►
From YouTube: GitLab Kubernetes Agent and Runner UX
Description
Rayana, Maria and Thong discuss UX for GitLab Kubernetes Agent, and Runner
A
A
A
I
was
just
talking
about
how
what
the
differences
are
between
runner
and
agent,
and
the
first
thing
is
that
they
are
not.
They
are
too
completely
distinct
processes
and
we're
still
working
out,
which
places
make
sense
to
connect
and
which
places
makes
does
not
make
sense
to
connect,
and
currently
the
primary
use
case
of
the
run.
The
agent
is
to
deploy
a
kubernetes,
manifest
files
to
the
kubernetes
cluster.
A
B
A
So
it's
at
the
end
of
devops
process
it
doesn't.
It
doesn't
wait
for
runner
to
run
or
anything
like
that
it
currently
doesn't
it
just
sees
there's
a
new
commit
in
get
delay
or
in
the
repository
and
it
deploys
we
don't.
We
don't
wait
for
it.
That's
because
our
current
use
case
is
that
there's
not
many
ci
jobs
that
we
can
think
of
for
kubernetes
processors
right.
B
A
A
A
A
C
A
So
and
the
other
thing
that
that
we're
trying
to
do
is
that
we're
trying
to
be
flexible,
we're
trying
to
have
the
ability
to
to
configure
the
agent
for
kind
of
cross
project
settings
where
the
idea
is
that
there's
like
an
application
project.
And
then
the
application
project
will
trigger
a
cross
project
pipeline
to
the
manifest
projects
where
which
contains
all
the
kubernetes
manifest
resources,
and
also
we're
trying
to
be
together
for
the
other
scenario,
where
everything's
in
one
one
repo,
which
is
definitely
simpler.
A
Has
less
separation
of
permissions
or
duties,
because
the
stereotypical
teams
is
that
kind
of
sre
team
will
work
on
this
project
and
they'll
have
their
own
permission,
structure
and
access
to
the
cluster?
And
then
the
development
team
would
have
another
project
with
the
app
right
and
they
are
not
allowed
to
deploy
anything
and
currently
gitlab
is
just
not
set
up
for
those
two
teams
mixing
together.
All
in
one
project.
B
A
A
Yes,
they'll
have
to
deal
with
agents
and
yes,
they'll
have
to
do
runners
because
they
might
have
their
own
runners
to
link
the
kubernetes
files
or
whatever
right
they'll.
Have
that
and
and
then
they
will
have
some
sort
runner
to
consume
the
the
container
images
that
was
built
by
the
application
to
the
developers.
A
B
I
think
it
depends
on
how
the
teams
are
structured
right.
They
can
be
whatever
the
engineer
can
be.
A
sre
managers.
There
are,
for
runners,
for
example,
that
these
people
that
are
responsible
for
managing
groups.
So
you
have
you
go
like
a
higher
level,
but
I
think,
if
financial
correctly
for
runner,
for
example,
this
is
more
like
on
an
enterprise
level
right
when
you
have
bitlab
admin
instance,
and
you
have
a
more
broken
down
hierarchy
for
these
responsibilities.
A
C
A
As
a
user,
I
want
to
use
the
runner
for
fine,
I'm.
I
guess
I
guess
the
the
first
like.
If
we
look
at
the
devops
things
as
a
user,
I
want
to
use
the
runner
to
build
my
container
because
that's
the
first
one
cloud
native,
let's
assume
we're
cloud
native,
that's
that's
kind
of
the
first
job
and
then,
as
a
user
I
want
to.
A
C
A
C
C
B
C
You
have
you're
using
gitlab.com
and
you
are
an
sre
and
let's
say
you
take
care
you're,
an
a
devops
engineer:
sre
you're,
the
combination
of
ops,
personas
right
so
you're,
the
only
person
who
takes
care
of
the
infrastructure
and
deployments.
That's
my
scenario,
your
man,
one
one,
one
team
and
you
see-
and
you
want
to.
C
A
A
A
A
C
A
C
C
A
A
B
A
C
C
A
A
A
A
And
then
you
can
set
set
your
own
specific
service
account
to
run
this
right,
but
it's
not
it's
it's
pretty
fixed.
You
can
only
run
this
one
service
account.
So
if
you
want
to
run
as
another
service
account,
you
you
have
to
install
another
runner
basically,
which
is
doable
like
it
depends
on
your
specific
needs
and
what
your,
what
you're,
trying,
what
how
flexible
you
want
to
be
and
how
secure
you
want
to
be
right.
So
I
feel
like
that.
A
D
A
A
If
I,
if
I'm
coming
in
as
an
engineer
in
large,
organize
from
large
large
company,
for
example,
and
then
I
read
the
documents,
it
says,
I
have
a
kubernetes
cluster,
just
install
the
agent
and
you
can
it
will
take
you
everything
for
you
great.
That
sounds
great
to
me.
I
recommend
its
cluster.
I
come
from
a
large
organization.
A
B
C
C
A
D
A
C
C
A
A
B
A
B
Yeah,
you
still
need
runners
to
set
up
well
ci,
because
you
need
to
write
the
jobs.
You
need
to
write.
Everything
right
then
pass
everything
related
to
runner
gear
and
then,
when
you
are
specifically
deployed
to
your
kubernetes
cluster,
the
whole
shibang,
then
it's
where
we
go
to
to
the
next
step.
I
think
that's
where
maria
talked
about
making
the
clear
differentiation,
but
also
saying
this
is
another
step
you
can
take
in
the
process.
A
B
B
I'm
no
kubernetes
expert,
but
we
are
talking
about
like
how
to
manage
clusters
and
how
to
have
like
a
a
high-level
view
for
the
admins
and
for
the
sres
about
everything
that's
going
on
with
our
clusters.
Oh
sorry
with
the
runners,
so
maybe
this
is
also
another
thing.
You
know
that
we
can
discuss
further
how
if,
if
they
are
in
a
way
part
of
a
single
user
flow,
if
they
could
be,
how
do
they
connect
at
the
higher
level
at
that
admin
level
at
the
for
these
enterprise
customers?
C
A
C
A
That's,
like
you
know,
there's
the
higher
level
flow,
which
is
the
kind
of
a
summary
flow
and
also
the
kind
of
the
control
flow
for
maybe
like
sras.
They
want
to
know
if
there's
any
issues
or
anything
like
that
after
things
have
been
deployed
or
a
health
dashboard
flow
for
example,
so
all
those
happens
afterwards,
which
neither
the
agent
nor
the
runner
handles
pretty
well,
so
the
opportunity
is
there
potentially.
A
D
A
A
No
yeah
he
could
it
could
it
could
it
could,
but
the
agent
can
sort
of
like
automatically
set
up
things
or
something
like
that.
That's
kind
of
the
vision,
well,
we're
thinking,
even
even
bigger,
in
that
the
agent
can
automatically
set
up
things
and
push
things
to
a
generic
api
that
is
not
kubernetes
specific
and
in
the
future,
anyone
can
set
up
their
own
automation.
A
A
A
So
this
is
something
that
victims
trying
to
find
out,
whether
is
that
a
big
enough
market
or
not
to
do
that
or
whether
we
need
to
solve
this
whole
issue
of
automatically
set
up
outer
box
experience
for
monitoring
and
so
on
in
a
better
way,
because
if
you
only
solve
the
outer
box
experience
for
kubernetes
clusters,
sure
it's
a
win,
but
it
might
be
a
very
small
win
or
might
be
great.
But
we
don't.
C
C
But
according
to
well
at
least
a
few
people,
I've
spoken
to
they
use
more
vms,
some
still
use
on
prem.
Basically,
I
think
now
is
the
time
that
people
have
migrated
or
are
still
migrating
from
on-prem
to
cloud
solution.
But
what
I'm
hearing
is
most
people
still
use
vms
and
fewer
people
use
kubernetes,
because
there
is
a
lack
of
expertise.
C
A
A
A
C
A
I
think
it's
kind
of
size
dependent,
so
if
you
have
only
under
50
barbers,
which
equates
to
about
maybe
four
teams
for
developers,
it's
not
worth
it
to
set
up
kubernetes,
probably
because
it's
probably
cheaper
to
set
up
your
own
set
of
vms
and
you
can
customize
it.
You
can
get
all
the
benefits
of
customized
flexibility
without
having
to
kind
of
standardize
right.
A
Thing
to
deploy
to
which
is
kubernetes
gets
its
own
benefits
because,
beyond
that
size,
beyond
100
developers,
you
you
have
the
budget
to
employ
two
to
three
people
to
take
care
of
it,
because
then
it
is
and
then
your
developers
don't
have
to
wait
before
vmcp
provision
anymore.
Kubernetes
does
that
for.
B
A
A
A
A
A
It
doesn't
have
the
experience
of
one
single
dot
gillab
sarah
yama
found
before
we
run,
for
example
this
one
where
I
need
to
set
up.
We
need
to
know
the
name
of
the
agent
and
then
we
need
to
set
up
a
folder
for
you
for
each
agent.
If
I
only
have
one
agent,
then
it's
it's
a
lot
of
work
and
then
I
have
to
set
up
the
name
of
the
id.
C
This
is
quite
frustrating
I'm
trying
to
find
a
video
from
victor
who
explained
so.
I
posted
the
solution,
architect,
video,
which
is
14
minutes,
I
think,
quite
short,
but
I
think
it's
a
bit
different
to
the
one
victor
shared
with
me,
but
yeah.
I
think
it
can
be
well.
I
don't
know
the
kubernetes
commands.
That's
my
problem
because
you
use
grenades,
specific
commands
right,
not
commands
for
the
installation
you
use
both,
but
you
also
have
the
configuration
file
where
you
define
some
stuff,
so
my
only
the
only
alert
was
for
the
two
agents.
C
A
B
If
I
can
suggest
something,
since
you
already
have
this
in
the
application
right,
you
have
a
problem
you
can
validate.
I
would
do
a
maturity,
scorecard,
validation
idea,
because
then
you
can
just
choose
just
one
scenario
and
then
just
see
how
people
configure
this
like,
for
example,
we
did
for
secret
management
and
yeah
everything
everything's
there,
and
the
expectation
is
that
you
have
your
vault
configuration
you
can
just
bring
to
gitlab.
People
could
not
complete
that
you
know
one
hour
and
then
someone
is
still
learning
how
to
configure
going
through
the
docs.
A
C
B
B
A
B
I
think
that's
the
for
for
these
configurations
where
we
assume
yeah
it's
there.
People
can
just
figure
out.
It's
super
painful
and
git
lab
docs
are
yeah,
are
super
complex,
so
if
you're
not
you
know
that
viktor
might
not
invest
in
ui
right
now,
and
you
know
from
engineer
point
of
view
you
can
improve
this.
This
flow
improve
the
configuration,
yeah
I'll
suggest.
A
B
A
I
think
the
num
we
don't
necessarily
have
to
have
a
ui
but
like
because
the
ui
might
even
increase
the
number
of
steps
right.
I
think
the
number
of
steps
is
already
quite
high,
so
that's
and
and
the
complexity
of
understanding
is
quite
high
as
well.
So
I
think
those
are
valid
points
to
consider
without
looking
at
the
solution.
First.
C
B
Yeah,
but
you
can
figure
out
the
pain,
points
right
and
then
like
what
you're
saying
to
find
the
job
should
be
done.
You
have
the
jungles
to
make
make
configuration
easy
or
to
whatever
do
everything
one
emo
file,
you
know
and
not
necessarily
do
ui,
because
if
you
cannot
configure
like
seamlessly
from
the
api,
then.
C
At
least
we
can
polish
on
the
user
side.
So
that's
what
I
always
say
to
developers,
because
you
have
built
it
in
a
specific
way,
so
you
know
how
it
works
and
you
feel
most
developers
feel
they
should
be
truthful
to
the
user,
with
the
way
that
the
thing
is
built.
But
you
can
lie
pretty
easily
in
the
ui.
C
A
A
A
C
A
C
Again,
it's
linking
that's
the
problem
with
the
agent
you
link
it,
so
you
link
to
the
manifest
you
link
it
to
its
own
configuration.
If
you
need
to
link
it
also
to
a
different
yaml
or
maybe
that's
also,
you
have
a
different
yaml
for
security
stuff
and
you
add
it
to
the
agent
of
configuration
manifest.
A
A
A
A
A
Yeah
could
be
at
one
point
I
was
contemplating,
but
it's
really
really
hard
to
predict
what
groups
align
at
point
of
time.
So
most
of
the
time
groups
don't
have
much
alignment
yeah
yeah,
because
you
know
if
you're
not
working
on
the
same
features
at
the
same
time
they
don't
that
don't
match.
Then
then
you
don't,
then
you
don't
have
olaf,
but
if
we
then,
if
you
do,
then
we
start
we
do.
We
do
ad
hoc.
So.
C
A
A
A
C
We
have
five
comments,
make
clear:
the
agent
is
specific
for
kubernetes
and
runner
can
be
any
type
of
infrastructure
name
the
agent
properly
communicate.
Well,
that
was
mine,
communicate
in
the
docs
I'll
communicate
in
the
docs,
which
one
you
can
choose
and
use
depending
on
the
use
case,
and
I
put
a
sub
note,
look
at
the
docs
and
see
how
understandable
and
usable
they
are
and
for
me
to
do
a
category
maturity
card
scorecard,
ux
scorecard
for
the
agent
installation
experience
adding
here.
B
C
So
all
the
steps
that
we
have
described
above,
I
would
like
us
to
go
through
gitlab
and
see
that
would
be
ux
scorecard
type
of
thing
and
evaluate
the
user
journey
and
maybe
then
try
to
figure
out
well
see
how
you
can
do
it
now.
If
you
can
do
it
and
if
you
can't
maybe
work
on
a
vision
on
how
you
would
be
able
to
do
it.
C
B
B
I'm
writing
the
runner
ux
page
and
a
lot
of
the
information
shared
here
is
going
to
help
me
kind
of
try
to
understand
a
little
bit
more,
the
not
only
the
personas,
but
what
this
thing
is
and
if
I
have
any
questions
I'll
bring
you
or
any
suggestions.
Yes,
I
learned
a
lot
from
you
that's
today,
so
thank
you
so
much.
A
C
A
A
Yeah
so
like
it's
pretty
easy,
like
I'll,
try
and
link
to
the
example
project
that
I
did.
This
is
what
I
do
like
most
days
now.
I
just
create
new
projects
with
ci
ymo,
so
it
can
be
as
simple
as
this.
This
is
a
project
that
I
did
to
kind
of
play
with
artifacts
and
dependencies,
so
I'll
put
it
in
the
wrong
place.