►
From YouTube: Kubernetes SIG-Scheduling Weekly Meeting for 20200227
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
A
B
B
A
B
Yeah,
so
this
would
be
a
new
strategy,
so
basically
would
allow
a
user
to.
If
you
turn
on
this
strategy,
the
pod
pod
lifetime
strategy
max
pod
lifetime
strategy.
It
would
allow
you
to
specify
a
maximum
lifetime
for
pods
and
pods
that
are
older
than
the
specified
lifetime
would
be
evicted
essentially
and.
A
I
think
it's
good
to
some
description
for
the
contacts
for
the
for
the
use
case,
and
so
you
want
unlimited
pod
to
be
evicted.
Just
if
it's
running
on
exist
define
the
max
power
lifetime
hours
right.
That
is.
Yes,
that's
correct,
yeah,
I'm,
not
sure
about
the
regular
process
of
proposing
why
in
discovery
so
Mike,
so
I
think
we
need
to
open
the
document
similar
like
like
carob
in
disk
address.
So
there
are
two
putting
this
things
forward.
A
C
B
D
B
D
D
D
A
Know
one
side
wants
that
question
for
the
schedule,
so
if
we
define
multiple
strategies
and
what,
if
two
strategy
conflict
it
what's
their
design
result
example
for
this
strategy
was
to
you
make
this
kind
of
path
for
something
you're,
not
a
strategy.
We
pretend
the
pre
resolve
this
kind
of
path
so
who
wins.
C
So
if
one
strategy
wants
to
preserve
the
spod
and
then
another
strategy
wants
to
evict
it
it'll
end
up
the
victim
d
schedule,
it
runs
it's
the
best
way
to
assume
okay,
but
I
agree
that
that
is
that
is
kind
of
a
an
issue
that
I've
noticed
with
the
D
schedulers
that
it's
not
clearly
defined.
What
priority
you
want
to
give
to
different
strategies,
which
could
be
a
good
improvement
for
it
in
the
future
sure
you.
A
D
D
C
A
A
D
A
D
D
E
Guys
jet
record
I'm
I'm
sorry
for
joining
late,
but
I
think
you're
having
a
good
discussion
on
this
is
one
of
the
questions
we
might
ask
ourselves.
Is
there
a
reason
not
to
do
version
one
to
do
option
one,
and
likewise,
is
there
a
reason
that
we
would
not
want
to
do
for
option
two
I
think
we
just
talked
about
why
we
wouldn't
choose
option
two:
what
about
option
one.
D
So
the
advantage
of
so
why?
Wouldn't
we
want
option
two?
The
main
reason
is
that
it
adds
too
much.
If,
if
a
user
wants
to
to
set
the
extenders
in
all
of
all
of
the
profiles,
then
they
have
to
copy
and
paste
everything
which
is
a
lot
of
burden,
and
our
thinking
is
that
for
the
most
cases
they
they
would,
they
would
want
extenders
to
be
available
in
all
profiles.
D
E
D
So
all
the
all
their
workloads
need
to
go
through
the
extenders,
but
that's
already
something
that
that
happens
today
so
like
today,
I
mean
117
and
before
we
only
had
one
profile,
we
didn't
have
a
concept
of
profiles
anyways,
so
all
workloads
who
would
have
to
go
through
the
same
API
to
selectively,
say
I.
Don't
want
this
extender
to
run.
For
this
part,
so
I.
E
D
A
So
in
the
upstream
you
know,
runtime
class
is
defined
as
a
top-level
API
and
it's
reference
bag,
while
runtimes
class
name
in
the
past
back
so,
but
in
this
case,
of
course,
we
can't
deform
the
extender
API
as
a
top-level
I
mean
in
the
same
level
of
the
component
config,
instead
of
just
in
the
root
level
inside
of
the
component
config
dam
in
each
profile,
we
can
reference
the
top-level
extender
fire,
the
external
name.
So
there
not
option
yeah
I'll
leave
a
comment
there
yeah.
A
A
And
the
third
item
is
because
we
are
close
to
the
typical
phrase,
so
there
are
some
testing
flakes
opinion
is
our
attention.
So
here
are
less
the
issue
about
and
entry
antenna
stats
lake.
So
that's
a.
There
are
three
symptoms
inside
these
issues.
Why
is
the
intended
preemption
issues?
I,
think
that
is
because
of
Burke
defining
some
index
out
of
the
bounds
so
that
it
should
be
a
easy
fix.
The
second
one
is
on
the
second
one
is
also
on
the
preemption
preemption
warrant,
but
there
Lucas
has
actually
not
identified
through
my
local
test.
A
I
found
that
if
I
don't
specify
the
Spector
no
name
directly
and
that
can
resolve
the
issue,
so
we
haven't
go
to
the
bathroom
after
issue,
but
I
believe
this
kind
localized
this
issue,
so
I
recipe
are
and
livered
to
do
there.
So
I
don't
want
to
block
name
this.
That
is
for
preemption
one
and
second,
one
second
entrance
is
a
little
interesting,
so
that
is
in
your
interest.
Introducing
this
release,
which
is
the
part
purchase
trailer.
A
So
basically,
what
I
did
in
the
interim
test
is
I
used
a
replica
set
to
subpoena
for
parts
and
then
I
used
a
plain
part
to
schedule
against
the
existing
full
pass,
so
that
I
want
to
check
whether
the
passer
schedule
even
way
or
not
like
to
test
that
the
pathologist
breath
feature.
But
interested
thing
is
that
the
label
are
defining
the
plain
pod
exactly
matches.
A
The
replica
says
the
label
selector
so
that
the
replica
said
there
is
a
I
think
there
is
a
go
routine
running
background
Ronnie
in
infinitely
and
pick
up
any
part
in
it
thinks
is
his
own
part
that
it's
a
doctor,
iPod
and
and
the
steady
some
reference
aspect
on
with
it.
So
in
that
case
the
prima
heart
was
adopted,
and
then
it
exists.
Sorry
it
exists.
The
Maxima
replicas,
so
is
there
will
immediately
be
deleted,
so
the
intuitive
flakes
kit
Cunha.
A
That
is
an
interesting
bug
and
third
wise
about
the
new
feature
overhead
is
actually
not
a
bad
is
because
the
entrant
has
to
check
into
a
code
base
but
filter
gate.
Not
checking
so
erica
is
working
on
a
fix
for
that.
So
finally,
I
think
we
are
very
good
on
the
status,
so
it
is
for
the
same
thing
so.
A
E
A
So
is
not
I
think
it's
not
clear
documented.
There
is
only
a
document
in
the
pod
template
saying
that
be
aware
of
label
selector
and
make
sure
if
you
don't
want
a
part
to
be
controlled
by
the
part.
Template
don't
make
the
labels
matches
that
part
and
place
the
labels
lecture.
So
if
you
check
the
issue,
8
4,
4,
1,.
A
In
the
second
issue,
if
you
killed
down
second
issue,
885
5,
6,
I'm,
sorry,
the
first
one,
eight
eight
five,
five
two
and
I
lasted
issue.
I
thought
it
was
a
issue.
So
in
the
special
notes
to
the
reviewer
section,
I
said
it's
entering
bar:
instead
of
the
synapse
proxy
885
5
0,
you
can
see
my
debugging
process
and
the
water.
Let
look
at
comment
it
there
I'm
putting
it's
an
interesting
symptom,
I
found
so
I
guess
a
lot
of
people
also
have
this
confusion.
Yeah,
okay,.