►
From YouTube: 2017-05-18 Kubernetes SIG Scaling - Weekly Meeting
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
B
A
All
right
one
one
thing:
one
thing
for
today
did
you're
gonna
take
notes.
One
thing
is
I
wanted
to
report
on
a
report,
get
some
feedback
on
the
cluster
config
sizing
for
scalability
testing
conversations
I've
been
having
and
actually
have
a
couple
of
slides
I
wanted
to
show
any
any
other
business
decide
that
so.
C
A
That's
still
work,
okay,
so
my
name
is
claimer
yeah,
I,
know
PowerPoint,
but
I
needed
to
draw
pictures
and
I
already
have
some
materials
built.
So
I'll
get
this
into
a
different
dock
format,
but
this
was
the
most
efficient
way
to
see
things.
Okay.
So
just
starting
point
here.
This
is
the
basic
three
node
configuration
that
we're
talking
about
for
this
configuration.
I
think
that
the.
A
B
B
So
there
was
a
question
when
we
were
had
this
original
transition
period,
which
is
like
last
release
cycle,
where
we
uncovered
that
you
know,
there's
race
conditions
on
certain
scenarios
where
you
could
get
stale
data
and
that
should
no
longer
exist
with
the
latest
versions
of
NCP
and
the
I/o
performance
problems
which
are
in
3-1.
Also,
why
tech,
just
to
clarify
what
version
are
we
shipping
for
about
today?
Are
we
shipping
for
the
one
7
release,
I.
B
A
C
B
A
C
D
B
B
Three
one
three
one
anything
greater
than
three
one
no
longer
has
the
I/o
problem
on
coronaries,
okay,
and
the
reason
why
that's
so
bad
is
if
you're
running
something
that's
like.
If
you're
running
a
crap
version
of
I
up
limited
Amazon
account,
you
can
run
out
UPS.
You
know
if
you're
thrashing
a
CD
which
that
will
basically
cause
you
to
lose
meters
periodically
and
have
your
schedulers
and
controller
managers
bouncing
all
over
right.
D
D
It's
like
for
the
scalability
test,
I'm,
pretty
sure
we
just
end
up
a
single
master
with
everything
on
it
right,
so
that
wouldn't
necessarily
important
tree
problem.
But
if
you
want
to
make
sure
you
have
an
API
servers
up,
why
would
you
do
that?
Well,
our
scalability
testing
proves
it.
You
don't
have
to
do
that
for
scalability
reasons,
but
for
availability
reasons
to
get.
A
B
Yeah,
so
what
upstream
ships
with
and
qualifies
with
is
currently
it's
3
or
17
stones
at
krei
protect
my
major
that's
the
one
they've
tested,
but
the
client
side
is
already
the
client
side
tooling.
Inside
of
the
repository
has
already
been
enabled
for
3-1,
so
anybody
can
swap
in
you
know,
exit
III
one.
They
should
have
a
problem
right.
A
A
B
Be
honest,
I
have
never
met
besides
Google
in
the
cloud
I
had
never
met
anybody
who
runs
at
the
scale
of
a
thousand
plus
that
doesn't
have
a
5
node
environment.
So,
if
you're
doing
several
hundred
and
you
care
about
things,
three
is
common,
but
once
you
get
to
the
point
where
your
hat
you,
you
care
a
lot
about
availability,
it's
like
a
trade
off
availability
and
upgrade
ability,
I've,
never
seen
I've,
never
seen
in
the
wild.
Anyone
at
the
thousand
note
scale
that
doesn't
use
five.
So
do.
A
B
You
want
to
say
it
like
say
say:
these:
are
the
trade
offs
for
3,
&
5,
you
know,
I
would
say
like
I
would
say,
depending
upon
the
people
and
how
much
they
care
they
might
want
to
choose
a
or
b
I.
Don't
think
we
should,
which
is
your
space
when
you
say,
if
you
care
about
hi
Adele
better
than
one
right.
A
A
F
C
B
A
C
A
C
C
A
A
Is
we
had
a
meeting
with
some
of
the
AWS
folks
and
wanted
to
get
their
input
on
this
specifically
and
so
I
went
through
this
scenario
with
them
and
one
of
the
one
of
the
things
that
we
started
with
is
we
have
a
lot
of
experience
on
AWS,
so
what
we
did
was
try
to
pick
a
note
that
was
similar
to
the
scope
of
the
previous
one
and
say
well.
Why
don't
we
use
that
and
it's
not
the
biggest
thing
that
amazon
has.
A
But
it's
probably
you
know
upper-middle
it's
an
upper
middle
class
node,
you
might
say,
and
it's
a
c4
8x
large,
and
so
we
think
that's
a
good
starting
point.
Aws
folk
said
yeah
that
looks
about
right.
I
mean
as
long
as
we're
as
long
as
we
were
able
to
change
it
later.
If
it
looks
like
there's
an
issue
going
on,
the
AWS
team
is
also
willing
to
review.
So
when
we
make
a
little
bit
further
in
terms
of
documentation,
they're
happy
to
be
involved-
oh
great,
so
so
related
to.
A
A
So
this
is
a
little
bit
more
Amazon
specific.
We
can
comment
on
the
Google
case
a
bit
more,
but
the
the
notion
of
setting
the
two
SED
clusters
up
on
dedicated
I,
ops
volumes,
I
I,
guess
we
felt
in
the
Amazon
folks
both
felt
that
this
would
make
the
configuration
a
bit
more
predictable
and
easy
to
debug.
Can
you
question
real
quick?
Why
do
you
have
six
city
blocks
in
this
picture
because
you're
running
to
XE
d,
the
horizontal
layers
there
I
split
at
CD,
a
three
node
cluster
for
events
and
one
proof
cluster
state?
B
G
G
A
You
know
Tim
has
said
that
on
this
call
a
whole
bunch
of
times,
like
you
know,
probably
more
than
a
year
of
agreeing
with
that
point
so
and
I
think
we
would
view
it
the
same
way
in.
In
effect,
this
configuration
is
like
a
baby
step
in
that
direction,
to
try
to
separate
the
event
stream
into
something
else.
G
B
I
quote
by
closed
one
of
them,
which
is
the
web
hook
like
I,
wanted
to
add
a
web
hook
and
Daniel
pooh-poohed
it
and
every
time
I
try
to
like
mock
the
API
there.
You
know
so
territorial,
but
the
I
originally
wanted
to
have
a
very,
very
simple
web
hook,
similar
to
how
we
do
like
hooks
for
everything
else.
So
that
way,
just
completely
offload
it
from
the
main
system.
E
A
B
Because
you
answer
is
pretty
simple:
is
that
people
don't
if
you
don't,
have
a
certain
level
of
multi-tenancy
you
care
about,
as
you
just
create
another
cluster
right,
so
the
answer
that
most
people
have?
Is
they
don't
want
one
large,
multi-tenant
environment
all
the
time?
Is
it
just
create
multiple
clusters
right.
H
B
Are
suggesting
so
like
if
the
reason
for
splitting
events
on
a
very
large
cluster
is
that
you,
you
can
run
out
of
space.
Well,
especially
in
that
CDC
scenario,
you
could
have
run
out
of
key
space,
because
you
had
degenerative
cases
where
you
were
standing
so
many
events
we
just
fill
up
to
be
right
and
and
also
there's
a
strict
constraint
right.
So.
C
Yeah,
the
initial
was
a
reason
why
we
did
it.
It
was
basically
the
latency
and
throughput
of
xcd
like
to
add
CDs
to
separated
CDs
were
handling
it
much
better,
I'm,
not
sure
if
it's
still
the
case
with
3.0
or
3.1.
We
didn't
measure
it
if
we
can
get
to
read
of
it,
but
yeah
I'd,
like
that.
The
reason
that
we
mentioned
about,
like
filling
the
space,
is
potentially
yet
another
reason.
Yeah.
B
C
I
started
what
I
wanted
to
say:
I
think
it's
not
that
we
once
we
decide,
we
can't
change
it.
We
can
change
it
easily
or
relatively
easy,
so
I
think
yeah
start
eating
it
as
a
starting
point
and
like
potentially
revisited
in
like
a
quarter
or
something
if
that
doesn't
make
sense
appear
that
it
doesn't
make
sense.
It's
like
pickles
would
wake
yep.
A
There
was
my
pitch
for
doing
it
this
way:
okay,
so
provider
input.
So
we
got
some
input
from
AWS.
That's
great
I
reached
out
to
both
Gabe
and
Brendan
at
a
juror
and
ask
if
they're
interested
in
having
some
input
on
this
I
assume
they
can
find
somebody
on
your
team
that
would
be
interested
as
well.
I
haven't
actually
gotten
the
input
yet
but
I'm
putting
it
on
the
queue
of
things
to
work
on.
A
A
Maybe
this
may
roughly
be
to
say
review
anyway
and
then
a
question
which
is
I've
got
this
Google
Doc,
which
is
I've,
got,
who
I've
got
words
in
a
Google
Doc
and
pictures
in
PowerPoint
and
I
need
to
use
the
which
is
just
a
bit
of
that
didn't
work
for
me.
Where
should
I
put
this
should
I?
Do
we
are
on
the
goals
or
create
a
new
page
and
link
off,
or
what
do
you
think
was
it.
C
I
think
there
is
like,
or
we
started
some
some
like
page
in
16
community,
we
all
with
how
we
would
think
the
setup
should
like
how
we
would
think
with
SLO
should
be
defined.
So
I
think
there
is
a
place
for
like
setup,
crafter
setup.
How
should
caster
setup
look
like
there,
so
maybe
there
but
I
will
I
will
try
to
look
into
it
and
will
can
you
like?
Add
it
to
the
meeting
notes
and
other
comment
to
me.
I
would
try
to
create
and
links
to
you.
C
A
A
D
C
C
From
what
I
know
from
golfing,
they
will
decide
whoever
there
will
be
a
pusher
ease
or
not
either
this
week,
which
apparently
still
didn't
happen,
or
next
week
like
by
the
end
of
next
week.
There
should
be
a
decision
whether
there
will
be
a
putt
released
or
not
of
go,
and
if
there
will
be,
then
we
will
definitely
upgrade
to
new
batteries.
If
there
won't
be.
We
will
stick
to
this
pot
like
this
part.
B
C
D
B
Don't
like
distress
can't
do
that,
so
mr.
benders,
which
include
like,
but
not
excluding
core
OS
but
like
a
Debian
RedHat
variants,
can't
do
that
nope,
there's
very
strict
policies
and
guidelines
and
how
they
build
and
how
how
source
is
maintained
throughout
the
whole
process
in
the
build
tree
and
everything
else.
Super
strict.
F
B
C
D
Just
wanted
your
documentation
of
how
to
do
it.
The
pattern
here,
then,
that
includes
not
sound
like
needs
to
include
more
than
just
a
run
with
the
doctor.
I
spilled
process
design,
creates
authorize,
build
process
for
boxtroll
this
season.
I
didn't
realize
its
limitation
for
other
districts
with
that's.
Why
I
asked
I.
D
C
D
This
is
fact
that
if
we
go
to
the
Alliance
it
it
turns
out,
we
have
to
like
alter
the
code
base
that
compiled
with
it
sort
of
stuff.
Yes,
but
I.
Think
Kyrgyz
are
wrong,
I
think
like
Jeff
Grafton
or
somebody
expressing
it
just
on
the
test
in
Versailles
helping
this.
But
we
have
the
open
question:
how
do
we
maintain
a
a
tree
that
works
with
little
anxious,
while
still
allowing
the
main
development
trees
to
cover
those
details
against
the
vetted
version
of
villain,
right,
I?
Think.
C
D
If
you
get
a
girl,
I
gotta
know
like
look.
This
is
really
going
to
hold
up
the
humanities
release.
If
you
really
really
really
do
it
actually
sort
of
response
it,
because
you
really
really
really
have
to
get
go
earlier,
so
we're
not
in
this
situation.
I
just
want
to
make
sure
we
have
a
story
for
them.
Well,.
B
D
B
Nice,
if,
like
some
of
the
go
laying
people
getting,
this
is
the
largest
golang
project
that
exists
right.
So,
like
it'd,
be
nice
if
some
of
the
people
who
work
at
a
company
that
rhymes
with
frugal
kind
we're
possibly
to
invest
some
folks
that
could
that
help
write
the
language
spec
to
be
involved
in
the
project
so
that
they
can,
because
you
know,
everybody's
time
is
limited,
I,
don't
I,
don't
know
boy
text
gonna
have
enough
time
to
dig
through
every
issue
against
tip
alright.
That
occurs.