►
Description
Discussion for https://gitlab.com/gitlab-org/charts/auto-deploy-app/-/issues/70.
A
A
B
A
Customers
which
support
a
future
request
of
coloring
where
support,
and
we
need
to
introduce
architecture,
change,
to
resources.
This
Roxy
Blake's
compatibility.
However,
we
see
what
to
ship
this,
which,
as
a
new
standard,
is
slowly
maglite
users
from
old
version
to
version.
So
let
me
explain
this
issue
a
bit
before
study
this
issue.
A
This
is
there
are
bunch
of
issues
about
like
enabling
nginx
sorry
canary
in
glass,
into
the
oddity
Bob's
deployments
and
to
allow
users
and
to
use
advanced
request
control,
for
example
like
here,
a
couple
of
parameters
can
be
assigned
to
canary
ingress,
for
example,
if
users
set
specific
header
a
canary
of
a
prince
of
Canary
noble,
then
the
requests
always
goes
to
kind
of
reused
us
and
it's
a
kind
of
similar
to
kill
up
next
right.
You
said
cookie
and
then
you're
always
access
to
canary
instance.
A
The
the
similar
thing
is
achieved
in,
as
you
begin
today's
dirty
Bob's
as
by
controlling
the
weight
of
parts,
the
like
scaling
up
parts
in
production
and
scaling
it
down.
Assessors
scaling
down
the,
for
example,
roll
incremental
rollout
situation,
we
at
first
scaling
up
canary
parts
and
then
next
scaling
down
the
the
parts,
a
production
as
the
stable
truck.
A
A
Yes,
so
so
that's
a
point
why
we
want
to
introduce
contouring
less
and
actually
we
have
a
community
contribute.
Are
three
community
contributors
here
like
it
is
a
actually
trying
to
install
canary
in
class
into
our
charts.
As
we
had
a
discussion,
it's
bit
tough
I
guess
I
mean
what's
the
biggest
concern.
Is
that
likely
this
breaks?
A
Because
this
is
in
setups,
so
it
breaks
backward
compatibility,
and
now
we
just
have
discussed
for
a
long
time,
but
we
haven't
had
any
solutions
yet
so
maybe
like
we
can
almost
conclude
that,
but
we're
breaking
changes
almost
inevitable.
So
we
need
something
in
the
new
new
architecture
to
introduce.
Let
us
easily
introduce
a
flaky
change
by
versioning
yeah.
B
A
So
architecture
changes,
so
this
is
like
today's
architecture
and
Joe
managed
up
in
G,
but
first
is
a
interests
or
engineering
class.
It
is
in
X
English
in
G,
L
managed
a
namespace
and
it
reads:
G
production,
a
namespace
and
it
has
English
and
interprets
this
service
and
its
roots
request
pots
and
stay
with
track
or
kind
of
a
truck
or
a
roll.
A
truck.
A
And
the
proposal
architecture
is
the
basically
installed
canary
English
here.
So
here's
the
same
jail
managed
to
have
engineering
singles
and
it
reads
the
production
of
space
in
class
and
it's
the
same
and
here's
a
new
one.
We
install
Connor
in
class-
and
it
is
kind
of
English-
basically
does
a
bunch
of
like
if
the
conditional
switching
so,
for
example,
if
canary
heads.
A
Like
a
canary
in
a
board
parameter
is
exists
in
weakest
header,
then
it's
routes
request
canary
track
if
it
doesn't
exist
and
which
to
stay
blue
truck
or
maybe
like
there's.
A
weight
based
strategy
is
defined
here
so,
like
40%
of
traffic
goes
to
Henry
in
a
success.
60
percent
goes
to
stable,
something
like
that
and
then
the
basically.
This
is
also
a
new
one.
A
A
A
A
A
A
This
probably
likely
be
deprecated
and
he
said
we
maintain
the
charts
in
how
to
deploy
which
project
we
no
longer
release
new
charts
to
charts
Caleb
I.
Oh
yes,
the
the
currently
published
charts
are
sorry.
Let
me
ask
a
question
like
today:
Oh
dirty
plumage
fetch
the
charts
from
charts
Caleb
I.
Oh
I,
understand
that
this
is
because,
when
we,
when
we
introduced
or
to
deploy
how
to
DevOps,
it's
already
relied
on
chart
skill
of
IO.
But
you
cannot,
you
know,
I,
understand
correctly
it's.
C
No,
we
don't
have
to.
In
fact
we
are
not
yeah.
We
have
an
option
for
people
to
have
their
own
charts
checked
in
the
repo
in
under
slash
charts.
So
if
they
already
checked
in,
we
use
that
local
version
so
I
think
this
is.
This
is
a
relic
history
thing.
Where
previously
we
don't
have
a
docker
image.
We
just
do
everything
in
gillip
CRM,
oh,
so
that's
why
we
need
to
fetch
it
because
we
don't
have
a
place
to
put
it.
A
C
B
C
A
B
A
B
B
A
A
C
A
B
A
C
D
A
B
A
Yeah
so
here's
my
thought
on
this
diversion
image.
Tocqueville
deeply
machine
trouble
to
play
the
octopus
template
points;
tv0
not
like
exact
version,
a
specific
version,
but
just
v
zilla,
and
we
just
it's
a
latest
version
of
zero
point:
zero
point:
X,
dot,
X
and
their
version
of
OD
blemish
in
template.
It
could
be
updated
to
be
one
and
at
the
killer
of
major
or
minor
relays,.
A
The
major
versions
between
previous
step
and
the
ongoing
setup
doesn't
much
if
the
major
versions
doesn't
match.
Then
we
show
a
warning
to
users
which
are
in
a
pipeline.
Fears.
I'll,
show
two
options
up
create
is
the
latest
major
version
by
following
a
manual
upgrade
guide,
I
yeah,
so
this
is
interesting.
I
didn't
realize
that
we
have
such
things.
B
B
A
B
C
D
We
can
also
use
whip
in
the
thing
we've
been
talking
about
in
a
different
thread
around
putting
new
users
on
the
new
thing
as
soon
as
we
feel
it's
ready
for
that.
So
you
know
when
you
make
a
breaking
change,
it's
not
actually
a
breaking
change
if
the
person
has
never
used
the
thing
before
so
that
would
give
us
the
ability
to
iron
out
a
lot
of
problems
before
having
to
switch
people
who
have
stable
work,
clothes.
B
C
A
D
At
the
moment
we
can
make
American
changes
whenever
and
like
if
you're,
following
the
strict
settler
contract,
it
doesn't
matter,
I
think
like
putting
it
to
1.0
is
just
saying
we
think
this
is
stable,
like
so
I
think
we
could
release
1.0
right
now
of
all
of
them
and
say
this
is
the
like
and
actually
say
from
now
on,
it's
frozen
only
bug
fixes
this
is
stable,
but
now
we're
gonna
and
then
the
next
one
that
we
bundle
with
thirteen
two
might
be
no.
No.
D
C
Both
alter,
deploy,
image
and
out
of
app
will
become
one
point
over
okay.
It
just
it
just
means
that
the
our
detecting
detection
script
would
have
to
detect
for
V
0
and
V
1.
Well,
that's
fine!
We
just
actually
we
don't.
We
just
say
it's
not
unless
than
two
that's
fine.
It's
not
a
decimal
point
at
all.
I,
like
it
I
like
the
idea,
so
we
bundle
and
we
bump
we
bumped
it
to
V
1
for
the
whole
thing,
for
the
whole
bundle
make
sense
to
me.
D
Maybe
maybe
the
first
time
we
bundle,
we
bump
it
to
something
that's
larger
than
both,
but
is
still
0
point
something
just
because
V
1
makes
it
sound,
very
stable,
but
I
think
it's
more
of
a
semantic
thing.
I,
don't
think.
Actually
is
that
means
that
it
needs
to
be
perfect
they're
also
pumping
into
the
one
should
be
fine.
It
just
means
that
we
should
follow
cember
very
strictly
or
more
strictly
than
we
have.
We
have
been
following
it
quite
strictly
already.
C
D
I
mean
that's
actually
why
sambar
is
somewhat
meaningless
to
us,
because
we
we
are
bound
to
much
stricter
stability
requirements
than
somewhere.
Okay,
because
our
users
aren't
upgrading
like
if
even
with
you
know
like
on
Caleb,
calm
users
are
still
auto,
upgrade
it
anytime
that
the
default
template
is
updated
and
that's
like
a
separate
issue
that
or
maybe
that
that's
one
in
one
way
we
can
solve
breaking
changes
is
by
you
know
not
auto
upgrading
the
templates
for
people
who
are
already
using
them
and
saying
like
there's
a
new
template
out.
D
C
B
C
D
C
D
C
Okay,
this
is
for
we
addresses
our
question.
How
do
I
know
integral
is
one
point
of
point
over
if
software
is
being
used
to
in
production.
You
should
probably
already
be
one
point.
No
point
know
if
you're
worrying
a
lot
about
backwards
compatibility.
You
should
probably
already
be
one
point
no
point.
So
that's
fair
enough
to
me.
D
D
Exactly
that
one,
this
one
is
in
the
namespace,
but
so
so
when
I
have
like
I'll
do
I,
when
I'm
rolling
out
just
production
I
go
to
just
stable
track.
How
do
we
make
sure
that
the
change
is
that
I'm
rolling
out
do
not
change
anything
in
the
ingress,
for
example,
or
I
mean
what
I'm
just
doing
canary
sorry,
I'm
just
doing
canary
or
one
of
the
unstable
tracks?
How
do
we
make
sure
that
that
doesn't
break
the
angrist
so.
D
B
A
D
Yes,
so
I'm
feeling
like
it
would
be
better
to
manage
this
resource
independently
of
the
application
deployments,
because
otherwise
we
need
to
ploy
your
application
or
your
canary.
Like
I
know,
we
have
to
have
a
special
condition
that
says,
should
I
be
updating
the
canary
ingress
inside
the
chart,
but
yeah.
So
just
wondering
good,
because
there's
like
separate
boxes
around
stable
truck
canary
track
roller
track.
So
I
was
wondering.
Maybe
that
was
your
intention
with
the
graph
you
have
like
to
kind
of
extract
the
canary
or
the
external
ingress
into
a
separately
managed
resource.
A
Yes,
so
like
at
first
I
was
thinking
that
maybe
we
can
create
a
kind
of
English
before
kind
of
a
track
or
that
resource
where
Connery
were
roll
out
is
created
at
a
when
this.
You
know
these
ones
are
gonna,
be
removed
when
several
track
is
freely
roll.
We
rolled
out
so
I
thought
that
we
can
also
remove
this
kind
of
reading
glass
when,
like
this
verso
States
are
gone
and
then
switcher
like
just
update
this
in
Class
G
point
to
stable
track.
C
A
A
Yeah,
if
we,
we
always
create
kind
of
being
less
rigorous
like
this
cannery,
where
Rolla
track
is
created,
then
we
set
way
to
0%
because
they're
nothing
to
you
with
you.
So
all
the
traffic
goes
to
stable
it.
When
kinda
truck
is
created,
then
we
can
just
update
a
parameter.
The
weight
value
to
like
10%
or
20%.
A
C
Think
I
think
that's
okay,
I
think
we
can
test
this
up
so
harder
as
well.
I
think
what's
gonna
happen
is
that
we're
gonna
do
the
same
thing
as
we're
currently
doing
now,
which
is
that
we
have
one
chart,
but
we
will
have
multiple
releases,
so
there
will
be
a
production
release
and
an
OB
production
can
release
and
then
episod.
D
D
B
D
It's
kind
of
what
I
was
thinking
I'm,
not
sure
that
really
makes
sense.
I
mean
that
will
definitely
increase
the
complexity
of
auto
deploy
image
like
which
is
reaching
a
point
where
I'm
thinking.
Maybe
we
should
be
using
something
like
file
if
we
really
want
to
double
down
on
helm
charts
for
it.
Oh.
B
D
C
D
I
think
help
file,
just
like
kind
of
is
they
already
ironed
out
a
lot
of
like
ways
to
call
helm
correctly
with
multiple
charts
but
yeah,
otherwise,
I'm
not
sure,
like
personally
I'm,
not
super
convinced
that
helm
is
the
best
solution
for
something
like
the
auto
deploy
like
in
general.
Alright,
that's
also
another
long-running
discussion
that
we
should
kind
of
that's
something
that
we
might
want
to
discuss
separately
shaniyah
or
for
like
a
future
handover.
I.
A
D
A
D
There
it
like
the
chart
has
become
a
DSL
that
is
more
complicated
than
the
resources
that
it
is
defining
and
actually
you
know,
if
you
render
the
chart
for
majority
of
users,
it's
gonna
be
like
three
files
or
something
like
it's
something
super
simple
like
resources
that
the
user
could
just
be.
You
know
versioning
themselves
so,
like
the
value
is
there's
some
things
there
like
the
parameterization
is
useful
to
us,
because
we
have
this
Gatling
integration
that
has
to
work,
and
it's
a
common
operation
that
everyone
needs
to
do.
D
But
then
you
know
there's
other
customizations
there
like
the
ports
and
the
liveness
probes
and
stuff
that
Caleb
really
doesn't
care
about,
but
the
user
needs
to
be
able
to
customize,
and
it
would
be
much
easier
if
they
could,
just
you
know,
apply
the
kubernetes
customization
without
having
to
consult
our
DSL.
But
this
is
like
a
generic
issue
with
all
helm
charts
and
it's
not
really
us
like
it's
a
you'll,
see
many
chart.
Authors
complaining
about
this
Oh,
so
there's.
B
D
There's
a
I've
spoken
a
bit
with
me
Kyle
about
this
as
well,
and
he
pointed
me
towards
some
solutions
that
I
haven't
had
time
to
actually
look
at
that
could
but,
like
one
of
us
called
kpt
I
think
it's
a
play
from
Google,
that's
kind
of
like
a
helm
rival,
but
it
uses
a
thing
called
customize,
which
is
a
Cuban
Eddie's
native
lightweight,
templating
solution.
It
doesn't
let
you
it
doesn't.
Let
you
do
everything
and
it's
it's
a
bit
different.
D
B
D
C
Think
the
focus
is
customization,
because
now,
at
the
end
of
the
day,
we
still
need
something
to
deploy.
So
home
works
as
well.
The
forum
is
the
customization,
so
there
are
tools
out
there,
for
example,
kpt,
which
allows
customization
in
a
much
simpler
form,
and
it
doesn't
require
us
to
have
a
fair
statements,
so
people
can
just
say
include
a
folder
and
that
customizes
the
resource
that's
already
in
the
default.
So
that's
that's
a
whole
separate
issue.
I
think
think.
D
C
Okay,
I
want
to
come
back
to
the
version
number
because
and
we're
talking
about
Visio
architecture,
and
if
we,
if
we
decide
to
bump
it
immediately
to
1,
then
the
Visio
architecture
is
now
the
v-0
and
v1
architecture
and
then
and
then
now
we
are
talking
about
a
v2
architecture.
Now
right
so
I'm
going
to
be
clear
on
that
I.
C
A
A
C
B
C
That's
one
change.
We
need
I,
think
it's
up
there
in
one
get
the
first
paragraph,
but
even
though
it's
a
major
version
bump
there's
no
breaking
change,
nothing
happens.
So
that's
one.
That's
the
first
point
and
then
at
some
point
from
now
to
14
point:
oh,
we
can
introduce
v2,
as
that
includes
the
architecture,
changes
you're
breaking
change,
but
it
will
not
break
people's
pipelines
because
it
will
not
be
the
default.
Yeah
will
be
often
so
we
can
introduce
v2
architecture
at
any
time
and
v2
major
version
or
the
impact.
So
again.
A
C
A
So
yeah,
like
ambitious
users,
can
use
latest
image
version
right
and
when
we
introduce
like
major
versions,
we
don't
say
that
it's
stable
yet
right
until
it
becomes
a
default
in
ot,
deploy
temperate
quick
sup.
What
we
introduce
like
v1
were
like
later
beat
you
yes,
but
we
don't
say
that
it's
a
stable,
it's
more
like
age
right!
Yes,
right.
B
A
C
D
D
How
do
we,
how
do
we
do
this
I
feel
like
like
with
the
chart
not
vendored?
We
could
start,
we
could,
you
know,
make
release
to
it
that
are
well.
We
can't
actually
make
breaking
changes
to
it
because
of
the
you
know,
because
people
are
using
latest
we'd
have
to
rename
the
chart
and
publish
it
under
a
different
name,
but
if
we
under
it
we
bypass
that,
but
then
we
have
to
update
the
image
for
people
to
be
able
to
use
the
new
chart.
So
how
do
we?
D
D
A
C
A
B
A
Thinking
about
like,
can
we
check
their
chart
version
before
proceeding
the
deployment
yes
saying
if
the
chart
parser
is
different?
Yes,
like
ot
blue
bar
chart
is
like
scenic
Sigma
significantly
different.
Then
we
sure
a
large
users
like
fail
just
failed
the
pipeline
and
then
which
Persian
charts.
Do
you
want
to
keep
using?
If
you
don't
wanna,
go
through
the
manual
application
I'm
on
your
upgrade,
then
just
specify
the
environment
variables
T
example.
Then
this
becomes
our
variable.
B
C
D
Guess,
yes,
that's
that's
kind
of
like
what
it
would
be,
so
so
it's
really
a
lot
like
the
Postgres
one,
except
that
here
we
are
doing
it
with
the
image
version
itself.
Instead
of
as
like
a
variable
inside
the
script
and
and
then
we
fail
it,
we
check
somewhere
early
inside
the
image.
What
the
previous
layer
like
deployment
was-
and
you
say
now
we're
gonna
break
and
my
only
question
there
is:
is
it
acceptable
on
point
release
of
kid
lab
to
break
people's
deployments
to
show
them
that
message?
D
I
mean
it
will
only
happen
once
because,
as
soon
as
somebody
does
see
that
message
and
decides
to
use
the
old
version,
they
will
be
stuck
on
that
version
forever.
When
we
break
it.
Sorry
well,
like
you
know
how
we
did
the
Postgres
version
for
30.
No,
we
there
was
the
first
time
somebody
rolls
out
on
30
know.
If
they
were
using
the
in
cluster
Postgres
pipeline
will
break.
It
will
fail,
so
you
say
you're
doing
like
a
critical
security
fix
or
we're
scaling.
D
Your
rollout
or
something
I
plan
failed
now
with
buskers
we're
pretty
safe,
because
I
don't
think
anyone
was
really
using
it
for
production,
I'm,
hoping
not
many
people
are
using
it
for
production
and
also
I
was
13.
Oh,
it
was
a
braking
release
and
like,
though
that
was
kind
of
expected,
but
is
the
same
okay
to
do
in
minor
release.
D
B
C
D
D
A
D
Had
the
only
like
complaint
I
saw
from
users.
Well,
we
had
actually
complaints
from
our
own
s.
Eries,
who
ran
into
this
and
also
at
the
same
time,
ran
into
other
deployment
problems,
and
it
was
hard
for
us
to
tell
what
was
causing
the
failures
and
we
had
to
actually
fix
like
several
things.
Until
we
actually
got
things
working
again
and
then
there
was
somebody
who
was
on
a
custom
chart,
and
that
was
a
use
case.
We
forgot
about
4:30
know
where
they
just
had
deleted.
D
Postgres
from
their
custom
chart
thing
in
cluster
1,
so
they
didn't
actually
have
one,
but
we
said
we
detected
one
and
then
we
failed
their
pipeline
when
we
shouldn't
have,
but
that's
like
a
really
hard
situation
for
us
to
deal
with
because
we're
not
like
we
can't
yeah
the
person
charts
are
always
gonna,
be
that
kind
of
a
second-class
citizen.
You
know
that
sense,
but
that
was
the
main
consideration.