►
From YouTube: Kubernetes SIG Testing 2019-01-08
Description
A
To
the
cloud
okay,
hi
everybody-
this
is
the
kubernetes
sig
testing
weekly
meeting
today
is
Tuesday.
January
8th
I
already
gave
you
the
spiel
about
our
code
of
conduct,
so
please
adhere
to
it.
On
today's
meeting
agenda.
I'm
gonna
give
Patrick
about
ten
minutes
worth
of
time
to
talk
about
a
number
of
PRS.
He
wants
to
discuss
around
the
end
testing
framework
and
then
I
would
like
to
see
if
we
can
talk
about
our
evil
plans
for
the
v1
14
month
timeframe.
So
what
do
we,
as
the
sake,
want
to
try
and
accomplish
this
quarter?
A
B
Muted
and
I
was
on
the
wrong
screen.
I
was
actually
looking
at
my
document
and
zoom
dog
zum-zum
Bruno
was
elsewhere.
Okay,
so,
let's
get
started
ten
minutes
isn't
that
much
time,
but
what
I
wanted
to
achieve
anyway
is
just
primarily
to
get
some
some
attention
towards
fees,
open
pool
requests,
because
some
of
the
change
is
touch-free
end
to
end
framework
and
for
the
first
one
I
thought
that
it
might
also
be
useful
to
discuss,
perhaps
other
use
cases
for
the
change
that
I'm
making.
B
The
advantage
is
that
test
registration
code
can
check
command
line
parameters
like
on
which
on
which
host
system
am
I
running
or
what
what's?
What's
the
operating
system
thinks
chest
test
fit
traditionally
have
been
done
at
run.
Time
can
now
be
done
before
even
adding
a
test
that
then
can't
run
later.
B
That
might
actually
be
the
right
thing.
For
example,
some
of
the
tests
are
for
the
GCE
driver
that
doesn't
read
and
run
on
my
local
cluster,
so
I
don't
care
whether
that's
that's
registered
and
I.
Don't
want
to
see
skip
messages
at
all,
but
then
it
might
also
be
a
bit
sometimes
a
bit
harder
to
understand
that
some
tests
aren't
locked
down
on
the
test
suite
and
that's
that's
why
we
were
discussing
how
to
lock
that
information
I
try
to
implement
that.
But
it's
currently
not
really
possible.
B
In
a
consistent
way,
because
for
tests
call
that
skips
basically
an
its
statement
and
it
function
call
doesn't
know
of
a
final
test
name.
That
information
is
only
available
inside
ginko
and
there's
no
API
at
test
registration
time
to
retrieve
that
so
I
could
try
to
work
with
Kiko
upstream
to
to
end
to
fix
that.
But
in
the
short-term,
that's
probably
not.
C
B
Name
and
the
command
line
parameters.
No,
but
if
you
want
to
lock
something,
it
would
be
useful
to
know
what
the
test
actually
is
and
because
it's
the
skipping
happens
deep
down
in
a
describe
context.
Something
blahblah
call
stack
for
code
that
actually
has
the
if
check
and
knows
that
it
is
skipping,
something
doesn't
know
the
context.
It
doesn't
know
all
the
current
names
that
qing-kuo
been
concatenates
together.
But
if
you
use
the
actual.
C
B
A
B
D
C
B
C
B
You
could
you
basically
give
it
its
statement,
then,
would
need
to
call
something,
but
it
when
it
knows
that
it
needs
to
run
that
it.
When
initially
does
v-initial
leave
a
big,
the
initialization
that
completely
turns
around
the
pattern
that
normally
is
used
with
before
each
alright
I.
Don't
think
that
would
work
well,
we
don't.
B
C
B
Would
it
would
be,
it
would
be
perhaps
doable
I'm,
not
I'm,
not
a
particularly
big
fan
of
that
approach.
I
don't
figured
it
would
make
for
testing
simpler.
Let
me
let
me
give
another
example
where
I
think
that
skipping
tests
had
in
it
time
would
be
useful
right
now.
I
know
that
people
get
confused
when
they
run
a
conformance
test,
and
they
see
lots
of
skipped
tests,
or
at
least
bet
has
been
the
situation
a
while
back.
I,
don't
know
what
about
what
what's
currently
happening
and
I
think
it.
B
B
You
could
have
a
command-line
argument,
saying
I'm
running
conformance
tests
for
a
single
node
cluster
only
run
the
tests
that
are
suitable
for
that
situation
and
then,
instead
of
having
just
having
to
it
to
skip
tests
at
runtime,
where,
when
someone
manually
needs
to
check
whether
that,
whatever
it
was
okay
to
skip
the
test,
you
would
immediately
see
that
or
you.
It
would
immediately
be
obvious.
But
this
is
a
complete
test
run.
Putting.
C
In
a
different
way,
I
don't
think
it
makes
sense
to
do
test
registration
at
long
time
because,
like
if
you
do
that,
then
your
test
only
exists
like
how
am
I
supposed
to
know
what
tests
are
supposed
to
exist
unless
I
happen
to
be
on
like
the
correct
environment
that
allows
those
tests
to
exist.
That
sounds
like
very
painful
to
debug
in.
D
D
You
say
like
this
error:
these
are
all
the
entire
full
suite
of
tests
that
should
be
run
for
a
given
environment,
and
then
you
actually
do
the
run
and
the
run
gives
you
a
different
set
of
results
because
it
had
to
do
a
skip
list
for
X,
Y
or
Z,
because
you
do
are
running
on
cloud
provider
X.
You
know
you
are
doing
these
other
things
now.
Obviously,
conformity
should
never
do
that
and
conformance
shouldn't
run
time
skip.
Anything
like
there
was
originally
a
skip
list,
but
that
should
have
been
fixed
now
for
awhile
right.
B
A
And
the
details
of
this
but
I
agree
that
I
have
concerns
around
the
implications
for
conformance.
So
if
there
are
people
here
who
could
step
up
and
help
carry
it
forward,
I
would
appreciate
that
if
you
can't
find
those
people
here,
I
think
other
forums
for
discussion
could
be
architecture
or
the
conformance
working
group
yeah.
B
No,
that's!
That's
fine
I'll!
Try!
This
I'll
try
without
delay
it
test
registration
if
I
can't
get
it
to
work
sufficiently.
I
might
have
to
come
back
or
ask
ask
for
help.
But
let's
let
me
try
something
else
first
and
see
well
how
high
I
get
button?
Let's,
let's,
let's
move
on
to
the
next
port
topic,
the
next
poll
request,
and
that
is
about
the
test,
see
the
channel
files
that
are
primarily
used
in
in
six
storage
for
things
like
installing
the
host
path
drive
host,
puff
CSI
driver.
B
B
We
were
trying
to
replace,
or
we
were
considering
replacing
using
local
files
with
downloading
by
on
the
URL
at
runtime,
but
then
both
I
and
other
people
on
the
six
storage,
a
select
channel,
had
concerns
about
more
flaky
tests
and
the
external
dependencies.
So
what
I
came
up
with
in
this
pull
request
is
an
approach
where
the
test,
the
framework
itself
is
giving
is
given
a
URL,
but
it
has.
There
are
helper
scripts
that
use
that
information
to
populate
a
local
cache
which
gets
committed
together
with
URL
update.
E
B
At
runtime,
what
happens
is
that
the
file
is
a
being
used
from
from
the
local
cache,
like
any
other
of
e,
manifest
files.
But
then
we
have
another
update
script
under
hack.
That,
after
URL
change
makes
sure
that
the
cache
is
consistent
and
there
is
a
verify
script
that
I
intend
to
add
to
a
prowl
job
so
that
only
if
anything
in
that
directory
changes
it
will
check
bad.
The
cache
is
indeed
consistent.
B
A
D
That
most
of
the
data
for
the
test
were
hermetically
sealed
enough
to
be.
It
gets
weird
when
you're
starting
to
download
the
more.
We
and
I
take
a
look
at
the
exact
details
of
what
you're
doing
and
why
but
I'm
generally
concerned
with
wise
storage
special
case
in
these
two
different
scenarios,
because.
B
E
B
Have,
for
example,
of
his
VV
external,
provisional
external
attacher:
these
are
images
that
are
released
by
vi
storage
as
docker
images
and
for
using
those
images
you
need
to
have
certain
about
Firebug
rules
installed,
for
example,
and
those
rules
gets
documented
and
published
in
the
individual
repositories
for
each
of
these
components.
Why
is
the
test
installing
these?
Because
not
all
CSI
drivers
are
installed
in
a
normal
cluster?
You
basically
have
to
install
the
driver
before
you
can
use
it.
The.
F
A
I
guess
a
lot
of
this
feels
like
it
is
discussion
that
is
centered
around
the
e2e
testing
framework
and
to
me
this
is
begging
the
question
of
whether
or
not
the
testing
Commons
sub-project
is
a
more
appropriate
forum
to
steward
this
discussion,
and
if
we
want
to
pick
that
thing
back
up
because
I
feel
like
we
haven't
had
that
meeting
in
quite
a
while
and
it's
a
slightly
more
European
friendly
time
zone.
So.
E
A
What
I'm
hearing
from
Patrick
are
requests
for
help
from
people
who
have
sufficient
experience
and
knowledge
experience
with
the
framework
and
historical
knowledge
of
the
project
to
understand
where
all
the
bodies
are
buried
and
I
feel
like
Ben
is
relatively
occupied
with
kind
it's
not
something.
He
can
provide
full-time
I'm,
pretty
locked
up
with
the
114
release.
It's
not
something.
I
can
dedicate
full
time.
I
know.
Tim.
St.
A
A
Don't
really
see
a
lot
of
time
for
a
big
long-winded
discussion
on
that
and
and
I
think
Patrick
raises
very
valid
points
that
Andrew
you're
right
there
like
this
problem
will
only
grow
in
complexity
as
we
look
to
separate
out
cloud
providers
and
the
configuration
for
a
cluster
under
test
increases
in
complexity,
doesn't
decrease
so
like
finding
the
right
patterns
to
make
it
easier
for
everybody
to
copy
paste
and
rub
your
stand.
Tests
out
would
be
would
be
helpful.
I
just
don't
know
if
this
is
the
right
forum
to
figure
that
out.
E
B
E
C
Mean
so
for
all
of
those
I
would
expect
you
have
some
tool
like
so
for
a
bunch
of
RC
IQ
tests,
that
is,
that
is
calling
the
cluster
of
and
some
other
things
and
in
the
test
or
whatever
you're
using
on
it
works
reasonably
or
1q
Batum,
and
at
that
point
you
would
install
act
like
it
would
be
like
an
add-on
and
I
wouldn't
expect
any
dot
test
install
that
I'd
expect
whatever
was
calling
e
to
e
dot
test.
Do
that
so.
A
E
A
B
Yeah
but
then
you
also
have
to
have
some
feedback
loop
between
I'm
intending
to
run
these
tests,
and
these
tests
need
this
setup
and
we
trust
the
setup
there
needs
to
know
basically
yeah
what
what
kind
of
what?
What
kind
of
setup
is
needed
for
the
tests
that
are
intended
to
be
running.
It
might
not
be
a
full
test,
for
example,
and
likewise
tests
then
need
to
be
able
to
check
that
the
setup
has
been.
It
has
been
done.
Well,
they
might
be
able
to
observe
that
so,
but
that
might
be
covered.
Why.
B
C
To
get
both
right
if
I
want
to
run
that
I
didn't
get
a
test
right.
So
the
thing
that
is
running
the
test
can
do
that,
though
we
can
say:
I
want
to
run
the
host
path
test
and
I
can
say
I'm
going
to
install
this
I
mean
like
imagine
for
like
the
end
user
workflow.
If
they
want
to
test
their
cluster,
I
would
not
expect
as
an
end
user,
that,
if
you're
testing
some
drivers
that
you
also
went
and
installed
those
drivers
as
part
of
the
test.
B
A
B
E
I'd
like
to
talk
with
you
more
about
this,
it's
exactly
what
we're
doing
it.
Vmware
and
I
understand
what
you're
saying
right:
it's
it's
a
test
matrix!
It's
not
it's!
No
longer
just
a
two-dimensional
graph.
We
weave
in
with
external
providers,
we've
introduced
a
new
access
and
it's
just
a
bevy
of
new
points
of
test
and
I.
Like
Aaron,
you
said
like
a
pre-conditioner
like
a
pretest
and
I
would
I
would
almost
even
say,
but
it
could
also
be
considered
a
post
cluster,
because
these
now
are
part
of
the
cluster
right.
E
A
That's
all
fair
and
just
to
be
clear.
The
reason
I've
mentioned
Canadian
was
because
I
think
comedian
has
tried
to
be
very
explicit
in
defining
a
linear
set
of
phases
that
involved
that
gets
you
to
standing
up
a
cluster
and
I'm.
Trying
to
think
of
this
in
that
mindset
like
what's
the
next
linear
set
of
steps
and
where
do
those
end
before
we
start
kicking
off
tests,
so
just
some
administrative,
since
we're
gonna
lose
our
room
in
that
four
minutes.
A
This
sounds
like
it
merits
enough
to
talk
in
a
separate
forum,
I'm
totally
happy
to
host
a
breakout
discussion
on
this
zoom
or
the
testing
Commons
meeting
might
be
a
more
appropriate
place,
I
feel
like
Patrick
and
Andrew.
Could
you
guys
start
a
discussion
on
the
mailing
list
to
figure
out
what
makes
the
most
sense
for
folks?
Second
I
would
like
to
actually
have
a
discussion
about
our
play
and
v1
14
work,
so
I'm
gonna
kick
that
to
next
week
and
try
and
keep
next
week
open
for
that.
A
If
you
do,
you
have
things
you
think
this
sake
should
be
focusing
on
or
you're
planning
to
work
on.
On
behalf
of
this
sig,
please
apply
them
114
milestone
of
those
issues
I'm
going
through
and
scrubbing
all
the
things
this
week
Tim
you
looked
like
you
had
something
to
come.
Some
clarity,
look
that
you
had
something
to
say:
I'm.
D
A
F
So
we
have
a
bunch
of
you,
know,
automation,
to
update
dependencies
with
the
hack
up
to
update
depths
and
update
basil
and
there's
you
know
some
support.
Essentially
basil
will
also
support,
go
modules,
and
so
I
would
like
to
you
know,
assuming
that
we
get
all
of
the
same
automation,
working
with
go
modules.
I
would
like
to
switch
over
to
Cheston
for
repo
to
be
using
the
go
modules
rather
than
depth
and
yeah.
C
F
C
You
test
installs
I,
believe
go
modules.
Has
some
different
behavior
around
things
like
good
good?
The
other
thing
that
I
would
really
look
out
for
which
is
by
having
associated
products,
yet
is
that
go
modules
doesn't
use
vendor
I
think
you
can
vendor,
but
it
won't
read
it
so
for
like
test,
so
we
will
probably
need
to
run
a
module
cache
like
Athens.
We
might
want
to
start
looking
into
that.
E
Sorry
Andrew
going
in
oh
yeah
I
when
I
saw
this
on
the
magenta,
actually
put
a
comment
about
20
minutes
ago
on
this,
because
I
was
concerned
that,
like
I,
think
this
is
a
good
idea,
but
there
are
still
at
least
to
my
knowledge,
pending
Python
scripts.
So,
for
example,
like
the
the
s3
work,
they
are
the
the
I
forgot.
What
it
was
it
was
when
it
gets
the
bits
and
I
was
gonna,
enhance
it
for
s3,
but,
like
oh,
it's
Python,
we
don't
want
to
do
new.
Python
works
want
to
convert
it.
To
go.
F
E
This
was
but
at
least
this
particular
thing
was
a
special
case.
You
have
a
go
program
for
it,
but
you
haven't
started
using
it
yet
and
I'm
saying
if
there's
outstanding
go
work
that
you
want
to.
Maybe
we
can
take
this
offline.
Yeah
I'm,
familiar
with
go
modules,
have
strong
feelings
on
them,
because
I'm
friends
with
some
Poirier
and
yeah
but
I,
think
I
think
they're
interesting.
But
in
this
particular
case
it
was.
E
There
was
a
go
program
that
you
wanted
to
start
using
to
get
bits
off
of
the
the
buckets,
but
you
were
using
Python
and
you
hadn't
converted
the
go
program
yeah
and
when
I
raised
the
topic
of
enhancing
the
go
program,
he
said
well,
we're
not
using
it
yet,
and
then
you
had
no
timeline
for
migrating
from
Python
to
go,
even
though
there
was
a
go
program
for
it.
It's
the
test
grade
scraper
link.
A
I
couldn't
remember
what
it
was
for,
but
we're
not
actually
using
to
scrape
things
out
of
TCS
and
some
of
over
the
test
grade
live
ring
so
anyway
yeah
we
should
take
this
offline.
Please
never
use
the
words
bash
and
vendor
again
things
like
that
scare
me.
I
shouldn't
have
dependencies
like
that,
but.