►
From YouTube: Kubernetes SIG Testing 2018-09-11
Description
A
Ben's
gonna
talk
us
through
a
discussion
of
moving
kind,
kubernetes
and
docker
local
clusters
out
of
the
testing
for
repo,
which
I
think
is
kind
of
long
overdue,
and
then
we
have
a
couple
other
things
that
hopefully
we
can
get
to,
but
I
will
stop
rambling
and
hand
it
over
to
James
cool
thanks
yeah.
Can
everyone
see
what
screen?
B
Think
the
release
repo
I
looked
into
doing
a
similar
thing
and
finally
managed
to
split
everything
apart
from
the
upstream
test
in
front
so
get
one
of
our
configuration
out
in
the
meantime
as
well.
I
also
managed
to
get
everything
kind
of
powered
through
basil
once
again
inspired
by
how
you're
doing
the
crown
cluster
deployments
so
yeah.
This
is
kind
of
our
new
repository
first
thing,
jet
sucks
at
testing.
So
the
test
in
for
a
repo
is
now
just
matching
master
of
the
upstream
and
we
can
merge
in
additional
things
into
there.
B
So
this
is
used
like
testing.
Throw
to
deploy
Pro
itself
with
the
production
fly
rule
got
some
documentation
for
it
all
here
too,
if
you
need
to
put
together
in
science,
deploying
a
new
version
of
crap,
so
yeah
as
I
say
it's
all
driven
through
basil,
so
well,
I
might
quickly
do
is
just
deployed
or
upgrade
our
crowd,
because
I
did
do
it
earlier.
I
know
now
it's
in
about
20
commits
or
so
just
to
show
you
that
workflow
so
I'll
be
quick
with
it
as
well.
Of
course,
I'm
sure
everyone's
got
lots
to
do.
B
B
B
Humanities
upstream
there
I
would
normally
do
a
pull
request,
but
okay,
yeah,
okay,
so
there's
a
few
changed
in
there
just
kind
of
very
trusting
everyone
else,
because
otherwise
I'm
gonna
be
spending
the
rest
of
this
cool
rolling
back
these
changes,
but
yeah
I
can
go
ahead
then,
and
push
that
to
our
upstream.
B
I'm
gonna
hit
there
and
just
update
that,
and
just
by
doing
basal
run
of
the
production
don't
apply
job
that
will
actually
now
go
through.
It
might
kill
my
Internet's
I've
got
really
bad
into
it,
so
I
might
have
to
cancel
it.
They're
cool
starts
going
funny,
but
that
will
now
actually
go
ahead
and
go
and
build
all
of
those
images
of
prow
generate
the
new
manifest
and
roll
them
out.
B
Okay,
yeah
I'll
get
to
that
in
a
second
and
so
I'll
go
ahead.
I,
don't
know
how
my
Internet's
gonna
go,
but
I
think
yeah
that
will
actually
push
up
the
images
to
to
our
own
buccal
registry.
I
won't
leave
that
running
because
as
I
say,
my
internet
is
really
bad.
You
I
think
others
have
seen
where
that
goes
from
there
and
it
goes
ahead
and
pushes
up
an
image.
So
it's
pretty
smooth
in
that
respect.
I
also
managed
to
get
our
actual
CI
images
building
through
this.
B
So
it's
the
equipment
of
the
images
directory
and
the
testing
for
repo
it
after
I
got
everything
else
in
Basel
and
inside
to
bother
me
that
more
wasn't
so.
We've
got
the
our
images
directory
here,
for
example,
if
I
go
into
this
certain
manager
e
to
e
directory,
we
just
have
a
build
a
zoom
to
begin
with,
and
that
reference
is
an
image
here,
which
is
such
such
images.
Basel
build
code
on
image,
which
is
our
could
run
up.
The
basic
build
image
like
up
straining
has
so
you
can
see
which
is
just
here.
B
In
order
to
do
this,
though,
we
had
to
add
support
for
building
from
jakka
files,
this
building
for
a
doc
file.
This
kind
of
high
definition,
particularly
emetic,
because
of
the
things
like
the
run
commands,
and
no
one
really
wanted
to
have
to
do
everything
like
dropping
all
of
these
things
in
through
basil
and
so
I've
written
an
extra
workspace
rule
that
basically
allows
us
to
take
this
and
expose
it
contain
a
load
target
in
our
workspace
I.
B
And
yeah,
so
you
can
see
here
that
particular
image
we
actually
we
can
expose
that
to
the
rest
of
the
basil
project,
using
it
container
underscore
docker
file
rule
which
we
define
there
and
that
can
do
things
like
build
matrix
two.
So
we
can
do
multiple
different
versions
of
basil,
just
by
specifying
more
in
there.
It
will
also
go
ahead
this
by
defining
which
sorry
yet
from
there.
B
We
then
have
our
push
targets
to
find
as
a
standard
rules
underscore
docker
push
target,
so
we
can
just
go
ahead
and
push
all
of
our
images
and
they'll
get
automatically
tagged
which
yeah
it's
made.
It
a
lot
easier
to
manage
things
instead
of
having
to
switch
between
make
files
and
basil,
and
this
that
and
the
other
and
keeping
things
up
to
date.
B
We
then
use
the
decoratives
I
had
to
port
over
the
bootstrap
script,
to
point
to
this
repository
and
which
wasn't
too
difficult,
but
then
also
at
the
same
time,
we
switched
over
the
majority
of
our
repositories
to
the
decorated
job
config
stuff,
which
is
working
really
well
actually,
and
we've
managed
to
get
that
set
up
for
a
private
repository.
Now
too,
as
of
today,
after
a
few
issues,
yeah
there's
not
really
a
big
crescendo
to
any
of
this
I
mean
we
are
talking
about
configuration
management
which
I,
don't
think
was
ever
gonna,
be
tickity
exciting.
B
B
D
B
A
B
E
E
Any
time
you
write
a
Basel
rule
that
starts
calling
out
to
things
on
the
host
you
you're,
like
kind
of
asking
to
circumvent
how
hermetic
Basel
is,
and
that
can
cause
problems
for
caching,
which
is
part
of
why
they
typically
don't
ship
rules
like
this
themselves,
but
I
know
that
one
of
the
maintainer
is
issue
of
rules.
Docker
has
wanted
to
do
something
like
this
for
a
while,
because
it
is
so
hard
to
take
a
lot
of
really
useful
docker
files
and
get
rid
of
all
the
rotten
statements
in
them.
B
Young
kind
of
so
far,
I
haven't
actually
seen
my
workspace
file.
Dependencies
being
cached,
so
haven't
quite
hit
something
there,
and
there
is
obviously
here
all
the
time
of
my
images
changing
over
time,
because
Debian
swamp
the
same
thing
but
I
suppose
that's
kind
of
on
me.
But
it's
back
more
precisely
in
my
images.
A
So
you
mentioned
liking
the
base.
The
Basel
experience
has
been
pretty
nice
for
you,
I'm
curious.
It
sounds
like
maybe
your
pain
point
was
you're
flipping
back
and
forth
between
basel
and
make
files,
and
you
have
you-
maybe
had
a
choice
to
choose
between
doing
everything
with
the
set
of
consistent
make
files
versus
doing
everything
with
Basel.
What
made
you
choose
to
go
the
Basel
route
and
what
sort
of
life
lessons
do
you
feel
like
you
learned
along
the
way,
I.
B
Think
that's
good
yeah
I
mean
I,
think
a
well-made
set
of
make
files
probably
would
and
kind
of
and
consistently
made,
or
he
would
work
quite
well.
I
guess
not
and
I
don't
particularly
enjoy
make,
and
also
it
was
the
it
was
the
case
of
dependency
version
kind
of
drifting
and
I
need
to
keep
track
of
those
so
I
think
or
just
having
to
keep
tracking
the
track
of
them
somewhere
centrally.
B
Of
course,
that
could
just
be
a
flash
group
checks
the
bin
directory
and
make
sure
some
sha
are
correct,
but
it
was
kind
of
just
looking
for
a
consistent
way
to
model
this
I
also
I
kind
of
use.
This
is
a
bit
of
a
pilot
for
switching
certain
manager
over
to
this,
because
we
have
run
into
issues
there
and
we
have
quite
a
lot
of
different
generated
files
and
we
quite
regularly
get
users
who
are
generating
things
with
three
different
dependency
versions
off
them
and
get
quite
complicated.
B
A
So
I
think
you've
done
a
really
great
job
of
splitting
those
things
up,
I'm
kind
of
curious,
whether
or
not
we
could
follow
your
path.
My
only
little
bit
of
hesitancy
is
like
well.
I
am
now
sold
on
basil,
especially
in
a
world
where
you're
trying
to
keep
track
of
dependencies
in
different
languages
or
different
artifacts
I'm
still
unclear
on
how
friendly
it
is
to
the
newcomer.
So
it
sounds
like
it's
made,
it
kind
of
like
a
seamless
experience
for
you,
but
it's
like
you
know
thing
to
install
yeah.
E
B
Yeah,
that's
kind
of
worked.
Another
thing
I
was
looking
at
is
actually
making
use
of
multiple
prior
clusters,
so
we
can
stage
things
and
more
easily,
because
I
think
we
can
we're
pretty
close
to
being
able
to
get
it
down
to,
like
you
say,
just
a
couple
of
extra
lines
to
say
when
it
this
context
and
deploy
there
as
well.
B
A
Guess,
but
my
last
question,
if
nobody
has
anything
else,
is,
is
one
of
the
things
we
use
a
lot
in
testing,
for
is
the
config
updater
plug-in,
which
lets
us
avoid
a
lot
of
redeploys.
So
when
a
PR
merges
boom,
a
config
map
gets
updated,
and
that's
all
that
we
care
about.
How
do
I
weld?
Is
your
setup
play
with
that
yeah.
B
That
works,
hey,
I,
think
I
might
be
looking
in
the
wrong
file,
but
yeah,
so
that
works
just
fine.
We
just
enabled
own
thing
updated
like
this
and
then
Jess
access
testing
has
the
plug-in,
enabled
and
I
to
be
honest.
Prom
is
pretty
well
set
up
for
doing
this
than
splitting
out.
I
know,
I
heard
that
you're
also
trying
to
split
out
some
of
the
basically
configuration
we've.
E
We
should
really
look
at
that.
I
think
we've
mostly
been
coming
at
it
from
the
point
of
we
want
to
keep
testing
for
all
the
config
and
extra
group
and
we've
proud
out,
because
it's
the
one
of
the
more
polish
dish
things,
but
it
might
be
interesting
to
look
at
it
from
a
what
if
we
move
the
conflict
out
and
then
we
meet
the
brow
out
since
moving
the
config
out
might
be
more
doable
and
at
least
remove
the
pain
point
of
trying
to
track
upstream
and
both
way.
Merge
conflicts
themselves.
B
One
very
very
last
thing:
I'll
mention
because
I
noticed
it
I
always
found
it
bear
the
pain
upstream
with
the
bootstrap
images
is
managing
that
bootstrap
top
py
file
and
making
sure
you
have
a
consistent
one,
so
I've
gotten
rid
of
it
from
our
docker
file.
Now,
and
it's
basil
drops
that
in
we
have
our
bootstrap
image
bundle.
I
went
all
in
on
writing
my
own
rules
as
well,
and
that
drops
in
all
so
dependencies.
E
Big-Big,
plus
one
on
that,
where
we
were
going
to
move
it
into
the
image
and
then
we
decided
we
wanted
to
just
kill.
Bootstrap
sin
is
finally
going
after
that
effect,
we
just
piloted
moving
it
into
and
test
to
eliminating
bootstrap
out
of
the
out
of
the
layers
of
scripts.
That
executed
in
to
end
tests
just
last
night,
and
it's
mostly
working.
D
B
E
Okay,
so
so
I've
demoed
it
a
little
bit
before
so
I
might
not
necessarily
go
into
all
that,
but
in
the
imagist
it
lets
you
run
a
local
kubernetes
cluster
using
doctrine,
docker
containers.
E
E
Things
are
really
concerned
with
us
being
stable
and
we
have
a
bunch
of
nasty
basketballs
like
decent
right
now
that
we
depend
on
for
nearly
everything
and
we
kind
of
want
to
move
away
from
that,
the
other
thing
being
that
some
of
the
ones
like
mini
to
you
have
a
very
polished
experience,
but
they
have
a
little
bit
more
overhead.
We
can't
do
nested
virtualization
on
our
current
set
up
and
it
doesn't
really
support
building
from
source
and
it
doesn't
very
trivial
to
do
that.
E
It
really
does
released
kubernetes
versions
where
it's
really
a
lot
of
the
time
want
to
be
testing
against.
Like
some
koreans
version.
Brother
confronts
us
so
kind
targets
to
burn
it
ease
1.11
plus,
so
it
has
QA
dome
config
and
can
do
that
properly
and
the
goal
is
to
run
at
the
very
least,
a
conformance
suite
and
make
that
very
fast
and
cheap
and
easily
locally
replicable.
We're
pretty
much
to
that
point.
There
are
some
flake
issues
with
some
of
the
conformance
tests,
but
I
think
I
might
have
that
end
down.
E
Finally,
so
up
until
this
point,
we've
been
incubating
it
in
testing
for
as
a
follow-up
to
some
previous
doctor
and
doctor
work
that
we've
had,
but
to
some
of
the
earlier
points,
testing
for
a
sort
of
this
monolithic,
repo
of
every
random
tool,
we've
built
for
testing
kubernetes
I
would
like
to
move
it
out.
I've
posted
too
list,
and
so
far
it
looks
like
the
leads
are
okay
with
this,
but
I
wanted
to
bring
it
up
a
meeting
and
run
it
hi.
E
Everyone
I'm
hoping
to
move
it
to
a
tripping,
e-cigs
repo,
so
that
I
can
start
doing
some
actual
blocking
in
to
improve
singing
it's
on
the
kind
codebase
itself,
rather
than
adding
those
two
and
making
testing
for
PR.
Significantly
so
I'd
also
like
to
be
able
to
stay
believe
version
it
back
to
cube
test
so
that
we
can
use
it
without
worrying
about
like
head
breaking
it
immediately,
because
then
we
can
actually
have
been
on
a
bit.
E
C
I
got
a
quick
question
on
this,
so
I
know
that
cube
tests
already
had
your
different
cluster
providers
to
do
all
the
setup-
and
you
know
like
I've,
got
one
that
we're
working
on
for
blowing
clusters
on
de
jure,
but
it
kind
of
seems
like
you've
sort
of
flipped
this
around
and
said.
Let's
build
a
cluster
first
and
then
run
cube
tests
inside
of
it.
C
E
You
need
it
sorry
I'm,
going
to
treat
this
like
a
normal
provider,
but
to
this
point,
I've
not
wanted
to
add
the
code
to
cube
test,
because
it's
not
there's
nothing
forcing
this
to
be
stable,
I'm,
actually
considering
even
just
having
cube
just
call
out
to
it.
So
we
can
avoid
calling
you
too
much
yeah
I,
don't
think
it
should
be
different
than
any
other
provider.
E
The
reason
we've
wanted
this
is
that,
most
basically,
all
the
providers
we
have
now,
if
you
want
to
test
kubernetes,
you
need
to
be
able
to
rinse
some
cloud
resources.
Yeah.
The
spin
of
time
can
also
be
pretty
bad
cops
and,
and
things
can
easily
spend
like
10-15
minutes
getting
and
I
read
node
cluster
with
metadata.
C
E
I've
been
able
to
run
everything
I've
been
able
to
in
parallel,
excluding
the
miners,
those
serial
tests
that
we
have
marked,
as
conformance
run,
the
other
conformance
tests
and
15
minutes,
including
checking
out
kubernetes
building,
making
no
damages
booting
a
cluster
running
the
tests
tearing
it
down,
and
this
is
something
that
you
can
also
just
run
on,
like
a
laptop
yeah.
C
Yeah
cuz,
the
other
thing
I
was
thinking
about,
is
like
if
I
want
to
add
in
Windows
node
to
this
it
would
be
relatively
easy
to
call
out
to
vagrant
bring
up
a
VM
and
now
I've
got
Windows
and
Linux
nodes
there
there's
it
would
be
relatively
straightforward
to
do,
but
I
guess.
My
question
is:
do
you
see
that
being
in
separate
tools
like
kind
or
is
that
something
that
makes
more
sense
for
cube
test
itself?
So.
E
The
ends
are
definitely
something
that
I
think
we
want
at
some
point.
We've
had
they're
a
lot
slower,
I
think
James
can
actually
point
to
that
a
bit
and
for
our
CIO
at
least
we've
been
leveraging
Hey,
which
lets
the
handful
of
us
that
we
have
maintaining.
All
of
this
not
focused
much
on
keeping
the
cluster
up
anymore,
like
we
were
thanked,
but
has
a
trade-off.
We
actually
don't
have
nerds
nested
virtualization.
We
can
get
that
through
some
funky
business.
E
C
E
Yeah
yeah
I'm
not
too
familiar
with
doing
it
on
Windows,
yet
I
am
actually
intending
to
try
to
get
the
Linux
nodes
to
boot
on
docker
for
Windows.
At
some
point,
I've
got
it
working
on
docker
for
Mac
I
think
the
biggest
problem
with
all
of
them
is
going
to
be
when
I
reintroduce
multi
node
support.
Networking
is
going
to
be
really
fun
on
Mac
and
Windows,
but
yeah
should
be
possible.
E
A
Yeah,
as
like
one
of
the
members
of
the
github
admin
team
and
also
is
the
state
testing
lead
like
I'm
super
in
favor
of
this
breaking
out,
and
then
the
only
thing
I
really
require
is
a
decision
from
this
sake
that
we
agree.
We
want
to
create
a
new
repo
and
I
was
going
to
suggest
creating
a
Rico
called
kind
in
kubernetes
6,
because
we
don't
have
to
go
through
sig
architecture
or
anything
like
that.
Kubernetes
cigs
is
free
to
play
for
6.
A
A
Cool
all
right,
that's
pretty
much!
That's
pretty
much!
All
I
need
yeah.
Oh
my
god,
just
like
been
first
approached
me
saying,
like
hey
mr.
github
labels
since
you've
had
fun
like
reducing
all
these
kind
labels
with
it
like.
How
much
would
your
head
explode
if
I
called
my
new
project
time?
It's
like
you
know,
whichever
answer
is
gonna
make
you
actually
go
through
with
this
cuz.
E
I
think
probably
I
can
I
can't
believe
it
games
for
that.
Quinton
Lee
had
done
quite
a
bit
of
work
on
prototyping
that
our
previous
iteration
on
this,
and
it
was
just
called
dent
for
doctor
and
doctor,
just
not
the
most
imaginative
name
and
also
one
that
is
confusing.
It
used
all
over
the
place,
I
think
in
the
long
term,
I
might
regretting
a
little
bit
having
docker
there,
but
I
think
we
can
get
away
with
just
calling
it
kinda
friendly
local
cluster
or
something
and
deal
with
that
later.
B
E
A
C
A
So
this
might
be
the
right
group
of
people,
but
I
know
I
personally
am
really
heads
down
for
the
current
outgoing
release
and
so
I
don't
anticipate
this
PR
getting
merged
before
then
like.
If
it
was
something
so
the
phrase
the
way
you
described
it
doesn't
make
it
sound
like
we
just
need
a
little
push
to
get
it
over
the
line.
It's
more
like
we're
kind
of.
We
need
some
suggestions
on
how
to
redesign
this
to
support
our
use
case.
Well,.
E
E
A
A
Bit
of
etiquette
is
like,
maybe
so
your
reviewers
tone
of
voice
didn't
necessarily
imply
like
this
is
a
hard
stop
until
you
redesign
it's
just
like
I
think
we
talked
about
this
last
time
and
if
you
respond
back
like
this
Sarris
like
this
really
is
infeasible.
This
is
too
much
he's
generally
been
a
pretty
reasonable.
Guy
I
know
that
sometimes
PR
reviewers
tend
to
wander
off
and
disappear
for
days
at
a
time.
I
am
guilty
of
that.
Pinging
them
how
to
band
on
slack
or
email
can
sometimes
help.
A
C
E
Like
continually
pinging
people
as
one
of
the
unfortunate
ones
you
doing
tests
in
particular,
I
think
most
of
the
people
to
pay
attention.
These
things
are
doing
too
many
things
and
they're,
not
enough
people
that
are
spending
a
lot
of
pens.
Click
on
these.
So
it's
really
hard
to
get
good
with
use
you'll.
If
you
can
just
like
keep
reminding
people
and
yeah.
C
C
Yeah,
it's
just
for
a
time
perspective.
You
know
we're
basically
trying
to
get.
You
know
the
windows
tests
stable
and
if
the
tests
don't
get
merged
until
you
know
feature
freeze,
then
that
means
we've
got
two
weeks
of
test
runs
to
fix
bugs,
and
so
it
really
impedes
our
ability
for
signaling
us
to
be
able
to
move
forward
with
fixing
bugs
if
we
can't
get
the
test
go.
So
that's
why
I'm
trying
to
like
I
I,
know,
release
now
so
like
my
ass
could
be
like.
A
100%
agree
with
that
and,
like
I
said,
I
tend
to
have
a
much
more.
I
can
be
a
lot
crazier
if
this
is
about
like
improving
test
coverage
and
stuff.
So
for
me,
it's
just
I'm
unclear
how
risky
this
is
in
terms
of
it.
Breaking
existing
tests.
I
have,
are
the
pre
submits
to
go
on,
but
if
you
can
convince
me,
I
can
be
an
ally
on
your
side
and
we
can
push
forward
on
this
and
that,
as
Ben
said,
like
the
release,
team
can
be
reasonable.
So
you
know
a
lot
of
human
talking
stuff.
E
Wholly
agree,
I
would
really
like
to
see
this
go
forward.
I've
been
trying
to
continually
follow
up
on
the
PR.
Some
testing
for
related
to
this
I
think
this
is
an
effort
we
definitely
want
to
see.
Go
through
I.
Just
don't
know
that
necessarily
that
people
here
have
all
the
bandwidth
and
background
and
I'm
not
really
sure
who
does
on
things
like
how
the
test
flags
work.
Ya
know
if
there's
anyone
paying
good
attention
to
that
historically,
but
I
know
that
people
are
looking
into
fixing
it
now.
You.
A
Might
be
able,
then,
if
or
anybody
in
the
Sunnyvale
office,
if
you
can
maybe
tap
Jeff
Grafton
on
the
shoulder
and
see
if
he
could
give
this
a
once
I've
heard
that
might
help
just
as
he's
somebody
who
kind
of
overlaps
that
area
property
niches
and
he
used
to
have
some
need
to
be
framework
to
expertise.
I
to.
B
A
A
D
Okay,
so
my
name
is
Ophir
I'm
gonna
present
this
next
time,
but
today
I'm
just
gonna,
mention
it
I'm,
developing
a
code
review
tool
that
is
pretty
special,
it's
multi
repo,
so
you
can
have
one
review
for
a
couple
of
changes
to
in
a
couple
of
repositories
and
every
repository
can
actually
be
hosted
in
a
different
place.
So
one
can
be
in
git
lab
one
can
be
in
github,
one
can
even
be
on
Prem
and
they
can
all
be
part
of
the
same
review
and.
E
D
Can
also
help
companies
collaborate
on
open-source
projects,
so
every
company
has
its
own
private
red
bow
and
there's
this
common
open-source
common
one,
and
so
you
could
have
like
one
PR
that
spans
all
of
the
repos
and
like
the
code
that
his
private
stays
private,
so
yeah.
Those
are
like
the
cool
things
about
it,
so
companies
can
actually
collaborate
on
open
source
being
at
head.