►
From YouTube: OCI Weekly Discussion - 2021-05-26
Description
Recording of the weekly OCI developer's call from 26 May 2021; agenda/notes here: https://hackmd.io/El8Dd2xrTlCaCG59ns5cwg?view#May-26-2021
B
B
I
basically
went
around
to
a
bunch
of
people
at
the
lab,
and
I
talked
to
a
bunch
of
people
that
I
know
that
are
outside
the
lab
on
slacks
and
I
kind
of
collected
a
lot
of
feedback
of
like
what
would
what
would
you
want
for,
like
you
know,
for
containers,
if
you
could
like
do
anything,
and
I
took
a
lot
of-
I
mean
I
came
up
with
a
lot
of
sort
of
like
ideas,
and
I
don't
want
to
discuss.
B
I
kind
of
want
to
propose
that,
like
there
could
be
an
hpc
working
group,
because
it
makes
sense
that
this
this
huge
community
has
kind
of
some
representation
in
oci,
and
it's
not
to
say
that,
like
anything
that
we've
this
group
would
come
up
with
would
be
accepted,
but
I
you
know,
I
think
it's
important
to
be
part
of
the
conversation
and
maybe
there'd
be
like
a
new
spec
or
project
that
came
up
with
so
but
yeah.
I
don't
have
a
like
a
formal
presentation
or
anything
I'm
just
here.
A
No,
that
sounds
good.
I
I
guess
that
I
was
queuing
up
this
figure
time
allocation,
but
I
hope
that
wasn't
the
full
thing
I
mean
I
mean
there
has
been
like
the
singularity
work
has
been
done.
I
haven't
followed
up
with
them
recently,
but
way
to
go
mike.
B
But
I
wanted
to
kind
of
get
people's
feedback
about
if
this
is
something
that
seems
like
it
would
be
reasonable
and
useful
to
do.
And
if
that's
the
case,
then
I
can
go
down
that
you
know.
I
read
the
tob
pr
99
or
whatever
it
was
that
talks
about
the
process
for
submitting
a
working
group,
and
I
can
pursue
that
and
I'll
go
out
and
I'll
recruit
a
lot
of
people
to
sort
of
to
join
it
and
go
from
there.
E
This
is
shane
cannon,
I'm
from
berkeley
lab
and
I've
been
on
the
oci
mailing
list
for
a
long
time,
but
I
have
not
managed
to
make
it
to
any
calls.
But
when
I
saw
vanessa's
proposal
I
wanted
to
make
sure
I
joined
this
one.
I
think
that
there
would
be
a
lot
of
it.
There
already
is
a
lot
of
interest
from
the
hbc
community,
so
I
think
we
could
get
a
good
bit
of
participation
and
there's
a
lot
of
synergistic
groups
already
out
there.
E
B
Yeah
shane,
I'm
glad
you
came,
you
know
I
didn't.
I
don't
think
I've
ever
seen
you
I've
interacted
with
you
for
years,
but
so
I'll
just
say
it's
it's
nice
to
see
you
and
nice
to
finally
meet
you
sort
of
virtually.
E
I
think
we
met
one
time
at
a
thing
at
boulder,
but
it
was
like
four
years
ago.
Oh.
F
Yeah
vanessa,
this
is
brandon
calling
with
the
sandia
national
labs.
I'd,
definitely
be
helping
or
wanting
to
help
you
make
that
happen
for
us
from
the
oci
side.
F
Obviously,
like
steve,
you
said
with
the
singularity
you
know
for
us
in
sandia,
I've
kind
of
been
our
main
champion
of
that
since
2014-15
with
my
colleague
john
miner,
for
for
the
efforts
that
we've
been
doing
with
cloud
and
containers
at
sandia,
and
so
even
like
2015,
I
was
the
first
one
to
put
benchmark
our
one
of
our
supercomputer
clusters
against
a
hpc
virtualization
cluster.
F
I
put
up
in
aws
with
a
cfn
cluster
at
the
time,
and
then
you
know
sg
and
slurm,
but
then
I
containerized
all
that
too,
and
it
was
still
you
know
that
was
so
foreign
now
it's
coming
back.
F
I
I
spent
a
time
at
amazon
as
well
aws,
but
coming
back,
it's
it's
interesting
to
see
where
the
community
is
and
isn't
still
for,
the
labs
as
a
whole,
and
so
I
think
you
know
all
of
us
having
having
the
ability
to
help
the
open
community
like
we
love
doing
from
the
laboratory
side
would
be
really
good.
H
D
I'm
trying
to
think
of
where
that
might
actually
live,
and
it
there
isn't
really
directly
within
cncf,
for
that
maybe
app
delivery
would
probably
be
one
of
the
more
direct
tags
for
that.
G
Yeah,
I
was
just
I
mean
what
it
seems
like
to
me
like
for
some
of
this
working
group
proposals.
It
sounds
like
there's
kind
of
an
overarching
kind
of
what
would
be
in
kind
of
a
sig
today
to
kind
of
discuss
the
needs
of
this
particular
community
and
then
kind
of
spin
out
some
of
these
ideas,
like
a
lot
of
the
the
ideas
that
vanessa
had
mentioned
here,
like
each
one
of
them
itself
could
almost
be
a
working
group.
So
the
working
groups,
as
they're
defined.
G
Kind
of
tightly
scoped
threads,
whereas
it's
not
really
designed
to
like
keep
a
community
of
interested
parties
together.
B
B
B
We're
we're
doing
similar
things
so
like
when
you
say
like
work
is
so
if
you're
running
something
in
the
cloud
you're,
probably
using
kubernetes,
we
also
might
have
a
container
cluster
at
an
hpc
center,
but
typically
we're
using
like
a
job
manager,
but
we're
also
a
batch
scheduler,
but
we're
also
using
containers
with
the
batch
scheduler.
B
So
I
guess
how
why
is
it
so
hard
to
work
together?
I
I
feel,
like
I
feel
like
there
need
to
be
people
in
the
room
from
both
sides
when
different,
specs
and
oci
are
being
discussed
to
be
like.
Okay
is
this:
is
this
driving
with
my
use
case,
and
you
know,
and
what
do
I
have
in
common
with
this?
It's
something
that
I
just.
I
think
we
need
to
do.
I
don't
as
far
as
I'm
aware
that
hasn't
really
been.
E
Done
yeah,
I
think
I
think
I
would
agree
that
there's
just
a
difference
in
some
of
the
priorities
between
the
two
communities,
and
it
doesn't
mean
that
we
can't
you
know
collaborate
or
work
together.
It
just
means
that
you
know
that
maybe
pushes
it
down
the
the
priority
stack
at
times.
A
I
think
what
I'm
hearing
is
there's
there
looks
like
there
might
be
some
hpc
groups
and
maybe
in
cncf-
and
I
think
the
question
is
what
what
would
we
do
here
to
help
facilitate?
Is
it
like?
How
do
you
store
hpc
stuff?
You
know
how
do
you
run
hpc
stuff
like
what
was
the?
A
B
I
mean
I
think
it
could
be
a
way
to
foster
collaboration,
but
I
think
when
you
kind
of
think
of
details,
so,
for
example,
let's
say
in
the
hpc
use
case-
there's
we
want,
I
don't
know
metadata
about.
We
want
more
metadata
about
layers,
something
about
the
layer.
Then
we
would
go
and
we
would
see
if
there's
an
ability
to
to
maybe
add
that
to
the
current
spec.
B
If
there's,
if
a
bunch
of
hpc
people
come
together
and
they're
like
you
know,
we
really
need
some
kind
of
orchestrator
for
fetching
containers
and
a
batch
scheduler.
Well,
that's
like
a
totally
different
project,
but
it
should
be
developed
not
just
like
a
you
know
like
a
fun
project,
but
it
should
be
made
like
a
standard,
so
it
doesn't
just
work
with
you
know:
slurm
and
sun
grid
engine,
but
also
like
flux.
B
Like
the
same,
the
question
is
like:
do
we
want
to
combat
this
from
like
a
let's
make
a
project
together
or
let's
try
to
better
make
a
standard
for
the
entire
community
in
the
same
way
that
the
other
specs
are
already
standards?
So
that's
how
I
sort
of
distinguish
between
the
two,
like
something
being
an
oci
versus
like
some
other
group
that
works
on
container
or
cloudy
or
hpc
things.
C
I
So
I
just
want
to
say
that
metadata
is
something
that
I'm
working
on
as
well,
but
I
but
I
hear
what
you're
saying
vanessa,
because
I
have
to
be
in
several
different
communities
in
order
to
align
with
applying
metadata
to
the
container
use
case,
and
perhaps
perhaps
that's
something
you
want
to
be
clued
into,
but
it
seems,
but
I
see
like
yeah,
it
seems
to
me,
like
you
know,
is
if
someone
has
a
particular
use
case.
I
B
Yeah,
I
don't
think
the
owner
should
be
on
like
one
person,
I
guess
that
if
there
was
an
hpc
working
group,
that
would
be
a
venue
that
all
these
people
from
you
know
the
national
labs.
Academia
that
are
vested
in
this
use
case
would
come
together
and
think
about
the
current
specs
and
specs
that
don't
exist
in
context
of
that
use
case,
and
you
know
worst
case
scenario,
we
accomplish
nothing,
we
present
ideas
and
for
new
projects
or
additional
specs
and
oci
says
nope.
That
is
like
out
of
scope
of
what
we're
going
to
doing.
B
B
That,
I
think,
is
a
an
accomplishment
and
at
the
best
case
we
like
make
some
kind
of
spec
that
may
be
specific
to
hpc,
but
maybe
some
of
these
hpc-ish
things
that
are
being
run
in
industry
on
the
cloud
would
also
benefit
too.
So
I
guess
I
don't
see
that
there's
really
any
drawback
to
like
trying
it
out.
B
I
I
feel
like
I'm,
you
know
having
deja
vu
with
this
conversation,
because
I
think
I
brought
it
up
with
regards
to
build
as
well
like
software
build,
and
the
response
was
the
same
thing.
There's
oci
itself
doesn't
have
a
working
group
structure.
Cncf
doesn't
really
have
any
kind
of
special
interest
whatever.
Where
does
build
fit
in
yeah.
C
C
G
Yeah
from
my
perspective,
cncf
is
better
at
fostering
kind
of
these.
These
communities,
like
mentioned,
like
the
I
guess,
they're
called
tags
now
not
sigs,
but
basically
groups
of
individuals
who
have
a
very
common
interest
and
from
the
working
group
proposal
it's
defined
to
be
tightly
scoped
and
you
would
picture
like
if
you
had
one
of
these
tags
or
sigs
or
whatever
they're
called
producing
kind
of
ideas
and
brainstorming
and
then
coming
to
cncf
and
creating
working
groups.
G
Basically,
you
could
use
that
to
sponsor
these
individual
work
group
items,
but
at
least
how
the
working
groups
have
been
proposed
today,
they're
they're
intended
to
be
short-lived
in
lifespan,
whereas
I
don't
think
that,
like
from
a
community
perspective,
you
just
don't
put
like
a
lifespan
on
the
community,
you
just
want
to
potentially
produce
multiple
different
working
groups
out
of
that
or
multiple
different
changes,
so
that
that's
that's
kind
of
from
what
I
gather.
What
vanessa
was
really
saying
like
having
a
working
group
that
creates
working
groups?
If
I
understood
that
correctly,.
B
Yeah,
I
guess,
reading
the
the
proposal.
I
didn't
really
read
it
as
like.
This
is
just
a
quick.
I
mean
it
could
be
a
temporary
thing,
but
I
don't
see
why
a
working
group
like
if
there's
going
to
be
a
consistent
working
group
for
something
for
a
specific
project
in
progress.
I
don't
see
why
there
couldn't
be
a
working
group
just
to
kind
of
kick
start
a
couple
of
like
projects
for
hpc,
because
we
need
we
need
that
structure
and
I
don't
agree
that
this
kind
of
thing
fits
with.
B
Cncf
cncf
is
cloud
native
and
you
know
well
that
does
include
containers,
but
you
know:
hpc
is
not
really
the
same
thing
as
as
cloud
native
at
least
I've.
There's
probably
there
are.
There
are
similarities,
but
I
wouldn't
call
hpc
cloud
native
and
I
think
it
fits
better
with
us
with
oci,
because
oci
and
I
suppose,
if
it's
just
an
issue
of
like
the
technicality,
that
the
working
group
needs
to
be
like
a
specific
scope
project.
B
Well
then,
we'll
find
some
other
assembly
method
and
instead
of
and
we
would
then
propose
actual
projects
for
oci,
but
to
me
that
feels
like
it's
not
a
very
inviting
thing
like
there's
a
bunch
of
people
from
the
hpc
community
that
want
to
be
involved
in
this
and
to
tell
them
nope,
you
can't
make
a
working
group,
but
you
have
to
kind
of
come
to
us
only
with
specific
projects
that
feels
kind
of
it
feels
very
exclusionary,
especially
because,
like
it
doesn't,
doesn't
seem
like
there's
any
risk
to
giving
people
a
chance.
C
Yeah,
sorry,
no,
no,
it
wasn't
a
no.
It
was
just
pointing
out
that
the
the
people
that
you
need
to
be
involved
also
include,
probably
kubernetes.
If
you
want
to
do
scheduler
plug-ins
for
your
stuff
but
yeah.
If
you're
only
talking
about
the
features
that
are
enabled
by
the
standard
right,
the
standard
specs
that
we
own
in
oci
or
some
minor,
you
know
extension
from
that,
then
that's
fine,
that's
fair!
B
C
B
G
It
comes
down
to
just
like
technicalities
and
structure
like
I
don't
I
don't
think
it's
like.
I
think
we
should
figure
out
a
way
to
do
that
and
that's
kind
of
how
the
working
groups
came
out
in
the
first
place.
Is
we
wanted
to
do
something
and
we
didn't
have
a
structure
for
it,
so
we
created
structure
so
like
if
we
want
to
have
the
structure
for
it
like
we
can
create
the
structure
for
it.
It's
not
a
matter
of
whether
we
should
or
shouldn't
do
it
like.
G
It
said,
like
I
think,
cncf
has
more
of
a
structure
around
community
than
than
oci
has
traditionally
and
that's
been
kind
of
one
of
the
problems
we've
had
as
as
the
oci
community
has
grown.
G
So
I
I
don't
think
that's
that's
unreasonable
but
like
if,
if
the,
if
the
point
of
it
in
the
first
place,
is
that
you
need
structure
around
it
and
that
structure
needs
to
be
defined,
then
we
need
to
make
sure
that
it
fits.
I
guess
is
the
point
like
there's
nothing
stopping
you
today
from
creating
a
community
around
it
and
just
just
doing
it
but
yeah
it's
it's
about
having
that
structure,
I
think.
H
B
H
A
And
we've
had
like
kind
of
a
wide
range
of
here's,
a
specific
thing
we
want
to
add,
because
x,
y
and
z,
so
there's
some
context
added
to
this,
a
very
specific
addition
to
the
spec
or
here's
a
wide
ranging
scenarios
and
something
we'd
like
to
do-
or
here
is
some
new
new
enhancements
to
existing
stuff.
So
all
of
that
is,
you
know,
kind
of
the
range
of
stuff
that's
come
up.
Sometimes
it
could
be
as
simple
as
a
routing
to
hey
here's,
a
group
that's
already
working
on
it.
A
A
Did
you
need
mark?
Do
you
need
a
lot
of
coverage
from
a
lot
of
people
in
marketing
support
like
what
kind
of
project
was
it?
Is
it
a
project?
Lots
of
people
are
going
to
contribute
and
you
want
to
be
on
the
the
landscape
and
so
forth
versus
there's
a
couple
of
people
working
on
it
that
you
know
a
dozen
more
implement
it
and
then
the
whole
world
consumes
it.
A
H
Yeah,
I'm
thinking
steve
you're,
probably
better
to
talk
about
this
than
most
of
us
when
we
are
taking
distribution
from
docker
and
moving
that
over
into
cncf.
Was
there
a
thought
of
whether
it
belonged
in
oci
at
that
point,
because
it's
very
similar
to
a
lot
of
stuff
we're
doing
over
here.
But
it's
not
a
spec,
it's
an
implementation.
So
I'm
just
wondering
if
that
was
kind
of
the
dividing
line.
A
One
of
them
there's
been
a
lot
of,
I
mean
others
should
speak
to
it.
Also
that
oci
was
more
around
specs
and
cncf
more
about
reference
implementations,
but
then
we
have.
We
have
some
implementations
in
oci
and
whether
that's
something
people
would
redo
again
that
you
know
but
yeah.
I
think
I
think.
G
History
there
he
has
a
little
longer
history
there
it
started,
I
think,
either
before
around
the
same
time
as
cncf
I
mean
I
think,
if
oci
was
created
today,
it
probably
would
have
been
created
as
like
a
something
sub
of
cncf,
specifically
around
like
specifications
but
yeah
at
the
time
it
was
specification
and
reference
implementation,
but
over
time
most
implementations
have
gone
into
cncf
because
there's
not
just
one
implementation
of
things
like
at
the
time.
G
B
Yeah,
it
could
be
that
it's
nicely
split.
It
could
be
that,
like
an
actual
kind
of
spec
is
designed
here,
but
then
for
an
actual
implementation,
then
we
go
to
cncf
to
actually
implement
it,
so
we
don't
kind
of
create
something
that,
as
you
just
said
and
then
like
bless
it
like
this
is
the
implementation.
F
My
first
oci
meeting
was
dockercon
2016
in
the
upper
room
when
everybody
met
and
got
together
and
then
tagged
along
as
I
could
and
then
life
personally
yeah
personal
life
got
in
the
way
of
some
things
and
being
able
to
get
back
in
this
past
few
months
have
been
really
fun
with
everybody,
but
I
take
it
vanessa
the
same
way
where
you
know
we're
kind
of
saying
oci
we
want
to
make
sure
interfaces
are
standardized,
and
so
can
we
have
a
standardized
interface
for
slurm
for
usgs
that
they
can
go
back,
work
with
and
that's
compliant
and
then,
as
a
working
group
we
can
be
out
of.
F
B
C
That
right,
when
you
talk,
you
start
talking
about
adding
another
specification,
then
we've
got
to
get.
You
know
some
permission
and
approval
for
the
new
spec
oci
has
a
limited
balance.
You
know
a
little
imbalance
to
what
we're
trying
to
work
on
and
that's
why,
when
you
mention
kubernetes,
I
was
like
well
yeah,
there's
already
some
work
going
on
on
adding
new
plugins
for
that
scheduler.
I
just
happen
to
know.
You
know
a
few
of
the
features
and
caps
that
are
being
worked
on.
C
B
C
H
A
Mean
I
think,
just
yeah.
The
suggestion
would
get
a
group
of
people
together
that
want
to
collaborate
on
it
and
work
on
some
proposals,
and
if
you
need
some
some
structure
and
cover
that's
you
know
that's
what
amy
and
chris
and
others
are.
You
know
we're
trying
to
set
up.
You
know
because
lf
kind
of
spans
both-
and
I
think
you
figure
out
what
it
is
that
you
want
to
accomplish,
get
the
people
and
then
what
are
the
gaps
in
that
that
pr
99
kind
of
starts
to
outline?
A
How
do
you
define
the
group
in
a
way
and
obviously
that's
a
pr
there's
a
discussion
still
on
it?
I
would
focus
on
what
you
want
to
achieve
first
and
then
figure
out
how
to
work
where
to
fit
the
structure.
B
B
Actually
already
have
a
small
list,
I'm
going
to
find
more,
but
I
you
know
I've
already
started
doing
that.
So
would
you
recommend
putting
together
basically
that
proposal
and
then
maybe
in
the
proposal
we'll
just
kind
of
put
down,
I
can
kind
of
reiterate
some
of
these
ideas
and
projects
and
then
on
the
pr
we
can
kind
of
discuss
like
what
that
looks
like
for
a
working
group.
A
If
that's,
what
helps
you
facilitate
the
team
to
come
together,
like
I
put
one
together
for
the
reference
type
stuff
as
when
this
whole
thing
was
coming
up
with
the
work
groups
just
to
try
to
to
test
it
out,
and
it
helped
me
actually
provide
some
clarity
and
some
scope
and
some
things.
So
if
that
helps,
you
get
your
group
together,
because
obviously
the
purpose
of
a
working
group
is
you
have
a
bunch
of
people
with
varied
opinions
and
you're
trying
to
find
some
solidity
solidity
around
a
focus.
A
B
E
I
wonder
vanessa
if
it
would
work.
You
know
if
cncf
is
a
better
place
to
create
a
special
interest
group,
we
could
always
create
sort
of
the
sig
hpc
component
and
we
could
use
that
to
say:
hey,
let's
discuss
amongst
ourselves.
What
are
things
we
want
to
prioritize
for
the
oci
community
bubble,
those
up
and
then
those
might
become
more
concrete
kind
of
working
group
proposals
or
something
like
that.
E
I
still
think
it
would
be
useful
for
us
to
be
able
to
come
in
and
kind
of
give
some
overview
presentations
just
to
help.
You
understand
design
choices,
we've
made
in
our
environments,
and
you
know
why
that
happened,
and
you
know
it
might
help
you
think
through.
Oh
this
is
you
know.
This
applies
to
other
communities
besides
hpc,
but
that
sounds
great.
B
A
I
mean
think
you
have
people
here
that
are
obviously
passionate
about
this
space
and
we
work
on
various
pieces
of
it
and
you
know
sometimes
seeing
somebody
do
a
presentation
spawns
an
idea
that
oh,
what
if
we
did
this
that
could
make
up
something
simpler,
that
you
might
not
have
been
thinking
about,
or
it
might
be
something
that
hey.
If
you
just
add
this,
look
at
how
powerful
what
you're
doing
can
be.
I
mean
like
this
is
kind
of
how
we
did
some
of
the
stuff
back
in
the
artifact
stuff
was
like.
A
We
were
working
on
how
to
implement
helm
in
registries,
and
we
saw
a
pattern.
We
were
kind
of
experimenting
with
some
ideas,
and
then
we
came
to
this
group
and
said
hey
what
if
we
did
this
and
then
that
conversation
iterated
a
lot
and
if
we
had
working
groups,
then
we
probably
would
have
formed
a
working
group
then,
but
the
idea
was
it
was
a
specific
thing.
The
working
group.
I
think
that
what
we're
what
we're
proposing
here
is.
A
It
is
a
more
actionable
thing
that
has
a
results
in
a
spec
or
some
kind
of
output.
Of
a
project,
whereas
I
think
what
the
conversation
was
was
if
it's
a
long-term
tag
as
opposed
to
special
interest
group,
I
don't
know
another
difference,
but
the
that
you
know
you
could
do
that
in
cncf
and
then
say:
hey
in
cncf,
we've
got
a
bunch
of
people
hey.
If
we
only
did
this
one
thing
in
run
c
or
distribution,
look
what
we
can
get
hey!
A
Okay,
there's
a
working
group
to
make
that
ask,
or
if
there's
a
larger
spec,
that
you
want
to
drive.
That's
a
new
thing
that
would
be
the
up.
That
could
be
the
output
of
that
group
that
comes
over
here
and
says:
hey.
We
want
a
working
group
to
do
this
thing.
Here's
a
special
interest
group
over
and
say
tax,
the
cncf
that's
been
forming
it.
B
E
That
I
think
that
could
work
as
well
and
we've
got
you
know,
there's
a
cj's,
you
know
community.
He
has
there's
a
couple
of
others
that
you
know
they're,
certainly
the
ones
that
through
exascale
ecp,
but
I
think
we'd
want
it
broader
than
that,
but
we
could
probably
build
out
from
from
there
if
we
needed
to
it's
probably
worth
looking
at
cncf
and
in
the
it's,
the
research
computing
or
something
like
that
and
see.
G
Yeah,
I
think
if
you
get
together
and
you
have
a
set
of
some
of
the
goals
you
want
to
accomplish,
I
think
it'd
be
easier
too,
because
because
maybe
we
could
just
immediately
create
some
of
the
working
groups
and
and
go
from
there
as
well.
If
there's
some
kind
of
high
priority
items
that
you
already
agree
on,
that
that
they
need
to
get
in
but
yeah
I
just
I
just.
I
just
want
to
make
sure
it's
like
from
a
if
you
want
to
maintain
like
a
long-term
community.
G
I
I
think
cncf
has
a
better
structure
for
that
right
now.
I
think
I
still
kind
of
want
to
hear
amy's
opinion
on
on
all
of
that
or
because
she
knows
the
rules
and
structures
better
than
anyone,
but
I
think
that's
right.
D
All
right
I'll
step
back-
and
I
was
letting
everybody
just
kind
of
like
run
around
in
this
so
in
large
part
like
this-
could
absolutely
work
within
cncf.
We
do
have
an
older
working
groups
kind
of
thing
in
here,
but
I'm
much
more
likely
to
be
able
to
see
this
as
like
a
tag
that
focuses
around
how
this
could
connect
into
am
kind
of
academic
working
group.
This
might
also
separately
be
part
of
that
research
group
that
I
think
it
was
brandon
brandon.
D
I
think
it
was
brandon
yeah
brandon
mitchell's
talking
about
as
well,
but
also
none
of
this
has
to
be
solved
like
today.
Just
getting
the
idea
of
hey,
we
want
to
be
able
to
like
do.
This
is
good
enough
and
derek.
I
don't
know
if
I've
completely
cut
you
off.
B
E
You
know,
probably
not
in
the
next
week
or
two,
but
I
think
we
could
try
to
get
our
you
know
we
could
convene.
We
could
kind
of
think
what
points
we
think
would
be
most
important
to
bring
up
and
yeah.
We
I'd
be
happy
to
to
help
co-present
something
and
present
it.
You
know.
I
certainly
know
of
a
few
things
that
come
to
mind
that
I'd
want
to
just
let
others
know
about
so
they're
aware
of
it.
B
Dad
for
the
scripture,
oh
sorry,
and
for
the
pr
99,
I'm
wondering
so
hold
on
one
second,
I'm
looking
at
it.
D
B
Yeah,
I
think
there
should
be
something
stated
that,
like
this
is
for
a
like.
Very
this
is
not
for,
like
a
community
like
a
working
group
is
not
really
a
community.
A
working
group
is
like
a
group
of
people
that
want
to
get
a
specific
thing
done,
because
I
that
went
right
over
my
head
when
I
was
reading
about
this
working
group
thing.
G
I
think
the
goal
is
to
keep
everything
pretty
tightly
scoped
and
defined
initially
until
we
need
more.
If,
if
there's
a
lot
of
overlap
with
kind
of
tags
in
cncf
already,
then
it
it
doesn't
make
too
much
sense,
because
oci
is
already
kind
of
like
a
tag.
It
doesn't
really
exist
in
the
same
like
it's
it's
more
of
like
a
sibling
organization
to
cncf,
but
in
actuality
it's
it's!
G
It's
more
like
more
like
how
the
tags
operate.
It's
it's
smaller,
we're
very
tightly
scoped
in
terms
of
like
what
we're
trying
to
do
and
I
think
we're
trying
to
prevent
oci
from
becoming
too
largely
scoped
in
terms
of
like
and
in
specific
dimensions
like
in
terms
of
like
the
specifications
like
we
want
to
kind
of
grow
out
from
what
we
have
don't
want
to
just
like
gobble
up
a
lot,
that's
kind
of
what
cncf
has
done,
and
you
know
that
yeah.
B
And
that
totally
makes
sense.
My
suggestion
is
just
like
in
that
document.
If
someone
comes
that
is
like
marginally
familiar
with
oci
and
they're
reading
it,
it
should
just
be
very
specific
about
that
aspect
of
the
working
group.
If
that
makes
sense,
and
also
so,
okay
sega's
special
interest
group,
you
keep
saying
tag.
What
does
tag
mean.
H
D
Yes,
happy
to
help
so
once
upon
a
time
there
were
such
things
known
as
the
special
interest
groups,
we
had
both
special
interest
groups
in
kubernetes
and
in
cncf
people
got
very
confused
across
the
land,
about
which
thing
what
belongs
where
and
so
as
of
cubecon
in
early
may,
we
are
now
tags
for
cncf.
It's
the
technical
advisory
group.
B
L
G
D
Correct
over
there
they
have
a
toc
and
yes,
their
power
flows
from
the
toc
proper.
The
toc
is
the
one
that
like
says,
we've
convened
this
particular
tag
to
be
able
to
focus
on
whatever
area
it
is.
There
are
co-chairs
that
run
those
particular
groups,
and
then
there
are
tech
leads
that
kind
of
work
around
in
there
they're
intended
to
be
kind
of
longer
term
working,
just
like
bodies
they're
not
really
set
up
to
be
like
the.
D
G
The
purpose
of
them,
though,
isn't
necessarily
like
because
I
I
know
with
like
the
tag
runtime
the
one
I'm
more
involved
in
it's
a
lot
of
reviewing
other
projects
and
seeing
which
ones
are
recommended
to
the
toc.
But
is
there
other
scope
to
it
where
you
would
say
in
this
case
like
convene,
but
you
may
review
projects
that
you
suggested
toc,
but
you
may
also
suggest
working
groups
to
oci.
D
See
and
that
gets
complicated
because
part
of
what's
going
on
with,
like
the
the
tag
structure
that
is
totally
different
from
where
we'd
be
over
in
oci,
is
like.
They
have
projects
that
are
actively
working
on
trying
to
graduate.
We
don't
tend
to
be
able
to
have
any
structure
here
around
like
the
certain
levels
and
things
that
you
get.
D
So
it's
not
really
the
same
kind
of
attraction
in
the
the
same
kind
of
voice
and
part
of
what
the
tags
were
originally
charged
with,
was
being
able
to
come
in
and
review
projects
and
and
then
advise
the
toc
on
fitness.
For
does
this
fit
in
cloud
native
and
if
it.
D
C
You
know
associated
with
these
tags
for
fixing
holes
across
cncf,
and
you
know
brought
various
projects
that
are
seen
as
high
level
and
there's
some
funding
that
comes
with
it
and
that
sort
of
stuff.
So
and
you
get
communication,
so
it's
sort
of
it
would
it
would
it
may
help
in
this
case,
but
it
may
not
if
you,
if
you
guys
don't
want
to
work
with
that,
you
don't
want
to
create
a
new
working
group.
You
know
associated
with
sick
run
time
and
whatnot,
then
that's
fine.
That's
fair!
C
D
G
I
don't
know
if
there's
any
specific
structure
that
really
defines
this
group
like
as
we're
meeting
today
either
like
I
don't
know
how
formal
it
necessarily
needs
to
be
like
if
we
wanted
to
have
a
separate
version
of
this
group.
That's
focused
on
specific
topics,
that's
not
necessarily
a
problem.
I
mean,
I
said.
I
don't
think
we
necessarily
need,
like
a
tob
vote
on
scope
and
stuff
just
to
just
to
like
add
items
to
the
calendar
and
set
agendas.
C
C
B
Yeah,
definitely
in
the
hpc
community
I
mean
people
are
pretty
skeptical
of
containers.
I
mean
you
sort
of
have
like
two
sides.
I
guess,
even
though
containers
are
usually
used
in
most
centers.
E
F
J
Coming
back,
I
I've
been
popping
on
you
guys,
weren't
here
last
week,
at
this
time
I
I
haven't.
I
have
meetings
all
day
and
I'm
able
to
get
get
a
few
minutes
in
per
week,
so
basically
yeah
yeah,
so
just
saying
hi
I
read
through
vanessa's
proposal,
looks
looks
fine.
I
don't
know
if
you
want
me
to
go
comment
it
topic
by
topic
or
anything
like
that,
but
I
think
the
the
one
thing
to
divide.
J
I
didn't
hear
the
whole
conversation,
but
I
think
the
one
thing
to
figure
out
is
like
all
right,
like
figure
out,
what's
truly
ap
hpc
like
what's
unique
to
hpc
versus
not
eating
to
hpc
and
then
what's
specification,
changes
or
specification
requirements
versus
like
what's
a
service
requirement
right
like
like
a
build
service,
might
be
a
good
cncf
project,
for
example,
whereas
like
a
or
an
orchestrator
for
fetching
like
I
like
that
idea
like
that,
could
be
built
into
container
d
or
something
like
that
or
some
other
project
in
cncf.
J
But
maybe
there's
like
some
changes.
We
need
in
the
oci
spec
to
make
the
like
communication
between
that
orchestrator
work
or
something
I
don't
know,
I'm
kind
of
pulling
it
out
there,
oh
and
then
the
arbitrary
blob
ones.
Yeah
we've
been
doing
that
for
years.
B
J
Yeah
yeah
I
mean
there
was,
I
think,
brandon
mitchell
here
linked
to.
I
think
it
was
john
johnson's
request
where
it's
just
descriptor
level
tags,
which
I
think
is
a
good
idea,
probably
cat
pics
in
there
yeah.
So
I
mean
that
that
looks
good
and
then
what
was
the
other
one
container
metadata,
there's
the
so
so
you're
aware
of
the
the
oci
like
annotations
are
there
missing
is
the
issue
that
there's
missing
annotations
or
like
like?
J
Oh,
you
don't
need
to
answer
now,
just
like
we
can
figure
out
which
annotations
are
truly
missing
and
where
they,
which
objects,
maybe
maybe
there's
certain
objects
that
we
don't
have
annotations
for
that
might
be
good
to
kind
of
figure
out
so
and
then
the
blob
uncompressed
sizes
that
one's
tough,
like
I'm,
I'm
always
like.
I
it's
a
weird
opinion
like.
I
think
that
compressed
layers
are
kind
of
a
something
we
shouldn't
have
done,
maybe
and
just
let
like
transport
and
storage
compression
takeover,
because
that
would
be.
J
I
think
that
would
be
generally
better
because
then
you
could
calculate
on
disk
sizes.
But
if
there's
like
a
proposal,
you
have
like
an
easy
way
to
do
it
that'd
be
interesting.
I
like
the
only
problem.
There
is
the
difficulty
in
calculating
that
size
and
then,
like
you,
have
to
do
a
mish
so
like
on
one
machine
or
the
other.
You
could
have
a
very
different
disk
usage
profile
right,
depending
on
like
whether
it's
with
layers
or
like
with
butter
fs
like
do.
J
We
have
a
way
of
that's
like
oci
platform
agnostic
to
calculate
the
disk
usage,
or
do
we
have
disk
usage
per
storage
methodology?
That
was.
That
was
the
thing.
The
line
of
ticking
I
had
but
like
adding
up
the
uncompressed
layer,
sizes
and
or
reporting
or
just
turning
off
compression
and
going
to
transport
compression
are
both
kind
of
options
there.
That
might
make
that
a
little
bit
easier.
J
J
Yeah,
it's
kind
of
nice.
If
you
know
because
you
can
put
a
limit
reader
if
it
says
like
oh
I'm
10
bytes,
and
when
I'm
unpacked
I'm
20
bytes,
you
can
limit
the
unpacking
to
20
bytes
and
if
it
goes
over
because
of
some
nefarious
magic,
you
could
really
limit
that
so
yeah.
But
all
in
all,
I
think
it's
you
know.
There's
some
good
stuff
in
here.
B
Cool
thanks
and
I
agree
that
we
will
that
we're
going
to
sort
of
come
together
and
figure
out
kind
of
break
it
down
into
pieces
what
could
fit
into
a
current
spec?
What
could
be
a
new
spec?
What's
better
in
cncf
for
a
project,
I
think
I
think
shane
and
I
will
kind
of
assemble
a
team
and
figure
out
how
to
do
that.
B
And
then
I
won't
go
into
detail
about
this,
but
for
metadata
I
think,
there's
sort
of
a
larger
it's
much
larger
than
just
like
creating
labels.
I
think
I'm
kind
of
thinking
also
about
like
api
compatibility,
so
that's
something
that
I'm
working
on
right
now
at
the
lab,
we're
looking
at
binaries,
but
I
also
think
there's
a
use
case
for
containers.
So
I
don't
have
like
an
answer
at
this
point,
but
you
know:
you'd
want
to
be
able
to
take
a
container
and
then
say:
okay,
is
this
api
compatible?
B
Is
there
api
compatibility
and
the
strongest
example
from
hpc
computing
is
just
with
like
mpi
libraries,
which
you
have
to
get
it
like
exactly
right,
the
two
versions
on
the
host
and
the
container,
or
else
it
won't
work.
J
This
came
up
in
arm
with
because
there's
like
systems
that
are
like
64-bit,
but
they
have
32-bit
pointer
types
and
and
so
you're.
The
kernel
at
the
the
kernel
abby
actually
changes
in
between
the
different
container
types,
and
this
also
came
up
in
the
context
of
windows
containers
as
well.
So
it
might
be
good
to
look
at
what
was
done
there
at
the
platform
level
and
see
if
that
is
totally
like
different
from
what
you're
thinking
and
then
or
it's
just
it
might
just
be
insufficient.
J
But,
like
I
assume
you
have
some
sort
of
library,
that's
provided
by
the
host
system
that
when
you
run
the
container,
it's
expected
to
be
present
like
a
like
a
certain
abbey
version
or
a
certain
library
that
you
get.
Then
you
mount
into
the
container
and
try
to
run
your
code
against.
E
J
Yeah
so
yeah,
this
is
a
this
is
a
problem
yeah
I
I
would
start
looking
at
at
the
arm
in
windows,
containers
to
see
what
was
done
with
abby
there
and
then
kind
of
work
work
backwards.
That
makes
sense,
let's
see
kind
of
what
the
delta
is
versus.
What
you
guys
are
thinking.
F
E
B
Well,
I
mean
like
once
you
like,
so
I
I
guess
it's
easier
to
talk
about
singularity
the
way
singularity
works.
Is
it
basically
downloads
the
layers
and
then
just
explodes
them
onto
the
file
system?
And
we
don't
know
what
size
it's
going
to
be
after
it's
kind
of
exploded
like
that,
and
it
would
be
really
great
if
we
could.
E
B
E
B
Tell
you
how
many
times
where
I've
had
someone?
That's
you
know
developing
a
container
and
they're
doing
some
kind
of
multi-stage
builds.
They
want
to
get
it
small
and
they
you
know
they
they're
inside
of
it
and
they
do
du
to
see
the
size
and
they
look
at
docker
hub
and
they're
like
what
like
this
is.
This
is
this,
is
not
this
doesn't
match
up?
I
don't.
I
don't
even
know
what
that
size
means,
because
you
sort
of
assume
the
size
you
see
on
docker
hub
is
like
the
actual
size,
but
it's
it's
not.
G
J
J
Yeah
yeah,
I
mean
I
mean
what
yeah
you
end
up,
might
have
having
to
post
process
the
layers
too
right
and
like
look
at
the
tar
records
without
landing
on
disk
anyways.
Yeah
yeah
come
up
with
something
propose
it
and
then
we'll
figure
we'll
figure
out
what
what
the
best
we
know.
We
can
figure
out
a
couple
of
options
and
then
walk
it
back
from
there.
G
I
think
maybe
the
references
work
would
probably
come
up
there
because
yeah
some
of
the
signatures
and
being
able
to
attach
some
of
this
other
type
of
metadata.
This
would
be
interesting,
yeah.
E
Yeah,
that's
it
area
is
where
there's
maybe
a
lot
of
potential,
because
there's
information
we'd
like
to
start
capturing,
but
rather
than
it
being
sort
of
an
ad
hoc
or
let's
throw
a
label
in
this
way.
If
there
was
some
standard
for
it,
I
think
then
other
tool
providers
could
start
to
adopt
that
and
implement
it.
J
E
But
on
the
metadata
there's
other
types
of
metadata
that
would
be
useful
to
be
able
to
capture
for
images
that
schedulers
could
then
make
use
of
that
to
know
what
resources
they
can
put
it
on.
For
example,
like
does
this
image
require
a
gpu
or
does
it
require
an
architecture
with
this
particular,
not
just
family,
but
maybe
you
know
particular
feature
you
know
cpu
feature.
We
have
a
lot
of
cases
that
kind
of
pop
up
like
that,
and
I
don't
think
it's
unique
to
hpc,
but
it's
more.
Maybe
it's
more
strongly
felt
with
htc.
H
H
I
Nisha,
I
just
wanted
to
know
what
the
status
of
all
of
those
manifest
things
are.
I
just
want
to
say
really
quickly
that
the
the
recent
u.s
government
executive
order
is
weighing
on
me
right
now
and
I
would
love
to
be
able
to
stick
an
s-bomb
somewhere.
A
Steve,
I
think
that's
your
cue
yeah
I
I
was
curious
when
you
were
looking
for
this
one.
I
mean
obviously
there's
this
concept
of
how
to
support
reference
types
and
I've
actually
been
working
on
a
paper
and
talk
and
examples
of
the
prototypes
we've
done
around
the
reference
type
work,
because
the
other
one
that's
come
up.
There's
just
a
bunch
of
secure
supply
chain,
artifacts
that
people
want
to
be
able
to
add,
and
then
do
you
change
the
existing
artifacts.
A
A
There
are
some
reference
examples
out
there
and
of
how
this
works
and
we're
kind
of
actively
working
on
enabling
that
approach
that
you
can
add
content
to
a
registry,
and
you
don't
need
to
know
about
that
new
content.
You
can
say
give
me
all
the
references
reference
types
to
a
particular
thing
and
you
can
get
back
the
signature
you
can
get
back
an
s-bomb,
then
you
can
ask
for
the
signature
to
the
s-bomb.
A
One
of
the
more
recent
ones
that
come
up
was:
what
are
the
scan
results
like?
Can
I
go
and
ask
an
image?
Hey?
What
are
this?
What
are
the
latest
scan
results
for
this
image
and
we
would
be
able
to
support
that
as
a
reference
type
as
well.
The
big
question
there
is-
and
what's
the
schema
of
that,
but
there
is
an
example
of
an
open
ssf
group.
That's
been
working
on
that,
so
I
I
saw
your
your
thing
coming.
A
I
saw
your
thing
this
this
afternoon,
just
as
I
was
prepping
for
this
meeting,
so
I
I
had
some
demos
I
can
show
today,
but
obviously
not
left
in
one
minute,
so
we
can
talk
about
next
week
if
you
want,
or
I'm
trying
to
create
a
recording
so
that
people
can
just
watch
it
on
their
own
time.
J
There's
a
project
called
crane
that
uses
some
sort
of
well-known
tag,
location
that
you
might
have
a
look
at.
That
would
be
perfect
for
this,
because
a
lot
of
these
things
you
don't
want
to
have
to
do
when
you
actually
push
the
image
you
want
to
put
them
in
later,
so
you
don't
want
to
bundle
them
into
the
image
itself.
I
So
no,
I
wasn't
looking
at
bundling
it
into
the
image,
but
one
of
the
issues
that
well
we're
at
time
now
there
are,
I
mean
I
actually
like
the
whole
like
image
index
down.
It's
almost
like,
if
you
needed
to
move,
you
know
your
constellation
of
artifacts.
You
can,
if
you
just
stick
the
root
of
the
tree
and
all
of
its
references,
but
I
can
also
see
the
attraction
of
just
keeping
them
well.
I
Keeping
them
in
well-known
tags
is
well
and
good
if
the
tags
are
well
known,
but
I
get
the
feeling
that
we
would
be.
There
are
a
lot
of
artifacts
that
need
to
be
attached
like
that
need
to
be
tr.
You
know
that
need
to
travel
around
with
the
image,
so
well-known
tags
may
not
be
interoperable
from
like
client
to
client.
I
I
Yeah,
I
think
I
mean
I
just
wanted
to
know
like
where
people
are
at
right
now,
but
it
seems
like
it's
still
ongoing
discussion.
I
am
happy
to
push
that
next
week.
C
Good
discussion
to
have
yeah
thanks
for
providing
the
link
to
the
reference
to
s
bombs
and
being
published
by
the
white
house.
That's
very
interesting.
It's
been
an
interesting
couple
of
weeks
for.
A
A
Finally,
when
you
first
emailed
me
that
there
was
an
executive
order,
I
was
like
which
vp
is
that
debt?
Is
it
azure,
like
you
know,
somebody
else
mentioned
so
about
government
of
like
wait,
the
government
wait
what
so.
We
all
know
what
technology
looked
like
in
congress
over
the
last
couple
years,
so
yeah.