►
From YouTube: Kubernetes SIG Apps 20180730
Description
No description was provided for this meeting.
If this is YOUR meeting, an easy way to fix this is to add a description to your video, wherever mtngs.io found it (probably YouTube).
A
A
B
This
slide
is
so
with
this
regular
CD.
It's
a
get
up,
skew,
bring
a
nice
deployment
tool
which
Jesus
get
repositories
as
the
source
of
truth.
It's
declared.
If
and
unless
you
choose
from
a
variety
of
configuration
management
tools,
see
the
most
basic.
Is
you
have
a
directory
of
just
plain
Yama's?
We
support
case
on
it
helm
and
we're
evaluating
gauging
interest
in
support
for
customized.
It's
implemented
as
a
kubernetes
controller
and
Ciardi,
and
it
has
enterprise
grade
features
like
single
sign-on
are
back
audit
and
compliance
and
security.
B
B
Okay,
so
this
is
our
go
CD.
The
first
thing
you
might
notice
is
that
we
have
single
sign-on
integration.
It's
implemented
using
decks
from
core
OS,
which
allows
us
to
integrate
with
all
major
forms
of
single
sign-on
like
Oh
F,
OID,
C,
sam'l,
2.0
LDAP
I
happen
to
configure
this
instance
so
that
any
members
of
the
Argo
project
had
team
can
log
in.
B
So
this
is
the
repository
that
we
maintain
for
our
tutorials
and
and
demos
just
to
illage
illustrate
some
of
the
features
of
our
go
CD
and
it's
intended
that
this
repos
intended
to
be
fort.
So
you
can
connect
it
to
your
AR,
go
see
the
instance
and
play
around
with
you
know,
commits
and
pushes
and
stuff.
B
I'll
connect
the
repository,
it's
public,
so
it
has
no
auth
all
right.
Once
it's
connected
I'll
select
an
app
from
that
repository,
it
just
scanned
for
all
the
helm,
charts
and
caisson
apps.
Inside
this
repository.
You
can
also
specify
just
a
directory
of
VMs,
also
the
case
when
I
add
most.
We
detected
this
case
on
an
app
that
has
three
environments.
All
of
three
are
pointing
to
the
internal
kubernetes
hostname,
which
we
use
for
the
most
part,
we'll
example,
four
demos,
but
you
can
actually
deploy
to
external
kubernetes
clusters.
B
B
B
Okay,
so,
as
you
can
see,
we,
you
can
see
the
live
state
changes
as
they're
happening,
we're
actually
streaming
the
this
resource
state
changes
from
the
koomer
API
server,
and
so
with
that
we
actually
are
able
to
show
things
as
they
change.
You'll
also
know
this
this
nice
tree
view
and
as
we
get
the
resources
associated
with
the
application,
we're
able
to
reconstruct
the
parent-child
relationship
between
using
the
ownership
references
of
objects.
B
Okay,
so
because
sync
happened,
we
see
that
the
new
replica
set
came
out,
but
right
lips
right
away.
Something
went
wrong,
so
you
know
this
health
status
is
not
green
anymore.
It's
it's
a
blue
sage,
so
we
see
that
this
container
just
had
a
it
has
some
having
some
problem.
So
if
you
click
on
the
container,
you
can
see
what's
going
on,
and
this
is
saying
it's
having
a
image
pulled
back
off.
You
can
see
the
events
with
the
resource,
and
here
we
see
that
the
container
is
having
a
trouble
pulling.
B
The
image
I
happen
to
know
that
this
image
doesn't
really
exist.
So
let's
go
ahead
and
roll
back.
We
do
keep
the
history
of
the
last
five
sync
successful
stinks
that
happen
and
we
are
gonna
roll
back
to
the
previous
one,
but
it
really
a
rollback
is
really
a
convenience
wrapper
around
a
sink,
so
you
can
actually
sink
to
any
state
and
check
them
into
your
git
repository.
B
Okay,
so
that's
pretty
much.
What
I
wanted
to
cover
I
didn't
want
to
mention.
We
do
have
a
CLI,
G,
RPC
and
rest
api.
Some
of
the
additional
features
I
didn't
cover
it
today
or
Argus.
City
has
a
concept
of
pre
post
and
sync
hooks,
and
these
are
the
same
concept:
exact
consulate
helm
hooks
where
you
can
rent
jobs,
and
we
have
support
for
hardware
workflows
to
execute
at
a
certain
stage
of
a
rollout
okay.
I'd
be
happy
to
take
any
questions.
B
Oh,
the
question
about
manual
sink
manual
sink
is
currently
required,
but
we
have
upcoming
feature
called
Auto
sink,
which
is
basically
what
you
think
it
should
be,
which
is,
if
you
I,
make
a
commit
to
the
git
repository
and
then
the
Argo
city
controller
know
this.
Is
that
change?
It
would
go
ahead
and
sink
that
automatically.
So
there's
no
need
to
invoke
the
CLI
or
UI
to
to
perform
that
sink.
That
you
just
saw.
B
B
B
Yeah,
so
it's
not
so
much
I,
wouldn't
call
it
a
pipeline
on
a
case
on
that.
App
is
just
a
opinionated
directory
structure
to
for
you
to
organize
your
your
manifests
in
in
environments
so
like
that
guestbook
this
guestbook
out
this
case
on
it
has
like
you
put
all
you
basically
at
your
manifest
system
here.
This
is
your.
B
A
D
A
Okay,
if
there's
no
more
questions
or
comments,
and
we
can
move
on
to
the
next
section
Jessie.
Thank
you
very
much
for
your
demo
was
awesome.
Thank
you
and
if
there's
any
other
follow-up,
you
guys
can
always
use
this
to
gas.
Flap
channel,
shake
ass,
slack
channel
or
the
Google
Groups.
A
A
E
The
topic
I've
been
thinking
about
a
little
bit
lately,
which
also
ties
into
the
our
goal.
I
guess
more.
The
CI
project
than
the
CD
project
is
how
people
are
kind
of
testing
their
CI
CD
pipelines
locally
on
their
machines.
E
F
B
I
can
help
I
can
answer
their
question
yeah.
So
so
yeah
there's
a
lot
of
confusion
about
like
the
agro
kind
of
family
of
products
and
so
for
now,
just
think
of
there's,
just
Argyll
workflows
and
there's
Argos,
CD
and
I.
B
Won't
I
won't
even
talk
about
our
global
events,
so
in
terms
of
relationship
between
Argo
and
Argos
Argo
we're
clean,
Argo
CD.
The
current
relationship
that
have
is
the
name
Argo,
there's
one
understanding
of
CD
of
workflows
inside
Argo
CD,
which
is
that
similar
to
how
a
home
understands
async
hook
for
like
a
job,
a
special
understanding
like
okay,
a
job
takes
a
while
and
it
has
a
completion
and
we've
also
we've
done
that
in
Argos
CD.
But
we
also
then
extended
out
to
Argo
were
close
and
saying.
B
So
you
are,
you
are
asking
about
testing
workflows
before
they
are
like
in
a
like,
checked
into
a
get
pipeline.
If
Margo
workflows
can
be
submitted
ad
hoc,
so
if
you
have
a
thought,
yeah
I'm
out,
like
an
Argo
workflow
mo
you
can
submit
that,
like
you
know,
keeps
you
feel
apply
that
I
know
we're
in
the
work
just
like
a
kubernetes
job.
So
exactly
like
a
kubernetes
job.
E
C
B
Did
I
find
it
somewhat
unreliable,
so
yeah
I
find
I
have
to
actually
delete
and
yeah
do
we
can
recreate
mini
cube
quite
a
bit
so
not
particularly.
B
C
G
With
with
your
local
kind
of
talk,
for
instance,
so
mini
cube,
you
know
if
you,
if
you
want
to
test
an
energy
kind
of,
have
to
low,
overload
the
mini
context
and
then
build
your
image
or
push
up
that
image
to
a
registry
or
something.
Whereas
with
this,
you
can
just
build
an
image
and
then
use
it
directly.
So.
A
I
actually
have
a
question
I
generally
test
qu
renée's
itself
on
live
clusters
using
to
be
test
or
like
cluster
up
you
you
use
local
lustre.
Is
that
actually
leveraging
Mini
Cooper?
Is
that
just
spawning
off
the
API
server
controller
manager
sed
and
goes
what
actually
goes
up
onto
using
the
bash
script
underneath,
like
I've,
never
looked
at
batch
script
or
a
little
cluster
to
determine
what
it
actually
does?
Does
that
leverage
me
cubital.
E
A
A
A
How
many
people
are
actually
using
see
ICD
on
top
of
kubernetes
in
their
production,
environment
or
even
a
development
environment
right
now,
is
it
something
that
I
mean?
Obviously
the
guys
looking
to
it
are
doing
it?
There
are
many
people
that
I'm
aware
of
doing
it,
but
is
not
doing
GIC
do
kubernetes
still
a
thing
I.
A
A
Okay,
there's
a
few
open-source
solutions
out
there
now
that
are
trying
to
reduce
the
number
of
permissions
better
necessary
in
order
to
build
containers
inside
of
your
kubernetes
cluster.
Some
of
you
don't
actually
need
to
run
the
dr.
demon
in
order
to
do
that,
but
I'm
not
sure
which
ones
are
dominant
at
this
point
or
actually
I'll
leave
it
at
the
maturity
level.
It's
not
something
I
do
on
a
regular
basis.
There's
no
they're.
H
Not
mature
and
they're,
but
they're
all
worth
following,
so
a
good
example
would
be
just
for
those
IMG
project.
It
works
if
you
patch
this
this
this
this
this
miss.
So
my
god,
that's
that's
too
hard.
Oh
really,
we'll
be
end
up
doing
is,
depending
on
what
cloud
provider
we're
running
one,
so
we're
running
a
TCP.
We
just
I'll
just
use
on
the
build
registry
or
the
container
builder
service
that
day,
which.
A
H
It's
just
it's
about
managing
complexity.
I
mean
my
whole
company,
my
engineering
partners
like
4050
people
with
the
most
you
know
you
could
have
that
point.
So
it's
just
about
choosing
what
we're
going
to
do.
First,
ultimately,
I
think
we'll
have
something
that's
better
and
we
definitely
want
to
build
and
cluster
it's
not
there
yet,
and
someone
posted
that
build
is
getting
there
and
it
is.
But
the
problem
is:
is
that
if
there's
caveats,
it's
not
my
knowledge
or
another
senior
engineers
knowledge,
it's
the
cumulative
knowledge
of
the
entire
group.
So
wherever
you
have.
I
One
thing:
that's
that's
hard
about
CI
CD
is
that
everybody's
process
is
gonna
smoke,
so
you
end
up
with
lowest
common
denominator
being
the
share
tool,
and
then
those
don't
have
a
great
experience
of
what
you
see
is
then
teams
or
companies
or
projects
taking
on
one
particular
slice
of
opinionative,
the
ICD
and
making
that
a
happy
path,
so
that
doesn't
end
up
working
for
for
everyone.
So
I
think
that's
where
you
see
the
breadth
of
CI
CV
tools
that
we're
seeing
today.
I
H
That's
a
good
point,
because
whenever
we
we
can
read
the
starting
with
Jess
Humboldt
book
on
continuous
delivery,
watching
what
Jenkins
has
done
watching
what
other
like
circle,
CI
other
players
have
done
seeing
what's
come
out
of
spinnaker
and
then
you
look
at
your
kubernetes
cluster
know
like
well.
I
can
do
all
those
things,
but
do
they
match?
Are
they
bright
for
this
environment
and
we're
still
figuring
out
how
that
works?.
A
A
We
didn't
build
some
type
of
community
of
interest
around
the
particular
subject
and
like
in
the
same
of
an
application
resource,
try
to
set
up
a
set
of
CR,
DS
and
just
externalize
individual
components
so
that
there's
a
common
interface
that
people
who
want
to
build
pools
in
a
space
don't
have
to
repetitively.
You
know
produce
each
time.
I.
A
Think
a
lot
of
you
guys
have
already
said,
though
kind
of
brings
up
the
challenges
and
doing
that
and
that
there's
so
many
there's
so
many
different
bespoke
build
processes
getting
that
API
right
and
even
getting
the
functional
decomposition
of
the
CIT
CD
system.
That
needs
like,
let's
say
80
to
90
percent
of
the
target
use
cases
of
most
of
our
users
might
be
very
difficult.
H
H
Things
that
we
could
think
about
and
just
to
go
back
to
what
you
say
about
CR
DS
can
every
good
example
of
doing
CD
or
TI
on
on
kubernetes
will
ultimately
probably
go
to
a
CRD
I.
Just
think
that
just
makes
sense.
It's
declarative!
It's
it's!
How
we
talk
to
our
clusters,
so
that's
gonna
happen.
My
belief.
A
F
A
Ci
CD,
and
even
the
way
people
view
it
came
earlier,
was
asking
about
pipelines,
which
is
generally
how
I
think
about
the
CI
CD
space
and
jet
plane.
You
know
it's
a
billable
pipeline
but
come
to
find
out.
Not
everybody
actually
thinks
of
it.
In
those
terms,
so
even
common
denominator
might
be
somewhat.
H
To
step
up,
this
will
be
a
great
topic
for
con
and
just
I
think
it
would
be
a
great
topic
to
to
show
how
we
are,
instead
of
focusing,
on
the
back
end,
we're
focusing
on
the
the
user
level
with
actual
CRT
itself
and
then
how
we
build
software
that
works
in
the
cluster
and
a
kubernetes
native
way,
not
just
bolting
on
something
to
happen
for-
and
you
know,
I'm
only
saying
this
because
you
know
I've
had
conversations
in
the
past,
but
I
just
think.
It's
a
great
topic.
J
Hi
there
I'm
Ben
from
the
scale
factory
we're
consultants
in
London,
and
we
were
miss
of
having
a
lot
of
similar
thoughts
to
what
you've
just
been
talking
about.
Now,
we're
looking
to
get
into
into
this
area
rather
than
building
yet
another
opinionated
sea
icd
pipeline,
that's
their
own
version
of
helping
run,
we're
starting
off
a
bit
of
user
research
in
them
talking
to
people
who
are
already
customers
of
ours
or
the
contacts
to
try
and
build
up
an
idea
for
from
a
developer's
perspective.
What
is
it
that
people
actually
want
from
CI
CD
pipelines?
J
A
J
So
yeah
we're
just
sort
getting
going
on
that.
We've
got
about
a
dozen
names
on
our
list
of
people
that
we're
talking
to
so
we
could
present
back
on
that
in
about
the
next,
don't
know
about
the
next
meeting,
but
feels
the
13th
one.
Maybe
oh
I
suppose
we
could
make
it
a
bigger
enterprise
and
how
other
people
see
results
all
back
together.
A
A
One
of
the
interesting
things
about
it,
though,
when
the
problems
of
solves
is
or
they
try
to
solve
with,
it
is
CIT
CD,
so
I
want
to
chase
them
down
to
have
them,
give
a
presentation,
at
least
someone
from
their
team
and
give
a
presentation
inside
of
say,
gaps
as
well,
because
we
download
a
lot
of
bad
software
here
and
be
good.
Is
you
know?
Native
gave
a
demo
that
too?
Maybe
we
can
also
have
them,
give
their
thoughts
on
yes,
Unni.
H
H
And
all
the
all
the
words
around
you
know
ingress
not
being
suitable.
It
doesn't
have
that
the
abstraction
layer
is
not
good
enough
and
then,
when
I
was
at
next
last
week,
I
saw
talk
about
extensions
that
the
GCP
team
did
to
ingress,
so
they
could
actually
not
store
data
in
an
annotation
I
think
there
is.
There
should
be
some
words
about
that,
like
that
pattern
is
pretty
neat
and
I
know
Joe
bata,
actually,
if
anyone's
interested
thrown
up
like
wrote
up
something
about
this
back
in
March.
H
A
H
A
I
mean
that's,
definitely
something
we
can
do
inside
of
the
gaps
and
like
either
come
up
with
a
set
of
recommendations.
I
can
get
some
of
the
guys
who
are
like
from
cookie
builder,
or
we
can
reach
out
to
some
of
the
guys
at
Korres
who
are
working
on
operator
kit
who
are
kind
of
to
the
heaviest
users
of
CR
DS.
There's
some
other
teams
inside
of
Google,
but
I
can
reach
out
to
and
talk
about
how
we're
using
them.
So
we
can
definitely
have
that
conversation
I.
F
Think
one
of
the
things
that
I
would
like
to
see
more
around
series
is
just
making
people
aware
that
CR
these
are
for
you
and
you
don't
have
to
be
a
hardcore
current
developer.
You
don't
have
to
be
developing
something
like
an
sto
or
something
like
that.
You
can
just
say:
hey.
We
were
in
this
work
load
all
the
time
and
we'd
like
to
be
able
to
specify
it
in
a
compact
way
in
our
cluster.
F
A
Mean
this
is
really
lengthy,
so
there's
also
a
difference
between
the
CRVs
and
like
the
combination
of
a
custom
controller
and
the
CRTC
really
this
year.
He
is
basically
just
a
way
to
teach
the
api
server
about
speaking
about
structured
data.
It's
just
generally
that's
not
necessarily
very
useful,
except
in
the
context
of
using
it
with
a
custom
controller
that
actually,
you
know,
tries
to
realize
some
state
based
on
modifications.
Drds,
oh
okay,
I
mean
lately.
H
And
whenever
imports
them
in,
it
actually
creates
a
CRD
of
the
webhook
that
expected
and
I
don't
know
if
it's
a
great
idea,
but
it's
nice
being
able
to
consume
that
instead
of
everyone
having
to
know
how
to
talk
to
everyone
in
a
namespace
or
group
of
namespaces
having
to
know
how
to
talk
to
github
to
get
events
out.
So
it's
a
good
representation
of
state,
and
it
may
be
to
be
careful
that
we
don't
get
many
thousands.
But
it
is
a
nice
pattern,
at
least
from
my
experimentation.
So
far,.
A
That's
another
kind
of
pattern
that
is
kind
of
tough
to
do
it
really.
When
do
you
so
you
created
this
year,
do
you
you're
going
to
register
your
CRD
with
the
existing
API
server
and
not
using
an
extension
API
server
at
what
point?
How
many
objects
can
you
start
creating
before
you
start
overwhelming
the
control
plane
based
on
different
usage
patterns
and
indefinitely?
What
do
you
do
you
go
to
an
API
extension
server
and
back
it
with
a
different
storage
medium
in
order
to
ensure
isolation
from
the
control
plane?
Sets
you
do.
A
H
A
Don't
well
I
mean
in
all
honestly
I,
don't
know
as
a
community,
we've
done
a
great
job
about
having
the
scalability
limits
or
various
operations
under
various
different
workloads,
so
they
mean,
like,
obviously
you
say,
thousands
well,
thousands
when
they're
doing
what
you
can
store.
Thousands,
never
touch
them,
and
obviously
that's
not
going
to
cause
any
problems.
What,
if
you're,
really
what,
if
you're
right
and
what,
if
you're,
updating
the
leading,
frequently
so
forth,
so
understanding
that
better.
If
any.