►
From YouTube: Kubernetes SIG-GCP meeting, 2017.11.09
Description
Notes at bit.ly/k8s-sig-gcp
A
Okay,
I
guess
wrong.
First
of
all,
thank
you
all
for
showing
up
and
sig
TCP
is
a
new
special
interest
group
around
the
Google
cloud
platform
and
I've
got
an
agenda.
I
hope
you
can
all
you've
all
opened
it.
It's
been
linked
to
it
through
the
the
email
announcements
and
it's
on
the
slack
channel,
header
and
stuff,
and
but
we're
going
to
go
through
briefly.
A
What
the
point
of
this
is
and
the
meetings
themselves
should
serve
as
a
place
where
people
who
are
kind
of
intimately
tied
to
have
kubernetes
works
on
Google
cloud
platform
can
find
out
what
work
is
underway.
What
changes
might
be
being
made
if
they
have
proposals
for
work
that
they
want
to
undertake?
They
should
be
able
to
bring
them
here
and
get
some
useful
feedback,
and
then
this
should
really
be
light
at
the
central
forum
for
that
kind
of
activity
and
the.
A
If
there's
an
on
goal
for
this,
for
this
meeting,
I
think
what
worried
me
a
little
when
I
was
thinking
of
setting
it
up
was
it
could
easily
turn
into
a
place
where
people
just
show
up
with
their
Google
cloud
platform.
Problems
and
I
think
everybody
who's
used
this
as
bump
into
kind
of
like
a
pretty
long
list
of
problems
and
if
it
just
became
like
office
hours
for
fixing
you,
you
know
you've
the
issues
you're
facing
with
Google
Cloud
I.
A
A
You
know
people
won't
answer
questions.
This
isn't
a
place
to
come
for
help,
but
I
think
the
real
balance
needs
to
be
struck
and
that
kind
of
lead
me
to
be
the
grid
that
I
had
below,
where
I'm
explicitly
asking
people
to
kind
of
look
as
they
introduce
themselves,
not
simply
to
say
what
they
want
for
the
sake,
but
also
what
we're
hoping
to
provide
and
to
the
sig
and
people
should
be
every
need
as
many
givers
as
takers.
If
the
thing
is
to
balance
and
be
successful,
I
feel
oh.
A
The
other
thing
I
mentioned
before
I
kind
of
yield
and
see
if
other
people
are
comments
and
all
of
this
and
some
basic
proposals
for
how
the
group
should
work
together
and
I
think
these
are
very
common
to
pretty
much
kubernetes
as
a
whole
and
all
of
the
other
special
interest
groups.
But
it's
important
for
them
debate
and
discussion
be
very
respectful.
We
should
like
appreciate
that
everybody
else
is
trying
to
do
work
with
good
intentions,
trying
to
do
the
right
thing
and
is
I.
A
Think
one
of
the
big
places
is,
you
know,
assume
people
are
not
idiots.
People
have
reasons
for
thinking
things
we
should
find
them
out
and
that
kind
of
also
leads
to
kind
of
be
inclusive,
as
many
viewpoints
as
possible
lead
to
a
much
kind
of
stronger
product
and
that's
kind
of
that's
an
important
thing.
A
dissident,
you
don't
understand,
or
we
don't
understand,
we
should
seek
to
find
out
what
it
is,
rather
than
just
assume
it's.
A
It's
not
relevant
and
the
other
thing
and
I
emphasize
this,
because
I'm
actually
I've
got
terrible
hearing
I
find
and
the
kind
of
video
chats
highly
enough
as
it
is,
and
but
if
we
are
talking,
if
you're
not
talking
I
should
say,
please
mute,
I
think
I
again,
that's
a
common
pattern.
It
is
important
the
voice.
We
lose
a
lot
of
time
and
so
I'm
I'll
shut
up
for
a
minute
in
case
anybody
on
the
call
or
anybody
around
the
table
and
wants
to
talk
about
what
they
think
GCP,
maybe
should
be
or
shouldn't
be.
C
So
hi
everyone
quick
overview
of
what
the
cloud
provider
effort
is.
Currently
today
we
have
a
cloud
interface
in
a
cloud
plug
in
that
cloud.
Plug-In
goes
into
the
cube,
API
server,
the
cube
controller
manager
and
the
cubelet
itself
and
is
used
to
make
calls
out
to
the
hosting
cloud
provider
to
do
various
things
like
provide
routes,
provide
IP
addresses,
provide
custom
volumes,
determine
whether
or
not
a
given
node
is
ill
consider
to
be
present
in
the
cloud
etc.
C
We
have
seven
entry,
implementations
of
a
cloud
provider,
the
biggest
three
pretty
obvious,
Google,
Amazon
and
Microsoft,
but
we
do
have
a
few
others,
I
think
like
VMware
or
one
from
red,
add
etcetera
and
due
to
the
fact
that
go
doesn't
have
a
particularly
good,
dynamic,
plug-in
mechanism.
That's
production-ready!
C
All
of
those
have
to
be
loaded
at
at
one
time
for
kubernetes
to
work,
irrespective
of
which
cloud
provider
you're,
using
or
in
fact,
even
if
you
are
using
cloud
provider.
So
that
means
we
have
a
lot
of
extra
code
running
that
isn't
needed,
which
is
bloating
up
certain
things.
It
also
means
that
for
the
cloud
providers
in
question,
they
have
to
go
and
get
their
code
reviewed
for
something
that
only
affects
them
and
so
for
all
of
the
cloud
providers.
This
ends
up
being
a
bit
of
a
slowdown
for
development
where
they
go.
C
I
want
to
make
this
change
to
the
way
I'm
a
cloud
provider
calls
and
it
has
to
be
reviewed
by
the
entire
open
source
community,
and
we
can't
do
very
quick
fixes
whether
you're
Google
Amazon
read
at
Microsoft,
so
that
the
idea
was
kit
that
we
came
up
with
was
hey.
Let's
attempt
to
shift
all
of
that
burden
into
an
isolated
cloud
code.
We
were
generally
at
this
point.
C
So
that's
the
plan.
We
generally
now
refer
to
it
as
stage
one
in
stage
two.
The
stage
one
plan
is
to
take
is
to
create
the
cloud
controller
manager,
entry
and
move
the
seven
big
in
repo
cloud
providers
into
it
and
prove
that
it
can
be
done
and
isolate
out
that
cloud
provider
code
and
then
Stage
two
is
actually
then
taking
that
and
moving
it
out
completely
out
of
tray.
C
We've
already
seen
some
benefits
so
not
particularly
great,
not
ones
we
see
for
like
Google,
but
all
new
cloud
providers
today
actually
start
out
in
quote-unquote
Stage
two.
So
if
you
look
at
companies
like
ranch,
there,
they've
got
working
kubernetes
in
stage
two,
where
they're
just
running
all
their
cloud:
specific
code
in
their
own
cloud,
controller
manager
and
everything's
working.
So
now
the
effort
is
for
the
big
seven
to
try
and
move
their
code
out
so
that
it
becomes
a
nicer
world
for
everyone
to
live
in
specifically
I'm
working
on
the
Google
portion
of
this.
C
So
a
couple
of
things
obviously
going
on
in
flight
one
is
that
we
need
a
new
repo.
So
at
some
point
in
the
future
there
will
be
an
additional
repo
where
all
the
Google
Cloud
specific
code
is
going
to
live
I.
Imagine
it's
also
going
to
be
where,
for
instance,
all
of
the
scripts
that
currently
live
under
cluster
GCE
are
going
to
move
etc.
C
So
you
know
at
this
point
there
are
a
cut
if
people
are
interested,
there
are
definitely
a
couple
of
documents.
I
can
link
for
the
various
progress
we've
seen.
There
are
also
several
PRS
out.
There's
a
work-in-progress
PR
that
I
have,
which
is
the
proof
of
concept,
so
we
actually
have
a
working
stage.
One
GCE
version
of
this.
There
are
some
hacks
in
it
which
I'm
not
hugely
fond
of,
but
it
at
least
proves
that
everything
can
be
done
and
then
allows
me
to
break
it
into
small
pieces.
C
The
other
other
interesting
points
of
note
on
that
there
are
going
to
be
some
we're
gonna
see
some
changes
as
things
move
along
I'm
in
the
process.
I,
don't
know
how
familiar
everywhere,
one
is
with
the
code
but
I'm
in
the
process
of
doing
things
like
breaking
down.
Node
controller
node
controllers
fairly
huge,
and
there
are
three
GCP
calls
in
it,
and
it
seems
unfortunate
to
move
the
entire
node
controller
over
to
the
cloud
controller
manager
if
we
can
avoid
it.
So
there's
going
to
be
some
Ryoga
texting.
Look
for
things
like
that.
C
C
So
the
full
move
out
may
be
delayed
on
that
I've
been
working
on
some
ways.
We
could
try
and
move
out.
Early
flex,
for
instance,
is
an
option.
It's
not
an
option
that
I
love,
especially
since
it's
been
around
for
a
lot
longer,
but
it
is
still
in
alpha
and
I.
Don't
think
anyone
really
wants
to
take
it
to
beta
with
CSI
coming.
So
that's
one
interesting
problem
that
we
have.
C
The
other
interesting
thing,
of
course,
is
that
we
are
still
going
through
and
finding
all
of
the
interesting
places
where
we
see
a
call
that
in
fact,
I
found
one
just
yesterday
in
cubelet.
That
says
it
well,
someone
else
found
it
pointed
out
to
me,
but
l
equals
GCE,
do
something
special,
and
so
it's
not
actually
a
cloud
call,
but
it
is
still
has
GCE
specific
code
and
all
of
that
needs
to
be
dealt
with
as
part
of
this
migration.
So
that's
the
top
level
view
I'm
happy
to
talk
about
any
details
that
anyone
has.
C
If
there's
anything
specific
you
want
to
know
about.
If
you
want
to
take
a
just
quickly
on
how
one
actually
goes
about
this,
the
three
systems
I
talked
about
the
cube,
API
server,
the
cube
controller
manager
and
the
cubelet
all
have
a
plugin
cloud
plugin
flag,
which
today
would
be
set
to
GCE
the
way
we
sort
of
experiment
with
this
is,
if
you
set
it
to
external,
then
that
service
can
no
longer
make
X
calls
to
the
cloud
plugin.
C
Obviously,
then,
you
also
need
to
bring
up
the
cue,
the
Cloud
Controller
manager,
which
the
cluster
up
scripts
today
do
not
do,
but
when
I
send
out
the
work-in-progress
I
can
RB
documentation.
I
can
also
link
the
work-in-progress
that
shows
the
four
other
significant
set
of
cloud
and
cluster
up
changes
that
are
needed
to
set
those
flags
and
get
the
whole
process
going.
D
Yeah
two
questions.
The
first
is:
this
is
all
fantastic
in
fact,
I'm
starting
the
newcomers,
the
whole
top
of
everything
that's
going
to
get
such
a
substance,
an
update.
Generally
speaking,
I
go
to
the
kubernetes
community
repo
to
find
out
about
all
of
the
working
groups
and
States
they're
involved
in
kubernetes,
and
today
the
cloud
provider
working
group
is
not
listed.
There
I
think
a
lot
of
the
information
that
you
just
talked
it
up
would
be
get
a
lot
more
traffic,
if
you
put
it
there
and
then
sort
of
broadcast
like
hey.
D
C
D
C
C
So
I
I
made
this
comment:
I
need
an
owner
for
the
controller
manager
for
this
project
to
have
any
value.
This,
apparently,
is
how
one
volunteers
to
become
the
owner
so
from
Google's
perspective
on
the
Google
side
of
this
picture,
I
now
own
controller
manager.
Okay,
obviously
we
need
owners
outside
of
Google
I
mean
even
inside
Google.
We
like
to
refer
to
the
what
what
if
someone
wins
the
lottery?
Oh
so
my
first
question
is:
okay,
great
I
own
controller
manager.
What
if
I
owned
what
if
I
win
the
lottery?
C
For
it
right,
and
so
where
I
come
to
would
be
what
if
I
win
the
lottery,
we
discussed
it
quite
a
bit
and
the
team,
the
sig
that
is
probably
most
affected
by
this
is
were
close.
Okay
and
so
I've
had
discussions
with
the
sig
workloads
team,
at
least
within
Google,
and
they
have
agreed
to
back
me
up
in
terms
of
you
know,
becoming
our
owners
becoming
very
knowledgeable,
etc.
On
that
code,
and
in
fact
a
majority
of
controllers
in
the
controller
manager
already
belonged
to
that,
so
to
be
clear
when.
D
C
Cool
any
other
questions,
all
right
and
sort
of
one
of
the
I
I
realized
it
wasn't
actually
asked,
but
just
as
an
implication
like
there
are,
a
lot
of
changes
that
are
gonna
be
coming
through.
I
have
been
working
extensively
with
dims
wlan0
and
Tim
Hawken
on
all
the
architectural
changes
I'm
trying
to
push
through-
and
there
are
I
mean
there's
about
nine
members
of
the
cloud
provider
working
group
at
this
point
so
and
I
think
I'm,
actually
the
only
person
from
Google
in
that
working
group.
C
So
there
is
somewhat
of
a
broad
representation
across
kubernetes
there.
The
other
thing
that
I
will
mention.
We
are
going
to
actually
need
a
lot
of
effort
from
outside
of
Google.
We
have
six
other
entry
providers
that
we
need
to
have
actually
start
pushing
this
through.
For
this
to
be
successful,
and
so
I
was
talking
with
Ken
from
say
gaps,
and
we
were
coming
up
with
a
plan
where,
in
110
we
were
going
to
start
actively
trying
to
find
an
owner
for
every
one
of
the
entry
cloud
providers
to
try
and
help
push
this
through.
C
D
He
will
find
tremendous
support
for
the
steering
committee.
In
that
regard.
You
talked
about
wanting
to
create
a
new
repo.
I
would
go
talk
to
save
architecture.
If
you
already
aren't
a
lawyer,
that's
probably
the
best
path
to
go
there
and
your
talk
of
eventually
removing
where
the
cluster
scripts
lives.
We
have
just
made
sure
that
you
get
in
touch
with
the
appropriate
people
and
say
testing
to
make
sure
we're.
You
know
where
we
should
be
pulling
the
appropriate
scripts
to
build
and
test
clusters.
D
C
Agree
with
that,
and
in
fact
I
don't
know,
but
this
is
getting
into
into
the
weeds.
But
what
are
the
issues
I
have?
Is
that
there's
a
fair
amount
of
start-up
control
that
is
actually
already
out
of
out
of
trees,
so
kubernetes
tests
infra
has
a
separate
set
of
startup
controls
that
control
the
sizes
of
all
the
pieces
and
how
things
work
when
doing
scale
and
performance
testing
and
I've
already
been
baiting
a
couple
of
times
there
when
I
start
up
an
additional
cloud
controller
manager.
C
Since
right
now
it
has
all
of
the
plugins
in
it
we're
seeing
a
slight
in
flight
of
the
size
of
the
master
such
that
it
no
longer
fits
into
n1
standard
one.
It
requires
an
n1
standard,
and
that
means
that,
right
now,
all
the
performance
and
scale
tests
fail
to
start,
because
I
have
no
easy
way
of
changing
the
startup
scripts
in
tests
infra.
Just
for
my
work
in
progress
so
that
it
will
work
yeah.
C
This
is
getting
a
little
off-topic,
but
we're
happy
to
take
that
up
over
to
testing
and
I
also
participate
in,
say
scalability
I
know
sounds
pretty
actively
involved
in
the
scale
and
performance
tests
and
as
the
CSA
only
I
they're
failing
for
their
turn
builders
to
turn
a
setting
like
so
awesome.
So
anyone
else
have
any
questions.
Anything
that'd
like
to
chat
about
the
cloud
provider
work.
C
No
part
of
this
is
intended
to
sort
of
fracture
kubernetes
and
in
fact,
one
of
the
things
that
we
need
to
spend
some
time
on
is
working
out
exactly
not
just
how
to
do
testing
and
make
sure
that
all
the
cloud
providers
continue
to
work
on
OSS
builds,
but
also
that
all
of
the
custom
cloud
provider
builds
are
not
only
are
working,
but
our
conformance
to
the
kubernetes
standards
of
what
it
means
to
be
kubernetes.
We
are
not
trying
to
fork,
or
you
know,
provide
different
differentiators
here
right.
B
C
That's
actually
an
excellent
question,
so
this
project
began
in
1:6,
which
is
actually
before
my
time
so
I
know
Tim,
Hawke
and
Brian
grant
from
Google
and
then
Siddarth
money
from
Rancher
did
a
lot
of
the
initial
push
for
this.
There
was
already
a
cloud
provider
interface
where
the
cloud
provider
plug-in
comes.
C
At
this
point
you
cannot
become
entry,
Sagarika
tech,
chure
said
no
more
entry
cloud
providers,
and
so
this
working
group,
but
a
lot
of
the
motivation
for
this
working
group,
is
giving
a
solution
for
out
of
tree
cloud
providers,
but
along
the
way
it
became
clear,
hey
if
we're
gonna
be
do
moving
everyone
out
of
tree.
We
need
a
level
playing
fields.
We
should
make
it
so
that
everyone
is
out
of
tree
and
in
fact
getting
everyone
out
of
tree
helps
everyone
both
in
that
I.
C
Don't
have
to
worry
about
being
accidentally
broken
by
Amazon
I
mean
it
function
in
Amazon,
where
they're
trying
to
set
themselves
up,
which
is
invoked
because
it's
go
and
init
functions
get
invoked
and
I.
Also,
you
know,
like
you,
can
shrink
the
size
of
my
code,
I'm
less
likely
to
be
impacted
by
them,
and
so
the
general
feeling
is
that
this
is
good
for
everyone,
both
in
the
abstract
and
in
the
particular
x',
but
yeah.
The
long
winded
way
of
saying
I,
don't
know
I
I
assume
that
all
happened
before
my
time.
C
At
this
point,
though,
you
know,
every
change
going
through
is
as
a
PR
and
I'm
being
as
many
of
the
Google
engineers
are
and
being
very
conscious
of
trying
to
make
sure
that
every
one
of
my
PRS
is
being
reviewed
by
at
least
one
non
Googler,
because
I
don't
want
to
have
any
perception
that
yo.
This
is
a
Google
project.
We
are
trying
to
be
inclusive
of
the
community
as
a
whole
and
so
I'm
trying
to
make
sure
that
every
one
of
my
PRS
related
to
this
is
being
reviewed
by
an
outside
member.
A
C
A
D
In
fact,
you
know
what
how
about
I
go
ahead
and
share
my
screen,
so
I
can
show
you
guys
why
why
I'm
driving
late?
What
problem
try
solve
here?
So
if
I
can
go
to
test
grid,
I
go
to
say:
Greece
I
go
to
the
city's
master
button,
I'm.
Looking
at
the
summary
tab,
which
shows
me
every
single
job
on
this
dashboard
I
expect
should
be
green,
or
at
least
not
failing
before
I
cut
her
release.
This
is
the
way
our
automation
scripts
are
supposed
to
work
right
now.
D
You
really
want
to
get
to
the
case
where
everything
should
be
great,
so
right
now,
I'm,
looking
at
a
summary
tab
that
shows
I'll
just
pull
the
text
up
a
little
bit
just
can't
see
all
the
gke
jobs,
a
family
g
c,
IG
k,
g:k
device,
plugin
GKC,
slow,
gke
cereal,
and
what
I,
what
I
want
to
find
is
somebody
who
owns
gke
I
can
ask
hey
what's
going
on.
Why
is
this
failing?
Is
this
a
known
issue?
D
Is
somebody
working
on
this
because,
effectively
if
these
jobs
are
failing,
if
the
tree
is
following
in
the
woods
and
nobody's
paying
attention,
then
why
is
it
even
there
if
these
jobs
aren't
actively
being
fixed?
Why
are
we
bothering
to
block
the
release
lemon
family?
It's
doing
nobody!
Any
good,
so
sick
GCP
seems
like
a
pretty
natural
place
to
to
to
look
to
for
people
who
are
experts
in
gke,
and
you
know
what's
going
on
and
how
to
fix
this
up
until
the
existence
of
this
thing.
D
I
I,
don't
really
know
what
the
public
facing
way
is
to
get
in
touch
with
the
gggg
KA
people.
I've
tribally
like
go
in
the
cluster
lifecycle,
because
I
know
like
Robert,
H,
Daley
and
a
couple
other
Googlers
hanging
out
there.
But
what
I'm
looking
for
is
like
an
official
point
of
contact,
so
in
a
similar
basis,
like
I
know,
if
the
AWS
job
fails,
I
go
pokes
a
gate
in
the
US,
because
I
know
there
are
people
there
who
know
about
AWS
and
they
also
know
about
cops,
which
is
the
way
that
AWS
is.
D
If
the
scaleability
job
is
failing.
I
know
cokes
a
scalability,
because
I
know
they're
interested
in
what's
different
about
their
cluster
configuration
and
what
they
can
do
to
tune
it
or
like
what
special
cases
they're
hitting
so
that's
effectively.
What
I'm
trying
to
do
here
is
get
boners
for
gke
I'm,
also
trying
to
get
coners
for
all
of
the
GCE
tops
and
we'll
take
a
look
at
the
GC
I
choose
to
each
other.
That's
currently
those
for
this
place.
D
So
this
is
a
job
that
just
stands
up
a
cluster
in
Google
compute
engine
and
runs
a
bunch
of
end-to-end
tests.
You
can
see
on
the
left
here
that
the
end-to-end
test,
the
individual
test
cases,
are
owned
by
a
bunch
of
individual
states,
the
links
a
gap
zones.
This
job
should
not
run
to
completion
state
network
as
this
proxy
version.
Sync
instrumentation
as
this
test
about
cluster
level
block.
D
There
are
a
couple
tests
here
that
are
not
owned
by
any
individual
sick,
like
variable,
expansion
should
allow
substituting
values
or
after
that
means,
and
then
there's
also
no
state
listed
for
like
the
job
as
a
whole
like
if
the
entire
cluster
were
to
fail
to
come
up
for
some
reason,
or
some
particular
way
about
the
GCE
cloud
provider
was
something
about
it
was
causing
the
cluster
to
fail
like
no.
No
sig
has
historically
over
that.
This
also
seems
like
a
thing
that
I
would
expect
see.
D
D
D
So
basically,
if
I
see
that
the
job
is
failing
in
one
of
those
meditations
I'm
going
to
find
the
safe
that
owns
that
job.
If
it
turns
out
that
it
test
cases
failing
in
that
job,
I'm
not
going
to
bother
the
job
owner
if
I'm
going
to
bother
whoever
that's
the
test
case,
assuming
I
can
find
that
if
that
test
case
is
feeling
just
didn't
that
chop,
it
up,
I'm
gonna
chase
down
both
the
test
case
owner
and
a
job
owner
and
a
state.
You
please
figure
this
out.
D
So
when
I
talk
about
chasing
people
down,
what
do
I
mean
I
mean
I'm,
going
to
be
finding
issues
and
get
hooked
and
I'm
going
to
be
labeling
them
with
priority
critical,
urgent
and
priority
failing
tests.
So
here's
an
example
of
me
finding
a
job
and
notifying
CCPP
that
the
sorry
suit
cluster
lifecycle
in
this
case,
because
I
know
that
you
baby
em.
Here's
an
example
of
me
notifying
them
that
this
job
is
failing.
D
What's
going
on,
I
link
to
them
the
particular
release
that
this
is
blocking
I
link
them
to
test
period,
I,
try
and
give
as
much
information
about
hey
this
is
this
is
happening.
Is
anybody
working
on
this?
It
turns
out
actually
I
know
somebody
is
working
on
it.
I
find
it.
So
I
get
to
them
and
then
they're
free
to
close
the
issue
to
let
you
know
that's
something
I
no
longer
have
so.
The
hope
is
that
just
here's,
an
example
of
how
that
would
work
for
individual
test
case
as
well
right
so
I'm.
D
Signet
work
because
I
know
they
own
this
particular
test.
That's
failing
I'm,
following
in
a
Nietzsche
failure
and
I'm
trying
to
provide
information
about.
You
know
where
this
is
happening,
using
all
the
tools
that
our
disposal
to
show
that
this
is
failing,
a
constant
number
of
jobs,
etcetera,
etcetera,
so
I'm
I'm,
effectively
signing
sick
GCP
up
to
helped
me
triage
bugs
to
the
appropriate
people
who
know
how
to
fix
them.
D
If
it's,
if
it's
an
instance
of
a
test
case
failing
and
there's
someone
to
say
that
on
one
side,
I'll
probably
just
go
through
that
directly,
but
if
it
turns
out
it's
not
time,
I'm
gonna
work
with
the
same
GCP
to
figure
out
who
should
own
this
and
for
TKE
is
non-googlers,
really
don't
have
insight
or
visibility
into
what's
happening
within
the
balls
of
people
too
often
what
I
have
heard
it
historically
is:
oh
yeah,
we
know
that's
an
issue
and
there's
nothing.
I
can
link
to,
and
it
really
helps
build
trust
within
the
community.
D
If
there's
like
issues
that
show
that
the
problem
is
known
as
being
worked,
you
can
link
to
when
our
Tiaras
aren't
working.
We
can
say:
oh
my
hair's,
not
working
because
of
this
thing,
just
trying
to
find
ways
to
elevate
attention,
keep
top
screen
and
keep
everything
passing
and
shipping
on
a
more
frequent
basis.
Any
questions.
A
Got
a
few
responses
not
too
much
as
questions
sure
the
the
first
things
kind
of
an
observation,
I
think
about
a
lot
of
these
tasks.
Jobs
in
particular
I.
Very
much
accept
the
point
that
that
gke
stuff
is
a
bit
of
a
black
hole
and
it
kind
of
does
need
an
entry
point
and
I
can
kinda
I.
Think
we
need
to
do.
We
need
to
figure
that
out
as
a
whole,
and
so
that's
true.
A
The
other
observation
about
them,
though,
is
that
more
often
than
not
a
gke
test
suite
is
basically
a
clone
of
other
tests,
seen
simply
running
critique,
a
cluster
creation
path,
and
so,
if
the
boats
read
a
kind
of
particular
thing,
it's
probably
not
necessarily
sigh
I
guess
what
I'm
saying
is.
It
feels
a
lot
of
those
cases
that
failed
should
actually
go
to
the
the
group
that
that
cares
about
that
test,
suite
as
a
whole-
and
you
might
be
saying
I-
guess
for
some
of
these-
that
there
is
nobody
who
cares
about
the
test.
A
C
A
I
mean
I
mean
I
I
was
the
the
branch
manager
for
one
a
I
have
got
a
lot
of
empathy
for
digging
through
and
release
master
it
so
I
used
to
call
it
the
lucky
8-ball
I'm
kind
of
approach
to
analysis,
but
one
thing
that
crossed
my
mind
is
the
idea.
One
thing
I
could
easily
sign
up
for
even
in
the
current
state
of
absolute
TCP
is
something
like
a
g
c
e
and
a
g
ke
happy
path,
test,
job
which
doesn't
do
a
great
deal,
but
it
establishes
clusters,
maybe
pokes.
A
You
know
a
simple
hello
world
kind
of
a
job
through
and
validates
it
can
do
a
few
things
and
then
shuts
down.
I
could
imagine
a
test
job
like
that
running
them
as
soon
as
that
goes
red.
That's
clearly,
a
one
hundred
percent
like
a
DCP
issue,
doesn't
do
anything.
I
wouldn't
bring
clusters
own,
and
that
could
be
like
a
canary
for
some
of
these
other
wider
issues.
A
For
example,
the
gke
thing
that
was
the
load
image
program
or
whatever
that
variable
was
that
just
took
out
all
the
gke
jobs
for
the
past
week
or
two,
and
that
would
have
been
caught
by
this
kind
of
approach
quite
early
and
exactly
how
we
sign
up,
for
this
is
a
little
TBD
man.
You
can
see
that
we
don't
have
very
many
participants,
they're,
certainly
not
Colonel,
full-time
staff,
team,
undercity
sepia.
To
some
extent
the
number
of
humans
are
going
to
be
tasks
of
all.
A
We
have
Sigma
TCP
work
kind
of
depends
on
how
the
group
matures
and
what
kind
of
pieces
of
work
it
takes
on
and
what
kind
of
base
it
has
and
there's
some
hoping
to
get
to
develop
that
in
its
current
state.
It's
hard
to
it's
kind
of
me
and
that's
why
a
lot
of
these
I
told
you
from
the
overload
of
them
failing
test
jobs
in
the
past
handful
of
days.
That's
not
American.
We
need
to
rapidly
respond
to
Eva
and
I.
D
I
want
to
live
in
the
same
world
where
all
of
the
cloud
providers
are
out
of
tree
and
each
sort
of
each
provider
is
responsible
for
the
quality
of
their
code.
I
love
to
be
able
to
delegate
all
of
the
testimony
related
to
that
cloud
provider
to
the
sig
that
owns
that
cloud
provider
and
sorry.
I
would
say
that
the
GCE
cloud
provider
is
kind
of
the
default
of
the
de-facto
cloud
provider
for
most
of
the
tests.
D
You
run
that's
why
I'm
sending
everything
in
this
thing
I
completely
empathize,
with
the
fact
that
this
is
a
Mason
Singh
and
it's
under
staff.
What
I'm
asking
for
is?
Can
you
help
me
understand
who,
as
a
member
of
the
Lewis
team
right,
but
you
really
mean
our
lives
are
collected?
We
made
a
lot
better
if
the
right
tests
are
in
there,
which
is
kind
of
a
separate
orthogonal
issue,
but
that
the
tests
that
are
in
there
are
actively
watched
and
maintained,
and
it's
kind
of
too
much
to
expect
somebody
who's
filling.
D
The
CI
signal
role
to
also
know
exactly
who
to
co
can
do
to
go
chase
down
within
the
walls
of
companies
that
they
may
not
necessarily
work
at
so
like
using
the
cloud
provider
as
the
abstraction
for
the
individual
cloud
provider.
Problems
is
one
and
then
assigning
a
say
to
each
of
those
platform.
Editors
is
one
way
of
doing
it.
A
But
I
I
kind
of
I
think
we
can
grow
toward
what
you're
talking
about
and
I
kind
of,
like
us,
I
can
imagine
like
the
TKE
level
stuff
that
the
gke
variant
of
a
test
suite
is
failing
was
the
GCE
one
is
happy.
It
looks
like
it's
obviously
will
gk
e
IV
not
appropriate
to
kind
of
win
sig
tcp,
I
gaza,
I've
guys.
You
were
doing
recently
and
I'll
start
to
try
and
make
time
to
look
at
that
and
relay
it
and
try
and
get
some
action
on
that
I.
A
Don't
look
I
consist
feel
quite
the
same
about
what
we
call
the
GCE
test
jobs
just
because
almost
everything
is
a
TC
test
role
and
so,
like
I
I,
don't
it's
that's.
Why
I
die
I'd
rather
be
on
the
hook
for
a
GCE
canary
that
the
validates
of
clusters
can
come
up
and
down
and
there's
nothing
systemic,
going
wrong
right.
D
Well,
this
is
why
I
was
trying
to
articulate
bet
most
often
on
going
after
the
individual
test
cases.
It
won't
agree
as
a
whole
is
failing
or
I
can't
identify
who
this
test
case
belongs
to,
but
I'm
going
to
come
and
talk
to
you,
but
it's
like
the
gke
stuff,
since
its
GK
specific.
That's
really
where
I
kind
of
I
can't
talk
about
that.
That.
A
D
D
What
I'm,
also
happy
to
do
is
to
see
if
I
can
find
other
venues
to
push
forward
hey.
This
is
the
new
world
that
I'm
trying
to
work
within,
as
the
CI
signal
lead
to
get
additional
people
and
Eagle
to
come
help
you
out
if
I
can
say
that
this
is
the
authoritative
touch
point.
Can
you
please
send
more
resources
to
the
sig
so
that
you
can't
actually
be
responses
like
I'll?
Do
what
I
can
in
there
cool.
A
Appreciate
up
okay,
so
the
the
one
thing
that
says
right
now
do
feel
free
to
ping
me
directly
to
kind
of
like
to
do
stuff
because
there's
a
there's,
a
flood
of
issues
coming
through
and
I
Conicelli.
If
there's
something
urgent
like
I
missed
the
gke
once
it
was
buried
with
eight
others,
but
really
I
should
have
jumped
on
that
because
I'm
that
was
like
a
right
systemic
thing.
Anything.
B
D
A
Thanks
for
the
work
you
do
trying
to
make
tests
actually
pass,
it's
it's
a
very
non-trivial
thing
and
we've
got
10
minutes
before
the
community
meeting
starts
and
I
want
to
keep
a
whole
like
a
hard
deadline
for
that,
so
that
we
don't
overlap
them.
But
before
we
go,
does
anybody
else
on
the
call
have
anything
else
they
want
to
add
or
ask
or
suggest
or
request
I.
E
A
question
oh
sure,
yeah.
My
name
is
Alberto
I'm,
currently
working
on
running
a
self-hosted
Q&A,
the
song
on
DCP
qu
M,
so
sometimes
I
find
issues
not
with
Q
Nettie's,
but
with
DCPS
specific
issues
with
Google
Cloud.
So
the
question
basically
is:
is
this
the
right
place
to
talk
about
those
issues
as
well
or
or
maybe
not.
A
Potentially
I
mean
I
kept
saying
earlier:
I
I
I
fear
that
if
all
we
do
is
talk
about
problems,
mgcp
normally
people
will
start.
You
know
it
is
not
very,
very
engaging
thing.
I'd
like
to
also
talk
about
what
people
are
doing
and
kind
of
like
benefits
and
stuff,
but
certainly
and
as
a
mailing
list
you
should.
You
should
feel
free
to
kind
of,
like
your
suggestions,
see
bring
up
and
routed.
I
could
only
be
kind
of
like
yeah
win,
sick
TCP
and
we
won't
answer.
A
A
A
A
At
e1
and
natural
face-to-face
meeting
we'll
see
how
that
goes,
but
certainly
for
now
it's
every
two
weeks
and
next
week,
I'm
hoping
that
I
can
get
just
in
Santa
Barbara
to
come
and
talk
about,
cops
and
he's
been
rewriting
parts
of
cops
said
it
works
not
just
for
AWS,
but
also
for
TCP,
which
is
potentially
quite
an
interesting
thing
to
do,
and
so,
if
we
can
get
him
to
come
and
talk
about
that,
that
would
be
great.
If
there
are
any
things
you
want
to
hear
about,
and
please
either
contact
me
directly.