►
From YouTube: kubeadm operator discussion 2019-10-14
B
B
B
The
plan
is
to
go
and
get
a
first-world
true
to
the
commitment
through
the
coop
Amina
330i
I,
see
in
the
code
that
there
are
some
people
new
to
the
discussion
about
the
crater,
so
I'm
quickly
going
to
to
share
what
we
discussed
at
tribulus
previously
about
this
topic.
So
the
idea
of
the
Cabal
minnow
territory
is
something
that
we
want
to
make
it
a
penny
in
in
Cooperman.
But
there
are
a
lot
of
questions
to
be
addressed.
B
So
up
to
now,
we
created
a
cap
which
is
provisional
just
to
get
the
work
started
and,
and
we
divided
the
logically
the
work
into
part.
The
first
part
is
implementing
something
that
looks
at
the
cluster
configuration
and
detect
the
difference
between
the
solid
state
and
current
state
of
the
day
of
its
configuration
and
then
basically
kick
off
a
sequence
of
action
that
the
pareto
has
to
do
in
order
to
reconcile
the
state,
and
the
second
part
is
the
what
what
they
call
it.
B
The
orchestrator,
which
is
basically
the
piece
of
the
controller
that
will
take
charge
of
such
operation
and
go
on
all
the
know.
The
respect
in
the
sequence
and
make
if
the
things
happen
so
this
first
iteration
is
focused
on
the
orchestrator,
and
we
also
agreed
that
we
go
through
fast
iterations.
So
what
we
try
to
achieve
today
should
not
be
fair,
perfect,
but
bathroom
should
be
a
step
up
free
in
the
right
direction.
B
So
a
little
bit
more
contest.
The
desserts
died
that
were
shared
one
or
two
weeks
ago
in
the
cocoon
of
this
tower.
So
this
was
basically
the
first
proposal
to
Ford
API
to
put
forth
operator
on
the
left.
You
can
see
an
example
of
what
is
a
coupon
be
more
flow.
In
this
case
the
coop
I
mean
worth
upgrade
workflow
and.
B
B
So
basically
the
idea
was
to
make
these
in
true
three
different
object.
The
eye
level
object
is
the
operation
which
more
or
less
match
with
the
cuban
min
overflows.
Then
we
have
task
group,
which
is,
we
should
be
a
sort
of
task
template
that
should
be
applied
to
a
group
of
nodes.
For
instance,
I
want
to
apply
this
group
of
tasks
to
the
contraband
machine.
B
The
major
concern
that
we
had
was
that
this
was
two
type
of
concern
on
the
first
days
that
this
API
is
trying
to
make
somehow
something
which
is
imperative
our
flow
into
an
IP
I,
which
is
declarative,
which
is
something
that
we
have
to
deal
with
and
the
second
part.
The
second
comment
was
to
the
fact
that,
with
this
API
we're
exposing
a
lot
of
detail
about
the
/
other
/
Oteri's
working.
C
B
B
B
B
Only
one
of
them
can
be
specified
and
each
subtype
Israel
is
a
descriptor
of
a
quad
mean
action.
The
prior
the
prodigal
prior
art
for
ease
is
the
is
our
volumes
are
defined,
so
no
new,
we
cover
native
and.
C
B
B
B
D
B
They
basically
are
in
full
control
to
a
scenario
where
the
operator
will
take
care
of
everything
so
I
the
gate
is
not
simple,
so
I
think
that
we
should
offer
some
way
to
the
user
to
to
counter
what
operator
is
doing,
that
the
simplest
option
is
to
offer
our
opposed
filter
that
can
be
set,
so
the
operation
cost
immediately
stops
or
Rossetti
and
operational
start.
This
is
again
nothing
new
in
kubernetes.
B
B
Alto,
which
is
by
the
fool
Oh,
fixing
all
the
step,
one
in
a
given
sequence:
drivin,
so
just
just
make
a
dry
run.
So
we
see
that
everything
that
at
least
the
setup
is
is
properly
done,
but
without
doing
damage
and
control
and
control.
Basically,
what
we
do
will
force
the
operation
to
automatically
pause
after
each
task,
so,
for
instance,
in
a
when
doing
a
grade
I
to
the
first
node,
then
I,
passer
and
I
wait
for
the
operator
to
tell
me
to
the
star
to
be
falling
before
moving
to
the
next
node.
A
B
B
B
F
So
pause
it
so
you
you
mentioned
that
this
is.
This
is
taken
from
another
deployment
API,
yes
right,
so
in
employment,
API
post,
the
post
boolean
is
used
for
posing
changes
in
the
spec.
So
when
you
make
changes
in
the
deployment
spec,
you
need
in
a
post
deployment.
Those
changes
you
cannot
be
applied,
they
will
apply
the
only
ones.
You
want
positive
moment.
B
F
B
B
B
I
also
think
that
we
need
properly
to
manage
invalid
groups,
because,
when
in
kubernetes,
when
you
manage
a
guarantee
of
objects,
may
be
that
the
user
do
something
the
world
and
you
get
to
manager
or
funds
or
some
inconsistencies
and,
in
my
opinion
operator,
should
stop
when
we
need
it,
they
delegate
such
conditions,
but
this
is
resistivity.
It
is
really
depend
by
the
implementation.
B
B
E
D
E
Spec
eighty,
so
what
happens?
If
do
we
have
to
handle
this
scenario?
Somebody
pulses
and
then
changes
the
descriptor
or
the
execution
mode,
or
something.
B
E
B
E
B
B
B
So
that
first
Rupa
definer
a
task
that
template
that
apply
to
a
group
of
north.
So
it
is
a
somehow
it
is
similar
to
deployment.
The
big
difference
is
that
you
have
to
apply
your
task
to
a
group
on
node
and
this
group
of
know
that
it
is
basically
water
should
be
changed
by
this
operation.
So
as
part
of
a
task
group
that
the
first
field
is
an
unknown
stalactite
and
these
are
lost
to
specify
which
node
you
have
to
should
be
impacted
by
this
group.
B
B
E
B
B
C
B
C
B
C
B
What
we
ever
is
move
to
create
strategy
play
strategy
again
is
a
numerator.
Now,
it's
port,
only
sequential
that
dense
mean
that
basically
I
am
going
to
create
task.
One
know
the
time.
So,
let's,
for
example,
I
needed
to
upgrade
ten
workers,
and
now
I
have
only
the
possibility
to
upgrade
worker
1,
then
worker,
2
and
so
and
so
on.
Maybe
in
the
future
we
want
to
allow
something
more,
let
me
say,
aggressive
upgrade
or
the
worker
in
parallel
or
something
even
more
smarter
upgrade
like
disruption,
so
I
upgraded
maximum
three
three
nodes
at
time.
B
B
Exactly
exactly
HR
comics,
serger
and
sameiras,
so
this
is
a
place
order
for
implementing
such
such
strategies.
Okay.
Finally,
this
is
the
the
part
where,
where
we
are
creating
tasks
to
be
recruited
on
the
node,
this
is
the
task
template,
and
this
is
the
selector
that
we
use
to
search
the
tasks
which
are
managed
by
this
controller.
B
And
finally,
we
have
the
status
for
the
task
group,
which
is
really
similar
to
the
action
status
start
time
paused
in
this
case
it
reflected
the
operation
puzzle.
So
when
I
stop
the
peroration
eyebrows,
all
the
tasks,
all
the
deadlines
and
task
groups,
completion
stage
and
then
the
counter,
so
the
list
of
nodes
which
are
targeted
by
vistas
group,
how
many
of
them
for
how
many
of
them?
There
is
a
running
task
to
see
that
one
faded
and
again
I've
added.
B
For
instance,
let
us
assume
that
the
user
changed
man
will
accept
some
tasks,
so
I
have
a
task
that
don't
match
and
don't
tell
the
upper
the
expected
level,
and
so
this
is
mostly
the
case
that
you
face.
For
instance,
when
you
implement
something
like
machine
controllers,
my
machine
deployments
or
bother.
B
B
B
B
Why
more
than
one
command?
Because
typically,
when
I,
for
instance,
when
I
do
a
creator,
I
go
to
an
order
and
then
I
have
to
execute
to
free
step
on
the
same
node
before
moving
on
the
next
one
and
for
the
common
again
we
have
a
common
descriptor
which
is
similar
to
the
operation,
ascription
descriptor,
so
a
type
with
a
lot
of
subtypes
all
optional.
Only
one
of
them
can
be
specified.
C
Hey
Fabrizio
I
have
a
question
about
the
list
of
commands
running
on
a
node.
Yes,
is
there?
Are
there
any
workflows
where
commands
have
to
be
run
across
nodes
in
a
particular
order?
And
if
so,
just
you
know
what
I
mean
like
there's
a
dependency
where
I
run
a
command
on
node
one
and
then
I
command
under
two
and
then
a
command
back
again?
On
the
note,
one.
B
E
B
D
B
B
B
B
C
B
B
B
E
E
B
B
E
B
Ok,
so
then,
what
what
we
ever?
We
have
the
status
again
start
time,
completion
time
positive,
which
is
the
reflection
of
the
operation
pulse.
We
have
the
current
commander,
so
I
have
a
five
command
I'm
executing
commander
one
two,
three
or
four
and
command
progress
which
is
I
added
these,
because
this
is
a
textual
representation.
One
of
five
one
of
of
twelve
five
is
just
for
a
better
visual
representation
of
status.
I
will
not
comment
later.
E
A
E
C
B
C
B
Yes,
exactly
so,
basically,
there
are
three
controllers,
which
are
the
operation
controller,
the
task,
loop
controller
and
the
task
controller.
So
the
task
controller,
the
first
two
with
this
applied
as
usual
as
a
deployment,
and
so
that
it
basically
exists
only
one
one
instance
of
those
controlling
the
cluster
or
two
in
case
you
two
or
three
in
case
your
redirection,
why?
The
the
task
controller
will
be
deployed
at
runtime.
So
when
an
operation
start
I,
look
at
the
nose
word.
B
B
D
B
B
E
Yes,
but
imagine
that
you
have
a
lot
of
loads
in
only
a
few
control
plane
nodes.
You
can
upgrade
all
the
control
plane
notes
like
let
the
first
task
and
then
you
might
decide
like
hey
our
our
board,
upgrading
the
workers
completely,
but
the
worker
upgrade
already
started
so
should
it
be
like
a
menu,
keep
corrode
delete
operation
or
should
we
add
some
sort
of
a
mechanic
to
abort
the
process.
I.
B
B
B
B
A
E
So
go
back
to
the
like
is.
The
major
concerns
is
that
we
are
exposing
too
much
granularity
and
what
people
are
looking
at
is
something
that
we
should
have
a
going
to.
Your
first
slide
like
a
a
different
approach.
That
is
the
declare
fully
declarative
approach
with
you
feed
an
object
that
basically,
the
coastal
configuration
of
cue
medium,
and
if
you
modify
this
coastal
configuration,
the
operator
is
going
to
figure
out
things
itself
and
perform
the
action
which
is
like
quite
complicated
to
do
right
now.
B
Even
if
we
we
designed
this
higher-level
logics
that
decide
which
list
of
operation
you
have
to
perform,
then
is
still
old.
The
need
to
perform
this
operation
so
some
sort
of
orchestration.
You
need
some
sort
of
the
orchestration
and
the
orchestration
is
complex
and
you
cannot
fit
everything
in
the
status
of
I.
Don't
know
of
the
cluster
configuration.
B
Leave
it
because
one
idea
that
I
consider
at
the
beginning
is
to
have
only
the
operation
object
and
keep
the
list
of
tasks
of
notes
in
the
status
of
of
this
object,
but
it
it
was.
It
is
almost
unreadable
as
soon
as
a
number
of
node
accrues
and
also
it.
It
does
not
make
a
good
the
separation
of
concern,
because
now
I
have
a
operation
which
is
basically
responsible
to
to
lay
down
the
list
of
tasks
groups,
and
this
is
immutable,
so
it
does
not
change
by
cluster.
He
knows
that
for
doing
upgrade.
B
You
have
to
do
three
task
groups.
That's
it
then
I
have
task
groups
that
basically
is
responsible
to
look
at.
My
cluster
choose
see
how
many
nodes
I
ever
and
created
tasks,
and
then
I
have
the
task
controller,
which
is
the
agent
that
goes
to
the
node
so
by
by
splitting
the
tree
operation.
The
the
problem
of
the
orchestration
in
is
free,
CRT
object.
I
got
an
easier
separation
of
concern,
even
if
I
I'm
getting
in
some
problems
like
embodied
objects.