►
From YouTube: Kubernetes SIG Cluster Lifecycle 20181205 - Cluster API
Description
Meeting Notes: https://docs.google.com/document/d/16ils69KImmE94RlmzjWDrkmFZysgB2J4lGnYMRN89WM/edit#heading=h.sp1hfrs780gt
Highlights:
- Adding automated checks to the code repository
- Local defaulting
- Cluster autoscaler backend
- 2019 roadmap
- How to contribute?
- Planning for webhook development
- Update on breakout session next week
A
Hello
and
welcome
to
the
Wednesday
December
5th
meeting
of
the
cluster
API
sub-project.
First,
a
cluster
lifecycle
I
want
to
start
today's
meeting
by
just
announcing
I,
don't
know
how
our
this
was
disseminated,
but
Peres
sent
out
an
email
saying
there
was
a
security
vulnerability
in
zoom
and
that
everyone
should
please
update
to
use
the
latest
client.
So
they
don't
have
the
problem.
I
think
it
affects
Mac,
Windows
and
Linux,
so
you
can
just
go
into
zoom
and
say
check
for
updates
and
make
sure
you
have
the
latest
one.
A
C
C
A
We'll
come
back
excellence
all
right,
so
I
stuck
a
couple
of
things
onto
the
agenda
as
I
was
going
through
and
triaging
issues
this
week,
I
closed
a
number
of
issues
and
there
were
couple
I
thought
might
be
worth
talking
about.
So
the
first
one
was
number
43,
which
was
putting
more
automated
checks
in
the
repo,
so
things
that
you
know
we
just
like
automation
to
take
care
of,
so
that
we're
not
you
know
sending
PRS
to
fix
little
things
that
we
catch
them
before
code
gets
submitted.
D
A
D
Think
it's
a
mixture
of
all
of
them,
but
it
still
misses
a
lot,
but
I
think
it
tries
to
look
at
everything
I
think,
but
it
only
what
it
does
is
actually
has
common
misspellings,
rather
than
so,
it
will
be
like
while
smelling
is
obviously
a
word
but
like
knowing
it
doesn't
have
Cooper
Nets,
which
is
the
one
I
always
do
but
like
it
will
have
other
common
type.
That's
what's
the
name
of
the
tool.
I
will
look
it
up
right
now
and
put
in
chat
okay,.
E
A
E
E
F
A
A
Yes,
Daniel
pointed
to
another
tool
which
helps
us
verify
that
we
aren't
ignoring
rivers,
which
is
a
great
thing.
That's
something
I
always
look
for
in
code.
Reviews
is
to
make
sure
we're
not
either
assigning
errors
to
underscores
or
calling
functions
of
return
error
and
just
not
not
catching
the
value
that
would
be
nice
cool
so
so
Justin
can
you
go
update,
not
just
chat,
but
also
the
issue
with
a
link
to
the
spelling
tool
we
use
for
4kk
and
maybe
also
put
some
some
links
to
the
other
tools.
A
People
are
putting
in
check
and
then
they
know.
Can
you
update
the
issue
and
put
a
link
to
the
error
check
thing
as
well,
and
then
also
if
people
want
to
help
add
any
of
this
automation
to
the
repository
that
would
be
great
I
know,
Alvaro
has
been
adding
some
of
its
I.
Don't
know
how
much
bandwidth
he
has
to
add
the
rest,
so
people
are
excited
about
helping
us
do
this.
That
would
be
a
great
contribution
and
I
think
a
pretty
easy
thing
without
having
to
necessarily
understand
and
in-depth
any
of
the
code
itself.
A
So
I
think,
as
this
already
issued
already
marked,
is
sort
of
a
good
first
issue
or
good
new
contribution,
but
that
doesn't
mean
that
somebody
who's
already
already
contributing
shouldn't,
also
feel
inclined
to
take
it
as
well
and
Steve
mentions
in
chat
that
he
might
build
up
else.
That
would
be
great
I'll.
F
G
A
B
A
B
D
Benefit
is
that
it's
it's
faster,
it's
easier
and
that
I
think
we
could
use
the
same
code
is
the
hope
ie
to
like
bounce.
The
expectations
on
this,
like
I,
originally
just
hit
a
couple
of
things
that
I
was
hitting
personally
and
then
other
people
like
alpha
Rho
I,
don't
actually
own
a
printer
name.
I'm
sorry,
but
like
gave
me
a
link
to
like
the
existing
code
and
Chris
fork
wriggle,
and
it
gave
me
like
some
pointers
as
well,
but
right
now
it
only
does
machine
deployments.
D
A
Jason
to
your
question
now
also,
my
understanding
is
with
CRTs
we're
supposed
to
be
able
to
get
basic
validation
and
potentially
basic
defaulting
through
schema
definitions
and
more
complicated,
advanced
defaulting
in
how
to
use
web
book
4.
So,
ideally,
we
wouldn't
have
to
write
a
mutating
web
hook
to
do
simple
things
like
it
should
be
set
to
0
by
default
right,
which
is
I,
think
part
of
what
Justin's
PR.
H
D
D
A
The
difference
is
basically,
if
unspecified
controller
should
do
best
practices
based
on
the
controller
code
which
may
change
over
time.
The
best
practice
may
change
over
time,
whereas
if
you
actually
do
the
faulting,
then
the
defaulting
value
gets
inserted.
The
controller
must
then
respect
that
value,
because
it
can't
tell
us
the
defaulting
code
put
it
in
or
if
the
user
specified
it
exactly.
D
And
so
as
a
concrete
example,
for
example,
suppose
you
decided
that
anonymous
author
on
API
server
was
no
longer
safe
today,
so
the
default
previously
might
have
anonymous
true,
unless
both
true,
but
we
decide
in
the
future,
and
all
myself
should
be
false
unless
you
actively
like
do
that.
So
that's
the
sort
of
thing
where
the
distinction
matters
is.
A
A
Yeah
for
existing
objects,
you'd
still
have
the
old
values,
but
I
think
that's
actually
not
bad
either,
because
if
you,
if
you
have
an
existing
object
and
then
all
you
do
is
change
your
controller
binary
for
that
behavior
of
that
existing
object
to
change.
It's
also
kind
of
surprising
to
users,
yeah
well,
I,.
D
A
So
wouldn't
just
say
the
next
time
I
run
my
reconcile
loop,
I'm
gonna,
like
redo
everything
with
the
new
best
practice
values.
It
would
be
when
I
make
some.
You
know
other
change
where
you
might
expect
things
to
be
shuffling
around
anyway,
like
I'm,
upgrading
your
control
plane,
then
I
would
reshuffle
things.
Then
a
default
focus.
That's.
D
A
Okay,
I
think
we
should
all
right
I
think
we
should
circle
back
to
this
and
we
get
to
Vince's
topic
of
offenses
made
it
home.
Yet
so,
let's
keep
turning
through
the
other
parts
of
the
agenda.
So
next
one
is
cluster
autoscaler
cluster
API
back-end
looks
like
there's
a
pull
requests
to
the
autoscaler
repair
linked
here.
Yeah.
E
That's
me
I
just
to
introduce
myself
quickly
I.
My
name
is
Olaf
I
work
for
a
holster
in
Berlin
Germany,
and
we
use
provide
managed
communities
to
our
customers,
including
the
Machine
controller
and
cluster
API
and
the
autoscaler.
And
what
we've
done
here
now
is
to
just
provide
a
cloud
provider
for
autoscaler
so
that
it
uses
the
machine
controller
and
the
cost
rate.
The
idea
of
a
CD
private
controller
to
scale
up
and
down
machine
provides
automatically
for
old
auto-scale.
E
Yeah,
and
so
we
it
works
on
our
cluster
pretty
nicely
and
pull
request
is
linked,
and
maybe,
if
it's
our
hope,
you
guys
could
Rita
so
I
have
some
comments
on
I
would
be
nice
again.
We
didn't
have
any
major
problem.
Well,
one
problem:
we
did
have
a
potential
issue.
Is
that
be
nice
if
there
were
some
cloud
independent
or
technology,
independent
providers
that,
like
the
Machine
cept
machine
Department
provider,
what's
the
provider
spec,
which
contains
to
provide
the
cloud
specific
that
the
cloud
configuration
for
the
phone
for
new
frontier
machines?
E
A
So
things
like
you,
know,
number
of
cores
amount
of
the
machine.
That's
you
know,
memory
and
and
cores
reserved
based
on
you
know
what
version
of
terminators,
you're,
running,
etc.
Right
so
I
think
the
idea
was
that
obviously
need
to
parse
the
common
bits
of
the
machine
class
to
understand
what
the
node
it's
going
to
create
is
going
to
look
like
and
then
the
other
big
missing
piece.
We
have
it's
the
autoscaler
when
it's
scaling
down
needs
to
be
able
to
delete
a
specific
machine
or
machine
set
right
now,
there's
no
way
to
do
that.
A
So
I
mean
those
are
two
big
to
big
pieces
that
that
I
know
that
we're
missing
and
maybe
since
you've
been
actually
deep
in
the
code
you
found
some
other
pieces.
I
know
that
hardik,
who
I
heard
on
the
line
today
also
has
a
fork
of
the
autoscaler
that
he's
been
getting
working
on
top
of
the
gardener's
machine
controller
manager,
which
is
also
adopting
the
cluster
API,
and
so
he
may
have
some
other
thoughts
on
other
things
that
were
missing
and
would
be
great.
If
you
look
at
your
PR
too.
Well,.
I
I
I
Cannot
because
right
now
they
are,
they
are
depending
on
the
load
objects
or
so
these
are,
in
my
opinion,
there
are
so
many
caps
at
the
moment
which
needs
to
get
tested
for
sleep,
a
part
and
another
scale
effort
can
only
the
integration
would
be
smoothly
possible.
So
what
would
be
nice
is
to
address
those
issues
first,
one
by
one.
Maybe
the
KP
was
open,
so
we
could
probe
me
which
try.
J
I
A
I
Ahead,
a
loss-
yes,
just
the
last
point:
I
just
pushed
the
issue
which
was
created
for
the
same
city
of
good
ideas
around.
How
can
we
stall
this
thing
at
the
post
API
level?
It
would
be
really
nice
if
you
can
put
put
some
comments
on
how
we
can
result
in
Hong.
Kong
can
define
a
good
strategy
for
the
deletion
of
the
machines,
no
missions.
E
A
Classes,
yes,
the
machine
class
does
still
contained
the
provider
specific
structure,
the
ideas
that
we
would
also
promote
fields
into
machine
class
itself
that
are
common
across
all
providers,
like
I,
think
the
structure
that
the
node
uses
is
something
like
resource
allocation
yeah.
We
chose
allocatable
and
you
can
fill
that
in
with,
like
my
CPU
is
this
my
memory?
Is
this
etc
right
so
and
then
the
only
problem
with
that
is.
A
If
we
have,
we
have
that
field,
and
then
we
have
the
field
inside
the
provider
spec
and
it
basically
we
up
to
the
actuators
to
make
sure
that
those
match
right.
So
it
would
be
possible
to
create
a
machine
class
or
you
put
you
know
in
the
in
apart
the
autoscaler
reads:
I
have
16
cores
and
then
put
in
the
provider.
A
Spec
I
want
a
four
quart
machine
and
of
course,
that's
not
going
to
work
very
well
with
goddess
Keller,
because
it's
gonna
assume
that
it's
getting
a
lot
more,
you
know
coerced
and
it
actually
both
and
so
we'll
have
to
figure
out
how
to
make
sure
those
things
stay
in
sync.
But
the
idea
was
to
extract
some
of
those
common
fields
up
so
that
the
autoscaler
could
read
them.
Okay,.
E
A
We've
already
got
things
like
labels
and
paints,
which
I
know
or
other
things.
Data
scaler
is
really
interested
in
and
I
think
there
were
a
couple
of
other
fields
again,
there's
a
there's,
a
doc
that
did
where
we
wrote
down
a
lot
of
the
requirements
quite
a
while
back
and
we've
been
slowly
trying
to
figure
out
sort
of
the
right
way
to
put
them
into
the
API.
Okay.
E
A
Okay,
did
you
anything
else
about
that?
A
lot?
First,
we
move
on
excellence,
okay,
so
next
I
put
a
link
in
here:
Lucas
created
a
doc.
He
created
one
last
year
for
the
2018
roadmap
for,
say,
cluster
lifecycle.
This
is
a
link
to
the
2019
version
of
that
doc
and
in
particular,
I
wanted
to
there's
a
small
section
here
called
cluster
API,
where
he
sort
of
started,
stifling,
like
you
know
his
idea,
or
maybe
maybe
ten
contributors
to
like
where
we
think
the
cluster
API
is
going
to
be
sort
of
in
the
next
year.
A
The
reason
I
bring
this
up
now
is
because
there
is
a
number
of
talks
happening
in
queue
con
related
to
our
sig.
The
first
talk
is
during
the
contributor
summit
attempt
st.
clair
is
going
to
tell
the
rest
of
people
as
a
contributor
submit
about
sort
of
the
state
of
state
cluster
life
cycle
and
he's
gonna
talk
about
where
we've
been,
and
what
we're
going
and
part
of
where
we're
going.
A
Here's
some
things
that
might
go
on
this
slide,
because
I
think
Tim
was
looking
for
help
putting
it
together,
but
I
didn't
want
to
sort
of
pull
people
and
say
like
what
do
you
guys?
Where
do
you
guys
think
will
be
in
a
year?
What
do
you
want
to
see
happen?
What
are
you
excited
to
I?
Get
added
to
the
project
and
I
have
a
couple
ideas,
but
I
want
to
pull
the
group
before
I
talk
about
the
things
that.
A
K
F
A
F
I
can
tell
you
from
more
of
a
company
selfish
standpoint
that
next
year
our
focus
will
be
on
the
CNI
add-ons.
How
I
mean
that
I
know?
We've.
We've
talked
a
lot
about
that,
but
for
us
to
actually,
if
we,
if
other
projects
in
the
company
wants
to
adopt
cluster
API,
that
needs
to
be
solved.
Another
topic
that
I
know
that
it
probably
doesn't
affect
any
of
you
guys,
but
we
have
a
high
interest
in
isolated,
pods,
sandbox
pods
as
Tim
I'll
Claire
from
signal
collared.
We.
F
So
we
I
worked
on
a
previous
project
on
isolated
containers
at
the
p.m.
where
we
built
a
one
time
that
that,
basically,
those
containers
in
with
wrapped
in
an
isolation
of
a
VM
didn't
have
dock
or
anything
in
there.
So
we
we
have
a
strong
interest
in
isolation
and
where
order
monitoring
the
sandbox
pods
work,
do.
F
No
I
think
you
how
how
we
deploy
I
mean
for
for
for
that
to
work.
We
we
need
certain
one
times
install
we
we
need
the
you
know,
because
cluster
API
is
going
to
deploy
a
cluster.
It
needs
to
make
sure.
If
we,
if
they
let's
say
the
user,
wants
sandbox
pods,
then
we
need
to
be
to
deploy
the
the
correct
one
times:
the
right,
correct
couplet,
so
that
that
is
an
area
that
we
have
an
interesting.
D
F
I,
don't
I
can't
I
couldn't
tell
you
because
I
haven't
since
I've
joined,
closer
API,
a
my
attention
have
been
but
I
know
in
the
Kamel
Claire's
sandbox
pod,
it's
a
different
one
time
and
as
requires
changes
to
the
COO,
but
I
can't
say
talked
about
in
the
stuff
that
we're
working
on.
But
you
know
it.
It
may
require
a
different
couplet
may
require
setting
up
I
just
I.
Couldn't
I
couldn't
tell
you
the
answer.
I
can't
tell
I
couldn't
answer
that.
F
H
D
I
have
a
I
have
side
like
I
have
something
coming
which
so
obviously
so
many
cube
can
actually
run.
We
have
the
alright,
let's
buy
Minich,
you
can
actually
run
with
I
think
the
mode
is
called
none
where
it
runs
in
docker,
so
it
doesn't
need
a
VM.
I
haven't
tried
that
personally,
what
I've
been
experimenting
with
is
running
API
server
and
Etsy
D
and
the
controller's
themselves
directly.
The
reason
for
that
being
that
seems
to
be
closer
to
how
everyone
was
running
or
other
people
were
actually
running.
D
Cluster
API
I
could
seemed
like
everyone
was
relying
on
a
pre-existing
cluster,
so
I
felt
like
it
would
be
good
to
have
all
configurations
you
supremes
us
in
cluster,
but
maybe
make
that
super
easy.
So
it
wasn't
like
a
you,
have
to
go
set
up
the
cluster,
so
I'm
experimenting
with
that
and
I'd
love
to
collaborate
with
you
on
that.
If
but
that's
sort
of
what
I'm
looking
at,
but
that
might
not
even
be
one
for
next
year.
With
luck.
A
Yeah
that
was
actually
something
I
put
put
in
the
road
back
to
was
thinking
about
ways
that
we
can,
you
know,
get
rid
of
/,
replace,
mini
cube
or
change.
Sort
of
the
bootstrapping
cloaks
I
think
the
mini
cube.
Bootstrapping
flow
is
pretty
convenient
for
maybe
a
developer
that
is
sort
of
experimenting,
but
it's
not
necessarily
a
great
end
user
experience
and
it's
also
as
long
as
point
out
number
of
times,
not
necessarily
something
that
we've
sort
of
wanted
to
want
to
deliver
to
enterprise
customers.
A
A
I
think
AJ
is
something
also
that's
sort
of
on
the
list
for
next
year
is,
if
you
do
have
an
existing
cluster
sort
of
a
management
cluster
is
a
lot
of
people
are
using
to
deploy,
think
deploying
a
che
underneath
that
isn't
necessarily
too
difficult,
but
if
you're
starting
from
scratch-
and
you
just
want
to
not
have
to
have
a
cluster-
to
create
a
cluster,
how
do
you
deploy
an
AJ
cluster
from
scratch?
I
think
is
a
pretty
difficult
problem
to
solve.
F
A
All
right
so
I'll
just
mention
a
couple
other
things
that
I'd
put
on
the
slide.
If
people
haven't
pulled
it
up,
so
one
is,
you
know,
obviously
getting
API
versions
through
sort
of
the
alpha
and
beta
beta
process.
As
Tim
mentioned
last
week,
part
of
that
is
figuring
out
how
we
do
conversion
between
different
versions
and
sort
of
what
the
migration
path
for
people
is
I
think
once
we
have
a
way
to
move
people
between
versions,
we
can
start
cutting
versions
that
people
can
depend
upon,
so
we
can
move
them
forward.
A
Auto-Scale
integration,
which
came
up
earlier
in
the
talk
I,
think
that's
a
big
thing.
We're
gonna
want
to
solve
next
year,
because
that's
in
my
mind,
one
of
the
big
benefits
of
the
cluster
api
is
is
you
can
get
an
auto
scaler
for
free
if
you're
using
a
cluster
api
and
another
thing
that
sort
of
on
my
personal
wish
list
is
to
get
us
into
a
place
where
we
can
use
the
cluster
api?
A
Not
just
have
itself
I
mean
on
the
site,
assess
tests
and
release
automation,
I
think
that
could
be
interpreted
as
testing
Lee's
automation
for
the
cluster
API,
but
to
use
the
cluster
API
for
kubernetes,
test
automation
and
in
particular,
getting
rid
of
Cuba
and
finally,
deleting
the
cluster
directory
from
the
kubernetes
repository,
which
it's
been
a
goal
for
a
while
I.
Think.
If
you
look
at
some
of
the
broader
themes
across
the
sig,
it's
becoming
clear
that
you
know
people
will
make
a
change
to
cube
up
and
that'll
become
part
of
their
release.
A
Notes
is
like
oh,
we've
switched
to
core
DNS
and
you
know
1.13
and
people
go
like
well.
What
do
you
mean
like
cube?
Admins
switched
last
release
when
it
stopped
switching.
How
does
Cuba
Korn
doing
it,
like
you
know,
sort
of
tying
like
because
it's
part
of
the
release
and
built
into
the
cottage
release?
It
becomes
part
of
the
release,
notes
in
a
way
that
doesn't
necessarily
make
a
lot
of
sense,
and
so
I
would
love
to
see
us.
Both.
D
We
can
still
use
like
there's
a
bundle
spec
or
like
some
form
of
bundle
to
give
the
the
progress
or
they
like
not
make
it
a
lot
more
work
and
have
everyone
sort
of
more
consistent,
I
guess
she
doesn't
require
one
to
use
the
bundle,
but
I
think
it
would
be
nice
to
let
or
do
something
bundle
ask
like.
Let
people
use
that
I
was
actually
thinking
about
adding
my
most
supports
to
do
the
storage
class
in
GCP
well,
starting
there.
So
yeah.
A
A
D
B
I
was
gonna
say
on
the
multiple
provider
topic
in
particular,
this
might
be
somewhere
where
we
can
extend
kind
of
the
webhook
approach
to
kind
of
handle
it
as
well.
Instead
of
right
now,
where
we're
deploying
multiple
control,
we
would
have
to
deploy
multiple
controllers
that
implement
specific
actuators.
B
We
could
actually
potentially
leverage
the
web
host
support
that
if
there
is
kind
of
common
logic
that
could
be
handled
directly
in
a
cluster
API
provided
controller
and
then
the
provider
specific
things
can
be
managed
as
web
hooks
that
interact
in
addition
to,
and
now
we
can
order
the
web
hooks
in
a
way
that
makes
sense.
Yes,.
A
I
think
there's
there
been
a
couple
proposals
for
us
to
basically
bring
the
Machine
controller
back
into
the
common
repo
and
not
actually
link
it
with
the
provider
code.
There's
web
hooks,
there's
G,
RPC
or
number
of
ways
we
can
sort
of
tracked
just
the
providers
to
the
code
out
and
make
that
through
its
own
thing
and
I,
think
that
would
that
would
also
be
a
great
thing
for
us
to
shoot
for
next
year,
because
if
we
could
say
here
is
the
machine
controller,
you
run
it
and
then
you
register
your
providers
to
the
code.
I
Yes,
just
to
add
on
that
point,
so
it's
recently
implemented
something
very
similar
for
in
cm,
so
a
separate
geography,
server,
implementing
only
three
methods
to
create
machine
and
Lister
machines,
and
when
he
learned
from
the
CSI,
a
lot
of
things
were
I'm,
not
even
of
I'm,
not
even
providing
the
tube
config
to
the
Jail
visit.
I
was
running
outside
so
that
gob
cedras
world
justice
too.
So
we
contact
between
the
shared
controller
share
it
the
promotion
controller
and
each
your
service.
I
I
Think
that
brings
a
lot
of
interesting
benefits
in
different
ways,
because
well
one
of
the
thing
which
I
am
seeing,
which
is
which
will
eventually
evolve
in
a
way
that
the
separate
controllers
which
are
not
going
to
have
the
cube
config
axis,
they
start
having
practices
of
messing
up
with
their
pasting
machine
objects
and
like
both
and
machine
status,
would
be
arbitrarily
managed
by
different
cloud
providers.
Adding
new
you
messages
and
stuff
like
that,
that
can
actually
create
a
lot
of
chaos,
so
I'll
implement.
J
I
A
And
I
think
that
goes
back
to
earth.
Locke
was
saying
about
like.
If
we
really
want
to
be
able
to
run
a
management
cluster
that
manages
multiple
environments,
then
the
providers
so
different
barman's
have
to
be
consistent
enough
that
we
can
manage
them
all
the
same
way
and
if
they're
all
have
their
own
machine
controllers
to
do
something
vastly
different
with
machine
status
and
their
tree
machine
status
very
differently.
It
becomes
hard
to
sort
of
combine
them
into
a
single
world
where
we
can
seamlessly
run
them
sort
of
together.
D
I
do
have
I
got
concerns
about
the
amount
of
work
that
that
will
entail
and
I
think
we
can
probably
start
with
you.
Justin
go
interphase
been
during
and
see
like
where
that,
where
the
code
is
I'm
like
where
the,
where
the
actual
like
logic
is
like,
how
much
is
how
much
we
actually
save
versus
like
if
we
created
your
PC
interface,
it's
sort
of
a
contract
right
so
feels
like
it
could
be.
A
lot
of
work.
Sign,
I'm,
not
unsupportive,
but
I
am
wary.
D
I
Something
really
so
I
was
also
under
the
impression
that
it
could
turn
up
a
lot
of
work
to
have
a
completely
new
contract
and
so
on.
But
it
shared
link
of
the
driver,
implementation,
weather
utilities
and
we
started.
And
if
you
look
at
under
the
app
and
the
package,
you
will
see
that
the
only
thing
required
to
add
a
new
club.
I
A
Straights
well,
if
anybody
else
has
other
thoughts
about
what
they
think
we
should
be
doing
in
419
feel
free
to
send
em
to
me,
on
slack
or
plus
me
into
a
comment
on
the
meeting
notes,
if
you
want
them
to
be
considered
for
these
slide
presentations
that
are
happening
next
week,
I'd
like
to
move
on
to
running
a
little
bit
low
on
time.
The
next
question
is
from
Jaeger
asking
for
good
first
issues
to
contribute.
A
It
sounds
like
Jaeger,
you
must
are
you?
Are
you
new?
Are
you
here?
Let's
see
maybe
just
reading
the
notes.
I
will
add
some
notes
here.
We've
talked
about
this
a
couple
of
times
in
the
past.
If
you
look
for
issues
that
have
the
Help,
Wanted
or
good
first
issue
label,
that's
a
good
place
to
get
started
things
we
mentioned
earlier
in
this
meeting
about
adding
automated
checks
to
our
repository
for
things
like
spell
checking
security
scanning,
etc.
A
Those
are
the
great
things
to
get
started
because
they
don't
require
you
to
have
any
sort
of
context
on
what
the
code
is
actually
doing.
It's
more
sort
of
getting
the
repository
set
up,
but
it
gets
you
familiar
with
the
flow
of
contributing
code
and
will
probably
require
you
to
like
run
through
tests
and
actually
sort
of
see
how
things
fit,
which
will
get
you
ready
to
start
tackling
some
encoding
issues.
We
can
always
use
help
on
things
like
documentation.
C
I
just
drum
writing.
It
was
only
10
minutes,
so
I
opened.
This
umbrella
issue
should
just
like
track
in
general
genetic
ideas.
This
is
my
understanding
of
its
like.
It
might
not
be
the
whole
truth.
I
like
what
I
would
like
to
see
the
Brotherhood.
That
would
transform
it
right
now,
like
a
validation
and
I.
Think
conversion
I've
talked
with
Jason
a
little
bit
about
how
we're
gonna
upgrade,
for
example,
the
status
definition
later
and
seems
like
at
the
conversion
of
Web
books
before
it.
C
It
in
113
is
one
way
to
do
it
and
then
there's
validation
and
mutating.
My
book,
which
I
heard
some
visible
philosophical
talk
before
from
Justin
around
the
mutating
and
defaulting
and
which
it
would
be
nice.
What
one
of
my
proposal
here
will
be
to
have
a
package
in
instead
the
cluster
API
that
actually
like
the
stay
ordering
and
just
like
it,
gives
like
a
bullet
code
for
other
providers
to
add
web
books
in
and
make
that
like
just
plug
in,
so
that,
if
we
have
a
one,
could
change
that
logic?
B
A
D
On
the
idea
of
whether
it
should
live
in
common
or
controller
I,
think
you
raise
a
good
point
that
the
sorry
that
per
cloud
repose,
I,
guess
I-
think
you're
is
a
good
point
that
there's
gonna
be
defaulting
inside
the
provider
spec
as
well
and
validation
of
the
provider
spec.
And
so
we
definitely
need
something
in
each
per
cloud.
Repo
I
think
it's
sort
of
like
the
pattern,
which
was
I,
think
we're
doing
with
plaster
CGI,
where
we
have
like
common
code
in
the
in
the
court
in
backwards.
B
Just
that
way,
instead
of
the
the
cluster
cuddle
model,
seems
to
in
some
ways
cause
more
more
harm
than
than
good,
with
having
different
versions
for
different
things,
so
I
think
having
a
common
shared
one
that
that
way,
you're
not
having
to
match
between
the
version
of
the
web.
Hooks
that's
deployed
with
containers
that
are
deployed.
Those
would
match
directly
from
the
project
versus
you
know
having
to
worry
about
moving
them
in
step
between
the
project
and
the
common
controllers.
A
C
No,
you
can
you
actually
also
need
to
specify
a
chain
so
like
the
order
of
multiple
things,
so
my
idea
was
to
just
put
the
this
logic
besides
like
at
the
order
of
things
inside
the
cluster
get
upstream
and
then
all
the
providers
will
actually
use
that
package,
but
the
package
will
also
kind
of
offer
some.
You
know
built-in
validation
web
hooks
for
internal
things
too
close
to
API
they're,
not
specifically.
A
C
H
Have
one
quick
question
is
to
be
sure,
I
and
I
thought
that
correctly,
so
it
is
possible
to
sort
of
force
an
order
of
multiple
workbooks
that
validate
the
same
thing.
I
know
it's
possible
to
have
more
than
one
but
I
think
it's
not
possible
to
order,
or
at
least
I,
don't
know
how
it
would
be
possible.
Like
say,
I
want
always
this
one
workbook
to
be
called
first
and
then
this
other
and
I.
C
Think
you
can
specify
like
the
the
order,
but
it
will
be
kind
of
like
a
sequential
that
said:
I
don't
know
to
Robert
a
question:
I,
don't
know
if,
like
it
will,
for
example,
if
you
return
an
error
message,
if
we
will
drop
it,
the
first
one
or
but
very
much
like
the
chain
will
keep
going
until
the
end.
I
would
suspect,
for
example,
something
for
validation.
C
A
H
A
Well
so
I
think
take
away
as
Jason
was
saying.
Is
a
lot
of
us
haven't
dove
deep
into
web
hooks
yet,
and
we
need
to
do
some
some
reading
and/or
experimentation,
it's
great
that
you're
you're
spearheading
this
and
and
there's
some
questions
that
we
would
like
to
know
a
couple
more
answers
for,
as
we
move
forward,
I
think
you
know,
starting
with
something
simple,
like
validation
or
defaulting,
makes
a
lot
of
sense.
A
K
B
A
J
A
So
now
the
Sunday
before
the
conference
there's
like
a
event
as
part
of
the
contributor
summit,
where
people
are
going
to
tribute
summer,
they're
getting
together
in
the
evening
on
Sunday
before
the
conference.
My
guess
is
that
the
contents
of
conference
ends
on
Thursday
log
killer,
probably
flying
home.
A
J
A
I
think
the
other
thing
is
that
I
think
a
lot
of
people
on
the
call
are
based
in
Seattle.
Maybe
not
a
lot
me
and
Chris
are,
and
we
can
just.
We
can
certainly
get
together
outside
of
this
at
this
time
it
doesn't
work,
for
me
sounds
good
with
that.
We're
two
minutes
over.
So
if
Jason
said,
if
you're
interested
I
don't
think,
there's
a
way
on
the
doodle
to
say
that
I'm
interested
but
I
can't
be
there
in
person,
because
I
know
a
lot
of
people
were
interested
in
dialing
in
via
zoom.
A
If
you
are
interested
in
down
in
via
zoom,
it
would
also
be
useful
to
fill
out
the
doodle
and
maybe
just
shoot
Jason.
A
note
saying
like
I'll,
be
remote
or
something,
and
we
can
sort
of
take
that
timing
into
account
as
well,
if
you'd
like
to
be
able
to
at
least
listen
and
participate,
and
with
that
we're
gonna
call
it
hopefully
we'll
see
a
number
of
folks
at
Q,
Khan
or
related
events
next
week,
and
otherwise
we'll
be
back
at
this
meeting
in
two
weeks.
So
how
do
you
guys
soon.