►
From YouTube: Kubernetes sig-aws 20181019
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
Hello:
everyone:
it
is
Friday
October
19th.
This
is
sick.
Aws
I
am
your
moderator.
For
the
day,
just
in
Santa
Barbara
I
work
at
Google.
We
have
a
couple
of
items
on
the
agenda.
The
agenda
should
be
in
the
chat,
that's
also
on
the
calendar.
Invite.
If
you
would
like
to
talk
about
something
it's
great
to
put
on
the
agenda,
so
we
can
be
sure
to
get
to
it
and
there
is
also
an
attendee
list.
A
If
you
would
like
to
add
yourself
to
that
list,
the
ideally
we
have
a
note-taker
I,
don't
have
no
one
wants
to
take
notes.
I
will
do
my
best
to
do
so
if
no
one
else
steps
up,
but
otherwise
I
suggest.
We
start
up
on
the
first
item
on
the
agenda,
which
is
Seth
an
s3
bucket
for
AWS.
Sorry,
it
was
instance,
type
JSON
file,
yeah.
B
So
this
came
up
before
I
think
in
some
of
the
cops
meetings
and
I
kind
of
agreed
to
do
it
and
haven't
started
yet.
But
there
are,
there
are
a
bunch
of
AWS
projects
in
kubernetes,
so
like
the
cluster
autoscaler,
the
eks
ami,
the
BBC
CNI
cops
that
all
needs.
Some
sort
of
information
about
ec2
instance
types
like
how
much
CPU
memory
ionize
max
pods
stuff
like
that,
and
then
when
you
want
to
use
a
new
ADA
Lewis
instance.
B
B
Yet,
and
then,
once
you
get
past
that,
then
you
have
to
update
the
cluster
autoscaler
and
all
of
those
projects
have
that
information
baked
in
so
that
you
need
a
new
release
on
every
single
project
before
you
can
start
using
that
new
instance
site.
So
we
had
this
idea
of
maybe
moving
that
information
to
like
a
central
place
where
all
those
projects
can
share,
like
let's
say,
pull
from
a
JSON
file
in
s3
and
then
just
pull
the
information
at
runtime.
So
you
don't
have
to
have
new,
really
is
to
use
a
new
instance
type.
B
A
Yeah
I
mean
I.
We've
talked
about
this
in
the
past,
I
think
the
e
I
think
it's
a
great
idea.
I
think
the
canonical
source,
at
least
initially,
is
likely
to
be.
The
pricing.
Api
I
think
there's
still
some
stuff
that
isn't
even
in
the
pricing
API
so
we'll
have
to
like.
Add
to
it.
So
you
to
me
this
would
make
sense.
I'm,
sorry
I,
don't
so
one
else
is
trying
to
talk
now.
Just
my
echo
I'm,
sorry
I
to
me
it
would
make
sense
as
a
a
sig,
sub-project
I.
A
Don't
know
whether
we'll
have
more
instances
of
this
sort
of
thing,
but
it
doesn't
seem
unreasonable
to
do
in
a
sig,
sub-project
I.
Think
if
you
wanted
to
get
started
starting
in
any
of
the
consuming
projects,
I
mean
you'd,
be
very
wondering
cops
but
like
I'm,
pretty
sure
any
of
the
projects
would
would
welcome
a
little
tool
script
to
build
that
sort
of
JSON
file.
C
A
Think
that's
very
fair,
so
maybe
mm
one
way
to
do
that
would
be
to
so
currently.
I
guess
everyone
is.
Everyone
is
effectively
like
building
a
well.
If
so,
if
everyone
vendored
in
the
library
and
the
library
contained
the
sort
of
fullback
I,
guess
that
would
work,
it's
sort
of
more
prescriptive
than
I
was
imagining
but
I
think
that's
that's
it's
a
it's
a
reason.
Yeah
I
think
that's
a
good
idea
having
a
good
fullback
is
important
and
I.
Think
I
think
this
would
be
a
very
natural
6l
project.
A
B
A
Yes,
but
I
mean
I
think
that's
always
gonna
be
the
case
for
a
fullback
right,
I,
don't
know
I'm,
sorry,
I
think.
What's
one
thing
that's
interesting
is
we
are
also
talking
in
there's.
Another
group
called
Kate's,
infra
and
I
think
they
have
a
side,
channel,
Cates,
infra
team
and
currently
not
a
lot
more
I.
A
Think.
As
part
of
that
you
know,
we've
I've
suggested
that
we
have
a
way
to
download
like
binary
artifacts
like
the
cops
artifacts.
This
would
be
another
there's
also
another
set
of
people
that
are
working
on
like
how
do
we
download
images?
This
would
be
more
like
a
binary
artifact
and
that
it's
not
a
docker
image
as
long
as
you're
not
going
to
docker
pull
this.
It's
unlikely
you're,
gonna
dock
or
pull
this
database
of
of
machines,
but
one
of
the
things
I'm
trying
to
get
to
is
the
idea
that
we
have
multiple
sources.
A
A
The
idea
being
that
you
know
s3
is
a
mirror
of
GCS
or
GCS
is
a
mirror
of
s3,
but
also
you
know
if
you
want
to
run
the
air
gaps
mirror
somewhere.
That
should
be
something
we
we
support
and
don't
make
too
hard.
So
maybe
between
those,
the
full-back
will
be
less
and
less
important.
I
guess
the
problem
with
a
fullback
in
my
mind,
is
it's
slightly
annoying
behavior.
If
you're
running
an
old
version
of
plus
or
API,
you
know
the
the
primary
source
goes
down.
A
C
Yeah
it
you
know,
the
only
thing
is:
is
that
I
know
that
I've
worked
with
some
customers
in
the
past
that
have
strict
requirements
on
what
external
kind
of
services
they're
hitting,
and
you
know
if
it's
always
actively
hitting
some
public
service.
You
know
those
customers
are
going
to
have
a
pretty
bad
experience,
yeah
that.
A
Is
reasonable,
I
think
I
think
having
the
option
to
have
it
baked
in
is
a
great
one.
I
think
we
can
see
if
there
are.
If
there
are
things
which
can
operate
without
external
dependencies,
then
I
think
it'd
be
great
to
have
it
like
have
a
flag
or
something
on
them
so
that
they
could
just
run
entirely
internally
and
I
think
if
they
are
things
which
are
always
gonna
need
something
like
tying
it
to
that
other
thing
that
they
need,
anyway,
might
be
a
reasonable
thing,
but
this
is
a
very
long
enough.
A
I
think
I
think
the
short
answer
is:
let's
do
a
cig
project,
6m
project
and
that's
anyone
will
do
a
straw
poll
now.
If
anyone
disagrees
with
that
idea
and
I
think
in
terms
of
getting
started,
if
you
start
it
anywhere
like
cops
or
cluster
API
or
any
of
the
consuming
projects,
it
should
be
very
easy
to
move
or
even
on
your
own
github,
repo
or
whatever.
A
A
B
A
I
think,
like
there's,
there's
some
of
this
in
cops
now
today,
right,
I,
think
so
I
think
if
we
pulled
up
pricing,
if
we
have
a
tool
that
runs
and
pulls
the
pricing,
API
cause
the
pricing
API
and
then
like
ads
whatever
other
information,
I
guess,
hard-coding
it
and
I
guess
those
hard-coded
things
would
have
to
be
updated.
Whenever
we
did
it,
there
was
a
new
instance
family,
probably
the
you
could
then
run
that
job
in
CI,
which
would
be
interesting.
A
I,
don't
know
that
we
would
trust
it
to
run
every
night
and
like
update
the
live
version,
but
you
know
we
could
we
could
we
could
look
at
putting
them
to
CI
so
that
it
I
would
start
with
just
having
it
run
manually
and
someone
has
to
run
it
manually,
but
it
would
also
be
interesting
to
run
NCI
and
have
it
squawk
at
you
if
it
was
different
or
eventually
once
we
trust
it.
Oh
to
update.
B
A
A
C
Definitely
because
you
know
I
definitely
don't
want
to
manage
that
list.
You
know
statically
by
any
means.
I.
A
A
A
B
A
Okay,
I,
don't
think
you
need
to
I
sorry,
so
no
I
think
I
think
cops
should
parse
the
JSON
file,
because,
ideally
we
would
just
get
rid
of
the
hard-coded
uncoded
go
object,
instantiation
x'
and
we
just
replace
it
with
a
embedded,
go
file
that
if
we
wanted
to
fullback
for
cops
I,
don't
know
if
it
matters
I,
guess
it
does
matter
yeah
for
cups.
We
can
still
run
offline,
so
yeah
I!
Guess
we
do
need
it
well.
Sort
of
fun,
I
mean
if
you're
I
don't
know
well.
B
D
B
A
We
can
I
mean
we
can
also
that's
I,
guess
that's
good.
We
can
also
just
like
look
at
whether
it's
easier
just
to
two
pars
it
pars
the
Jason
on
the
fly
is
every
time
it
might
be.
Okay,
oh
yeah.
That's
great!
Thank
you
for
taking
that
on
I.
Think
that'll
be
really
awesome.
It
will
address
a
lot
of
like
it's
also
frustrating
with
like
one
thing
is
updated,
but
the
other
one
isn't
yeah.
You
get
props
to
install
that
night.
A
C
So
I've
already
added
a
link
in
the
notes:
we've
had
our
initial
release
of
cluster
API
provider
for
AWS.
It's
pretty
rough
at
this
moment,
not
very
configurable,
but
we
wanted
to
get
something
out
there
sooner
rather
than
later,
to
give
people
a
chance
to
kick
the
tires
right
now.
We
are
currently
actively
working
on
updating
the
implementation
to
use
the
new
CRD
based
approach
that
is
recently
merged
in
the
main
cluster
api
project.
C
So
if
you
take
a
look
at
the
provider
now,
the
current
state,
a
master,
is
pretty
broken
as
we
continue
to
work
through
some
of
those
fixes,
but
we
hope
to
definitely
have
those
kind
of
sorted
out
because
we're
planning
on
demoing
this
CRD
based
provider
at
the
upcoming
community
meeting
next
week.
So
hopefully
we'll
have
a
nice
new
demo.
C
Maybe
some
new
functionality
I,
don't
know
yet,
but
we'll
at
least
be
on
the
CR
debased
cluster
API,
so
no
more
aggregated,
API
server
yeah.
So
that's
pretty
much.
The
current
state,
the
CRD
migration,
has
taken
us
a
little
bit
longer
than
we
had
initially
anticipated,
but
otherwise
I
think
things
are
moving
along
pretty
well.
A
E
I
was
dealing
with
an
issue
with
EBS
volumes,
own
awareness
scheduling
and
how
the
autoscaler
fits
into
that
and
while
I
was
digging
through
github
issues
on
the
autoscaler
repo
I
noticed
that
none
of
the
maintainer
--zz
actually
have
AWS
accounts,
and
so
there's
no
one
testing
in
any
of
the
AWS
code,
and
so
I'm
wondering
if
someone
from
AWS
or
Sega
did
with
us
would
be
able
to
provide.
You
know
either.
Maybe
a
test
infrastructure
or
this
time
spent
on
the
project.
D
E
A
A
Interestingly,
there's
actually
to
be
something
we
can
do
in.
I
guess
this
would
be
caught
so
there
are.
There
are
currently
all
the
EDT
tests
are
run
on
cops
or
run
by
a
cops
on
AWS,
and
that
will
likely
change
in
future.
But
if
we
install
the
cluster
autoscaler
as
part
of
either
us
either
he
or
some
because
it
would
so
if
we
got
the
class
to
auto,
splitter
and
sold
on
there.
I
believe
there
are
ete
test,
for
my
guess
is
we're
skipping
them
because
cops
is
not
installing
the
autoscaler.
A
I
think,
but
I'll
have
a
look
at
that
as
well,
and
I
don't
know
if
you
had
a
particular
if
you're
facing
a
particular
issue,
I
would
hope
that
the
volume
zonal,
the
volume
and
zonal
logic
should
be
the
same
in
GCE
and
AWS
like
it's,
basically
just
labels.
So
if
there
is
something
that's
interesting,
my.
E
Issue
was
how
I
set
up
my
yes
I,
had
one
is
G's
spanning
three
availability
zones
and
the
way
that
cluster
autoscaler
works
as
it
builds.
What
calls
what's
called
a
node
template
of
what
a
node
would
look
like
if
it
was
newly
launched
in
an
auto
scaling
group
and
I've
happened
to
be
a
different
easy
than
the
one
I
needed,
and
so
it
said
this
is
easy.
E
This
ASG
won't
create
a
node
in
the
required
ASG
in
require
easy,
so
I'm
not
going
to
scale
it
up,
and
the
solution
is
to
create
one
ASD,
deeper,
AZ
and
feed
them
all
to
the
cluster
autoscaler.
And
then
it's
able
to
pick
the
correct
it
is
cheap
to
scale
up.
But
that's
not
documented
anywhere.
I
was
gonna
open,
a
documentation
for
requests
in
the
autoscaler
Evo
and
that's
when
I
noticed
that
none
of
them
are
actually
using
a
native
us.
A
My
guess
is,
the
same
thing
would
happen
in
GCE
I'm,
reasonably
sure
that
makes
which
other
it
was
a
su
given
can
like
span
zoom.
So
it's
not
nice
or
even
a
like
it's.
Not
this
area
specifically
a
problem
here
are
with
it
legislative
arrest,
coverage
and
I.
Think
the
the
the
other
I
guess
piece
of
good
news
is
that,
like
this
is
some
of
what
cluster
API
is
going
to
hopefully
address
is
having
the
ability
to
either
directly
launch
the
exact
machine
we
want
or
I
guess.
A
This
is
actually
an
interesting
piece
of
feedback
on
if
we
use
aSG's
to
back
machine
deployments
and
machine
sets
in
the
machines
API
we
have
to,
we
still
have
to
be
aware
of
this
sort
of
thing.
The
the
fact
that
we
don't
just
we
can't
just
scale
up
the
AFC.
We
have
to
in
fact
launch
an
instance
and
add
it
to
the
ASG
I
guess,
which
is
a
bit
of
a
bummer.
F
I
misclicked
early
I'll
just
reading
the
issue
that
was
linked
and
the
issue
is
around
credentials
so
and
I
think
it's
more
of
a
user
question.
In
this
case
and
maintainer
saying
we
don't
have
any
WS
account
statement
begin
to
help
you
so
I
think
this
is
more
like.
A
We
should,
at
the
very
least,
make
sure
that
we
have
e
to
e
coverage.
It
is
harder
for
people
inside
Google
to
like
use
use
AWS
just
like
it
is
harder
for
a
toe
sweep
on
shorty.
Why
not
Google
the
politics
of
tricky
and
the
legal
is
like
stuff?
Is
tricky,
I'm,
gonna
fit
actually
drinking,
or
just
that
people
worry
that
it's
tricky,
but
that's
sort
of
how
it
is,
and
so
I
don't
know.
If
there's
anyone
on
autoscaler
who
is
not
employed
by
Google
I'm,
just
like
wondering
about
the
list
of
owners.
A
Anyway,
I
will
try
to
figure
out
who
who
those
people
are
and
yeah
it's
a
good
point,
and
we
should
at
least
make
sure
that
either
in
cops
or
in
other
tests
et
test
they're.
Adding
that
we
make
sure
that
we
have
coverage
of
orders
came
in
on
a
double
yes
and
undo,
see
if
we
don't
have
it
already,
but
I'm
pretty
sure
we
have
a
double
G,
CD
I.
C
Want
to
see
if
it
would
be
possible
to
start
recording
and
posting
the
cig,
AWS
meetings.
I
know
in
the
past,
I've
tried
to
catch
up
on
past
meetings,
and
sometimes
some
of
the
context
doesn't
quite
make
it
into
the
meeting
notes.
So
having
the
recordings
available
would
be
helpful
for
people
to
kinda
catch
up,
I
think
so.
A
Good
news
and
bad
news:
well
good
news,
bad
news
and
very
bad
news.
The
good
news
is
I've
been
recording
them.
As
of
others.
The
bad
news
is
I
have
not
been
posting
them
yet
because
I've
been
told
to
post
them
in
order
the
better
news
there
is
I've
started,
posting
and
I
think
I'm
up
to
August,
2017
I
guess
the
very
bad
news
is
there
are
about
26
hours
of
videos
about
to
hit
YouTube.
So
if
you're
planning
on
catching
up,
you
need
to
like
block
off
your
weekend.
A
But
but
yes,
you
are
absolutely
right.
We
should.
We
should
do
that.
The
the
other
problem
is
I've
been
told
to
post
them
in
order.
Maybe
I
can
see
if
I
can
get
a
little
waiver
on
that,
because
it
is
annoying
because
I
feel
like
so
I'm
missing,
like
mid
August
2017
I'm,
like
is
it
worth
like,
should
I
just
post
this
weeks
or
should
I
like
find
August
2017
right,
the
sort
of
the
common
sense
answer
might
be
in
conflict
with
my
instruction.
So
far
so
I
will
I
will
clarify
those
instructions
all
right.