►
From YouTube: Kubernetes kops office hours 20191011
Description
Recording of the kops office hours meeting held on 20191011
A
Alright,
hello,
everyone
and
thank
you
for
joining
us
for
cops
office
hours
this
week.
My
name
is
Mike
and
I
will
be
your
moderator
just
as
a
reminder.
This
is
being
recorded.
This
is
open
to
the
public
and
will
be
posted
on
YouTube,
hopefully
later
today
or
soon-
and
you
know
we
abide
by
the
community
policies
and
best
practices
and
community
meeting
guidelines,
so
please
be
respectful
to
each
other,
and
with
that
we
will
kick
off
the
meeting
so
the
way
we
generally
take
this
is
we
have
our
agenda.
A
A
A
We
got
to
look
good
on
the
kubernetes
slack
or
yoku
Bernays
reddit,
so
they
they
thought
it
was
related
to
this
other
kubernetes
issue,
which
I
mean
it
almost
seems
like
this
kubernetes
issue
or
cops
issue
is
a
become
somewhat
of
a
catch-all
of
a
cluster.
You
know,
no,
that's
not
joining
a
cluster
there's
so
many
reasons
that
that
can
be
so
I
was
trying
to
help
them
figure
out.
What
was
happening,
and
the
best
thing
I
can
figure
is
that
someone
at
their
company
use
the
new
version
of
cops.
A
It
didn't
realize
it
and
you
know
kind
of
a
collision
there
of
cops
versions,
kubernetes
versions
and
someone
having
updated
something
but
no
really
got
to
the
bottom
of
it.
I
don't
know
if
anyone
has
any
comments.
I
just
wanted
to
throw
it
in
there.
In
case
the
doubt
to
anyone
that
it
could
have
been
related
to
the
11110
enforcement,
as
Justin
pointed
out
on
there,
I
don't
think
it
is
as
well
but
I,
just
kind
of
you
know
and
as
a
food
for
thought
thing.
I
don't
know.
If
anyone
has
any
comments.
B
A
Do
have
an
issue
for
that.
I
know
that
a
few
people
have
mentioned
that
to
me
as
before
as
well
of,
like
you
know
at
least.
Obviously
we
could
let
them
go
backwards,
but
you
know
kind
of
first
path.
You
know
I
stopped
like
hey
by
the
way
this
cluster
has
already
been
applied.
You
know,
let's
say
114
and
you're,
trying
to
run
112
bad
things
might
happen,
and
then
you
know
give
them
a
flag
to
work
around
it,
but
I
personally
have
never
gotten
time
to
implement
it.
A
C
C
A
At
first,
when
I
saw
it,
I
just
was
like
wait,
do
cops
111
in
110,
just
break
so
I
went
downloading
your
old
version,
110
or
111
cluster
or
whatever,
and
everything
still
worked
fine.
So
it's
like
at
least
it's
not
like
some
sort
of
crazy
progression,
and
you
know
images
going
missing
something
yeah.
C
D
I
was
looking
into
some
protests
failing
and
I
noticed
that
we
don't
have
logs
for
the
new
cops
controller
in
the
crowd,
job,
artifacts
and
I'm
wondering
you
know.
Currently,
our
cube
sister
art
static,
pods
that
we
create
they.
Their
command
is
a
tea
that
appends
to
a
bar
log.
You
know
a
log
file
in
bar
log
on
the
host
using
a
post
path,
amount
I'm,
wondering
if
we
should
do
same.
C
C
In
theory
like
it,
would
be
good
to
get
more
of
these
pod
logs
in
anyway,
but
the
I
guess
that,
like
these
are
actually
like
really
special
cases,
and
we
should
keep
them
on
the
house
path
like
if,
basically,
as
we
add
more
stuff
into
that
controller,
it
will
become
it's
gonna.
Imagine
it'll
become
more
of
an
equivalent
/
cute,
but
just
sort
of
a
more
an
easier
way
to
run
part
of
pretty
cute.
Oh
yeah.
Do
you
want
to
join.
C
D
D
D
Yeah
so
more
and
then
testing
stuff
in
prowl,
just
seeing
how
we
create
clusters
for
and
end
test,
we're
mostly
doing
it
using
kind
of
assembling
the
command
line,
flags
that
we
pass
to
Cox,
create
cluster,
and
some
of
the
more
advanced
functionality
in
cops
doesn't
support
flags.
You
need
to
provide
a
you
know,
a
manifest
cluster
manifest
so
I'm
wondering
if
there
would
be
a
way
for
us
to
provide
cluster
manifests
to
the
ETD
jobs.
D
So
one
of
them
would
be
like
the
mixed
instance
groups.
So
where
did
the
ASG
use
of
multiple
instance
types?
That's
something
all
right.
I
know
we
have
the.
What
is
it
like
the
set
flag
that
lets
you
specify
arbitrary
fields
in
the
manifests,
but
they
need
to
be
explicitly
supported
and
cops,
so
just
wondering
if
there'd
be
an
easier
way
to
to
test
I.
C
C
Make
that
so
that,
like
it
was
more
I,
think
we've
established
a
grammar
for
that
so
to
make
it
not
have
to
be
hard-coded,
but
it
is
also
super
tempting
to
just
keep
adding
them
and
they
will
eventually
become
unit
tests.
So
if
you
want
to
just
add
another
one,
I
think:
that's
that's
absolutely
fine,
I
think
yeah!
Don't
if
you
invent
any
new
syntax,
then
we
have
to
look
at
it
more
closely.
C
D
C
I,
just
the
I
don't
know
other
way
and
proud
to
specify
a
final,
so
I
don't
see
how
we
would
do
any.
We
can
specify
flags,
but
as
far
as
I
know,
we
can't
specify
files
so
I
mean
we
could
always
like
put
them
in
a
bucket
something
I
her,
but
that
feels
even
like
feels
like
they
will
not
be
happy
about
that.
A
Alright
and
to
our
final
item
that
we
should
probably
go
over
the
road
map,
should
we
Justin
do
you
think
we
should
review
the
releases
and
the
release
before
jumping
to
the
road
map?
Sure,
okay,
cool
well
as
I
imagine
most
of
the
five
of
us
on
this
call
are
aware,
cops,
1:14
shift.
Thank
you,
everyone
for
your
hard
work
and
all
the
PRS
and
the
cherry
picks
and
Justin
specifically
for
doing
all
the
you
know,
release
cutting
and
everything
and
Peter
had
a
ton
of
fantastic.
A
You
know
commits
and
pr's
and
all
everyone's
helped
guy.
You
had
a
bunch
as
well
and
Rodrigo.
You
know
you
are
always
around
kicking
butt,
doing
a
lot
of
great
work
for
us.
So
thank
you,
everyone
for
all
your
hard
work
on
that
as
well.
In
the
next
couple
weeks,
we
have
cops,
115
beta1,
hopefully
cut
116
alpha-1
with
a
few
PRS
that
are
holding
it
up.
A
It
looks
like
and
then
promoting
some
am
eyes
from
alpha
to
stable
and
promoting
conversions
from
alpha
to
stable
the
one
I
guess,
since
we
last
chatted
was
the
eld
bug
in
cops
or
crew.
Today's
one
14-7
that
GOP
support
basically
breaks
on
server
C
load
balancers,
so
as
a
kind
of
a
heads-up
to
everyone.
That's
encourage
people
not
to
use
one
14-7
at
the
moment.
Wait
for
114
eight
114
six
is
perfectly
fine,
which
is
what
we're
recommending
users
use.
C
Out
I,
don't
think
so.
Yeah
I
think
they're
I
think
the
plan
is
to
cut
it
today.
Okay,
because
we
do
have
a
mechanism
and
I
haven't
actually
seen
this
one
I,
don't
know
if
we
have
a
mechanism
to
force
or
encourage
people
onto
a
newer
version,
I
don't
think
we
have
a
mechanism
to
skip
one
particular
minor
version.
A
Well,
yeah
and
a
bunch
of
people
were
pushing
to
get
a
new
version
cut
and
so
I
try
to
jump
in
and
say
we
know
it's
hard
to
cut
a
new
version
of
kubernetes.
So
why
don't
we
just
get
a
release?
Note
saying
that
this
is
a
known
issue
and
you
know
I'm
solve
it
socially.
We
can
kind
of
avoid
using
it
and
then
114
today.
A
A
You
know
let
it
bacon
in
beta
for
a
bit,
but
you
know,
since
since
115
is
not
not
many
significant
changes
and
116
has
quite
a
few,
it
might
be
beneficial
for
us
to
get
115
going
sooner
and
then
we
can
put
a
ton
of
effort
into
116
and
let
that
one
really
good
good
good.
So
at
least
that's
my
thoughts
at
the
moment.
If
anyone
else
has
any
thoughts.
C
That
sounds
great
and
we
can
also
we
can
also
cut
116
and
do
things
for
117
that'd
be
great.
We
just
got
to
get
the
cops
controller
working
across
backward
right.
Yes,
so
actually
I
think
this
one
there's
a
debate
about
the
notes
which
I
think
is
important.
I
think
I
think
the
status
is
without
cops
controller
support.
C
The
masters
are
still
labeled
with
the
correct
node
role,
because
that's
needed
for
bootstrap,
but
the
nodes
will
not
be
so
it's
not
perfect,
but
it
is
not
terrible.
Like
most
people
are
probably
not
tolerated,
not
deliberately
steering
workloads
to
node
roles,
but
I
do
think
that
no
notes,
if
they
are
steering
they're
steering
to
masters
I,
would
imagine,
but
I
do
think.
We
also
have
broken
I,
do
think
it
breaks
any
custom
labels,
custom,
kubernetes
labels
on
instance,
groups,
so
that
is
probably
the
bigger
one.
It
does
break.
C
Justin,
do
you
want
to
take
that?
Yes,
thank
you.
So
the
sick
coaster
lifecycle
is
going
to
present
a
road
map
on
a
lisa
cluster
lifecycle
sub-projects
at
coop,
con
North
America
in
November
next
month,
Wow
and
the
really
should
we
should
basically
figure
out
what
we,
what
our
road
map
should
look
like
or
what
we
want
to
put
on.
Our
road
map.
I
have
personally
added
the
items
which
I
think
I'm
going
to
try
to
work
on
over
the
next
well
foreseeable.
C
It
imminent
future,
and
if
people
want
to
add
other
items
or
tell
me,
the
other
items
are
important
or
other
things
are
unimportant
or
anything
like
that.
Then
please
feel
free
to
do
so,
but
I
am
in
in
the
117
timeframe,
which
is
like
the
next
two
to
three
months.
I
would
like
to
get
add-on
operators
or
I
first
add
an
operator
going
at
least
I'm
going
to
move
this
down,
but
some
more
work
on
secure
by
default.
C
Looking
at
no
bootstrap
tokens
like
basically
taking
the
great
work
done
on
the
AWS
note,
authorizer
and
extending
it
to
other
clouds,
I'm
able
to
do
GCE,
but
I
will
probably
need
to
help
on
OpenStack,
but
it
shouldn't
this
shouldn't.
Be
that
different
from
what
we
have
done
to
date,
which
is
what
we've
already
have
to
do.
It
should
be
a
very
similar
flow.
Basically,
you
have
an
IP
and
you
sort
of,
or
at
least
an
IP,
maybe
some
other
stuff.
C
I
would
look
at
like.
Can
we
use
it
for
masters
as
well,
and
can
we
actually
like
bring
up
a
whole
cluster
that
way
I?
Can
we
go
100
percent
server-side
or
support
a
mode
where
we
run
a
hundred
percent,
server-side
and
yeah
I
think
also
container
D
I'd
like
to
try
to
get
that
in
because
yes,
docker
and
maybe
some
of
these
darker
things
look.
C
A
Need
to
find
my
list
I
have
a
list
of
stuff
that
I
was
gonna
work
on
in
the
next
few
months.
The
one
I'm
really
hoping
to
close
out
is
the
documentation
stuff
that
I
started
the
website,
so
I
I,
don't
necessarily
have
a
timeline
for
that,
but
I'm
going
to
try
to
get
that
merge
soon
and
then
start.
You
know
getting
a
public
site
up
for
us
for
that.
A
A
E
I
mean
don't
carpet,
I
make
I'm
excited
for
the
container
d
yeah
one
one
thing:
I'm
only
gonna
get
her
into
finding
some
time
to
actually
getting
do
marker
and
capacity
optimized
and
get
not
done
because
for
us
we
want
to
get
that
in
before
Black
Friday,
for
obvious
reasons,
and
to
try
and
improve
their
resilience
ape
spot
and
should
should
have
something
time.
Nice
office
area
schematic
shouldn't,
be
a
difficult
battle.