►
Description
A
You
you,
we
go
just
try
different
button,
see
if
it
works
good
morning.
Thank
you
for
coming
today
to
the
I
think
this
must
be.
The
6th
agro
were
close
community
meeting
of
2020,
and
so
this
has
changed
a
bit
over
time
and
we've
started
to
do
more
stuff
around
agro
vents
and
what
we're
talking
a
bit
more
about
that
today
and
this
community
meeting
is
really
a
great
opportunity
for
people
to
find
out
what's
coming
up
in
our
NGO
workflows,
what
we're
working
on
today.
A
We
do
this
monthly
on
the
third
Wednesday
of
each
month
and
for
those
of
you
who
don't
know
what
our
go
workflows
are.
Our
events
are
they're
part
of
a
cloud
native
ecosystem
for
scheduling
and
executing
workflows.
So
by
culminating,
obviously,
we
mean
kubernetes
myself,
I'm
Alex
I'm,
a
principal
engineer
on
Argo
workflows
and
previously
I
worked
on
Argo
CD
as
well,
so
I
know
the
Argo
software
quite
well
from
a
technical
standpoint.
A
That's
that's
fine,
of
course,
and
now
a
little
bit
about
the
agenda
today
and
Derek
has
been
doing
a
lot
of
kind
of
work
on
Argo
events
to
make
it
easier
to
use
more
of
us
more
reliable
and
more
secure
and
he's
going
to
give
us
a
demo
today
of
a
new
and
custom
resource
called
event.
Bus
and
you'll
be
talking
a
bit
about
that
I'll
be
showing
a
few
slides
around
some
positive
changes
for
Argo
workflows
3.0.
A
A
Thank
you,
and
one
on
one
last
thing
here
is:
please
add
your
name
we'd
love
to
know
which
code
which
companies
are
doing
and
working
with
our
go
here,
and
so
just
add
your
name
as
an
attendee
here,
and
that
will
make
my
colleague
happy
as
well,
because
she
likes
to
know
who
comes
to
meetings.
Okay,
Derek.
Are
you
ready
to
take
over
yes,
six
correct
I
will
hand
over
to
you
now
all.
C
C
You
can
see
my
square
right,
okay,
so
let's
take
a
look
at
the
buzz
spec.
This
is
simplest,
even
bus
back,
you
give
a
name.
You
give
a
right
now,
a
stubborn
ass
and
we
sober
to
a
native
nest,
which
means
we're
gonna,
install
and
that
service.
For
you
or
there's
another
field
named
exotic
here,
which
means
you
can
provide
own,
that's
configuration
so
the
simplest
and
even
bus
configuration
and
just
keep
an
Arab,
and
you
can't
get
the
empty
here
and
then
we're
creating
that
service.
D
C
So
after
I
created
this
even
post,
it
will
spin
up
and
Nath's
screaming
wrist
wrist
force
net.
Miss
a
minimal
replica,
says
three,
alright,
so
right
now
the
three
powers
are
running
and
then
I
have
a
let's
look
at
the
so
first,
let's
look
at
an
accepted,
even
bus,
and
then
it
get
the
config
in
the
status
and
get
its
NASA,
and
then
it
has
access
secret.
C
This
is
test
and
ask
test
box,
and
then
I'm
gonna
use
to
come
out
on
here
to
publish
some
message
to
Chism.
That's
suburb
in
vain
I
used
as
subscriber
time
here
to
subscribe
messages.
Someone
just
want
to
make
sure
that's
service
is
working.
Okay,
so
I'm
gonna
use
the
stand.
Pup
a
stay
actually
is
the
reverse
side
of
an
S,
which
means
sister
not
screaming
non
that
service
and
that's
that's,
has
to
come
comes
necessary.
Nostril
working
is
not
screaming,
which
brings
the
assistant
in
history.
C
History
message
features
so
I
use
a
stamp
up
and
I
kept
a
nice
ask
me:
is
this
necessary?
You
are
and
then
here
because
we
are,
we
are
using
strategies
so
I'm
going
to
get
the
token
in
the
secret.
C
C
C
And
then
I
got
all
the
three
message
and
published:
this
is
just
to
make
sure
the
nest
service
is
working
and
then,
in
the
future,
with
so
right
now,
I'm
working
on
two
to
make
change
to
the
sensor
in
merging
and
give
away
a
brand
stores
and
in
the
future,
in
the
you're,
not
even
source
configuration
event
configuration
you
don't
need
to
give
the
subscriber
as
a
in
the
subscription
party
or
even
you
know,
get
away
and
the
censored
part
you
you
don't
need
to
give
anything
just
a
party
program.
C
C
It
can
look
at
more
aspects
about
the
event
plus,
so
if
you
want
to
give
more
replicas
just
keep
your
number
here
and
you
want
to
make
that
and
help
in
any
body
for
this
force.
If
you
get
any
penalty
true
and
then
you
know
just
follow
the
rules
and
infinity
to
schedule
car
and
you
also
keep
a
persistent
you
keep
your
starch
constant
access
mode
or
one
inside
by
the
forest
in
volume
size
to
5.
C
Gig
by
this
is
for
native
NASA
installation,
and
if
you
have
a
existing
NASA
streaming
server
installs
somewhere,
it
also
can
keep
it.
Thank
you
Miss
configuration
there
and
then
will
connect
to
the
existing
Ness
service
and
we
also
need
with
your
neural
network.
We
will
also
support
Kafka
in
the
future,
but
we're
not
gonna
create
an
item,
the
copic
installation,
but
we're
gonna
just
support
you
keep
in
the
cop
car
configuration
and
we
we
use
it.
The
cop
car
and.
C
Here
in
the
in
the
in
the
dock
you
can
get,
you
can
see
the
full
specs
and
the
same
aspects
here
and
there
now
right
now
we
have
any
uni
where
the
even
buzz
and
sensor
and
even
source,
but
we
want
to
give
them
demo
in
at
once.
You
just
want
to
get
your
feedback
earlier
to
see.
If
you
have
any
comments
on
the
on
the
specs
and
then,
if
you
have
any
question
you
can
pay
me
on
snack
or,
oh
sorry,
do
you
have
any
question
right
now.
C
A
C
C
So,
for
example,
this
this
calendar
gateway.
You
need
to
give
a
subscriber
as
a
a
chat,
endpoint
or
given
as
server
here
and
with
the
new
event
post.
You
will
not
need
to
give
this
part,
so
all
the
message
will
are.
The
events
were
sent
to
the
remembers
by
the
forum
and
sensors
read
the
message
from
the
even
bus.
A
B
A
Workloads,
if
you
want
to
add
that
the
community
meeting
document
that
would
be
absolutely
fantastic,
okay,
I'm
going
to
share
some
slides
with
you
just
these
kind
of
reflects
some
of
the
discussions
that
we've
been
having
at
in
the
core
team
around
the
kind
of
future
of
Argo
workflows
and
what,
where
we
might
go
next
with
it.
You're,
probably
aware
it's
changed
pretty
quite
a
lot
over
the
last
six
months,
or
so
with
the
additions
of
features
such
as
workflow
templates,
cron
flows
and,
obviously
the
persistence
feature
for
uploading,
those
statuses
and
workflow
archives.
A
A
A
We
always
try
and
avoid
them
if
we
can
do,
but
sometimes
we
find
out
certain
situations
where
we
are
maintaining
a
lot
of
existing
code,
that
we
don't
really
need,
don't
reuse
and
we
want
to
get
rid
of
it,
and
so
that
would
be
the
removal
of
some
of
these
deprecated
fields
in
the
configuration
of
specification
that
date
back
some
time,
such
as
configuration
around
the
executor
and
around
metrics,
and
around
templates
template
arguments
and
so
forth.
So
these
are
the
some
of
the
fields
we
will
look
at
removing
as
part
of
that
transition.
A
A
Trying
to
go
through
is
much
quicker
than
I
expected.
I.
Think
here
we
go.
Another
thing
we
want
to
do
with
our
workflows
is:
have
a
tighter
integration
with
Argo
events,
so
people
often
use
arc
events
with
Argo
workflows
and
they
want
to
be
able
to
answer
questions
such
as
you
know.
Why
was
my
workflow
triggered?
A
So
the
goal,
one
of
the
goals
we
would
like
to
have-
is
making
a
bit
clearer
why
our
workflow
was
executed,
and
this
also
ties
in
with
the
fact
that
Argo
vents
doesn't
have
it
any
kind
of
user
interface
or
in
fact,
CLI
everything's
done
with
coop
CTL,
and
this
will
give
us
opportunity
to
build
out
some
user
interface
features
that
would
support
Argo
events
better
and
also
in
a
cohesive
fashion
in
the
same
user
interfaces
Argo.
What
flows.
A
The
the
probably
the
biggest
potential
item
on
this
is
multi
cluster
and
multi.
Cluster
invariably
means
multi-tenancy
at
the
same
time.
So,
rather
than
having
one
user
interface
deploy
into
each
cluster,
we
would
have
a
single
user
interface
that
allows
you
to
act
as
a
overview
and
potentially
control
plane
for
all
your
different
clusters,
and
that
really
necessitates
at
that
point
kind
of
multi-tenancy
at
that
point.
So
the
current
role
we
use
for
our
go
workflows
is
that
we
essentially
have
a
service
account.
A
But
if
you've
got
multiple
clusters,
you
probably
have
multiple
teams
and
if
you're
an
enterprise
environment,
then
you
almost
certainly
want
to
separate
the
data
in
such
a
way
that
one
team
can't
access
another
team's
data.
So
you
need
some
kind
of
role
based
access
controls.
To
do
that,
we
have
a
relatively
simple
model
today
in
the
user
interface
either
you
can
use
a
north
mode
called
server,
which
essentially
means
that
I'll
go
it's
a
anonymous,
unauthenticated
logon
and
you
just
get
whatever
permissions.
A
So
we
introduced
MV
2.9,
a
single
sign-on,
and
that
allows
you
to
connect
to
your
single
sign-on
provider,
which
is
would
be
an
oid
co-author
to
provider
and
login,
and
we're
going
to
provide
some
additional
arm.
Our
back
bakes
controls
we're
interested
to
see
what
people
will
want
there.
One
thing
I've
learned
to
about
here
is
potentially
new
code
base,
so
you
can
run
the
agro
user
interface
now
we're
in
PI
running
at
the
Argos
server
commands
and
it's
built
into
that
code
base.
A
But
potentially
we
might
want
to
move
that
to
a
new
code
base.
So
the
ink
import
shared
code
from
both
the
our
gove
ends
and
our
go
workflows
code
base
and
operate
in
different
ways.
So
that
would
be
a
breaking
change
that,
therefore
you
would
not
have
a
user
interface
that
will
come
with
our
go
workflows
by
default.
That
will
get
a
separate
piece
of
software.
A
A
So
Daniel
Herrmann
is
asking
with
a
move
23.0.
Should
we
take
the
opportunity
to
make
the
reap
coke
compatible
with
gold
modules,
since
you
can't
actually
use
go
get
that's
very
interesting
question.
The
answer
to
that
is
yes.
Yes,
we
would
do
that.
There
is
a
reason
that
we've
not
done
that
effectively
so
far.
It's
because
the
go
modules
system
is
quite
quite
opinionated
about
what
v2
and
v3
means.
If
you
want
on
B,
0
or
V
1,
then
you
just
have
a
git
repository.
A
You
go
get
that
git
repository
if
you're
on
v2
or
any
other
v3
v4,
then
what
you
have
two
options
there:
either
you
create
a
subdirectory
called
v2
and
or
you
create
a
branch
called
v2
and,
as
you
can
mention,
if
you're
using
get
all
your
tooling,
is
based
around
repository
parts
and
versions
and
also
all
your
automations
arranged
around
that,
so
it's
actually
extremely
it's
basically
impractical
to
to
migrate
a
module
to
v2
I,
don't
I!
Don't
in
my
humble
thing
is
not
a
very
good
system.
A
A
So
another
question
so
Jaya
balan
has
asked
you
just
mentioned:
SSO
is
implemented
in
the
recent
version.
There's
agro
support
security
matrix
like
in
Jenkins
the
the
short
answer
to
that
is
no
it's
so
so
people
obviously
talk
about
SSL
mean
you
wish,
and
you
assume
that
it's
gonna
come
with
some
access
controls,
because
SS
are
almost
always
comes
with
some
kind
of
access
controls.
A
The
SSO
implementation
doesn't
come
in
any
access
controls
because
the
implementation
is
actually
separate
from
that
and
we're
still
trying
to
understand
what
the
right
way
to
implement
those
access
controls
is
our
go.
Cd
has
a
sophisticated
model
based
around
projects,
so
it
was
built
from
the
ground
up
to
support
multi-tenancy
and
multi
cluster,
and
we
have
a
number
of
learnings
with
that
and
it
uses
a
piece
of
a
ghost
awkward
called
Kasdan
which
allows
you
to
create
those
controls.
A
A
And
so
is
another
question
here
from
daniel
herman
again,
this
is
not
specifically
a
3.0
request,
but
have
we
considered
collaborating
with
the
Tecton
team,
while
Tektronix
focused
specifically
on
the
seat
unused
case?
There's
still,
undoubtedly,
although
that
projects,
maybe
an
orchestration
engine
project
similar
to
the
Argo
CD
stroke
flunks
effort,
so
we
do
hear
about
people
using
Argo
workflows
to
see
I
use
cases
that
we
don't
strongly
index
on
that,
because
the
CI
CI
ecosystem
is
incredibly
busy
I'm
wondering
luckily
Kerr.
A
B
So
we
we
do
talk
to
Christy
and
team
Christy
and
tact
on
product
management
team.
Actually
a
lot
of
overlap
with
me,
though.
Okay.
So
even
recently
we
met
Christie
Wilson,
who
is
the
maintainer
of
Tecton
and
and
she
again
I
generated
that
their
focus
is
mainly
on
CI,
whereas
our
focus
is
mainly
on
theta
and
ml
based
workforce
even
before
they
started
the
Tecton
project.
Actually,
we
are
go
for
existed
at
that
time
and
we
we
try
to
build
it
together,
but
I
think
Google
wanted
to
build
a
new
project.
B
So
we
are
trying
to
work
together
collaborate.
We
have
integrations
with
tech
down
around
Argos
Edie,
but
you're
right.
It
is
to
workflow
inching,
but
they
are
specifically
focusing
on
CI
and
we
are
focusing
more
on
ml
and
data,
but
of
course
our
go
workforce
can
be
used
for
CI
use
case
as
well.
So
actually
this
this
this
will
help.
Actually,
if
in
the
meeting
notes,
you
can
also
add
like
you're
using
it
for
CI
new
skills
or
others,
we
keep
asking
this
question.
E
B
E
A
Already
sure
the
moment
around
that's
a
good
question:
it's
just.
It
was
just
the
very
first
testing
of
the
water
to
see
what
people
thought
of
these
ideas
and
see
if
any
hands
will
put
up
or
any
red
flags
were
raised
immediately.
So
I
don't
know
else,
tears,
I,
don't
know
sorry
I
get
a
long
answer,
but
it
could
just
be
I.
Don't
know.
A
So
we've
just
got
one
discussion
topic
today,
I
mean
you
can
always
nominate
other
discussion
topics
if
you
want
to
even
even
now
in
the
meeting
this
has
come
up
recently
around
the
Java
and
Python
SDKs,
and
keeping
keeping
SDKs
in
general
up
to
date
and
I.
Think
why
and
if
I
spot
your
names
correctly
and
you
wanted
to
table
how
what
I
asked
for
the
best
way
to
do
this
was:
is
that
right.
D
Yes,
yes,
so
so
the
our
use
case
is
that
we
are
trying
to
leverage
the
existing
Python
SDK,
but
currently
it
there
are
a
couple
of
issues
with
the
existing
version
of
it.
First
of
all,
it
only
supports
our
go
up
to
2.5
our
C
10
or
something,
and
so,
if
you
want
to
use
the
latest
algo
workflows
that
won't
and
if
you
want
to
use
the
latest
features.
D
So
we
would
like
to
see
if,
if
it's
possible,
to
move
the
SDKs
to
the
same
level
as
other
workflows,
so
that
we
can
always
Auto
generate
all
the
API
specs
and
so
on
so
make
sure
it's
always
consistent
with
the
awkward
photos.
And
this
way
we
can
also
publish
releases
in
the
same
schedule.
So
one
thing
that
confuses
me
is
that
oftentimes,
a
fan
I
found
it
really
hard
to
match
the
pattern,
SDK
version
with
corresponding
our
workflows
version.
So
so,
because
of
those
reasons,
it's
nearly
unusable
at
this
moment.
A
I've
got
a
comment
here
from
Kristen
I'll.
Come
back
to
your
comment,
come
up
from
Michael
Crenshaw
here,
so
he
says
that
he
built
some
one-off
versions
of
the
jar
SDK
and
he's
going
to
look
into
building
an
automated
system
for
the
company's
use
and
he's
hoping
it'll
reasonable
to
contribute
back
I
guess:
there's,
there's
kind
of
an
implicit
I'll
ask
here:
is
anybody
interested
in
and
helping
maintain
the
Python
SDK
for
the
community.
D
Yeah
that
our
team
from
and
financial
is
definitely
interested
in
maintaining
we
are
actually
open
source,
a
pattern
DSL
for
a
DSL
project
to
the
community.
Soon,
hopefully,
I
can
share
the
details
soon,
so
that
would
that
we
all
depend
on
the
Python
SDK.
So,
ideally,
if
it's
something
that
can
be
maintained
well
either
from
us
or
from
other
community
members,
it
will
be
very
beneficial
to
the
community
in
the
long
line.
D
A
So
the
way
that
we
typically
support,
whether
the
core
team
typically
supports
community
projects,
is
we'll
set
you
up
with
a
repository
with
the
appropriate
access
controls
and
access
to
kind
of
continuous
integration,
and
we've
also
got
support
promoting
those
projects
as
well.
So
people
are
aware
of
them
and
they
go
and
find
how
resources
they
need
to
find.
That's,
that's
typically,
the
way
we
do
it
and
pretty
quite
a
good
example,
is
the
Argo
helm
repository.
A
D
Exact
issue
of
the
existing
Python
SDK
repo
is
that
they
are
just
the
main
team
is
not
responsive
right
now,
I
know
it's
a
difficult
time
at
this
moment.
If
we
really
want
to
use
it
and
we
better
come
up
with
a
decision
whether
we
want
to
move
these
to
our
workflows
on
the
same
in
the
same
repo
so
that
it
can
be
validated
and
checked
by
the
existing
CIA
and
make
sure
everything
that
works
all
the
time.
G
H
But
like
reference
it
to
the
workflows,
nothing,
but
a
reference
of
the
template
is
that
the
feature
you're
referring
to
yeah,
yeah
and
yeah
I
was
gonna.
Suggest
that
the
first
thing,
if
you
weren't
taking
advantage
of
the
template
reference
feature.
That's
like
the
first
thing
you
could
do
from
a
auditing
and
security
perspective.
The
the
next
feature,
that's
actually
in
development.
Right
now
is
right.
Now
you
can,
you
can
submit
a
workflow
and
it
could
be
a
fully
defined
workflow
where
you
have
a
spec
and
templates
and
everything
is
defined
in
the
workflow
itself.
H
In
that
way,
the
templates
could
be
controlled
by
say,
operators,
and
then
the
workflows
are
just
submitting
like
instantiations
of
that
of
that
workflow,
and
so
that's
one
secure
the
enhancer
and
I
would
say
that
is
allows
or
prevents
people
from
something
arbitrary
code.
So
to
speak,
but
that's
something
that's
in
development
right
now.
F
F
G
F
H
But
since
you
are
really
using
templates,
it
sounds
like
you're
ready
using
a
template.
Referencing,
yeah
I.
Think
the
really
then
every
if
with
that
in
and
that
new
feature
turned
on,
then
all
auditing
actually
is
on
the
templates.
You
can
control
that
centrally,
and
so
then
you
need
not
be
afraid
of
people.
Submitting
workflows,
arbitrary
were
close
anymore
yeah.
A
This
is
it's
all
kind
of
ties
back
into
kind
of
learning
more
about
what
people
need
from
kind
of
security
in
our
back,
so
Yankee
workflows
excels
at
executing
workflows,
but
you
currently,
you
do
need
to
build
more
around
that
to
get
it
to
you,
get
what
you
need
out
of
it.
We
do
know
some
of
the
engineers
here
at
Intuit.
They
use
kind
of
a
get-ups
base
model
where
users
don't
actually
have
the
permissions
to
create
or
update,
workflows
directly.
It's
done
through
a
platform
that
they've
built
and
the
platform
uses
git
ops.
G
A
A
Okay,
I've
got
why
I'm
looking
in
the
chat
here
so
Jaya
balan
has
asked
general
question:
Argo
workflows,
I
am
totally
new
to
this.
We
want
to
implement
Argo
as
a
replacement
for
Jenkins
long-running
jobs
in
our
company
longer
any
batch
jobs
I
mean
jobs
our
own.
It's
quite
a
long
question
and
run
Python
PHP
scripts
have
run
for
15
hours
and
Jenkins
few
jobs
run
for
20-plus
hours.
We
want
to
provide
a
chain
scalability
in
this
job's
case.
Based
solutions
can
help
us
achieve
that.
Can
I
go
be
the
home
to
these
jobs.
A
I
understand
with
our
goes:
I
really
developed
the
ML
data
science
tasks
and
we
have
two
thousand
six
jobs
and
Jenkins.
They
want
to
convert
to
our
workflows.
It's
interesting
to
hear
a
lot
of
people
talking
about
CI
here,
because
I
think
Jessie.
Obviously
you
were
involved
with
project
originally
and
see.
I
was
actually
one
of
the
key
original
use
cases
for
it,
and
certainly
you
know
if
it's
a
generic
workflow
execution
platform.
We
do
have
some
examples
around
running
CI
based
jobs,
so
it
should
be
possible
to
use
our
go
workflows
for
that
I'm.
A
I
Okay,
this
is
Dee
here
thanks,
so
we
have
north
job
they're,
not
actually
see
a
jobs,
they're
just
one-liners
crane,
that's
going
to
be
executed
on
the
cron
schedule,
so
it's
like
every
day,
I'm
going
to
run
this
crib
and
it's
going
to
take
five
or
ten
hours
for
me.
So
we
are
drawing
some
jobs.
Not
all
of
them
are
CA
jobs
or
not
a
lot
them
more
using
github,
or
something
like
that.
I
I
H
Ours,
what
I
would
say
is
if,
if
that
is
your
use
case,
and
then
a
workflow
is
actually
no
better
than
a
goober
nice
job
in
that
in
circumstance,
because
you
just
think
in
its
is
as
reliable
and
as
a
che
as
a
kubernetes
job
would
be
where
work
clothes
is
more
excels.
It's
actually,
if
you
were
to
split
that
single
entry
point
like
long
20-hour
script
into.
H
Essentially
they
got
chain
reaction
of
like
hundreds
or
dozens
or
of
connected
pods,
like
maybe
that
script
get
to
want
to
work
in
one
pod,
split
a
fan
out
and
do
some
more
work
in
like
100,
parallel,
pods
and
then
fan
back
in
and
then
you
know
continue
if
you
were
able
to
break
that
up
and
workflows
would
be
able
to
leverage
your
your
kubernetes
clusters,
essentially
as
a
compute
grid
of
nodes
and
with
respect
to
HJ
and
scalability.
The
each
individual
pod
is
only
as
reliable
as
the
underlying
cluster.
H
Okay,
so
with
like
preemptable
and
system
gke,
they
might
just
shut
down
nodes
from
underneath
you,
and
so
that's
nothing
that
either
Cabrini's
jobs
or
workflows
is
going
to
prevent
that
from
happening.
But
as
long
as
the
underlying
platform
remains
up
like
those
those
pods
will
write
were
close,
will
ensure
that
it's
it
doesn't,
it
creates
at
most
once
or
at
least.
H
Actually
it
should
be
exactly
one
syntax
where
our
semantics,
where,
if
you
create
a
a
pod,
we
will
not
try
to,
and
let's
say
the
controller
crashes
before
persisting
the
say
it
will
actually
try
that
when
it
comes
to
a
couple,
try
to
create
that
same
pod
with
the
same
name,
and
so
in
that
respect.
This
item
potent,
and
so
it
has
a
exactly
when
semantics,
on
on
the
creation
execution
of
the
nodes
in
your
workflow.
H
I
So
the
reason
why
we
consider
our
goal
was
it
provides
a
lot
of
features
that
Jenkins
provides
like
the
slack
notification
and
the
dependencies
like
after
execution
of
one
workflow
it
can
based
on
the
results.
I
can
execute
another
workflow,
but
some
of
the
things
are
not
clear,
like
in
Jenkins,
I
can
go
to
the
UI
and
I
can
look
at
the
console
output,
lively
like
what
is
happening
in
my
scrim,
but
and
we
have
a
trend,
2000
users
in
Jenkins,
and
we
have
20
insta
Lester.
I
So
if
you
want
to
move
to
cargo-
and
we
have
like,
we
are
targeting
2,000
jobs
now
and
let's
say
we
are
splitting
it
into
five
clusters
with
like
400
jobs,
each
and
then
I
want
to
provide
access,
2000
users.
So
how
am
I
going
to
handle
that
like?
How
can
handle
the
user
management
here
and
also
I
would
like
to
know
how
the
console
output
works
here
in
our
group,
because
all
the
UI
is
kind
of
still
in
the
development
stage.
Right.
H
For
that
scale,
so
as
I
like
solute
to
earlier,
so
we
we
recently
just
implemented
single
sign-on
to
workflows,
but
the
implementation
is
that
is
such
that
once
someone
gets
past
a
single
sign-on
they're,
actually
just
using
the
the
same
auth
that
anyone
got
that
got
past.
That
signal
sign
on
with
so
everyone
is
basically
using
the
same
service
service
account.
So
we
don't
have
the
sophisticated
our
back.
A
The
biggest
thing
I've,
certainly
found
to
make
it
easier
to
do
is
to
ensure
that
your
kind
of
CI
jobs
are
kind
of
robust
to
being
rerun.
It's
also
kind
of
applies
to
our
go
workflows
as
well.
You
know
they
run
fundamentally
on
an
unreliable
system.
Kubernetes,
you
know
with
even
if
you're,
providing
very
large
clusters
and
you've
got
pod
disruption
budgets,
it's
still
possible
for
a
step
within
your
workflow
to
fail
for
things
that
are
outside
your
control.
A
So,
having
you
know,
very
long,
running
pods
as
part
of
your
workflow
means
that
that
workflow
will
always
a
risk
and
having
those
pods
being
killed
at
some
point
during
the
process
and
there's
nothing
worse
than
waiting
five
hours
for
a
workflow
to
fail.
Because
of
something
unrelated
to
anything,
you
do
what
what
and
what
kind
of
you
might
want
to.
Try
is
breaking
down
workflows
into
smaller,
shorter
steps
and
making
its
kind
of
ident
ie
if
they've
run
before
they're
kind
of
smart
enough
to
know
that
they
run
before
by
looking
at
the
app.
A
A
Anybody
who's,
run
Jenkins
job
and
got
to
you
know
complex
Jenkins
job
with
many
many
steps,
and
it's
got
to
the
end
of
its.
You
know
to
our
job
and
then
the
last
step
fails
because
of
the
network
transient
network
issue,
kind
of
knows
that
kind
of
pain
and
then
being
able
to
deal
with
that.
It's
always
really
helpful.
A
H
Yeah
I
did
forget
to
mention
the
the
retry
feature
of
work.
Well
is
actually
pretty
powerful
in
that
like
when
you
get
to
that
same
scenario,
where
last
the
last
step
of
some
workflow
fails
when
you
retry
workflow,
all
of
the
preceding
steps
are
not
executed
all
and
then
they'll.
Only
the
last
step
is
actually
the
one
that
is
retried.
A
A
Well,
thank
you
very
much
for
everybody.
Who's
come
along
today
and
thank
you
for
Derek
and
everybody
else.
Who's
brought
up
the
discussion
topic.
They
want
to
ask
if
you've
got
any
questions,
you
want
to
ask
obviously
come
and
find
us
in
the
slack
channel
when
you
feel
about
you
want
to
provide.
Also
it's
about
the
SDKs.
Come
funds
in
the
SDKs
channel
on
slack
and
have
a
chat
there
and
I
will
give
you
four
minutes
back
of
your
day.
Thank
you
very
much
thanks.
They.