►
Description
Meeting agenda: https://docs.google.com/document/d/1KLsGGzNXQbsPeELCeF_q-f0h0CEGSe20xiwvcR2NlYM/edit#bookmark=id.obc6xsqrqj51
A
B
A
All
right
and
so
we're
talking
about
are
we
talking
about
this?
One
I
think.
B
Okay,
anyways:
let's
go
through
the
okay
yeah.
B
I
think
I
know
how,
but
it's
been
a
you,
have
a
Google
Cloud
account
you
can
use
for
it,
though,
or
no
I.
A
I
do
have
a
we
do,
have
a
gcp
account
that
I
can
use
to
create
clusters,
although
I
unfortunately,
like
just
yesterday,
I
got
an
email
that
apparently
like
our
their
Reef
they're
I.
Don't
have
a
key
right
now
that
I
can
do
it
with
because
I
guess
our
key
got
leaked
earlier
this
week,
so
they
had
to
decommission
the
key
that
we
were
using
and
you
know
we're
creating
new
ones.
So
I
I
can't
make
it
one
right
now.
Unfortunately,
okay.
B
Like
I'm
hoping,
we
don't
even
need
it,
because
if
this
PR
is
doing
what
we
think
it
should
basically,
it
should
effectively
be
updating
every
single
CI
job
that
runs
on
Google
Cloud
on
this
PR
to
uccm.
So
from
what
I
can
tell
in
the
past.
B
In
the
previous
run
like
it
was
all
the
tests
were
green,
so
I'm
assuming
it
works,
but
so
I
guess
like
the
next
thing
here
is
yeah
like
ask
Antonio
to
rebase
the
pr
or
just
like,
create
a
new
like
create
your
own
Branch,
just
like
all
of
his
ranked
and
just
like,
open
a
new
prison
and
then,
if
possible,
try
to
spin
up
a
cluster
yourself
on
gcp,
using
the
changes
here
and
see
it
just
like.
B
Do
a
sanity
check
on
the
cluster
make
sure
it's
configuring
everything
correctly
and
then
the
follow-up
to
that
would
be
for
all
the
pre-summit
jobs
that
ran
for
this
PR
cross-reference,
the
test,
the
test,
infra
config
for
the
proud
job
and
just
make
sure
that
it's
not
overriding
the
cloud
provider
flag
to
something
where
I
would
not
use
the
default.
Okay,.
A
A
What
is
what's
the
entry
point
to
running
these
tests,
then
yeah,
like
am
I,
just
going
to
use
kubetest
too,
or
something
like
that
to
well
I
guess:
I'd
already
have
a
cluster
up.
So
what
I
just
build
the
binaries
and
run
them
or
something.
B
So
you
have
to
Cluster
up
so
I
think
this
for
the
Sandy
chat.
I
was
just
thinking
just
use,
Cube
control
and
just
like
check
that
the
cloud
controller
manager
is
running
and
yeah
I
guess
it
would
be
good
if
you
can
access
it
into
the
master
VMS
and
just
check
what
the
flags
are
configured.
Actually
you.
B
Right
right
so.
B
A
Okay
and
I
could
like
I'm,
since
these
are
the
scripts
that
are
used
well.
I
guess,
like
I,
want
to
make
sure
that
I
exercise
these
scripts
right.
So
if
I
just
run,
if
I
just
use
like
Ginkgo
to
run
the
end-to-end
test
from
the
cloud
directory,
is
that
going
to
be
enough
to
exercise
these
scripts
or.
A
B
Yeah,
so
what
I,
what
I
run
is
I
just
run
cube
the
cube
up
script
directly,
which
should
stand
up
a
cluster
but
okay.
So
why
don't
I?
Show
you
okay,
so
this
is
this:
is
the
command
I
I
used
to
run
like
so
so
before
the
CL?
This
is
this
is
what
I
would
run
to
get
a
cloud
controller
manager
up
and
running?
Okay,
okay,
like
that
and
oh
you
know
what
let
me
put
it
on
the
stock
channel,
so
yeah.
A
B
B
This
is
what
this
is,
what
I
ran
like
before,
like
or
like,
if
I
wanted,
to
create
a
cluster
in
like
in
master
right
now,
that
is
using
control
manager,
that's
command,
I,
ran,
and
so
in
theory
that
should
still
work
assuming
like
nothing,
has
changed
significantly
and
so
I
think
what
the
expected
Behavior
should
be.
If
you
checked
out
Antonio's
branch,
you
shouldn't
have
to
specifically
specify
like
all
of
those
parameters,
I'll
accept
the
except
the
node
size
and
the
master
size.
B
Well,
actually,
okay,
so
the
the
no
case
so
the
node
size,
Masters,
are
irrelevant,
I
think
I,
just
use
bigger
values
because,
like
I
didn't
want
to
like
deal
with
Lego,
you
know
the
VM
is
too
small.
So
it's
like
slow
or
anything
like
that
right.
So
maybe
you
could
ignore
those,
but
the
feature
gate
override
and
the
Clapping
butterfly
override.
In
theory,
those
should
not
be
needed
anymore
or,
and
if
you
just
ran
like
Cube
up,
it
should
just
kind
of
directly
spin
up
CCM
clusters.
A
A
B
A
Okay,
all
right
cool
well,
like
yeah
I'll,
give
I'll
give
Antonio
like
a
day
to
respond
and
if
I
don't
hear
anything
from
him
by
like
end
the
day
tomorrow.
My
time,
which
should
be
like
plenty
of
time
assuming
he's
not
out
or
something
then
yeah
I
can
just
set
up
an
extra
PR
with
take
his
changes,
create
a
new
PR.
That's
rebased
and
yeah
like
give
it
a
try
out
on
Monday,
you
know,
and
if
it
works,
then
you
know
cool.
Maybe
we
can
just
go
ahead
and
yeah
see
what
tests
fail.
A
Like
am
I
gonna,
need
to
look
at
like
the
once
we
get
this
in
place.
Is
there
going
to
be
like
a
matrix
that
I
need
to
look
at
in,
like
the
yeah.
B
B
Yeah
and
there's
a
bit
of
like
a
timing
here,
because
129
just
kicked
off
so
right
or
not,
it
didn't
kick
off,
but
like
we,
we
kind
of
just
finished
up
like
128,
so
the
earlier
we
land
this
on
a
in
the
earlier
in
the
release
cycle.
Then
people
learn
like
if
we
break
test
like
it's
not
a
huge
deal
and
we
have
yeah
plenty
of
time
to
like
fix
it
totally
yeah,
whereas.
A
A
No
yeah,
no
I
totally
agree
with
you
like
now
is
a
good
time
for
us
to
do
this
so
like
I'll,
I'll,
try
to
like
I'll
just
try
to
prioritize
it
and,
like
probably
the
the
best
I
could
say,
is
probably
Monday
I
could
try
and
test
it
out.
Okay,.
B
A
A
B
A
B
The
the
other
requirement
is
from
the
cloud
provider
gcp
repo,
like
building
the
CCM
off
of
the
latest.
Oh
right
right,
yeah.
B
We
are
testing
against
like
some
fixed
known
version
of
ccm.orgs,
and
there
is
no
actual
test
coverage
anymore
for
if
the
latest
CCM
code
is
compatible
with
the
current
version,
and
so
the
awesome
Jordan
was
to
also
have
like
a
periodic
job
from
the
gcp
repo
that
is
periodically
like
pulling
latest
Master
like
rebuilding
CCM
with
that
and
then
testing
it
with
so
yeah
so
I
think
that's
the
other
thing
that
you
address
I
I
know:
Antonio
did
some
work
on
it.
A
A
I
guess
foreign.
A
There
was
a
question
that
I
had
from
a
colleague
well
this
was
this
isn't
related
to
what
we're
talking
about
now,
but
but
I'll
come
back
to
that.
But
from
here
what
you're
saying
is
like
somehow
we'll
need
to
build
a
version
of
this.
That
has
like
an
updated
Go
version
and
has
updated
like
to
0.29
I
guess.
B
B
B
B
So
there's
there's
basically
like
two
things
you
want
to
do
like.
First
is
we
want
to
update
all
the
pre-submit
jobs
and
post
them
jobs
in
KK
to
use
cloud
control
manager,
but
it
doesn't
have
to
be
a
cloud
controller
manager,
that's
built
on
the
latest
version
like
the
latest
source
code.
It
can
be
like
n
minus
one
version
of
it
as
long
as
it
brings
up
the
cluster
and
like
how's
it
working
right.
B
So
it's
basically
what
we
do
with
like
the
cni
or
the
CSI
like
we
can't
like
compile
like
the
cni
version,
based
on
like
the
latest
cni
apis.
We
kind
of
just
rely
on
the
fact
that,
like
we're
using
some
last
known
version
that
works
and
testing
that
against
the
latest,
KK
binaries-
so
that's
the
first
part,
but
because
we
now
lack
this
test
coverage,
we're
changing
some
CCM
code
isn't
actually
caught
by
the
test.
B
We
need
to
be
able
to
basically
periodically
like
check
that
the
some
latest
source
code
changes
in
KK
isn't
breaking
like
the
CCM.
The.
B
A
You
know
this
actually
fits
in
really
nicely
with
the
testing
work
that
I'm
doing
because,
like
what
I'm
doing
right
now
is
like
building
out
tests
that
you
could
build
from
the
cloud
provider
repo
so
like
I
think
ultimately
like
what
you.
So
what
you're
saying
now
just
to
make
sure
I
got
this
right
is
in
the
cloud
provider
repo.
We
should
have
a
pre-submit
test
that
basically
just
deploys
kubernetes
and
says:
okay.
If,
if
the
cluster
comes
up
and
is
in
a
healthy
State,
then
like
we're
good
to
go.
A
Basically,
we
haven't
broken
the
basic
assumptions,
so
that
should
be
like
our
first
test
and
then
like
the
work
that
I'm
doing
hopefully
eventually
will
get
to
a
place
where,
where
in
the
cloud
provider
repo,
not
only
will
we
run
a
pre-submit.
That
starts
a
cluster,
but
it
will
then
also
run
a
series
of
cloud
controller
tests
to
make
sure
that
everything
works
like
for
whatever.
That
means
for
the
for
the
cloud
provider.
A
A
B
A
Only
other
question
the
only
other
topic
I
had
was
I
just
wanted
to
bring
up
this
this
PR
just
to
see,
if
maybe
we
could
get
some
more
eyes
from
from
your
side
on
it.
There's
like
a
there's,
a
a
PR
up
to
remove
basil
in
favor
of
like
a
make
file
for
the
provider
and.
A
B
A
B
And
obviously
like
yeah
like
get
all
the
pieces
passing
and
then
yeah
I,
imagine
that
the
this
might
be
a
chicken
and
egg
problem,
but
I
I
bet
a
lot
of
those
pre-summit.
Jobs
specifically
invoke
like
a
bazel
command
to
build
and
publish
the
image
right.
B
A
B
A
A
Much
okay,
well
I,
guess,
have
a
good
day
then,
and
have
a
good
weekend
and
yeah.
We'll
talk
next
week,
see.