
top
level
v1
without
bending.
So
what
does
that
mean
that
actually
means
department,
so
that
doesn't
quite
look
clear
to
them.
So
it's
just
to
make
the
semantics
more
explicit
and
clear.
B
I
see
oh
yeah,
I
don't
have
anything
against
it.
I'm
just
not
sure
about
the
priority.
A
The
priority
yeah
given
it
has
been
existing
for
a
while.
It's
not
urgent
and
it's
just
a
good
hab
to
get
implemented
and
get
the
yeah
to
get
this
media
smoke
clear.
C
Yeah,
but
but
I
think
it
is
not
only
used
by
the
cookie
schedule,
it
also
may
be
used
by
some
people
use
certified
custom
scheduler.
They
also
use
this
this
binding
api
to
update
to
the
node
name
of
the
port
yeah.
But
if
we
change
it
at
the
api
level,
it
may
be
hard
for
the
people
to
integrate
with
the
coordinates
they
may
be
updated.
They
maybe
need
to
update
their
product
yeah
to
yeah
either.
A
So
basically,
it's
this,
it's
a
it's
a
it's
a
polishman,
but
it
may
have
some
sad
effect.
So
I'm
not
sure
if
we
put
this
into
the
seek
architecture
or
cigar,
pin
kpm
machinery
that
will
get
some
pushback
and
the
same
so
we
don't
have
much
benefit
it's,
except
for
the
for
making
the
things
more
more,
pretty
more
clean,
so
yeah
it
depends
depending
on
maybe
there's
some
api.
Also
api
implements
implementations
of
the.
A
So
yeah,
this
is
the
thing
I
will
comment
on
the
issues
to
see
who
is
interested
and
who
may
already
have
some
expertise
and
do
some
research
on
this
right,
this
one
and
other
than
that
it
seems.
There's
no
other
items
today
and
if
you
do
have
some
please
visit
that
and
otherwise
I
may
give
you
some
minutes
back
and
others
see
any
urgent
items
to
discuss
today.
A
B
No
well,
since
we
have
time,
maybe
we
can
chat
a
little
bit
about
the
one
I
just
posted
in
the
chat,
the
one
about
events
to
register.
A
A
Some
predict
so,
along
with
the
introduction
of
the
events
to
register
function,
so
each
plugin
can
claim
what
kind
of
cluster
events
is
correlated
with
the
failure
that
this
plugin
fail
particular
path
like
if
you
want
bonding
plugins.
So
you
fail
a
path
by
register.
Some
volume
events
like
if
a
new
pv
or
pvc
gets
introduced,
maybe
the
unscheduled
part
failed
by
your
volunteer
plugin
can
be
made
reschedulable.
A
So
this
is
good,
but
there
are
some
corner
cases,
no
qualities,
but
some
special
cases
is
that,
like
the
sample
shoe
mentioned
that
there's
some
categories
we
didn't
consider
properly
and
yeah.
If
you
want
to
look
at
just
look
at
this
comment,
I
won't
go
through
every
case
you
mentioned,
because
some
of
them
are
still
being
discussed,
but
some
of
the
point
is
valid
like
this
one.
A
So,
for
example,
if
failed
by
some
advanced
scheduling,
feature
very
advanced,
I
mean
some
features
that
consider
cross
node
path,
distributions
like
part
to
larger
spread
and
the
interface
affinity.
So
in
this
case
you
cannot
simply
think
that
a
single
path
event
will
will
not
impact
this
path
reschedulability.
A
So
this
is
the
thing
we
didn't
consider
properly
and
nowadays
like
like
for
now,
if
a
part
gets
get
added,
for
example,
maybe
the
node
you
previous
failed,
I
mean
not
fit
for
the
topology
spread
can
be
schedulable
because
in
another
part
in
another
note,
the
total
number
of
that
particular
net
of
that
particular
topology
region
get
increased
but
rhino
is
not
implemented
yet.
A
I
would
say
called
mixing.
Scheduler
directories
electives,
like
in
the
latest
implementation
of
the
particular
spread.
We
introduced
the
minimum
domains.
We
introduced
the
nodes,
inclusion
policy,
so
that
is
totally
one
area
we
need
to
consider
when
we
make
when
we
actually
mixing
some
some
some
scheduling
features
altogether.
So
this
is,
I
will
go
through
the
comments
today
or
this
week
and
see
whether
we
can
get
a
good
strategy
on
this.
B
C
A
So
the
I
think
the
pre
check
will
fit
for
the
regular
scheduling
failures.
I
mean
the
single
node
check,
scheduling,
failures,
like
resources,
selecting
like
no
power
like
some
others,
but
for
the
enterprise
scheduling
constraint
like
if
a
part
previously
failed
for
the
project
spread
and
a
new
node,
get
added.
Sorry,
a
new
sorry,
for
example,
if
a
node
get
edited
like
this
label
got
changed.
A
A
A
B
A
A
One
is
basic
scheduling
constraint,
which
is
fit
for
the
one
node
right
and
the
other
is
more
advanced,
the
features
which
is
considered
cross,
node
distributions
so
for
for
part,
project
spread
and
interprofinity
that
I
think
it's
the
the
most
special
two
features
we
have
now
so
for
these
two
features
we
need
to
do
some
additional
check
to
see
okay,
if
the
no
that
doesn't
if,
for
example,
if
a
null
get
introduced
and.
A
And
then
we,
inter
in
iterated
the
parts
right
if
a
part,
doesn't
go
through
the
pre-check,
but
it's
a
particular
part
failed
by
the
advanced
scheduling,
plugins.
Like
part
project
spread
and
the
proximity
we
still
get
that
chance
to
be
moved
back.
So
that's
a
very
rough
idea.
I
have
so
each
plugin
maybe
have
a
attribute
to
claim
this
as
a
advanced
scheduling,
plugin
or
a
basic
so
for
basic.
We
just
do
the
pre-check
basic
check
and
for
the
others
we
give
it
a
chance,
but
it's
very
rough.
B
Did
we
find
a
a
case
for
a
pot
affinity
or
antifinity,
because
we
for,
if
I
remember
correctly,
for
for
port
affinity
when
a
node
when
a
plot
changes,
we
check
that
if
it
matches
the
affinity
or
infinity.
A
B
Then
I
guess
my
my
question
is
whether
we
want
to
to
make
it
configurable
or
make
it
only
for
photopoly
spreading.
A
B
A
So
there
is
a
is
that
that
particular
we
can
call
that
part
the
critical,
the
part
in
the
critical
path
to
mark
the
finish
of
the
part
group,
I
mean
the
satisfy
the
minimum
member
of
the
apart
group.
So,
upon
the
arrival
of
that
part,
and
that
part
happens
to
be
schedulable,
we
move.
We
initially
activate
the
request
in
that
past
scheduling
process
so
that
it
got
the
chance
to,
during
that
particular
waiting
period,
that
the
other
n
minus
one
task
get
the
chance
to
be
moved
back
to
the
active
queue.
A
So
that
is
this
method
technique.
The
co-scheduling
plug-in
is
doing
like
it's
activated,
it's
proactively,
activate
its
sequence
of
which
belongs
to
the
same
power
group.
So
that's
the.
A
It's
it's
more,
it's
more
a
hook
in
the
framework
because
it's
using
public
defined
and
reserve
the
the
key,
the
string
called
the
past
two
active
and
the
index
in
the
skeleton
of
the
framework
code.
We
will
check
that
by
the
end
of
scheduling
cycle
and
the
bonding
cycle.
So
if
we
got
some
parts
in
the
state,
we
will
move
them
back
in
a
better
efforts
manner.
A
That's
that's
a
more
hurt
instead
of
public
sdk
api
or
apis.
A
Yeah,
it's
it's.
Its
goal
is
to
make
the
path
belonging
to
the
same
power
group.
Has
the
higher
chance
to
be
convert
to
be
diverged
one
by
one
and
to
be
rescheduled,
so
you
can
see
it's
for
the
performance
to
improve
performance
and
kind
of
confirmed
by
some
users,
like
the
opening.
C
B
Bots
that
use
the
required
spreading
right.
The
do
not.
C
B
A
A
One
other
thing
is
that
we
may
promote
the
schedule
component
config
to
v1
in
this
release.
U125.
If
there's
a
issue
there,
I
will
link
it
back
here.
If
there's
no
objection,
we
will
follow
a
path
and
there's
a
discussion
is
to
keep
the
we
want
beta
to
steal.
From
for
another
release,
the
market
has
deprecated.