►
From YouTube: SIG - Performance and scale 2023-08-03
Description
Meeting Notes:
https://docs.google.com/document/d/1d_b2o05FfBG37VwlC2Z1ZArnT9-_AEJoQTe7iKaQZ6I/edit#heading=h.tybh
A
These
address.
How
far
is
an
attendee
okay,
I
think
going
I
wanted
to
go
over
today
at
least
was
just
to
push
through
and
tracking
and
I'm
gonna
see.
If
there
was
any
progress,
I
think
Bobby
was
able
to
complete
his
release.
Overhaul
changes,
I
think
were
at
a
point
where
we
can
see
how
some
other
stuff's
going
I
think
this
looks
go
pretty
well,
so
well,
you've
got
the
first
one.
B
I
think
it's
I
have
addressed
all
the
comments
on
there.
Okay,
we
just
need
some
more
eyes
and
more
discussion
around
it.
I
can
set
up
some
time
this
week
for
us
to
go
through
and
decide
the
next
steps
here.
So.
A
A
A
I
think
we
might
get
some
interest
and
I
think
that
that
would
be
a
good
next
step
to
see
what
we
can
get
for
people
to
show
up
and
give
us
some
feedback
on
this
I
think
this
is
there's
like
a
lively
discussion
that
can
unfold
from
this
I
mean
I.
Think
there's.
A
You
know
at
least
a
lot
of
the
things
that
I
got
like
when
talking
to
some
of
the
maintainers
was
about,
like
you
know
the
process
and
sort
of
how
we
go
about,
maintaining
it
and-
and
you
know
one
of
the
responsibilities
and
so
I
think
that's
maybe
where
we
can,
we
can
go
to
get
some
more
interest,
because
I
think
at
least
to
me
I
think,
what's
here,
makes
sense,
I
think
where
we're
gonna.
A
It's
like
this
stuff,
like
reviewing
the
the
kind,
API
change
and
sort
of
organizing
the
a
group
to
do
that.
That's
the
sort
of
the
the
commitment
as
part
of
this
and
seeing
what
the
interest
is
I
think
there's
interest
to
do
it
I
just
think.
Maybe
we
need
to
have
a
forum
where
we
discuss
it
and
not
be
sort
of
the
way
to
get
the
interest
on
this
and
get
move
this
forward.
A
B
So
you!
So
if,
if
I
understand
correctly,
we
should
I'll
start
a
thread
on
keyboard
Dev
mailing
list
and
ask
for
next
steps
on
on
this,
so
it
would
be
along
the
lines
of
can
I
get
some
reviews,
some
more
discussion-
and
you
know
some
help
on
on
the
next
steps
here-
is
that
what
your
suggesting.
A
Like
some
something
like
this,
so
like
Lee,
has
done,
Leah
sent
out
an
invitation
for
this
I
I
didn't
I
couldn't
attend.
It
was
early
this
morning,
but
I'm
wondering
what
the
attendance
was
for
this
like
and
others
have
done
this
as
well
like
so
I'm
assuming
this
is
getting
some
interest.
I
mean
you've
seen
like
so.
You've
got
design
proposal
and
then
some
people
like
this
so
he's
something
like
that.
Where
he's
he
sent
out
the
design
proposal,
he
schedules
a
meeting
and
then
that's
that's.
A
What
I'm
saying
is
it
like?
Try,
try
and
go
this
path
because
as
a
way
to
get
some
people
to
get
opinions,
okay,.
A
Okay,
yeah
because
I
know
this
one.
This
one
has
a
lot
of
interest.
It's
just
we
we've
everyone's
busy
and
we
basically
just
need
to
reignite
the
the
topic
and
put
it
at
the
Forefront
of
our
for
some
folks.
So
I
think
it
could
be
a
way.
I,
don't
know
we'll
see,
I
guess
what
what
comes
of
it.
A
All
right
we've
got
deprecation
policy.
I
I
really
think
this
thing's
the
same
I,
don't
know
Adam
RC
seat
himself,
but
I
don't
know
if
I
know
this
has
had
a
lot
of
back
and
forth,
see
June.
A
B
Yeah
I
think
all
of
these
bullet
points
are
along
the
same
lines
right
so
I
wonder
once
we
start
the
API
review
discussion
like
all
of
these
can
happen
as
follow-ups.
A
B
Not
sure
if
they
are
related
wearing,
they
can
happen
in
parallel
or
they
need
to
go
in
the
same
order.
But
yeah.
A
I
I
mean
I
think
this
thing
has
been
happening
in
parallel
for
a
while
I
guess
what
I
was
one
of
saying
is
that
I
mean
it
would
be
nice
to
maybe
it
or
can
at
least
I
don't
know
after
you
have
your
discussion
or
whatever,
basically
spend
some
time
on
on
getting
some
like
I
think
these
in
the
same
place
as
you
so
I'm
trying
to
say,
and
maybe
he
needs
to
go
the
same
path
yeah,
because
this
is
right.
A
So
we've
been
around
since
June,
okay,
okay,
the
filter
release
notes
this
is
Anders,
but
then
we
need
to
this
means
a
pull
request.
A
B
A
B
Have
not
had
a
chance
to
make
progress
on
these
I'm
trying
to
focus
on
the
simulation
in
seeing
if
we
can
get
Advantage
from
the
fake
cubelet
stuff,
the
the
Quark
controllers,
okay,
so
I'll
plan
on
spending
some
time
this
week,
I
think
as
far
as
priority
is
concerned,
the
the
first
thing
that
we
discuss
last
time
that
needs
to
be
taken
care
of
here
is
the
tooling
documentation
so
that
we
can
go
ahead
with
that
blog
post.
B
So
I'll
spend
some
time
this
week
to
come
up
with
a
documentation
PR
for
the
tooling
and
then
some
rough
guideline
on
the
blog
post
that
we've
discussed.
A
Okay
sounds
good
I,
wonder
if
we
can,
one
of
Daniel
can
help
us
put
this
on.
Can
we
put
together
the
posting
account
for.
B
Us
I'll
take
that
as
action
item
two
so
I
need
to
I
need
to
create
an
issue
for
it
and
start
a
thread
on
keyword
Dev
to
to
seek
help.
I'll
start
that
at
least
this
week
and
see
how
far
we
get.
Okay.
A
B
That
page
is
Obsolete
and
that
that's
what
that
issue
is
for
so.
A
If
you,
if
I
go,
but
if
I
do
but
I
can
get
to
this
page,
though.
B
Oh
yeah,
you
can,
it
should
be
there
in
the.
A
The
link
you
had
is
obsolete.
I
know
that's
what
you're
saying
I
I'm
looking
for
I'm
just
going
to
change
it
I'm
going
to
edit
it
okay!
Well,
here
we
go
this
one
right.
This
is
oh!
No!
This
is
not.
It
is
a
rendering
yeah,
oh
well,
okay!
So,
okay,
that's
what
I
need
to
do
so.
I
just
need
to
change
this
to.
A
B
B
Yeah
I
can
give
you
a
quick
update
so
right
now
we
are
at
a
point
where
I've
added
support
for
BMI
running,
going
from
schedule
to
running
State
now
I'm
looking
at
I'm,
looking
at
all
the
vectors
where
load
can
be
generated
on
the
API
server,
so,
for
instance,
the
demon
sets
that
are
part
of
the
kubernetes
cluster.
They
would
generate
some
watch
load
and
get
put
Edge
update
load
against
the
API
server.
B
So
the
easiest
one
right
now
is
predicting
the
watch
because
that's
where
the
informers
for
all
the
controllers
start
so
I've
taken
summer
of
estimate
and
I'm
working
on
another
controller,
which
will
add
an
estimate
of
watch
requests
against
the
API
server.
The
initial
proof
of
concept
was:
it
went
well.
I
saw
a
memory
Spike
from
2
gigabytes
to
5
gigabytes
when
when
I
started,
all
those
watch
threads.
So
that's
in
line
with
our
approximation
but
I'm
running
into
troubles
right
now.
B
With
that
controller,
so
I
don't
have
anything
to
like
demo
right
now,
working
forecast.
That's.
A
B
Yeah
so
the
way
this
works
right
now
is
the
new
controller
watches
on
a
node
and
as
soon
as
a
fake
node
is
created,
the
current
Quark
controllers
will
make
it
go
to
running
state,
and
this
new
controller
will
see.
If
it
is
in
the
running
State,
then
it
will
start
a
bunch
of
watch
requests
so
that
will
simulate
the
cubelet
load
generator
and
so
that
will
simulate
requests
coming
from
that
fake
cubelet,
as
well
as
fake
pods.
On
that.
A
No,
all
right,
okay,
cool
okay!
Are
you?
Are
you
still?
Are
you
doing
this
working
committee
contributing
back
to
the
Kayla
project
or
how
are
you
like,
or
do
we
need
to
have
its
own,
like
plug-in
keyword.
B
I
am
not
sure
yet,
because
so
what
I
want
to
the
way
I'm
thinking
about
this
is
that
I
want
to
understand
if
this
gets
us
closer
to
the
actual
real
world
simulation
and
if
it
adds
value
to
the
project.
So
once
I
can
prove
that
this
is
like,
let's
say,
50
or
60
percent
to
the
real
world.
Stress
I,
I,
think
that
would
be
the
right
time
to
start
having
conversation
with
with
the
project
maintainers
yeah,
so
this
is
called
POC
work.
Until
now,
it's
it's
not
at
all
polished.