►
From YouTube: 2017-02-02 Kubernetes SIG Scaling - Weekly Meeting
Description
2017-02-02 Kubernetes SIG Scaling - Weekly Meeting
A
Just
stuck
something
in
the
sixth
scale
check
on
slack
room,
which
I
should
add
to
the
notes.
This
was
some
things
that
came
up
at
the
p.m.
meeting,
which
Joe
no
I
think
both
managed
to
attend.
So
do
you
not
have
stood
on
the
second
one
as
well?
Joe
I,
don't
recall
that
was
the
lemon
moved
to
Monday
morning
early
over
the
weekend,
so
yeah
anyway.
A
B
A
A
B
C
C
I
wanted
to
have
a
placeholder,
because
at
the
time
this
is
period
there
actually
was
no
placeholder
for
the
overall
scale
goal.
So
I
mean
we
originally
when
this
existed.
We
that
we
want
something
in
defeat.
Your
opponent
says
our
scale
goal
are
so
that
people
can
tie
the
individual
items
in
a
particular
release
to
it
and
then
the
visible
okay.
We
said
that
we
wanted
to
hit
this
goal,
so
that's
the
primary
reason
that
this
feature
exists.
A
A
B
Totally
reasonable
and
like
as
Merrick
and
I
chatted
before
this
release
and
all
honesty
to
try
and
make
this
release
have
more
metrics
of
goodness.
We
actually
haven't
laid
down
those
metrics
yet
because
we
both
been
sidelined
with
other
issues,
Rizzo
I
think
in
the
future,
we're
going
to
start
to
fill
out
a
tree
for
how
we
or
is
just
a
table
that
defines
what
are
the
metrics
that
we're
measuring.
Besides
the
standard
to
you
know
to
define
this
goals
for
the
future
right.
A
Yeah
I
think
so.
I
keep
just
point
there
and
chat
about.
We
should
only
be
putting
things
in
that
we
can
measure
I,
think
it's
to
the
degree,
at
least
that
this
kind
of
thicket
informs
the
release,
notes
and
other
things
the
users
will
come
along
and
see.
It's
probably
worth
a
little
bit
of
explanation.
Make
sure
that
people
understand
when
they're
reading
it
why
it's
it
may
be
more
limited
than
you
would
expect
coming
and
taking
a
look
at
it.
A
B
I,
don't
know
what
the
progress
is,
I
mean
we
like
white,
tic
and
I
have
already
enabled
everything,
and
it's
all
running
right
now
with
sp3
and
Ji.
Young
has
put
the
doc
up
for
people
to
look
at
and
we
need
to.
Cluster
Ops
has
apparently
looked
at
it,
but
I
haven't
talked
with
Rob
about
what
his
thoughts
are
on
it.
So.
A
B
A
Actually
gone
and
looked
through
this
one
recently,
but
when
we
were
scanning
through
it
during
the
meeting
there
were
some
required
checkboxes
left,
unchecked
I.
Think
part
of
the
role
of
the
meeting
here
was
to
encourage
the
people
that
were
working
on
it
to
get
through
the
checklist
that
they
needed
to
do
that.
The.
B
D
For
not
having
a
specific
link
to
refer
to
here,
but
I
feel
like
I
saw
something
come
across,
maybe
straight
testing,
or
not
one
of
the
notification
things
a
request
for
automated
sed
to
to
exit
III
upgrade
testing.
You
have
any
idea.
If
such
a
thing
exists
or
is
possible
within
the
one
six
timeframe
they
have
the.
B
So
I
believe
we
have
documented
inside
of
his
document
that
the
recommended
practice
is
to
snapshot
attempt
to
migrate
forwards.
If
you
fail
for
any
reason,
use
the
snapshot
or
still
stay
in
b2
mode,
is
it
automated?
No,
it
has
it
there's
a
explicit
section
in
the
document
that
specifies
that
you
need
to
do
like,
and
how
did
you
wind
up
planning
for
this
because
it's
a
you
know
be
careful,
we're
going
to
migrate.
All
your
data
type
saying
rule
yes,.
D
That's
sort
of
a
possible
I'm
just
pointing
out
that,
like
upgrade
testing,
was
a
real
big
pain
point.
So
the
last
release
cycle
and
folks
are
trying
to
get
a
little
more
pre-emptive
about
this,
and
so
just
moving
from
FTD
to
that
CD
3
sounds
like
an
upgrade,
so
it
might
get
caught
up
in
this
whirlwind.
Okay,.
B
Already
have
the
scripts
for
it
so
at
least
on
their
size
right,
so
the
community
is
probably
going
to
have
community,
which
is
like
this
broad
ecosystem
of
everything.
That's
brother,
it's
probably
going
to
have
to
take
the
jion's
document
and
turn
it
into
some
type
of
playbook
scripts
chef,
something
puppet
something.
What
you
know
pick
your
pick.
Your
favorite
configuration
management
system
and
you
know
have
to
handle
it
downstream,
but
it's
all
documented
in
that
main
dock.
D
A
Do
you
say,
let
me
move
in
by
acknowledging
that
the
checklists
on
the
features
have
been
well.
You
know
I've
been
sort
of
counting
the
drama
bank.
If
we
have
checklist,
we
should
use
them
for
tourism.
But
if
you
happy,
we
should
use
them
and
I'm
wondering
if
there's
a
upgrade
testing
upgrade
procedure
checkbox
somewhere,
they
needed
to
be
added
to
the
standard
list,
whether
that's
a
match.
D
It's
possible.
Let
me
see
if
I
can
dig
up
the
link
in
like
the
next
hour
here,
because
it's
I
got
a
lot
to
sift
through,
but
I'll
put
it
in
the
meeting.
It's
and
we
can
chat
about
this
later
at
work
because
I
like
I'm
kind
of
unclear,
whether
we're
using
the
new
feature,
checkbox
list
or
the
old
feature
checkbox
list,
and
if
we
should
be
suggesting
anything.
E
A
A
A
A
E
E
So
we
want
to
actual
our
data
first
right,
and
so
there
seems
to
be
a
few
main
sources
of
data
to
collect
from
will
just
close,
your
ops
and
the
capes
community,
internals
or
PR
testing,
so
I'm
going
to
share
who
to
contact
for
this
information.
If
you
guys
have
a
clue,
that'd
be
great
and
I'm
just
blending
right
direction.
E
There,
social
ops,
individuals,
Aaron,
was
able
to
find
some
information
about
clutter,
ops
or
whatever
their
career
clusters
are
including
some
information
from
February
16th
2016,
which
actually
used
avoid
any
clear
my
template.
So
we
have
several
questions:
we're
not
going
to
answer
first
to
get
an
idea
of
like
what
it
was
finally
trying
to
do,
based
on
what
we
know
currently
and
how
we
collect
all
the
data
and
get
something
useful
back
to
the
community.
E
So
and
Bob
has
ransom
comes
here.
He
could
actually
add
to
these
columns
of
great
if
I
can
go
and
expand
discreet.
A
little
bit
succeed.
Oh
sorry,
about
this,
all
right
so
like
what
what
are
the
metrics
that
we're
trying
to
evaluate
against
gathering
that
data?
How
do
we
go
ahead
and
know
go
to
the
community?
Tell
them
like
hey.
Can
you
give
us
some
information,
there's
questions
about
what
she
anonymous.
What
could
information?
We
should
collect
what
things
we
can
play
any.
How
do
we
could
solve
that
into
some
package?
E
D
A
A
In
other
words,
kind
of
all
of
the
pattern
like
we
have
for
the
quick,
the
feature
Rico,
where
there's
a
standard
checklist
of
items
and
the
lady
publish
information,
is,
you
hope,
initially,
don't
basically
fill
out
a
form
in
the
the
other
data
collection
items
are
all
and
I
mean
that
way.
It's
explicit
that
you
provided
your
data.
It's
you
need
to
make
sure
that
everybody
can
see
it.
A
You
would
certainly
be
able
to
go
and
run
some
scraping
on
top
of
that,
if
you
wanted
to
aside
from
that,
you
know
we're
starting
to
look
at
things
like
shared
Google
sheets
or
some
database
somewhere
with
you
know
various
access
and
permission
issues,
at
least
at
least
posting
them
to
get
help.
We
we
keep
fairly
straightforward
yeah.
E
See
it's
I'm
open
to
formal
way
right,
some,
some
way
that
people
can
go
see
you
can
a
follow
through
cetera,
and
if
you
like
it,
we
got
it
as
well
right.
So
to
follow
up
on
that,
if
possible,
so
I
propose
are
there's
two
ways
to
things
that
we
need
to
collect.
One
is
with
the
global
properties
of
a
cluster
and
the
other
one
will
be
the
know,
specific
properties
of
a
cluster.
E
B
E
B
Be
like
this
is
a
dead
flow
for
doing
this
type
of
evaluation
of
your
cluster
like
step.
Zero
would
be,
you
know,
get
the
properties
of
the
cluster
check,
conformance
right,
that's
the
zero
order
step.
Then
the
next
step
would
be
like
do
these
performance
set
of
performance
benchmarks
and
compare
them
against
what
upstream
numbers
are
given
their
profiles
right
run
it
on.
Oh.
A
D
B
Can
collect
some
of
the
information
to
have
access
to
all
the
cluster
data
and
all
the
data
is
introspective
enough
like
it
can
actually
go
to
the
nodes,
and
it
actually
does
go
to
the
notes
and
gather
information.
So
you
know
it
can
gather
node
wise
data
is
the
question
of
whether
or
not
the
couplet
API
exposes
enough
information
for
it
to
gather
it
today
gather
some
of
it.
It
doesn't
gather
out
there
like
I.
Have
you
know.
E
B
A
D
B1
well,
the
direction
we
want
a
head.
My
my
proposal
would
be
that
we
send
somebody
over
to
cluster
ops
and
get
cluster
ops
motivated
to
do
another
survey.
The
survey
that
I
collected
from
February
of
last
year
is
way
way
out
of
date,
and
maybe
it's
time
that
we
do
sort
of
an
annual,
hey
who's,
running
kubernetes
and
what's
requested
or.
A
That's
fine,
but
let
me
let
me
finish
my.
Let
me
finish
my
question,
which
is
that's
one
way
to
do
it.
The
other
way
to
do.
It
is
more
what
Tim
was
suggesting,
which
is
have
some
bits
of
stuff
that
you
can
run
on
the
cluster
that
can
gather
the
data
and
publish
it
back.
It's
on
form,
I,
would
say
obviously
to
the
human
in
the
loop,
but
I
think
on
a
production
cluster
someone's
happy
to
run
a
little
bit
of
code.
That
says:
hey
what
the
hosta
look
like.
A
A
B
There's
enough
capabilities
in
the
intent
testing
infrastructure
for
folks
to
grab
topology
information
as
well
as
some
configuration
data,
but
not
all
of
it
right
like
there's,
no
way
I'm
going
to
be
able
to
figure
out
that
your
Dockers
specifying
overlay
versus
you
know,
advice
matter
and
stuff
like
that.
But
how
many
know
this
is
the
master
example?
Yes,
you
can
you
can
get
out
idea,
I
guess.
B
A
I
think
I
think
this
be.
The
number
of
nodes
in
the
cluster
is
only
one
thing
of
interest.
I
think
the
day
we've
spent
a
lot
more
energy
on
I
know
is
wondering
how
people
have
their
control
point
configure
more
so
than
yeah
of
note.
So
you
know
Tim
off
off
popular
head
aside
from
like
the
STD
stuff.
Do
you
know
how
much
a
lot
together.
E
B
The
same
way,
that's
the
biggest
problem,
there's
like
everybody
and
their
brother
has
their
own
way
of
deployed
so
because
of
that,
unless
he
there's
no
unifying
principle
for
you
to
gather
cluster
state
like
that's,
that's
where
the
configuration
management
tool
comes
into
play,
you
have
to
like
create
a
playbook
that
says
you
know
rather
grab
this
grab
the
data
of
what
you've
configured
and
you
know,
punch
it
all
back
together
into
some
reports
and
I've
expected
us
on
our
side,
because
I
need
that
data
from
time
to
time.
What
what.
E
Is
that
just
scroll
down
so
I'm
not
entirely
sure?
If
what
we're
talking
about
here
is
a
little
bit
outside
of
scope
of
what
I
had
proposed,
we
can
definitely
discuss
that.
You
want
to
look
at
it
and,
as
far
as
the
update,
more
individuals
involved,
more
people
giving
me
questions
about
what
contempt
is.
E
There
seems
to
be
some
sort
of
disagreement
about
what
the
scope
should
need
for
this,
because
some
people
think
that
oh,
oh
mod,
all
components
should
be
able
to
provide
the
slash,
configs
end
point,
and
there
are
other
individuals
such
as
Daniel
long
life,
to
suggest
that
no,
that
should
not
be
the
case.
It
only
is
some
component
to
provide
that
so
more
individuals
have
some
questions
and
then
the
same
individuals
from
prior
adding
more
question
on
top
of
what
they
don't.
Really
we
done.
We
definitely
stepped.
D
In
the
middle
of
a
yeah,
it's
sorry
didn't
totally
stalled
out.
We've
had
you
know
strong
opinions
from
debts
to
K
and
B
grants
and
lava
lamps,
and
it's
just
kind
of
not
going
anywhere
so
they're
they're
two
prongs
here
right,
one.
We
had
sort
of
a
proof-of-concept
TR
for
how
you
could
maybe
do
this
for
tubulin
and
then
the
second
was
the
proposal
for
how
we
might
want
to
do
this
and
yes,
I,
do
see
a
high
degree
of
overlap
or
like
a
standard
way
of
describing
austere.
D
D
A
F
D
Only
be
useful,
I
agree
that
this
covers
an
introspection
we
need
to
add,
but
I
don't
even
like.
Can
we
just
see
what
the
state
of
cluster
data
collection
is
today
like?
How
much
can
we
get
and
what
are
we
missing
and
then
we
can
start
to
figure
out
where
we
need
to
add
intersection.
You
know
I
assume
every
we
make
this.
What
we.
E
F
E
Visibility,
we
can
definitely
collect
it
from
their
components
themselves.
Probably
don't
have
the
latest
orders
or
price
shittin,
so
that's
kind
of
Brian's
point
of
view,
but
again
it
may
be
that
we
component
should
have
by
themselves.
Supplier
is
kind
of
data
into
the
API
server
connectors
explosives.
A
Well,
I
still
think
I
mean
err.
Our
mission
here
in
the
short
run
student
that
weekend
minimal
impacts,
and
it's
not
going
to
be
everything
yep,
so
missus
comfortable
what
that
means
it
is
destined.
It's
definitely
been
interesting
how
much
this
topic,
kids,
but
sort
of
irritated,
varies
nerves
all
around
the
planet.
In.
F
Want
to
say,
I
think
one
of
the
missing
pieces
here
is
that
we
just
don't
have
you
know
a
uniform
identity
system
for
things
that
are
not
the
API
server
right.
So,
like
we
see
this
in
terms
of
the
way
we're
locking
down
communication
from
the
API
server
to
the
node,
you
essentially
need
to
do
certificate
route
right.