►
From YouTube: Kubernetes SIG Scheduling Weekly Meeting for 20230112
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
welcome
to
today's
six
scheduling
meeting
and
today
is
the
first
one.
The
new
year
happy
New
Year
everyone.
So
today
our
main
agenda
is
go
through
the
features
card
caps
in
the
127
release,
so
we
can
go
one
by
one.
The
first
is
for
five
has
clear
decision,
whether
or
not
they'll
be
targeting
in
the
127
and
as
a
clear
owner
related
to
which
is
the
owned
by
Patrick.
A
So
this
is
owned
by
santoshiha
can
say
this
is
actually
following
the
same
parading
with
the
the
same
mechanism
we
introduced
for
the
part
about
spread,
because
during
the
ruling
upgrade
there
can
be
out
paths
versus
New
Paths
and
they
are
showing
the
same
criteria
to
be
matched
by
the
label
sector.
Usually,
and
then
you
just
mix
mess
up
with
how
to
do
the
the
scheduling
directives-
semantics
among
the
New
Paths
only
instead
of
just
combining
all
the
New
Paths.
A
So
this
is
the
one
for
part,
affinity
and
part
and
type
infinity,
and
the
the
oh,
the
other
one
that
we
already
had
is
for
the
for
the
part
number
Express.
So
this
one
for
affinity
and
it's
a
new
feature,
we
we
agreed
to
to
make
it
offer
in
127
I,
will
be
reviewing
that.
The
second
one
is
just
like
I
said
it's,
the
the
other
match
label
keys,
but
in
another
scaling
directory,
which
is
on
the
auto
body
Express,
the
body
constraint,
I,
think
so
that
one
was
Alpha
in
125.
A
So
we
got
a
two
consensus
to
make
it.
Data
in
127
and
Abdullah
will
be
revealing
in
that
yeah
and
to
get
to
the
timeline
you
can
go
to.
A
So
you
go
to
the
Securities
page
I
think
the
enhancement
phrase
is
February
10th,
so
we
have
one
month
before
that.
Just
can
put
it
update,
so
the
third
one
is
surrounded
by
Abdullah
is
about
the
job.
Spec
yourself
can
contain
some
scheduling
directory
like
paraphernalia,
then
type
for
anything
and
also
as
the
schedule
Gates
join.
So
we
have
a
feature
game
regarding
the
whether
the
scheduling
directly
is
rotatable
on
the
job
or
not
so
because
a
suspended
job
doesn't
have
any
backing
parts.
A
So
we
can
change
the
job
securely
because
maybe
the
previous
version
of
job
spec
doesn't
fit
for
the
custom
cluster
situation.
So
we
give
the
controller
or
some
others
some
alternative
option
to
order
to
mutate
the
directing.
So
the
plan
is
to
make
it
GA
in
27
and
Abdullah
will
be
winning
that
and
I.
Will
that
really
like
that
I
think
there's
some
pretty
minor
change
like
to
let
the
scallion
Gates
go
through
yeah.
Let's
look
at
some
other
things,
so
the
other.
The
next
one
is
the
past
scheduling
readiness.
A
This
was
made
Alpha
we
in
20
06
and
we're
gonna,
make
it
120.
Sorry,
we're
gonna
make
it
beta
in
127,
so
you're
not
sure
the
feature
enabled
the
users
to
specify
a
couple
of
strings
working
as
the
gating,
the
parts
Readiness
so
that
before
those
gates
are
cleared,
this
part
will
not
enter
the
scheduling
skill
to
getting
its
chance
to
be
scheduled.
A
So
yeah
that's
more
details,
but
I
don't
want
to
mention
too
much
so
I've
been
owning
that
and
after
I
will
be
in
there
and
this
one
so
I'm
not
sure
who
put
a
note
here,
but
since
somebody
who
doesn't
have
the
capacity
to
get
to
it,
so
maybe
we
just
let
it
know
here.
B
No
I
think
the
problem
is
they
missed
when
they
graduated
to
Beta,
they
did
not
flip
the
flag
could
be
true
by
default
and
so
I
think
that
was
just
already
in
beta,
but
it's
still
disabled
by
default.
A
I
see
I,
see,
okay
got
it
got
it,
and
this
one
I
think
needs
to
so
it's
a
very
complex
PR
involves
Dynamic
resource
and
allocation,
involves
a
lot
of
components
and
I
think
auto
review
that
Primary
in
the
last
release.
So
maybe
there
is
something
to
do,
but
I'm
not
sure
Patrick
can
have
cycle
to
promote
to
play
in
English.
But
my
guess
is
no,
but
yeah
Patrick.
C
He's
gonna
be
working
on
multiple
changes
that
were
requested
between
Alpha,
but
there
is
no
graduation
to
Beta,
okay,
so
I'm
just
going
to
stay
off
yeah.
The
question
I'm
asking
mostly
to
Signal,
is
whether
they
want
to
still
track
it
as
part
of
the
enhancements
process,
because
the
changes
I
mean
it's
not
one
to
change.
This
like
there
is
a
big
list,
so
maybe
it's
worth
going
through
through
the
process.
Anyways,
but
still
it's
gonna,
remain
beta.
Sorry
remain
Alpha.
A
Excess,
so
for
the
last
one
is
that
some
users
reports
the
requirements
saying
that
okay,
the
default
preemption,
that's
honor,
the
pdb.
You
know
best
efforts
way
means
okay.
If
there's
one
candidates
who
is
passed
filing
in
the
PV
and
the
other
is
not
definitely
will
choose
the
one
that
doesn't
have
any
PDP
violation
out
there,
but
it
is.
It
is
the
best
advice
manner.
A
That
means
we
each
if
two
candidates
are
the
only
two
candidates
and
they
both
tie
with
violating
the
pdbs
that
we
just
go,
choose
one
to
preempt
the
one,
even
if
it's
pdb
protected.
So
there's
a
requirement
says:
okay,
given
epidemi
is
kind
of
formal
directory
to
ensure
the
user's
workers
not
disrupted.
Maybe
we
should
not
go
preempted
apart
if
they
appear
to
be
exactly
so
I
think
the
the
Alex
proposed
some
a
good
document
there
and
I
think
we
more
prefer
to
apply
a
field
to
the
pdb.
A
Spec
says:
okay,
preemption
enforce
us
or
something
like
that
so
to
honor
the
pdbs
semantics
during
the
default
preemption
in
the
enforced
manner.
So
this
can
is
not
that
prioritized
about
ever
touch
base
with
Alexa
offline,
and
then
we
will
pick
up
the
approval
too
see
how
we
go
forward
so
yeah
any
other
missing
features.
I
mean
cap.
We
have
or
formal
features.
You
are
aware
of
then
just
leave
it
here
or
to
spinos
in
the
slide.
C
Oh
yes,
yeah
a
company
if
there
are
no
other
topics.
Actually
there
is
a
number
of
tickets
like
this.
This
one
is
about
cost
scheduling.
Basically,
key
flow
is
thinking
whether
they
want
to
add
support
for
bot
group
as
it
is
currently
implemented
in
the
scheduling.
C
Plugins
repo
qflow
already
has
support
for
volcano,
so
they're
thinking
about
in
this
all
group,
I
left
a
question
there
mostly
well
to
Wei
on
Alex
whether
about
what
what
is
the
work
of
all
group
going,
Upstream
Upstream,
because
maybe
it
doesn't
make
sense
for
qflow
to
support
group
if
it's
not
Upstream
so
better.
Maybe
it's
better
to
think
about
how
how
we
kind
of
stream,
both
group
yeah
so.
A
A
C
C
It
would
be
possible
to
proceed
with
with
the
proposal,
and
another
thought
I
have
is
in
terms
of
performance
of
the
cost
scheduling
plugging,
because
currently
the
cost
scheduling
plugin
relies
heavily
on
the
Q
sort,
endpoint
yeah
and
that
doesn't
sound
that
efficient.
But
now
that
we
have
the
free
pre-inq
endpoint,
it
could
be
a
major
significant
performance
Improvement
for
the
coastline.
A
So
I
have
one
issue
tracking.
That
is
that
so
the
embarrassing
thing
for
the
part
group
you
start
is
that
we
just
use
the
part
groups
creation
timestamp
as
the
Sorting
criteria,
but
that
is
far
from
Perfection,
because
that
that
timestamp
is
fixed.
That
means
if
the
pi
is
created
just
by
some
time
it
holds
the
priority
ahead
of
other
November
Park
group
in
the
past.
A
But
we
don't
have
this
issue
in
the
regular,
odd
base
scheduling
in
our
entry
part,
because
every
time
we
reschedule
apart
once
the
part
is
being
re-que
the
the
queue
timestamp
is
to
refresh,
but
we
cannot
be
easily
doing
that
for
Part
Group,
because
we
just
don't
include
a
virtual
concept
of
pop
group,
but
with
the
green
queue
extension
that
is
I
would
say
it's
possible,
but
I
need
to
do
some.
Some
exercise
on
that.
A
That
is
there's
one
issue,
checking
that
and
if
we
can
make
so
issue
the
fairness
in
our
group
and
the
Power
Group
versus
other
regular
part,
we
can
sort
of
like
impulse
the
proper
back
off
for
the
part
group
itself,
instead
of
like
just
the
park
of
always.
It's
a
very
high
priority
comparing
to
others
yeah.
That
is
one
issue
to
to
resolve.
C
C
A
Yeah
anyone
who,
who
has
interest,
can
pin
me
or
Alex
pick
up
just
as
I
mentioned
for
now.
It's
stuck
has
two
points.
One
is
the
integration
with
customers
scalar,
the
other
is
to
fairness
of
the
past
group.
Part
groups
queuing
so
and
with
help
of
killing
the
increasing
queue
extension
Point.
The
latter
problem
may
be
result,
but
we
still
need
to
think
about
spend
time
on
the
integration
of
the
class.
So
let's
go.
A
Okay-
and
we
still
have
some
time,
yeah.
C
They
sometimes
I
would
like
to
mention
that
kill
flow
at
well,
not
qflow,
but
one
contributor
in
keep
flow
is
proposing
a
number
of
changes
so
that
keep
flow
is
more
compatible
with
kubernetes
and
the
other
issue
that
was
open
not
related
to
this
but
related
to
job
is
that
they
want.
Basically,
they
are
proposing
to
use
the
index
job
as
the
underlying
mechanism
to
to
to
create
pots
into
flow.
C
I
didn't
include
that
in
the
docs
I
can
search
for
you
and
put
it
there,
but
then
that
also
has
some
nice
implications,
because,
let's
say
if
Q
flow
migrates
to
job,
then
maybe
we
don't
need
a
pod
group.
Maybe
we
can
just
use
job
as
the
you
know,
as
the
grouping
I'm.
C
D
Oh
hi,
everyone,
so
my
question
is
more
like
a
general
question,
so
I
don't
know
if
you
have
like
Financial
or
this
this
section
of
questions
that
I
can
ask
for
a
general
content.
D
Okay
sounds
good,
so
I'm
new
to
this
interest
group
I'm
right
now,
I'm
reading
the
scheduler's
code,
I
wonder
like
will
be
the
best
way
to
seek
for
some
helps
when
I
I,
you
know,
don't
understand
some
code
details
or
you
know,
don't
understand
some
decent
decisions.
I
didn't
really
found
much
like
documentations
about.
Like
the
you
know,
the
code
structures
in
general,
so
I
wonder
if
you
can
give
me
some
pointers
or
like
yeah
or
like.
If
there
are
any,
you
know,
discussion
forum,
yeah.
A
Oh
I
think
that
the
best
way
to
get
started
with
I
suppose
just
from
a
very
brand
new
to
kubernetes
and
open
source
development.
This
website
kubernetes
kubernetes
dot.
Dev,
is
pretty
nice
website
and
you
will
see
a
lot
of
ton
of
the
resources
out
here
like
there's
someone
to
get
your
study
like
it
just
goes
through
the
yeah
get
started
to
get
your
study
from
the
very
beginning
and
the
ETC
code
commission
blah
blahness.
A
A
A
Yeah,
like
a
hierarchy
overview
here,
also
is
kept
up
to
date,
and
but
if
you
find
some
out
of
that
information
just
raise
a
PR
to
right
to
to
crack
it
I
think
this
one
is
very
good.
Yeah.
D
I'm
I'm,
honestly
that
I
I
cannot
right
now
get
some
high
level
ideas
about
how
this
close
calculator
work
right
now,
I
probably
have
more
like
a
detailed
question,
for
example,
how
those
plugins
are
like
triggered
in
order
so
I'm
trying
to
find
the
code
for
for
that,
but,
like
it
probably
takes
me
some
time
to
really
find
like
how
this
orders
are
plugging
actually
be
so
I.
Wonder,
like
you
know,
for
this
kind
of
question.
What's
the
best
way
I
can
you
know
find
some
help,
yeah.
A
D
A
I
think
me
and
some
others
get
give
about
the
to
schedule:
parkings,
okay,
I
can,
it
could
be
me
and
I
can
give
you
the
link,
offline
and
also
I
I
suppose
you
know
there
is
a
repo
called
schedule,
plugins
and
there's
a
lot
of
plugins
written
here
and
you
can
turn
to
because
this
is
so
so-called
Auto
three
plugins
and
also
you
can
turn
to
the
Upstream
for
the
intern
entry
plugins
for
their
implementation.
But
before
that
maybe
the
the
cube
contact
is
a
good
stuff
for
you
to
get
to
know.
A
B
B
D
C
I
would
say
so:
trolley
is
best
to
start
in
the
community
level.
The
first
link
the.
D
C
Way
sure
sure
and
as
you
go
through
all
the
other
documents
that
people
have
shared
or
even
if
you
go
to
the
slack
Channel
as
you
find
these
answers,
maybe
it's
good
if,
if
you
have
some
time
just
add
them
to
the
devil.
D
D
C
So
that
other
people
can
benefit
in
the
future,
they
are
reasonably
up
to
date.
What's
the
okay,
2021,
okay,
A
bit
old
now,
but
yeah,
it
feels
actually
reason
that
people
some
people
did
this
rearrange
this
this
documentation.
So
it's
somewhat
up
to
date.
So
if
you
go
through
the
through
all
these
questions-
and
you
get
answers,
please
open
up
your
here.
D
A
B
I
can't
even
ask
me,
but
we
can
just
tease
it
so
we're
looking
at,
like
we
had
this
nice
dashboard
for
open
issues
that,
like
in
under
set
under
scheduling
that
we
can
use
for
triaging,
probably
in
the
next
meetings.
B
The
only
problem
with
it
is
that
by
default,
there's
no
like
automated
way
of
populating
that
dashboard,
but
it
seemed
that
someone
wrote
a
bot
that
we
can
use
like
Sig,
auth
I
think
they
wrote
a
bot
that
helps
automating
that
so
maybe
we
will,
if
anybody
has
like
a
knowledge
of
how
to
set
up
bots
on
GitHub.
Please
let
us
know
if
you
can
help
us
set
up
this
tool
for
our
dashboard.
B
So
what
yeah?
What
way
is
showing
here
is
the
tool.
All
right
is
the
basically
the
yeah
the
tool
and
I
think
this
can
be
set
up
as
a
bot
on
GitHub.
A
Yeah
yeah
a
little
background
is
that
a
GitHub
has
I
think
it's
a
beta
feature
called
GitHub
project
and
this
is
a
sort
of
a
dashboard
and
you
can
put
their
issues
on
whatever
you
want
to
track
here,
but
for
now
it's
not
automatic,
so
you
have
manually
put
it
there,
but
someone
likes.
You
asked
written
some
good
tools
to
Auto
populate
the
the
just
open
the
new
issues
and
the
PRS
here.
So
we
want
to
follow
the
same
process
and
have
some
kind
of
Bot
to
Auto
Import.
A
C
How
how
do
you,
who
has
the
power
to
currently
add
new
items
here.
A
C
Okay
and
I
guess
for
everyone
listening.
Yes,
this
applies
to
any
kubernetes
repo.
C
So
if
you
have
things
for
for
enhancements
or
for
code,
we
can
start
tracking
it
here.
A
All
right,
I
think
this
pretty
much
for
today's
meeting.
I
will
start
recording.