►
From YouTube: Kubernetes SIG Apps 20190701
Description
No description was provided for this meeting.
If this is YOUR meeting, an easy way to fix this is to add a description to your video, wherever mtngs.io found it (probably YouTube).
A
A
A
B
A
B
So
we
have
CR,
DS
can
be
used
for
a
whole
bunch
of
things
and
a
lot
of
times.
We
think
of
them
is
a
cluster
extension,
but
one
of
the
ways
people
are
starting
to
use
them
is
in
the
way
they
deploy
and
run
applications
such
as
operators,
and
so
here's
the
use
case
I,
have
in
mind.
You
have
two
people
with
two
entirely
separate
namespaces
who
don't
know
about
each
other
and
each
of
them
installs
an
operator
say
my
sequel
or
Postgres,
or
something
like
that.
B
It's
the
same
one,
using
the
same
CRD
now
you've
got
the
one
C
or
D
in
who
really
don't
know
about
each
other.
One
of
them
wants
to
upgrade,
and
you
might
do
things
I
change
the
store
version.
Things
like
that.
How
do
you
handle
the
case
of
managing
these
differences?
What
are
the
rules?
What
do
we
tell
people
to
do?
How
do
we
explain
this
I.
C
B
Certainly,
and
so
this,
but
this
is
a
real-world
problem
of
go,
install
it
and
you
get
somebody
with
cluster
admin
to
do
it
and
then,
for
example,
you've
got
one
Operator
who
upgrades
their
operator
and
they
need
a
new
version
of
the
CRT.
In
order
to
do
that,
maybe
they
went
from
v1
to
v2,
or
you
know:
V
1,
alpha
1
to
a
few
Hyundai
2
1,
and
you
still
got
somebody
else
using
the
V
1
alpha
1.
A
B
C
Sorry
I
was
trying
to
understand
how
you
can
manage
to
install
tools
like
if
you
are
installing
the
second
one.
You
will
not
be
able
to
trade
the
CRT,
because
it
will
already
exist,
which
should
probably
be
Poynter,
that
it
is
already
installed
in
the
cluster
and
those
are
customized.
This
is
not
a
controller
per
user.
This
is
control
per
cluster.
B
C
A
A
B
So
to
restate
consider
the
case
where
you
use
a
tool
that
installs
the
CRT
for
you
and
has
silently
ignores
that
it's
already
there
and
doesn't
tell
you
that's
a
possibility,
in
fact
that.
B
B
So
say
people
are
asking
for
tools
that
may
silently
ignore
if
it's
already
there
and
just
keep
running
right,
that's
something
we
have
had
asked
for
over
on
him,
silently
ignore.
If
the
CR
D
is
already
there
and
just
install
the
code
base
and
keep
going
so
you
can
install
something
more
than
once
and
not
run
into
that
error.
B
A
C
You
at
least
the
part
before
you
start
asking
for
the
advice.
It
was
clear
to
us:
I
don't
and
the
current
state
CR
DS
arnott
moved
it
in
and
they
are
meant
just
for
cluster
admin,
which
kind
of
means
not
Moodle.
If
you
are
having
a
tool
like
how
to
install
it,
you
have
probably
given
cluster
admin
rights
to
every
person
on
the
cluster
using
hell.
We
are
we're
not
a
good
thing.
I
guess
at
all,
I
think.
A
Having
party
departments
there
isn't
really
a
good
way
to
I,
guess,
distribute
and
install
operators.
Today,
you
know
other
than
a
tool
like
just
using
something
like
home
or
just
you
know,
installing
a
set
of
manifests
using
QCT
and
oftentimes.
It's
done.
That's
usually
done
by
developers,
especially
in
the
case
of
hell.
Hundreds
will
install
things
in
their
namespace
and
you
can
easily
run
into
these
kind
of
conflicts
as
a
result.
B
Yeah
and
so
if
we
don't
tell
people
or
give
them
guidance
on
what
to
do
or
where
they
could,
you
know
cause
get
into
trouble.
Then
they're
gonna
run
into
it
and
they're
already
asking
to
route
around
the
warnings
and
errors
that
are
there
and
there's
no
guidance
and
I.
Don't
think
we
expected
people
to
use
them
quite
the
way
they
do.
B
C
Currently
CR,
these
are
meant
for
extending
the
cluster
rich
resources,
and
that
means
on
the
costella,
like
I
own,
the
cluster
I'm,
going
to
add
this
resource
there.
They
are
not
meant
for
a
user
who
is
not
concerned
and
basically
a
developer
or
tool
that
the
developer
uses
to
be
expanding
the
cluster,
because
you
are
trying
to
make
it
multi-tenant
and
you
are
ignoring
the
privileges
that
are
constrained
for
the
CR
DS
I
know
people
are
doing
it.
I
don't
have
a
better
answer,
then
you
shouldn't
be
doing.
C
A
A
Think
CRE
is,
and
you
know
much
like
hosta
robe,
awnings
and
other
subscribe.
Resources
are
a
opps
level.
Concern
and
I.
Don't
know
if
it
is.
If
there's
a
place
that
we
can
document
how
we
expect
people
to
manage
the
ADIZ,
maybe
that's,
maybe
that's
something
I
could
live
in
a
cube,
Nettie's
documentation,
that's
what
I
was
gonna.
B
Say,
there's
nothing
communicating
this
now,
that's
a
problem
right.
If
you
don't
say
anything,
people
will
just
do
whatever
they
wherever
they
want
to
go
and
I,
not
aware
of
any
guidance
on
this
at
one
point
there
was
some
something
in
a
mailing
list
that
I
was
on
that
thread
and
I
would
probably
have
trouble
finding
that
item
again.
It's
not
clear
where
what
we
tell
people.
B
B
A
A
Okay,
if
there's
nothing
else
on
that,
we
can
move
on
so
I
want
to
give
a
quick
update
on
kubernetes
115,
so
that
was
released
last
week
and
some
of
the
features
from
the
cig,
apps
side
surround
pod
disruption,
budget
there's
now
support
for
custom
resources
and
also
the
immutability
has
been
relaxed.
So
it's
now
possible
to
mutate
PDP's,
and
then
there
was
also
a
fix
for
the
incorrect
rock
mountain
defaulting
and
I
was
introduced
in
112.
A
A
D
E
C
Yes,
since
I
see
the
broke
mount
figs
here,
it
reminded
me
we
should
probably
sing
well
next
time
we
cancelled,
but
maybe
the
time
after
that
we
should
probably
sing
Kum
it
since
the
case
of
mutating
admission,
basically,
which
cannot
hurt
the
dog
for
which
we
should
probably
talk
for
a
long-term
fix.
So
the
spirits
are
allowed,
don't
happen
again
well,
they
are
happening
now
for
mutating
emissions,
but
this
was
an
instance
of
it,
so
we
should
probably
sink
in
the
future
on
a
long-term
solution.
C
C
E
C
Okay,
so
for
those
who
don't
know
about
it,
if
there
is
a
mutating
admission
that
make
sure
it
changes,
the
boots
back,
like
the
controller
sees
the
boots
back,
is
something
and
mutating
admission
as
a
fielder.
The
resources
that
actually
compare
hey,
I'm,
well,
I
figure,
I,
think
deployment
has
for
region.
E
F
A
A
F
Can
you'll
see
so
thinking
once
how
we
introduce
this
new
feature
and
it's
still
offered
today
and
the
feature
is
about
how
we
garbage
collect
jobs.
So
if
you
create,
if
you
have
a
something
like
a
script
that
keeps
creating
jobs
and
eventually
you
want
to
somehow
clean
it
up
and
today,
there's
no
easy
way
to
clean
made
up,
and
so
this
feature
is
to
allow
you
to
set
a
TTL
on
your
jobs
so
that
you
can
have
it
cleaned
up.
F
Have
it
cleaned
up
after
it's
either
complete
or
failed
after
some
time
so
I
have
a
cap
here
and
it's
in
C
gaps
and
then
I
want
to
bring
this
feature
to
better.
So
I'm.
Looking
for
contributions
to
help
me
and
bring
this
feature
to
better
and
I
also
want
to
introduce
this
feature
to
finished
paws
as
well,
and
the
code
is
pretty
simple.
It's
under
kubernetes
package
controller
at
retail
after
finished,
and
the
controller
code
is
about
300
fish
lines.
F
So
basically,
it
and
I
have
some
more
details
for
that.
This
in
this
cap
is.
F
C
C
G
D
C
F
Originally,
this
proposal
is
for
for
those
resource,
like
secrets
and
config
maps,
I'm
thinking
you
can
set
a
TTL
after
those
resources
are
not
referenced
or
used
by
any
other
resources.
But
that
requires
basically
something
like
a
reverse
on
the
reference.
I
proposed
this
to
API
machinery
and
they
have
some
concern
about
the
mobility
yeah.
C
F
We
can
simply
just
use
the
TTL
for.
C
C
F
A
C
One
in
metadata,
one
inputs
back
and
you
were
handle
it
differently
as
well,
so
I
guess,
promoted,
I,
guess
making
this
for
work
for
thoughts
and
then
promoting
it
to
beta
one.
That's
stable,
I
guess
that
sounds
good
to
me
and
we
can
address
the
Jared
case
in
the
metadata
early
if
it
comes
to
it,
yeah.
G
F
F
F
E
E
Had
one
more
question
so
I
mean
to
the
proposal,
so
my
understanding
is
that
the
pod
management
policy
right,
which
is
parallel
or
non
or
remember
what
the
other
option
is,
but
it
only
applies
to
creation
of
stateful,
said
donation
and
the
scale
operation
right
it
has
nothing
to
do
with
rolling
update.
Is
that
from
the
standing,
correct.
E
E
I
want
to
make
it
only
work
for
rolling
updates,
because
that's
the
whole
reason
like
we
want
to
make
rolling
updates
go
faster
for
stateful
sets
which
can
support
that.
So
my
understanding
is
that
I
I
actually
really
don't
care,
whether
it
work
for
the
rapid
management
policy
case
or
not.
But
if
you
all
think
that
that
should
also
be
supported.
Yeah,
please
comment
and
see
and
I
think
my
basic
understanding
is
that
I
don't
really
see
a
need
for
Mac's
and
available
to
be
supported
for
that
those
cases
cool.
That's
it.