►
From YouTube: SIG Cluster Lifecycle 2021-05-18
A
A
All
right,
so
I
have
one
topic
for
today
and
I
asked
somebody
from
vmware
who
is
involved
in
that
to
join
the
call
and
discuss
it,
but
I
don't
think
they
will
join.
So
I
guess
they're
busy.
Basically
sick
testing,
as
you
may
know,
had
a
super
project
that
is
called
testing.
A
Commons
testing
commons
was
responsible
for
maintaining
the
end-to-end
test
framework
and
refactoring
it
to
make
it
better
suitable
for
external
use,
but
this
framework
still
resides
in
kubernetes
kubernetes
test
and
eventually
they
wanted
somebody
to
take
ownership
of
the
whole
effort
to
create
a
new
framework
from
scratch
because
they
didn't
like
the
original
design.
A
I
I
think
a
lot
of
people
agreed
with
that
that
ginkgo
and
parts
of
the
design
were
not
were
not
ideal,
so
I
think
that
volume
vivian
from
vmware
took
some
charge
to
propose
a
new
framework
design,
and
this
is
the
design
document
it's
located
in
kubernetes
six,
this
framework,
I
read
the
design
doc
and
it
left
a
good
impression
on
me.
A
You
should
check
it
out,
I'm
not
going
to
review
it
here
in
detail,
but
it's
basically
the
tldr
is
that
it
creates
features,
features,
go
into
futures
sets,
and
then
you
run
tests
from
the
feature
sets
and
each
feature
has
a
set
of
steps
that
you
can
execute
to
test
this
particular
feature.
A
A
A
A
This
will
become
the
official
framework
that
is
supported
by
sikh
testing,
and
it
would
be
nice
that
a
lot
of
projects
move
to
it
eventually
and
then
the
of
course.
The
most
important
part
is
that
it's
consumable
as
a
repository
and
you
don't
have
to
fork
kk
to
use
a
framework.
A
B
It's
very
interesting.
You
should
mention
that,
so
one
of
the
things
we
have
been
doing
in
chaos
is
moving
to
cube
test
two
and
with
that
we
have,
we
have
at
last
gained
the
ability
to
run
our
own
ede
tests.
B
In
addition
to
the
so
you
know,
we
regularly
run
the
kke
tests
and
like
peter
in
particular,
and
some
others
are
doing
great
work
to
do
this
peter
rifle,
but
they
we
can
now
run
our
own
tests
as
well,
and
I
think
we
they
added
one
recently,
which
runs
the
I
think
the
aws
csi
tests.
B
So
we
have
this
ability
to
run
tests
that
are
distinct
from
the
upstream
ede
tests
and
we
are
not
bound
to
use
like
a
framework
of
any
sort,
and
I
think
currently,
I
don't
think,
there's
a
standard,
that's
sort
of
the
answer,
I'm
writing
a
few
that
are
not
using.
I,
like
I
like
go
test
to
work
as
it
were.
You
know
like
to
just
work,
so
I
tend
to
just
use
go
test,
and
inevitably
I
am
building
up
my
own
framework
as
it
were.
B
Like
you
know,
you
can't
write
e
tests
without
refactoring
out
common
code,
but
it's
not
a
testing
framework,
it's
more
a
library
of
functions
to
to
interact
with
the
the
kubernetes
api
in
a
way,
that's
more
convenient
I'll,
put
a
link
to
that
in
the
agenda
when,
when
I'm
done
talking
in
a
second,
but
I
mean
I
think
I
like
I
like
this
approach,
and
I
think
I
think
the
the
lesson
I
would
take
away
from
this-
is
we
don't
have
to
like
as
a
kk
as
kubernetes
like
come
up
with
an
official
next
generation
test
project,
and
you
know,
do
big
design
up
front.
B
We
can.
We
can
have
a
bunch
of
projects
that
try
a
bunch
of
different
things
and
we
can
see
what
works
and
what
doesn't-
and
we
can
use
this
group,
for
example,
to
like
collaborate
and
discuss
what
works
and
what
doesn't
work
well,
and
if
you
want
to
try
this
framework
in
in
kubernetes
or
in
the
other
project.
I
think
that's,
that's
wonderful,
and
I
think
you
know
that's
that's
good.
I
certainly
share
the
concerns
with
jinko.
I
guess
I
think
I
think
it
just
wasn't.
B
A
Yeah,
I
think
this
is
pretty
much
the
current
state
everybody's
trying
to
see
what
works
for
them.
Inevitably,
all
these
separate
efforts
will
have
to
create
the
set
of
common
utils,
which
is
like
how
do
you
like
apply
timeout
on
a
certain
port?
How
do
you
make
sure
that
a
certain
deployment
has
a
number
of
replicas
the
expected,
so
these
essentially
will
become
very
common
and
they
already
reside
in
the
kk
framework.
B
Sorry,
I
failed
it
soon.
The
ones
in
the
in
the
central
in
the
kkk
library
again
have
sort
of
evolved,
and
there
are.
There
are
a
number
of
functions
that
deliver
that
functionality,
rather
than
there
being
a
library
that,
like
makes
it
easy
to
interact
with
the
with
client,
go
effectively
from
a
test,
and
so
I'm
I
will
put
a
link
to
this
sort
of
abstraction
that
I'm
putting
together,
but
I
I
certainly
agree
with
the
idea
of
creating
a
nicer
abstraction.
B
I
don't
know
that
it's
I
don't
know
that
it
matters.
To
be
honest:
if
we
have
five
of
them,
it
would
be
better
or
ten
if
it
may
be
better
if
we
had
one,
but
I
don't,
I
don't
think
it's
a
big
deal
to
have
a
bunch
of
them
and
and
then
eventually
like
converge
on
something
else
like.
I
know
that
now
that's
a
little
inefficient,
but
it
is
open
source
and
I
think
I
think
open
source
tends
to
produce
duplication
of
effort
over
or
over,
like
trying
to
like
converge.
A
Yeah
definitely
it
is,
I
mean
also.
This
is
also
very
new.
This
new
framework,
I
really
don't
know
what
is
going
to
happen,
but
I
was
hoping
that
void
can
join
this
meeting.
For
instance,
your
feedback
is
valuable
that
you're
already
starting
something
like
your
own
framework
that,
if
maybe
void,
implemented
your
ideas
in
such
a
way
that
you
can
agree
to
consume
his
framework.
B
I
mean
for
tests
it
doesn't
it
doesn't
really
matter
like
I
mean,
certainly
if,
if
there's
a
nice
library
that
works,
I
think
people
will
use
it,
but
I
think
yeah
it's
hard
when
everyone's
sort
of
getting
started,
but
yeah
I'll
definitely
take
a
look
at
this
and
and
see
if
we
can
switch
to
it
I'll,
put
a
link
into
the
it's
just
a
pr
at
this
stage
for
our
our
go
tests.
A
All
right
any
more
comments
of
this
entering
this
framework.
A
Today,
cool
moving
to
subproject
updates
for
cube
adm,
I
really
don't
have
anything
to
announce
in
this
meeting.
We
are
working
on
the
v1
beta
3
api.
If
you
have
any
feedback
for
what
we
have
planned
for
v1p3.
You
can
add
this
comment,
your
comments
to
this
particular
issue.
We
have
a
cap
that
we
updated
for
122..
A
So
basically,
the
idea
is
to
have
a
new
api
version
in
one
twin
cubed,
m1
122,
like
most
of
the
high
priority
items,
are
already
merged,
and
now
we
are
left
with
some
less
lesser
priority
items,
but
basically
the
these
are
the
best
effort.
At
this
point,
we
will
see
what
we
have
the
time
for,
and
it's
a
good
thing
that
we
have
a
an
extended
release
cycle
this
time
you
know,
as
you
may
know,
the
release
schedule
of
kubernetes
moved
to
four
months
per
release,
which
is,
in
my
opinion,
an
improvement.
B
Thank
you,
yeah,
I'm
we
we.
Basically
we
did
a
few
releases
recently.
We
have
our
121
release
coming
soon,
which
is
basically
where
we
want
to
be
in
terms
of
schedule,
so
we
are
ahead
of
where
we
normally
are.
B
I
guess
on
schedule
and
our
122
branch
is
open
earlier
in
the
schedule
than
it
normally
is,
and
so
we
are
hoping
to
get
in
some
bigger
features
like
add-ons
and
things
like
that,
but
nothing
massive
has
has
yet
landed,
but
so
we
are,
but
we
are
hoping
for
a
a
happy
122
cycle
with
lots
of
fun
stuff
coming
in.
A
Thank
you.
I.
I
joined
one
of
the
seek
off
meetings
and
I'm
not
sure
if
they
contacted
you,
but
basically
they
were
complaining
that
one
of
the
cops
jobs
I
think
it
was
cops,
gce
master
or
something
like
that
is
very
flaky.
I
think
it's
due
to
resource
cons,
constraints
on
the
job
setup
itself
and
it
was
flaking
a
lot
of
sea
golf
tests
because
those
require
apparently
more
resources.
A
Did
they
ping
you
on
slack
or
something
like
that.
If.
B
A
Yeah,
I
mean
it
was
jordan,
so
he
might
also
ping
you
on
the
internal
slack.
I
don't
know.
Basically
let
me
try
to
find
this
job
because
they,
I
think
this
is
the
second
time
they
are
discussing
this
flag,
because
what
what
happens
is
that
they
triage
their
inter
their
end-to-end
tests
based
on
overall
flakiness
in
the
entire
kubernetes
project.
Apparently,
if
some
some
of
the
jobs
are
very
flaky,
they
try
to
contact
the
people
that
are
responsible
for
the
jobs.
A
In
this
case
it
was
a
cops
job
and
they
conducted
us
about
a
cubic
year
problem
in
the
past
as
well,
but
I
guess
it
should
be
in
the
the
cops
does
work.
Maybe.
B
The
cops
actually
the
couch
cesar
is
pretty
big,
so
you
might
not
want
to
go
for
that
one
but
I'll
reach
out
to
jordan
and
try
to
figure
out
which
what
this
is
about,
because
I
haven't
seen
it
but
yeah
we
can.
We
can
try
to
figure
out
what's
going
on
there.
A
C
C
Hello,
I'm
stephen
I've
never
been
here
before,
but
hello.
All
of
you
nico
hasn't
been
here
in
a
while.
So
we
had
two
releases
119
120.,
some
of
the
biggest
things
we
had
an
auto
pause
feature
to
reduce
cpu
usage.
So
it
detects,
I
think,
after
five
minutes
nothing
hits
the
api.
C
It
will
kind
of
go
into
a
sleep
mode,
reduces
the
cpu
and
if
it
detects,
if
it
gets
a
new
api
ping
and
then
it
will
wake
itself
back
up
so
the
first
start
after
it
goes
to
sleep,
takes
a
little
bit
longer.
But
after
that's
back
to
normal,
we
had
a
mini
cube,
build
command
because
currently
before
this.
C
You
always
had
to
have
an
adapter
on
your
machine
if
you
wanted
to
build
things.
So
what
this
does
is
you
should
do
mini
cube,
build
and
then,
like
the
image
or
the
docker
file,
you
will
use
mini
cube's
internal
docker,
which
will
prevent
the
user
from
having
to
have
docker
installed
on
their
host
machine
itself.
C
We
also
had
an
add-on
to
support
gcr
credentials.
I
don't
know
a
ton
about
that
and
then
we
also
added
a
mini
cube
cp,
so
you
can
easily
copy
files
directly
into
mini
cube
without
having
to
do
any
ssh
craziness.
A
C
Yeah
I
was
actually
well
yeah.
I
was
experimenting
a
bit
with
the
build
when
we
were
actually
having
to
like
get
the
actual
like
socket
and
everything
so
that
there
was
a
lot
more
steps
involved
to
do
it
kind
of
manually,
we're
manually
doing
like
build
kit
commands
ourselves
just
to
kind
of
like
test
the
prototype,
so
the
mini
cube
drill
command
is
much
much
simpler.
It
doesn't
require
all
the
setup
and
ss18
and
getting
a
bunch
of
stuff
set
up.
A
Let's
see,
okay,
it's
lower
race.
I
I
think
kubernetes
kubernetes
has
a
similar
problem.
We
currently
as
part
of
our
hack
scripts.
We
are
currently
requiring
the
user
host
to
have
docker,
and
I
think
ben
elder
from
the
kite
project
will
try
to
eventually
modify
that
to
basically
use
buildkit.
Instead,
it,
the
original
request,
came
from
the
the
kite
project.
I
think.
A
But
this
this,
this
set
of
updates
seems
very
nice.
I
also
like
mini
cube
copy.
It's
a
nice
extension
appreciate
it.