►
From YouTube: CDF SIG MLOps Meeting 2020-02-27
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
A
Revised
comparison
between
the
queue
for
pipelines
are
go
and
Tecton,
and
then
you
know,
question
let's
go
through
you
know,
I
would
say
the
work
in
progress,
designing
specifications,
talk
and
and
the
further
work
you
have
done
in
terms
of
you
know,
updating
the
compiler
that
your
flow
pipeline
compiler
to
generate
spectrum
pipelines
over
that
and
yet
do
you
want
to
go
first
and
get
started
so
by
the
way.
Jeremy's
here
you
know
he's
from
the
cue
flow
team.
He
leads
the
cue
flop
project.
A
C
Cool
yeah,
so
I'll
just
get
the
comparison
again
this
time,
starting
from
more
of
the
50s
DK
side.
So
the
previous
comparison
was
more
centered
in
comparing
Harbor
City
features
against
actin.
Then
we
started
from
what
is
actually
used
in
the
compiler
and
mapped
into
what
is
on
the
harder
side
and
what
it
could
be
objective
side,
of
course,
the
compiler
package,
the
first
package.
C
We
know
that
there
is
a
compiler
for
Argo
and
there
is
a
prototype
compiler
on
Tecton,
which
is
work
in
progress,
so
in
terms
of
just
just
go
through
the
list
here.
So
there
are
some
functionality
to
create
a
container
operation
in
the
SDK
from
a
Python
function
or
loaded
from
from
a
file,
and
that
is
easy,
easily
implemented
in
the
in
Tecton.
So
that
would
be
a
task
intact
to
the
side.
C
So
in
terms
of
container
operation,
I
think
there
is
a
good
support
intact
inside
one
thing
that
we
don't
have,
but
there
is
a
workaround
is
the
set
image
full
secret?
Actually,
we
may
have
discussed
adding
it
to
to
the
TAS
directly.
The
way
Tecton
handle
this
is
is
by
specifying
a
service
account
which
has
the
required
image
pool
secret
attached.
So
that
would
be
the
way
to
do
it
intact
on
today.
C
C
In
fact,
Tecton
has
a
catalogue
of
tasks
like
usable
tasks,
and
one
way
for
work
would
be
to
define
those
tasks
in
the
catalogue,
you're,
also
working
on
extra
features
that
will
allow
us
to
appoint
to
a
catalog
in
the
task
by
reference,
for
instance,
so
that
it
doesn't
need
to
be
imported
first
into
the
into
your
cluster.
So
we
could
just
reference
to
it
and
that
would
make
the
compiler
work
even
easier.
A
We
will
also
need
to
addresses
you
know
the
components
which
we
are
creating
for
queue
flow
pipeline,
which
are,
you
know,
essentially
docker
containers,
but
there
is
a
components
toriyama
file
which
defines
you
know
the
input,
the
output,
the
packing
docker
image
for
this
component.
If
we
can
start
doing
that
in
a
generic
way
as
well,
so
that
you
know
they
can
be
reusable
across
both
Tecton
as
well
as
queue
flow
pipeline
right
and
so
I,
don't
know
whether
that's
something
you
know.
C
Yes,
so
that's
absolutely
so:
that's
I
think
in
terms
of
SDKs
is
implemented
as
load
component
from
Yammer,
for
instance,
which
is
on
line
8
and
yeah.
So
that's
that
can
be
mapped
into
into
detect
on
tasks.
So
I
mean
from
the
SDK
point
of
view.
The
component
is
loaded
from
from
a
file,
and
that
would
be
the
same
format,
I
believe,
regardless,
whether
it's
our
goal
or
tectum.
C
So
we
could
have
the
same
component
for
motto
and
then
it
would
compile
into
our
goal
workflow
image
or
attacked
on
attacked
on
tasks,
something
that
you're
working
experimented
with
right
now
is
the
ability
to
store
tasks
into
the
registry
so
into
the
container
image
registry
as
artifacts.
So
you
know
CI
format
and
that
would
allow
us
to
pull
things
like
are--some
components,
using
a
specific
version
like
a
char
specific
Shou
like
it
is
done
for
container
images.
D
C
E
A
Home
from
our
go,
it's
more
like
you
know,
providing
Techtron,
also
as
a
target
backing
to
run
these
pipelines.
So
I
think
specifically
again
speak
on
behalf
of
IBM
and
Red
Hat
right,
so
we
have
standardized
on
packed
on
as
the
native
CI
CD
pipeline
for
the
cloud
and
that's
the
motivation
on
our
side
over
that.
A
You
know,
if
that's
something
that
is
already
going
to
be
embedded
in
OpenShift
as
a
supported
pipeline
on
our
IBM
cloud
as
a
supported
pipeline
engine
it,
the
the
internal
teams,
are
requesting
that
you
know
for
something
like
you
flow,
it
would
make
sense
sitting
on
top
of
Tecton,
rather
than
something
like
Argo
right
for
which
they
will
lead
to
figure
out.
You
know
getting
people
trade,
my
particular
skill
set
and
supporting
that
so
I
think
that's
the
primary
motivation
on
our
side.
Jeremy.
Do
you
have
any
comments
on
the
motivation
from
your
side?
Yeah.
E
I
think
I
think
there's
two
I'm
Tom
Jeremy
I
work.
A
lot
on
kukuku
go
I.
Think
there's
two
I
think
this
is
just
my
two
cents.
So
the
first
is:
if
you
look
at
pipelines
right
now,
Argo
is
not
a
first-class
citizen,
and
so
I
think.
One
thing
that
pipelines
is
missing
is
a
well-defined
resource
model
that
provides
a
little
bit
more
abstraction
between
the
Python
SDK
and
the
actual
underlying
resources,
and
so
I
think
you
know
the
one
question
is
you
know:
should
it
be
Argo?
E
Should
it
be
tacked
on,
should
it
be
some
new,
well-defined
concept,
new
resource
model,
right
and
so
I
think
you
know,
NMS
just
mentioned
pipelines
defined
its
own
sort
of
llamó,
which
has
for
a
component,
which
is
basically
the
notion
of
a
reusable
task.
Now
that's
something
that
I
don't
think
Argo
really
had
originally,
but
what
well
it
did
happen.
Tecton
did
have
that
in
the
form
of
the
task,
resource
and
so
I.
E
Think
to
me
you
know,
one
of
the
questions
is,
you
know:
does
pipelines
really
need
to
introduce
a
new
notion
of
a
reusable
task,
or
should
it
just
be?
You
could
have
just
use
or
that
just
use
you
know
a
Tecton
attacked
on
tasks
and
then
I
think
I
think
Argo
I
haven't
paid
attention
recently
to
are,
though,
but
I
think
in
the
past
year,
or
so
they
might
have
introduced
something
more
akin
to
a
task.
E
I
don't
know
so
maybe
maybe
that
gap
has
been
closed
and
so
maybe
that
could
be
used
in
in
or
ago
and
then
I
think.
Also
just
you
know,
as
anima
says
you
know,
certain
companies
are
getting
kind.
Various
solutions
like
Tecton,
like
so
Google's
heavily
invested
in
found
as
well,
so
I
think
so
you
know
come
on
from
that
perspective.
You
know
we're
aligning
with
you
know
a
huge
team
at
Google,
that's
working
on
tech,
doc.
C
If
I
can
add
something,
it
doesn't
add
too
much,
but
from
a
technical
community
point
of
view,
Tecton
really
aims
to
be
like
a
common
denominator
and
continuous
delivery
scenarios,
which
is
the
ICD
but
also
delivery
of
machine
learning,
artifacts
so
I
think
there
is
a
general
willingness
from
the
Tecton
folks
to
integrate
as
much
as
possible
with
solutions
on
top
of
Tecton
and
to
be
like
yeah
come
on
integration
the
year.
There
is
a
thing
that
fits
with
that
aim:
that
direction
as
well
yeah.
A
I
mean
even
Jenkins
X,
you
know
they're,
never
aging
Tecton
under
the
cover.
So
if
you
think
or
see
you
know
the
trend,
this
cloud
native,
CIC
Lee
community
is,
you
know,
lining
around
Tecton,
so
I
think
it
makes
sense
I'm
just
looking
at
the
time
and
we
all
right.
So
there
are
a
couple
of
things
you
would
want
to
call
out
which
you
think
are
or
can
be
long-term
blockers
in
terms
of
getting
the
parity
at
a
functionality
level
between
orgo
and.
C
Long-Term,
not
really
so
the
two
things
that
are
marked
yellow,
so
the
access
handler
and
the
the
right
one,
the
recursion
I
think
the
the
exit
endler
is
something
that
is
actually
really
close.
So
it's
been
actively
worked
on
and
we
are
very
close
to
having
that
functionality
in
parallel
is
under
design,
so
we
were
discussing
about
it
so
we're
discussing
whether
to
do
its
own
task
level
on
step
level
or
both.
So
this
is
big
design,
but
I.
Don't
think
it
will
be
a
long
term
to
implement
that.
C
A
At
this
point,
I
would
say:
I
mean
I
haven't
seen,
I
mean
90
was
another
cue
floor.
95%
of
queue
flow
pipeline,
recursion
is
not
used,
so
I
would
say
you
know.
That
is
one
of
those
cases
where
we
probably
don't
bother
too
much
at
this
point
things
like
yes,
volley.
More
volume
snapshot
right,
the
exposing
the
Native
communities,
capabilities,
running,
sequential
parallel
jobs,
right.
E
A
C
A
Andrea,
so
please
keep
on
updating
this
right
and
we'll
you
know,
as
we
move
through
some
of
this
work.
I
will
keep
on
prioritizing
some
of
these
requirements
we
find
out.
You
know,
which
should
be
intact
along
with
that
Christian.
Do
you
want
to
share
your
screen
and
let's
do
a
quick
walkthrough
of
what
you
are
being
up
to
and
Jeremy
a
Christian
and
Andrea
right?
So
Andrea
is
our
tectonic
emitter
from
IBM
in
the
Techtron
community.
Christian
works
with
us
on
the
house.
B
B
F
Yep
I'm
trying
to
click
through
the
zoom
share
screen
buttons,
and
it
wants
me
to
go
to
my
system
preferences
and
one
second
I
don't
know
I
can
share
it
on
my
and,
if
you're
having
trouble,
you
know
it
wants
me
to
quit.
Zoom
and
restart
I'll
do
that.
Anyway,
maybe
you
can
start
with
the
Google
Docs
image.
A
So
I
think
you
know
Jeremy.
This
is
something
we
discussed
or
it's
so
part
of
it
was
you
know,
essentially,
yes,
putting
up
a
plan
and
figuring
out.
First,
you
know
the
gaps
in
Tecton
which
is
needed
for
a
full
featured.
You
know,
cue
flow
pipeline
checked
on
compatibility,
so
you
know
that's
an
exercise
which,
which
has
been
going
on.
A
The
other
thing
is,
which
is
essentially
some
of
the
requests
which
we
are
getting
from,
for
example,
from
internal
teams
or
more
like
you
know,
they
just
need
the
compiler
right,
because
UI
is
a
big
piece
of
it.
If
you
see
the
two
flow
pipeline,
UI,
it's
heavily
dependent
on
the
Argo
channel,
the
way
it
is
constructed.
The
way
it
is
you
know,
displays
their
dad
acceptor.
A
All
right
so
I
think
that's
an
aside
piece
of
work
which
will
require
you
know
pretty
extensive
investigation
in
in
general,
but
even
beyond
that
I
think
the
minimum
thing
which
you
know
some
of
the
teams
we
talked
to
they
want
it
was
you
know:
okay,
I
mean
in
any
case,
you
know
a
lot
of
the
products
that
they
are
only
ye
can
be.
You
know,
take
that
compiler
generated
sorry,
the
DSL
and
generator
tech
phone
equivalent
right,
so
that
has
been
the
main
driver.
A
A
A
Our
initial
set
of
instructions
have
been
around.
You
know,
I'll
just
jump
to
you
know
the
items
he
has
been
looking
at
right.
So
essentially
we
were
copying
part
of
the
compiler
code
here
into
the
KP
Techtron
repository
generating
that
sequential
pipeline,
which
was
mapped
to
the
steps
right.
So
quite
a
few
things
wrong
there
right,
so
he's
been
fixing
it
a
adding.
A
A
E
Oh
thanks,
I
think
this
is
really
great.
I'm,
certainly
excited
about
tech
time.
I
mean
I.
Think
to
me,
I
think.
The
big
question
is,
you
know,
is
the
way
to
start
incrementally
introducing
into
kupo
so
that
you
know
people
start
seeing
it
and
people
start
using
it,
and
you
know,
because
just
replacing
pipelines
with
backgrounds,
a
pretty
big
chasm
so
I
think
one
one
thought
that
you
might
want
to
look
into
is
I
I,
think
sharing
low-hanging
fruit
in
terms
of
starting
to
replace
some
of
the
functionality
and
fairing
with
tech
Don.
E
A
Okay,
make
sense:
oh
I
was
I
was
conversing
with
Janie.
Yesterday
right
we
were
actually
discussing
sharing
a
bit
part
of
the
reason.
Is
you
know
in
terms
of
I
think,
as
you
have
also
tried,
the
roadmap,
but
in
general
right
I
wanted
to
see
the
applicability
of,
or
involvement
in
it
with
service.
You
know
the
the
business
drivers
or
so
to
say
so,
I
think
the
Techtron
work
probably
will
or
can
act
as
a
business
driver
on
that
side
right
so
I.
E
So
you
can
have
one
text
on
packages
that,
as
a
pea
processor
converts
a
notebook,
you
can
have
another
Tecton
task
that
uses
Kanaka
versus
build
kids,
actually
build
it
and
then
sharing
so
it
just
becomes
like
a
Python
SDK.
On
top
of
that,
you
know
some
sugar
for
the
phonetics
treat
that
with
pipeline
would
Python
it
supposed
to
just
do
in
raw
mo
and
then
that
sort
of
merges
those
two
pretty
nicely
I.
Think
no.
A
No
I
hear
you
and
I
think
one
of
the
things
which
you
mentioned
regarding
in
other
notebooks
does
that's
something
which
we
are
doing
internally.
We
have
a
cue
flow
pipeline
component
which
can
take
a
notebook
and
and
run
it.
You
know,
and
that
component
is
essentially,
you
know,
has
Peppermill
under
the
covers
right,
so
the
the
that
will
also
align
you
know
one
of
the
things
which
I
need
to
think
is
you
know,
so
we
have
this
catalog
of
cue
flow
pipeline
components
right.
A
So
now,
when
you
talk
about
you
know
the
catalog
of
Tecton
tasks
right,
we
will
need
to
figure
out
how
to
do
them
in
a
generic
way
so
that
you
know
they
are
reusable
across
both
you
can
take.
You
know
KP,
as
is
and
run
it
as
inaudible,
because
in
the
end
these
are
docker
containers
right.
So
so,
if
we
be
a
little
bit
more
intelligent
about,
you
know
constructing
the
metadata
around
it.
Hopefully
we
can
reuse.
You
know
in
a
consistent
manner
across
both
of
them
yeah.
C
A
F
It's
some
spotting.
My
screen
is
this:
sharing
the
Google
Doc
right
now,
my
entire
screaming
here:
okay,
fantastic
I'm!
So
yeah
thanks!
You,
you
walk
through
what
you
do
share
your
entire
screen.
So
you
know
that
yes,
I've
been
careful
not
to
share
pictures
of
my
wife
and
daughter
here,
but
yes,
how
that
was
the
easiest
I
had
too
many
screens
open.
Thank
you
right.
So
I've
been
focused
more
on
the
very
low-level
tasks
of
getting
the
q4
pipeline
to
kV.
F
Tecton
project
that
you
might
have
seen
may
have
seen
into
a
shape
that
developers
can
collaborate
on
that
right.
So,
if
you
go
to,
if
you,
okay,
if
you
checked
on
it's
a
very
small
project,
still
has
a
few
samples
and
has
the
SDK
and
the
SDK
part
is
the
the
part
that
I've
been
working.
Working
on
I
have
been
doing
so.
I
thought
that
was
a
step
too
far.
I've
been
doing
Python
development.
F
For
you
know
a
couple
of
years
now,
I'm,
coming
from
a
Java
background
and
I'm,
really
keen
on
being
able
to
you
know,
go
to
github
clone
a
project
and
you
know
try
it
out
and
chipped
away
and
that
wasn't
possible
with
this
project
before
mainly
because
you
had
to
do
things
like
you
know,
get
the
cue
flow
pipelines
project.
You
know
clone
dead
somewhere
locally,
build
it
copied
the
good
files
from
this
particular
KP
tacked
on
compiler
over
there
q4
pipeline
project.
F
You
know
and
then
built
a
give
a
Python
project,
and
then
you
know
use
it
as
you
would
use
kfp.
If
the
caveat
that
you
basically
eliminate
the
KP
functionality
in
terms
of
the
compiler
and
now
only
produce
tacked
on
so
I
did
make
an
effort
to
change
a
few
things
in
this
project
and
make
it
so
that
people
can,
you
know,
basically
go
and
collaborate
and
not
having
to
copy
code
around
and
worry
about.
F
The
versions
of
the
kfp
of
the
KP
version
that
the
original
code
was
being
copied
from
and
copying
that
back
and
that's
just
a
just
a
headache
right.
So
in
that
I
think
I
have
a
PR
open
to
make
many
of
those
changes,
and
now
the
steps
are
more
streamlined.
F
So
typical
you
create
your
virtual
environment,
you
have
a
pip
install,
you
know
to
get
your
project
installed
and
then
you
can
run
the
test
from
the
command
line
or
you
can
run
it
in
your
IDE
and
I
separate
it
as
the
command
line
executable
for
DSL
compile
I
created
these
occupy
detect
on.
So
you
can
run
it
in
parallel
to
your
standard
cube
flow
SDK
compiler.
F
Steps
in
the
pipeline
to
Tecton
steps
inside
one
task
and
instead
mapping
the
individual
private
operations
into
a
task,
so
they
can
run
in
parallel
and
we
can
then
also
make
some
progress
on
on
using
worst
places
and
sharing
sharing
data
for
input
and
output
parameters.
So
this
is
where
I
am
right
now.
The
next
bigger
concern
I
have
is
how
to
stage
that
work
in
a
way
that
we
can
feed
it
back
into
the
queue
flow
pipeline,
SDK
and
compiler,
and
maybe
in
the
future
habits
innovated
the
queue
flow
pipeline,
SDK
and
compiler.
F
They
have
a
a
command
line,
flag
way
and
say:
I
want
to
generate
Tecton
instead
of
the
Argo
llamó
and
the
code
would
all
live
in
one
repository
to
get
to
that
point.
I've
made
an
effort
to
to
make
make
sure
that
the
methods
that
we
had
to
override
to
generate
detect
on
Yama
they
follow
the
same
project
structure
as
in
db2
flow
pipelines,
SDK
or
the
same
class
name,
same
method
names
and
for
now
I'm
using.
A
That's
probably
the
time
you
want
to
get
Jesse
and
team
involved
from
the
cue
flow
pipeline
and,
as
Christian
mentioned
right,
align
that
so
that
you
know
we
start
working
in
a
common
goal
that
at
some
point
we
want
to
merge
into
the
main
stream
through
flow
pipeline
repo
and
give
that
option
for
a
flag
right.
So
that's
I
think
there
is
going
to
be
a
milestone
fairly
soon
where
we
will
need
to.
You
know,
have
a
conversation
with
the
queue
for
pipeline
team
right.
A
So
hopefully
you
know,
maybe
in
a
couple
of
weeks
or
so
right,
so
I
think
that's
something
you
know
we
are
hoping
you
can
facilitate
on
that
side.
I
also
intend
to
get.
You
know
that
that
probably
is
also
the
right
time
to
get
Red
Hat
it
more
in
as
well
that
you
know
this
is
something
which
can
now
start
turning
into
something
real
thanks.
Jeremy
I
do
need
if
you
I,
do
need
a
minute
of
your
time.
You
know
for
discussing
miss
of,
is
you
know
something
miss
Avista
announcement
right.
A
A
Me
Jeremy
yeah,
so
say
my
service.
You
know
the
mail
I
sent
us
here
was
more
around.
You
know
the
coordination
I
know
that
I
mean
the
technical
part
is
done
right,
I
mean
she
was
talking
about.
Yes
in
blog,
but
in
terms
of
you
know
being
having
to
coordinate
across
you
know
some
of
the
blogs
plus.
You
know
if
she
is
doing
any
press
release
right.
So
that's
what
my
meaning
take
was.
A
Even
it
was
not
I
mean
for
when
I
had
sent
her
a
draft
of
the
blog
right,
which
now,
given
the
time
frame,
we're
going
to
for
the
most
part
right.
We
are
going
to
publish
right
what
I
wanted
to
get
her
eyes
on
it,
but
I
didn't
hear
anything
back
right.
So
there's
what
I
was
asking
you
you
know:
I
can
ping.
You.
E
A
A
You
know,
but
in
general
the
thing
was,
you
know
what
we
wanted
was
a.
We
want
to
hyperlink
to
the
main
blog
right,
which
is
coming
from
your
side,
so
that
that's
one
thing
I
wanted
from
her
is
you
know?
Is
there
a
standard
URL
to
which
you
know
which
we
can
put
in
right
now
so
that
you
know
when
this
goes
live
you
know,
thou
shalt
be
able
to.
A
We
shall
be
able
to
point
to
the
other
thing.
Was
you
know?
Do
we
need
any
other
coordination
because
from
our
mail
it
up
sheared,
you
know
she
wanted
to
coordinate.
Quite
a
bit
of
you
know,
activities
across
orgs
right,
but
given
that
it
is
Thursday
and
Monday
is
when
this
is
happening
right
so
I'm
a
little
bit.