►
From YouTube: Kubernetes SIG Testing 2018-11-27
Description
A
B
B
So
that's
kind
of
there's
two
things
I
wanted
to
talk
about.
One
is
this
is
behavior
changing
and
that's
currently
Mexican
currency
would
effectively
be
one.
So
I
want
to
make
a
script
to
update
any
existing
cron
job
to
be
nice
like
add
the
Mexican
currency
one
and
then
anybody
else
using
prowl
would
have
to
update
their
depth.
Job
definitions
or
potentially
have
Mexican
currency
go
over
1
zeros
infinite.
There
was
some
discussion
on
the
PR,
but
yeah,
so
I
think
people
are
ok
with
it.
I
just
wanted
to
kind
of
get
consensus.
A
C
So
you
know
to
be
serial,
there's
a
number
of
them.
So
if
you
have
like
say
a
scalability
test,
sometimes
it
takes
longer
than
expected
so
we're
like
basically
scheduling
things
as
quickly
as
possible.
Another
one
is
actually
say
like
build
jobs.
Typically
for
those
we
just
set
some
extremely
low
interval
instead
of
a
cron
and
we
lean
on
it
won't
run
them
more
than
like,
actually
continuously.
We
don't
want
to
be
running
like
ten
builds
at
once,
but
we
do
want
to
build
as
quickly
as
we
can.
C
B
C
I
think
I
think
that's
a
feature
that
we
want.
The
only
question
is:
do
we
want
the
cleaner
route
of
the
default,
just
being
zero,
the
go
style
and
take
the
you
know
effect
that
people
who
have,
probably
intentionally
or
unintentionally,
been
leaning
on
the
existing
behavior,
get
bitten
by
the
change,
or
do
we
want
the
ugliness
of
like
some
magical
default
value
I
mean
all
we
need
is
to
put
it
into
announcements,
markdown
and
then
it's
fine
right,
I
think
so
this
one's
probably
fairly
breaking
of
like
reasonably
common
jobs.
C
I,
don't
have
a
good
way
to
grasp
like
how
much
say
like
SEO
depends
on
the
stabilization
or
not
because
the
thing
that
I
think
about
this
is
that
you
could
very
easily
depend
on
this
without
knowing
about
it
as
well.
So
if
we
make
an
announcement,
we
need
to
make
it,
as
is
like
widely
as
well
like.
We
might
want
to
consider
more
than
just
announcements
that'd
be
for
this,
because
is
this
tea
of
building
their
own
version
of
the
images
or
are
they
just
using
whatever
you
this
poem?
C
A
A
C
That
now
is
that
not
I
think
we
should
probably
reach
out
to
the
people
that
we
know
yeah
publicly
and
mention
like
hey
when
you
update
you're
gonna
want
to
check
this.
B
So
the
other
thing
is:
I
only
did
this
for
Kron
style,
not
interval
style.
Do
we
want
to
do
it
for
interval
style
as
well?
That's
kind
of
there's
a
comment
that
says
it's
deprecated,
so
my
thought
was:
if
you're
not
updating
things
yourself,
then
maybe
you
get
the
old
deprecated
behavior,
but
Bing
being
a
Googler,
defecate
deprecated
just
means
stable.
B
C
Force
everyone
to
think
about
Crohn's,
there's
a
lot
of
jobs
that
just
kind
of
like
and
run
it
once
an
hour.
They
don't
really
care
the
exact
schedule
and
I
kind
of
actually
helps
avoid
people
doing
stuff,
like
scheduling
all
of
the
jobs
to
run
at
the
beginning
in
an
hour
or
something
right,
so
I
think
we
probably
don't
actually
want
to
deprecate
that
also
because
the
functionality
is
pretty
simple
and
I
think
that
we
do
probably
want
this
behavior
for
it,
because
it
was
full
and
it.
D
C
B
The
thing
that
I
think
makes
it
ugly
is
that
they're
the
base,
proud
job,
has
a
max
concurrency
field
which
is
just
ignored
by
the
interval
jobs,
and
the
comment
in
there
says
zero
is,
you
know,
infinite
concurrency,
so
it
currently
like,
if
we
implement
it
to
where
one
job
has
one
default
and
the
other
ones
have
another
default
than
when
you're
using
the
base
job.
It
gets
interpreted
differently,
based
on
which,
like
embed,
you
know
what
it
is
embedded
in
seems
like
a
very
strange
behavior
yeah.
It's
something
else.
B
D
C
And
there's
not
too
many
of
those,
but
they
didn't
need
to
explicitly
define
it
before
and
they
definitely
need
it.
The
other
thing,
I
guess,
is
just
that
if
this
were
kubernetes
I'm
pretty
sure
the
way
this
would
work
is
you
would
have
a
new
API
version.
Your
the
old
API
version
would
keep
doing
what
it
did
and
we
don't
have
that
going
on
in
proud.
So
we
have
this
v1
API,
but
we're
going
to
change
how
it
behaves.
That's
what
we
do
that
we
do
have
I.
D
I
would
disagree,
we
do
have
it
because
the
API
string,
you're
referring
to
is
not
defined
anywhere.
You
can't
actually
find
it
if
you
want
to
go
look
for
it,
so
it's
just
like
kubernetes,
because
you
don't
know
where
the
new
ones
defined
either
I'm.
It's
a
joke.
Right!
Try
to
fight,
go,
find
index
of
all
those
API
strings
in
Cooper,
so
I
think.
A
Bend,
the
other
thing,
too,
is
like
aside
from,
like
being
being
more
diligent
on
the
versioning
of
the
erd
I.
Think
we
at
some
point
should
probably
just
create
a
C
or
D
in
versions
actual
config
yeah
that
might
be
nicer.
It
might
out
that
we
might
also
be
able
to
get
around
some
of
the
issues
with
because
at
some
point,
I
think
the
larger
our
deployments
are
going
to
start
hitting
the
config
map
size
limits.
I
would.
C
Also
say:
yeah
we've:
we
have
some
problems
that
we've
applied
limit
yeah,
so
we've
hit
apply
limit
and
soon
I
think
our
job
definitions
are
going
to
hit
the
completion.
I
would
also
have.
If
we
do
this,
we
should
probably
start
with
V
1
alpha
1
yeah
V
1,
oh
yeah.
Well,
when
I
made
that
I
have
we
had
that
conversation
in.
C
C
C
C
D
So
this
has
happened
before
I
forgot
what
it
was
I
think
I
think
Steve
actually
Steve.
You
may
have
raised
the
issue
before
to
do
with
regular
expressions
matching
tests
is
there.
Is
there
may
be
a
potential
Phyllis
philosophical
behavior?
We
might
want
to
modify
where
we
avoid
defaults
or
unspecified
defaults
or
employ
rather
implied
defaults
and
try
to
be
more
explicit
so
that
people
do
think
about
these
things
moving
forward,
or
at
least
if
they
don't
have
to
think
about
it,
because
it's
just
there's
it
just
seems
like
this.
A
Know
I
think
I
think
the
change
here
like
moving
it
to
having
no
zero
implying
unlimited.
Does
that
so
I
think
that
does
move
into
like.
If
you
need
concurrency,
you
should
have
explicitly
said
it
at
the
same
time.
I
think
like
with
the
way
to
the
configuration
is
structured
right
now.
If
we
like
there's
so
much
for
paucity
and
there's
so
much
repetition
that
if
we
made
everything
was
it
it
would.
Oh.
D
A
D
C
D
Think
I've
seen
zero
more
frequently
than
that
I
I
guess
what
I
was
suggesting
I
mean
that
empty
not
be
taken
to
me,
necessarily
the
zero
value
it
being
taken
to
me
to
mean
empty,
and
in
this
case
there
could
be
the
opportunity
to
say.
Oh,
you
didn't
specify
that's
on
us
to
define
that
behavior
and
you
shouldn't
count
on
what
that
means
moving
forward.
If
we.
C
A
A
C
A
C
I'm
fine,
with
going
forward
with
either
change
as
long
as
we
just
make
a
we
make
a
beyond
the
not
so
inside
of
deal
is
contact.
Some
people,
yeah
I,
think
that's
enough
to
diligence.
Consider
and
let's
consider
improving
this
in
the
future,
possibly
to
something
like
a
separate
global
defaulter,
I.
E
B
E
B
You
if
you
puck
the
plank-
and
let's
say
you,
set
your
max
concurrency
at
1:
it
will
queue
up
all
of
those
jobs
and
you
can
end
up
with,
like
new
jobs,
never
actually
being
run,
because
you
just
you're
just
continually
adding.
So
if
you
have
a
job
that
runs
every
minute
and
it
takes
10
minutes
to
run
your
your
job
queue
will
just
grow
infinitely.
We
can't
Q
maximum
one
job.
We
should
well
that's
that's
what
her
role
of
game
is
doing.
B
C
E
Sure
yeah
so
I'm
working
on
my
greeting
out
a
community
center
in
tests
from
boos
I
think
I
got
them
running.
The
last
missing
piece
I
think
is
T.
Currently
booth
graph
merges
this
method.
Json
generated
from
q
passed
in
json
and
those
data
are
used
by
both
goob
nader
and
test
grit.
To
show,
for
example,
teaming
number
and
build
number.
Oh
I
guess
mostly
most
important
is
the
timing
number
of
the
example
the
version
of
canary
is.
E
You
are
testing
so
I'm
trying
to
let
sidecar
mimic
their
similar
behavior
merged
metadata
JSON
into
finish
the
song
Steve
mentioned
he
has
a
artifact
uploader
controller
and
which
can
handle
pot
logs,
but
I.
Just
discussed
with
and
looks
like.
The
only
thing.
I'm
not
sure,
is
how
that
controller
can
handle
like
artifact,
Pyro's,
yeah.
C
Well,
specifically,
logs
I,
don't
know
if
it
makes
sense
for
deep
aisles,
yeah,
I'm,
not
sure,
but
but
it,
but
like
totally
independent
of
that,
I
would
really
like
us
to
move
forward
on
light.
I've
talked,
like
both
sides
doesn't
really
matter
to
me
who
I
don't
have
been
with,
but
someone
who's
working
at
this
problem,
like
I'd
like
to
get
in
particular
test
great
working
with
the
commit
ideas.
A
Yeah
I'm,
not
certain,
it's
gonna,
be
much
more
work
in
either
direction,
but
I
think
one
of
the
things
that
something
that
lives
outside
of
the
job
like
one
of
the
benefits
of
that
is
you
know.
If
you
have
a
full-fledged
client,
you
can
look
at
the
entire
project,
see
Rd
I
mean
I,
think
we
had
talked
about
a
couple
times
like
actually
uploading
the
Gamo
for
the
proud
job
and
for
the
pod.
C
Are
those
like
the
the
one
that's
under
test
so
like
the
the
primary
it's
like
out
just
like
plumbing
through
that
commit
ID
is
super
useful,
like
we
can
lose
stuff
like
the
custom
for
one?
That's
not
super
important,
but
the
one
that's
under
test
that
test
grid
gives
you
the
functionality
to
like
drag
between
key
right.
E
E
Would
would
use,
for
example,
say
communities
testing
for
a
master.
It
should
be
that's
ID
you
and
want
they're
not
to
commit
it
resolves
you.
We
want
to
commit
okay,
also
there's
for
the
communities
interim
task.
It's
actually
extracted
from
cue
past,
because
we
don't
actually
clone
the
canary
repo.
We
download
the
table
all
and
we
keep
tests
will
populate
the
metadata
that
JSON,
with
a
common
number
and
hatch
from
the
table.
C
Well,
I
mean
we
can
come
back
later
and
figure
out
some
way
to
pass
that
to
the
sidecar,
but
again
I
think
that
would
be
easy
enough
to
changed
later
or
potentially
we
can
even
replace
the
entire
TCS
Leon
at
some
point.
It's
just
that's
a
lot
more
investment
right
now
we're
mostly
doing
this
thank
bootstrap
was,
but
we
dropped
that
field
and
that
field
is
super
useful
for
CI.
So,
besides
even
the
custom
metadata
that
one
is
like
blocking
me
from
booth,
a
call
apart
again.
A
A
We
should
and
I
mean
I
didn't
really
have
any
blockers
in
the
PR,
but
maybe
we
should
put.
Can
we
update
upload
the
just
the
metadata
and
stuff
and
then,
unless
you,
because
you're
also
doing
this
like
the
downward
API
suspect
right,
which
is
a
subset
of
the
proud
job
spec?
Unless
you
need
that
as
a
specific
yellow
file,
maybe
we
should
wait
on
doing
that
until
we
could
just
upload
the
actual
projects
like
itself.
D
Is
there
a
reason
to
not
just
try
to
keep
all
of
that
data
I,
don't
think
it
would
be
too
large,
as
opposed
to
trying
to
decide
what,
for
example,
commit
ID
to
use
based
on
the
situation
say.
Oh,
if
there
are
three
different
types,
have
three
different
fields,
and
if
it's
one
of
those
it
would
be
populated,
we
keep
all
of
that
in
another.
A
D
A
C
F
That
only
takes
30
seconds,
probably
it's
just
sort
of
a
shout
out
hey.
We
exist
where
there's
been
some
discussion
of
supports
and
long-term
support
and
what
that
could
mean
in
terms
of
what
users
want
and
implications
to
the
project
where
we
to
even
go
there.
So
the
workgroup
is
forming
it's
just
a
place
to
talk
about
these
things.
Catalog
enumerate
the
problem.
Space
figure
out
what
might
be
possibilities
so
obviously
were
any
changes
to
be
proposed.
They
would
dramatically
impact
testing
and
y'all
would
be
a
key
blocking
stakeholder
on
that.
F
But
we
would
like
you
to
not
just
at
the
point
that
some
cap
is
bubbled
out
to
be
like.
Oh
now,
that's
horrible
ridiculous
could
never
work
but
to
be
involved
in
the
discussions
and
help
in
shape
guide
what
we
might
do
for
improvements
for
users
of
the
project.
So
there's
a
link
there
to
the
meeting
agenda.
It's
Tuesdays
9:00
a.m.
Pacific,
alternating
weeks
with
sig
cluster
life
cycle
and
the
PR
link
for
sig
zamel.
F
That
also
includes
like
a
work
in
progress,
draft
charter,
and/or,
readme
and
some
sort
of
links
and
general
information,
but
we're
trying
to
did
the
conversations
go
in
towards
some
type
of
action,
because
this
has
been
talked
about
a
lot
and
there's
a
gap
versus
what
users
want,
but
we
need
to
turn
that
into
some
concrete
actions
to
improve
the
situation
for
operators
and
users.
So
that's
basically,
it
I
think
questions
well,.
C
I
mean
I,
guess
the
only
thing
that's
going
to
come
up
at
some
point.
That's
not
exactly
cig
testing,
but
we
wind
up
being
involved
in
a
lot
is,
is
like
which
artifacts
are
the
community
or
not
like
I
know
recently
the
Debian's
and
the
RPMs
are
a
bit
awkward
because
of
the
way
they're
hosted
and
there's
even
been
some
discussion
that
I'm
sorted
in
favor
of
at
this
point
that,
like
maybe,
we
should
try
to
host
all
the
ways
of
installing
it
I'm
sure
that'll
get
interesting
in
that
group
too.
Yeah.