►
From YouTube: Making Plugins Mainstream: Developing a Plugin Manager for Kubectl - Ahmet Alp Balkan, Google
Description
Join us for Kubernetes Forums Seoul, Sydney, Bengaluru and Delhi - learn more at kubecon.io
Don't miss KubeCon + CloudNativeCon 2020 events in Amsterdam March 30 - April 2, Shanghai July 28-30 and Boston November 17-20! Learn more at kubecon.io. The conference features presentations from developers and end users of Kubernetes, Prometheus, Envoy, and all of the other CNCF-hosted projects
Making Plugins Mainstream: Developing a Plugin Manager for Kubectl - Ahmet Alp Balkan, Google
Kubectl Plugins are an extension mechanism to add custom commands to Kubernetes developers’ favorite tool: kubectl. As the plugin mechanism became stable recently in Kubernetes 1.12 and people started to develop plugins, we realized a challenge is upon us: How to make these plugins easy to develop, package and distribute. This talk is a Kubernetes SIG CLI subproject "krew" and how we have designed and built a "cross-platform package manager" that makes plugins accessible to vast amount of Kubernetes users easily. This talk will go through: * Kubernetes project’s emphasis on extensibility * Introduction to kubectl plugins and writing a small plugin * Designing Krew: how to write a plugin manager for kubernetes without writing a fully-fledged package manager * Life of a kubectl plugin managed by Krew * Challenges of building a curated catalog of plugins as Kubernetes OSS community
https://sched.co/Uab4
A
So
welcome
everyone
and
excuse
me
hello
to
everyone
from
YouTube,
so
we're
gonna
talk
about
make
cube
control
plugins
today,
but
first
of
all
who
had
tacos
in
San
Diego.
Can
we
have
a
quick
show
of
hands?
Okay
for
those
who
didn't
eat
a
taco,
you
should
probably
go
eat
a
taco
before
you
leave
okay,
so
we're
gonna
talk
about
cube
control
today,
as
I
pronounced
it
throughout.
Throughout
this
talk,
you're,
probably
gonna
hear
different
pronunciations,
that's
my
subconscious.
A
Soul
consciousness
kicking
in
so
you
know,
cube
control,
as
you
know,
is
probably
the
single
most
important
client
of
kubernetes.
This
is
how
developers
interface
kubernetes.
Most
people
do
not
use
API
directly.
Most
people
do
not
use
the
dashboard
directly.
This
is
the
main
interface,
so
cute
control.
So
a
little
bit
about
myself,
you
can
find
me
on
Twitter
and
github
as
a
VP
in
the
past
I've
done
certain
open
source
projects
like
I,
worked
on
porting
docker
to
Windows
some
parts
of
it.
I've
been
a
doctor,
maintainer
and
I
started
as
a
container
registry.
A
So
recently,
I've
been
at
war
advocate
Google
cloud
I'm.
Still
there
I'm
working
on
service
stuff,
like
Google
cloud,
run
and
Kay
native,
if
you're
interested
about
that,
we
can
chat
about
that
too.
But
overall
I'm
looking
at
open
source
developer
tools,
landscape
so
relevant
to
this
talk,
I've
actually
developed
some
Q
control,
plighted
tools
that
you
might
be
using
out
there,
which
is
like
QC,
txq
venez.
There's
this
weird
thing:
I
think
that
was
demoed
on
the
lightning
talks
on
Monday.
A
There
was
puke
control,
Alice's
and
I've
developed
the
control,
player-manager
crib,
so
I
maintained
that
project
actively
and
we're
gonna
be
talking
about
a
little
bit
about
that
today.
So,
first
of
all,
I've
structured
to
talk
into
two
like
we're
gonna
talk
about
using
plugins
as
a
user
and
using
plugins
as
a
developer.
So
the
agenda
is
basically
what
are
keep
control
plugins
and
like
what
is?
A
How
would
you
use
a
plug-in
manager
to
get
plugins
from
the
internet
and
why
did
we
need
plugins
in
the
first
place
and
then
in
the
second
half
of
the
talk
we're
gonna
talk
about?
You
know
maybe
like
hands
on
developing
your
plugin
and
then
we're
gonna.
Look
at
the
plugin
runtime
how
plugins
are
executed,
and
you
know
why
you
should
consider
developing
a
plugin
and
how
to
distribute
plugins
and
if
we
have
time
which
I'm
not
sure
we're
gonna
get
to
best
practices.
A
Okay,
so
who
here
actually
has
used
the
control
plug
in
before,
can
I
see
her
show
a
fancy?
That's
awesome!
That's
awesome!
So
you
know
cube
control
plugins
in
a
nutshell,
in
its
current
form,
it
spin
around
C's,
Perez
112,
and
you
know
it's
an
extension
mechanism
that
lets
you
build
a
nuke
sub
command
to
keep
control.
Essentially,
so
here
you
know
either
I
have
the
skill
of
the
world
command
called
my
command,
and
it
does
funny
things
like
printing
printing
colorful
text
right.
A
So
in
a
nutshell,
how
the
mechanism
under
the
covers
works,
is
it's
pretty
simple.
You
write
a
program
just
like
you're
writing
any
program
in
any
programming
language
right.
You
get
an
executable
at
the
end
and
you
name
it
cute
controlled
as
something
so
that
Q
control
prefix
is
the
key
right
and
then
you
put
it
somewhere
on
your
path.
Environment
variable.
A
You
know
you
just
make
sure
your
computer
can
find
it
when
you
type
the
thing
and
then
you
can
invoke
it
as
cube
control
space,
something
so
at
that
point
that
becomes
a
cube
control
command.
So
that's
pretty
easy
right.
So,
as
a
user,
you
might
be
wondering
world
I
find
plugins.
So
there
are
a
lot
of
answers
you
might
find
them
on
github
right.
A
You
can
search
for
cube
control,
plugins
on
Google
and
you
can
land
to
a
plugin
now
I'm,
not
sure
if
you're
gonna
find
all
of
them,
there's
no
guarantee,
and
there
are
lists
on
the
internet.
That
lists
you
know
the
cube,
control
plugins
that
are
known
to
the
humanity
and
possibly
there's
maybe
a
central
registry
through
some
package
manager-
and
this
is
where
crew
comes
in
so
crew
is
a
plugin
manager
for
cube
control.
A
A
It
lets
you
discover:
plugins
install
plugins,
keep
the
plugins
up
to
date,
which
is
pretty
important
and
then
remove
the
plugins
cleanly.
So
if
you
actually
have
downloaded
plugin
from
a
random
place
on
the
internet,
you're-
probably
not
keeping
it
up
to
date
and
you're
not
able
to
like
greenwood
cleanly,
as
you
would
think
so,
crew
actually
started
as
a
as
an
interim
project
in
2018
I
had
an
intern
named
loop
orchard
at
Google
cloud.
A
So
we
did
this
as
a
totally
fun
project,
but
we
had
the
idea
for
some
time
and
I've
been
chatting
some
sick,
CLI
folks
about
it
and
we
brought
it
to
life,
and
then
we
started
to
get
more
plugins
so
turns
out.
People
actually
want
to
use
something
like
this,
and
earlier
this
year
we
have
donated
the
project
with
CLI.
It's
now
governed
under
60
Li,
and
there
are
maintained
errs
that
are
totally
enough
from
Google.
So
that's
the
four
world,
so
essentially,
crew
is
not
an
actual
package
manager.
A
If
you
think
oh,
this
guy
totally
developed
a
new
package
manager,
that's
the
that's
the
worst
idea,
you're
totally
right.
We
did
not
move
any
package
manager
because
that's
a
really
hard
thing
to
do.
Soku
is
actually
like
a
glorified
batch
script
that
lets
you
download,
plays
from
the
internet,
keep
them
up
to
date
and
delete
them
so
that
there's
not
a
whole
lot
more
than
that.
So
it's
actually
a
pretty
simple
project.
A
It
doesn't
handle
dependencies
if
your
plugin
has
dependencies,
users
should
install
them
separately,
so
number
of
control,
plugins,
open
crew
today
is
actually
over
60.
So
this
chart
to
me,
I
pulled
out
last
night
is
pretty
impressive,
because
every
time
there
is
a
queue
come
with
double
our
plugin
count
and
it
doesn't
seem
like
we're.
Gonna
stop
anytime
soon,
so
that's
fun.
So
in
the
last
cube
con
in
EU,
we
actually
had
something
like
30
plugins
and
the
previous
one.
When
we
first
introduced
crew
in
Seattle,
we
have
like
fifteen
or
sixteen
seventeen
plugins.
A
Something
like
that.
So
it
seems
like
this.
Trajectory
is
growing.
So
if
I
were
to
draw
a
trend
line,
you
can
see
that
it's
going
for
a
up
and
we're
sticking
to
the
trends
so
keep
the
looking
plugin
plugins,
please
so
I
want
to
give
a
hands
on
try
to
crew
and
some
keep
control
plugins
right
now.
So
here
I'm
gonna
switch
to
my
demo.
A
Okay,
alright,
let's
bring
up
a
terminal
cool
okay,
so
you
know,
as
you
might
know
you,
you
need
to
install
crew
first
of
all,
so
crew
is
this
package
manager,
you
plug-in
manager,
I,
don't
say
package
manager,
but
here
it
says,
package
manager.
So
I
can't
lie
to
you
about
that.
Sorry
about
that,
so
you
installed
crew
one
time,
there's
a
command
that
you
run
to
install
we're
trying
to
find
a
better
alternative
to
this.
A
But
it's
actually
there's
a
weird
install
here
that
is
not
like
through
homebrew
or
something
like
that,
which
is
a
OS
Native
package
manager
is
because
crew
itself
is
a
plugin
that
manages
that's
managed
by
crew,
so
crew
updates
itself,
just
like
any
other
plugin.
Okay,
so
we're
gonna.
It's
already
installed
it
so
that
we
don't
run
into
conference
Wi-Fi.
A
So,
let's
see
so
I've
installed,
crew
I'd
have
cube
control
crew,
as
you
can
see,
it's
a
plug-in
itself
and
it
has
all
these
sub
commands
right.
So,
as
a
user,
if
I
wanted
to
find
a
plug-in,
I
would
type
Q
control
crew
search.
You
can
see
that
all
right
right,
all
right.
So
when
I
type
this
I'm
gonna
resize
it.
As
you
can
see,
there
are
a
lot
of
plugins
here.
If
you
wanted
to
install
something
from
this
list,
all
you
need
to
do
is
just
copy
the
name
here.
A
Let's
install
the
first
one:
excess
matrix
right,
cube,
control,
crew,
install
xx,
matrix
all
right,
so
right
now,
what's
happening
is
crew
is
going
to
the
internet
now
Ward
downloading
the
package
looking
at
the
package
manifest
and
just
you
know
essentially
unzipping-
that
package
I'm
printing
here
some
instructions
that
the
plugin
author
has
created.
You
know
it
says
to
use
this
plugin
around
this
command.
That's
pretty
simple!
Okay,
so
reinstall
cube
control
access
matrix
so
now
we're
executing
it
as
a
plugin.
So
this
is
a
binary
that
is
right
now
in
my
path
right.
A
So
this
is
gonna.
Take
a
while
I
assume
because
of
the
internet,
but
essentially
what
this
plugin
does
is.
It
shows
you
which
API
groups
you
have
permission
to
operate
on,
and
it
seems
like,
as
a
user
I
have
I'm
the
administrator,
so
I
have
permission
to
do
everything.
So
this
is
a
functionality
that
I
did
not
everyone
keep
controlling
for
and
now
I
have
this
functionality
so
cube
control.
If
you
wanted
to
see
like
all
the
plugins
you
have
you,
you
could
actually
run
control,
plugin
lists,
you
know,
I
have
some
plugins
here.
A
As
you
can
see,
crew
itself
is
a
plugin
access.
Matrix
is
a
plug-in,
and
you
might
note
that
there's
actually
a
path,
some
path
here
that
I
added
to
my
environment
variable
earlier,
so
that
this
is
where
the
plugins
are
installed.
This
is
where
how
your
computer
actually
finds
them?
Okay,
okay!
So
what
do
we
have
here?
So,
if
I
wanted
to
delete
that,
for
example,
I
can
do
like
cube
control
crew
upgrade
so,
let's
see
if
any
of
the
plugins
actually
have
updates.
A
Okay,
it
seems
like
a
plugin,
has
an
update,
so
we're
actually
downloading
an
update
right
now.
I,
don't
know
how
long
it's
gonna
take
on
the
conference
Wi-Fi,
it
seems
like
I
upgraded,
a
plugin
called
Who
am
I,
that's
cool,
so
just
like
that.
I
get
that
package
manager
experience
without
having
to
influence
a
package
manager
fully
right.
Okay.
A
So
let's
go
back
to
the
slides
okay,
so
why
did
we
need?
Could
control
plugins
in
the
first
place?
You
might
wonder.
First
thing
people
want
to
add
new
colonize
the
cube
control
Oh.
All
the
time.
Look
at
all
these
issues,
I,
don't
know
pulled
a
mute
from
them,
so
feel
free
to
take
your
time-
and
you
know,
look
read
actually
the
titles
from
some
of
these
issues.
They're
all
valid
they're,
all
essentially
saying
I
have
this
functionality
that
it
seems
like
you
know,
we
should
have
definitely
having
tube
control.
So
why
don't?
A
We
add
it
to
the
queue
control
core
directly,
so
this
makes
perfect
sense.
Right
and
second
thing
is
a
lot
of
people
have
opinions
about
how
things
should
be
done
right.
You
can't
help
it.
This
is
all
full
source.
So
look
at
this.
This
is
a
bunch
of
tools
that
are
developed
around
cube
control
that
the
one
thing
they
feel
kubernetes
logs
for
I
do
have
all
these
tools.
Actually,
these
are.
This
is
an
incomplete
list.
This
is
only
the
projects
that
have
like
more
than
hundred
stars
that
do
this.
A
There
are
a
ton
more
that
you
know
has
less
than
fewer
than
hundred
stars
that
do
the
same
thing.
There
are
100
tools
out
there
to
tail
kubernetes
locks
because
everyone
thinks
they
should.
We
should
do
it
differently
right
and
the
third
is
reaching
consensus
in
the
open
source
is
not
trivial.
This
is
this
is
a
big.
It's
not
an
issue,
it's
just
by
design
we're
trying
to
make
something
work
for
everybody
right.
A
So
here's
an
issue
that
my
colleague
at
Microsoft
actually
called
Macon's
opened
in
2016
I
was
there
at
Microsoft
and
Cole
complained
this
about
this
issue.
To
me
many
times
he
says
why
don't
we
make
an
environment
variable
so
I
can
set
my
current.
You
know
my
current
context
and
name
space
to
an
environment
variable
in
my
shell,
so
that
I
don't
have
to
type
context
such
name
space
all
the
time.
It
makes
perfect
sense,
so
this
issue
was
opened
in
2016.
A
When
did
you
think
we
closed
this
issue
like
one
year
any
hands
two
years
three,
three
years
more
than
three
years?
That's
correct!
Okay!
So
we
closed
this
issue
in
August
this
year,
and
you
know
there
are
all
the
support
is
there
there
are
all
these
uploads
all
these
hearts,
nothing
negative
about
it
and
that's!
Okay!
You
know.
Maybe
this
is
not.
You
know.
We
can't
do
everything
in
the
core
and
that
makes
sense.
We
can't
just
keep
adding
environment
variables
right.
A
We
can't
keep
adding
more
surface
and
because
we
have
to
maintain
it
forever
right.
So
you
see
a
problem
here.
I
see
an
opportunity,
I
wrote
a
tool
called
QC.
Tx
is
one
of
the
most
popular
tools
in
the
Q
control
ecosystem
and
that's
fine.
You
know
like
if
people
want
to
have
a
different
experience.
Maybe
there
should
be
a
way
for
people
to
express
themselves
differently
and
we
cannot
solve
everything
in
kubernetes
core
and
that's
perfectly
fine
too.
The
core
is
meant
to
be
for
everybody
right.
A
We
if
we
try
to
solve
everything
in
the
quarry,
would
not
be
quarry
anymore.
We
do
pretty
much
everything
right.
So,
for
example,
let's
take
a
look
at
you
know
the
coronaries
stack
so
essentially
at
the
bottom.
You
have
the
notes.
Right
notes
have
applicability
model,
you
install
storage,
plugins,
you
install
network,
plugins
and
then
higher
you
go
in
the
stack.
You
can
install
extensions
to
kubernetes
itself.
You
can
develop
controllers,
you
can
develop
extension,
API
servers.
You
can
write
your
own
CRTs.
A
You
can
write
Admission
web
books,
you
can
write
I,
don't
know,
authentication
authorization
plugins
into
the
API
server
and
you
know
on
the
top.
There
is
cube
control.
So
why
not
keep
the
same
accessibility
model
go
on
to
the
client
side
as
well
right.
So
this
is
why
what
we
have
a
few
control
plugins
and
the
LAN
I
think.
Another
reason
is
curious.
Core
moves
slower,
I
think
this
is
a
big
frustration
that
you
know
when
you
join
the
kubernetes.
A
You
know
the
community,
you
realize
all
my
PR
takes
forever
to
merge
or
something
like
that,
and
that's
ok.
So,
for
example,
if
you
were
to
write
an
official
command
to
cube
control,
you
would
write
a
proposal.
First,
you
will
get
the
proposal
approved
and
that's
ok,
you
would
write
you
know
you
would
prove
that
your
program,
your
command,
is
actually
useful
and
it's
stable
and
it
works
and
people
definitely
want
it
right
and
then
you
have
to
write
it
in
code,
go
only
so
that
it's
hosted
in
the
kubernetes
codebase.
A
So
there
are
these
sort
of
restrictions
and
you
will
be
tried
to
coronated
release
cycle.
You
have
a
bug,
you
got
away
for
a
release
and
then
you
know
it
has
to
be
consistent
with
the
rest
of
cube
control.
You
can't
just
invent
like
weird
named
flags,
and
you
know
you
can't
print
emojis
in
the
command
line.
You
know,
but
that's
not
a
thing
to
do
and
then
essentially
a
new
command.
It
takes
like,
sometimes
years
to
stabilize,
like
you,
control
plugins
mechanism
actually
took
over
a
year
to
stabilize.
A
If
you
you
got
to
be
patient,
you
got
to
put
you
know
your
heart
into
it,
to
make
it
happen,
but
on
the
plug-in
side
there
are
no
restrictions.
You
can
go
home
right
and
pull
again
today,
distribute
it
like
I
did
on
several
occasions
and
then
you're
free
in
ten
minutes.
You
can
have
a
public
plugin,
that's
used
by
everyone,
you
know.
So
this
is
another
reason
and
then
sixth
reason
is
pretty
interesting.
People
are
actually
creative.
A
You
know
this
was
not
that's
wide.
Yeah
I
was
trying
to
show
this.
So
you
know
people
write
plug-ins
that
have
I,
don't
know
like
weird
I
know:
previews
and
people
create
tables
on
the
command
line
and
people
they
should
take
it
a
step
further.
There
are
your
base
plugins.
Now
there
is
a
project
called
cui,
which
is
K
UI.
It's
trying
to
create
we
based
plugins.
So
this
is
pretty
experimental,
but
you
know
that's
the
whole
point.
We
actually
want
to
allow
safe
experimentation
without
hurting
the
core
itself
right.
A
So,
for
example,
there's
a
command
there's
a
there
has
been
several
plugins
that
are
trying
to
solve
the
debug
problem.
You
know
we
want
to
have
a
debug
sub
commanding
cube
control.
Maybe
there
has
been
proposals,
someone
added
like
a
ephemeral
container
support
to
the
API
and
the
node,
and
then
you
know
they're
trying
to
figure
out
there
should
be
a
cube
control
sub-command.
So
there
has
been
proposal
getting
kept
proposals
out
there,
but
essentially
to
make
sure
we're
doing
the
right
thing.
A
We
need
to
get
a
lot
of
user
input
and
this
is
the
plugins
is
where
it's,
but
it
can
happen,
you
can
write
a
plugin
put
it
out
there
and
then
you
would
get
user
feedback
immediately.
In
fact,
there's
I
think
maybe
I
heard
from
someone
there's
a
cube
control,
debug
plugin.
That
will
show
we
showed
on
the
keynote
stage
today
or
something
like
that.
So
you
know
we
want
to
let
people
experiment
and
we
want
to
graduate
successful
ones.
You
want
to
get
to
ideas
and
figure
out.
A
What's
the
right
thing
to
do
and
then
move
on
to
the
next
thing.
So
let's
talk
about
developing
plugins,
you
know,
I
talked
about.
You
know
how
to
write
a
plugin
earlier.
You
need
to
write,
you
know
your
application
and
then
you
need
to
make
it
executable
name.
It
cube
ctrl,
something
and
then
put
it
in
your
path,
and
then
you
need
to
invoke
that
as
cube
control
space,
something
so
this
is.
A
You
know
in
a
nutshell,
how
plugins
work
so,
let's
developer,
playin
I'm
gonna
do
that
in
a
second,
but
essentially
I
want
you
to
check
out
this
repository
sample,
CLI
plugin.
This
has
a
pretty
good
go
repository
with
most
things
already
set
up,
all
the
command
line,
flags
handling,
cube,
configure
environments
variable
and
most
of
the
things
are
already
in
place.
Okay,
so,
like
you
should
just
fork
it
and
try
developing
your
plugins
with
that.
So
here,
I'm
gonna
do
something
different
I
think
I
want
to
write
a
bash
based.
A
Plugin
bash
is
one
of
my
favorite
languages
for
sure.
So,
let's
see
ok,
so
right
now,
I'm
in
this
directory,
I
think
I
will
create
a
play
called
kid
control.
What
should
we
call
it?
Can
we
call
it
tacos?
Is
that
cool?
Alright,
let's
call
it
travel,
okay,
keep
control,
taco,
okay
and
I'm
going
to
say
you
know
this
is
a
batch
program,
user
banana.
What
is
my
syntax
highlighting
not
working
terrible?
Okay?
Here
we
go
and
then
I'm
gonna
say
hey,
let's
write
the
executable
name
like.
A
What's
the
you
know,
the
program
is
you
know,
dollar
sign
zero.
It's
passed
to
your
application.
Is
that
and
then
your
argument,
let's
print
them,
and
then,
let's
see
what
we
have
here.
So
next
thing
is
I.
Make
this
thing
executable
right,
I,
say
mod
+,
eggs,
I,
say
cube
control
this
and
that's
the
next
thing.
I
need
to
actually
add
this
current
directory
to
my
path,
so
I
do
export
path,
equals
current
working
directory
and
rest
of
the
path
cool.
A
So
let's
do
cube
control
plugins
lists
this
plugin
lists
cool
I,
see
the
cube
control
taco
here.
So
what
am
I
gonna
do
next?
Can
anyone
guess,
yes,
I'm
gonna
type
cube
control?
Taco,
let's
see
Q
control,
taco!
Okay,
it
worked.
It
seems
like
that's
the
program,
that's
the
argument.
If
I
pass
like
things
into
it,
like
I
love,
tacos,
okay,
see
that's
the
argument
that
I
get
right.
That's
cool!
That's
awesome!
Alright!
So
let's
do
something
more
interesting.
Now,
let's
say:
I
want
my
plug-in
to
print
nice.
A
Colourful
output
to
the
command
line,
so
I'm
gonna
do
I
was
a
cold
cube
control
with
the
rest
of
the
arguments,
but
pipe
the
result
to
this
program
called
locat
which
actually
makes
it
fun.
Okay.
So
let's
try
cube
control
tacos.
Now,
maybe
control
get
namespaces.
How
about
that
you
control
taco
get
namespaces
there
we
go!
That
looks
fun
right,
so
you
can
do
stuff
like
this
using
plugins
system
fun.
Alright,
I'm
gonna
go
back
to
my
slides.
So
what
happened
right?
There
was
essentially
when
I
type
the
command
called
queue
control.
Abc.
A
The
first
thing
happens
is
the
queue
control
needs
to
know.
Where
is
the
plugin?
So,
first
of
all,
if
a
in
this
example
is
a
built-in
command,
that's
not
gonna
be
treated
as
a
plugin.
Okay.
If
you,
if
you're
trying
to
write,
you,
know
override
the
existing
subcommands
of
queue
control,
that's
not
going
to
work
and
there's
a
good
reason
for
that,
namely
security
and
other
things.
So
the
next
thing
that's
gonna
happen
is
queue
controls
trying
to
find
the
player
name
cube
control
ABC,
because
that
might
be
a
single
plugin
right.
A
You
can
actually
write
your
sub
comment
of
the
sub
commands
as
a
plug-in
as
well
like
you,
don't
have
to
have
one
executable.
You
can
have
multiple
executables
and
that's
good
and
if
that
fails,
we're
gonna
look
at
queue,
control,
A,
B
and
then
finally,
we're
going
to
look
at
queue,
control
a
and
that's
what
happened
here
so
we're
trying
to
come
from
the
longest
prefix
to
the
shortest
prefix
that
is
treated
as
a
command
that
is
treated
as
a
plugin.
A
So
the
next
thing
that
happened,
you
know
when
queue
control
found
this
thing
in
the
path
was
it's
going
to
exec
this?
So
it's
gonna
make
an
exact
system
call
on
POSIX
at
least
right.
It's
gonna
replace
the
queue
control
process
with
the
plugin.
This
means
you're
actually
inheriting
the
log
stream,
such
as
standard
out
standard
error
from
queue
control
itself,
and
you
are
now
determining
the
fate
of
that
queue.
Control
like
invocation
in
Windows.
A
It's
a
bit
different
because
it's
windows
in
Windows,
your
plugins,
are
actually
executed
as
a
sub
process,
but
you
know
still
you're
determining
the
exit
code
and
everything.
So
this
is
what
happens
when
you
run
a
plugin
cube
control
goes
away
and
your
plugin
is
the
actual
thing
that's
running
as
queue
control
now.
So
why
would
you
as
your
developers
here?
Why
would
you
develop
a
cube
control?
Plugin?
A
Can't
you
just
like
write
your
own
I,
don't
know
like
program
just
like
I
did
with
queue
CTX
or
something
and
roll
with
that
you
totally
can,
but
you're
probably
has
a
weird
name
right.
Your
tools,
name
is
probably
not
as
memorable
as
you
would
think.
You'd
develop
that
to
proceed,
and
you
remember
it,
but
others
will
remember
it.
So
that's
a
marketing
problem.
You
know
so
I.
There
are
all
these
successful,
plugins
they're,
all
pretty
successful,
but
their
names
are
not
like
great.
A
So
when
you
put
a
plugin
under
queue
control,
you
can
give
it
a
more
natural
name.
Okay,
so
it's
it's
part
of
like
it
feels
like
it's
part
of
the
cube
control
experience.
If
you,
let's
say,
devolve
a
plug-in
for
developers
at
your
company,
they
will
think
oh
I,
don't
know
who
developed
this.
They
won't
even
realize
it's
a
plugin
honestly,
they
don't
think.
Oh,
this
is
part
of
queue
control
right
natural.
So
that's
what
we're
trying
to
go
with,
and
the
second
reason
is,
you
know
the
most
of
the
times.
A
Cube
control
is
missing
a
feature
and
that's
okay,
you
know
queue.
Control
cannot
do
everything
you
gotta
get
used
to
that.
So,
for
example,
it's
in
here
in
I
have
some
issues.
That's
basically
saying
can
we
have
you
know?
Can
we
add,
like
a
dash
dash
user
flag
to
the
cube
control
exec
because,
like
we
need
that
feature
right,
we
need
like.
We
need
to
execute
something
need
to
run
a
shell
as
a
user.
A
That
makes
perfect
sense
but
turns
out
someone
already
figured
out
a
way
to
do
that
and
like
they'd
what
they
quickly
hack
together,
a
bash
plug-in,
and
it's
actually
used
a
lot
nowadays.
So
that's
pretty
cool
someone
asks,
can
I
have
a
plugin
that
tells
me
which
user
I'm,
authentic
ated
to
my
cluster
right
and
there
we
go.
Someone
actually
wrote
a
plugin
by
intercepting
the
token
from
the
request
and
trying
to
figure
out
who
the
user
is.
That's
cool
again.
You
know
this
is
one
of
Brian
grants
issues.
A
You
want
same
thing
that
does
pew
control
export,
which
is
you
know
you
have
the
deployment
yamo
that
comes
from
the
API,
but
there
are
all
these
static
objects
and
bunch
of
other
like
unnecessary
annotations
and
auto-generated
fields.
You
don't
want
any
of
that,
so
we
have
multiple
plugins
in
crew
actual
today
that
lets
you
export
an
API
object
without
all
the
weird
parts
of
it.
So
that's
pretty
nice
and
then
sometimes
puke
control
does
exactly
what
you
want,
but
it
doesn't
do
it
very
efficiently.
A
Okay,
so
on
the
Left
I,
have
this
cube
control
off?
Can
I
come
in.
So
this
tells
you
whether
you
can
do
something
in
the
API
or
not.
So,
for
example,
I
ask
cube,
control
off,
can
I
delete,
namespaces,
delete
users
install
you,
create
users
and
stuff
like
that.
It
says
yes,
yes,
yes,
but
I
don't
want
to
go
through
every
single
capability.
I
want
to
see
I
wanna
see
what
I
can
do
the
whole
thing
in
one
place.
So
there
is
a
plugin.
You
know
the
excess
matrix
that
I
just
showed
right.
A
It
shows
the
whole
thing.
Maybe
it
takes
a
long
time
to
query
it
took
several
seconds
here,
but
at
the
end,
I
have
this
full
view
that
I
didn't
get
from
queue
control
before
and
that's
okay,
if
you
control
provides
the
primitives
and
someone
builds
on
top
of
it
right,
that's
great.
So
another
example
with
queue.
Control
of
can
I
write,
for
example,
I
want
to
say,
can
I
delete
a
namespace
as
Jeff
at
google.com
or
Tom
at
google.com,
but
I'm
trying
to
figure
out
who
can
delete
a
namespace?
That's
what
I'm
trying
to
write.
A
So
someone
wrote
a
plugin
that
shows
you
who
can
delete
namespaces
and
it's
literally
called
Q
control
who
can
delete
namespaces
right
and
in
this
case,
I'm
C
I'm
listing
all
the
users
I'm
just
matching
all
that
users
from
the
are
back
database,
I'm
measuring
them
with
the
API
groups
and
objects
and
I
can
merge
them
together
to
see
who
can
actually
do
the
namespaces.
So
brilliant
plug-in
idea.
That's
awesome!
Let's
see
so.
Let's
say
you
have
a
plugin
right,
let's
say
you've
a
plug-in.
What
do
you
do?
A
What's
the
next
thing
to
do
you
have
this
great
idea,
you're
totally
implemented
it.
It
works
perfectly.
So
the
next
thing
is
to
bring
it
to
your
users.
Okay,
so
plugin
developers
have
these
problems.
How
can
people
hear
about
my
plugin?
This
is
more
important
than
you
would
think.
If
you
don't
want
your
project
to
die
and
be
a
fun
thing
only
you
need
users
right.
You
need
to
get
feedback,
you
need
to
make
it
a
real
thing.
A
There
are
a
lot
of
successful
plugins
out
there
today,
by
the
way,
there
are
like
almost
maybe
plugins,
with
tousle
stars
out
there
and
because
they're
super
useful.
For
example.
How
would
you
package
your
plugins
for
others
to
install?
Have
you
thought
about
this
before?
Like?
Would
you
make
them
do
go
get,
but
if
they
don't
have
dog
like,
but
if
it
should
they
download
a
bad
script?
Should
they
I,
don't
know
like
how
they
write
your
download
your
plugin
installed
in
written
in
Python?
A
Should
they
use
pip
install
there
are
positions
here
and
how
can
I
deliver
updates
to
my
users?
You
know
it's
not
one-time.
Almost
no
software
is
delivered
one
time
unless
you're
the
author
of
connect
or
utils,
or
something
even
they
need
to
update
things
on
your
computer
right.
So,
if
you're
a
kid
control,
plugin
developer,
this
is
exactly
where
crew
plugin
manager
can
come
and
help
okay,
so
you've.
What
you're
trying
to
do
is
get
your
plugins
to
the
user,
but
in
reality
you
have
users
all
over
the
place.
Your
users
on
Mac
and
Windows.
A
Is
anyone
ever
packaged
something
for
chocolaty
package
manager
on
Windows
to
people
exactly
right?
You
don't
want
to
leave
all
these
windows,
people
out
right
and
then
Linux
I
personally
have
never
packaged
anything
for
Linux,
because
I
don't
want
to
learn
about
all
these
packaging
things.
I,
don't
work
at
the
Red
Hat
or
anything
like
that.
Canonical
I,
don't
understand
the
package
management
package
managers
at
all
I'm,
pretty
sure
I'm
gonna
mess
it
if
I
get
into
that
world.
A
So
I
don't
want
to
learn
that
so
this
is
a
real
crew,
helps
the
package
of
four
crew
ones
and
then
your
users
get
it
and
we're
gonna
see
how
that
happens.
So
without
crew.
You
know,
there's
a
plug-in
here
that
was
called
Cube
control,
grep
it.
What
it
does
is
it
you
know
every
time.
Do
you
remember
yourself
doing
cube
control,
get
pods
and
something
and
then
pipe
that
to
grab?
This
is
exactly
that.
Okay,
it
can
run
on
multiple
namespaces
at
a
time
it's
pretty
cool
plugin,
but
essentially
without
crew.
A
Look
at
all
these
installation
instructions.
It
says:
do
this
W
get
on
TARDIS
thing
based
on
the
operating
system
name
and
then
move
it
to
use
your
local
bin.
Hopefully
it's
writable,
and
then
you
got
two
more
it
right.
There's
no
mention
of
how
would
you
update
it
how
to
uninstall
it
I
mean
obviously
you'll
delete
the
file
nice
because
it's
written
in
go,
but
what
if
it
was
not
written
in,
go
right,
so
it
screws
all
these
problems
pretty
effectively.
A
Essentially,
as
a
plugin
user
crew
helps
you
to
install
plugins,
find
plugins
right
and
keep
them
up-to-date
and
remove
them
cleanly
as
a
plugin
developer,
crew
makes
your
plugin
discoverable,
it
solves
the
packaging
problem,
you
will
don't
go
write
package
managers
for
like
20
different
things,
multiple
platforms
and
thank
you.
Deliver
updates
write
a
super
important,
so
crew
today
operates
on
a
centralized
plugin
index.
This
means
that
we
have
a
github
repository
on.
You
know
coronary
cigs
organization,
and
there
are
all
these
playin
manifests,
and
this
is
where
we
what
we
do
today.
A
This
is
something
that
we
are
trying
to
move
away
from.
You
want
to
give
you
the
ability
to
host
your
own
plugins
manifest,
or
maybe
you
would
use
a
plugin
index
that
is
a
private
git
repo
in
your
company,
so
that
your
plugin
you're
could
control
users
can
get
the
stuff
from
there.
This
is
this
will
make
sense,
we're
just
not
there
yet
so
to
package
and
distribute
a
plugin
with
crew.
It's
actually
pretty
simple,
too.
You
just
need
to
make
a
tarball
or
zip
file.
A
A
There
has
to
be
a
mo
okay,
there
has
to
be
a
mo,
and
this
is
the
mo
that
we
use
so
with
every
plugin
has
a
plugin
manifest
that
describes
what
that
plug-in
is
called
and
how
to
download
it
for
from
where
and
how
to
actually
install
that
plug
into
the
system.
So
in
this
case,
I
have
an
actual
plugin
manifest
that
shows
the
plugin
is
called
excess
matrix.
It
has
a
version
number.
The
version
number
is
totally
used
to
actually
deliver
upgrades.
A
If
you
bump
the
version
number
just
like
I,
you
know
download
the
upgrade
here
on
the
conference.
Wi-Fi,
you
will
get
the
new
version
2
and
then
at
the
end
you
know,
there's
this
platform
section,
that's
kind
of
weird.
Ok,
so
let's
take
a
look
at
what's
going
on
there
right.
So
this
is
the
platform
section.
This
is
one
platform
of
a
plugin.
It's
still
not
very
complicated
right
on
the
top.
I
have
a
selector,
so
this
selector,
all
it
does,
is
to
match
the
user's
computer.
A
If
the
user's
computer
is
running,
Linux
will
install
a
different
binary
if
it's
running
Mac,
when
we
started
from
binary
right,
you
can
have
all
sorts
of
like
label
selectors,
just
like
you've
been
doing
on
on
proven,
alias
pods,
etc.
This
is
all
available
here.
You
can
do
say,
hey
install
it
if
it's
Linux
as
well
as
Mac,
you
can
do
that
too.
You
can
separate
windows,
you
can
deter
it
with
64-bit
and
next
I
have
this
URL
field
right,
so
that
URL
field
actually
is
a
public
URL.
A
That's
a
tarball
here
in
this
case
and
then
just
below
that
I
have
a
checksum.
This
checksum
is
important
because
we
don't
want
people
to
change
the
archives
without
actually
changing
the
version
number,
and
this
is
for
all,
for
you
know,
security
reasons,
etc,
because
we
actually
review
plug-ins
and
at
the
end,
you're
gonna,
see
that
you
know
there's
this
files
section.
The
file
section
is
pretty
simple
too.
You
just
say
all
the
files
that
need
to
be
copied.
You
can
use
like
wild
cards.
You
can
use
the
asterisk
here.
A
Essentially,
what
I'm
doing
here
is
and
I
say
inside
my
tar
bowl
there's
this
practice:
Linux
Amalie
64
binary
extracted
as
records
and
then
I
copy
the
license
out,
because
you
know
you
should
distribute
your
license
right
and
then
at
the
end
of
the
file.
I
have
this
bin
field,
so
that
pin
field
is
my
plugins
entry
point?
Ok,
that's
what
gets
symbolically
linked
to
users
path?
That's
not
a
final
aim,
though.
That
name
can
be
anything,
but
essentially
the
plugins
name.
A
Whatever
the
users
will
use,
is
determined
by
the
metadata
name
here
and
file
name
as
well.
So
some
reason
piece
for
success
are
basically
you
should
name
your
plugin.
Clearly,
ok,
no
one
likes
weird
name,
someone
likes
to
memorize
your
acronym
and
stuff,
like
that,
and
you
should
reuse
flags
from
cute
controls.
We
have
this
really
nice
repository
called
CLI,
run
timing,
cue
control,
anytime,
you're,
writing
a
command
line.
So
let's
say
you're
writing
a
command
line
tool
for
sto.
There's
gotta
be
service
fashion
here,
just
kidding
so
anytime,
you're
writing
it.
A
Let's
say
a
command
line
so
for
easier.
You
should
actually
use
that
you
know
CLI
runtime
repository
because
it
has
all
these
tools
that
help
you
read
the
cue
config
files.
You
know
how
you
know.
Have
you
all
these,
like
common
flags
as
cute
controls
such
as
you
know,
there's
namespace,
there's
cute
config,
you
know
user
whatever
right
so
this.
This
is
something
that
you
can
already
use
and
to
do
that,
obviously
you
need
to
use,
go
but
whatever
right
and
then
finally
there's
this
notion
of
off
plugins.
A
So,
for
example,
if
you're,
if
you
have
a
cluster
on
Azure
or
GK
or
YDC
clusters,
you
actually
need
to
import
off
plugins
and
that's
basically
just
an
import
statement
that
solves
everything.
But
essentially,
unless
you
do
that
cloud
clusters
cannot
use
your
plugin
and
we
encountered
this
pretty
frequently
because
people
develop
plug-ins
for
their
mini
Q
or
something
like
that.
Okay,
so
that's
a
wrap
up.
A
We
want
you
to
write
plugins.
We
want
you
to
join
us
in
currency
CLI
so
that
we
get
your
input
as
well.
As
you
help
us
develop
this
community
of
you
know.
What's
next
for
in
what's
next
in
the
company's
client
landscape,
all
right,
you
can
be
a
plugin
Volpert,
that's
pretty
easy!
We
have
over
60
plugin
developers
today.
I
think
there
are
some
120
plugins
out
there
on
get
up
today,
big
crew,
pretty
much
as
like
half
of
them,
and
then
you
can
submit
your
plugins
to
crew
to
make
them
discoverable.
A
You
can
actually
join
the
crew
project.
The
good
thing
about
curve
is,
I
don't
want
to
just
say:
hey
did
come,
join
me
and
become
a
maintainer,
but
the
cool
part
of
crew.
Is
you
don't
need
to
know
rest
of
kubernetes?
You
know
it's
a
standalone
thing
to
on
its
own.
It
has
a
pretty
young
and
fresh
codebase.
We
almost
never
get
bug
reports,
which
is
pretty
weird.
A
We
had
a
release
for
over
a
year
and
no
one
has
reported
a
bug
about
it
and
it
worked
like
a
champ,
so
we
know
we're
not
trying
to
bring
you
in
tutorial
we're
actually
trying
to
bring
you
into
creativity.
Okay,
we
need
ideas,
we
need
to
figure
out
what's
next
and
you
can
join
crazy,
sick
CLI.
We
have
a
meeting
like
every
other
Wednesday
or
so
you
can
find
more
information
on
the
internet.
So
thank
you.
This
is
my
talk.
A
B
A
Now
we
do
not
have
auto
completion
and
that's
pretty
sad
right,
because
the
whole
thing
that
makes
acute
control,
pretty
usable
and
effective
day
to
day
is
auto
completion.
So
I
think
what
we're
trying
to
do
is
something
like
why
don't
we
have
the
coop
controls?
Completion
command,
maybe
ask
the
plugins
to
return
some
completion
data,
so
if
that
was
a
thing
that
we
could
do,
we
can
do
that.
But
right
now
I
mean
a
lot
of
people
at
the
cube
control,
zsh,
complete
zsh
or
bash
commands
into
their
bash
RC.
A
A
I
mentioned
that
briefly,
we
currently
do
not
support
that.
We
want
to
support
that.
We
have
several.
We
have
a
proposal
out
there
for
multiple
plugin
indexes.
Essentially,
we
want
people
to
run
private
key
triples
in
the
enterprise
environment.
Just
you
just
described
so
the
corporate
developers
they
can
get
the
plugins
automatically.
D
A
Yeah
so
yeah,
so
here's
the
thing
honestly
partly
to
blame
here,
I'm
trying
to
find
you
know
time
between
the
other
Google
work
and
finally
in
Korea.
So
it's
actually
I
have
a
proposal
in
mind
and
several
other
people
do
as
well.
We've
gone
through
some
drafts,
but
essentially
we
have
not
settled
on
something
that
could
work
out
of
the
box
directly
without
impacting
the
public
open
source
plugin
landscape.
A
So
we
would
probably
like
have
a
separate
file
where
you
would
describe
other
plugin
indexes,
but
then
we
run
into
issues
like
what,
if
there's
a
name
conflict
like
name
overlap
with
the
between
a
public
plugin
and
a
private
plane.
So
how
do
we
know
where
a
particular
plugin
comes
from
where,
when
a
time
when
the
time
comes
to
upgrade
that
plugin,
another
thing
is
like?
We
actually
have
certain
learnings
and
the
centralized
index,
because
we
know
how
to
operate
the
plugin
manager
as
well
as
the
index.
A
So,
for
example,
there
was
a
time
for
a
long
time.
If
you
deleted
a
plugin
from
the
index,
this
will
break
all
the
users,
so
we
actually
never
deleted
a
plugin
from
the
index.
So
but
you
know
we
addressed
all
these
stuff
already,
so
we're
trying
to
probably
go
somewhere
that
people
can
write
their
own
indexes.
It's
been
a
P,
1
or
P
zéro
for
the
longest
time.
I
can
tell
you
that.
D
D
A
Honestly,
like
there's
a
feature
that
we
support,
but
it's
not
exactly
what
you
want
so
I'll
quickly
show
it.
Essentially
if
you
have
chrome
installed
today,
you
can
do
this,
you
can
do
QC
tail
cruellest
right.
This
is
the
plug-in
list
and
if
I
actually
pipe
that
to
a
file,
let's
say
foo.
If
I
looked
at
foo,
it
actually
is
the
list
of
plugins
right
I
can
delete
all
these
plugins
and
I
can
read
like
back
them
up
and
restore
them.
A
Basically,
by
doing
I
can
do
keep
control
crew
install
and
I
can
pipe
it
back
like
this.
I
mean
obviously
they're
all
installed,
but
essentially
this
would
install
it
again.
So
this
is
there's
like
a
backup
mechanism
in
place,
but
it's
not
quite
what
you're
looking
for
I
think
right,
but
yeah
I
think
we
didn't
think
of
like
a
you
know,
artifact
story
or
anything
like
that.
Essentially,
if
you
want
think
so,
I
think
come
from
get
or
something
like
that.
But
if
you
have
ideas,
let's
check
alright.
E
A
F
F
Yeah
customized
was
well,
it
was
customized,
it's
a
separate
topic
because
customized
became
essentially
a
part
of
the
one
of
the
sub
projects
for
460
of
I
and
the
core
mechanism
are
ended
up
being
part
of
the
cube
CTL
itself,
I
haven't
seen
any
commands
going
from
the
building
commands
outside.
There
is
a
proposal
like
Ahmed
mentioned
about
debug,
and
if
everything
goes
well,
we
will
combine
all
the
debug
functionality
from
several
sources
into
a
full
full-blown
debauch
under
cube
CTL.
F
I've
noticed
when,
when
Ahmed
was
listing,
the
plugins
and
one
of
the
plugins
is
exit
cron
job,
which
is
actually
possible
to
do
with
the
built-in
commands
because
create
job
has
a
from
cron
job
flag.
So
that
was
funny
because
it
means
that
people
are
not
looking
at
the
building
commands
and
that's
there
pretty
much
I
would
say
at
least
a
year
or
even
two
years
old,
because
I
remember
writing
it
a
long
time
ago,
so
yeah
but
yeah
that
that's
more
or
less
how
it
looks
like.