►
From YouTube: SIG - Performance and scale 2023-06-08
Description
Meeting Notes:
https://docs.google.com/document/d/1d_b2o05FfBG37VwlC2Z1ZArnT9-_AEJoQTe7iKaQZ6I/edit#heading=h.tybh
A
A
All
right
all
right
performance,
job
results,
I
merged
your
PR,
so
we're
good
to
go
here.
B
I'm
sorry
I
need
to
drop
and
join
right
back.
Sorry
about
that.
A
B
Awesome
yeah
sorry
I
missed
some
of
what
you
were
saying
earlier.
A
So
all
the
same
as
the
the
the
poor
quests
you
created
was
merged,
so
I
don't
know.
If
there's
anything
we
want
to
review
on
this
or.
B
A
B
So
I
was
doing
some
work
so
yeah.
Let
me
give
an
update
or
on
what
are
the
things
that
are
needed.
So
first
I
have
a
comparison
of
of
previous
data
with
a
newly
added
job
results.
B
Okay,
so
this
is
so,
there
are
three
charts:
currently
a
periodic
end-to-end
BMI
periodic
end
to
end
VM
and
density
100
pm.
B
Now
we
regressed
a
little
bit
in
terms
of
getting
the
historic
data.
The
plots
right
now
are
only
for
weekly
data
and
I.
Have
an
issue
open
for
this
aggression.
Bill
will
soon
fix
this,
but
what
I
wanted
to
do
today
is
take
this
historic
comparison
that
is
on
the
right
and.
B
Compare
the
recent
weekly
data
and
see
how
we
are
doing
Ryan.
Can
you
still
hear
me.
A
B
I
got
you:
okay,
just
checking
okay,
so
first
thing:
let's
look
at
the
p50
for
VMI
is
as
noted
earlier.
We
saw
that
the
time
is
actually
going
down,
p50
and
I
think
be.
So
if
we
look
at
this
range
right,
we
were,
we
were
below
20
close
to
15
range.
B
Here
we
are
seeing
that
some
dots
are
are
coming
up
in
that
are
above
20,
but
all
in
all
we
are,
we
are
doing,
okay
in
in
terms
of
average.
B
Let
me
see
if
I
can
do
this
okay,
so
this
will
be
more
recent,
so
these
had
nice
groupings
of
of
15,
16
I
think
it.
It
seems
to
me
that
we
are
regressing
a
little
bit
222
here
so
because
of
that
it
would
be
good
to
also
take
a
look
at
P95.
C
B
B
A
Okay-
maybe
let's
see
so
this
is
a
week
to
week.
Maybe
let's
see
if
the
trend
continues,
let's
see
if.
A
Kind
of
as
we
slowly
Zoom
this
out,
we
keep
an
eye
on
this
data
and
we'll
see
if
it's
yes,
if
this
is
pointing
to
a
pattern.
C
A
What
was
the
date,
it
was
like
May.
So
first.
B
B
June
4th
is
when
we
are
seeing
that
upper
dot
in
b50.
C
C
A
B
No
I
I
was
saying
that
I
couldn't
find
anything
on
anything
that
stood
out
on
the
charts
below
this.
This
was
the
only
observed
anomaly,
I'm
just
scrolling
through
to
see
if
anything
else,
pops
out.
Oh.
B
B
B
Okay,
so
with
this
we
can
go
to
the
next
chart.
C
B
We
are
at
18,
so
same
performance
regression
that
we
saw
for
VMI
around
three
to
four
seconds.
Excuse
me
is
observed
here
as
well.
So
if
you
look
there
were
there
were
no
dots
closer
like
Beyond
20..
Here
we
are
seeing
some
dots
consistently
go
above
20..
Maybe
that's
bringing
the
the
averages
up
a
little
bit
so
I
I
think
both
for
VM
as
well
as
VMI,
be
regressed.
A
little
bit.
B
B
Patch
BMI
same
story:
here
we
have
improved
a
little
bit
in
averages.
C
C
B
Okay,
so
with
that,
we
can
move
on
to
the
density
comparison,
so
this
one
was.
It
was
different
to
the
performance
job.
Here
we
actually
saw
the
numbers
climb
when
we
last
reviewed
the
graphs,
so
we
just
need
to
make
sure
that
comparing
this
is
is
actually
compatible.
B
We
were
seeing
close
to
50
did
the
average
in
the
last
week
when
we
tracked
this
now,
if
you
look,
we
have
gone
down
in
those
averages.
We
are
consistently
hitting
in
mid
40s
and
the
the
average
for
this
week
is
actually
43..
So
this
is
an
oh
actually
I'm.
Looking
at
the
wrong
metric,
let's
see
so
consistently
hitting
high
50s
in.
B
Yeah,
so
the
trend
continues.
We
did
not
improve
so
the
last
time
we
talked
about
this.
B
B
B
B
A
C
B
Weeks
to
one
week
the
this
is
me
so
when
we
transition
from
my
for
my
Benchmark
collecting
repository
to
this
CI
benchmarks
repository,
there
was
a
first
instance
that
collect
collected
all
history
and
then
from
there
on
I
have
been
looking
at
weekly
charts.
So
there
is
a
feature
that
is
needed
in
the
graph
plotting
tool
that
can
accept
a
date
from
which
we
need
to
plot
the
graph.
So
that
is
missing
and
and
that's
why
this
graph
only
plots
the
data
that
is
in
the
repository,
which
is
for
one
week
so
I.
B
B
B
Yeah,
so
once
we
have
this
fixed,
this
will
be
a
continuous
graph,
so
you'll
be
able
to.
Instead
of
comparing
it
left
and
right,
you'll
be
able
to
see
it
in
in
a
straight
timeline
on
a
single
chart.
Okay,.
A
Okay,
all
right
I'll
check
that
thanks,
bye,
okay,
okay,
sounds
good,
then
so
we've
had
so
I'll
have
some
changes
that
I'll
come
in.
So
what
about
us?
Let's
review
some
of
the
stuff
for
V1.
So
we've
got
these
two.
A
I
think
we
just
need
to
for
this
one:
hey
we
just
pinged
in
I
can
just
ping
Daniel
or
maybe
sort
of
thread
with
Daniel
and
let's
see,
if
we
can
take
this
over
from
him
and
then
I
mean
it
seems
to
be
done,
I
mean
just.
We
need
the
changes
that
incorporates
the
the
pr
you
did
and
then
I
think
we're
good
I'd
like
to
get
this
like
if
Daniel's
busy,
I
don't
want
to
block
on
this.
Let's
just
start
a
Threading.
Let's
let
them
know
we'll
take
this.
A
Oh
okay,
okay,
so
maybe
what
we
can
do
is
like,
let's
just
sync
with
him
and
just
say:
hey
we'll
take
this
and
you
can
put
push
the
commit
push
the
commit
that
corrects
this
and
then
maybe
you
can
approve
it
and
we
can
move
on
Harmony
Louisville
can
do
it.
B
Yeah
I
think
that
would
be
great
because
I
have
to
spend
at
least
30
minutes
before
every
meeting
to
get
this
run
manually
and
set
up
everything
so
it'll
be
good
to
you
know,
get
unblocked
on
this
and
get
that
time
back.
So
we
can
look
at
other
things.
A
Okay,
all
right
now,
okay
from
this
stuff,
let's
see
what
do
we
have
left?
We
had
a
bunch
of
stuff
merged
and
then.
B
Yeah
that
so
the
last
block
you
see
is
some
more
scoping
I
did
for
for
V1.
So
this
is
the
the
issue.
I
was
talking
about
earlier,
where,
when
we
switched
branches,
sorry
repositories
there
wasn't
any.
There
is
an
enhancement
needed
in
the
weekly
craft
subcommand,
and
that
will
get
us
to
a
point
where
we
don't
have
to
compare
charts
like
this.
The
way
we
did
today,
we
will
get
us
a
single
graph
from
a
date.
B
So
let's
say:
if
we
decide
that
the
past
two
releases
was
on
January
1st
2022
and
we
want
data
from
there,
so
the
graph
tool
should
accept
a
date
from
which
it
will
plot
all
the
data,
and
this
issue
is
for
that
I
will.
This
is
what
I
will
work
on
and
we
need
to
get
this
first.
A
B
And
if
you
go
back
to
the
main
tracking
issue,
okay,
so
once
we
get
that,
then
I
think
we
need
an
ability
to
render
this
HTML.
That
is
created
by
the
first
issue,
and
then
we
need
an
ability
to
screenshot
that
HTML
and
save
it
in
the
repository
once
we
have
it
in
the
repository,
we
can
put
it
in
the
badge
like
how
the
the
CI
CI
has
it
CI
deposit.
A
B
I
yeah,
that's
something
we'll
have
to
figure
out.
I
hadn't,
like
I,
had
thought
that
we
needed
a
separate
repo,
but
I
need
to
still
dig
into
the
all
the
GitHub
notice,
like
all
the
GitHub
settings,
to
see
if
it
is
like
we
can
render
that
HTML
or
not,
but
for
the
third
one
we
definitely
need.
B
C
B
If
you
get
some
Cycles
and
if
you
want
to
take
a
look
at
second
or
third
like
I,
imagine
that
will
be
the
only
thing
that
is
needed
and
I
think
there
is
one
more
when
we
have
all
of
this,
we
will
need
to
put
it
in
periodic
jaw,
but
I've
not
scoped
that
out.
Yet
there
is
a
prerequisite
to
that,
which
is
this.
Repository
needs
to
be
configured
with
raw,
so.
A
A
That,
okay,
we
need
this
one.
This
is
a
keyword,
okay!
Well,
we
need
to
get
this
in
ASAP
because
the
branch
is
like
next
week,
so
this
needs
to
be
done.
First,
okay!
Well,
it's
already
approved,
but
I
don't
know.
What's
going
on
with
the
testing,
you
might
need
to
rebase
this
I,
don't
know
what
is
going
on
here.
A
A
Yeah,
this,
it's
not
so
just
rebase
and
I-
think
that'll
resolve
this
and
I'll
prove
this
and
we'll
get
this
we'll
get
this
in
before
next
week,
and
that
should
that
should
give
us
enough
time.
So
then,
okay,
then
the
rest
of
this.
This
leaves
us
a
month,
it's
actually
less
than
a
month
because
of
holidays
and
stuff.
So
it's
about
three
weeks.
So
is
this
achievable
in
three
weeks?
So
this
would
the
end
result
of
this
would
be.
A
Be
the
past
eight
months:
okay,
yeah,
okay,
and
that's
what
we
say
and
may
we
maybe
what
we
do
is
we
just
do
it
as
a
release?
Note?
Maybe
we
don't
need
this
stuff
like
like?
Maybe
we
prioritize
just
getting
this
one,
and
then
we
put
this
in
a
release
note
so,
okay,
then
this
would
need
to
be
as
a
well
actually
I'm.
B
A
A
Yeah
I
agree
with
you,
I'm
saying
this
PR
sorry
I
had
a
highlight
that
was
a
little
misleading
I
mean
when
we
say
when
we're
saying
we
want
to
publish
that
our
P95
is
whatever
for
the
past
eight
months,
and
this
is
what
we're
going
to
publish
it
as
for
V1,
how?
How
would
we
do
that
right?
Like
I,
was
saying
a
release
note,
but
at
least
that
we
would
need
a
pull
request
to
hubercuber.
B
B
So
if,
let's
say
automation
is
taking
too
much
of
time,
then
we
I
will
still
manually
screenshot
it
upload
it
to
the
I,
mean
you
push
it
and
we
can
have
the
badge
for
that
release
and
have
a
future
tracking
issue
to
automate.
It.
A
B
A
B
A
Yeah
or
see
this
is
where
I'm
this
is
where
I'm
struggling.
It's
like
it's
like
the
way
we
communicate
this
right.
It's
the
badge
is
good.
The
badge
is
important
right
people
show
up
to
the
repo
it's
in
get
I,
guess
that
could
happen
at
any
time.
Right,
it's
not
necessarily
well,
it
would
be
nice
to
be
in
the
B1
Branch,
but
the
the
other
part
of
it
is
that
I'm
thinking
is
like
how
do
we
spread
this
around
Beyond?
Just
the
badge?
Is
it
a
release?
Note?
A
Is
it
a
document
that
we
say
we
publish
in
the
V1
Branch?
You
know
in
the
cube,
cuber
docks
somewhere
and
we
have
at
least
though
it
says
this
is
the
P95
or
we
are
something
where
we
have
a
release.
Note
something
about
scale
for
V1.
You
know
what
I
mean
like
how
do
we?
How
do
we
do
that
Beyond?
Just
the
batch,
the
badge
is
important,
but
also
we
want
to
get
the
message
out
there
into
you
know
all
the
channels
so
that
this
can,
you
know,
be
promoted
in
other
ways.
So.
B
That's
where
I
think
I
need
a
little
bit
of
help
in
understanding
the
process
and
the
timeline.
So
let's
say
we
got
a
release
Branch
right
next
week
that
will
be
Cube
word:
release
B1
from
there
yeah
I
think
the
way
to
kubernetes
does
it?
Is
there
that
there
is
an
alpha
release
card?
Then
there
is
Alpha
two,
and
if
the
testing
on
that
those
delays
Cuts
go
well,
then
eventually
that
there
is
a
a
release
card.
So
my
question
is
from
the
branches
being
separated.
B
Is
there
still
a
process
for
bringing
important
bug,
fix,
slash
data
back
into
the
release,
and
if
the
answer
to
that
is
yes,
can
we
take
like?
Can
we
create
a
placeholder
for
that
data
and
fill
in
retro
fill
fill
that
in
like
during
the
time
of
release.
A
Yeah,
maybe
I
think
that
would
work
so
the
I
guess
it
would
be.
We
created
documents
and
the
badge
and
then
and
then
we
say
we're
gonna
we're
gonna
come
back
and
fill
it
in
before
July,
like
in
two
weeks.
B
That's
correct
because,
ideally,
we
would
also
like
to
have
the
the
data
from
the
release.
Candidates
right,
oh
but
you're,
saying
because
the
branches
will
cut
it
main-
will
no
longer
track
the
release
code.
So
the
only
short
of
us
getting
data
is
this
week
before
the
branches
before
the
branch
is
get
cut.
A
Yeah
they'll
be
bug
fixes,
but
for
the
most
part,
the
thing
that
we'd
want
to
have
is
for
like
next
week,
like
that's
kind
of
it's
actually
yeah
like
next
week
is
sort
of
is
going
to
be
our
deadline
more
or
more.
So
for
anything
that
we
want
to
merge
into
this
repo
and
anything
we
want
to
test
against
like
we're
testing,
it's
Maine,
we
have
one
more
week
to
do
it.
B
A
Yeah
they
should
be
there
still
yeah,
so
we
could
like
start
a
pull
request
against
the
V1
Branch
or
something,
and
then
it
should
triggering
okay.
B
Got
it
so
then,
that's
that
that
helps
us
right
so.
B
We
can
scrape
the
data
before
the
branches
get
cut,
so
that
will
be
the
the
Baseline
for
this
release.
For
all
the
bug
fixes,
we
can
make
sure
that
regression
like
the
the
Sig
performance
job
research,
is
not
regressing
and
then
publish
the
scrape
data
from
this
week
as
B1.
A
So
then,
okay,
so
let's
take
the
data
all
right.
So
how
long
will
that
take
to
get
the
data
and
publish
it
done.
B
B
I
will
I
had
a
question
for
this
docs
art.
Shall
we
bring
this
up
for
discussion
in
YouTube
World
Community
call
on
where
exactly
would
would
be
a
good
place
to
you
know,
share
these
numbers
I
think.
A
I
I
think
people
are
going
to
defer
to
us
to
be
honest,
I
think
I
think
it's
going
to
be
yeah
I
mean
I,
think
it's
sort
of
people
are
gonna
defer
to
our
opinion.
I
I
think
this
is
like.
So
this
has
like
things
like
usage
things,
Cloud
internet
conformance,
debugging
right,
so
it
can
make
perfect
sense
to
have
performance
in
here.
I
think
it
I
think
it's
pretty
reasonable
like
this.
Is
these
aren't
features,
freeze,
guests,
FS,
local
storage,
right
I
mean
this
is
performance
would
fit
perfectly
right
here.
A
It's
just
yeah
I
mean
I.
Think
this
is
the
right
place.
I
think
what
we
got
to
figure
out
is
what's
going
to
go
in
it
I
think
it's
you
know.
We
have
I,
think
the
badge
makes
sense,
I
think
maybe
the
chart
a
chart
or
something
showing
the
last
eight
months
and
and
then
a
summary
and
that's
it
and
kind
of
I.
Think
that
would
do
it.
B
A
B
To
give
a
glimpse
of
how
performance
results
are
doing
across
time
right,
so
if
you
want
to
get,
if
you
want
to
get
a
live
numbers,
you
go
to
the
badge,
but
if
you
want
to
get
the
numbers
for
an
existing
cut,
release
or
crafts
for
that,
then
you
go
to
that.
Sig
Docs!
Sorry,
the
docs
folder
that
that
you
were
sharing
right.
So
that
will
be
a
snapshot
of
historical.
A
B
B
We
need
an
issue
to
like
create
a
pull
request:
placeholder
pull
request
where
okay
actually
do
we
need
a
placeholder,
or
can
we
just
put
data.
A
B
A
No,
no,
that
would
be
let's
see
so
well.
This
is
strictly
because
of
okay
because
of
the
release.
Change
is
going
to
overlap
here,
so
the
release
Cadence
changed
so
I
think
so
this
is
the
monthly.
So
what
does
this
push?
It
put
us
at
zero,
five,
nine
March,
so
we're
looking
at
this
is
three
months
so
up
until
so.
Let's
why
don't?
We
just
do
the
last
three
releases
again,
but
it's
not
going
to
be
the
full,
the
full
amount
of
time
that
this
would
normally
take,
which
would
be
eight
months.
A
A
That's
eight
months
right:
six
and
yeah
October
I'd
say
a
month:
okay,
so
zero,
five,
eight
October
12th
right
here,
I
think
that's
our
answer!
That's
there.
So
this
is
the
right
Cadence,
then
so
the
that's
two
releases
that
makes
sense:
zero:
five,
nine,
zero,
five,
eight,
okay,
so
zero
five,
a
must
have
been
our
first
release
on
the
new
Cadence.
Then
no
I
wasn't.
This
was
your
five
nine
once,
but
it
just
happened
to
work.
Okay.
A
B
Yeah
I
mean
the
timeline,
doesn't
matter
right,
so
what
we
are
going
to
say
is
just
that
this
is
past
three
releases,
how
we
did
five,
eight
five,
nine
and
Viva
that
study
yeah
and
the
date
I
will
pull
the
data
from
is
October
12th.
A
Should
we
post
so
like
we
had
some
interesting
ones
which
was
like
okay
as
I'm
thinking,
patch
and
stuff,
but
people
are
going
to
misunderstand
that
I
think
I'm,
afraid
they're
gonna,
misunderstand
that,
like
we
don't
have
enough
data
on
scale
to
say
like
patch
change.
This
is
gonna
mean
this
for
you,
we
just
we're.
Just
gonna
show
the
data
and
people
are
gonna,
see
it
and
I,
don't
know
what
they're
gonna
say:
maybe
we
just
P95
and
what
was
the
other
one?
We
said
P95
and
P.
B
Yeah
yeah
so
I
think
yeah,
that's
an
interesting
question
so
another
we
can
so
two
thoughts.
One
is
that
yes,
P95
and
b50
makes
sense,
and
a
second
thought
that
comes
to
my
mind
is
when,
when
cube
word,
does
a
release?
Does
it
usually
do
some
kind
of
blog,
post
or
call
out
domain
features
because
I.
A
Know
someone
there
should
be
I
I
think
there
is,
but
this
one
this
one
there
definitely
will
be
I,
don't
know
where
the
website
is.
But
let's.
A
Two,
maybe
we
do
one,
maybe
maybe
we
should
do
one
so
here
we
go
so
there's
it
says
the
release
changes.
Okay.
Maybe
we
should
do
one
for
performance
at
the
very
least.
If
there's
not
I,
think
there
will
be
one,
but
if
there
is
not,
maybe
we
should
do
one
for
performance
or
at
least
contribute
to
the
V1.
If
there
isn't
one
planned
yeah.
B
Yeah
that's
what
I
was
thinking
is
that
we
can
have
all
the
crafts
in
that
document.
Sorry,
what
I
mean
is
that
we
looked
at
the
docs
folder
right.
We
can
have
more
detailed
graphs
in
there,
but
then
supplement
that
with
a
blog
post.
That
explains
some
of
it
right.
B
A
So
let
me
see-
let's
take
a
look
at
this
for
a
second
I'm,
going
to
go
to
your
previous
ones
that
are
rendered.
Let's
find
one
there.
A
You
go
okay,
so
just
looking
at
one
of
these
for
a
second,
so
so
P95
I
I
think
we
got
to
do
like
the
so
a
mix
of
so
some
of
the
ones
we
think
are
most
important,
so
P95
p50
and
once
I
look
consistent
too,
like
this
one's
I,
don't
know
how
to
something
that
we
can
kind
of
show
like
people
can
read
it
and
take
a
glance
and
be
like
okay,
I
understand.
Okay,
there
was
a
change
here
and
maybe
some
ones
that
are
constant,
like
so
P95
p50,
maybe
some
well.
B
That
next
week
we
will
have
more
accurate
data
of
what
changed
yeah.
So
we
can
decide
that
once
everything
is
in
a
single
part
like
which
ones
to
pick
yeah.
A
Here's
a
good
one,
so
patch
counts
for
VMI.
This
is
like
this
is
the
one
we
want
to
see.
This
is
a
great
for
this
is
made
Taylor
Made
for
a
blog
post
right
here.
Yes,
okay,
yeah,
all
right!
That's
fine!
Let's
do
that!
We'll
decide
next
week,
all
right,
so
so
in
terms
of
priorities.
Let's
do
so
we
got
to
get.
We
got
to
get
this
Qbert
quivert
change
in
it's
centered
here,
let's
where's
our.
A
A
We
can
just
stub
it
out
and
and
like
if
you
don't,
if
we
don't
have
data
for
it,
just
right.
If
you
want
to
wait
for
the
data
and
we'll
just
let
let's
tag
fobby
and
Dave
and
others
just
let
them
know
our
intention
and
then
we
need
to
talk
to
Daniel
and
about
getting
this
I
think
this
is
just
some
when
you
need
a
minor
change
to
get
this
all
in
all
right,
a
bunch
of
things,
and
then
this
is
the
other
one.
B
B
I
will
take
I'll,
take
three
action
items,
there's
creating
a
placeholder
PR
for
data
to
be
filled
in
next
week
and
rebasing.
This
yeah.
A
Okay,
this
one
all
right.
Let
me
talk
to
Daniel,
then
I,
don't
know
what
needs
to
be
changed
here,
though.
I
have
not
I,
don't
know
what
the
command
is,
but
you
can
tell
me
and
I
can
I
can
push
this
through.
B
So
for
this
one,
so
two
things
we
need
to
make
sure
is:
that
is
the
script
present
in
that.
So
can
you
open
the
files
yeah.
A
A
So
three
action
items:
you
got
that.
Can
we
base
this
the
open,
the
docs
stub
and
then
this
one.
B
A
Okay
sounds
good.
All
right,
I
do
want
to
review.
We
got
12
minutes.
I
wanted
to
look
at
this
I
made
a
bunch
of
changes
to
this
so
very
quickly,
so
I've
written
a
few
I've
kind
of
formalized
this
a
little
bit
more
I,
have
not
completed
my
research
and,
of
course
said
it.
It's
really
an
design
document
state
but
I
think
it's
better
communicates
what's
going
on
here.
A
So
oh
tldr
is
I'll,
sell,
riddle
summary
about
this
and
about
our
thesis
and
and
how
we
want
to
record
timestamps,
and
then
it
goes
in
the
requirements.
Basically,
this
is
what
we
want.
We
want
a
granule
per
object
metric
in
Prometheus
and
then
we
also
want
a
granular
per
object
metric
on
the
object.
I
have
this
sp1
I
just
think.
If
it's
P0
it's
going
to
distract
people,
so
we
will
leave
this
as
I
think
it's.
A
Finally,
this
is
P1
and
so
use
cases,
for
this
is
just
we
want
to
isolate
bottlenecks,
and
we
want
to
see
our
intent
creation,
that's
pretty
straightforward,
so
high
level,
so
kind
of
the
way
approach
I'm
taking
with
this
is
cuber.
The
reason
this
approach
makes
a
lot
of
sense
for
cubert
is
because,
like
we
have
an
API
server,
we
have
controllers
and
we
have
an
API
that
we
serve,
but
it's
sort
of
it's
you
know
we
can't
even
call
keyword
clustered
on.
A
Excuse
me
a
cluster
add-on,
but
it's
sort
of
more
than
that.
It's
it's
really
keyword
is
no
different
than
any
CRT
right.
It's
just
the
controller
right
we're
using
the
kubernetes
API
server,
every
crd,
every
API.
Technically,
you
can
look
at
it
as
just
a
sort
of
the
guest
API,
that's
in
the
cluster,
serving
whatever
requests
from
the
user,
and
so
the
thesis
is
that
really
kubert's
perspectives
on
this
is
it
should
be
no
different
than
really
any
api's
perspective
and
that
any
API
wants
to
be
efficient.
A
He
wants
to
understand
its
effect
on
the
kubernetes
API
server.
It
wants
to
understand,
on
the
effect
on
the
total
scalability
performance
in
the
cluster
and
so
having
this
sort
of
granularity
when
looking
at
how
it
is
being
used,
so
the
different
phases
it
goes
through,
how
quickly
it
goes
through
them
and
it's
overall
effect
will
be
important
for
actually
viewing
and
discovering
issues,
and
so
that's
the
general
idea
that
I'm
proposing
here
is
that
this
is
the
idea
of
a
phase
transition.
A
Timestamp
is
something
that
can
be
used
anywhere
for
any
crd,
any
project
that
wants
to
use
the
kubernetes
API
server,
in
addition
to
the
core
apis
and,
and
so
that
would
be
helpful
for
viewing
these
General.
These
granular
or
granular,
look
at
phases
and
I
am
being
very
specific,
with
phases
and
I
know
that
people
don't
like
phases,
maybe
in
kubernetes
or
they're,
looking
to
go
away
from
it.
A
That's
fine,
like
I,
I,
think
it's
just
the
way
it
is
now
so
nothing
communicating
that
way
would
make
the
most
sense
when
talking
about
it
and
maybe
a
changes
in
the
future
and
it's
fine.
This
can
adapt
in
any
way
that
it
changes
so
I
mean
that's
pretty
much
the
perspective
that
I
changed.
It
is
that
it's
we're
sort
of
talking
about
this
from
we're.
Looking
at
this
from
just
a
slightly
different
perspective,.
B
Yeah
I
I
think
the
gist
of
how
this
helps
make
sense.
So
we
can,
you
know,
rely
on
the
community
to
give
us
feedback
makes
sense.
The
only
thing
I
had
a
question
on
is:
if
you
can
go
up
at
the
yeah,
so
here
right,
I
think
it
would
be
good
to
have
a
discussion
on
what
kind
of
benefits
does
up
a
power
object
metric
have
because,
at
the
end
of
the
day
we
will
be
able
we
will
be
aggregating.
B
Those
when
dealing
with
thousands
of
objects
right,
so
maybe
at
Prometheus
level
or
object,
doesn't
make
a
lot
of
sense
as
long
as
each
object
is
contributing
to
aggregation.
B
No,
what
I'm
suggesting
is
that,
let's
say
for
those
use
case
right
if
you
want
to
isolate
a
bottleneck
in
let's
say
called
creation
time.
So
if
you,
if
you
have
all
the
tools
in
the
world
right
like
how
would
you
go
about
it,
so
you
look
at
the
Pod
data
in
inside
of
the
Pod
data.
What
exactly
is
taking
more
time,
so
you
would
look,
let's
say
at
PVC
or
network
data
right,
but
that's
not
the
starting
point
of
your
debugging
Story.
The
starting
point
is
an
aggregation.
B
A
Let
me
see
so
I
think
what
you're
saying
is
that
this
is
the
the
granularity
allows
us
to
expose
the
use
cases
where,
for
example,
I,
could
isolate
the
problem
to
one
piece
of
Hardware
in
my
entire
cluster,
whereas
this
would
not
be
captured.
If
we
just
did
aggregation.
B
Well,
I'm
I'm,
saying
the
the
other
way
around
yeah,
it's
good
to
look
at
aggregation
where
each
object
contributes
things
so,
for
example,
in
in
Prometheus
right.
Let's
say
you
have
data
for
each
part
in
Prometheus.
That
would
be
too
much
data
going
into
Prometheus
right
versus.
If
you
had
end-to-end
creation
to
running
for
as
actually.
A
B
But
what
does
not
exist
is
breaking
down
that
Aggregate
and
what
is
contributing
to
regression
in
that
right.
A
B
B
And
from
that
aggregation,
then
we
can
chase
the
performance
problem
of
of
Single
part
or
something
so.
Single
port
data
comes
next
I.
Think
from
an
aggregation
point
of
view,
we
need
more
insights
into
what
regresses
that
aggregate.
A
A
Like
that,
and
then
let
me
do
this
one
as
user
I
want
to
identify,
if
a
is
a
let's
say,
or
this
faulty
piece
of
hardware.
A
Something
like
that,
that's
another
one.
You
can
go
on
forever,
like
this
I
mean
this
is
this
illustrates
the
point
right.
B
Yeah
and
so
I
think
a
concrete
example
of
this
would
be
right
there.
As
you
said,
the
aggregation
of
the
creation
to
running
at
cubelet
level
is
present,
but
if
that
goes
bad,
let's
say
tubulatory
crashes
in
in
that
creation
to
running
time,
is
there
data
to
point
at
PVC
or
network
or
or
any
of
the
other
sub
tasks
that
cubelet
performs?
Now
that
doesn't
have
to
be
a
power
object
statistics
as
long
as
we
can
at
least
pinpoint
a
pattern,
and
then
we
can
change
a
power
object.
A
Okay,
I
also
want
to
say
for
here:
I
want
to
okay
I'll.
Think
about
that.
So
here's
here's
what
I'm
thinking,
because
we
only
have
two
minutes
left.
So
this
is
good
I.
Think
I
I've
not
gone
into
like
the
technicals
of
like
volumes
and
how
we
can
accomplish
this
and
I,
don't
even
know
if
I
want
to
just
yet,
because
I
mean
we
only
have
30
minutes
when
talking
with
the
let's
say
kubernetes
group
anyway,
so
I
don't
think
we
have
a
chance
to.
A
But
basically
what
I
want
to
do
is
to
just
review
this
stuff
and
take
questions.
I,
think
that's
what
I'm
gonna
do
and
then
I
want
to
take
this
out
and
we're
gonna
we'll
take
this
up
in
the
future.
The
future
one
and
I'll
just
leave
the
existing
metrics
in
place.
I.
Think
that
illustrates
the
point
this
is
going
to
be
our
requirements
documented,
we'll
see
what
we
get
I.
Think
that's
yeah,
that's
gonna
be
useful
I,
so
I
can't
I
can't
add
a
suggestion
in
the
document.
A
For
some
reason,
I
still
can't
do
it
so
I
either.
Maybe
you
can
help
me,
do
it
or
I
can
ask
we'll
check
and
he
can
let
them
know
we
can
do
another.
I
can
put
a
stamp
out
of
place
for
us
in
the
document
or
something.
B
A
Yeah
I'll
send
it:
oh
it's
in
the
it's
in
the
notes:
okay,
yeah
I'll,
get
it
yeah
right
here,
yeah
just
add
it
in
there
and
say:
like
I'll,
do
the
requirements?
Let
me
move
this
out.
I'm
gonna,
take
the
volume
stuff
and
let
me
pull
this
out.
I'm.
C
A
Yeah
go
ahead
and
yeah
go
ahead
and
share
it
in
there
and
we'll
I'll
take
some
time
on.
C
B
Yeah,
do
you
want
me
to
phrase
this
as
enhancements
to
Performance
metrics.
A
I
think
they
kind
of
know
us
as
like
the
keyword,
six
scale
group.
We
say:
kubert,
sixth
scale,
performance,
metrics,
enhancement,
sure
something
like
that
sounds
good.
I.
Think
we'll
know
what
we're
talking
about.