►
From YouTube: What Goes in a Release: Kubernetes 1.14 in Code, Conversations, and Hearts - Maria Ntalla, Pivotal
Description
Join us for Kubernetes Forums Seoul, Sydney, Bengaluru and Delhi - learn more at kubecon.io
Don't miss KubeCon + CloudNativeCon 2020 events in Amsterdam March 30 - April 2, Shanghai July 28-30 and Boston November 17-20! Learn more at kubecon.io. The conference features presentations from developers and end users of Kubernetes, Prometheus, Envoy, and all of the other CNCF-hosted projects
What Goes in a Release: Kubernetes 1.14 in Code, Conversations, and Hearts - Maria Ntalla, Pivotal
During the run of a Kubernetes release, a huge group of people comes together to suggest features, agree on implementation details, develop, integrate and ship the next version of Kubernetes. People from all over the world, from various companies and backgrounds come together to make it happen. This talk will look at publicly available data from github, slack and mailing lists to tell a story about the makeup and diversity of the Kubernetes community, as well as the code, the interactions, the effectiveness and the dedication of the work that goes into building a Kubernetes release.
https://sched.co/MPZS
A
Hello
good
afternoon,
my
name
is
puia
from
transform.
I'm
gonna
be
the
host
for
this
session
and
the
next
one.
This
is
the
community
track
and
I
would
just
ask
you
of
one
thing.
We
have
a
Q&A
in
the
end,
if
you're
leaving
before
that
leave
very
quietly,
because
it's
gonna
influence
the
recording
and,
like
I'd,
ask
me
just
to
respect
speakers
that.
B
B
Alright
good
afternoon,
everybody
welcome
to
this.
So
thank
you.
So
much
for
stopping
by
I
know
that
you
probably
have
better
things
to
do
in
Barcelona
at
4:00.
I
really
appreciate
it
good.
Just
a
couple
of
words
about
me,
my
name's
Maria
and
I
am
a
software
engineer
and
engineering
manager
or
pivotal
in
London.
B
You
can
find
me
on
Twitter
and
I
am
Maria
very
creatively
on
the
kubernetes
slack
I've
been
working
in
the
open
source
humanities
community
for
about
a
year
now
in
a
number
of
different
areas,
but
mainly
in
cigarette
leaves
and
I
was
part
of
the
stake
so
part
of
the
release
team
for
the
last
two
or
three
releases,
I
look
a
bit
of
an
intro
to
the
talk
and
I
guess
like
the
motivation
behind
it.
I
personally
find
given
it
is
very
interesting
part
of
it
is,
of
course,
the
software
itself
in
the
technology.
B
B
I
also
find
it
very
interesting
from
a
perspective
of
I
guess,
like
human
collaboration
and
community,
and
what
I
mean
by
community
in
that
context
is
all
the
people
from
all
over
the
world
from
a
whole
different
bunch
of
companies
and
time
zones
that
in
some
way
or
another,
get
together
and
pretty
consistently
shape
working
software
every
three
months
without
fail
and
then
go
on
to
support
it
for
another
few
months.
So
it's
quite
curious
to
see
what
does
that
look
like
in
practice?
How
do
these
people
communicate?
How
do
they
collaborate?
B
B
Okay,
so
before
we
get
started,
I
just
have
a
very
wise
person
being
recorded.
I
have
a
couple
of
disclaimers
or
rather
reminders
that,
but
I
wanted
to
make
her
go
through,
be
the
content
head.
The
first
one
is
that
what's
coming
up
is
a
whole
lot
of
averages
and
summaries
and
I'm
very
conscious
of
the
fact
that
any
attempt
to
put
numbers
around
humans
and
their
behavior
is
a
very
inexact
science
and
therefore
will
probably
leads
to
imperfect
results.
B
B
Okay,
so
oversimplifying
slightly
kubernetes
is
largely
made
up
of
code
pretty
much
entirely.
So
my
first
question
was:
how
is
that
code
shaped
up?
How
does
it
change
through
three
time
I
figured
that
was
a
pretty
central
part
of
any
kubernetes
really
so
I
wanted
to
dig
into
it
a
bit
more,
so
the
kubernetes
codebase
is
organized
in
an
assortment
of
repos.
B
They
all
live
in
on
github
on
the
on
the
kubernetes
org
at
the
time
of
writing
this
at
this
point
in
time,
the
kubernetes
or
it
costs
sixty
eight
public
repos,
and
they
all
if
I'm
not
mistaken,
had
some
sort
of
activity
during
the
115
released
by
activity,
I
mean
either
issues
created
or
PR
is
open
and
I
don't
be
mistaken
and
not
all
humanities
related
activity
happens
within
the
kubernetes
org.
In
fact,
if
you
run
a
search
for
cuber,
Nettie's
arcades,
just
don't
get
hub,
you'll
come
back
with
I.
B
B
Now
over
the
course
of
114
over
a
thousand
thousand
38
issues
and
2047
PR
student
hands,
kubernetes
were
open
and
that's
not
counting
anything
that
was
opened
before
that
was
working
during
the
release
as
well.
So
that's
just
things
that
were
opened
during
the
those
three
months:
all
these
thousand
and
a
bit
issues.
Four
hundred
almost
half
of
them
are
still
open
today
and
about
three
hundred
of
the
pairs
that
were
opened
during
1:14
are
still
being
worked
on
today,
and
a
total
of
three
hundred
and
seventy
contributors
participated
in
in
opening
them.
B
Okay.
So
let's
look
at
the
activity
in
a
bit
more
detail.
As
we
all
know,
a
good
contribution
is
measured
by
the
lines
of
code.
End
of
sarcasm,
so
I
I
was
actually
quite
curious
to
see
just
how
like
what
the
size
and
the
complexity
of
the
contributions
themselves
where
and
there
were
close
or
close
to
12
million
lines
of
code
added
and
about
6
million
removed,
leaving
KK
about
five
and
a
half
million
lines
of
code
heavier
or
richer
in
lines
of
code.
B
B
B
Okay
in
terms
of
what
these
changes
were
a
bit
more
specifically,
it's
kind
of
hard
to
know
without
going
into
the
commit
tips
in
a
bit
more
detail.
But
there
were
a
total
of
31
thousand
total
files
change,
some
of
them
duplicates
or
it's
about
6,000,
unique
ones
and
of
those
two
and
a
half
thousand
were
calling
files.
Three
hundred
were
about
sixty
one
were
markdowns
recommendation,
usually
and
a
healthy
120
ml
files,
which
is
quite
I,
guess
usual
to
see
in
a
cloud
software
project.
B
An
additional
angle,
but
I
always
find
interesting
to
observe,
is
the
source
code
versus
test
codes,
balance
and
so
in
the
just
over
six
thousand
files
that
were
touched
with
PRS
during
one
fourteen,
two
thousand
and
two
hundred
of
them
had
tests
somewhere
in
the
name,
so
I
think
they
indicate
that
they
are
some
sort
of
test
at
some
level.
Looking
at
go
files
in
specific
among
of
the
two
and
a
half
thousand
go
files,
855
word
tests.
B
This
is
just
something
that
I
tried
to
put
together
to,
because
I
was
quite
curious
to
to
look
at
the
intention
behind
the
changes
that
would
like
the
code
is
always
interesting,
but
what
is
actually
changing-
and
this
was
this-
is
some
sort
of
the
stuff
that
pops
out
in
terms
of
the
titles
of
the
PRS
issues,
so
you'll
see,
there's
a
little
fix.
There's
a
lot
of
tests
as
well.
B
Of
course,
I
didn't
remove,
there's
quite
a
bit
of
whip
there
at
the
bottom
right,
which
is
all
fantasy,
and,
of
course,
I
was
also
I
kind
of
find
interesting
to
see
terrific
there
in
fact,
I
think
about
two
hundred
and
something
PRS.
So
just
over
ten
percent
of
the
pairs
that
were
opened
during
one
fifteen
were
cherry-picks
to
previous
round.
B
In
this
area,
as
well
across
the
over
3,000
issues
and
PRS
that
were
opened
during
130
at
114,
excuse
me,
there
were
18,000
plus
comments
that
made
up
conversations
underneath
them.
A
lot
of
them
came
from
bots.
Actually,
they
were
not
only
humans
and
both
sides.
You
may
have
seen
them
interacting
in
github.
Repos
are
very,
very
but
I
guess
like
they.
They
make
a
lot
of
comments.
They
they
take
a
lot
of.
B
They
take
care
of
a
lot
of
things
in
the
PRS
and
it's
pretty
common
to
see
activity
rates
of
up
to
like
50
comments
an
hour
from
the
KPI
row.
But,
for
example,
who
is
the
pace
of
the
urban
ABC
I
need
to
get
home
very
often.
What
that
made
me
remember
is
the
cases
that
opening
the
PR
is
not
the
the
end
of
the
game.
B
So
this
week
it's
not
the
end
of
the
conversation,
but
it's
typically
the
point
where
consensus
in
principle
starts
becoming
decisions
and
implementation
starts,
becoming
thinking
about
edge
cases
are
becoming
agreement
in
details
and,
in
fact,
PRS
that
were
opened
during
114
have
a
mean
duration
or
I
mean
like
time.
I
guess
from
from
open
to
and
our
marriage
of
about,
550
two
hours
so
we're
looking
at
north,
both
collectively
a
million
hours
of
work
quote
unquote.
B
B
Of
course,
it's
more
than
words
that
make
up
our
conversations
and
there
were
tons
of
not
word
reactions
to
to
messages
or
like
communication
in
non
word
matter
in
kkap,
RS
830.
Four
of
these
twelve
hundred
plus
reactions
where
it
comes
up,
oopsie,
double
click
finger
comes
out,
but
very
quickly,
44,
where
Lux
or
I
don't
know,
56,
prays,
41,
curious
eyes
and
141
where
hearts
a
whole
lot
of
the
conversation
happens,
as
you
may
have
guessed
all
on
slack,
the
Burnett
it's
like
lives
at
paid
starts
like
the
IO.
B
It
holds
its
267
public
channels
and
and
all
but
appeal
them
had
activity
during
the
114
timeline.
The
most
active
was
success.
Thing
followed
by
signal
ease
during
the
three
months
of
114,
a
grand
total
of
180,000
conversations
happened
and
slack,
and
if
the
number
sounds
somewhat
big,
keep
in
mind
that
it's
actually
summarized
so
it
holds
resonate.
It
holds
messages
that
were
posted
near
each
other
on
a
given
day.
So
I
think
we
could
even
multiply
that
by
three
and
it
wouldn't
be
far
off
the
actual
number
of
messages
that
were
posted.
B
Sixteen
thousand
messages
had
my
again
all-time
favorite
assertion
to
make
and
to
hear
which
is
I.
Don't
know
there
were
twenty
eight
thousand
occurrences,
oh
I,
like
it
or
I,
agree
agreeing
with
somebody,
but
then
exactly
fifty-six
results
of
I
disagree,
which
I
thought
was
perhaps
way
too
low,
319
shoutouts
during
the
114
timeline
and
129
bike,
shed
alerts
or
invitations,
I'm,
not
sure,
okay,
so
compared
to
flock
and
github
a
lot
less
than
still
pretty
significant
communications
happen
through
email.
B
During
the
114
timelines,
there
was
a
total
of
300
emails
or
there
abouts
that
were
sent
in
sig
mailing
list
special
interest
with
mailing
lists.
The
conversation
length,
as
we
measure
that
by
the
email
chain
length
varies
from
anywhere
between
just
the
one
email
to
the
broadcast,
announce
something
you
guys
should
show
the
actual
site
to
scribe
lengthy
conversations
with
around
50
50
to
60
emails,
but
averaging
at
around
3
to
4
messages.
So
it's
a
lot
I
guess.
B
Engagement,
perhaps
is
a
lot
less,
of
course,
compared
to
github
or
or
lock,
viewership
I
haven't
run
the
exact
numbers
on
how
many
people
open
and
read
those
emails
I've
seen
that
about
that.
Typically,
it's
around
that
20
to
30
mark.
But
of
course
it
really
depends
on
the
specific
topic
on
the
specific
sig.
So
it's
very
common
to
see
a
big
announcement
for
a
cigarette
project
to
actually
reach
hundreds
of
people
and
I
actually
wanted
to
run
the
same
thing
as
I
did
before
for
the
for
the
PR
titles.
B
But
this
time
for
the
email,
titles,
cuz
I
was
quite
curious
to
see
the
folks
talk
about
similar
or
different
things
in
emails
and,
as
you
can
see,
there's
a
whole
lot
of
things
around
design
like
it
kept.
Your
net
is
100
proposals
for
other
types
of
proposals
approving
things
and
then,
of
course,
talking
a
lot
about
meetings
and
invitations
to
edit
clients.
B
B
So,
moving
on
to
real-time,
a
total
of
140
sig
working
group
committee
and
other
public
meetings
happened
during
the
course
of
one
protein,
or
at
least
those
were
the
ones
that
I
could
find
somewhere
on
the
interwebs
and
they
a
month
to
a
combined,
almost
100
hours
or
synchronous
communication
viewership
for
the
recordings
ranges
between
0
and
around
300.
Again,
it
depends
on
the
meeting
and
the
content
on
the
agenda
and
there
and
the
second
how
many
members
it
has.
B
Last
but
not
least,
something
that
I
feel
is
not
as
visible
sometimes
is
that
it
wasn't
only
humans
that
did
the
work.
Machines
were
also
very
busy
during
the
114
release
outside
alongside
their
human
friends,
as
I
mentioned,
before.
A
pretty
significant
amount
of
messages
and
github
conversations,
for
example,
come
from
bots
that
helped
manage
issues
issue
and
PR
lifecycles
bots
take
on
everything
from
reminding
folks
to
sign
their
clas.
This
will
eventually
vote.
B
It
was
a
screenshot
of
Kate
CIE,
both
reminding
a
user
insider
feel
like
rerunning
known
latest
beta
if
I'm
pronouncing
his
name
correctly.
Bots
is
about
that
has
many
things,
but
one
of
them
is.
It
holds
as
far
as
I
know
context
as
to
which
which
test
dopes
are
known
to
be
quite
flaky
and
reruns
them
for
folks,
so
that
they
don't
get
stuck
on
on
their
PRS
I
find
even
posting
ponies,
which
in
fights
out
that
it's
not
going
to
load,
because
it
was
quite
a
lovely
picture.
B
Outside
github
computers,
also
the
the
heavy
lifting
of
patiently
building
and
testing
code
and
merging
contributions,
I
counted
a
total
of
a
thousand
one
hundred
and
twenty
five
different
jobs
test
jobs
that
run
at
some
point
of
the
life
of
code
that
you
contributed
to
kubernetes.
So
that's
like
either
pre
submit
before
your
code
has
been
merged
or
post
submit
or
periodically,
and,
of
course,
they
produce
a
whole
lot
of
arty
parts
logs
and
test
summaries
at
the
end.
B
So
the
numbers
here
are
a
little
bit
different,
partly
because
they
were
so
large
that
it
was
hard
to
weather
over
the
entire
day
later
than
the
first
two
weeks
kind
of
fun,
but
also
the
results
don't
go
as
far
back.
So
they
refer
to
the
last
two
weeks
from
today,
but
I
think
they're
comparable,
as
these
two
weeks
are
not
particularly
eventual
right.
B
So
over
over
two
weeks
there
are
a
hundred
and
sixty
one
thousand
and
a
bit
of
test
runs,
which
amounts
to
one
hundred
and
twenty
three
thousand
and
a
bit
hours
of
building
testing
testing
code.
Keep
in
mind
that
a
typical
release
runs
for
about
twelve
to
fifteen
weeks,
if
I'm
not
mistaken,
so
you
can
multiply
that
by
six
six
to
seven
times
for
the
course
of
out
release
opens
about
40%,
which
amounts
to
around
I
guess,
like
thirty
five
to
thirty
six
thousand
hours
a
week
where
failed
run.
B
B
B
That
is,
that
the
time
that
Universal
spent
building
the
code
with
the
time
that
humans
have
spent
putting
it
together,
okay,
so
now
I
would
like
to
go
back
and
answer
my
original
question,
which
was:
how
is
it
that
this
set
of
people
that
maybe
don't
personally
know
each
other,
they
maybe
haven't
said,
sat
next
to
each
other
ever
or
in
different
countries?
How
did
they
get
together
and
ship
software
every
three
months?
B
Secondly,
with
lots
and
lots
and
lots
of
conversation,
iteration
different
views
coming
in
the
game,
asking
questions
offering
different
viewpoints
and,
last
but
not
least,
with
a
whole
lot
of
help
from
their
robot
friends.
Thank
you
very
much
for
coming
and
let
me
know
you
have
any
questions
or
thoughts,
ideas.
B
Yep,
the
question
was
that
there
were
quite
a
pretty
much.
Every
community
meeting
is
recording
it,
but
I
thought
it
was
very
challenging
because
it
what
there
wasn't
an
archive
of
everything.
So
it's
it's
interesting
because
literally
every
video
is
recorded
and
everything
is
on
YouTube.
What
I
struggled
with
is
that
as
far
as
I
know,
there
isn't
a
list
of
all
the
videos
ever
recorded
in
one
place
and
they're,
not
they
weren't
named
in
a
way
there
isn't
like
a
conventional
way
of
naming
it
so
that
I
could
build
up
a
search
query.
B
What
that
meant
in
practice
is
I
was
like
well,
I
think
that
most
of
them
will
have
super
Nettie's
or
case
in
the
name,
the
hub
meeting
in
the
name
there
have
sig
or
working
group
or
WG
in
the
midnight,
and
so
I
switched
all
these
results
together
and
hopefully
I
think
I
got
most
of
them.
But
that's
what
it
looked
like.
D
B
Of
it
is
pretty
reproducible
I
would
say
there
is
a
set
of
pretty
nasty
Python
scripts,
but
just
a
little
bit
numbers
so
I'm
thinking
a
pretty
reproducible
with
the
exception,
maybe
all
YouTube
recordings
which
weren't
the
ones
that
I
had
to
fetch
the
data
manually,
to
be
honest,
I,
think
everything
else
to
be
producible.
That's
a
good
point.
Actually
I
would
like
to
perhaps
tidy
up
a
bit
and
included
in
the
slide
a.