►
From YouTube: Kubernetes SIG Testing 2019-02-12
Description
A
Everyone
this
is
the
success
ting
weekly
meeting
for
February
12th.
This
is
going
to
be
recorded
and
shot
up
to
YouTube.
At
some
point
future.
We
have
three
topics
for
conversation,
so
hopefully
a
short
one
I
am
taking
the
first
one
I
just
kind
of
want
to
have
it
open,
open,
open,
mic
questions
comments
concerns
current
state
of
proud
the
repo
right
now
we
do
have
a
broken
git
client,
and
so
it's
not
possible
to
update
the
current
set
out.
There
was
a
conversation
yesterday
about
issuing
the
revert.
A
C
A
D
A
Getting
master
working
is
hurting
the
I
think
the
issue
with
with
this
one
specifically
to
it's
like
their
I
guess,
untouched,
the
pool
right
now
and
so
the
stuff
that
went
in
just
kind
of
went
in
right.
So
if
we
I
mean
I
agree
that
it
might
be
good
to
hold
off
online,
we
factors,
but
at
the
same
time,
do
we
then
add
features
to
untrusted
code.
That's
core
to
pow.
D
Well,
that
there
were
factors
to
make
things
more.
Testable
are
certainly
the
first
thing
we
should
address
after
we
get
master
working
again.
I
think
I
agree
that
yeah
we
should
be.
If
we
have
significant
areas
that
are
causing
outages.
When
we
make
breaking
changes
or
when
we
make
changes,
then
we
need
to
be
refactoring
those
to
be
testable.
I.
D
A
And
so
that's
also
similar
to
the
plan
config
loading
code
as
well.
We're
like
look
the
code
right
now
calls
out
to
you,
know:
OS
get
ends
and
there's
no
real
thing
for
that.
Similarly,
the
git
client
right
now
literally
calls
exec
command
and
there's
nothing
for
that
either.
So
it
doesn't
does
seem
like
the
current
flow
system.
E
I
mean
calling.
There's
the
you
know,
reader
interfaces
to
deal
with
files
and
whatnot
I
think
there
are
smaller
changes
that
we
could
do
to
make
things
testable
rather
than
refactoring
everything.
Although
refactoring
things
might
be
a
good
solution,
but
I
would
be
more
comfortable
doing
that
once
we
get
in
the
unit
test
that
we
have
already
shown
had
caused
us
outages.
E
I
mean
no
I
mean
or
potentially
yeah.
If
we
don't
want
to
revert,
I
mean
I,
think
reverting
is
fine,
I
think
also,
if
it's
like,
oh
hey,
we
goofed
and
here's
a
one-line
fix,
and
here
is
a
unit
test
that
cut
acted
like
Tessa.
This
exact
scenario,
I
think
that
is
also
fine
but
I
think
they're
responding
to
that
by
not
reverting,
but
instead
sending
the
two
thousand
line
PR
to
refactor
everything,
I
think
that
is
potentially
and
most
likely
a
good.
E
Mean
yeah
I
mean
in
you
know,
general
right,
like
I,
think
all
the
things
that
we're
doing
is
necessary.
It
sort
of
just
I
think
would
be
useful
to
get
to
a
working
State
for
a
day
or
two
and
get
to
where
you
Steve
and
us
can
be
somewhat
close
to
head,
because
I
think
right
now
we're
still
back
in
January
for
our
proud
employment
and
so
be
a
good
thing
to
fix.
Before
we
continue
on
our
relentless
refactoring
bandwagon.
I
think.
D
A
So
if
we
revert
this-
and
we
say
like
it's-
a
good
idea
not
to
be
largely
factors,
do
we
like,
as
a
policy
do
we
want
to,
for
instance,
accept
a
PR
that
adds
you
know
some
small
subset
of
testing
and
changes
that
github
enterprise
support
and
accept
the
risk
that
that
might
have
not
added
it
enough
testing
to
actually
be
sufficient,
or
do
we
say,
let's
not
add,
features
to
this
until
we're
like
in
a
place
where
we
can
actually
test
the
functionality
of
the
fire.
I
think.
D
A
Any
other
topics
on
the
ordered,
updates,
yeah
I,
think
generally
I'm,
also
having
important
updates,
not
a
good
time,
I
feel
like
do.
We
because
also
I
mean
we've
had
a
couple
conversations
recently
too
about
like
sort
of
v1
compatibility
like
backwards
compatibility
sort
of
stuff.
Do
we
think
we're
at
a
point
where
we
can
say
like
like
we
can
never
have
a
breaking
update
again
in
the
future.
I
feel.
E
Like
historically,
the
pattern
that
we
followed
is
to
make
sure
that
whenever
we,
you
know,
add
new
functionality
or
change
existing
functionality,
there's
some
sort
of
migration
period.
Where
we
support
you
know
both
patterns,
which
often
can
you
know,
is
technically
more
complicated
and
I
feel
like
that's
worked
well
for
us,
so
I
feel
like
we
could
just
continue
doing
that
and
I
don't
exactly
know
how
that
would
work
for
this
particular
example.
You
know
when
we're
migrating
from
status.
E
The
pattern
that
we've
been
going
and
that's
more
or
less
well,
because
it's
giving
people
time
to
upgrade
and
given
us
time
to
upgrade
without
regressing
any
functionality.
And
so
then
we
can
get
to
where
the
new
world
is
and
then
we
can
remove
the
deprecated
functionality.
I
feel
like
that
pattern
has
been
working
and
I
feel
like
in
the
last
month,
or
so
we
haven't
been
as
rigorous
in
supporting
that
and
I.
Think
that's
where
we
run
into
trouble.
Yeah.
A
D
E
E
Although
to
you
know
to
be
fair,
I
feel
like
the
primary
people
complaining
mostly
have
been.
You
know,
I
feel,
like
we've
been
inflicting
most
of
the
pain
on
ourselves
that
it
hasn't
been
a
lot
of
complaints
from
you
know
it's
either
red
hat
or
the
case
instance.
That
is
feeling
the
brunt
of
things
and
not
so
much.
You
know
complaints
from
other
people
saying
hey
you
guys
hit,
though
it's
more
like
hey.
We
need
support
for
a
github
enterprise
and
you
don't
have
it
Lynden.
Like
hey.
D
We
have
a
lot
of
the
complexity,
though,
and
we
also
have
enough,
like
user
volume,
that
we
probably
run
into
or
discover
most
of
the
issues.
First,
that's
a
good
point
yeah.
Maybe
we
should,
let's
maybe
should
we
just
try
and
be
more
noisy
about
updating
announcements,
MB
and
like
also
pinging,
some
other
channels.
While
we
do
that.
A
D
One
thing
that
was
discussed
was
staging
prowl.
We
discussed
that
yet
again,
I
think
the
two
things
were
discussed
want
a
been
mentioned,
potentially
something
like
the
subreddit
simulator
kind
of
just
having
a
bunch
of
bots
like
spamming
in
their
own
little
prowl,
repo
or
whatever.
That's
that's,
that'd,
be
kind
of
cool.
You
might
also
be
good
to
have
like
a
more
kind
of
fraud
canary
kind
of
deal
where
we
have
tests
tests
infra
on
its
own
crowd,
applaud.
C
C
C
D
C
Argue
in
so
I
guess
the
main
problem
is.
We
need
something
that
we're
okay,
caring,
but
we
also,
if
we
do
it
to
like,
you,
know
proud,
sulfur
to
test
them
for
them.
We
have
a
kind
of
a
problem
or
when
we
break
it,
we
have
problems,
fixing
it
because
it's
broken
since
it
kind
of
automates
itself.
So.
A
D
E
Think
that
might
accident
I
feel
like
we
either
we
should
be
I
mean
the
the
problem
is
were
proud,
was
not
really
designed
to
have
multiple
browsers
running
in
managing
the
same
repo,
and
so
we
would
either
need
a
way
to
black.
You
know
to
vote
to
not
have
the
primary
prowl
instance
handle
the
whatever
either
the
test
info,
repo
or
other
canary
repo,
or
else
we
would
have
to
use
a
different
port,
because
a
lot
of
our
plugins
and
stuff
are
configured
to
target
org
level.
It.
C
E
E
Mean
we
so
we
have
the
three
jobs
right
that
perm
that
pushes
the
new
image
on
each
commit
and
then
creates
a
that
then
applies
whatever
the
current
configures
on
these
commits
and
then
creates
a
PR
to
update
the
config
that
it
will
apply
and
we
could
essentially
have
a
fourth
step.
That
does
something
that's
a
diverse
and
yeah,
so
I
mean
I
feel
like
if
we,
if
we
did
all
that,
then
we
could
essentially
get
better
protection
and
make
sure
the
test.
E
A
E
Yeah
I
think
there's
more
complicated
things,
but
you
know
for
the
most
part
we
could
just
make
the
PR
creator
say,
but
let
go
check
this.
You
know,
give
it
four
links
and
make
sure
to
open
those
and
make
sure
they
look
same
last
out
is
I.
Don't
know
that
right,
you
know
how
much
load
we
generate.
You
didn't
say,
diverse
I,
don't
know
it
did.
It
would
actually
have
caught
the
issue
right.
D
I
feel
like
if
we
have
something
like
they'd
have
like
a
diverse,
it's
gonna.
It
would
certainly
be
helpful,
but
I
think
we're
gonna.
Let
miss
like
a
ton
of
issues
just
kind
of
like.
Historically,
if
you
look
at
like
the
kubernetes
security
repo
and
how,
though
there's
been
see
you
there,
I
would
imagine
that
there'd
be
similar
types
of
skew
with
a
bot.
You
know
just
a
bot
run
repo.
D
A
A
A
B
B
We
demoed
this
I
think
about
six
months
ago
when
it
was
kind
of
barely
existent
and
it's
now
better
and
could
possibly
replace
Governator
in
the
next
few
weeks.
Maybe
so
I'm
going
to
show
you
what
we
we
have
and
then
people
can
complain
about
why
it's
not
good
enough
or
something
if
I
can
find
the
share
button.
Why
do
I
have
the
sixteen
different
options
here?
No.
E
B
So
this
is
what
spyglass
looks
like
running
on
my
workstation,
because
prowl
is
still
in
January.
So
basically
you
get
a
page
like
this,
which
has
your
links
to
various
pages.
I,
don't
know
you
can
see
under
there
as
well
as
the
same
metadata,
but
gubin
a
Turkish
handily
summarized
into
a
sentence
or
you
can
see
all
of
it
like
goober.
Later
we
show
you,
the
tests
have
failed
somewhat
more
compactly.
You
can
expand
them
for
the
same
detail.
Google
Nature
gives
you
and
you
can
click
this
link
below
there.
B
We
go
and
see
the
associated
logs
with
no
coloring,
because
it
turns
out,
we
store
corruptive,
anti
colors
in
our
j-unit
files,
unless
we
have
recipes
shorter
at
that
is
less
broken
now
because
we
fixed
it
and
we
showed
we
build
Logs
with
highlighting,
and
we
also
do
the
same,
highlighting
good,
related
bars
of
words
that
sound
bad,
which
is
really
kind
of
useless
and
should
definitely
change
how
that
works
someday.
B
But
that's
not
a
common
aim
and
you
can
expand
skip
lines
and
you
can
see
the
whole
thing
like
goober
later
it
loads
foes
progressively
and
also
cause
positioning
on
my
machine,
it's
slower
than
it
would
normally
be,
and
that
should
open
in
your
window,
but
doesn't
so
as
a
bug.
So
there
is
feed
brief
demo
of
spyglass,
but
you
can
see
it
loading
progressively
instead
of
everything
having
to
be
ready,
like
Cuban,
eight
of
those
other.
B
B
This
is,
we
do
yes.
Well,
actually
we
don't
have
a
PR
dashboard
and
that
feature
of
goober
now
so
we
will
not
be
replacing.
Are
we
gonna
keep
running
the
river?
We
will
keep
running
goober
later
for
that
feature,
but
possibly
with
other
parts
out
of
it,
or
at
least
with
suggest
you
not.
So
we
have
this
page,
which
shows
that
this
job
doesn't
work,
and
similarly,
we
have
this
page,
also
inspired
by
our
uber
later
having
a
similar
page.
B
Have
been
mode
locks
I
do
not
currently
intend
to
implement
when
old
logs.
Unless
someone
explains
to
me
who
is
using
them
and
why,
to
my
current
knowledge,
nobody
does.
Although
we
have
seen
one
good
feature:
request
and
I'm
trying
to
chase
up
people
who
might
know
who's
using
them
other
than
that
we
don't
have.
Oh
and
I
haven't
actually
implemented
these
links
at
the
top
for
real
I.
Think
from
for
his
demo,
so
I
need
to
build
that
too
I
mean
that
I
don't
see
any
blocking
issues
and
I.
D
B
Idea
is,
but
we
would
switch
the
links
from
github.
There
would
be
a
banner,
probably
the
top
of
all
this
but
says
we
are
now
using
spyglass
and
sort
of
Luminizer.
You
can
go
to
google
native
here
contacts
suggesting
if
you
need
it
new
features
here,
because
Governator
will
eventually
go
away
in
that.