►
From YouTube: Cassandra Kubernetes SIG Meeting 2020-05-21
Description
Discussion on how to move forward with a combined operator plan.
In the call we agreed that the CEP text was important and at a good place, but the real work for a Kubernetes based operator needs to happen in the Custom Resource Definition. All of the existing operators have some form of CRD which are very similar based on how you deploy Cassandra in Kubernetes. John Sanda and Patrick McFadin will be looking at how to gather the various CRD artifacts already in GIT repositories for further discussion.
Example CRD: https://github.com/datastax/cass-operator/blob/master/operator/deploy/crds/cassandra.datastax.com_cassandradatacenters_crd.yaml
Moving the CRDs to a GIT repository will then allow for changes to be tracked as PRs to a final state.
One the CRD is completed, then the discussion of code donation can move forward and CEP can be completed.
A
All
right,
okay!
Well,
let's!
Let's
start
this,
just
I,
don't
know
if
you
all
saw
there
was
a
I
had
an
episode
on
accelerate
about
the
Cassandra
operator.
I
thought
it
was
very
good
thanks
to
Chris
John
got
involved
in
it
as
well,
but
I'm,
sorry
that
hit
the
cutting
room
floor
cuz
we
had
to
shorten
it
up.
There
was
also
some
good
contributions
and
there
from
Yelp
and
from
Sky
it
was
a
great
little
community
chat
about
kubernetes
and
cassandra.
So
if
you
get
a
chance
to
watch
it,
it's
really
kind
of
cool.
A
What
I
want
to
talk
about
right
now
is
some
of
the
things
that
I
mean
I've,
been
talking
to
some
folks
like
just
Apache,
Cassandra
folks
and
generally
in
community.
It's
like
alright.
Where
do
we
go
now?
I,
I'm,
thinking
that
what
we
need
to
do
at
this
point
is
not
put
a
lot
of
pressure
on
the
Apache
Cassandra
project
to
know
hey,
we
want
to
dump
some
code
here,
figure
it
out.
I
think
this
is
a.
This
is
a
separate
effort,
and
by
that
I
mean
getting
together.
A
The
groups
that
are
interested
in
actually
contributing
code,
creating
a
separate
repo
somewhere
doesn't
matter
I
could
create
one
right
now
working
around
that
and
really
kind
of
building.
This
thing
this
because
of
CEP,
as
it
sits
right
now,
I
think,
is
in
a
fairly
it's
an
ironed
out.
It's
got
a
lot
of
content
in
it,
but
for
this
to
move
forward,
we
need
to
have
some
code
together.
If
there
is
code
now
here
are.
A
My
concerns
is
that
that
means
that
who
wants
to
contribute
code
and
is
there
a
possible
path
to
making
that
happen,
and
this
is
the
real
talk.
Part
of
this
is
that
we
have
a
lot
of
parties
that
have
created
and
have
built
operators
over
years
literally-
and
this
isn't
like
this
is
brand-new.
This
is
World
Greenfield
here
with
code.
There
isn't
a
lack
of
it,
there's
a
lot
of
it,
and
this
could
potentially
turn
into
a
really
nasty
fight
in
community
about
all
the
potential
contributors
here.
A
I
wish
that
we
had
instant
cluster
here
today,
but
I
will
make
sure
to
get
them
on
here
next
time,
because
they,
you
know
they're
interested
in
contributing
as
well,
but
I
want
to
open
this
up,
for
discussion
is
what
is
our
next
step
moving
forward?
What's
our
best
way
to
move
forward,
I
would
love
to
hear
some
ideas
around.
This
I've
floated
a
few
ideas,
but
I
think
this
is
a
good
discussion
point,
and
if
anybody
wanted
to
take
the
floor
right
away,
it's
yours
I
mean.
B
I
can
take
my
I,
have
my
opinion,
I
think
regarding
cats
cooked
we
as
I
said
last
last
time,
we're
happy
to
donate
it,
so
one
path
would
be
for
people
to
look
at
it
and
say:
is
that
acceptable?
I
can
contribute
all
the
list
of
features
it
has.
We've
just
updated
the
documentation
we
just
did
yesterday,
so
there's
a
brand
new
website
with
the
documentation.
That
explains
all
the
steps
that
you
can
go
through
to
configure
everything
that
you
want
to
configure
in
it.
B
B
Maybe
we
look
at
it
and
say
is
that
acceptable
and
we
are
working
with
instructors
on
on
Casca,
so
it
could
backup
and
restore
right
now
as
we
speak,
we
work
together
and
they
have
modified
a
sidecar
to
do
the
rest.
All
in
their
online
rest
offer
table
that
we
asked
for
because
it
was,
it
only
did
restore
on
a
new
cluster.
So
now
it's
going
to
do
rest
on
an
existing
cluster
thanks
to
Stefan
who's
supposed
to
come.
B
So
the
other
thing
that
we
need
to
do
with
Casper
would
be
to
include
the
data
stacks
sidecar.
That
would
be
one
way
to
look
to
get
work
together.
Is
that
easy
to
add-
and
that
would
be
very
nice
and
I
mean
as
I
said,
Casco
has
nothing
orange
in
it.
It's
from
the
start
was
based
for
the
community.
We've
tried
to
put
everything
on
the
on
the
table
and
now
I
couldn't
understand
it.
It's
not
accepted
I'm,
not
saying,
should
I'm
just
saying
that
it
should
be.
B
It
actually
works
pretty
well,
but
it's
not
perfect,
but
it
does
I
mean
I'm.
Looking
at
I,
look
at
the
the
most
of
the
commits
that
are
on
the
other
operators
and
there's
something
like
six
or
eight
months
behind
so
okay,
we
can
wait
eight
months
that
can
be
I'm
quite
sure.
You
guys
are
gonna
catch
up
pretty
quickly,
but
anyway,
that's
one
plan.
I
think.
A
And
there
was
a
mention
of
the
sidecar
that
was
actually
brought
up
in
the
contributor
meeting
the
sidecar.
There
is
a
sidecar
project,
a
sidecar
CEP,
which
is
turning
out
to
be
a
pretty
hard
dependency
for
the
operator,
and
that
was
something
that
we
discussed
in
the
Cassander
contributor
meeting
as
I
understand
it
that
that's
something
we're
gonna
have
to
manage
soon.
The
sidecar
project.
This
seems
just
stalled
out.
Some
there's
not
there
was
a.
A
D
What
I
would
ask
is
I
mean
from
working
on
the
Cassano
project,
quite
a
bit
that
we
keep
running
into.
Is
this
question
of
like
if
we
have
seven
different
operators?
Are
they
expressing
seven
different
points
of
view
like
if
a
user
knows
nothing
about
the
ecosystem?
They're
like
I
just
want
to
operate
at
orbit
Cassandra?
Could
we
enumerate
seven
different
operators
and
each
of
them
have
a
reason
to
exist,
or
is
the
Venn
diagram
they're,
basically,
all
overlapping,
with
each
other
all
the
way
down
in
terms
of
how
they
look
to
the
user?
D
First,
the
former
makes
sense
just
to
have
them
all
available
and
have
an
agreed-upon
standardized
API
for
all
these
things
interact
in
the
ecosystem.
If
it's
the
latter,
that's
where
I
would
I
would
Rhian
reiterate
Brandon's
point,
but
if
we
have
a
bunch
of
different
software
defects
that
are
essentially
doing
the
same
thing
in
the
end
in
terms
of
their
presentation
to
the
user,
it
doesn't
really
add
a
ton
of
value
to
any
of
us
much
less.
The
people
trying
to
use
the
ecosystem
to
have
a
bunch
of
different
options
to
do
the
same
thing.
B
I
think
under
the
target
target
is
one
operator
that
is
supported
by
the
community
for
everybody
to
use
I
think
we
all
agree
on
that.
The
thing
is,
what's
the
starting
point,
do
we
start
from
scratch?
Do
we
start
from
Gasca?
Do
we
start
from
data
stacks
operator
or
inspector?
Whoever
else
is,
you
know,
I,
don't
know
all
the
others
looking
quite
good,
but.
D
Yeah
I
mean
Kevin
naive
approach,
and
things
we've
run
into
on
the
cassano
project
itself
is
just
say:
look
what
is
the
end
state
looked
like
and
which
one
of
the
operators
is
closest
and
start
from
that
one,
and
they
have
everybody
else,
merge
into
it
he's
in
theory.
That
would
be
your
shortest
path
to
whatever
the
destination
is
now,
whether
or
not
that's
like
politically
feasible
or
if
whatever
institution
wrote.
D
That
operator
has
the
bandwidth
to
do
with
emerge
cycles
or
there's,
there's
all
kinds
of
other
eCos,
some
questions
that
come
out
from
that,
but
at
least
from
kind
of
tiebreaking
on
a
starting
point.
It's
like
if
you're
gonna
run
a
race
you
get
to
choose
where
you
start
in
the
race
you
may
as
well
choose
whichever
starting
point
is
closest
to
the
finish
line
and
go
from
there.
E
E
E
Did
you
do
this
in
a
way
to
like
drag
new
users
who
know
kubernetes,
but
don't
know
Cassandra
in
vs.,
like
people
who
know
Cassandra
thoroughly
well
like
feel
comfortable,
standing,
200
bits
of
the
ML
configuration,
but
you
know
that's
a
lot,
a
lot
of
burden
for
a
new
user,
so
I
just
know
how
we
built
our
operator.
I
was
working
on
it
since
pretty
much
day,
one
and
now
I'm
the
engineering
lead
on
that
team
and
I'm
I'm
kind
of
pro
for
some
coordination.
I
just
don't
know
how
to
get
there
right
now.
E
And
I
was
trying
it
I
was
going
through
we've,
just
just
to
throw
it
all
out.
There,
we've
looked
at
gas
cop,
my
videos
off
we've
looked
at
cask
up,
we
started
our
team
like
and
like
wrote
up
some
design,
Docs
and
started
coding
for
like
a
week
and
then
and
then
you
all
released
cast
off.
So
we
looked
at
them
like
oh
there's,
a
lot
of
the
cask
up
design
matches
our
design.
Pretty
close.
This
is
good,
like
let's
keep
going.
E
E
Do
these
things
side
by
side
and
I?
Think
there
is
some
standard
stuff
that
we
should
try
to
get
ready,
but
it's
hard
to
figure
out
how
we
do
that.
That's
er,
DS
matching
without
specific
behaviors
matching
I
think
we
have
more
tests
written
now,
like
we've
spent
a
lot
of
time
on
testing
and
I
Frank
I,
don't
know,
maybe
there's
more
tests.
You
all
have
that
aren't
in
that
repo,
but
I
did
try
to
go
through
them.
E
F
A
F
F
B
Well,
we
could
we
could.
We
could
compare
the
existing
CR
DS,
see
what's
in
each
and
see
what's
good
in
each
and
say
I
want
I
need
this.
This
so
I
don't
need
this
I
mean
we've
got
with
we've
added
quite
some,
some
stuff
in
it
recently,
because
it
was
requested
by
the
by
the
ops
team
because
they
wanted
other
thing
that
our
plugin
to
do
to
the
operator.
They
wanted
to
add
their
own
little
stuff.
That
was
running
in
parallel,
their
own
sidecars
or
the
place.
B
F
We
have
the
that
that
little
doctors
out
there
now
with
this
EP
content,
should
we
add
a
section
to
that
for
the
various
series
that
we
want
to
describe
and
that
way
we
have
a
way
to
comment
and
discuss
the
various
fields
and
what
their
purposes
are,
that
kind
of
thing
and
converge
into
something.
That's
useful.
B
Personally,
I,
quite
like
the
idea
to
restrict
the
debate
on
the
CRT
itself,
because
that's
that's
the
main
thing
I
mean
that's:
what
the
user
going
to
use
user
going
to
going
to
create
a
CRT
and
just
apply.
Does
it
work
that
it
doesn't
work?
That's
all
they
want
to
know.
So
if
we
agree
on
this
EOD
I
suppose
the
the
rest
is
is
the
rest
is
an.
F
For
what
it's
worth,
that
I've
done
a
lot
of
research
into
other
operators,
CR
DS,
just
to
see
kind
of
the
shape
of
them,
what
they
call
various
fields,
what
a
tee
is
or
or
data
structures
they
use
that
are
built
into
kubernetes
sources,
building
them
themselves,
such
as
like
resources,
conditions
that
kind
of
thing
and,
in
my
opinion,
I
hope,
I'm,
not
alone
in
this
I,
think
having
a
consensus.
There's
this
really
going
to
be
a
solid
foundation
as
we
as
we
build
the
engine
portion.
F
A
G
I
mean
kubernetes
is
when
I
talk
about
this.
You
know,
plug
abilities
is
the
name
of
the
game.
So
it's
that's,
not
I,
don't
think.
That's
a
crazy
idea
of
any
stretch,
but
I
think
at
some
point
that
the
discussion
around
the
yeah,
the
discussion
we
thought
needs
to
be.
It's
gonna
have
to
be
under
this
decisions
made.
F
A
F
G
Yeah,
yeah
and
and
I
and
I
think
for
for
the
edification
from
everybody
who's
not
involved
with
their
respective
projects.
It
would
be
good
as
we
move
forward.
Looking
at
that,
there's
been
discussion
about
the
sidecar,
the
management
sidecar
about
the
rather
than
just
well
that
okay,
where
that's
what
we're
gonna
use
for
my
like
to
know:
okay,
what
are
the
various
components
and
so
forth?
What
are
the
pros
and
cons
of
you
know
a
particular
so
looking
at
different
pieces,
one.
D
Thing
that
might
be
valuable
because
we
run
into
these
kind
of
problems
a
lot,
not
just
good
software,
but
like
in
the
world
in
general,
where
you
don't
have
clear
metrics
to
measure
things
against
each
other.
It's
much
more.
Some
things
are
like
orthogonal
and
completely
unrelated
here
in
the
apples
versus
oranges
situation.
But
here
it's
like
you
could
ask
the
questions
of
okay.
Do
we
all
agree
on
what
good
code
is?
Do
hell
agree
on
what
measures
of
quality
are?
Do
we
all
agree
on
what
measures
of
testing
are?
Do
we
all
agree
on?
D
If
we
have
a
shared
set
of
values
around
how
to
measure
the
value
of
things,
then
we
could
apply
those
measures
to
these
different
code
bases
that
might
give
us
some
more
objective,
insight,
I
suppose,
into
what
starting
point,
because
completely
rationally,
if
you've
got
multiple
groups
of
people
that
have
offered
multiple
things,
each
of
them
is
going
to
say.
I
spent
a
lot
of
time
and
energy
on
this
I
don't
want
to
just
punt
this
thing
to
use
something
else
right.
So
there's
there's
a
piece
of
this
worth
like.
A
And
I
worry
about
this
is-
and
this
is
just
me
being
old
software
guy
in
the
room
is
this
could
descend
into
a
lot
of
not
invented
here
fast,
and
it
could
cause
a
lot
of
rifts
in
the
community.
It
will
cause
an
F
ton
of
rifts
in
the
community
if
it
turns
into
a
civil
war
between
insta
cluster
or
on
a
data
stack
over
one
stupid
commit.
D
And
well
there's
a
very
easy
solution
to
that.
Patrick,
which
is
you
have
the
operator
project.
You
have
multiple
sub
projects
which
are
everybody's
operators
and
you
just
say,
punting
on
having
a
reference
operator,
because
we
don't
have
a
shared
value
system
or
should
be
all
right.
You
avoid
the
Civil
War
at
the
expense
of
it
having
a
more
difficult
entry
point
of
the
community
for
people
that
are
specialists,
so
I
mean
this,
isn't
like
a
we're.
Gonna
define
Hills
to
die
on,
but
I
know.
C
D
D
Maybe
that's
what
I
would
pause
it
and
weave
the
thing
is:
we've
run
into
this,
a
ton
on
the
cassandra
project
itself,
right
like
if
you're
looking
at
consensus,
algorithms
and
you've
got
the
raft
party,
the
Paxos
party
to
eat
Paxos
party,
and,
if
everybody's
just
gonna
stick
their
flag
and
a
hill
and
be
like
I
hate
all
the
rest
of
it.
I
want
to
do
this
thing.
You
never
can
move
like
you
got
to
figure
out,
okay.
What
is
what
is
it
than
we're
trying
to
achieve?
D
A
You
know
it's
like
here.
I'm
trying
to
be
a
good
zoom
citizen
and
get
busted
I
was
what
I
was
saying
is
it
sounds
like
what
we're
trying
to
get
around
not
I,
appreciate
everyone
showing
up.
This
is
the
first
thing
that
we
have
to
do
right
and
no
ping
out
is
okay,
but
it
doesn't
go
move
us
forward,
but
the
CRD.
A
If
we,
if
we
agreed
and
I,
think
you
know,
that's
fine
notice
what
we
do
in
that,
but
I
also
sensing
it
that
we
need
to
have
a
clear
direction
from
the
citecar
project
before
we
embark
on
this,
or
is
that
is
that
dependency
gonna
be
a
problem,
because
if
it
is
then
I'm
happy
to
put
my
energy
towards
making
this,
you
know
trying
to
combine
these
efforts
and
to
move
forward.
I
can.
B
Reply
for
casco
I
mean
we
were
very
happy
to
see
the
sidecar
coming
and
we
didn't
know
it.
So
we
didn't
plan
to
include
it,
but
now
we
have.
It
means
that
next
step
after
the
backup
and
restore
is
made
we
plan
to,
but
it
will
depend
what
happens
with
the
community,
because
we
don't
really
want
to
maintain
our
our
thing
forever.
It's
not
we're
not
in
the
in
this
game.
B
Okay,
we
are
telco,
we're
not
here
to
maintain
we're
happy
to
donate
and
participate,
but
but
if
we
have
I
would
like
as
owner
to
basically
add
the
features
that
you
have
in
the
in
the
data
sacks
sidecar,
because
there
are
things
that
we
need
and
we
were
actually
wanted
to
do
one
and
say:
oh
no,
we're
not
gonna.
Do
it.
We're
just
gonna
use
this,
but
we
need
to
include
that
and
we
haven't
done
it
yet.
So
for
me,
it's
it's
a
real
bonus
to
edit
and
it
should
come
with
the
operator.
G
Here's
a
thought,
there's
a
you
know,
I
just
said:
no
one
wants
to
go
ahead
and
throw
away
the
the
hard
work
and
effort
that
they've
for
them
they've
done
what
what
if
you
have
a
rather
than
somebody
having
to
be
forced
into
making
that
decision.
You
agree
on.
You
know
for
particular
feature
of
pieces.
What
needs
to
be
done
and
start
thinking
about
a
LT
yeah,
a
code
base
from
scratch.
Essentially,
but
it's
bringing
in
you
an
agreement
on
okay,
you
know:
here's
the
here's,
the
library
we're
going
to
use.
G
Here's
we're
gonna,
do
for
writing
our
end-to-end
tests.
Here's
what
we're
gonna!
Do
you
know
here's
what
what
framework
were
used
for
the
just
generating
the
scaffolding,
but
now,
whether
it's
operators
dkq
build
or
whatever
and
and
so
reach
the
agreement
at
that
level?
Well,
yes,
it's
obviously
duplicating
work.
That's
already
been
done,
but
at
least
in
theory,
you'll
be
able
to
iterate
fast
or
somewhat
I
mean
because
people
have
already
blaze
these
trails
and
leave
the
technical
debt
behind
and
you're,
not
forcing
anyone.
D
To
throw
away
work,
it's
a
lot
easier
to
merge
on
smaller
scopes
that
way
where,
for
like
here's
a
southern
feature,
we
have
three
implementation:
let's
merge
the
three
together
and
it's
a
lot
less
from
scratch,
and
it's
a
lot
more
from
like
Big
Bang
merging
the
whole.
Damn
project
is
that's
too
big
of
an
elephant
to
eat,
know
that
this
is
gonna
work,
I,
think
I'd,
advise
ya,
identifying
specific
functionality
and
saying
here's
all
the
different
options
and
their
points.
D
B
Main
worry
I
would
have.
Is
that
made
on
our
side?
We
have
some
features
that
may
not
be
really
only
for
orange.
Obviously
there
would
be
from
ops
people
over
the
place,
but
I
mean
it.
May
it
may
be
a
bit
funny
to
hear
when
we
say
we
want
to
have
specific,
a
specific
side,
cows
for
log
log,
log
analyzing
for
example,
so
they
have
we've
done
that
with
put
volumes
and
our
up
guys
just
popped
up
there,
their
tail
tail
sidecar
there
or
anything
mean
if
you
if
we
have
D.
B
If
we
put
a
PR
with
this-
and
they
say-
oh
no,
so
it's
it's
too
funny.
Your
ideas
are
too
funny.
We
don't
need
that
and
that
then
it
doesn't
work
anymore,
see.
That's
that's
the
ID,
because
I
do
believe
that
we
have
the
big
chance.
We
have
that
we
we
all
on
the
same
base.
We
all
work
from
the
operator
SDK,
so
we
all
do
the
same,
reconciling
their
different
names
from
objects,
but
they're,
basically
the
same
stuff,
we're
doing
and
and
we
generate
community
subjects
and
the
main
things
is
dog
back
abilities.
G
B
E
I
think
that
feedback
and
that
knowledge
forget
who
just
said
this
moves
like
what
John
was
saying.
That's
what's
really
valuable
is
like
what
did
we
try
and
what
didn't
work
and
like?
Why
is
it
so
difficult
for
them,
so
yeah
we're
using
our
operator
in
Astra
where
it's
like
total
automation?
So
it's
hard
there's
there
is
user
feedback,
but
it's
a
different
sort,
and
you
know
we're
also
going
down
like
trying
to
recommend
data.
Sacks
customers
use
this,
but
those.
E
Get
more
feedback
from
that,
so
it's
it
feedback
from
people,
yeah
I've
been
I've,
been
trying
to
solicit
feedback
from
anybody
at
data
stacks.
Like
do
you
want
to
spin
this
up
and
tell
me
even
this,
even
the
kind
of
like
least
knowledgeable
person
about
kubernetes
can
be
like
well,
I
tried
this,
and
this
is
where
I
saw,
and
you
say
oh.
This
is
why
you
got
confused
like.
A
Everyone's
capturing
a
really
important
moment
here
when
it
comes
to
usability,
almost
none
of
us
have
an
important
opinion
in
that
regard,
because
we
we've
really.
We
have
expertise
at
a
different
level,
I'm
thinking
of
so
here's
the
person
that
I
want
to
be
impressed
by
and
the
end
state
of
what
we're
trying
to
accomplish
here,
someone
who
sits
down
for
the
first
time
to
use
Cassandra
and,
if
wants
to
do
it
in
kubernetes
and
finds
it
a
really
simple
and
richest
and
honest.
A
B
H
To
kind
of
expand
on
that
point
a
little
bit,
so
this
kind
of
goes
back
to
what
Kristen
saying
earlier
about
looking
at
other
operators
and
that
end
user,
that
you're
describing
really
isn't
somebody
who's
just
looking
to
deploy
Cassandra.
So,
in
my
case,
though,
you
know
my
DreamWorks,
my
team
were
just
you,
know
five
DBAs,
but
we
operate
fifteen
different
types
of
databases,
and
so
a
important
perspective
for
us
was
how
do
we
have
the
same
tooling
and
the
same?
H
How
do
I
limit
the
number
of
operators
in
the
number
of
things
that
my
small
lean
team
needs
to
learn
so
the
way
that
the
way
that
we
operate
extend
our
database
should
really
reflect
very
closely
how
we
do
Couchbase
or
how
we
do
elasticsearch
or
anything
else
right
so
and
again,
Chris
was
kind
of
mentioning
this.
With
going
with
that
CRD
approach.
First
and
maybe
looking
at
how
do
we
do
things
in
a
more
even.
G
H
A
Ameri
I'll
of
you
man,
I
think
that
is
exactly
where
my
head
is
is
like,
and
so
it's
interesting
I've
been
talking
to
a
few
folks,
Jim
and
I
were
actually
just
talking
about
this
I
actually
feel
like
the
operator
is
only
one
part
of
the
puzzle.
Yeah
I
mean
I.
Think
kubernetes
has
a
lot,
there's
a
lot
of
work
in
kubernetes.
That
needs
to
be
done
to
make
it
work
better.
What
databases
period
well.
B
But
that's
we
actually
I
think
the
the
the
debate
we're
having
about
the
starting
point
is
a
waste
of
time.
Basically,
we
need
to
go
forward,
make
something
work
for
the
users
and
just
make
it
usable
and,
for
example,
our
problem
is
that
we
are.
We
are
running
an
internal
cloud,
so
we
not.
There
is
not
infinite
resources.
There
really
finish
resources
limited
and
that's
where
we
struggle
all
the
time
because
says:
oh
no
accounts.
Can
you
do
stuff?
You
wanted
out?
Oh
no,
you
don't
have
enough
CPUs!
Oh
you
like
some
memory.
B
Oh
no,
the
discs
are
awful.
Sorry,
you
can't
do
that
and
they
say.
Oh.
Why
can't
the
operator
tell
me
beforehand
that
you're
my
schedule
is
not
going
to
work
say.
Well,
you
can't
do
that.
If
you
need
to
do
the
dry
run
thing,
it
just
says:
your
thing
is
really
written
correctly.
It's
gonna
work
you're
just
going
to
stay
pending
for
until
the
the
guy
installs
a
machine,
and
that
takes
six
months
because
try
to
get
a
hardware
these
days
with
a
covet
thing.
B
So
it's
difficult
so
I'm
saying
that,
even
if
guys
that
are
in
the
in
the
cluster,
like
you
I,
like
clusters
and
iteration
of
stuff,
there
is
a
lot
to
add
to
the
operation.
For
it
to
be
useful,
I
mean
people
on
open-source
thing
can
play
with
CR
these
and
match
them.
If
they
don't
want
to
pay
an
extra
I
mean
what
we
want
is
a
nice
UI.
That
said,
oh
I'm
dropping
a
clustering
booty's,
and
that
makes
that
gives
you
a
nice
blister.
B
A
Well,
it
should
work
in
a
variety
environment,
whether
you're,
where
you're
racking
up
your
own
equipment
or
you're,
renting
it
from
Amazon
or
Google
or
Microsoft
or
poly
cloud
or
whoever
it.
It
just
needs
to
work.
So
I'm
I'm
really
happy
about
this
conversations
going
I
think
we're
all
kind
of
like
we're,
starting
to
circle
around
like
progress
to
getting
things
done,
which
is
at
this
point
I'm
hearing.
Let's,
let's
do
the
CRD
first.
So
what
do
we
need
to
do
start?
A
repo
I
mean
another
doc.
B
B
B
A
B
A
I
mean,
let
me
explain
how
the
cep
as
I
understand
it
needs
to
go
this
we
need
to
get
to.
We
need
to
start
creating
things
as
a
part
of
this,
which
is
the
CP,
of
course,
is
an
important
document,
but
there
also
needs
to
be
like
a
combine
like
this.
Is
here's
the
outline?
Here's
the
potential
code
for
it
there
like
we,
we
need
to
put
together
a
package,
and
the
code
is
really
important.
A
C
A
My
site
maybe
consider
just
working
outside
of
the
project.
That's
what
I
was
talking
about
from
the
beginning
is
is
I,
think
we
need
to
do
a
lot
of
work
outside
of
the
project.
The
product,
the
Apache
Cassandra
project,
is
not
a
kubernetes
project,
so
I
think
what
the
work
we're
doing
now
happens
outside
of
the
project.
For
now,.
F
H
And
and
I
think
I
think
that's
the
right
approach.
I
mean
as
far
as
like
targeting
CR
days
versus
two
EPs,
because
we're
talking
about
CPG
really
is
looking
at
defining
functionality
when
you're
coming
at
it
from
a
CRD
perspective,
you're,
really
looking
at
it
from
a
design
architectural
perspective,
which
is
really
probably
how
we
should
be
talking
about
this.
And
secondly,
you
know
if
you
look
at
kubernetes
enhancements
as
well
like
they
start
with
the
design
proposal,
they
start
with
what
would
discuss
some
resource
look
like
before
they
start
getting
into
implementation
right.
E
E
I,
don't
understand
globally,
but
like
what
other
operators
do
for
this
problem,
because
you
know
we
can't
be
the
only
people
adding
a
validating,
webhook
and
I,
don't
want
to
do
it
and
some
like
operator
SDK
has
support
for
it,
but
it's
very
poor
support
and
leaving
you
to
kind
of
document.
How
a
user
turns
this
thing
on
and
like
what
kind
of
privileges
they
need
on
their
cluster
is
kind
of
a
a
big
mess
like
operator.
E
Sdk
honestly,
is
the
more
we've
used
it,
the
more
we
find
it
underwhelming
and
and
we're
working
on
a
another
yeah
banding
casts
ever
but
we're
working
on
another
operator
thing
and
for
something
else
and
we're
looking
at
just
straight
queue
builder.
It
seems
like
more.
The
community
is
going
to
straight
cube
builder,
so
looking
at
that
kind
of
the
kubernetes
community,
I
would
say
would
probably
get
us
to
the
best
solution
with
the
least
amount
of
head-butting
friction
dad.
E
E
When
you,
when
you
are
released
it
operators,
TK,
was
the
best
one
and
we
talked
to
some
people
at
Google.
The
guy
who's,
like
he's
like
the
number
to
commit,
or
whatever
he's
like
the
designer
of
cube
builder,
which
is
really
then
the
designer
a
lot
of
this
stuff
cuz,
all
the
operator
SDKs
on
top
of
cube
builder
anyway,
where
I'm
going
this
is,
he
was
like
oh
yeah,
I
use
cute,
but
it'll
be
good
use,
Cupid
alert,
it'll
be
good.
E
He
works
when
she
ke
and
we
looked
at
queue
builder
and
the
documentation
was
not
good
at
all
and
the
examples
didn't
work,
we're
like
no.
We
made
the
right
choice.
Let's
keep
going
ahead,
then
go
back
six
to
seven
months
later
got
a
lot
better
and
I
think
he
was
speaking
off
of
like
internal
to
Google
version.
I
can't
be
sure
about
that,
but
I.
E
B
E
G
E
Some
kind
of
funny
initiative-
I,
don't
quite
remember,
but
what
I'm
going
with
this
is
like
I
do
think
that
if
we
try
I
think
getting
the
CRT
I
would
appreciate
the
chance
to
talk
about
it
and
go
over
some
choices
and
why
we
did
some
some
things
we
did
and
I.
Don't
if
you
feel
the
same
way
or
not,
and
I
heard
I
hear
what
you
said
before,
Frank
that
you
know
there's
some
stuff,
you
added
where
it's
like
yeah.
E
B
B
We
did
what
you
said:
I
mean
we
looked
at
what
other
operators
were
doing
and
when
they,
when
they
requested
something
else
to
say.
Oh,
they
do
that
for
the
Kefka
thing,
so
we
just
made
some
copy
and
paste
adapted
to
do,
and
we
could
do
it
again
on
another
on
a
double
base
code
I
mean,
let's
be
I
mean
what
we
want
is
in
our
years
time
to
all
of
the
same
operators
and
if
possible,
we
not
care
where
it
comes
from
the
Indian.
A
Anyway,
okay,
so
I
wanna
I
definitely
want
to
make
sure
we
have
some
action
items
in
here
like
okay,
cuz
I
know
what
I
love
meeting
with
all
of
you
like
every
couple
of
weeks,
but
you
know
I'm
also
an
engineer
I
like
to
ship,
but
it
sounds
like
we're
getting
and
but
I'm
also
being
super
careful
about
this
I
don't
want
to
just
come
out
and
say
all
right:
we
got
to
ship
something,
let's
get
it
together.
I
mean
we.
This
is
the
hardest
part.
A
I
think
this
is
probably
one
of
the
harder
open-source
projects
that
have
or
project
things
that
have
worked
rounds
because
there's
just
so
much
existing
code
out
there
usually
yeah.
It
makes
it
a
huge
challenge
and
I'm
happy
that
I'm
really
happy
people
are
showing
up.
I'm
gonna.
Keep
reiterating
that.
Thank
you.
If
we
could
turn
this
into
a
C
or
D
discussion,
great
I'm,
gonna,
I,
think
John,
you
and
I.
Look
why
don't
you
and
I
spend
a
little
bit
of
time
getting
it
like
finding
a
home
for
this
to
be
discussed
like
here.
A
It
is,
let's
discuss
it
just
so.
We
can
we'll
put
the
energy
behind
that.
What
I?
What
I
worry
about
is
we're.
Gonna
start.
You
know
in
order
to
have
the
discussion
we're
using
Google,
but
like
there's
using
the
Cassandra
dev
mailing
list
almost
seems
like
a
bad
idea
right
now,
so
you
got
an
extra
hand.
There
Jim.
B
B
C
C
A
C
A
C
C
G
A
That
and
I'm
just
gonna
say
well
because
the
cameras
rolling,
I'm,
gonna
I
just
hit
record
so
I'm
gonna
go
ahead
and
create
a
slack
channel
for
this,
and
that
will
be
our
gathering
place.
I
will
post
it.
Everyone
could
see
it.
It
will
be
able
we'll
be
able
to
have
more
discussions
as
we
go
along
and
it
won't
be
populating
the
dev
list
which
has
stuff
going
on
these
about
stuff
like
tokens
and
compaction
strategies
in
there.
Well,
no
we're
talking
about
important
stuff
like
urban
eggs.
So
by.
C
B
A
Like
a
brand
is
already
angry
about
that,
what
we're
not
using
Java,
it's
e
time
for
a
new
slack
channel-
hey,
no,
no
I'm,
not
a
Java
fan,
don't
don't
get
me
wrong.
I
know
back
in
the
day,
Perl
guy,
so
we
have
I
think
we
have
this
an
action
item
right
now.
Let's
get
let's,
let's
rally
around
the
CRT
like
now:
okay,.
B
B
I
A
B
A
B
A
Need
to
get
get
us
aligned
up
or
figure
out
what's
going
on
in
the
sidecar
side,
because
there
is
a
there
is
a
dependency
there.
We
should
figure
out
like
because
that
was
brought
up
by
Binet
who's,
doing
the
sidecar
project.
It's
like
hey,
there's
a
lot
of
site
cards
and
operators
and
we
have
a
sidecar
project
and
ok,
let's
talk
about
that
too,
but
yeah
all
right!
A
B
I
A
B
A
B
B
A
E
E
What
Frank
just
said,
like
not
being
a
Cassandra
user,
I,
didn't
use
Cassandra
before
I
started
working
at
data
stacks
and
I'm
still
inside
of
my
first
two
years,
I
think
that's
a
lot
like
what
I'm
gonna
put
your
name
mayor
said
where
it's
like.
If
you
want
to
use
this
in
kubernetes,
maybe
you
shouldn't
have
to
care
you.
Maybe
you
should
know
that
cassandra
has
seed
nodes,
but
the
operators
are
just
like
not
exposed
so
much
of
the
guts
you
know.
So
that's
where
I
think
we're
all.
E
E
But
if
you
said
the
right
best
practices
as
much
as
you
can
any
operator
like
now,
you
have
to
really
go
and
meddle
with
those
underlying
kubernetes
building
blocks,
so
getting
the
CRT
right
and
like
some
cooperation
and
getting
the
feedback
like
you
mentioned,
if
it
is
hard
still
for
your
operations,
team
I'm
curious.
Why?
Because
our
operator
looks
a
lot
like
yours.
So
what
is.
G
E
B
I'm
not
saying
they're,
not
good
thing
sometimes
is
I
mean
our
internal
communities
as
nodes
of
constraints
that
are
not
public
you're
not
allowed
to
have
roots.
You
don't
have
root
access.
You
can't
it's
read-only
stuff,
you're,
not
cluster
wide.
You
don't
have
any
cluster
wide
rights.
You
can't
do
a
cube,
CTL
top.
For
example.
That's
nodes.
B
That's
not
allowed
a
lot
of
thing
that
you
can't
do
so
it's
a
bit
of
a
pain
for
the
users
and
that
if
you
want
to
have
a
multi-tenant,
because
that's
what
we
have
multi-tenant
cluster,
then
it
will
be.
You
have
six
your
data,
so
you
can't
see
what
the
other
people
is
doing.
So
that
means
restrictions
and
that's
you
have
to
have
your
operator
works
in
that
of
environment
as
well.
A
A
Think
a
good
thing
has
we
keep
talking
like
I
said,
thank
you
for
showing
up,
and
let's
just
keep
doing,
that
we
have
some
reinforcing
action
items
to
tackle
if
we
need
to
get
on
a
call
sooner
than
that,
I'm
more
than
happy
to
schedule
that
or
get
it
on
the
board
I
like
the
energy,
though,
let's
keep
it
going
thanks.
Everyone,
okay,.