►
Description
Red Hatter Gerald Nunn shares his experience with ArgoCD Rollouts Canary Strategy and OpenShift Routes.
Learn More:https://argoproj.github.io/argo-rollouts/features/canary/
https://docs.openshift.com/container-platform/4.13/networking/routes/route-configuration.html
https://www.linkedin.com/posts/gerald-nunn_ive-been-trying-out-the-argo-rollouts-canary-activity-7090740177572163586-jODm?utm_source=share&utm_medium=member_desktop
A
A
A
B
B
But
yeah
I
there's
a
lot
of
good
things
coming
out,
so
when
openshift
I'm
bringing
that
up
because
you
know
like
openshift
is
One
release
behind
usually
so
we
might
I
think
we'll
see
that
in
4
15
and
so
then
that
kind
of
segues
into
414
is
going
to
be
released
in
a
couple
months.
There's
gonna
be
some
API,
deprecations
and
stuff
like
that.
So
make
sure
that
you're
keeping
up
with
those.
So
that
way
you
don't
get.
You
know
blindsided
by
something
that
you're
not
expecting.
A
A
Yep
right
well,
so,
let's
just
let's
just
get
the
show
on
the
road,
because
Gerald's
sitting
here
waiting
patiently
for
his
turn
so
Gerald
Nunn
joins
us
today
he
is
our
colleague
at
Red,
Hat
I'll.
Let
him
introduce
himself
a
little
bit
more
and
but
real
quick
first.
What
happened?
Was
he
posted
something
interesting
about
get
Ops
on
LinkedIn
I
was
like
that
looks
like
an
episode
topic
and
he
was
like
yeah
happy
to
do
it.
So
I
held
him
to
his
word.
A
A
C
So
we
get
the
same
weather
as
Seattle,
basically,
if
you're
trying
to
think
of
in
terms
of
climate
and
other
things
which
makes
it
one
of
the
warmer
spots
in
Canada
during
the
winter,
which
is
why
I
moved
out
here.
So
my
role
at
Red
Hat
is
I.
Am
the
get-offs
technical
marketing
manager
I'm?
Essentially
the
new
Christian
Hernandez
that
took
over
in
January
after
Christian
left
God
bless
him.
C
A
C
It
no
I
mean
so
in
openshift
get
Ops
1.9.
We
introduced
Argo
rollouts
as
a
tech
preview
feature.
So
if
you're
not
familiar
with
Rolex
I
got
a
few
slides
that
we'll
go
through
that
kind
of
talk
about
rollouts,
what
it
does
why
you'd
be
interested
in
it,
as
well
as
some
of
the
limitations
with
the
tech
preview
just
to
put
people
on
that
page
and
then
we'll
talk
about
how
to
use
rollouts
in
its
current
form
and
state
in
openshift,
without
service
mesh.
C
So
just
using
going
against
normal
openshift
routes,
how
that
works.
Some
of
again
what
some
of
the
limitations
are?
The
ins
and
outs
of
that
and
we'll
do
a
bit
of
a
demo
after
and
to
get
to
see
some
new
UI
in
the
openshift
console
that
I
have
been
pocing
as
well,
both
for
git,
Ops
and
rollouts.
So
a
bit
of
fun
on
that
front,
maybe-
and
it's.
A
Pretty
shiny,
I
I
forget
we
have
a
UI
right,
I
forget
we
have
one,
oh,
that
I'm
in
the
the
you
know,
openshift
on
Azure
space.
I,
really
don't
know
what
the
UI
stuff
looks
like,
because
it's
a
whole
different,
even
type
of
access
than
it
was
in
the
open
shift
on
AWS
things
that
I
used
to
work
on
so
I.
Forget
that
it's
there
I'm
excited
to
see
what
it
looks
like
these
days.
I,
don't
remember
the
last
time
I
accessed
it
really.
A
It
is
so
full
the
UI
is
so
helpful
for
learning
the
Paradigm
of
kubernetes
and
yeah.
It's
we
do
a
really
good
job
actually
on
our
UI
of
like
bubbling
up
and
aggregating.
The
important
logs
from
the
various
places
like
I
think
that
that
sometimes
gets
a
little
underappreciated
in
the
in
the
some
of
the
rest
of
the
noise
that
exists
in
the
UI.
A
This
is
how
you
do
all
the
things
you're
used
to
doing
in
the
UI
on
the
command
line,
and
then
that
was
just
like
you
said
it
was
a
very
natural
progression
to
pure
yaml
and
manifest
and
and
never
looking
at
the
UI
again,
but
I.
Don't
think
that
it
would
have
been
as
easy
to
mentally,
just
like
rock
without
that
visual
visual
aid
at
first
it
was
a
real
like
it's
an
accessibility
piece.
It
really
is
yeah.
B
It
gives
you
it
gives
you
a
picture
right
like
it
helps
you
paint
the
picture
of
what
your
application
looks
like
or
what
your
deployment
looks
like
if
you
have
more
than
one
application
and
I.
Think
that,
like
being
able
to
see
some
of
the
dependencies
right
like
oh
I've,
got
a
replica
set
that
might
not
be
spinning
up
a
pod
or
I've
got
a
deployment
that
might
not
be
spinning
up.
C
A
If
there's
an
operator,
you
can
see
that
okay,
this
Opera,
this
pod,
is
actually
controlled
by
this
operator
over
here
or
it's
controlled
by
this
other
thing
over
here.
That's
so
helpful
to
when
you're
going
through
the
debugging
steps
to
figure
out
where
certain
behavior
is
coming
from
to
know
kind
of
that
hierarchy
of
ownership,
yeah.
A
B
Sorry
I'm
sorry
I
was
going
to
say
a
Shameless
plug
for
for
backstage,
right
or
Janus,
IDP
or
or
red
hat
developer
portal.
Right
like
it's,
that
that
also
gives
you
that
ownership
of
the
resources
right
so
you'll
have
like
hey
here's
all
these
things
and
they're
owned
by
they're
everything's
owned
by
Gerald,
so
call
Gerald.
If
there's
a
problem
yeah,
that's
yes,
that's
all
I'm
gonna
label
at
2AM!
Yes,.
B
C
C
C
Yep
so
then
I'll
just
do
the
the
slideshow
here
and
you
can
still
see
them.
Yes,
all
right,
so
I'm
just
going
to
do
a
kind
of
for,
because
not
everybody
is
necessarily
going
to
be
familiar
with
Argo
rollouts
and
what
it's
good
for
and
why
you
should
use
it.
I
figured
we'd
just
do
a
quick
five
slide
overview
of
it
and
talk
about
it
just
to
set
the
stage
for
the
at
the
demonstration.
C
So
what
is
Argo
rollout,
so
Argo
rollouts
is
really
it's
a
drop
in
replacement
for
kubernetes
deployments,
but
with
a
lot
more
advanced
capabilities
around
Progressive
delivery.
So
you
think
of
a
kubernetes
deployment.
There's
really
two
strategies
in
a
deployment
replace
and
roll
out.
Argo
rollouts
adds
two
more
blue,
green
and
Canary,
which
are
common
deployment
strategies
for
deploying
your
application
and
we'll
see
each
of
those
in
the
the
subsequent
slides
it
can
integrate
with
a
variety
of
traffic
Management
Solutions.
C
So
if
you
deploy
this
against
service
mesh,
for
example,
it
can
use
service
mesh
to
automatically
shift
traffic
around
depending
on
whether
using
blue,
green
or
Canary.
It
will
manage
all
of
that,
for
you
automatically
right
so
I,
remember
back
in
the
old
days
when
I
was
doing
blue
green
by
you
know
two
data
centers
and
managing
the
load
balancer
by
hand.
C
You
test
your
rollout
and
make
sure
that
as
your
Canary
is
progressing
or
as
you
want
to
switch
from
blue
to
green,
that
what
you
deployed
is
actually
working
correctly.
So
that's
really
the
the
why
of
our
go
rollouts
in
terms
of
deployment
strategy.
So
there's
two
that
I
mentioned
blue
green
and
Canary.
So
blue
green!
Really,
you
know
you
have
initial
deployment,
the
blue
one.
Let's
say,
then:
you
go
ahead
and
you
deploy
a
new
version,
which
is
the
green
one,
but
nobody's
actually
connecting
to
it
from
a
user
perspective.
C
You
test
that
out.
You
make
sure
it's
good
to
go,
and
then
you
switch
the
load
balancer
over
to
use
the
green
one.
And
then,
when
you
get
to
the
end,
you
just
blow
away
the
blue
one
and
you've
only
got
the
green
one
left
right
and
then
you
can
do
another
deployment.
In
that
case,
new
deployment
might
be.
The
blue
one
Canary
is
really
more
of
an
advanced
version
of
blue
green
and
the
idea
behind
Canary
is
rather
than
just
doing,
a
sudden
switch
between
the
two
versions.
C
What
I'm
going
to
do
is
do
a
progressive
switch
I'm
going
to
step
the
number
of
users
between
each
deployment.
So
again,
I
start
with
the
blue
one.
That's
my
initial
version.
I
deploy
the
green
one
and
in
this
case
I
send
10
of
the
users
there
and
I'm
like
the
canary
in
the
coal
mine,
which
always
sounds
a
bit
cruel.
You
know
like
likening
this
to
the
canary
in
the
curl
mine
with
where
they
go
down
and
make
sure
there's
no
no
bad
gas
or
bad
air
and
the
canary
dies
in
this
case.
C
You
know
you
send
10
of
the
people
there
and
you
hope
that
that
doesn't
die
at
this
application.
Everything's
good
to
go
and
everything
is
good
to
go.
You
continue
progressing
to
more
and
more
users
until
you
finally
get
to
100
and
once
again
the
older
version.
The
Blue
version
goes
away,
and
this
is
all
managed
automatically
by
Rolex.
That's
really
the
beauty
of
it
is
I
kind
of
alluded
to
earlier,
trying
to
do
all
this
manually
with
a
load
balancer
and
doing
that
yourself
can
be
quite
a
painful
experience.
C
C
So
what
analysis
template
allows
you
to
do
is
integrate
the
rollout
with
a
metric
solution
right
or
a
metric
provider,
so
something
like
Prometheus
datadog,
a
webmetric
provider
Etc
that
you
can
use
to
actually
determine
the
status
of
the
role
in
so,
for
example,
when
I'm
rolling
a
canary
in
the
10
of
the
people
are
hitting
it,
I
can
use
an
analysis
template
to
go
hey.
Is
that
10
actually
working
or
is
it
not
working?
C
But
you
can
control
what
that
success
rate
or
failure
rate
is
for
you
is
it.
You
know:
95
success,
98
success,
99
success,
Etc
and
tailor
that
for
your
particular
application.
So
when
you're
running
a
Canarian
analysis,
there's
a
couple
of
different
modes
that
you
can
run
in
back
background
and
inline
background
runs
while
the
rollouts
progressing
in
line,
it
runs
as
a
discrete
step
and
canaries,
as
I
mentioned
earlier.
It's
all
about
the
steps
right
and
that's
really
where
those
templates
can
come
into
play
with
the
blue
green.
C
The
analysis
template
is
basically
run
on
a
pre
or
post
promotion,
so
pre-promotion.
It
can
be
really
useful
in
blue
green
to
run
as
a
test
I.E
nobody's,
hitting
the
green
one
yet,
but
I
want
to
test
it.
I
want
to
make
sure
that
there's
no
problems
with
it
before
I
switch,
my
load
balancer
over
and
then
post
is
kind
of
a
validation
right.
Essentially,
I've
got
users
on
it.
Now.
C
A
B
C
Ahead,
no,
that's
a
great
point.
The
fact
that
you
know
you
do
have
to
plan
for
the
additional
resources
that
this
deployment
strategy
takes
because
in
a
traditional
deployment
rollout,
you
know
the
the
number
of
PODS
that
you're
surging
at
any
time
and
point
in
time
is
relatively
minimal,
whereas
in
a
blue
green,
it's
like
two
complete
parallel
Stacks
right,
whereas,
as
you
alluded
to
also
with
Canary,
it's
more
kind
of
a
slow
Step
Up
step
down.
So
it's
easier
to
manage
the
resources
or,
if
you
have
less
resources,
to
go
to
Canary
deployment.
C
A
And
it
also
sort
of
depends,
I
will
say
a
little
bit
on
how
ephemeral
your
environment
is
or
can
be.
So,
for
example,
when
we're
looking
at
the
managed
openshift
fleets,
we
don't
create
or
destroy
anything
when
we
upgrade
those,
but
we
still
do
it
in
you
know,
there's
a
canary
which
is
essentially
becomes
what
they
call
a
sector
right.
A
The
the
new
release
version
runs
then
we'll
start
progressing
out
to
to
other
sectors
based
on
like
Risk
analysis
is
actually
something
that
we
do
so
not
to
say
that
you
guys
aren't
exactly
correct
in
terms
of
you
know,
this
practice
and
application
can
can
create
additional
load,
at
least
temporarily,
but
it
does
not
necessarily
have
to
so
that
is
a
little
bit
of
a
practice
thing.
A
sectorization
thing.
A
The
other
thing
I
wanted
to
mention
when
you're
doing
this
and
you're
you're,
trying
to
figure
out
like
what
percentage
should
I
set
for
the
advancement
right.
There's
usually
some
standard
practices
like
you
do
in
10,
15
25
like
something
like
that
right
and
continue
to
iterate
up.
You
might
want
to
look
at
the
size
of
your
Fleet
or
the
size
of
the
things
and
determine
what
level
of
risk
really
comes
with
with
this
and
then
so.
A
You
can
kind
of
try
to
basically
you're
going
to
try
to
create
a
manageable
level
of
incidents,
because
exactly
what
Gerald
says
you're
going
to
have
like
90
of
it
be
fine
and
10
is
going
to
be
a
mess,
and
that
might
still
be
enough
to
continue
to
progress
or
roll
out
depending
on
scale
or
what
have
you.
But
somebody
has
to
action.
A
The
thing
that
failed
so
as
you're
planning
through
you
know
all
these
things
and
what
metrics
and
and
how
much
at
a
time
and
what's
the
acceptable
failure
rate
that
will
still
allow
us
to
progress
to
the
next
phase
of
our
rollout.
You
need
to
kind
of
think
about
that
and
keep
that
in
mind
and
again
I'm.
Looking
at
this
from
a
perspective
of
a
managed
platform
and
a
managed
application
or
just
an
application,
that's
not
managed
you're
going
to
have
like
very
different
considerations.
C
Yeah
absolutely-
and
the
nice
thing
is
Johnny
pointed
out
as
well-
is
that
with
rollouts,
while
this
can
all
be
automated,
you
can
go
from
zero
to
100
from
completely
automated.
If
you
want
to
put
a
pause
at
10,
for
example,
and
have
a
human
look
at
things,
you
can
absolutely
do
that
as
well
right.
So
you
get
a
little
more
manual
intervention
if
it's
needed.
B
Yeah,
it's
it's
cheap,
CI,
CD
right!
So
we're
talking
about
a
little
bit
in
the
pre-show.
It
gives
you
a
chance
to
put
gates
in
and
go
and
check
things
out
and
then
like
okay,
let's
bump
it
up
to,
let's
add
a
little
bit
more.
Let's
add
a
little
bit
more
and
Tyler
made
a
comment
and
chat.
It
said
in
Bluegreen.
You
want
to
do
Network
mirroring
to
load
the
new
version
with
the
current
traffic
to
see
how
it
behaves
before
the
release.
Absolutely.
A
Yep
yep.
That's
definitely
that's.
Definitely.
Well
that's
one
blue
green
strategy.
It
depends
on
what
you're
trying
to
get
out
of
your
blue
green.
Exactly
so,
for
example,
you
might
do
blue
green
on
only
UI,
where
everything
else
is
the
same.
You
have
no
real
considerations
on
the
on
the
server
side
pieces
or
real,
significant
risk.
You
might
do
blue
green,
where
you
split
the
traffic,
because
you
want
to
see
if
you
result
in
better
conversions
right,
so
it
depends
on
depends
on
sometimes
a
blue,
green
or
Progressive.
C
Well,
everyone
else.
It
also
depends
on
whether
your
blue
green
is
talking
to
the
same
stateful
back
end.
You
can't
mirror
the
traffic
if
they're,
both
talking
in
the
same
database.
That's
just
going
to
lead
to
problems
right
so
generally,
the
blue,
green
I'm,
referring
to
from
an
application
perspective,
they're
all
connected
to
the
same
production,
stateful
back-ends,
so
mirroring
becomes
a
little
more
problematic,
I
think
in
those
scenarios,
but
always
interesting
hearing.
My
top
I
know
Tao
so
interesting
to
see
what
he
has
to
say
and
if
he
wants
to
comment
further
on
it.
C
B
C
C
C
A
C
Cool
okay,
so
the
in
terms
of
the
limitations
with
the
current
Tech
preview,
so
traffic
management
from
a
service
mesh,
Works,
100,
no
problem
at
all
and
blue
green
on
openshift
Route
will
work
fine
because
it's
simply
doing
switching
of
the
the
services
for
the
that
the
roads
connected
to.
So
there's
no
problem
with
that
as
well,
but
Canary
is
essentially
working
on
a
best
effort
perspective.
It
does
not
do
active
traffic
management
with
an
openshift
Road
at
the
moment.
That's
a
gap
that
we
want
to
close
before.
C
We
go
ga
with
the
product
what
it
means
by
best
effort
and
we'll
see
this.
The
demo
is
essentially
what
it'll
do
is
they'll
scale
pods
right.
So
if
you
have
a
a
stable
and
a
canary
and
the
canary
is
a
10
and
let's
say
your
pod
pool
is
10.
The
canary
might
get
two
pods
the
stable
get
eight
The
Next
Step
the
canary
gets
four.
The
stable
gets
six
and
so
on,
and
so
on.
C
Right
and
it'll
sort
of
do
an
artificial
Canary
based
on
that,
and
that
was
really
what
the
the
LinkedIn
post
was
highlighting
that
it
actually
worked.
Surprisingly,
well,
when
I
tested
it
out,
at
least
for
my
relatively
stateful
stateless
application,
that
I
was
testing
at
that
point.
So
in
terms
of
traffic
management,
that's
the
limitations
there
and
then
the
other
limitations
that
I
wanted
to
mention
here
is
that
we're
only
supporting
namespace
scope
for
rollouts
at
this
time,
rather
than
cluster
scope.
C
C
So
there's
some
benefits
to
that,
and
then
the
other
one
that
I
want
to
highlight
is
that
for
better
or
worse,
the
Upstream
rollouts
does
not
support
authentication
against
Prometheus
and
unfortunately,
well,
unfortunately,
for
rollouts,
but
good
as
a
practice
and
openshift,
we
require
authentication
to
interact
with
the
the
Prometheus
or
Thanos
in
an
openshift
cluster.
Now
you
can
use
the
web
metric
provider
as
a
workaround
for
that
which
essentially
goes
against
any
kind
of
web
URL
and
Prometheus
and
Analysis
in
the
day
or
just
a
rest.
C
Api
that
you
can
make
a
web
call
against,
but
it
is
a
little
more
convoluted
and
involved
to
do
that
versus
just
using
the
previous
metric
out
of
the
box
and
again,
this
is
something
that
will
be
addressed
before
we
go
ga
with
our
go
rollouts
in
openshift
get
off.
So
those
are
the
key
points
that
I
just
wanted
to
highlight
and
that's
it
for
slides
any
other
questions
on
this
Johnny
or
Hillary
before
I
move
on
when.
C
Don't
have
a
date
for
it,
so
our
product
manager
is
working
on
coming
up
with
a
calendar
of
a
bunch
of
the
current
Tech
preview
features
in
terms
of
when
we're
planning
on
gaing
them,
but
don't
think
she's
released
that
as
of
yet
so
I
have
to
get
back
to
you
on
that
one.
Okay,
but
you
know
if
I
was
evil.
I
would
commit
to
a
date
and
tell
until
I
Harriet
that's
the
date.
I'm.
A
C
C
Yeah
so
there's,
unfortunately,
there's
no
date
at
this
time.
There
are
a
number
of
things
that
we
need
to
close
from
an
Enterprise
perspective
in
rollouts
before
we
can
GA
it,
and
some
of
them
are
mentioned
here,
there's
a
few
others
that
I
haven't
talked
about.
So
it's
really
where
our
focus
is
going
to
be
in
the
next
few
months
to
try
to
close
close
those
gaps.
A
C
B
C
Name,
it's
just
a
CR,
a
custom
resource
and
the
nice
thing
with
rollouts.
It's
a
very
lightweight
controller.
It's
not
like
openshift
get
Ops
or
Argo
CD
where
there's
like
you
know
five
different
pods
that
need
to
run
in
order
to
do
something.
It's
just
like
one
pod
right.
So
it's
not
really.
From
my
perspective,
a
big
deal
to
do
that.
Gotcha.
A
A
A
B
You
know
one
thing
like
on
on
astronaut,
Bishop
Diamond,
but
I
feel
like
that's
something
that
goes
really
well.
It
comes
across
really
well
because
a
lot
of
people
are
interested
like
how
do
I.
How
do
I
fix
this
when
it
breaks?
You
know
what
I
mean
and
when
you
do
a
cooking
show
demo
and
you're
like
hey.
Look
it
just
works
you're!
Welcome
right!
You
don't
get
to
see
all
the
the
sausage
being
made
right,
so
don't
get
to
see
all
the
pain
and
the
typos
and
stuff
like
that.
B
C
So
for
the
demo,
I've
got
four
namespaces
that
will
be
focused
on
cicd
I've
got
a
pipeline
that
will
run
a
Dev
namespace
for
the
applications
deploy
just
a
regular
deployment,
no
rollout
a
git
Ops
namespace,
because
this
is
all
being
managed
by
Argo
CD
and
then
the
prod
name
space,
where
the
rollout
is
actually
working
or
running
the
rollouts.
So
you
can
see
here
Johnny
and
the
get
Ops
operator.
If
you
go
to
the
demo
prod
namespace,
there's
a
roll-up
manager
here
called
Argo
rollout.
We
can
go.
C
Look
at
that
yaml
and
you
can
see
it's
it's
a
a
nothing
burger
yaml
right,
super
easy
to
deploy,
ignoring
all
the
craft
that
kubernetes
likes
to
add
on
to
any
any
resource.
But
you
know
spec
wise,
there's,
really
nothing
there
right,
it's
more
of
a
marker
to
to
get
the
controller
to
deploy
it.
So
it's
really
easy
to
get
up
and
running
in
terms
of
the
controller
and
then
the
pods
will
work
from
there.
C
C
I've
got
two
rollouts,
a
blue,
green
and
a
canary
rollout,
and
it
gives
you
a
bunch
of
different
information
like
the
canary,
for
example,
you
can
see
the
different
steps
that
I've
got
set
up
in
that
Canary
Etc
and
you
can
generally
do
everything
you
want
in
this
UI.
But
it's
this
UI,
as
you
can
tell
up
here,
is
running
locally
right.
This
is
not
a
UI
that's
deployed
in
openshift.
There
is
an
extension
available
for
Argo
CD,
but
I
believe
at
the
current
moment,
nope
shift
get
Ops.
C
It'll,
be
amazing
and
you
can
see.
I've
got
two
rollouts
here,
blue
green
and
Canary
that
we
saw
earlier
and
the
idea
really
here
is
they
both
get
updated
with
the
same,
manage
you'll
both
see
them
go
at
the
same
time,
but
it's
really
so
that
when
I'm
doing
this
demo,
people
that
I'm
doing
the
demo
for
can
compare
and
contrast,
blue,
green
and
Canary
and
how
they
work
right
at
the
same
time
so
from
a
CI
perspective,
I'm
going
to
go
ahead
and
kick
this
process
off.
C
After
all
that
explanation,
but
any
questions
yeah,
that's
basically
going
to
be
covered.
Oh
maybe
I
should
go
in
this
a
bit
more
down
here.
First.
So
if
you
go
down
here
and
you
see
the
blue
green,
for
example,
it
gives
you
some
basic
information
about
the
the
rollout,
but
you
can
see
down
here
all
the
different
revisions
that
have
happened,
and
you
can
see
that
my
stable
and
my
active
services
are
pointing
to
the
same
revision
I.E.
This
has
all
progressed
and
everything's
good
to
go,
and
it's
the
same
with
the
canary.
C
If
I
go
look
at
the
canary,
we
can
see
the
revisions
here
and
you
can
see
my
stable
and
Canary
are
pointing
to
the
same
revision
from
a
service
perspective.
Nothing's
progressing
nothing's.
You
know
happening
right
now,
so
this
is
kind
of
what
it
looks
like
when
it's
all
stable
at
this
point.
So
right
now,
if
I
go
look
at
the
the
applications
deployed
from
a
route
perspective,
you
can
see.
C
I've
got
different
routes
here
for
the
active
and
preview
for
blue
green
and
then
I've
got
different
routes
for
Canary
and
stable
I
also
have
a
canary
all.
So
what
the
canary
all
does
is
it
selects
all
pods
for
the
canary
and
is
essentially
providing
the
view
that
a
user
hitting
the
application
would
see?
So
if
I
have
10
Canary
90
stable,
you
will
see
that
in
the
in
this
Canary
at
all,
whereas
if
I
hit
Canary
or
stable
I
would
see
either.
C
This
is
just
the
stable
or
just
the
canary
it'll
be
easier
when
I
do
the
demo
and
you
can
actually
see
it
running
but
from
an
application
perspective,
it's
the
standard,
rollouts
demo
app
you
get
these
colored
squares
when
you
deploy
a
new
version
of
the
application
we'll
deploy
it
with
the
green
you'll,
see
the
green
colored
squares
pop
up
here
instead
of
the
blue
ones,
as
we
go
through
it.
So
that's
just
kind
of
a
quick
run
through
of
what
we're
going
to
see
as
we
go
through
it.
C
B
C
Yeah
no
you're
exactly
right,
so
you
can
see
you
got
replicas,
you
got
selector.
The
strategy
is
obviously
replacing
the
deployment
strategy.
But
if
you
go
down
here,
you
get
the
Pod
template
exactly
the
same
as
the
Pod
template
for
deployment
right.
So
that's
really
why
I
kind
of
highlighted
as
being
a
drop
in
replacement
for
deployment.
Yes,
there
are
some
differences
between
the
two
in
terms
of
the
spec,
but
you
know
conceptually
it's
essentially
the
same
and
spec
wise.
A
lot
of
it
is
identical
between
the
two
awesome.
C
C
Yeah,
it's
important
to
understand,
though
it's
not
using
the
deployment
to
manage
the
rollout,
like
the
rollout,
will
still
do
its
own
rollout
to
have
its
own
pods,
its
own
replica
sets,
but
it's
just
sucking
in
that
template
from
the
deployment.
So
that's
why
I
say
I
don't
know
like
when
would
I
have
a
deployment
deployed
in
parallel
that
I
want
to
suck
in
that
template
from?
C
What's
the
use
case
for
that,
but
if
I
could
see,
maybe
if
you're
migrating
from
one
to
the
other,
where
it
could
be
useful
but
yeah
anyways,
neither
here
nor
there
if
you
do,
if
somebody
on
the
call
or
on
the
the
on
the
session
has
an
idea
with
or
can
outline
what
a
good
deployment
case
is
for
a
good
use
case,
for
that
is
feel
free
to
chip.
In
on
the
comments.
B
A
A
Oh
as
I
say
as
long
as
you
think,
I
think
I
mean
yeah
you're
onto
something
with
the
migrations
or
or
something
like
that
they're.
It
might
be
really
useful
for,
if
you
want
to,
if
you
don't
necessarily
own
the
thing
that
you
want
to
roll
out
right,
so
somebody
else
owns
it.
It's
it's
somebody
else's
like
open
source
project
or
whatever,
and
they
have
a
deployment
for
it.
But
you
want
to
go
through
a
rollout
strategy
with
it.
A
C
C
C
B
C
And
then
there's
pros
and
cons
to
that
approach
and,
like
I,
said
it's
a
whole
separate
subject
that
we
could
spend
20
minutes
on
no
problem.
So
let's
table
it
to
the
end,
if
there's
time,
I'm
happy
to
show
it
and
talk
about
it,
you'll
see
a
little
bit
of
it,
though,
as
I
go
through
the
demo,
all
right
so
I'm
going
to
kick
off
this
pipeline
here
so
currently
we're
on
blue,
as
we
saw
so
I'm
just
going
to
change
this
to
Green
and
I.
C
I
need
to
work
on
my
mic
more
apparently
yeah,
because
I've
got
a
nice
directional
mic
now
that
particularly
because
I
have
a
mechanical
keyboard
and
it
was
too
loud
for
the
mic.
I
was
using
before
anyways.
This
kicks
off
a
simple
pipeline
that
will
deploy
the
application
into
the
dev
environment
and
then
the
prod
environment.
It
basically
just
does
a
update
of
the
git
repo
in
in
Dev
and
then
has
Argo.
Do
a
sync
of
it
and
it'll
do
the
same
thing
and
prod.
C
C
So
now
we
can
see
I've
got
my
rollouts.
Are
things
are
happening
here,
the
canary
ones,
in
a
pause
State
and
the
blue
green
is
progressing.
So
we
go
look
at
blue
green.
You
can
see
that
I've
got
my
analysis
run
running
here
and
you
can
see
that
there's
a
preview
now
on
the
newest
revision,
so
the
preview
Services
going
against
that
and
the
stable
active
revision
is
still
the
the
revision
20..
You
can
see
this
thing
here,
that's
turning.
C
Tell
me
if
you've
got
any
400s,
500s
Etc
and
if
you
don't
it's
all
200s,
that's
great
at
pass
and
you
can
see
it's
doing
different
tests
here
and
there's
actually
four
tests
and
they
pop
in
as
they
happen,
the
other
one
that's
running
is:
let's
see,
does
this
highlight?
Oh,
let
me
go
up
here.
There
we
go
it's
it's
a
job,
that's
running
and
I'm
running
Apache
Siege
to
generate
load
against
this
application.
C
Now,
when
you
actually
open
this
application,
it
is
a
restful
application
that
makes
API
calls
it
gets
it's
back
in,
so
it
will
generate
its
own
load,
but
I
kind
of
wanted
to
have
the
siege
thing
running,
because
I
want
to
go
well
if
I'm
running
an
application
that
doesn't
generate
its
own
load,
which
is
most
applications.
How
can
I
do
that
right?
I
want
to
test
that
out.
C
You
can
see
that
the
analysis
run
past
and
and
now
I
can
look
at
these
and
I
can
see
the
preview
stable
and
if
you
go
look
at
the
routes
back
here.
Let
me
look
at
active.
You
can
see.
The
active
is
still
green,
but
if
you
go
look
at
preview
or
sorry,
it's
still
blue,
but
you
look
at
preview.
It's
green
right!
So
now
we're
actually
we've
got
no
user
traffic
actually
going
to
this,
because
all
the
user
traffic
is
going
to
stable.
C
C
So
you
can
see
it's
progressing
again
and
you
can
see
now
that
there's
a
post
analysis
phase
that's
happening,
but
the
active
service
is
now
the
blue
service
right.
So
that's
what
people
are
going
to
be
seeing
when
they
start
hitting
this
and
it'll
progress
and
then
once
it
finishes
the
post
testing
here.
Another
three
runs:
this
stable
will
move
back
up
there
and
we'll
be
all
blue
from
a
blue
green
perspective.
C
So
this
is
how
the
blue
green
is
actually
working
in
terms
of
you
know,
enabling
users
to
quickly
and
easily
run
those
parallel
stocks
have
infrastructure
and
have
be
able
to
manage
that
more
effectively.
If,
for
some
reason
there
was
a
problem,
you
can
actually
do
a
rollback
on
this
and
I'll
show
this
a
bit
later,
because
I've
got
to
turn
off
Argo
CD.
C
In
order
to
do
the
rollback,
otherwise
I'm,
you
know
fighting
with
Argo
to
try
to
get
it
to
do
it
to
go
against
git,
which
is
never
a
good
good
thing
to
do.
But
you
know
doing
the
rollback
is
pretty
quick
and
easily
the
other
nice
thing
I've
got
in
the
UI
here.
Is
you
can
see
what
image
is
deployed
with
each
revision?
C
So
here
you
can
see
it's
the
green
image
and
down
here
you
can
see
it's
the
blue
image
right
just
by
hovering
the
mouse
over
it
and
if
the
rollout
had
more
than
one
container
you'd
see
more
of
these
boxes
with
the
different
container
names
and
you
can
hover
over
them
to
see
them.
So
you
can
see
now
now.
This
is
all
done
so
the
next
one
to
look
at
is
then
canary.
C
So
we
go
look
at
our
Canary
here.
We
can
see
that
we've
got
the
the
old
revision
which
is
stable
and
the
new
revision
which
is
Canary,
and
similarly,
if
I
go
look
at
my
routes
again.
If
we
do
the
stable,
we
can
see
we're
blue
I'm
going
to
start
turning
some
of
these
off,
because
I've
got
too
many
of
these
things
going.
C
This
is
the
where
I
kind
of
grab
the
screenshot
right
this
graph
down
here,
showing
you
the
split
between
the
the
blue
and
the
green
services.
So
the
reason
why
the
canary
isn't
progressing
if
I
go
back
to
my
rollouts
and
we
go
look
at
the
yaml
is
I've
got
a
step
here.
That's
just
a
pure
pause
right,
so
I've
got
20
traffic
going
to
Canary
and
I've
just
got
this
pause.
So
in
order
to
move
past,
this
pause
I
actually
need
to
do
the
promotion,
so
go
back
here.
C
I've
got
in
each
of
those
steps
to
kind
of
give
the
system
time
to
accommodate
and
get
used
to
things,
and
as
it's
going
through
things,
you
can
see
that
the
green
squares
will
start
becoming
more
and
more
versus
the
blue
squares
right
until
finally,
it's
just
100
green
square
is
running
at
that
point
in
time.
Yeah
always
for
some
reason,
even
though
there's
no
traffic
Interruption
it
kind
of
blocks
out
here.
If
you
aren't,
don't
have
the
the
app
visible
but
anyways.
C
You
can
see
here
how
it's
progressing
right,
So,
eventually
we'll
get
to
100,
Canary
and
nope,
no
nothing
happening
on
blue
yep.
There
we
go
so
that
is
kind
of
just
quickly
showing
the
the
blue
green
and
you
know
and
the
canary
and
how
they
work
together
or
have
not
work
together.
But
the
the
difference
between
the
two
in
terms
of
the
style
of
deployment
that
you're
doing
right,
the
canary
is
really
nice.
C
In
their
terms
of
being
like,
we
said,
risk
management
being
able
to
step
up
the
the
change
that
you're
making
and
exposing
you
know
only
a
small
percentage
of
your
community
to
it,
so
that
you
know,
if
there's
a
problem,
if
you're
doing
something
that's
particularly
risky
in
terms
of
a
new
feature
or
a
new
technical
technology
change,
you
can
kind
of
manage
that
more
effectively
than
a
full-blown.
You
know
we're
just
going
to
do
a
hard
cut
to
the
new
version,
so
I'll
pause
there,
Johnny
Hillary
any
questions
so
far.
B
C
A
C
C
There's
no
pause,
nothing,
but
then
there's
a
pause
step
after
and
we're
saying:
okay,
hard
pause
here,
don't
do
anything
until
somebody
promotes
this,
and
then
you
can
see
as
we
set
the
weight
to
higher
numbers
and
increments,
we're
doing
slight
pauses
at
each
step,
just
to
kind
of
give
things
a
chance
to.
What's
the
word
bake,
you
know
when
you're
kind
of
making
changes
like
this.
Sometimes
you
just
want
things
to
kind
of
settle
out
a
bit
and
give
it
more
time
now
for
the
purpose
of
the
demo.
C
B
C
A
C
C
Yeah
there
we
go
so
now
you
can
see
a
little
better,
promote
undo
and
it's
just
it's
a
normal
Cube
cuddle
plug-in
right.
So
you
just
copy
the
command
into
a
directory
with
like
with
Cube
cuddle
or
it's
not
a
cube,
cuddle,
sorry,
not
with
Cube
cuddle,
but
it's
on
the
path
and
then
it'll
just
show
up
automatically
when
you
do
OC
or
cube
cuddle.
B
C
C
Yeah
so
on
the
analysis
run
just
to
show
that
a
bit
as
well
so
remember
when
I
said
that
the
I
have
to
use
the
web
metric
instead
of
the
yes
Prometheus
metric.
Well,
one
of
the
side
effects
of
that
is.
You
need
to
URL
in
code
the
parameters
that
you're
sending
with
the
webmetric
right.
You
can't
just
have
a
nice
kind
of
here's.
C
My
Prometheus
query
copy
it
right
out
of
the
the
metrics
tab
in
the
open,
show
console
paste
it
right
in
you
got
to
do
this
little
extra
step
of
URL
encoding
it
and
the
other
side
effect
of
that
is
that,
in
terms
of
getting
the
the
data
out
in
terms
of
the
success
condition
and
reading
it
there's
a
little
more
effort
because
you're
getting
a
Json
body
back
and
you've
got
to
kind
of
figure
out.
Okay.
What
is
that
Json
body
look
like
and
where
do
I
pull
that
metric
out
of
the
response
right?
C
C
C
C
A
Yeah
I,
don't
I,
don't
know,
I
have
no
idea.
Actually,
this
is
one
of
those
things
like
I.
Don't
understand
how
this
deviation
occurred,
at
least
with
the
like
the
lack
of
you
after
O's
in
American
spellings
at
least
I
know
like
the
history
of
that
one
right,
but
I,
don't
know
how
we
got
that
deviation.
That
was
just
that
one's
bizarre
to
me.
C
So
you
can
see
it's
progressing
and
they're.
Just
gonna
wait
for
it
to
get
into
the
the
pause
here.
But,
okay,
there's
one,
that's
positive!
You
go
look
at
I'll
show
a
bit
of
the
the
get
Ops
thing
here.
We'll
go!
Look
at
our
applications
under
get
Ops,
so
you
can
see
my
applications
are
here
and
the
prod
one
go.
Look
at
that
and
the
resources.
So
you
can
see
that
the
canary
resource
is
suspended.
C
So
Argo
understands
the
state
of
the
rollout
in
terms
of
what
that
pause
really
means,
and
that's
why
it's
showing
suspended
here-
and
this
is
all
done
through
Argo
CD
health
checks
right
so
I
think
you
both
are
familiar
with
Argo
City
health
checks
that
you
write.
You
write
a
little
bit
of
Lua
code
and
check
whatever
resources
you
want
in
terms
of
its
health
status.
C
So
there's
out
of
the
box,
Health
statuses
for
rollouts,
and
it
will
know
that
if
that
is
in
a
pause
state
that
it
will
return
a
suspended
status
to
Argo
CD
right
and
if
that
rollout
fails,
it'll
get
a
degraded
status
as
well.
When
you
go
to
deploy
it.
So
you
can
actually
see
it
interacting
with
Argo
CD,
in
terms
of
the
statuses
that
the
rollout
has
as
it's
going
through
this
this
process.
B
That's
awesome,
yeah.
The
other
thing
I
was
going
to
say
too
about
that
is
like
oh
in
the
actual
Argo
console
you
can,
you
can
pause
or
suspend
or
you
know,
roll
back
or
promote
the
actual.
You
know
test
right
like
the
canary
or
the
blue
green.
So
it's
it's
pretty
neat
like
you.
There's
a
little
button
inside
that
you
can
click
on
that
and
say
all
right,
yep
I'm
good
to
go.
I
wanna
I
want
to
oh
you're
gonna.
C
B
B
A
B
C
I
was
gonna
say
it's
funny,
though,
is
somebody.
That's
writing
a
UI
yeah
for
this.
It's
like
I,
look
at
their
UI
and
it's
like
amazing
how
much
it
does
actually,
when
you
start
drilling
into
it,
with
all
the
different
options
and
things
that
are
under
the
hood.
That
need
to
happen
to
present
these
views
right
so
but
yeah,
here's
the
different
options,
you're
referring
to
Johnny
right,
so
you
can
abort
it.
You
can
promote
it
full.
You
can
restart
it
resume
it
Etc.
C
B
Awesome
and
the
reason
why
I
brought
that
up
too
right
is
because,
if
you
give
somebody
through
our
back
like
access
to
this
right
and
say,
hey
look,
you
can
manage
your
applications.
You
can
manage
your
rollouts
and
stuff
like
that.
They
can
do
all
of
it,
but
you
don't
want
them.
Maybe
to
have
cluster
access
right.
You
only
want
them
to
have
Argo
access.
B
C
C
Okay,
let
me
go
back
here.
Let
me
go
back
to
the
Rolex.
Now
I've
got
to
progress
it
just
to
do
the
roll
back.
Let's
progress
both
promote
promote,
so
this
will
start
going
through
and
I'll
just
the
blue
green
one
of
them
will
hopefully
finish
relatively
quickly
and
then
we'll
do
a
roll
back.
But
to
do
the
rollback
as
I
mentioned,
I've
got
to
turn
off
Argos,
we'll
just
do
that
really
quickly.
C
Here
turn
off
the
automated
test:
sync
policy
down
there
so
that
we
can
actually
make
that
rollback
change
and
if
I
go
back
to
the
rollouts,
let's
see
if
it's
done
not
quite
yet.
This
is
the
problem.
When
you
try
to
pick
an
analysis
run
and
it
takes
too
long
yeah,
it's
like
come
on.
Oh
one
more
left
and
then
I
gotta
wait
for
the
job
to
finish.
I'll,
try
to
oh
finish,
the
job.
A
C
Oh
there
we
go
so
this
one,
that's
stable
is
yellow
now,
actually,
let's
go
and
have
a
look
at
that
real
quick
routes,
active
yep,
so
it's
yellow
if
I
go
back,
oops
go
back
here
to
the
rollouts
and
we
just
do
a
rollback
on
the
no
I
was
doing
on
blue
green
and
we
go
back
to
the
the
green
one.
So
I'll
do
a
roll
back
on
that.
So
you
can
see
what
it's
really
doing
is
just
doing
a
roll
forward.
C
So
it'll
go
through
the
same
analysis,
template
perspective
and
it'll
roll
that
out
as
the
new
version
and
again
this
will
take
a
few
minutes
to
to
go
through,
but
you
can
see
how
straightforward
and
simple
it
is
to
do
a
roll
back
just
like
it
is
for
deployments
right.
It
follows
the
exact
same
methodology
and
process.
So
if
you're
familiar
with
rollbacks
and
deployments,
you'll
be
familiar
with
them
in
nargo
rollouts
as
well.
B
A
A
C
So,
from
a
get
Ops
product
perspective,
we're
very
interested
in
feedback
on
rollouts
and
what
people
are
looking
for
in
terms
of
additional
capabilities
and
features,
you
know
closing
gaps
from
an
Enterprise
perspective,
because
there's
always
things
that
customers
need
that.
We're
not
may
not
necessarily
be
aware
of
at
first
blush
right.
So
if
there
are
folks
that
are
listening
to
this,
that
are
interested
in
trying
it
out
by
all
means,
do
do
please
do
try
it
out
and
do
please
feed
us.
C
You
know
your
feedback
in
terms
of
things
you're,
looking
for,
because
we're
definitely
interested
in
hearing
more
about
how
to
do
things
and
one
of
the
things
we're
interested
in
is
UI.
What
do
you
want
us?
What
do
you
want
to
see
for
UI?
Should
it
be
in
the
Argo
UI?
Should
it
be
in
the
openshift
console?
Should
it
be
in
the
red
hat
developer:
Hub,
there's
lots
of
different
places
where
it
could
end
up
right
and
with
limited
people
in
engineering
to
do
this
work
in
lots
of
different
priorities
floating
around.
A
C
C
We
doing
yeah
no,
if
you're
a
customer
open
a
support
case
and
just
provide
feedback
through
that
and
I
will
get
funneled
to
the
product
manager.
Alternatively,
you
can
reach
out
to
myself
directly
I,
don't
I,
never
mind
hearing
from
customers
or
you
can
reach
out
to
the
product
manager
Harriet
Lawrence
directly
as
well,
but
I
won't
post
your
email
address
here,
but
you
know
as
a
starting
entry
point.