►
From YouTube: CDF SIG MLOps Meeting 2020-06-18b
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
C
B
Quick
update
on
the
this
morning's
a
specific
session
I've
been
looking
at
various
areas
of
the
roadmap
document
over
the
course
of
the
development
on
technology
requirements.
But
the
discussion
today
was
primarily
around
the
legal
compliance
and
ethical
aspects
involved
with
machine
learning
projects,
and
that
raised
some
quite
interesting
discussion
about
the
need
for
a
mops
tooling
to
actually
have
some
level
of
understanding
of
some
of
the
challenges
in
that
space
and
actually
enable
governance
processes
which
go
quite
a
lot
further
than
those
were
typically
used
to
finding
in
conventional
clcd
environments.
B
B
B
There's
already
quite
stringent
regulation
around
obtaining
permissions
to
use
people's
data
to
train
models
and
that
regulation,
or
also
grants
the
right
to
withdrawal
that
permission
at
any
time.
So
we
need
to
consider
the
requirement
for
solutions
to
be
able
to
understand
is
a
lea,
a
database
of
permissions
and
the
fact
that
if
anyone
withdrawals
information
that
potentially
invalidates
deployed
models
and
triggers
a
need
to
retrain
and
redeploy,
you.
B
B
Typically,
what
we're
looking
to
do
with
the
the
technology
requirements
section
is
flesh
out
known
requirements,
so
the
bits
that
we're
looking
to
fill
out
this
document
should
be
phrased
in
the
form
of
a
requirement
and
then,
where
we
already
know
about
potential
solutions,
then
the
narrative
section
in
the
final
chapter
here
is
is
where
we
can.
We
can
discuss
what's
available
at
the
moment
and
how
have
those
things
might
potentially
change
over
time
to
reflect
you.
B
B
B
D
D
This
group,
you
know
just
by
googling
around
and
I,
was
really
interested
to
come
in
and
see
what
people
are
talking
about,
and
my
question
is
the
following:
you
know
is
this:
is
this
in
the
service
of
kind
of
developing
a
platform
to
do
these
things,
or
is
this?
You
know
more
of
a
codification
of
the
process
that
people
go
through
such
that
you
can
use
that
as
a
template.
B
What
the
what
the
requirements
are
in
terms
of
what
people
need
to
be
able
to
do
in
that
space.
What
what
the?
What
the
challenges
are
and
then
how
we
can
leverage
the
best
practice
that
we
already
have
and
some
of
the
tools
that
we've
we've
been
using
for
many
years
to
to
actually
close
the
gap
and
make
it
much
much
easier
to
merge
machine
learning
assets
as
first-class
citizens
within
a
a
software
asset
management
plan.
B
It's
a
lot
of
sensible
heat.
The
the
intention
of
the
the
roadmap
really
is
to
just
just
have
one
place
where
we
really
clearly
spell
out
what
the
challenges
are
and
then
paint
a
vision
for
what
we
might
be
able
to
achieve
in
terms
of
simplifying
and
automating
and
and
generally
improving
the
capabilities
to
deliver
those
those
individual
requirements
as
as
a
product
pipeline
and
then
do
that
in
such
a
way.
That
is
it's
open
for
for
anyone
to
to
come
in
and
and
get
involved,
and
you.
E
E
One
thing
sireesha,
I
would
like
to
add
is
that
if
you
are
interested
in
something
more
on
the
ground,
more
concrete
implementation
right,
so
another
activity
which
we
are
driving
in
the
seg
is
essentially
you
know
to
flow
pipelines
and
Tecton
coming
together
to
address.
You
know
a
large
number
of
these
use
cases
and
requirements
right
and
the
pretty
advanced
in
terms
of
you
know,
handling
a
bunch
of
these
requirements
which
are
originating
in
the
ml
ops
world
right,
but
definitely
you
know
there
is
a
lot
more
to
be
turned.
D
D
E
D
D
Think
it's
it's
useful
to
be
able
to
see
something
like
the
roadmap
to
point
to
issues
with
people
which
people
in
the
broader
software
community
are
experiencing,
that
we
may
not
have
been
to
scale
to
experience
yet
so
you
know
from
my
perspective,
the
roadmap
has
already
been
useful
in
terms
of
informing
what
we're
going
to
need
to
think
about
it:
ml
ops,
even
a
little
bit
down
the
road,
so
I
mean
thank
you.
Thank
you.
Thank
you
all
for
putting
that
together.
Thanks,
Mitch
Mitch
organization.
Do
you
represent
cities?
D
So
it's
called
kensho
technologies.
It's
a
wing
of
S&P
global.
D
B
E
Urinal
right
so
I
think
your
flow
deployments
on
Azure,
and
that
was
I
mean
the
point
is
intentionally
if
you
are
deploying
and
leaving
the
end
nodes
exposed
without
you
know,
securing
them
with
an
authentication
authorization
on
an
ingress
gateway
right.
That's
probably
one
of
the
problems
you
know
needing
to
be
exploited
right,
so
cue
flow
comes
in
different
flavors
right.
Where
one
of
the
flavors
is.
You
know
you
use
a
combination
of
sto
and
X
for
authentication
and
authorization
right
head.
E
Tightening
that
controllers
also
the
deployments
which
have
been
exploited
or
or
I
would
say.
You
know
the
freewheeling
clusters
of
cute
flow
and
and
part
of
the
reason
you
know
they
became
an
interesting
target
is
because
a
lot
of
the
people
do
more
heavy
duty.
You
know
GPU
oriented
stuff
under
the
covers
right.
So
so
the
nodes
are
amenable
like.
E
If
somebody
has
deployed
chip
flow
in
all
probability,
you
know
they
have
certain
nodes
which
I
keep
user
billable
and
those
are
the
kind
of
you
know
computing,
most
power
which,
which
you
know
these
Bitcoin
miners
need
right.
So
but
yeah
I
mean
there
is
a
discussion
on
it
on
on
the
queue
flow
thread
as
well
right
as
well,
as
you
know,
recommendations
if
and
more
than
a
recommendation.
I
think
you
know
based
on
this
incident.
It's
probably
your
requirement
now
that
you
know
should
appoint
you
flow
of
these
public
Cloud
endpoints.
F
E
F
Like
you
said,
my
name
and
I
appeared
just
to
be
incredibly
clear.
This
was
you
know
more
than
a
little
bit
my
fault
we
had
published
on
the
cupid
website,
documentation
on
how
to
set
up
keep
flow
on
a
casts,
the
azure
kubernetes
service.
In
it
we
had
a
you
know,
one
of
those
optional
insert
boxes.
That
said,
hey
if
you'd
like
to
open
this
dashboard
to
the
world,
here's
the
configuration
you
need
to
do
make
sure
to
secure
it
before
you
do
so.
F
You
know
using
Sto
and
enough
people
did
it
and
enough
people
didn't
secure
it.
Then
we
ended
up
both
removing
the
documentation
and
and
issuing
an
advisory.
But
this
is
no
different
than
you
know.
A
couple
years
ago,
kubernetes
had
a
dashboard
exploit
that
they
used
that
people
ended
up
mining
against
exact,
same
thing.
There
was
nothing
there
was
and
is
nothing
about
the
default
queue
flow,
install,
that's
insecure
it
just
you
know.
B
So
I
think
there's
there's
probably
an
interesting
learning
to
come
out
of
this,
that
we
might
want
to
include
in
in
the
roadmap,
because
again,
this
is
another
example
where
we
can't
necessarily
rely
on
the
fact
that
the
team's
deploying
these
tools
have
in-depth
security
knowledge
to
understand
the
implications
of
some
of
the
things
that
they're
doing
when
they're
deploying
so
so
we
should.
We
should
really
have
a
requirement
that
that
says
that
the
tooling
should
expect
that
that
it's
being
deployed
or
its
deploying
itself
in
such
a
way
that
is
intrinsically
secure,
yeah.
F
Point
I
think
it
might
be
interesting
to
expand
beyond
even
that
and
and
really
stress
that
the
CD
pipeline
I
give
quite
a
lot
of
talks
where
I
talk
about
the
CD
pipeline
being
critical
for
security,
meaning
not
not
that
it
is
inherently
secure.
But
at
the
moment
that
you
discover
something
is
is
insecure.
F
This
is
a
way
to
remediate
this
quickly
and
and
reliably,
because
you
have
this
pre-built
pipeline
that
you
can
say:
okay,
I'm
gonna,
make
this
configuration
change
and
push
it
out
and
nothing
else,
changes
which
is
obviously
quite
powerful,
so
yeah,
I,
completely
agree
and
I
think
we
can.
We
may
be
able
to
go
even
further
than
that.
Yeah.
F
E
Also,
you
know
the
next
in
the
set
with
what
we
have
integrated
is
you
know
the
metadata
in
the
lineage
right.
So
there
is
a
workflow
oriented
view,
which
is,
you
know,
go
to
step
ABCD,
but
then
now
there
is
also
you
know
the
concepts
of
dataflow
integrated
in
it,
which
is
essentially,
you
know,
track
back
the
lineage
and
look
have
a
tag
of
every
input
and
output
flowing
from
those
ABCD
as
well
as
trace
back
from
D
back
to
it.
So
all
that
has
been
you
know,
integrated
and
I.
E
E
Think
couple
of
things
which
we
are
still
you
know
hoping
you
know
the
check
tone
community
as
such
drives
further
on
as
the
two
big-ticket
items
is,
you
know
the
looping
support
which
I
think
and
maybe
I
don't
know
Andrea.
If
you
have
any
insights,
it's
probably
fell
through
the
cracks.
I
believe
Greg
started
something
on
it
designed
all.
They
were
supposed
to
be
comments
and
I,
don't
know
where
it
is,
it's
probably
more
than
a
month
and
I
think
the
seismic
condition
spec
moving
to
beta
right.
G
So
both
of
them
are
being
worked
on
and
didn't
really
fall
through
the
cracks
I
mean
there's.
There
is
a
design
that
Greg
worked
on
and
he's
able
to
make
some
progress
on
it.
There
is
a
dependency
on
different
feature
that
is
also
being
developed,
so
Greg
is
able
to
make
progress
on
that
and
in
parallel
at
some
point
there
would
be
a
dependency
on
the
other
feature
being
being
Brady.
So
we're
looking
in
what
is
the
timeline
for
that
and
but.
E
G
So
they're
they're
progressing
for
the
conditionals
we.
So
there
was
an
initial
review
cycle
and
there's
going
to
be
the
updated
version
based
on
all
the
feedback
in
the
next
API
working
group
on
Monday
and
yeah.
Hopefully
there
will
be
the
define
the
revision,
then
we
can
start
implementing
it,
but
definitely
there
is
a
focus
on
on
those
as
well,
because
we
started
discussing
also
GA
plants
and
those
both
are
well
at
least
the
conditionals
are
independent
dependency
for
for
that.
So
so.
E
Okay,
definitely,
you
know
paradise
condition
of
this
dimension
right
as
part
of
that
I
mean
more
so
than
the
triggers
I
mean
in
general
right.
There
is
Canada
which
exists
right
for
triggering
and
eventing
right.
So
I
don't
know
if
there
was
an
overarching
need
for
introducing
triggers
in
Tecton
as
well,
given.
D
E
You
know
they
are
part
of
the
same
or
similar
family
right,
I
mean
originated
from
Canada
sirens
and
skinny-dip
addresses
pretty
richly
quite
of
the
aspects,
our
own,
you
know
triggering
any
venting
but
I
think
like
things
like
loops
and
conditionals
right.
These
are
more
intrinsic
to
the
pipeline.
You
know
in
terms
of
the
priority,
as
opposed
to
you
know,
an
external
trigger
mechanism,
yeah
yeah.
G
Well,
I
guess
from
an
ml
ops
point
of
view
for
sure
so,
but
yeah.
G
G
C
E
E
G
Yeah,
so
we
just
introduced
that
so
we
just
finished
introducing
what
we
call
TPS
or
Tecton
announcement
for
some
process.
So
it's
the
process
itself
has
been
approved
and
finalized
now,
so
all
the
design
that
we
had
in
flight
will
be
now
moved
into
TPS.
So
with
that
can
get
the
official
sign
on
sign
off
on
them.
F
H
E
A
point
on
that
also
David
right
so
I
mean
we
did
have
your
name
on
the
metadata
side
of
cue
flow
right,
so
yeah.
If
you
want
to
think
about
you
know
what
you
do
because
currently
you
know
that
part
of
the
project
is
is
orphan
in
the
sense
that
if
we
don't
have
someone
you
know
who
has
a
strong
interest,
the
general
perception
is
to
archive
it
right
bird.
So
we.
E
And
I
think
the
the
one
way
to
look
at
it.
A
little
late
will
be,
you
know,
can
can
this
one
inside
the
chill
flow
becomes
your
bridge.
You
know
that
you
know
there
is
to
take
it
and
I'ma
it
with
your
external
work.
The
right
so
I
mean
there
are
a
bunch
of
specs
into
that
project.
Right,
metadata,
specs,
but
also
you
know
a
dashboard
to
look
at
the
middle.
You
know
backing
database
behind
the
scenes.