►
From YouTube: Kubernetes kops office hours 20200522
Description
Recording of the kops office hours meeting held on 20200522
A
B
D
I
think
they
should
be
done
at
the
same
time
anyway.
116
should
be
done
because
of
the
OpenStack
issue
that
well
it's
not
working
in
116
think
we
discussed
it
last
time.
So
only
because
of
that
which
is
the
biggest
issue
but
I
agree
with
John
1:17
is
weighted
by
pretty
much
everyone.
So
if
we
can
find
find
the
solution
to
move
it
forward
would
be
very
nice.
I.
E
We
should
prioritize
117,
but
it's
not
that
hard
when
I'm
doing
one
release
to
do
another,
so
we
can
talk
about
how
ready
we
have
ready
our
to
do
117
later,
but
I
think
we
need
the
116
release
anyway
and
I
think
we
have
another
topic
that
suggests
I
will
be
doing
other
releases.
They
might
not
be
done
zeros,
but
they
will
be
other
releases,
so
we
can
certainly
do
116
3.
At
the
same
time,
Christian
I've
spotted
a
flaw
in
your
iPad
hosting,
which
is
when
you
switch
to
the
agenda.
A
A
D
So
that's
me:
some
weeks
ago,
Justin
added
agreed
they're
me
and
Peter
well
helped
a
bit
very
little
and
I
try
to
help
with
this
so
big
to
get
the
grid
itself.
You
know
to
organize
the
bit.
We
discussed
that
we
kind
of
need
our
own
grid
group
right
to
have
the
tabs.
So
this
is
my
PR
trying
to
move
that
forward
and
I.
Think
Peter
took
a
look
so
right
now
it's
not
a
big
change
from
what
we
have,
but
basically
it
makes
us
owners
on
that
grid
group.
E
E
D
E
E
I
know
we
we
had
hoped
to
not
have
too
many
branches
open,
I,
don't
know
whether
this
will
actually
be
just
the
way
it
is
that
actually
like.
We
do
actually
need
a
bunch
of
branches
open
at
all
times.
Just
for
me
like
they
are
all
stabilizing
at
different
rates,
point
of
view,
but
I
guess
we
will
figure
that
out
yeah
if
we
don't,
if
we
don't
cut
the
branch
I
don.
Yes,
ever
seen
19,
so
it's
we
need
to
cut
the
bridge.
Yeah.
C
A
A
D
E
E
Mean
we
should
we
should
try
Trevor
you
that
what
if
you
try
a
please
do
feel
free
to
try
using
the
command
if
it's
appropriate
and
we
can
try
to
forget
the
permissions
at
that
point
like
we
can
ask
for
white.
It
does
not
work
for
this
person
type
thing,
but
yes,
anyway,
that's
a
sidebar
I've.
We
have
multiple
of
us
to
put
it
into
one.
E
I,
don't
I
have
one
point
against
at
one
point
for
the
point
against
is
I,
don't
think
spot
instances
is
working
with
it
last
time,
I
checked
and
I.
Don't
know
if
I
want
to
contradict
me
on
that
the
0.4
in
favor
it
is
that
it
does
look
like
Amazon
is
certainly
going
in
this
direction
and,
in
particular
the
instance
metadata
service.
C
D
D
D
C
C
F
So
we
we
use
it
with
next
instance,
policies
and
spot
instances
how
you
do
yeah.
We
do
that's.
F
G
A
Confusing
and
trying
to
figure
that
out
was
like
me
talking
to
Amazon
support
for
two
hours
on
the
phone
trying
to
get
technical
support,
get
it
working
it
the
way,
I
wanted
it
to
be
so
lunch
templates.
Is
you
really
need
to
know
what
you
need
to
configure?
It's
definitely
not
cops,
but
if
you
know
which
flags
to
set
that
it
eventually
works.
G
D
D
Also,
there
are
three
things
in
there:
its
launch
configuration
that
we
use
now
by
default,
its
launch
templates,
which
we
want
to
use
in
the
future
and
mixed
instance,
policy,
which
is
some
kind
of
launch
templates.
So
it's
a
bit
different.
It's
still
launch
templates,
but
in
order
to
use
mixed
instance
policies
you
have
to
use
launch
templates
and
everyone
that
used
it
until
now
got
it.
E
We
can
continue
once
it's
I,
don't
believe
anyone's
proposing,
typically
put
it
into
117,
so
it
doesn't
need
to
block
117
either,
and
we
can
like
like
make
progress
on
all
the
other
fronts
and
make
a
decision
that
will
hopefully
be
a
fairly
easy
to
a
jury,
pick
that
we
can
like
either
cherry-pick
a
turn
it
on
type
thing
or
not.
Do
that
and
yeah
the
docks
for
how
to
turn
on
spot
price
would
certainly
help
me.
D
C
So
I'm
refactoring,
some
of
the
circuit
issuance
and
I,
would
like
to
get
some
of
the
are.
There
are
currently
two
existing
certs
that
are
issued
by
node
up
that
have
one
year
lifetime,
so
I,
don't
think
currently
masters
that
are
greater
than
one
year
with
work,
and
I
would
like
to
extend
that
such
that
no
node,
that
is
older
than
one
year
would
necessarily
work
and
see.
If
that
is
fine
with
the
you
know,
whether
that
is
a
requirement
that
I'm
breaking.
A
E
And
I
believe
is
another
data
point
that
the
I
don't
know
it's
happened.
Yes,
but
it
looks
like
it
seems
pretty
clear.
There
will
be
a
one-year
support
policy
for
kubernetes
versions
as
part
of
the
LTS
work
stream
they're,
making
that
a
year
so
I
don't
think
that
means
it
will
be
I
kind
of
imagine.
There
will
be
too
many
notes
that
don't
need
any
updates
for
a
whole
year,
but
it's
not
inconceivable
that
such
a
thing
could
happen.
Maybe.
E
C
E
B
C
E
Also
integration
there
with
the
no
problem
detector
like
if
we
can
treat
this
as
another,
we
could
treat
it
as
a
no
problem,
whether
or
not
it's
integrated
in
their
project
or
not
like
the
same
exposure
mechanism
and
then
some
policy
to
auto
restart
nodes
that
have
been
problematic
for
some
period
of
time
or
that
are
more
than
n
days
old.
Where
n
can
be
your
personal
preference
like
I,
know,
I
know
a
lot
of
people
like
to
it
seems
like
a
good
idea
to
do
that
from
a
security
standpoint
anyway.
E
I
just
worry
that
the
I
think
it's
a
really.
If
we
can
get
all
of
that
in
I,
think
we
should.
We
should
do
that
and
I
think
that'd
be
great
I.
Think
in
the
until
we
have
that
I
worry.
If
we
I
think
we
should
maybe
make
it
a
little
bit
longer
than
a
year
just
because
of
the
LTS
being
a
year
as
well,
but
otherwise
I
think
if
you
were
to
say
15
months
or
18
months,
I
think
that
that
sounds
very
reasonable.
Okay,
because.
E
A
I
think
trying
to
stick
to
basic
kubernetes
as
much
as
possible
is
usually
a
good
idea,
because
people
will
get
confused
and
I
think
it's
more
important
like,
for
example,
D
no
problem
detector
to
do
it.
That
way,
simply
because
of
interest
visibility.
So
for
me,
as
an
administrator
and
I,
just
don't
worry
about
process
for
a
year,
which
is
certainly
speaking
for
itself
and
then
all
of
a
sudden
they
just
disappear
and
come
back
up.
A
E
So
I
mean
I
love
the
initiative.
I
love
the
like
the
skew
is
brilliant,
like
that's
really
good
I.
Think
like
we
have
a
roadmap,
we're
not
gonna
I,
don't
know
mess
you
pulla,
unless
you
call
Oh
John
John,
yeah
you're,
not
gonna,
like
we're,
not
gonna.
Have
it
like
tomorrow
so
like
in
the
absence
of
all
of
that
like
how
about
15
months
is
that
okay,
15
months
with
yours
figure,
that
seems
pretty
cool.
A
E
G
C
G
G
E
A
C
Yes,
so
what
am
I
rolling
update
tests?
Flaked?
It's
got
a
section
where
there's
a
race
where
the
test
has
to
win
the
thing
is
that
happens
several
times
throughout,
so
let
me
pull
up
my
ticket,
so
basically
we
can
add
more
delay
into
the
test
to
make
it
more
likely.
It
win
the
race.
But
the
question
is:
how
much
time
do
we
want
to
grant
to
the
rolling
update
or
part
of
the
integration
test?
C
E
Per
test-
and
that
seems
fine
right,
I
I,
the
the
big
win
and
a
lot
of
this
is,
if
I
recall
correctly,
the
one
of
the
slowest
things
in
our
you
know.
Tests
is
key
generation,
we
generate
a
lot
of
keys,
I
feel
like
that,
would
actually
get
us
the
more
than
the
10
seconds
back.
Have
we
ever
decided
it
was
or
that
whatever
is
going
to
be
so
I
I,
don't
see
any
reason
to
let's
do
the
pragmatic
it
here?
No,
that's
my
opinion
underfunding.
One
else
cares
about
this.
E
C
So
so,
basically,
this
is
a
proposal
are
I.
Rent
tricks
on
the
number
of
people
are
number
of
approvals
that
came
from
each
approver
and
number
of
lgd
ends
came
out
of
each
member
and
have
a
PR
90
105,
which
throws
a
fair
number
of
the
approvers
into
the
reviewers
list,
promotes
puts
myself
and
Ciprian
into
reviewers
and
take
some
of
the
less
active
approvers
into
America's
status.
A
E
Also,
like
that,
it's
very
data-driven
so
I
think
I
think
this
is
the
right
thing
to
do
to
reflect
the
reality
on
the
ground
and
if
anyone
objects
that
everyone
is
welcome
to
direct
here.
But
if
anyone
objects
like
at
a
later
date,
we
can
always
like
consider
them
as
a
special
case
in
in
any
direction.
I.
C
A
Yeah,
the
thing
is
what
I'm
thinking
about
for
me,
especially
I,
got
a
little
bit
lost
because
I
missed
the
emails
that
I
got
when
it
was
the
reviewer,
so
I
would
have
reviewed
many
more
if
I
would
have
received
email
said
there
is
something
to
review
which
drove
me
off
a
little
bit
so
I'm
thinking
of
potentially
I
find
the
time
to
actually
review
stuff
again,
if
I
get
out
of
vacations.
If
there
is
something
because
if
it's
not
a
black
box
anymore,
so
I
don't
have
to
be
approval
for
that.
A
A
A
I
mean
you
can
merge
that
already
I
will
just
create
another
one.
To
put
me
back
to
reduce
I,
wouldn't
really
mess
with
that,
because,
if
I
simply
can't
edit
pull
request
because
I'm
not
like
administrator
on
the
project
and
all
of
that
stuff,
it's
too
complicated.
So
just
noted,
like
that,
and
if
anyone's
interested
in
going
back
just
creating
pull
requests
and
we
can
like
fast-track
that
thing.
Yeah.
F
C
A
That
sends
also
for
reviewers.
I
would
encourage
them
if
you
find
something
that
you
know
this,
and
that
review
approve
as
potentially
knowledgeable
in
that
sense,
do
assign
them
to
the
issue
as
well.
So
I
think
you
can
really
make
it
visible
to
them
a
little
more,
that
there
are
something
that
could
be
proving
for.
H
H
B
A
H
I
brought
this
up,
I,
don't
know
if,
like
I
month
ago,
the
last
time
I
was
actually
on
a
call
and
I
think
John
you
had
some
I
think
you
had
some
other
things
you
were
looking
at
in
this
area
and
I
think
we
decided
not
to
emerge
it
at
that
time
and
I
just
wanted
to
see.
If
you
were
good,
adding
that
in
at
this
point,
if
you
had
more
concerns,
haven't
I
already
merged
something
like
that.
You
may
have
I
literally.
A
A
E
You,
yes,
just
this
PR
went
in,
which
is
I,
think
a
good
idea.
It's
basically
surfaces,
something
one
of
the
ec2
instance
lifecycle
attributes
on
the
nodes
which
is
nice.
However,
it
reused
the
prefix
node
role,
kubernetes
Leo
and
I
I
worry
that
that
is
a
potential
collision
and
like
node
role,
has
its
own
complicated
history
and
I.
Don't
know
whether
it
would
be
easier
to
avoid
reusing,
node
role
and
I.
C
E
G
D
Shortly
we
can
say
that
it's
a
contact
that
it's
used
for
this
or
termination
I
think
it's
a
popular
thing.
We
used
it
in
the
past
to
notify
us
on
slack
when
a
note
went
down,
for
example,
so,
okay,
thank
you
context.
I
will
read
more
and
I
understand.
Anyway,
it's
just
a
label
or
something
so
we
can
change
it
at
any
time.
We
don't
use
it
in
cups
or
anything.
E
E
But
there's
not
even
there's
not
even
yes,
there's
not
even
a
flag
that
someone
could
set
if
they
want
Sternoff.
So
that's
it.
That's.
That
was
just
my
concern.
That's
all
thank
you
for
the
context
and
I
will
I
will
look
at
that
a
little
bit,
more,
maybe
I'll
add
a
flag
or
something
always
another
flag.
E
It's
not
been
top
of
my
mind,
I
must
say
in
the
past
couple
of
weeks
we
are
getting
pieces
in
line
to
make
this
happen.
I
don't
think
the
pieces
are
there.
Yet
I
am
hoping
we
can
build.
Maybe
I
will
try
to
build
the
1:18
using
cloud
build
or
some
other
repeatable
process.
I
think
the
changes
there
were
a
couple
changes
I
got
enough
months
ago.
E
I
think
that
let's
go
now,
which
I
think
make
that
repeatable,
and
so
hopefully
that
will
be
so
make
up
reproducible,
and
that
is
a
good
step
towards
it,
the
the
idea
being.
If
we
have
I
guess
we
bristles
in
second
required,
but
the
this
the
the
flow
will
be.
We
build,
builds
continuously
using
CI,
which
I
think
is
in
place
today.
I
think
we
have
the
ability
now
to
use
those
CI
builds,
as
is
there
will
be
a.
E
We
will
create
a
PR
to
copy
some
artifacts
from
the
staging
repo
to
a
production
bucket,
which
is
our
release
bucket,
and
at
that
point
those
artifacts
will
be
published
and
the
idea
is,
there
will
be
a
small
set
of
people
that
are
able
to
approve
the
PR
and
that
PR
will
effectively
be.
The
release.
Are
these
so
that
on
the
binary
artifacts,
so
there's
a
lot
of
pieces
there?
E
I,
don't
I,
haven't
yet
seen
any
other
projects
to
my
knowledge
that
have
it
done
so,
there's
not
someone
we
just
totally
copy
and
there's
definitely
other
projects
yet
doing
binary,
artifacts
ie
the
cops,
binary
and
I.
Don't
think.
There's
anyone
else
uploading
to
github,
but
those
are
all
things
that
we
can
try
to
make
progress
on
I.
A
Yes,
they
have
the
ideas
you
have.
Their
idea
is
to
use
Python.
If
it's
not
the
Python
Google
is
using
what's
a
specific
word
or
the
the
way
Google
work
writes
python,
essentially
at
everyone
else
worked
at
Google
agreed,
but
other
than
that
it
was
a
plan
like
a
few
months
ago.
I
think
it
was
C.
116
releases
so
that
people
talked
about
that
and
I
don't
know
if
there's
been
a
lot
of
progress
in
that
regard,
but
it
wasn't
initiative-
and
there
is
a
like
cover
issue
for
that
as
well.
I
mean.
E
A
A
H
A
So
I
think
160.
Maybe
we
have
already
talked
about
potentially
giving
some
releases
in
the
prior
ones
of
the
agenda.
So
that's
pretty
discussed
blockers
yeah
blockers
for
117.
We
have
something
about
IP
table
alternatives.
We
have
something
about
buster
a.m.
eyes
which
potentially
could
be
the
default,
and
so
there
is
discussion
about
the
base
image.
D
E
E
E
E
I'm
not
sure
good
point,
one
of
the
one
of
the
other
things
I
feel
like
is
happening
is
like
a
move
to
lighter
and
lighter
weight
and
I
feel
like
that's.
The
advantage
of
maintaining
our
image
and
I
feel
like
a
lot
of
the
and
I
want
pushback
here,
a
lot
of
the
like.
We
have
this
project,
which
is
now
spinning
up,
which
is
building
images.
I.
Imagine
that
will
produce
a
lot
of.
D
Okay,
but
then
at
least
we
should
make
them
generic
or
something
because
we
don't
release
images
for,
let's
say
112
the
advantage
of
using
the
let's
say:
official
images
is
that
they
work
for
any
release.
So
if
Debian
release
is
a
new
pastor
with
updates,
we
can
just
use
it
in
any
release.
If
you're
doing
it
in
your
will
in
the
custom
pipeline,
then
we
don't
do
it
for
those
we
do
it
only
for
the
last
two
three
releases
I
think.
E
We
will
do
it
or
everything
I
think
we
did.
We
go
back
further
than
that,
but
yes,
I
I,
accept
that
point
and
I
think
also
like
they
are
images.
Sorry,
the
newer
images
should
work
for
older
ones,
but
I
know
we've
had
problems
with
like
uninstallation
or
an
installation
of
different
versions
of
docker
and
container
tea.
For
example,
yeah.
D
E
Yes,
look
at
this
in
the
version
that
we're
running
is
the
same
as
Google's
cause,
so
that
gke
team
will
be.
We
can
leverage
some
of
the
work,
the
GTA
eternal
team
does
there,
which
I
thought
was
maybe
an
advantage,
but
yes,.
F
E
D
For
sure
at
least
the
Amazon
Linux
guys
advanced
kernels,
but
and
they
get
the
newer
patches,
they
tried
to
back
port
as
much
as
possible,
but
at
some
point
they
just
move
on
Debian
for
sure
won't
do
that.
Also
in
regards
to
Buster,
there
were
a
ton
of
bugs
related
to
IP
tables
that
they
I,
don't
think
they
ever
fixed.
So
I
think
they
were
approached
in
various
ways
to
fix
I,
think
of
three
bugs
and
they
never
did
for
all
three
of
those
kubernetes
had
to
do
work
around
so.
E
D
E
E
E
D
My
lips
maybe
make
up
here
with
I,
think
Peter
did
something
similar
were
an
issue
not
PR
and
say:
let's
like
rocket
is
for
bastard
and
I
know
smiley
faces
for
Ubuntu,
and
anyone
can
add
some
opinion
and
I
would
say
that
we
should
do
this
for
118.
There
is
still
enough
time,
and
if
we
have
the
process
to
build
things,
we
know
that
both
buster
and
Ubuntu
are
pretty
stable
because
they
ran
for
well.
We
ran
it
in
production
and
also
in
test
grid.
I.
Think
pretty
much
everyone
here
used.
E
A
E
D
E
That
one's
really
good
as
well,
because
it's
not
just
it's
not
just
the
minute
we
shared
or
the
minute
we
save
directly,
but
also
like
the
back
off
of
API
servers
like
we
can
stop
API
started
going
into
back
off.
That
is
a
win
yeah
I'm
also
I've
also
realized
that
we
had.
We
used
to
have
a
similar
strategy
in
proto
cube
where
it's
a.
D
E
E
So
it
should
be,
and
it
should
be
much
better
and
yeah
I
think,
like
sometimes
you're
like
there's
a
long
tail
where
sometimes
like
bring
up
takes
much
longer
than
you
expect.
I
think
that's!
This
sort
of
thing
is
a
big
contributor
to
it
like
when
winning
yes
ever
goes
into
backup,
so
this
will
make
it
much
better.
E
E
I
will
talk
to
one
of
the
things
we
can
do
is
there
is
this
at
CD
a
DM
project,
I
wonder
if
we
can
do
the
same
thing,
we
did
with
the
image
mode
where
you
move
the
code
more
or
less
wholesale,
underneath
that
and
that
will
help
with
our
unification
of
the
code
and
will
also
mean
we
get
the
machinery
for
free
I
will
talk
about
to
be
at
CTA
via
maintains
about
that
which
and
I
suspect
they
might
be
amenable.
But
I
will
I
won't
speak
on
their
behalf.
C
E
A
E
A
D
A
G
A
Sounds
like
yes
decision
to
remove
it,
so
in
that
sense,
117
is
done
and
we
already
covered
pieces
of
118
as
well.
I
think
we
also
talked
about
removing
support
for
chorus
and
Jessi
in
that
sense,
but
any
final
words
for
118.
In
that
sense,
awesome.
This
able
static
tokens
by
default
as
a
vanities,
1:18.
D
A
A
That
does
like
the
product
they
sell
is
automatic
registrations
and
they
totally
rely
on
these
tokens
for
admin
uses
for
the
platform,
so
I
know
a
lot
of
people
that
actually
use
these
static
tokens
also
and
how
they
operate.
Just
removing
that
an
in
simple
way
would
probably
hijack
a
lot
of
operations.
So
many
people.
E
D
E
A
E
E
I
need
to
build
an
image
for
stretch,
but
I
think
we
decided
that
that
that
will
then
unblock
everything
else.
We
have
two
more
and
we
have
two
emergencies
on
yours.
I,
don't
think,
is
any
decisions
right,
yeah.