►
From YouTube: Kubernetes SIG Testing 2018-05-08
Description
A
Hi
everybody
today
is
Tuesday
May
8th,
welcome
to
the
cig
testing
weekly
meeting
I'm
your
host
for
today's
meeting
on
today's
agenda.
I
just
wanted
to
give
those
of
us
who
are
at
couponing
you
a
chance
to
debrief
real
quick.
If
there's
any
interest,
we're
gonna
have
a
brief
chat
about
dockerfile
vs.
basil
for
at
least
building
and
prowl
images,
maybe
a
broader
just
philosophical
discussion
and
before
I
press
the
record
button.
A
So
what
I
heard
was
that
it's
pretty
much
ready
to
go
and
we're
ready
to
live
in
a
world
where
we
never
write
a
job
that
uses
bootstrap
ever
again,
and
we
now
want
to
start
taking
jobs
that
either
weren't
written
with
bootstrap
in
the
first
place
and
augment
them
with
pod
details
such
as
some
of
the
jobs
that
fada
wrote
or
take
existing
jobs
that
were
written
with
bootstrap
and
replaced
them
to
use
pod,
utils
and
Steve.
You
were
saying
something
about
turning
off
cloning
as
a
first
step
on
that
yeah.
B
So
I
think
there's
there's
two
things
that
well
the
one
thing
that
I
think
we
do
need
to
get
finished
so
because
there's
more
moving
pieces
right
now,
there's
more
logs
to
keep
a
hold
of,
and
one
single
test
log
isn't
enough
anymore.
I
have
a
PR
which
McCall
is
legit,
send
for
a
controller
that
we
can
run
and
the
controller
just
uploads.
All
the
container
logs
that
stuff
gets
boosted.
Push
to
GCS
message
supposed.
Also
the
other
update
is
in
all
the
various
places
that
we
have
a
logs
link
in
dec.
B
C
A
B
Okay,
yeah
they're
only
tractors,
there's
issues
in
touch
them
in
the
chat.
The
the
two
outstanding
things
that
I
think
would
be
really
nice
for
UX
are
pretty
much
most
finished.
I'll
try
to
get
that
stuff,
at
least
to
a
merging
state
before
Friday,
and
if
I
can't
I'll
ask
the
close
to
so
push
it
through
all.
D
B
The
it's
funny
you
ask
you
and
Ben
convinced
me.
That
was
a
good
idea
like
two
months
ago,
a
conversation
and
I
think
the
biggest
thing
was
having
jobs
own
their
own
uploads.
Basically,
in
every
case
where
the
job
is
malformed
or
has
to
exit
early,
we
lose
artifacts.
So
I
think
the
conversation
was
less
as
foxy.
We
could
proxy
through
the
artifact
upload
through
a
server
to
GCS
and
and
then
also
we
have
the
long
to
be
able
to
learn
externally.
B
B
Right
so
yeah
right
now,
the
PR
that
adds
the
controller.
All
that
controller
does.
Is
it
watches
four
pods
created
by
proud
that
our
reference
to
a
proud
job
and
then,
when
a
container
in
that
pod
transitions
into
the
we
push
the
logs
up
to
GCS?
But
then
all
the
artifacts
and
like
test
stuff
is
still
pushed
from
the
top
and
I'll
link
a
link
that
umbrella
issue
too.
So,
in
the
umbrella
issue,
I
tried
to
sort
of
outline
what
you
and
Ben
have
been
talking
about.
In
that
we
we
think.
B
A
A
The
second
coming
that
I
know
that
got
picked
up
by
the
CN
CF
is
like
one
of
the
highlight
moments
of
the
third
day,
so
just
went
to
like
even
a
big
shout
out
to
them.
Also
out
doing
my
this
is
fine
dog
usage.
I
didn't
realize
there
were
so
many
other.
This
is
fine
dog
images
available,
I,
don't
know.
If
any
of
you
are
curious
to
see
anything
from
the
slides,
I
posted
videos
of
the
recordings,
the
whole
conference's
videos
have
been
posted
online
and
also
posted
links
to
our
slides
in
the
meeting
agenda.
A
Notes.
I
would
say
if
anything,
my
the
one
biggest
takeaway
I
had
in
looking
through
things
for
my
Docs
is
that
I
finally
found
the
graph
that
proves
that
there's
no
way
we
would
be
attaining
the
merge
velocity
that
we
have
for
kubernetes
kubernetes.
If
we
weren't
doing
batch
merging,
like
it's
just
theoretically
impossible,
so
merging
things
in
batches
is
definitely
saving
their
lives
and
I
was
pleased
to
hear
that
we're
talking
about
having
tied
sort
of
vary
the
size
of
batches.
A
It's
using
to
help
our
merge
velocity
going
forward
and
the
other
thing
that
I
saw,
which
y'all
are
probably
all
aware
of,
because
you've
been
helping
with
PRS
into
the
config
for
tied
and
proud.
But
we
have
a
lot
more
repos
other
than
just
kubernetes
kubernetes
that
are
using
proud
for
their
pre,
submit
jobs,
which
I
take
as
a
sign
that
we've
reached
some
level
of
maturity.
A
Project
wise,
where
folks
other
than
just
the
folks
in
the
communities
project
see
this
as
a
useful
thing
to
help
merge
their
code,
so
yeh
prowl
I
also
think
we
kind
of
need
like
a
better
proud
logo
than
the
one
we
have
for
what
it's
worth.
I
know.
We
have
like
these
these
two
circles
in
front
of
the
the
kubernetes
logo
that
I
think
are
supposed
to
represent,
like
the
wake
at
the
front
of
the
boat,
which
is
what
prow
is
but
I
think
we
can.
We
can
do
better
than
that.
A
B
A
I
am
I
did
get
a
chance
to
talk
to
Michael.
What's
how
do
you
pronounce
his
last
name?
Carjacking
yeah
carcass
he's
super
enthusiastic
about
prowl.
He
said
he's
like
two
or
three
presentations
already.
He
advocates
for
it
as
much
as
possible
within
Red,
Hat
and
he'd,
really
like
to
work
with
us
to
help
advocate
for
it
more
outside
of
just
open
shift
in
kubernetes,
which
I'm
a
huge
fan
of
so
yeah
I
mean
I,
think
people
some
of
our
teams
in
India.
A
A
A
You
know
I'm
not
sure
if
there's
anything,
that's
necessarily
preventing
us
from
we
rewiring
prowl.
As
an
app
I
know,
I
had
asked
about
it
in
the
channel
a
while
back
and
it
seems
like
if
nothing
else
like
it
would
give
us
more
tokens
to
play
with.
We
get
20
tokens
per
user
in
New
York
that
were
a
member
of
so
ballpark
we'd,
get
like
another
2,000
3,000
tokens,
at
least
in
kubernetes
kubernetes,
which
could
be
pretty
cool
anyway.
That's
like
I,
don't
know
it
was
really
it.
The
conference
was
roughly
the
same.
A
D
E
D
D
A
A
B
B
If
you
use
the
doctor
files
and
so
I
guess,
there's
kind
of
my
take
on
between
as
a
naive
developer,
showing
up
thinking.
Oh
okay,
this
is
go,
I'm,
gonna
use,
go,
build
or
go
install.
These
are
container
images.
There's
you
know.
There's
doctor
files
checked
into
the
repo
I'll
use
docker
build
if
I
showed
up
as
a
new
contributor
and
tried
to
do
any
of
that
today,
I
would
get
something
that
was
not
working
at
all
and
I
think
that
we
should
probably
figure
out
like
a
pieces
and
result
here.
B
One
of
the
comments
that
Christophe
made
on
slag,
like
for
better
or
for
worse
I,
do
think
right
now,
doctor
files
are
the
most
common
currency
for
declaring
what
a
container
looks
like
so
on
that
front,
I
think
we
should
keep
them
around
if
it's
possible
I
think
we
should
try
to
get
proud
or
try
to
get
Basel
to
do
what
the
doctor
files
weren't
doing
and
if
it's
as
simple,
as
you
know,
changing
some
paths
over
and
I
think
that's
like
a
hugely
important
like
I.
Think
it's
fine!
B
If
we,
you
know,
if
the
Basel
image
building
stuff
Peugeot
has
a
opinionated
place
where
it's
gonna
put
go
things,
we
can
go
ahead
and
update
our
doctor
files
to
do
that.
I,
don't
think
we
need
to
like
spend
a
million
years.
Making
sure
Basel
does
something
that
it's
not
used
to
doing,
but
we
should
at
least
keep
those
in
sync.
C
I
mean
one
thing:
I
feel
like
the
larger
problem
is
especially
if
we
are
wanting
other
organizations
to
use
it,
as
I
really
feel
like.
We
should
be
publishing
images
that
other
people
can
use
and
I
feel,
like
you
know,
okay
I
feel
like
that
would
solve
a
lot
of
the
problems
you
know
as
if
I
could
just
say.
If
I
could
just
write
or
customize
the
getting
started,
yeah
Mille
and
you
know
poke
whatever
point
my
container:
they
used
the
prow
image
and
then
go
from
there.
I
feel
like
that
would
be
useful.
C
C
F
C
B
C
B
A
A
Even
if
we
have
really
excellent,
clear,
concise,
read
nice,
which
we
don't
that
like
made
it
super
obvious
that
it
was
just
basil
and
here's
how
you
do
each
and
everything
I
still
feel
like
the
repo
would
gain
a
reputation
of
being
a
little
more
difficult
to
contribute
to,
because
you
have
to
learn
this
other
super
weird
way
of
doing
things.
That's
unique
pretty
much
just
to
this
one
project.
There
aren't
too
many
other
projects
out
there
that
use
basil.
A
That
said,
I
also
get
that
this
repos
and
experiment
for,
like
proving
that
basil
can
be
used
to
do
all
of
those
things.
And
in
my
limited
experience,
it's
been
working
very
well
there.
It's
it's
pretty
it
pretty
much
just
works
on
my
MacBook,
which
is
cool
but
like
this
is
kind
of
the
same
problem
that
kubernetes
repo
is
facing.
A
But
our
CI
does
use
like
basil
to
build
things,
so
something
got
checked
in
which
broke
one
pass
but
didn't
show
up,
because
we
were
testing
the
other
path
right.
So
I
hear
that,
like
two
paths,
ain't
so
great
but
I'm
wondering
if
there's
a
way
where
we
could,
we
could
start
with
basil
as
the
canonical
source,
but
make
sure
that
those
more
like
friendly,
open
source
ways
are
either
generated
for
a
basil
or
at
least
covered
under
CI.
So
we
know
that
our
practices
aren't
breaking
them.
D
G
I
was
gonna,
say,
kinda
the
same
thing,
but
with
the
the
reference
back
to
the
main
repo,
which
is
there
is
a
long
history
of
basil
and
make
getting
out
sync
with
each
other
in
kubernetes
crew,
readies
and
whether
it's
okay,
you
know
make
basil
release,
doesn't
include
these
artifacts
for
make
basil
releases
broken
because
it
includes
this
version
of
QQ
proxy,
but
make
release
hasn't
caught
up
to
that.
Yet
they
they
get
out
of
sync
quite
often,
but.
C
B
C
You
saying
you
know
with
the
idea:
I
mean
I,
think
I.
Think
the
the
you
know
I
think
is
I
mean.
Is
it
what
like
free?
You
know
if
I
have
a
go
image
that
I
can
have
go
image
spit
out
a
docker
file,
or
are
you
saying
that
have
the
go
image
rule
work
off
of
a
you
know,
Rita
have
to
read
a
docker
file
for
input
either
or
right,
I.
D
A
Like
I,
just
from
the
perspective
of
bringing
on
board
new
contributors,
a
lot
of
what
I
heard
you
know
and
have
it
more,
the
the
mentorship
space
and
in
the
conference
track
and
like
a
lot
of
people
are
like.
Where
should
I
go
to
contribute
new
code
and
I
really
want
to
advocate?
How
like
prow,
is
an
awesome
code
base.
That
seems
to
adopt
a
lot
of
adopt
a
lot
of
principles
that
work
well
with
sort
of
the
shape
of
a
cloud
native
and
kubernetes
centric
application.
A
But
if
they
can't
use
the
tool
chain
that
they're
familiar
with,
if
they
have
to
learn
all
this
extra
stuff,
I
can't
in
good
conscience
and
new
contributors
to
that
code
base
and
so
I.
Just
I'd
love
for
us
to
make
sure
that
you
know
the
simple
gold
go
tool
chain
and
even
if
they're,
like
whatever
I,
don't
think
it's
too
crazy
for
somebody
on
their
laptop
to
use
a
docker
daemon
to
build
an
image.
Even
if
it
is
ultimately
a
glorified
tarball.
And
there
are
many
ways
of
building
without
the
dr.
C
Communities
like
the
overarching
question
is:
are
we
comfortable?
You
know
asking
people
to
use
basil
and
if
you
know
maybe
we
shouldn't
be
using
basil
at
all,
but
you
know
me
if,
if
to
me,
it
seems
like
there
should
be
if
there
is
a
place
where
we're
comfortable
asking
people
to
use
basil,
it
seems
like
testing
for
is
a
good
place
to
do
that
and
if
I
mean
the
guy
I
do
think
I,
don't
I.
You
know,
I.
C
Think
me,
having
worked
at
Google
for
eight
years
and
already
being
familiar
with
Faizal,
like
you
know,
blade
or
the
internal
version
like
basil
is
infinite,
is
intimately
comfortable
for
me.
So
it's
sort
of
a
little
bit
hard
for
me
to
understand.
You
know
exactly
how
challenging
it
is
to
install
basil
and
get
used
to
it.
Yeah.
B
I
think
I'm
like
less
interested
in
specifically
what
tool
you're
using
for
just
the
ability
to
interoperate
and
I
guess:
that's
kind
of
the
the
biggest
difference
I
see
between
basil
from
what
I
see
now
I.
Honestly
I,
don't
know
if
you
know,
like
it's
very
surprising,
to
me
right
now
that
using
the
go
standard
tool
chain,
I
can't
compile
tested
for
what
using
basil
I
can
like.
That's
a
really
surprising
thing
that
I
really
didn't
expect
to
happen
and
I
think
on
a
lot
of
other
repos.
B
D
E
Can't
use
like
go
get
write
like
a
vast
majority
of
other
folks
who
wanted
to
like
do
extensions
or
superpositions
or
something
we
can't
go,
get
something.
We
can't
do
that
in
the
main
repo
either
or
you
can't
go,
get
kubernetes
it
just
implodes
right,
and
you
could
do
that
almost
every
other
go
project
that
exists
right.
C
Today,
go
install
fails
on
some
there's
different
problems
when
the
core
repo
that
doesn't
have
anything
to
do
with
basil.
It's
mostly
around
the
fact
that
we
generate
a
bunch
of
code
and
are
wanting
like
I
feel
like
in
the
main
repo.
The
fact
that
goat
get
doesn't
work
is
sort
of
a
larger
conversation
and
right
yeah,
and
the
fact
that
we
have
all
these
weird
staging
repos
and
the
staging.
How
critical
that
was.
B
C
B
C
Me
what
I
would
like
is
you
know,
maybe
if
we
could
have
to
me
I
guess
the
thing
that
I
like
is
about
the
go
image
rules
is
that
it's
sort
of
clean
and
it
prevents
people
from
putting
too
much
logic
into
the
dockerfile,
because,
like
really
all
that
we
want
to
do,
is
get
a
same
base
image
and
create
a
bill.
Compile
the
binary
and
put
that
into
the
image
and
I
feel
like
when
you
give
someone.
C
You
know
more
power,
there,
they're
gonna
tend
to
make
it
really
complicated,
and
so
I
would
be
attracted
more
to
the
idea
of
like
having
the
you
know,
figuring
out,
a
way
to
have
the
rule
generate
the
dockerfile,
and
so
that
we
know
that
what
we
are
generating
with
the
go
image
rule
is
the
same
aim
as
what
you
would
do.
If
you
ran
a
doctor
file,
yeah
I
mean.
B
That
makes
easily
might
it
might
get
a
little
bit
complicated
because
I
think
right
now
the
doctor
files,
what
they
expect,
but
the
way
that
they're
written
right
now
for
proud
is
I.
Have
the
compiled
binary
available,
I'm
gonna
copy
into
the
root
of
the
file
system
and
I'm
done
generating
one
docker
file.
That
does
all
of
that
in
a
way
that
the
resulting
container
image
only
has
the
compiled
binary,
it
doesn't
have
the
go.
Runtime
doesn't
have
any
of
the
compiled
artifacts
like
that
might
be
tricky
but
think
you
can.
B
B
A
A
So
what
better
repo
to
prove
that,
like
basil
works,
I
just
feel
that
we,
as
a
group,
haven't
done
a
great
job
of
educating
the
community
on
how
to
use
basil
to
do
all
the
things,
and
there
is
an
expectation
from
community
members
that
the
standard
go
tool
chain
could
work,
that
you
could
go
get
things
and
that
she
could
docker
build
things.
I,
don't
know,
Steve
that
your
case
sounds
like
them.
A
B
A
D
B
Maybe
just
edit,
as
opposed
to
bit
and
I
honestly
think
those
doctor
files,
I,
don't
think
they've
ever
changed.
I
think
they've
been
committed
once
and
never
touched
so
I
think
the
generation
is
nice
and
I
see
why
it
would
be
valuable
to
have
that
in
sync
all
the
time,
but
at
the
same
time
I
think
it's
not
a
huge
amount
of
time,
if
we
just
say
for
now:
let's
just
bring
it
into
sync
and
then
deal
with
it
later.
We
want
this
new
job.
C
B
C
I
guess
I
mean
something
that
I
would
be.
You
know,
I'm
curious
about
is,
you
know,
is
the
idea
of,
like
is
the
idea
of
making
testing,
for
you
know,
sort
of
having
a
dependency
on
Ted
on
Basel?
Is
that
like
an
AA?
Is
that
like
a
non-starter
or
is
that,
like
a
conversation
we
could
eventually
have,
and
what
would
we
need,
and
you
have
that
I
think.