►
From YouTube: Kubernetes SIG Testing 2018-02-27
Description
Meeting notes: https://docs.google.com/document/d/1z8MQpr_jTwhmjLMUaqQyBk1EYG_Y_3D4y4YdMJ7V1Kk
A
Hi
everybody
today
is
Tuesday
February
27th,
welcome
to
the
weekly
sake,
testing
meeting
and
your
bearded
host
here
in
cracking
burger.
This
is
a
publicly
recorded
meeting.
It
will
be
posted
to
YouTube
forever
and
ever
ever.
First
up
I
wanted
to
have
a
bit
of
follow-up
to
a
discussion
that
happened
here
last
week,
while
I
was
down
at
Elm
summit
and
then
I
think
followed
up
in
contributor
experience
sort
of
just
around.
How
can
we
responsibly
roll
out
new
automation
and
I?
A
Think
there's
consensus
on
this
based
on
like
the
thumbs
up,
emojis
I
have
the
comment:
I
posted
right,
but
history
as
I
understand
it
is.
There
have
been
a
couple
times
when
we
turned
on
things
without
giving
people
a
heads
up,
and
sometimes
that
can
be
seen
as
being
disruptive
or
things
are
just
happening
without
people
understanding.
Why
and
so,
I
kind
of
want
to
broadly
propose
that
going
forward.
A
A
Think
state
testing
thinks
a
lot
of
the
implications
of
how
that
ties
together
with
all
the
automation
that
we
already
have
you
get
that
proposal
there
you
implement
it
and
then
float
out
sort
of
the
what
this
is,
why
we're
doing
it
and
why
you
should
care
playing
a
Cubert
any
step
prior
to
going
out
and
ideally,
if
there's
some
way,
we
can
do
offed
out.
That
would
be
great
I
mean,
generally
speaking,
I
think
there's
a
couple
exceptions.
We've
been
pretty
good
about
turning
things
on
and
an
award
wide
level.
A
If
you
have
anything
to
clarify,
but
I
agree
with
his
point
that,
like
when
I
talk
about
broadcast
and
if
you
burn
Eddie's
death
prior
to
roll
out,
it's
not
to
get
everybody
to
say.
Yes,
let's
go
it's
more
to
have
somebody
say:
hang
on
I,
really
don't
want
this,
just
in
true
lazy,
consensus
fashion,
so
I'm
thinking
in
terms
of
on
the
order
of
weeks
and
if
there
are
processes,
changes
that
are
significantly
larger.
A
C
A
C
I
think
the
site
or
the
t-shirt
size
labels
are
actually
being
applied
automatically
for
us
by
prow,
and
so
those
Bo
or
the
PRS
that
are
coming
in
are
actually
being
used,
and
we
do
actually
use
that
as
part
of
our
contribution
to
experience
as
well.
We
do
have
that
in
our
contributing
dock,
so
we're
kind
of
embracing
as
much
as
we
possibly
can
a
stream
or
two
so
I
think
this
is
great.
Are
you
noticing
there
are
other
labels
that
are
not
being
applied
by
proud?
The.
C
C
A
A
related
note,
I
have
a
fork
somewhere,
where
I'm
trying
to
hack
up
that
labels
PMO
file
and
in
the
cod
feels
like
description
and
whether
it
applies
to
issues
or
PRS
and
I'm
trying
to
generate
dock
out
of
that
I.
Don't
really
know
where
the
best
places
for
that
to
live
long
term.
I
haven't
just
posted
somewhere.
That
I
think
that
I
would
post
around
I'll,
try
and
channel
some
folks
at
this
meeting
or
at
the
end
of
it.
A
A
E
A
F
This
is
something
that
I
think
read.
A
couple
PRS
recently
went
in
from
Cole
I
think
that
made
like
the
execution
of
different
jobs
for
different
branches,
that
report
under
the
same
status.
It
made
that
logically
flow
a
little
bit
better
and
I
guess
the
intent
there
is
if
we
have
different
implementations
or
different
configs
for
the
same
job
per
branch.
F
We
still
want
that
to
be
like
semantically
named
something,
and
we
were
just
setting
this
up
for
a
couple
of
the
OpenShift
origin
repos,
it's
kind
of
clergy
right
now,
I
didn't
know
if
you
know
maybe
better
call.
If
we
had
any
thoughts
about
like
do,
we
want
to
maybe
have
a
first
class
concept
of
like
here's,
the
mapping
between
branch
to
job
for
this.
D
I
think
I
would
I
wanted.
That
was
the
more
advanced
job
presets,
which
wouldn't
sort
of
look
like
that.
I,
don't
I,
think
we're
doing.
Okay
with
pod
presets
for
now
and
I.
Don't
think
the
coffee
pace
is
next
necessarily
that
bad,
because
you
do
want
like
an
explicit
config
for
this
branch
visually.
They
start
to
diverge
anyhow,
and
you
want
to
freeze
the
older
branches
in
place
so
like
more
of
a
first
class.
Okay.
F
A
G
Yeah,
so
this
new
user
Gaspar
is
we
enclose
every
awkward
PR?
Are
you
ki
are
for
now
our
so
here
is
so
like
too
many
information,
so
whether
the
tests
are
passed
fair
or
ten
days
and
the
second
informations
a
the
most
requirement,
so
the
my
recover,
it
is
display
a
table.
So
the
first
row
is
just
listing
all
of
the
label
that
you
are
currently
have
in
your
rip.
G
Oh
I
saw
your
PR
and
the
second
row
is
is
going
to
display,
like
a
curry,
are
for
each
queries
will
say
like
what
Kyle
a
bore
at
your
distance
order
for
the
require
label
and
which
what
our
label
that
you
have
is
forbidden
so,
which
means
you
shouldn't
have
it
so
you
want
to
know
more
about
the
curry.
You
can.
Click
on
this
burn
is
so
the
curry
is
so
it
it
like.
Every
label
about
a
curry
and
that's
a
and
about
a
tests
or
you
can
is
I.
G
Think
it's
pretty
much
like
browse
data
stories
include
every
time
that
word
on
the
PR,
so
you
can
click
to
the
squealing
tutor
turbinator
chopped
and
it
headers.
They
don't
indicators
or
in
case
if
there
is
any
pest,
but
fairies
will
show
you
the
test
of
tail
first,
but
if
all
tests
are
passed
it
just
turn
K
every
like
collapsed
on
your
house
is.
B
So
would
it
be
possible
to
just
as
a
feature
request
to
potentially
combine
as
a
reviewer
who
reviews
like
a
million
PRS,
knowing
the
labels
and
seeing
it
all
in
a
in
a
table?
Ish
fashion,
like
uber
Nader
on
droids,
would
be
super
duper
helpful
because
I
don't
even
know
how
many
reviews
I
do
in
a
single
week.
You
know.
B
H
In
because
the
this
one's
more
about
dads
answer,
the
question
of,
why
is
my
pure,
not
Margene,
which
this
make
you
currently
has
some
functionality
for
it's
less
about
you
is
or
like
as
a
reviewer.
You
should
not
be,
hopefully
not
be
worrying,
so
much
about
why
random
peers
aren't
merging.
You
should
just
be
doing
your
step,
but
what
what
labels
were
you
thinking
of
I
wanna,
see
because.
B
E
B
H
E
H
H
D
D
Ultimately,
responsible
for
making
sure
their
stuff
merges,
we've
had
some
existing
discussion
about
potentially
extending
it
to
just
show
PRS
and
default
to
your
PRS,
but
have
ways
to
tune
the
filtering
things
we
might
be
able
to
make
that
sophisticated
enough
to
tune
to
the
ones
you're
reviewing
I'm,
not
sure
yeah.
E
Prs
or
PR
they're
really
interested
in
why
this
ultimately
we'd
like
to
augment
the
tied
status.
The
link
to
this
that
shows
a
specific
status
first
or
sorry,
information
for
just
a
single
specific
PR,
so
that,
if
you're
on
any
PR
page
on
github,
you
can
click
the
details
link
and
it
will
take
you
to
this
page
for
that
PR
I.
D
Think
that's
going
to
happen,
but
right
now
the
in
pointless
is
only
designed
for
getting
your,
so
it
should
be
pretty
straightforward
to
expand
to
that,
and
definitely
the
focus
is
on
being
able
to
link
to
it
from
tide
and
for
users
to
be
able
to
see
their
PRS,
but
I
think
we
can
expand
it
to
and
also
we
can
make
the
cruises
fancy
as
we
need
because
it
uses
the
token
when
you
log
in
so
you're
only
dossing
yourself,
making
the
query
more
complex.
We
should
be
able
to
do
that
so.
B
I
think
the
Gili
constructive
feedback
than
I
have
would
be
to
solicit
feedback
from
new
users
such
as
those
who
are
kind
of
coming
in
from
code
rebec's,
who
are
getting
shepherded
because
they
would
be
the
primary
consumers
of
this.
Because
a
lot
of
times
I
see
folks
struggle
terribly
with
trying
to
get
their
PRS
in.
And
you
know
trying
to
retest
and
figure
out
what
labels
they're
missing
and
why
I.
D
Also,
I
don't
want
to
know
too
much
but
Aaron
your
labels
descriptions
if
you
could
make
that
consumable,
I'd
like
to
extend
this
so
that
you
can
hover
the
labels
and/or
get
a
link
to
that.
Okay,
because
this
tells
you
what
labels
you
need
to
fix.
What
the
next
step
is.
We
need
a
way
to
explain
how
do
I
fix
this
label.
We
have
a
little
bit
there
for
this,
but
I'd
really
like
to
give
more
detail.
Okay,.
A
A
Yeah
I
saw
the
proposal
go
by
and
it's
pretty
neat
to
see
it
in
action,
so
I
think
all
those
steps
forward
in
improving
me
in
refining
the
query
and
making
tide
ultimately
point
to
this
are
are
perfect.
This
seems
like
you
demo
it
in
front.
I
can
Trebek's
anytime,
you
want
to
and
there's
be
a
community
facing
demo
that
happens
like
a
week
or
two
do.
D
F
A
A
Yeah
yeah
I
understand,
and
so
that's
what
I
wasn't
sure
if
this
doesn't
it
doesn't
seem
to
me
like
rolling
this
out,
would
be
disruptive
of
the
existing
stuff
that
we
have,
but
I
defer
to
to
you
guys
if
you
think
that
this
would
disrupt
anything
more.
D
A
H
I
I,
don't
have
a
demo
handy
right
now,
but
it's
not
too
interesting
when
it's
working.
Basically,
we
have
we've
been
having
an
occasional
across
those
failures
where
the
windows
or
the
back
build,
will
just
fail,
because
someone
checks
in
code
that
uses
Linux,
Linux,
isms
or
whatever
I've
had
a
peer
out
for
about
a
month.
Now.
H
I
think
that
we'll
do
a
cross
type
check,
not
a
full
compile,
but
it
will
parse
and
try
to
see
if
the
semantics
work
for
everything-
and
this
can
verify
that
in
a
couple
of
minutes
for
the
whole
repo
for
the
kubernetes
repo.
And
hopefully
that
will
mean
that
we
can
prevent
breaking
the
windows.
Build
of
the
Mac
builds
all
together
and
there's
some
future
work
where
I
think
it
could
be
a
good
check,
because,
right
now
it
feels
like
right
now.
H
Building
and
testing
is
abstracted
from
dance
to
the
point
where
we
see
a
lot
of
PRS
that
don't
even
build
every
day
and
I'm,
hoping
that
I
can
get
fast
enough,
that
there
is
some
sort
of
pre
submit
level
in
two
seconds
check
with
the
files
that
you've
changed
are
still
buildable,
and
we
can
get
better
verification
like
that.
Just
so,
people
have
better
edit
debug.
E
H
So
you
can
avoid
all
that
it
takes
about
20
seconds
to
do
a
full
type
check
for
one
architecture
or
for
one
platform,
platform,
architecture,
combination,
yeah,
I,
don't
I,
don't
think
it
will
get
in
before
the
code.
Freeze,
there's
been
a
chain
of
bugs
in
go
and
ghost
libraries
for
type-checking
that
I've
been
discovering,
but
it
should
hopefully
be
in
soon
after
code
freeze
and
prevent
future
issues.
H
H
H
A
D
With
verify
is
somewhere
on
my
backlog,
we
should
not
expand
that
job,
but
not
only
should
we
run
it
as
a
separate
job,
we're
considering
if
we
get
things
stable
with
it,
making
it
a
pre
step
for
all
the
build
jobs,
so
they
can
fill
out
faster
and
using
it
to
using
the
full
cross-check
to
replace
most
instances
of
running
the
cross
build
okay.
That's
all
the
other
questions
answered.
H
D
D
A
D
Super
quick
update,
I've
got
the
basil
remote
cache
pretty
stable.
Now
there
are
few
things
that
need
working
on
it
works,
but
like
there's
some
performance
improvements
and
things
we
could
do.
We've
rolled
it
out
to
some
of
the
CI
jobs,
as
well
as
pull
kubernetes
basil
test,
which
runs
our
unit
tests
and
we've
seen
about
a
70
to
80
percent
reduction
and
test
time
for
basil
test.
D
So
we've
got
that
down
to
around
six
minutes
on
average
and
instead
of
like
twenty
five-plus
I'm
hoping
we
can
bring
it
down
further
and
I've
got
some
other
changes
headed
out
for
the
head.
So
when
we're
talking
about
people
making
blocking
jobs,
I'm
hoping
we
can
avoid
any
of
them
being
longer,
because
I
think
we
can
get
the
existing
jobs
down
to
run
a
lot
faster.
D
A
D
Some
things
I'm
going
to
do
for
that
I'm
going
to
at
some
point
we're
gonna
flip
this
on
for
build.
The
concerns
are
some
things
that
don't
catch
very
well
and
there's
some
other
fixes
that
need
to
be
made
first,
but
since
the
bill
will
be
a
little
bit
faster,
it'll
be
cached
that
will
speed
up
the
endurance
because
most
of
them
are
using
the
basil
book
now,
but
not
significantly,
compared
to
their
overall
test
link.
A
B
It's
telling
me
it's:
there
has
to
be
like
a
PR
for
every
single
release
cycle
into
the
jobs
list,
for
a
release
Parker
and
it
just
as
we
go
forwards
and
eventually
the
artifacts
are
built
from
Basel
it
just
it
would
I,
don't
know
if
anyone
has
thought
about
unifying
the
fact
that
we
always
do
end
through
n
minus
two
four
releases
and
instead
of
us
having
to
PR
this
super
long
modification
for
the
jobs
list
of
just
being
able
to
support
the
end
to
end.
But.
D
A
B
A
Yeah
I
think
the
one
six
jobs
for
the
biggest
thing
we
needed
to
roll
off
before
everything
had
followed
that
stable
one
stable
to
stable
three
prefix
I
think
that
thing
is
just
that
has
humans.
Humans
still
feel
a
lot
more
confident
and
understand
what
they're
looking
at
when,
like
the
test
grid
tabs,
have
the
actual
release
number
in
there
as
opposed
to
you.
What's
this
end,
and
one
thing:
what
does
n
mean
today
versus
tomorrow?
That
question
would
come
up
a
lot.
B
Yeah
cuz
the
test
grid
for
release,
still
shows
like
you
know,
we
are
into
110
booster.
Are
we
sick,
even
gonna
release
the
170
more?
That's
like
the
question.
It's
like.
Once
we
enter
in
free
zone
for
110,
we
start
doing
release
candidates
within
the
neck
coming
month.
Well,
when
is
the
drop
for
you
every.
A
Time,
I
get
really
antsy
about
this
sort
of
thing
the
pushback
I
get.
Is
you
still
want
to
make
sure
that
we're
capable
of
releasing
in
the
event
of
like
a
security
related
patch,
even
though
like
according
to
our
Doc's
yeah,
we
only
support
well
at
when
110
is
out
the
door.
We
technically
support,
110,
181,
nine
or
with
110
one
nine
one,
eight
like
we
still
kind
of
sort
of
have
this
fuzzy,
but
we'll
support
one
seven.
A
If
there's
something
super
critical
that
has
to
happen
so
we
usually
historically
I-
think
we've
waited
until
a
couple
weeks
into
the
next
release
cycle
before
we
drop
that
off.
But
I
love
to,
like
clarify
that
you
know
from
a
steering
committee
perspective.
I
think
that's
something
that
the
release
thing
gets
to
define
rather
than
the
project
as
a
whole,
but
it
does
have
like
and
user
indicate
implications
and
security
implications
and
stuff
like
that.
A
So
yeah
I
think
like
if
you
do
find
areas
to
him
where
that
that
seems
weird,
let's
try
and
identify
it
to
see.
If
we
can
smooth
that
out,
because
progress
has
has
been
made,
it's
just
sort
of
you
didn't
notice
that
I
guess
alright
Josh.
You
wanted
to
ask
a
little
bit
about
outreach.
He
also
were
three
minutes
a
minute.
I
We
participated,
kuben
A's,
participating
at
vici
on
the
last
round,
with
some
six
CLI
projects,
Paris
and
I
have
been
looking
at
doing
another
round
of
outreach
b-but.
This
hinges
on
finding
cigs,
who
have
projects
that
they're
interested
in
entering
somebody
on
I
was
thinking
that
test
infra
from
the
number
of
sort
of
open
to
do's
would
be
good
but
the,
but
the
the
issue
obviously
is
somebody
having
being
able
to
make
a
time.
Commitment
from
entering
mentoring
is
somewhat
time
consuming.
We're
saying
an
average
is
about
five
hours
a
week.
A
If
maybe,
if
you
like,
sort
that
list
by
date,
you
can
at
least
see
the
ones
that
have
been
touched
most
recent,
but
the
rest
of
them
might
be
in
need
of
a
get
scrubbing,
so
I'm
gonna
try
and
go
through
at
least
all
of
the
existing
things
that
are
tagged
as
Help
Wanted.
If
you
know
of
low-hanging
fruit,
please
just
slash
help.
It
I
think.
D
A
Know
we
had
one
and
it
got
removed
somehow
that
discussion
yeah
the
difference
between
Help
Wanted
and
four
new
contributors
and
honestly
I'm
totally
fine,
saying
that
four
new
contributors
Alain
will
come
back.
It
just
like
make
that
something
that
can
also
be
added
via
bottom
and
totally
happy
to
see
that
discussion,
taking
up
being
in
Trebek's.
Okay,.