►
From YouTube: Kubernetes Community Meeting 20190509
Description
We have PUBLIC and RECORDED weekly meeting every Thursday at 10am PT
See: https://github.com/kubernetes/community/blob/master/events/community-meeting.md for more details.
A
All
right,
like
let's
get
started,
it
is
May
9th
2019.
This
is
the
kubernetes
weekly
community
meeting.
My
name
is
Jorge
Castro
from
SiC
contributor
experience
and
I
will
be
your
host
today,
a
few
things
before
we
get
started.
This
meeting
is
being
live-streamed
to
youtube
and
is
publicly
available.
So
please
be
cognizant
of
what
you
say
and
as
always
remember
that
the
code
of
conduct
is
in
effect
we're
gonna,
have
a
quick
format.
A
Change
for
this
meeting,
so
first,
instead
of
a
demo,
is
gonna
actually
be
the
release
status
from
Claire
and
then
Stefan
will
go
with
a
demo
and
then
the
contributor
tip
of
the
week.
You
have
a
cap
of
the
week
this
week
and
then
we'll
have
some
sig
updates
and
with
that
Claire,
are
you
ready
to
take
it
away?
A
B
All
right
so
1:15
release
update.
We
are
almost
through
with
week
five,
since
this
is
an
11
week,
release
cycle
we're
at
about
the
halfway
point
in
the
release.
What
happened
this
week?
We
had
our
third
alpha
cut
and
next
week
we're
gonna
have
our
first
beta
cut,
the
115
branch
cut
and
the
115
jobs
will
be
created
and
the
111
jobs
will
be
removed.
B
For
enhancements,
we
are
currently
tracking
447
enhancements
and
we
will
see
how
that
number
changes.
As
we
get
closer
to
code
freeze,
I'm,
probably
going
to
now
start
reminding
the
community
that
code
freeze
is
coming
up
at
the
end
of
this
month.
It
is
Thursday
May
30th,
any
enhancements
that
are
going
to
be
in
the
115
release
need
to
be
code
complete
by
that
code,
freeze
date
and
additionally,
for
any
sig
leads
who
are
on
and
listening
now
is
also
the
time
you
should
start
thinking
about
different
themes
for
your
cig.
B
If
you're
sig
has
any
enhancements
that
are
going
to
be
delivered
in
the
115
timeframe,
the
release
lead
team
has
started
going
around
and
pinging
different
SIG's
that
have
enhancements
targeted.
If
you
haven't
heard
from
us,
yet
you
will
probably
hear
from
us
in
the
next
day
or
two,
and
these
themes
are
helpful
for
us,
as
we
start
to
think
about
the
different
communications
and
blog
posts
that
we
might
want
to
put
out
around
the
1:15
release
and
another
general
announcement.
If
you're
gonna
be
at
coop
Connie
you,
we
will
have
an
in-person
face-to-face.
B
A
C
D
D
Just
introduce
flag
or
a
little
what
what
this
operator
does
it
integrate
with
the
sunset
his
knife
solutions
like
SEO,
app
mesh
or
the
nginx
ingress
controller?
The
engineer
singles
controller
was
released
today
and
it
uses
these
service
knife
solutions
to
drive
traffic
to
control
the
routing
inside
the
cluster,
and
it
also
integrates
with
Prometheus
to
run
uncannily
analysis.
So
what
tracker
does
it
has
a
control
loop
like
any
other
operator,
but
instead
of
just
listening
for
kubernetes
events
or
out,
let's
say,
flag
lc/ld,
which
is
no
canary
object.
D
D
It
always
sells
at
some
point
right.
So
the
idea
is
to
restrict
the
impact
of
that
failure
and
have
an
automatically
rollback
if
there's
possible,
and
also
you
should
be
able
to
define
a
validation
process,
how
flag
how
this
automation
tool
can
validate
what
will
be
shipped
in
production
and
it
uses
a
couple
of
KPIs
built-in
KPIs
and
you
can
set
thresholds
for
that.
D
Another
goal
was
to
make
the
deployment
process
observable
so
Flagler
country,
the
graph
on
our
dashboard,
also
exports
its
own
Prometheus
metrics
has
built
in
other
things
system
with
slack,
but
you
can
also
use
a
lot
manager
and
have
pager
duty
or
whatever
you
are
using
as
your
auditing
system
and,
of
course
it
emits
kubernetes
events.
So
if
you
have
some
integration
with
with
the
kubernetes
events,
you
can
track
all
the
changes
that
flag
advanced
inside
your
custom.
D
The
goalie
flagger
is
instead
of
you
having
to
create
objects,
objects,
flagger
will
create
it
for
you
and
will
expose
your
app
inside
the
cluster
or
outside
the
cluster.
Based
on
some
specification
like
what's
your
application
port
and
what's
the
internal
and
external
domain,
you
want
to
expose
the
app
that's
it
and
plugger
does
on
all
the
rest
and
plus
then
at
least
is
is
the
way
we
call
get
ops
a
way
to
drive
configuration
for
kubernetes.
D
So
all
your
deployments
or
your
specs
or
young
walls,
these
whatever
there
will
be
in
a
git
repo
and
when
you
make
changes
that
git
git
repo,
those
changes
will
be
synchronized
with
the
cluster
and
I
will
be
showing
you
detox
pipeline
right
away.
It's
a
little
overview
of
what
Flagler
does
the
calorie
did
plan.
D
The
deployment
in
called
cannery
here
is
the
one
that
the
user
controls
you
are
in
control
of
that
you
you'll
create
this
deployment
and
based
on
that
deployment,
flagger,
for
example,
detects
if
your
deployment
has
secrets
or
config
maps
as
dependency,
Discover's,
all
all
the
diplomatic
dependencies
and
creates
a
clone
of
the
deployment
along
with
the
secrets
and
config
maps.
Why
it
does
that,
because
every
time
you
modify
something,
let's
say
you
modify
your
container
in
each
tag
or
you
modify
something
in
in
a
config
map.
D
That
change
will
go
through
analysis
phase,
so
you
are
in
control
of
the
cannery
and
that
can
is
not
exposed
to
your
clients.
The
primary
deployment
is
so
only
after
the
analysis
is
complete
and
if
everything
is
okay
only
then
any
change
in
mail
will
be
promoted
as
primary,
and
it
also
creates
all
these
cluster
IP
services
that
are
needed
by
the
underlying
service
measures
annotation
or
by
Ingenix
controller.
D
These
are
the
key
performance
indicators
that
tracker
uses.
So,
of
course
it
checks
the
deployment
health
status
by
looking
at
pods
and
all
this
stuff
it
has
built
in
metric
checks
for
request
success
rate
percentage.
What
that
means?
Let's
say
you
can
you
can
instruct
flagger
and
say
if
my
application,
if
one
percent
of
the
total
requests
are
resulting
in
a
five
hundred
response,
undo
the
undo
the
deployment
fail,
the
cannery
or
if
the
average
request
latency
is
above
five
hundred
milliseconds
I,
don't
want
to
deploy
this
in
production
roll
it
back.
D
Besides
these
two
built-in
matrix,
you
can
create
your
own
custom
checks
using
prom
care,
and
you
can
extend
validation
mechanism
with
web
hooks.
The
idea
behind
web
web
hooks
are,
let's
say,
flag,
will
call
a
web
hook
and
that
web
hook
will
start
an
integration
test
or
loading
test
smoke
tests
all
kinds
of
checks
that
it
will
be
impossible
to
just
put
all
this
logic
inside
flag
on
itself.
So
we
hooks
are
a
way
to
extend,
extend
this
mechanism
and
flagger
also
comes
with
the
web
hook.
D
D
This
is
how
the
cannery
CRD
look
looks
like.
So
there
are
some
discussion
at
the
beginning.
If
we
should
extend
built
up
on
the
on
the
deployment
stack
and
create
something
like
I,
don't
know,
cannery
deployment
object
that
inherits
all
the
all
the
deploy.
All
the
kubernetes
deployment
fields
and
I
started
with
that
idea,
and
then
I
I
backed
up
and
I,
backed
it
up
and
I
said:
okay,
we'll
create
the
can
you
see
and
inside
the
canary
she'll,
be
here
friends
a
diploma?
Why
are
you
that?
D
E
D
This
is
how
the
law
notice
looks
when
flagger
detects.
There
is
a
canary
change,
it
starts
the
load
test
and
afterwards
it
looks
at
metrics.
Just
fill
it
up
to
give
you
the
demo.
So
this
is
what
I'll
be
showing
you
I
have
a
deployment,
a
canary
object
and
a
horizontal
pod
of
a
scalar
definition,
my
deep
triple
when
I'll
do
a
modification
I
have
installed
on
my
on
my
cluster.
We
flux,
which
sinks
the
get
cluster
with
the
git
repository.
The
cluster
then
flag
a
defect.
D
4.1
I'm
gonna
commit
this
change
and,
like
half
of
me,
there'll
be
replicated
on
the
cluster
here.
On
this
side,
we
can
see
it's
a
tale
of
flagyl
logs
and
the
graph
arnolditch
world.
What
it
shows
on
on
the
on
the
left
side
is
the
primary
deployment
which
is
running
right
now
and
BC
its
traffic
from
this
one
and
on
the
right
side,
is
the
cannery
which
now
it
is
scaled
to
zero.
So
there
is
no
traffic
going
to
this
deployment
that
you
have
control
on
now.
D
D
It
checks,
the
the
health
check
it
waits
for
the
horizontal
photo
to
scalar
to
scale
it
up
if
useful
and
afterwards
he
starts
to
route
traffic
little
by
little
to
the
knee
instance.
So
now
it's
started
with
5%
and
before
it
started
it,
it
also
start
with
the
low
test.
As
you
can
see
now,
my
my
application
will
change
will
be
balanced
between
primary
and
under
Kennedy.
What
I
can
do
I
can
generate.
Let's
say
some
500
errors.
D
Now
Flagler
has
detected
that
the
the
number
of
errors
the
percentage
of
the
500
turns
in
the
last
minute,
is
over
99%
and
what
it
does
it
holds
the
the
roll
out.
So
the
traffic
right
now
is
at
20%.
If
there
is
continued,
let's
say
for
for
10
iterations,
it
will
roll
back
the
cannery
and
it
will
scale
it
down
to
zero.
If.
D
If
yours
will
disappear,
I
have
so
I'm
letting
you
have
a
toleration
here.
Let's
say:
I
accept
some
words,
but
if
it
fails
more
than
ten
time
or
if
it
spells
more
than
a
couple
of
minutes,
then
roll
it
back
so
hopefully
to
to
come
to
100%
and
you'll
continue.
Also
I
can
also
generate
latency
and
we'll
see
the
same
behavior.
It
will
pause
the
traffic
shift
and
it
will
wait
for
the
latency
to
go
below
the
threshold.
D
D
I've
said
the
canary
now
to
10-second
interval,
so
I
can
demo
it
faster,
but
in
a
production,
environment
you'll
probably
want
a
couple
of
minutes.
So
you
have
enough
data
in
Prometheus
to
make
the
right
decisions.
Okay,
now
it
hit
the
the
maximum
threshold
which
is
set
to
50%
traffic.
So
what
time?
What
I'm
saying
now?
If,
if
my
canary
can
do
50%
of
the
whole
traffic-
and
there
are
no
errors,
then
it's
okay
and
it
will,
it
will
be,
it
will
replace
the
production
to
plan
on
the
primary.
D
D
Why
I'm
not
using
the
cannery?
Why?
Let's
say
if
I
each
it
goes
to
50%?
Why
not
just
give
it
100%
and
and
kill
the
primaries,
because
I
don't
want
the
end
to
have
control
over
what's
running
in
production
and
that's
why
the
primary,
which
is
not
in
your
control,
is
owned
by
flagger.
That's
the
one
that
is
exposed
outside
of
the
canary
deployment
window
and
what's
happened.
D
D
Happened
now
the
old
pods
were
replaced
with
the
new
pods,
with
the
speck
from
the
from
the
cannery.
The
new
version
here
is
no
longer
recurring,
I'm
4.1.
Now
everybody
is
on
the
new
version
and
the
kennedy
deployment
has
been
scaled
to
zero,
but
this
is
how
how
flagger
does
general
deployments
now?
The
problem,
you
probably
notice
that
my
screen
has
triggered
right,
so
I
was
jumping
from
between
versions.
Normally
you
want.
C
D
D
A
Then
I'll
make
sure
we
put
a
link
to
the
slides
in
there,
so
you
can
follow
up
a
Stefan
if
you
have
questions
because
we
got
to
get
moving
here,
timewise
thanks
a
lot.
Moving
on
the
next
is
the
contributor
tip
of
the
week.
I
wanted
to
link
something
that
Christoph
Blocher
sent
to
the
contributor
experience
list
that
we
want
to
start
taking
advantage
of
the
github
status
on
github.
A
You
can
say:
I
am
NOT
at
work,
I'm
away
and
tie
that
into
our
two
links
so
that
when
it's
randomly
picking
who
to
assign
reviews
to
in
the
owners
files,
currently
that's
just
a
random
thing.
He
wants
to
make
it
so
that,
if
you're
on
travel
or
something-
and
you
set
your
github
status,
he
can
we
could
do
a
smarter,
Wade
assignment
and
review
things.
So
click
through
to
read
that
and
he
wants
to
turn
that
on
by
Wednesday.
So
if
you
can
give
him
some
feedback
this
week,
that
would
be
great
moving
on.
A
C
Hey
thanks
George,
so
I'm
gonna
go
ahead
and
share
my
screen
here.
That's
okay,
great!
So,
first
of
all,
thank
you
for
thank
you
to
Steven
for
reaching
out
and
seeing
if
I
would
present.
This
is
a
pretty
great
honor
and
I
really
appreciate
it.
This
is
my
first
cap,
so
speaking
of
Canaries,
you
know
this
is
a.
C
Maybe
you
can
give
me
some
feedback
on
how
I
did
and
I
built
off
all
the
other
cups
that
I
saw
and
and
took
a
stab
at
it,
presented
it
yesterday
to
cluster
lifecycle
or
sorry,
the
cube
ADM
working
group,
and
there
was
a
lot
of
enthusiastic
discussion.
So
you
know,
hopefully
this
will
move
ahead.
Just
real
quick
note,
I
did
put
this
for
1.15
I
know
it
was
closed
and
we'd
have
to
file
an
exception.
C
Cube
ad
and
working
group
decided
that
this
is
it's
probably
a
p2,
so
we're
going
to
try
to
do
alpha
in
116
instead,
I
just
haven't
updated
that
yet
here.
So
if
you
see
it,
you
you
don't
need
to
ask
us
now.
You
know
so
a
real,
quick,
TL
DR
on
what
this
is
currently
cube.
Abm,
when
you
do
a
cube
ADM
in
it,
it
spits
out
this
wall
of
text
and
you
get
a
cube.
C
Abm
join
at
the
bottom,
gonna
join
other
nodes
or
you
want
to
create
an
H
a
closer,
and
there
are
other
commands
that
also
that
also
emit
text
that
change
over
time
that
the
output
changes
and
that
a
really
good
example
is
that
cube,
ATM
joined
commands
and
112.
There
was
no
white
space
in
front
of
that
and
113.
There
were
two
white
space
characters,
two
spaces
and
what
114
those
spaces
were
gone.
I
was
building
some
automation
and
it
was
breaking
with
113
and
come
out
figure
out.
C
Why
and
that
was
it
so
I
was
like.
Oh
okay,
well,
I
wonder
if
there's
a
way
to
make
this
a
little
bit
more
structured
and
funny
enough
and
I,
don't
know
if
he's
on
the
call,
but
a
gentleman
by
the
name
of
Edward
bartosh
from
Intel
over
in
Finland
had
been
working
on
this
very
topic
and
so
I
reached
out
to
him
and
we
sat
down.
We
had
a
chat
and
decided
you
know.
Maybe
there
have
been
a
couple
of
efforts
at
this
and
they
haven't
really
been
successful.
C
So,
let's,
let's
try
to
formalize
this
into
a
cap,
so
we
worked
on
this
together
in
the
feast
not
on
the
car.
You
want
to
say
thank
you
very
much.
It
was
a
pleasure
with
with
whom
to
work
with
which
to
work,
so
the
motivation
is
pretty
simple.
You've
got
tooling,
like
terraform
upcoming
cluster
API
project
that
all
need
to
be
able
to
leverage
cube,
ATM
and
parse
its
output,
because
you
can't
always
invoke
cube
a
diems
library.
C
Sometimes
you're
gonna
be
running
it
in
cloud
in
it,
and
all
you
have
is
the
texts
or
other
like
ignition
for
core
OS.
For
example.
All
you
have
is
the
text,
so
it's
very
important
that
you
can
parse
that
output
in
a
deterministic
way.
So
up
until
the
can
the
cap
it
was
referred
to
as
a
machine
output,
I
started
trying
to
kind
of
convert
people
saying
structured
out,
but
because
I
think
that
is
really
the
important
thing
to
take
away
here.
It
needs
to
be
structured
and
deterministic
and
and
reliable.
C
So
ultimately,
the
goal
is
that
cube
ATM
should
support
structured
output,
including,
but
not
limited
to
and
I'm
not
going
to
go
over
all
the
different
types
you
can
see.
Read
them
yourself,
how
we
divide
it
and
to
buffered
and
unbuffered.
That
is,
that
is
a
little
different
than
you
might
think.
It
doesn't
mean
necessary.
C
It
doesn't
mean
necessarily
that
it's
streaming
every
piece,
every
line
that
you
get
just
so
people
understand
what
buffered
and
unbuffered
mean
here
is
a
quick
example
using
both
a
text
printer
and
a
JSON
printer
buffered
means,
if
you've
got
a
command
like
token
list
that
emits
ten
tokens
buffered
would
wait.
The
command
would
wait
until
all
ten
tokens
were
known
and
then
emit
them
all
at
once.
I'm
buffered
would
mean
admit
the
tokens,
as
you
discover
them
same
same
with
JSON
right.
C
You
can
emit
JSON
objects
as
you
discover
them
and
when
they
do
get
emitted
to
standard
out
it
actually
isn't
legal
JSON,
because
it
isn't
part
of
an
array,
but
there
are
ways
to
handle
it
see
there
that
buffer
is
part
of
a
number
I'm
buffer.
It's
not,
but
you
could
take
that
output
and
there
are
tools
that
know
how
to
handle
that,
such
as
JQ
anyway,
so
the
goals
are
structured
output
for
the
following,
but
not
limited
to
list
of
commands,
so
the
non
goals.
C
This
is
where
it
actually
gets
very
interesting,
and
we
had
a
long
discussion
about
this
yesterday
and
humidity,
and
working
group
I
need
to
go
through
and
update
the
cat,
because
I
think
there's
an
agreement
on
the
path
forward.
Now
the
first
question:
a
lot
of
people
ask
themselves
when
they
read
this
was
hey
Andrew
edy.
Why
didn't
you
use
CLI
runtime?
That's
what
Kim
kuddle
uses
CLI
runtime
has
a
printers
package.
Why
would
you
reimplemented.
C
All
three
of
these
links
go
to
the
same
explanation.
That
printers
package
requires
the
objects
that
are
being
printed
to
be
of
type
runtime
object,
which
means,
ultimately,
the
output
is
version.
Initially
we
thought
well.
That
actually
is
the
goal
right.
We
wanted
to
be
dependable.
Reliable
versioned
output
makes
sense,
but
we
also
did
not
want
to
in
define
out
of
the
gate
that
much
technical
debt
to
get
this
problem
solved.
C
C
So
I'm
going
to
be
updating
the
cap
today
or
tomorrow
to
reflect
that
that
the
initial
the
initial
design
of
this
is
going
to
be
the
translate
the
cube
ATM
commands
to
emit
version
two
objects
now
because
cube
ATM
is
already
in
beta
in
terms
of
the
the
API
version
right,
the
objects,
the
the
output
objects
would
belong
to
a
new
schema
that
that
is
an
alpha
about
two
minutes
Andrew.
Thank
you.
C
C
Adm
you'll
have
the
ability
to
output
as
go
templates
and
be
able
to
get
the
ca
Suresh
the
the
node
that
you
want,
as
well
as
all
of
the
different
tokens
and
yeah
I,
don't
know
how
much
more
to
go
over
really
only
two
risks
that
were
relevant
versions:
outputs,
no
longer
really
a
risk,
and
then
the
tab
writer
is
going
to
still
be
supported
or
the
tabular
output
will
still
be
supported
through
a
typewriter
and
yeah.
The
graduation
criteria
should
be
pretty
straightforward.
C
C
A
Cluster
lifecycle;
okay,
great!
Thank
you,
Andrew,
okay,
moving
on
the
sig
updates,
we
have
three
today.
Unfortunately,
me
she
had
to
cancel
last
minute.
However,
she
did
have
her
slides
from
the
sig
AWS
update
and
I've
gone
ahead
and
put
the
link
to
Justin
Santa
Barbara
sig
intro
session
at
this
upcoming
Q
con
cloud
native
con.
F
You
hello,
everyone
was
on
mute.
Can
you
see
my
slides
okay
good
to
go
alright
first
things?
First,
for
those
that
don't
know
me,
my
name
is
Paris
Tappan
I
work
at
Google.
I
am
a
co-chair
of
this
awesome.
Amazing
cig
called
contributor
experience
that
many
of
you
are
either
a
part
of
or
not,
but
still
because
you're
a
contributor.
So
luckily
for
us,
we
actually
have
a
planned
intro
at
cucumber
cilona,
so
these
slides
are
going
to
be
seen
there
as
well.
F
So
I
think
everybody's
going
to
get
a
little
preview
here
so
enjoy
contributor
experience
is
expanding,
which
is
awesome,
we're
scaling
out.
We
have
my
own
stones,
they
have
sub
project
boards.
This
is
how
we're
starting
to
organize
how
we
do
business
and
we
also
have
a
new
way
of
contacting
us
through
contributors
at
kubernetes
at
I/o.
This
is
part
of
the
G
suite
instance
that
the
Assyrian
Committee
has
and
we're
actually
looking
to
expand
that
and
I'll
talk
about
that
in
a
second
as
well,
but
we're
also
building
teams.
This
is
super
cool.
F
This
is
something
you're
saying
should
do
as
well.
I'll
get
to
that
into
a
second
too,
but
these
are
areas
that
we
feel
needed
some
attention
and
needed
some
some
plant
mentoring
opportunities
and
things
like
that.
So
we're
really
not
a
triage
team.
This
would
include
like
PR,
Wranglers
issue,
triage
captains,
etc,
an
events
team,
a
marketing
team
and
so
much
more.
It's
really
cool
stuff.
That's
going
on
with
our
crew.
You
should
definitely
get
involved.
If
this
is
something
that
interests
you,
we
have
26,000
contributors.
F
That
definition
would
be
based
on
people
who
file
issues,
PRS,
merge,
stuff,
etc.
So
yeah
come
and
help
us.
We
aren't
pausing
on
our
APEC
meetings,
though
this
was
something
that
we
tried
as
we
expand
in
different
markets
for
contributor
base.
We
feel
like
each
market
might
be
keen
on
having
someone
there
so
that
they
can
either
have
meetings
or
educate
folks
on
how
to
contribute
or
how
to
get
involved
things
like
that.
F
So
if
this
is
you
or
you
know
someone
who
would
like
to
do
this
in
a
pack
or
get
involved
with
a
pack
growing,
a
pack
et
cetera,
there
is
a
link
to
the
issue
on
this
deck.
This
deck
is
full
of
links
and
you
will
obviously
anybody
and
everybody
can
have
it-
can
have
access
to
the
deck
at
a
later
time.
F
So
one
of
the
sub
projects
that
we
work
on
is
mentoring,
we're
also
getting
into
succession
planning
and
contributor
growth,
anytime,
I,
say,
mentoring,
people
or
like
kind
of
go
and
call
in
a
corner
or
say:
oh,
my
gosh
I,
don't
have
any
time
or
say:
oh
my
gosh
I'm,
so
overwhelmed.
No
really,
this
stuff
is
important.
Definitely
listen
up
the
reason.
Why
is
we
don't
build
programs
based
off
of
old
concepts,
meaning
one
on
one
and
other
time,
burdensome,
mentoring,
tactics
and
concepts
for
building
programs
around
you
as
personas
your
time?
F
Building
easy
succession
plans
two
roles.
We
also
build
them
with
different
learning
styles,
so
documentation,
videos
not
just
words
and
then
also
delivery.
So
this
will
be
face
to
face
as
well
as
online.
What's
live
now,
we've
got
a
speed
mentoring
to
chat
to
sessions
at
Barcelona
this
just
opened.
Last
year
we
did
over
150,
we
had
150
people
there,
everybody
gets
special
swag
aka.
If
you're
a
mentor.
F
So
it's
important
we
pair
each
other
up
based
off
of
excuse
me,
another
great
opportunity
that
we
have
coming
up
is
the
Signet
greet
at
the
new
contributor
workshop
that
we
have
it's
at
Barcelona
as
well,
where
sakes
come
with
good
first
issues
and
folks
who
are
both
new
to
open
source
as
well
as
not
new
to
open
source,
especially
open-source
maintainer,
SATCOM
I
will
be
paired
up
with
those
folks
immediately.
F
Google
Summer
of
Code
is
also
just
launching
CN
CF
had
19
interns
approved
kubernetes
had
six,
so
we
will
have
six
starting
and
we'll
have
all
of
them
present
at
this
community
meeting
like
we
have
in
the
past
and
that's
the
super
fun,
and
then
we
also
have
meet
our
contributors.
This
is
a
once
a
month
program
where
we
just
have
general
Q&A
with
upstream
mentors,
who
want
to
help
people
with
their
journey,
whether
they're,
new
or
current.
F
We
actually
get
and
having
getting
more
current
contributor
questions
as
of
late,
especially
around
time
management,
and
how
to
become
an
approver
or
reviewer
how
to
get
into
an
owner's
file,
what's
an
owner's
file,
etc.
So
it's
really
super
neat
stuff.
The
playlist
link
is
in
here
we
service
anywhere
between
a
hundred
to
a
thousand
people
per
episode,
and
this
includes
code
base
tours
and
all
kinds
of
fun
stuff.
F
So,
maybe
more
about
that
on
the
next
update
succession
planning
via
teams
and
roles,
are
happening
again.
I,
just
I
briefly
touched
on
that,
but
we're
going
to
start
to
try
to
think
about
this
programmatically
and
see
how
we
can
implement
in
six
across
the
board
so
more
stuff
on
that
next
cycle
and
then
mentoring
if
scale
initiatives.
What
does
this
mean?
This
means
you
are
mentoring,
multiple
people
at
once
again,
not
one-on-one,
and
we
do
this
through,
like
tape
code
based
tours,
we've
already
vinokurov
any
scoober
Nettie's
we
just
did.
F
We
just
did
cube
control.
We
also
did
CSI
if
your
team
thinks
that
they
could
benefit
from
this
or,
if
you're
listening
out
there,
and
you
think
that
you
can
benefit
from
a
code
based
tour.
We
have
an
issue
linked
in
here
to
collect
suggestions
on
what
folks
want
we'll
probably
do
the
scheduler
and
a
couple
other
things
shortly
as
well.
We're
also
doing
live,
API
code
reviews,
triage
triage
sessions,
etc.
Why
is
this
important?
F
This
is
important
because
it's
also
showing
people
how
to
do
the
thing
that
you
do
just
so,
naturally
as
a
reviewer,
approver
or
maintainer,
and
then
we're
also
doing
new
contributor
office
hours
very
targeted
sessions
for
certain
things
that
are
looking
for
certain
kinds
of
contributors
like
for
examples
to
cluster
life
cycle.
Just
did
one
so
definitely
get
in
touch
with
us.
If
any
of
this
stuff
sounds
interesting
to
you
or
your
cig,
automation
and
github
management,
baqia
are
you
on
the
line
yeah.
G
So
yeah,
so
in
general,
we
are
looking
to
improve
the
automation
around
owner's
files,
because
it's
no
secret
that
we
have
an
insane
amount,
of
course,
files
across
all
our
humanities,
github
repos.
So
we
wanted
to
do
an
audit
of
these
files,
especially
in
proven
Eddie's
queries
report
to
make
sure
we're
keeping
these
files
up-to-date
and
also
that
we
are
promoting
you
folks,
who
are
doing
the
reviews
today.
G
G
The
other
thing
that
we
have
is
a
gorge
already
mentioned
about
this.
We
want
to
enable
a
feature
in
so
so
how
owners
files
actually
works
is
bad,
the
people's
listed
as
reviewers
in
an
owner's
files
or
requests
to
review
whenever
a
peer
comes
in,
but
if
the
reviewer
is
probably
on
vacation
or
apologies
that
if
they're
on
vacation
are
just
busy
and
they
set
up
their
github
status
as
busy,
we
will
not
request
a
review
from
them.
There
is
a
mailing
list
thread
on
the
second
Trebek's
contributor
experience
mailing
list.
G
If
you
have
any
feedback
or
questions
about
that,
please
chime
in
there.
This
will
also
be
posted
later
on
to
the
kubernetes
tab
mailing
list.
But
if
you
have
been
a
first
round
of
feedback,
we
would
be
more
than
happy
to
hear
it.
The
next
round
of
improvements
I
could
want
to
bring
on
is
making
sure
that
all
users
who
are
ads
being
added
in
earnest
files
are
actually
auric
members,
because
only
oil
members
can
do
an
LG,
TM
or
an
approved
anyway.
G
So
we
want
to
make
sure
that
all
of
like
just
to
make
sure
that
we
have
or
nurse
files
hygiene
can
you
go
to
the
next
slide.
Parris
cool,
so
we've
also
been
working
on
some
general
stream
whining
and
enabling
automation
across
all
our
github
boards.
So
we've
enabled
the
fade
about
issue
lifecycle,
automation,
if
you're
not
already
familiar
with
beta
bot,
it
will
add
a
comment.
It
will
add
that
comment
and
delete
label.
G
Excuse
me
which
says
that
an
issue
is
marked
as
lifecycle,
stale
or
and
then
as
website
cold
water
and,
ultimately
close
it.
If
there
hasn't
been
an
activity
for
a
certain
number
of
days,
we
enable
the
needs
to
be
based
again
across
all
arcs,
which
will
automatically
detect
if
a
PR
has
quick.
If
the
branch
has
conflicts
and
add
a
label
which
says
I've
made
three
base
and
also
add
a
comment,
so
the
other
comes
to
know
about
it.
We've
also
enabled
a
trigger
plugin
across
all
arcs.
G
So
we
want
to
make
sure
that
we
don't
people
and
we
won't
have
less
emails,
can't
go
to
the
next
site,
yeah
thanks.
So,
regarding
github
management,
we
so
contribute
experience.
Also
owns
going
to
have
management.
We
also
have
a
github
admin
team
which
who
have
owners
well
or
governments
for
all
our
communities,
github
arts
we're
also
responsible
for
creating
new
or
curating
your
repos
renaming
deleting
archiving
all
of
them.
So
since
the
last
time
we
had
a
contributor
experience,
update
and
the
community
meeting,
we've
created
almost
30
new
repos.
G
We
also
have
process
membership
requests
for
anyone
who
wants
to
join
a
communities.
Github
org,
a
huge
shout
out
to
Bob
and
Stefan
Bob,
is
on
this
call,
I.
Think
who
process
membership
requests
every
single
day.
This
is
a
lot
of
hard
work
and
we
also
have
team
management
through
config.
It's
a
Yama
file
in
Quran
ease
slash
our
grip.
Oh
please,
please!
Please
use
this
config
if
you
want
to
get
added
to
a
team
or
if
you
want
to
create
a
new
team
and
so
on.
G
Please
do
not
use
the
UI,
because
all
of
like
the
source
of
truth
is
this
config.
We
will
also
be
having
a
way
in
a
process
so
that
sub
projects
can
request
their
own
websites,
for
example,
kind
and
cube
builder.
Have
their
own
websites
today
same
journals?
Our
projects
can
have
that
and
we
will
have
a
proper
process
for
it.
G
That's
already
a
PR
open
again,
a
huge
shout
out
to
Bob
who
created
that
and
the
github
admin
team
will
have
access
to
them,
met
by
admin
stuff
and
will
be
helping
and
setting
up
the
necklace.I
configuration
so
that
all
sub
projects
can
create
a
website
easily
and
I.
Suppose
that's
it
for
automation
and
github
management.
Stuff,
yes,.
F
F
So
I'm
gonna
fly
through
these
because
I'm
looking
at
I'm
being
trying
to
be
cognizant
a
time.
We
also
have
several
other
sub
projects.
Community
management
events,
somebody's
chatting
me
now
commence,
communication,
etc.
One
of
those
being
slack.
That's
what
sort
of
like
get-ups
we're
doing
here.
The
special
thanks
to
Catherine
berry
for
that
she's
been
working
super
hard
on
just
making
slack
better
for
us.
We've
added
the
report
feature.
F
So
you
can
report
people
who,
like
are
breaking
code
of
conduct
and
then
also
different
kinds
of
additions
where
we
can
add
slack
channels
and
all
kinds
of
cool
things.
The
issue
of
PR,
oh,
that
isn't
the
deck
check
it
out.
Super
cool
stuff
we've
also
been
working
on
redesigning
the
community
site
which
launched
I'll,
go
ahead
and
do
a
little
live
deal
here.
So
we've
got
code
of
conduct.
F
We've
got
our
videos
with
our
in
our
different
playlist
themes,
so
office
hours,
community
meetings
like
this
one
and
then
like
meet
our
contributors
and
other
random
bits
and
stuff,
like
that,
our
discussion
forums.
Why
we
discuss
here
what
we
discuss
here
upcoming
events
find
a
meet-up
and
some
news,
so
that's
the
redesigned
community
site
and
then
for
community
more
community
management
stuff.
We
also
help
SIG's
and
working
group
spring
to
life
and
retire.
F
A
lot
of
working
groups
have
completed
their
journeys
as
of
late,
so
we've
been
helping
them
like
I,
guess,
D
up
or
D
operationalize,
if
that's
even
a
thing
I
just
made
or
if
I
made
that
word
up
and
we've
also
been
building
community
around
chairs
and
tech
leads
for
a
number
of
different
reasons.
First
and
foremost,
training
need
to
know
emails
for
information
that
needs
to
get
disseminated
to
humongous
SIG's
of
like
thousands
of
people
and
then,
of
course,
a
slack
channel
as
well.
F
Of
course,
we
do
events
we
have
one
coming
up
in
Barcelona.
We
definitely
have
seats
available
for
new
contributor
workshop.
If
you're
listening
to
this
and
have
friends,
family
co-workers,
who
would
like
to
know
what
it's
like
to
be
a
kubernetes
contributor,
this
is
your
chance-
will
also
have
opportunities
in
Shanghai
and
San
Diego
as
well.
Read
up
on
that.
F
Also
thank
you
to
our
all
of
our
moderators
yep,
more
than
doubled
in
size
and
all
of
our
platforms
also
be
zoom.
Youtube
slack
mailing
list
discussed
kubernetes
and
so
much
more.
We
have
some
cool
YouTube
stats
that
we'll
share
to
share
with
you
throughout
this
community
meeting
in
the
next
couple
of
months.
F
Contributor
documentation
is
always
on
our
forefront,
we're
always
reviving
the
living
editing,
making
better
the
contributor
and
developer
guide.
Even
our
outreach
e
stuff
outreach
intern
édouard
stuff
around
that
cool.
So
that
was
awesome.
What's
our
future?
Look
like
it
looks
real
bright.
We
are
trying
to
a
couple.
A
couple:
Google
Aaron,
particularly
I,
would
love
to
title.
Would
love
to
tie
cigs
Amal
to
all
the
things
we
have
a
contributor
site,
that's
being
worked
on,
the
link
is
in
there.
You
can
file
issues
against
it
as
well.
F
Once
we
once
we
go,
live
we're
continuing
to
build
mentoring,
programs,
we're
gonna,
give
out
another
contributor
experience
survey,
even
more
events,
formal
rewards
and
recognition
programs
and
even
more
training
how
you
can
contribute.
We
need
you,
you
can
either
be
a
moderator,
be
a
mentor
help
our
sake
spread
communications
far
and
wide
use
good.
First
issue:
Help
Wanted
labels.
Ask
your
sakes
what
they're
doing
for
mentoring,
chairs
you
can
upload
your
YouTube
meetings
fix
your
zoom
settings.
Don't
forget
about
the
your
community
meeting
updates
or
to
find
us
have
a
nice
day.
E
Hi
everyone,
my
name,
is
Bobby
salamat
I
am
co-lead
of
cig
scheduling
and
with
that,
let's
jump
on
sick
updates.
So
in
1:14
the
major
theme
for
114
114
for
scheduler
was
performance
and
stability
of
the
scheduler.
We
did
multiple
algorithmic
optimizations,
resulting
in
almost
3x
performance
improvement.
So
back
in,
like
2000
September
of
2018,
we
could
do
about
35
parts
per
second
in
5000
node,
cluster
x'
in
February
of
2019.
We
reached
a
hundred
parts
per
second
in
5000,
not
clusters.
This
was
a
huge
improvement
and
a
big
milestone
for
us.
E
Thank
you
for
our
thanks
to
our
contributors
for
all
their
work.
Another
important
feature
that
we
graduated
to
stable
in
114
was
potpie
orient
preemption.
This
was
this
was
actually
a
highly
requested
feature
from
a
large
number
of
users
and
is
already
being
used
by
various
parts
of
kubernetes
ecosystem,
as
well
as
our
user
workloads.
If
you
are
interested,
you
can
find
a
lot
of
documentation
about
how
to
use
this
feature
chances
are
you
may
already
using
it
actually
and
in
fact,
in
more
more
recent
version
of
versions
of
kubernetes.
E
E
What's
up
in
115,
the
main
theme
for
115
is
workload
workload,
reliability
and
scheduler
extensibility.
So
for
improving
reliability
of
workloads,
we
are
delivering
a
new
feature
called
even-odd
spreading
with
even
pod
spreading.
You
can
specify
how
your
parts
should
be
spread
among
different
failure.
Domains
example,
you
can
say
I
want
my
prize
to
be
spread
among
zones
or
among
notes
or
regions
or
whatever
you
know.
You
can
also
specify
a
custom
label.
What
is
different
between
even
participating
versus
anti
affinity
that
existed
before
is
that
anti-affirmative
are
more
accurately
hard-hard
anti
affinity.
E
Allow
you
to
put
one
part
in
a
particular
failure
domain.
For
example,
if
you
specify
that
your
web
servers
have
anti
affinity
to
one
another
on
the
same
node,
then
no
more
than
one
web
server
could
run
on
the
same
node
or
the
same
thing.
If
you
wanted
to
one
web
server
per
zone,
no
more
than
one
web
server
could
run
into
it
in
that
zone.
But
now,
with
even
party
spreading,
the
scheduler
can
look
at
the
number
of
zones
and
look
at
the
number
of
parts
and
spread
them
evenly
among
those
zone.
E
So
you
can
have
more
than
one
part
per
zone
or
whatever
failure
domain
you
specify.
We
are
also
working
on
improving
scheduling,
extensibility
and
customizability.
Basically,
we
are
creating
and
building
the
scheduling
framework.
If
you
are
not
familiar,
there
is
a
cap
for
that.
The
scheduler
framework
changes
the
scheduler
architecture
from
a
monolithic
architecture
to
a
pluggable
one.
Alpha
version
of
this
is
targeted
for
115.
We
are
building,
as
we
speak,
some
of
the
extension
points
for
this
scheduler
that
allows
plugins
to
be
built
for
these
extension
points,
and
we
have.
E
We
are
adding
a
new
configuration
mechanism
for
these
plugins
as
well,
so
we
expect
to
have
the
whole
scheduling
framework
built
in
116
between
115.
We
are
going
to
have
a
lot
of
important
extension
points
as
well
as
configuration,
so
we
are
also
working
for
better
pathway.
Early
support
for
batch
workloads
in
patch
workloads.
Priority
is
not
necessarily
putt-putt.
Priority
is
not
necessarily
always
great
and
the
way
that
it
is
today.
E
So
if
you
have
a
batch
workload
with
a
high
priority,
chances
are
that
your
workload
can
go
and
preempt
other
batch
workloads
when
batch
work
are
preempted.
All
the
work
done
by
those
were
closed.
Most
of
the
times
is
basically
gone,
and
you
have
to
start
from
scratch
again
next
time
that
the
workload
starts.
E
So,
instead
we
are
introducing
a
non
pre-empting
priority.
The
benefit
of
this
priority
is
that
it
actually
helps
those
worthless
first
of
all
not
get
preempted
by
other
workloads
when
they
are
running
and
also
if
they
are
not
running,
and
if
they
have
higher
priority.
You
go
to
the
head
of
the
queue,
but
they
don't
preempt
other
existing
workload.
This
is
needed
for
batch
processing.
We
are
expecting
to
have
this
feature
in
115.
E
So
how
these
plans
affect
you,
these
changes
are
expected
to
be
fully
backward-compatible.
The
only
thing
that
we
expect
to
happen
is
that
cluster
August
Kellerman
may
not
necessarily
work
very
well
with
the
scheduling
framework,
because
this
is
scheduling.
The
scheduling
framework
allows
the
scheduler
to
be
very
customizable
and
I
could
change.
Basically,
the
behavior
of
this
edge
Rick
could
change,
depending
on
the
configuration
and
the
plugins
that
you
use
so
cluster
autoscaler
has
this
mechanism
that
predicts
the
scheduler
decisions
and
that
logic
may
not
necessarily
work.
E
Well,
when
the
scheduling
framework
is
out
because
the
scheduling
framework
is
not
necessarily
always
had
the
exact
same
behavior
anymore,
we
are
working
within
the
cluster
autoscaler
to
read
the
configuration
and
apply
these
configurations
to
the
cluster
descaler
logic
as
well,
so,
hopefully,
that
that
work
all
will
be
delivered
soon
as
well.
On.
E
A
Alright
thanks
everyone
just
a
few
announcements
here
before
we
wrap
it
up
test
great
alerts
for
release
blocking
jobs.
The
deadline
is
Tuesday,
May,
14th.
Also
a
reminder.
This
is
the
cube
con
carne
de
casa
es
p
for
North
America,
that's
the
one
in
December
is
open,
so
the
CFP
is
already
open
for
that
and
there's
some
shoutouts
for
the
week.
So
the
way
this
works,
if
you
see
someone
going
above
and
beyond
the
call
of
duty
mention
them,
has
shout
outs
and
we'll
mention
their
great
work
in
this
community
meeting.
A
Dimms
would
like
to
thank
des
mini
satya,
Jim
angel
Zac,
Arnall,
SPE,
Breck
and
Jill
Rondo.
Congratulations
on
your
Google
open
source,
your
bonus
win
with
your
work
on
kubernetes
and
there's
a
link
there.
Whenever
you
would
like
to
shout
out
and
said
sparkles
too
salty
SH
for
the
most
amazing
codebase
walk
through
to
get
me
ready
to
share
with
the
new
contributors
at
cube
con.
Thank
you.
So
much
and
Jeremy
WX
would
like
to
do
a
big
shout
out
to
Adelina
to
Veni.
A
I
hope
I
got
that
right
for
helping
me
with
an
aks
engine
problem.
After
banging
my
head
on
my
chest
for
most
of
the
week
she
pointed
out,
I
was
using
a
version
with
a
bug,
my
head
and
my
desk.
Thank
you
very
much,
and
with
that
do
we
have
any
last-minute
announcements
or
anything
anyone
would
like
to
mention
before
we
break
yes,.
F
H
Else,
yes,
I
have
a
quick
announcement,
siga
sig
note
and
sig
scalability
I
am
coming
for
you.
You
have
jobs
on
the
release,
blocking
dashboard
for
the
master
branch
of
kubernetes.
You
need
an
email
address
for
me
to
send
alerts
to
all
of
the
other.
Six
have
responded
to
my
email.
You
two
have
not
I
know.
I
gave
you
until
next
week,
but
I'm
coming
for
you.
That's
all.