►
From YouTube: 2020-06-18 Cassandra Kubernetes SIG
Description
John Sandra walks through a Alpha version of a common CRD. Some important discussion on how Kubernetes and Cassandra see data centers and racks. Feedback form this meeting will be incorporated to the next version.
A
A
So
I'll
go
through
start
talking
and
feel
free
to
interrupts
and
at
any
point,
the
the
first.
The
first
type
that
you
see
listed
here
is
rack,
which
shouldn't
be
very
surprising.
It's
a
common
with
all
the
the
CR
DS
and
the
the
data
types
have
fields
for
maps
for
the
labels,
annotations
and
toleration.
A
A
A
You
know
with
what
granularity
does
it
need
to
be
done
at
the
rack
level,
the
whole
data
center
and
so
I
wasn't
I
didn't
get
hung
up
on
that
because
the
focus
was
on.
My
focus
right
now
is
on
topology,
but
it
was
a
question
that
popped
up
while
I
was
I
was
going
through
this.
The
next
thing
next
type
I
have.
B
A
C
A
C
A
A
I'll
start
at.
Let
me
I'll
start
here
at
the
data
center,
so
specifying
the
cluster
name,
the
data
center
name
and
the
nodes
per
rack
I
was
I
talked
with
you
know,
I
said
of
talking
yesterday
with
Jim
and
cereal
and
I
think
you
know
think
I
think
there's
probably
for
everyone
on
the
call,
probably
general
consensus
that
you
want
to
have
balance
balance
racks
in
terms
of
them.
A
The
I
both
there's
different
property
names
in
the
different
in
the
different
series
for
specifying
whether
you
want
like
cats
operator,
for
example,
said
you
know
for
the
affinity
qez
operator,
it
says
hard
or
soft,
which
kind
of
matched
to
you
know
the
the
language
used
in
the
community
stocks.
I
forget
some
had
what
the
the
term
is
in
its
cluster
operator
and
has
operator,
but
I
also
was
looking
at
one
of
the
more
polished
Postgres
operators.
A
There's
a
several
puskás
operators
as
well
and
I,
think
they
just
use
pod
anti
Efendi
so
and
I
said
well.
Let
me
to
make
it
pretty
explicit
there
with
the
accepted
values
being
what
they
mapped
to
the
try
to
make
that
more
make
itself
documenting
and
again
so
having
at
the
data
center
level.
Also
reading
declare
labels
annotations
and
toleration
and
emerging
them.
A
A
A
Your
top
level
type
is
the
data
center
and
you
would
simply
create
a
second
if
you
want
to
have
a
multi
DCE
cluster
II
simply
create
a
second
data
center
object
and
make
sure
they
have
the
same
cluster
name,
and
then
the
operator
takes
the
rest.
So
it's
trying
to
figure
out
or
come
up
with
a
proposal
that
and
both
in
those
conversations.
A
That
going
forward,
and
so
what
I
what
I
came
up
with
last
night
is
having
a
cluster
type,
and
it
consists
again
keeping
with
the
theme
of
being
able
to
define
your
labels,
annotations
toleration
at
each
level
and
then
having
them
propagate
down
and
be
merged.
But,
more
importantly,
you'll
see
here.
I've
declared
a
tight,
a
data
center
ref,
which
is
just
a
object
reference
and
then
I
have
an
array
of
data
center
objects,
and
an
array
of
data
center
rests.
A
A
If
you
look
at
a
example,
a
manifests
for
forecast
cop
and
then
also
see
a
also
making
use
of
the
data
center
rest,
where
it's
just
I've
only
specified
the
name
and
I
could
specify
the
the
the
minimally
you
know
minimally
specify
the
name.
The
resource
ID
could
be
specified
as
well,
but
minimally,
the
name
so.
D
Can
I
ask
a
question
sure
so,
what's
the
use
of
I
mean
what's
the
intent
of
having
a
different
object
for
the
Dana
Center
I
can
see
how
useful
it
could
be?
By
doing
so,
why
I
mean
it
seems
that
having
everything
in
one
place,
I
mean
one
object.
Where
you
see
exactly
the
topology,
there
makes
sense,
but
it
seems
that
you're
trying
to
make
both.
D
A
A
E
D
I
mean
you
can
have
one
object
or
multiple
object
in
even
the
same
llaman
far
there
won't
be
any
issue
I'm,
just
wondering
how
useful
it
is
to
have
those
objects
are
as
different
objects.
I
mean
I'm,
not
sure.
For
example,
I
mean
that
having
two
controllers
brings
anything
on
the
table.
I
mean
the
controller
of
the
cluster.
Wouldn't
be
sorry,
we
don't
do
anything.
D
I
mean
I
feel
like
it
will
just
tell
the
the
tenant
sinners
to
use
that
custom
name
or
maybe
that
configuration
file,
or
maybe
things
like
that,
so
in
a
sense
just
a
way
to
inherit
for
the
dress
and
some
properties.
That's
what
I
understand
from
what
I
read
I'm
trying
to
you
know,
get
a
sense
of
what
the
purpose
of
having
different
objects
and
and
split
the
cluster
in
different
objects.
D
Will
it
be
easier
to
just
duplicate
their
Center
for
another
cluster
I'm
trying
to
understand
you
know
how
useful
it
would
be
for
now
we
create
a
cluster.
We
have
the
configuration
of
the
data
centers
in
in
the
cluster.
Everything
is
in
there.
You
want
to
know.
What's
the
topology,
you
just
go
into
the
cluster
and
you
see
everything
you
don't
have
to.
Oh,
there
is
a
little
center
called
a
okay.
Let's
see
that
there
and
cynical
eh.
Oh,
this
is
a
configuration
that
we
don't
have
to.
You
know
walk
through
those
different
objects.
D
A
Sure
so
one
of
the
so
one
of
the
motivation
so
I'll
come
back
to
what
I
was
describing
before
and
but
then
I'll
mention
one
of
the
other
motivations
was
to
try
to
arrive
at
a
middle
ground
for
what
I
think.
So
what
I
was
hoping
in
talking
with
you
yesterday,
serial
with
Jim
was
finding
helping
that
reach
more
convergence,
which
I
did
with
some
of
the
questions.
A
Each
of
you
guys
and-
and
it
was
it
was
really
helpful
because
there
was
convergence
on
some
of
the
things
seeing
like
I
had
some
questions
about
different
fields.
I
wasn't
sure
about
the
doing
the
upgrade
strategies
and
canary
upgrades
and
seeing
there's
a
lot
of
similarity
there,
but
in
terms
of
the
topology
stuff.
A
So
it
was
my
way
of
trying
to
see
what
trying
to
come
up
with
a
way
that
is
incorporates
the
ideas
that
both
of
you
expressed
in
this
today
without
me,
taking
trying
to
maintain
the
middle
ground
with
with
all
so
so.
So
that's
what
the
one
point,
but
the
only
one
I
do
see
value
in
you
know
he
said
with
well.
D
I,
like
your
arguments
of
the
testing,
the
arguments
I
think
it's
a
good
one,
so
we
don't
have
to
create
a
whole
cluster
just
to
test
one.
Just
a
business
object,
for
example.
So
that's
that's
something
interesting
and
I.
Understand
that
you
want
to.
You
know,
make
two
different
words
work
or
together
using
the
same
so
yeah.
It
makes
sense
to
me
sorry
I
think
you
wanted.
Oh.
F
Yeah
sorry
I
did
want
to
interject.
You
know
not
to
come
out
of
left
field
here,
but
one
of
the
useful
elements
of
separating
different
sub
parts
into
standalone
kubernetes
objects
is.
You
can
use
some
of
the
built
in
status
capabilities
when
you're
operating
on
those
two
Burnett
eases
and
objects
as
well
right.
So
you
could
imagine
a
data
center
having
a
distinct
sort
of
status
from
the
rest
of
the
cluster
at
a
given
time.
Does
that
make
any
sense,
yeah.
G
Also
think
something
some
things
we're
trying
to
land
and
Cass
operator
is
like
getting
some
more
of
the
Cassandra
entities,
managed
declaratively,
like
we
already
merged
in
getting
users,
users
and
their
passwords
and
their
super
user
status,
declaratively
managing
kubernetes,
and
so
something
like
that
I.
You
know.
G
I'm
not
saying
different
different,
no,
not
different
kinds
and
kubernetes,
but
just
on
the
cluster.
Maybe
I
have
a
slice.
That's
like
users
right,
and
that
makes
more
sense
on
that
one
than
it
doesn't
a
data
centers.
But
we
don't
have
I.
Don't
have
this
split
right
now
and
guess:
I'm,
crazy,
okay,.
E
One
thing
that
would
make
me
nervous
is
of
its
forcing
topology
on
people
and
but
I
I
get
the
impression
that
it's
not.
This
is
somewhat
optional,
forcing
Cassandra
topology
yeah
Oh
like,
for
instance,
well
I,
don't
think
anyone
should
use
simple
snitch,
but
yeah
like
this
assumed
you're
using
a
snitch
that
does
data
center
awareness,
Rock
awareness.
E
G
A
Correct
I,
you
know
you'll
see
a
pretty
every
one
of
the
fields
except
for
the
names
that
they're
optional
you
have
the
omit
empty,
so
that
is,
that
is
correct.
Yeah
I
mean
yeah.
That
was
just
for
illustration
here
too,
because
if
I
wanted
to
include
the
labels,
since
the
focus
is
on
topology,
just
to
make
it
clear,
you
know
it
would
for
the
mapping
to
the
zones.
G
Yeah
so
I,
don't
yeah,
so
yeah
I,
don't
think
silencing
the
Cassandra
concepts
makes
sense
in
the
name
of
like
I.
Think
that's
over
indexing
on
making
it
too
easy
to
adopt.
It's
like
sorry.
We
can't
hide
the
idea
of
data
centers
and
racks
completely
like
you're.
Just
gonna
have
to
have
some
taupe
some
amount
of
mentioning
them,
but
if
you
don't
want
racks
just
to
find
one
rack
and
move
on.
E
G
A
Yeah
I
I
agree
with
that
that
the
I
think
it
was
Jim
made
really
good
point
us
we're
talking
today
that
you
know
you
is
a
distributed
database
cluster
database
with
high
availability,
you
download
it
and
everything
is
set
up
with
examples
of
default
string,
a
single
node
by
itself
you
downloaded
song
and,
and
so,
if
you're
completely
new
to
Cassandra,
it
would
be
nice.
You
know
and
and
I
know
you
know,
Santa's
gonna
loud.
You
know
some
of
the
tooling
is
there's
been
around
a
lot
longer
than
some
the
tooling
to
make
things
easier.
A
D
Sorry
so
you
have
Tara
sinners
objects,
so
you
have
external
and
in
inner
objects
inside
the
cluster,
so
bulldoze
the
inner
they're
a
sinners
object.
It
would
be
managed
by
the
data
center
and
controller
or
by
the
cluster
controller.
What
I
mean
by
that
when
the
cluster
sees
those
and
that
there
is
no
existing
object
for
them,
you
would
create
objects
for
them
and
then
the
controller
will
just
kick
kick
in
and
takes
kick
take
care
of
them.
Can
you
see
it
yeah.
A
So
so
what
I
I
think
if
I
understood
your
question
correctly,
I
think
I
put
my
thoughts
down
towards
the
bottom
here.
So
in
this
scenario,
where
so
you
create-
let's
you
starting
from
scratch.
I
have
my
data
center
yanil,
I
I
created,
so
the
data
center
controller
gets
the
notification
starts
doing
the
reconciliation.
A
A
A
D
Yeah
so
I
asked
that
question
for
you
know,
because
when
the
cluster
is
set
up,
then
how
do
you
update
the
objects?
Does
it
mean
that
you
have
to
go
on
the
cluster
to
update
the
II
objects
that
were
directly
added
to
the
configuration,
or
do
you
have
to
go
to
each
object
and,
as
in
that
case,
you
are
mixing
references
and
actual
object?
I'm
wondering
okay,
you've
got
the
cluster.
D
Everything
is
set
up,
I
will
to
add
one
more
node
to
their
sin
or
dc1
should
I
look
for
the
object
or
should
I
go
on
the
cluster
itself
and
what,
if
I,
wanted
to
do
the
same
on
this
III
can
I
do
the
same,
because
the
object
will
be
just
the
configuration
would
be
just
added
to
the
cluster
itself
or
in
that
case,
there's
only
one
choice
which
is
going
on
the
object
directly.
That's
why
I'm
asking
because
you
are
you
mixing
references
and
I
get
true
objectivity.
A
A
I
think
that's
a
really
good
question,
because
I
think
that
that
raises
some
points,
one
in
terms
of
making
the
intention
making
the
implementation
more
complicated,
complex
and
to
I
think
it
also
makes
decreases.
You
potentially
decreases
usability
as
an
end
user.
You
know
you're
trying
to
sort
out
okay.
What
fields
can
I
edit,
where
you
know
what
I'm
looking
at
this,
especially
as
it
gets
more,
you
know,
gets
more
involved.
You
start
throwing
in
other
other
things,
so
I
mean
I,
know
it's
there
times
where
I'll
I'll
go,
make
a
change.
A
D
Same
thing
is
today:
if
you
want
to
just
remove,
I,
mean
scale
down
or
remove
a
data
center,
we
go
on
the
cluster
object
and
we
just
delete
it
from
there
and
it
does
scale
down
and
delete
the
data
center.
What
would
happen
in
that
case
because,
in
one
case,
I
can't
just
delete
the
content,
but
in
another
case
this
is
just
a
reference.
We
just
will
kick.
You
know,
kick
off
the
deviation
of
the
data
center
or
just
remove
the
reference
in
the
cluster.
D
G
Maybe
this
is
a
question
that
a
particular
kubernetes
administrator
may
want
the
Ahlan
all
in
line
way
and
another
one
may
want
like
the
referential
way,
which
I
think
would
help
if
you're
using
multiple
namespaces
or
maybe
like
even
the
idea
of
multiple
kubernetes
clusters
right.
So
maybe
you
shouldn't
be
mixing
and
matching
them
in
one
cluster,
but
I.
Don't
know
that
that's
like
validation,
that's
easy
to
enforce.
G
A
So
maybe
maybe
enforcing
one
of
the
other
and
making
it
the
use
of
the
two,
usually
exclusive
and
and
then
the
other
thing
you
mentioned.
Jim
is
interesting.
It's
about
multi
cluster,
just
something
I
started,
trying
to
think
more
about
for
some
other
work
past
couple
days.
I
hadn't
thought
about
it
in
this
context,
but
but
that's
interesting.
A
A
So,
certainly
so
in
terms
of
the
I,
certainly
the
question
Cyril
to
I
make
a
lot
of
sense.
You
think,
really
you
know
what
are
the
implications
of
using
a
reference
said
if
you
remove
that,
simply
just
removing
the
reference
from
the
cluster.
What
what
is
that?
What
does?
What
is
the
behavior
that
what
does
that
mean?
That's
certainly
less
clear
than
if
you
haven't
defined
in
line
and
if
it's
a
reference
to
mate
yeah,
if
you're
using
references,
then
maybe
it's
the
matter
of
said.
A
H
A
A
H
Me
go,
let
me
pose
a
question
in
a
different
way.
I
would
I
would
I
would
argue
that
the
primary
benefit
of
using
references
is
so
that
you
can
a
couple
topology
from
you
know
the
underlying
the
underlying
database
nodes
right
and
the
only
reason
you
would
want
to
do
that
or
need
to
do.
That
is,
if
you
were,
you
know
if
you're
gonna
avoid
just
one-to-one
mapping
between
those
data
centers
and
those
this
databases.
So
if
that's
not
a
need,
maybe
this
is
introducing
an
unneeded,
complication
or.
H
G
I,
don't
think
that's
I,
don't
think
that's
the
case,
I
think
not
declaring
them.
If
you
declare
in
line
like
this
whole
object
needs
to
live
in
one
kubernetes,
namespace
right
like
and
if
you're
saying
this
is
a
multi
data
center
cluster.
What
does
it
mean
that
it's
in
one
namespace
like
do
all
the
data
centers
exist
in
their
namespace
to
like
I?
Think
that
you're
not
making
them
referential?
So
you
can
share
them
across
clusters
which
I?
Don't
we
don't
think
is
that
logic
I'm
assuming
what
ever
never
good
to
assume?
G
But
that's,
that's
not
very
logical,
like
you.
Don't
share
data
centers
across
different
Cassandra
clusters,
but
decoupling
them
from
being
all
all
packed
together,
means
that
maybe
I'm
gonna
run
one
data
center
under
one
name
space
with
some
sort
of
the
quota
restrictions
and
whatever
you
get
with
that
namespace
and
different,
wanted
it
in
a
different,
the
different
namespace
and
if
I've
got
the
network
topology
set
up
right
like
maybe
these
are
not
even
in
the
same
region.
I
know.
G
The
right
solutions,
so
I,
don't
know
for
me,
I
think
not
not
listing
if
you
want
to
do
like
really
good
so
distributed.
Support
like
tying
these
all
into
one
masterpiece
of
Y.
Amal
is
kind
of
gonna,
be
a
serious
hurdle
at
some
point.
You're
gonna
want
to
walk
back
and
say:
oh
now,
I
need
to
like
support
ya.
Multi
multi
kubernetes
cluster
multi,
whatever.
G
G
A
Think
I
forgot
I
went
one
of
the
last
things.
I
did
one
of
the
last
updates
that
I
made
to
the
dock.
Was
it
making
sure
added
the
the
cluster
name
or
the
cluster
field
to
the
data
center
object,
and
you
never
had
to
put
enough
in
there
to
uniquely
the
the
cluster
name
by
itself
may
not
be
sufficient,
but
enough
to
uniquely
identify
the
cluster
I.
H
A
It's
I
mean
it's
yes
and
no
I
mean
it's
simply
just
saying
the
the
name
of
the
cluster
as
far
as
the
data
center
controller
that
well
as
it
you
know,
is
that
what
I
described
in
my
doc
here,
the
data
center
controller,
you
know
it
would
see
that
name
or
anything
else
would
identify
the
cluster
and
and
simply
just
wouldn't
do
anything
with
it.
It's
just
okay,
it's
other
than
other,
only
other
than
just
validate.
It's
not
empty,
for
example,
but
then
it
would
be
the
the
cluster
controller.
A
D
A
G
Yeah
I
think
it's
pretty
challenging
to
design
these
things
because
you
don't
want
to
like
we
leave
aside
like
the
important
kubernetes
concepts
right
so,
like
you
say
like,
oh,
you
have
to
create
everything
all
in
one
namespace
like
which
I've
seen
a
couple
operators
do
that
weren't
that
maybe
weren't
even
Cassandra
operators.
If
it's
like,
oh
yeah,
we
really
only
intend
you
can
only
install
the
operator
in
a
namespace
name
this,
and
it
only
looks
in
that
namespace
to
do
stuff.
G
G
D
G
A
G
A
C
E
I
will
I
will
put
another
schedule
up
or
another
meeting
scheduled
at
this
time
for
next
week
and
then
work
on
getting
a
weekly
schedule.
It
will
probably
just
be
the
same
time
two
weeks
in
a
row
just
so
we
can
keep
that
that
rotation
going
but
yeah
I
mean
I,
I,
think
you're
right,
Jim
nailed
it
on
the
head.
This
kind
of
more
high
bandwidth
conversation
is
what's
needed,
then
we
can
give.
You
know,
John
really
good
feedback,
and
then
he
can
go
off
in
his
workshop
and
tinker
and
update
it
up.