►
From YouTube: GitOps AMA Panel - OpenShift Commons Gathering 2022
Description
OpenShift Commons Gathering 2022
Panel Moderator: Christian Hernandez (Red Hat)
Panelists:
Siamak Sadeghianfar (Red Hat)
Jaime Magiera (ICPSR at University of Michigan Institute for Social Research)
Cornelia Davis (Amazon)
Priyanka Ravi (weaveworks)
Full Agenda:
https://commons.openshift.org/gatherings/OpenShift_Commons_Gathering_on_GitOps.html
Learn more at OpenShift Commons https://commons.openshift.org
A
A
A
C
Yeah
yeah,
so
actually
you
know
pinkie
kind
of
kind
of
joined
us.
We
kind
of
added
her
last
minute.
So
you
know
we
we
kind
of
we're
like
okay.
Well,
she
can
do
the
panel
we
had.
We
had
room
on
the
panel,
so
real,
quick
pinky.
If
you
can
introduce
yourself.
I
know
we
heard
ciamac
and
cornelia
earlier,
but
if
you
just
quickly
introduce
yourself
yeah.
D
Yeah
so
my
name
is
priyanka
ravi.
I
also
go
by
pinky
and
I
have
been
a
user
of
flux
and
get
ops
for
I
think
two
and
a
half
years
now
I
actually
started.
I
started
with
get
ops
at
state
farm
where
I
was
a
on
the
platform,
the
get
ops
platform
team
and
I
recently
just
moved
to
weave
works
where
I'm
a
dx
engineer
now
so
awesome.
C
Awesome,
thank
you
yeah
yeah.
I
know
it
was
awesome
for
those
for
those
of
you
who
weren't
at
get
ups
con
in
in
north
america.
She
great
she
made
an
amazing
presentation,
one
of
my
favorite
presentations
there.
So
if
you
catch
the
the
playlist
there,
it's
really.
I
always
love
hearing
end
user
stories.
So
you
know
here
here
at
the
end,
we
we
kind
of
try
to
get
everyone
together.
I
think
I
think
we
couldn't
get
dan
because
of
his
schedule.
C
I'm
not
sure
I'm
still
trying
to
track
him
down
see.
Hopefully
he
can
make
it,
but
we
will
we'll
see
here
to
have
kind
of
like
a
q
and
a
session
I'll
I'll.
You
know
kind
of
for,
for
those
of
you
who
are
here
feel
free
to
drop
a
question
either
in
the
chat
on
the
event
tab
or
the
stage
tab.
I
keep
switching
back
and
forth.
I'm
not
100
sure
where
I
should
be,
but
I'm
switching
back
and
forth.
C
So
don't
worry
about
that
I'll
I'll,
see
you
guys
the
questions,
so
I
think
I'd
like
to
start
since
we
gave
all
the
other
presenters
time
to
speak.
I
think
I'll
start
with
with
pinky
with
with
my
first
question
here,
because
you
know
what,
let's
just
let's
just
start
with
pinkie,
because
she
wasn't
able
to
to
present.
C
It's
well
I
mean,
if
we're
going
to
be
honest
now,
so
you
know
I
actually
actually
this
this
question
came
came
to
mind
because
I
actually
attended
one
of
your
talks.
You
you
did
a
did
a
weave
works
presentation
and-
and
you
know
you
did
a
up
right-
get
ups
to
meet
up
with
weaveworks
and
this
question
pops
to
mind,
and
it
has
relates
to
the
kind
of
the
presentation
I
did
so
getup
seems
really
like
a
like
a
fully
automated
operation
model
for
your
your
entire
platform.
D
D
So
yes,
yes,
basically,
the
first
step
is
to
obviously
define
all
your
deployments
or
your
infrastructure,
whatever
you're
doing
declaratively
as
code,
so
that
it
and
that
in
of
itself
comes
with
so
many
benefits
right
because
one,
you
know
what
your
desired
state
is
by
just
going
and
looking
at
it.
Instead
of
all
these
manual
like
processes
and
then
also
it's
something
that's
repeatable
and
reusable
as
well
right.
So
there's
a
lot
of
benefits
to
just
doing
that,
one
step
and
then
the
like.
C
C
C
E
Exactly
I
think
that
it's
it's
helpful
to
start
out
with
those
small
pieces,
look
for
things
that
you
that
the
devops
or
devs
are
repeating
look
for
things
that
they
do
over
and
over
look
for
things
that
just
require
a
small
change
like
per
environment.
Right,
like
you,
just
need
this
variable
change
or
you
just
need
whatever
you
know
and
build
it
out
bit
by
bit
and
that's
the
technical
aspect.
C
A
C
Is
but
but
but
since
you
know,
you
were
involved
in
the
process
of
the
get
ops
principles,
you
know
you
and
I
were
in
that
meeting
a
ton
of
those
meetings
together.
You
know
I'm
kind
of
curious
so
like
what
what
considerations
were
taken
into
account
when
the
get-offs
principles
were
being
drafted
right,
like
I
kind
of
know
these
ques
the
answer
to
this
question,
but
I
kind
of
want
to
get
your
input
right,
like
what
kind
of
I'm
kind
of
think.
So
there
were
a
lot
of
opinions,
which
means
a
lot
of
obstacles.
F
Yeah
I
and
that's
a
really
great
question
to
dovetail
from
the
one
that
you
were
just
talking
about,
which
is
what
consistently
comes
up
is
hey?
Do
I
have
to
get
all
the
way
to
the
holy
grail,
and,
and
so
I
just
want
to
add
on
to
what
the
what
the
other
two
participants
already
said
in,
and
that
is
even,
if
you're
not
doing
git
ops,
adding
automation,
hugely
valuable,
do
it.
F
And
so
I
think
the
most
contentious
thing
when
we
started
building
the
principles
was:
do
we
really
need
to
specify,
for
example,
that
it
has
to
be
convergent,
or
do
we
really
need
to
specify
the
principle
that
you
know
git
is
the
only
way
for
you
to
affect
things
and
the
tension
that
we
fell
into?
Is
that
it's
easy
to
to?
F
Now
again,
that
doesn't
mean
you
can't
get
on
a
path
to
getting
to
full
get-ups
and
still
realize
value.
So
I
think
that
was
the
the
biggest
fundamental
tension
was.
Do
I
really
need
a
convergent
system?
Can't
I
just
check
things
into
git
and
have
a
script
run
and
call
that
get
offs,
and
the
answer
is
no,
let's
not
call
it
get
ops,
but
yes,
please
do
it.
If
what
you're
doing
today
is
clicky,
clicky
interfaces
by
all
means
create
a
a
script
that
triggers
off
of
a
git
check-in.
F
But
we
really
want
you
to
get
the
the
ultimate
values
which
is
which
comes
only
when
you,
when
you
embrace
those
four
principles,
or
maybe
it's
five
by
now.
But
we
really
tried
to
stay
very
kind
of
axiomatic
so
that
we
weren't
adding
things
that
didn't
provide
additional
value.
We
tied
those
things
to
value.
C
Yeah,
I
know
that
that's
the
I
think
you
touched
on
a
very
interesting
point.
This
is
like
the
value
right.
The
value
you
get
from
a
lot
of
this
automation,
I
think
very,
very
early
on
in
the
get
ops
working
group
is
like
well,
we
don't
want
to
like
exclude
people
from
the
club
right,
but
also
we
do
want
to
define
what
get
ops
is.
C
So
you
know,
I
think,
distilling
that
into
like
principles
of
like
you
know,
version
controlled
and
immutability
doesn't
necessarily
mean
get,
even
though
it's
get
ops
right
like
we
talked
about
get
up
because
we
didn't
want
to
exclude
other
other
methods
right
of
of
doing
like,
for
example,
like
object,
storage
right
that
is
immutable,
and
that
is
version.
You
can
version
those.
So
so
I
think
that's
very,
very
great
see
mac.
I
know
it's
late
for
you
and
I
don't
want
to.
C
I
don't
want
to
ignore
you,
but
that's
why
so
there's
quite
there
is
questions
coming
up,
so
I
think
maybe
I'll
start
this
question
with
you.
So
these
these
are
you
know,
I'm
not.
You
know,
I'm
not
putting
you
in
a
weird
position.
This
is
just
the
way
it
is
the
dice
rolled
right,
the
way
the
dice
rolled
in
the
questions
right.
So
maybe
you
don't
know
so.
The
question
comes
from
jeffrey
too,
about
using
getups
at
the
edge
right,
and
so
what
would
you
know?
C
B
C
B
So
there
are
like
a
couple
of
aspects
when
it
comes
to
edge,
which
affects
pretty
much
any
any
application
that
is
running
there.
Any
service
that
is
that
is
running
there,
including
how
you
approach
to
get
out
for
the
first
piece
of
that
is,
is
a
matter
of
scale.
Why
so,
like
I?
I
spoke
about
github's
picking
up
because
of
the
number
of
clusters
are
growing,
and
this
is
in
the
numbers
of
10
1500.
When
we're
talking
about
edge,
then
we
are.
B
We
are
speaking
about
thousands,
multiple,
ten
thousand
hundred
thousands,
endpoints
and
and
devices
we
talked
about
in
vehicle
services,
and
that
changes
the
perspective
of.
When
we
talk
about
get
ops.
There
is
git
ups
on
the
kubernetes
clusters.
There
is
always
the
pull
and
push
model
conversation.
There
is
a
choice
that
customers
have,
depending
on
their
security
and
compliance
or
some
of
the
other
aspect
they
choose
one
or
other
when
it
comes
to
edge
that
that
choice
really
disappears.
There's
only
one
way
to
think
about
this.
B
B
That
are
being
deployed
right
if
you
have
10
000
instances
of
a
device
coming
to
a
particular
git
repo
to
pull
their
configuration.
That's
not
something
that
necessarily
the
git
providers
are
ready
for
or
would
be
ready
for
so
the
way
that
this
works
also
might
need
to
change
on
the
on
the
githubs
part
on
the
engine.
That
is
doing
this.
So
there
are
multiple
aspects
of
like
that
that
flacky
infrastructure,
that
an
edge
device
is
and
then
the
nature
of
the
limited,
compute
and
and
spotty
network
that
exists
on
this
device.
F
I
would
love
to
pile
on
a
little
bit
on
this.
Just
a
moment
ago
we
were
talking
about,
are
all
the
principles
essential
and-
and
we
said
no,
yes,
that
we,
we
had
a
minimal
set
and
we
tied
it
to
value.
So
what
is
one
of
the
values
that
we
were
tying
these
principles
to?
Well,
it's
autonomy.
It's
independent
system,
autonomy
which
is
ins
absolutely
critical
on
the
edge.
F
C
But
first
of
all,
I
I
just
like
to
like
to
say:
having
argo
being
deployed
on
a
ship
is
just
like
perfect
right,
because
it's
like
it's
like
that's
what
he's
named
after
a
ship
so
but
so
some
some
of
the
things
that
we've
been
you
know
in
jamie
you've
been
part
of
some
of
these
conversations
in
the
get
ops
working
group.
Talking
about
the
idea
of
get
mirrors
and
using
payloads,
you
know
with
respect
to
the
source
of
truth
right,
because
you
can
always
just
tar
up
your.
C
C
If
you
want
to
learn
more
about
that,
so
this
next
question-
I
love
this
this
next
question
because
it's
going
to
go
to
both
jamie
and
pinky,
because
you
guys
are
end
users
so
I'll
pick.
Whoever
wants
to
chime
in
first
go
ahead,
feel
free,
so
there
are
so
I
think
maybe
I
should
this
is
by
anonymous,
which
is
makes
a
little
eerie.
C
Someone
anonymously
asked
this,
so
there
are
certain
instances
where
configuring,
the
cluster
are
not
possible
to
do
declaratively
or
in
in
this
person's
opinion,
declaratively
right
things
like
host
subnets,
ciders
machine
and
node
get
deleted
right.
So
some
tasks
are
are
almost
like
you,
you
can't
do
them
with
declaratively
right,
like
you
know,
like
vms
right,
like
nodes
like
networks,
so
you
know
I'm
curious
about.
You
know
both
of
you
pinky
your
time
at
state
farm
and
jamie.
How?
How
do
you
guys?
C
D
If
flux
can
do
this
and
it
actually,
I'm
pretty
sure
it
can
do
exactly
what
they're
saying
so
you
can.
You
can
patch
resources
and
stuff
like
that.
I
don't
know
argo.
Obviously
I'm
limited
to
my
experience
with
flux
but
yeah.
I
think
this
actually
can
be
done
and
there
are
maybe
some
like
things:
you'd
have
to
kind
of
finagle
to
do
it
but
yeah.
So
that
was
an
interesting
question
because
I
had
to
get.
E
D
C
C
That
sort
of
thing
right,
and
so
you
know
part
of
the
you
know
part
of
the
question
they
talk
about
like
networking
like
well,
you
can
technically
manage
your
network
configuration
so
like
I'm,
an
old
cisco
guy
right,
like
you,
can
have
like
your
configuration
of
your
cisco
switches,
you
could
technically
have
that
and
get
and
have
ansible
apply.
Those
like
it's
not.
You
know.
E
That's
you
mentioned
terraform
and
that's
actually,
what
we're
doing
right
now
is.
We
have
snippets
of
terror
form
saved
in
the
repositories
with
the
respective
apps
that
were
rolling
out
or
or
for
our
cluster
configuration
repositories
and
then
using
the
terraform
container
to
actually
apply
those
in
so
there's,
there's.
C
E
C
D
A
F
F
For
the
most
part,
that
code
is
golang
or
it's
java,
or
something
like
that.
What
we're
trying
to
do
with
git
ops
is
we're
trying
to
elevate
the
the
interface
to
the
developer,
the
devops
persona,
so
that
they
don't
have
to
deal
with
that
imperative,
the
heart
imperative
stuff,
and
so
there
is
no
rule
that
says
you
can't
do
some
imperative
as
a
part
of
your
automation
of
the
system.
F
It's
just
that
you're
taking
on
that
burden-
and
I
know
to
some
people
who
are
used
to
imperative
going
declarative
feels
like
a
burden,
but
in
the
end
it
ends
up
being
so
much
easier
because
you're
deferring
that
responsibility
for
the
hairy,
hard
imperative,
stuff
and
imperative
stuff
is
the
hairy,
hard
stuff
relative
to
declarative
you're,
deferring
that
to
a
system.
That's
good
at
that.
That
will
do
it
consistently.
That
won't
make
mistakes
and
all
of
that
stuff
and
that
that's
a
system
like
kubernetes
or
get
ups
or
something
like
that.
C
So
this
next
question
is
again
also
another
another
great
question,
because
I
honestly
don't
have
an
opinion
on
it.
So
let's
just
see
what
you
guys,
what
you
guys
think
about
it.
So
there's
there's
talk
about
pull
versus
push
right
so
or
pulling
a
configuration
versus
pushing
a
configuration
out
right.
I
know
that
flux
can
do
both.
I
know
argo
city
can
do
both
push
and
pull,
but
with
respect
to
get
ops
and
with
like
respect
to
this
is
general
to
everyone
right.
C
So
whoever
wants
to
chime
in
go
ahead
is,
what's
your
opinion
between
on
push
versus
pull
it?
Does
it
have
to
be
either,
or
is
this
just
something
that
is
like
whatever
works
for
you?
What
what's?
What's
the
you
know?
What's
the
consensus
on
that
who
wants
to
go
first
here?
Who
should
I
pick
on
see
you
much?
Do
you
have
an
opinion?
I
can
think
on
cm.
B
So,
with
the
push
model,
what
we
hear
from
customers
that
they
do
like
that
central
model
of
the
single
pane
of
glass
that
is
from
central
place,
they
can
see
all
the
status
of
everything
that
is
happening
across
their
their
fleet.
What
status
they
have.
It's
one
place
to
give
them
like
very
simple
visibility,
and
they
can
live
with
having
access
from
outside
to
all
those
clusters
right.
F
B
We
have
customers
that
this
is
an
absolute
no-no.
It's
there's
no
way
that
they
can
define
a
single
piece
of
network
somewhere
in
the
infrastructure
that
can
reach
everything,
and
so
they
they
cannot
live
with
that
and
and
the
pull
model
was
the
only
way
that
they
would
work
with,
and
we
talked
about
edge
where
some
use
cases
it's
just
impossible
to
go
the
other
way
around.
So
from
I
personally,
I've,
no
opinions
that
I
just
see,
I
see
advantage
disadvantages
for
each
of
them
and
each
of
them
fits
certain
use
cases.
E
Yeah,
I
would
just
add
to
that
so
in
in
a
very
specific
sense,
there
are
limitations,
physical
network
and
also
compliance.
So
if
you're,
an
organization
that
deals
with
compliance,
there
are
some
barriers
that
will
define
for
you
whether
you're
going
to
do
push
or
pull
based
on
access
and
whatnot.
So
that's
that's
another
thing
to
consider,
so
I'm
I'm
in
the
camp
of
right
tool
for
for
the
situation,
the
right
way
for
the
situation.
D
F
C
C
Yeah
for
those
that
that
know
me,
I
I
can
I
I
hold
very
strong
opinions,
but
sometimes
my
opinions
will
those
strong
opinions
will
change
depending
on
what
you're
doing
right?
Sometimes
I
get
asked
a
question
I
go
well
tell
me
what
you're
doing
my
opinion
might
change
depending
on
on
what
you're
doing
and
how
you're
delivering
it,
because
I,
you
know
it'll
be
a
strong
opinion,
but
it
might
change
so
to
you
know
to
be
cognizant
of
time
right.
So
you
know
to
respect
everyone's
time.
C
I
know
you
know
we
don't
want
everyone
sitting
in
front
of
camera
all
day.
I'll.
Ask
this
last
question
and
I
I
like
to
get
a
thought
from
each
one
of
you.
I
don't
know
where
you
are.
You
guys
are
with
respect
to
in
the
screen.
I
may
be
pointing
out
myself,
but
I'd
like
to
ask
the
question.
Actually
this
question
was
originally
meant
for
saying
I'm
going
to
ask
it
to
all
of
you,
so
you
guys
take
you
know
a
minute
or
two
to
answer.
C
I
will
start
with
cmak
and
we'll
go
so.
What
what's
you
know,
get
ops
is
so
new,
but
I
can't
help
to
think
about
what's
the
future
right
so
like
what
what's
what's
next,
where
does
get
ups
go?
Where
does?
Where
is
where
does
it
take
us?
What's
the
next,
you
know
splat
ops,
you
know
marketing
phrase.
That's
gonna
come
up,
so
where
do
you
see,
you
know
we
see
the
benefits
of
get
ups.
What
do
you
you
know
what
what's
the
future?
C
B
I
think
that
would
that
is
causing
a
little
bit
of
change
and
pushing
projects
and
products
toward
becoming
more
declarative,
so
that
that's
one
aspect
of
how
I
see
the
the
coming
year
would
play
out
that,
because
of
the
interest
in
this
model
of
working,
the
products
have
to
morph
themselves.
Projects
have
to
morph
themselves
into
something
that
plays
a
lot
better
with
the
with
the
github's
workflow,
and
this
is
something
that
we
had
read
has
gone
through.
Wework
has
gone
through,
and
I
expect
a
lot
of
other
products
that
do
the
same.
B
There's
a
lot
of
talks
about
security
as
code
and
and
compliance,
and
these
technologies
quite
often
are-
are
not
really
compatible
with
the
get
offs
model,
so
that
that's
one
aspect
that
I
see
that
things
would
change
with
with
projects
adapting
themselves
to
to
heat
ups,
which
is
an
interesting
thing
and
and
and
the
other
thing
is
that
get
ops
is
really
despite
being
around
for
a
couple
of
years.
It
is
at
in
the
very
beginning
of
its
adoption
cycle
or
the
hype
cycle,
whatever
we
can
call
it
right.
C
Yeah,
I
I
sort
of
think
it
like
it
almost
forces
your
hand
right.
I
see
if
I'm
looking
at
my
crystal
ball,
I'm
I'm
saying
that
there's
gonna
be
re-emergence
of
devops,
because
things
like
kubernetes
and
github
forces
your
hand
into
that.
So
cornelia,
real
quick,
so
you
know
take
a
couple
minutes.
What
do
you
think
is
that.
F
Yeah
I'll
make
it
even
quicker
than
a
couple
of
minutes.
So,
first
of
all
I
want
to
clarify
that
I
I
would
call
it
the
adoption
cycle,
not
the
hype
cycle.
I
do
think
that
we
are
it's
still
in
the
the
skinny
part
of
the
tail
on.
You
know
the
the
the
crossing
the
chasm,
I
don't
think
we've
crossed
the
chasm
yet
I
think
that
we,
because
we
spend
so
much
time
in
this
space.
We
we
kind
of
delude
ourselves
into
thinking
that
this
is
more
widely
adopted
than
it
is.
F
C
F
And
so
we
had
those
systems
before
but,
like
I
said,
I
used
to
have
to
teach
people
about
convergent
systems,
and
now
I
don't
have
to
do
that
anymore,
so
that
happened
with
kubernetes.
That
was
you
know.
Kubernetes,
obviously,
is
a
hugely
successful,
open
source
project
and
we
can
spend
a
lot
of
time
thinking
about
what
were
the
elements
that
made
it
successful.
C
C
C
C
Yeah,
you
can
start
because
machine
learning
can
like
predict
when
you'll
have
spikes
in
your
cluster
and
it'll
start
preemptively
scaling
your
cluster.
Let's
say
so!
That's
that's!
Really!
That's
really
interesting
is
there's
that
ai
ml
ops.
You
know
that
so
that
that's
that's
really
really
cool.
I
also
see
something
like
that.
Coming
up
so
pinky
I'll,
give
you
the
floor
at
the
end.
What
what
what
you
just
now
now
it's
yours!
What
do
you
predict.
D
F
D
D
So
I,
when,
when
I
first
started
with
get
ops
like
progressive
delivery,
wasn't
even
a
thing
that
was
really
implemented
yet
for
the
whole
process.
And
then
you
know
things
like
flagger
came
along
and
and
so
there's
there's
still
things
that
are
so
cool
and
innovative
that
are
still
changing,
like
what's
possible
with
get
offs
and
so.