►
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
So
I
guess
we
can
get
started,
looks
like
it's.
The
three
of
us.
A
Yeah,
so
you
know
for
this
week.
Unfortunately,
we
don't
have
a
demo
plan
or
anything
to
share
and
feature-wise,
because
we're
wrapping
up
the
2.1
release,
and
so
this
week
is
really
focused
on
testing.
So
you
can
set
dates
for
the
2.1
release.
A
Potentially
we'll
have
the
first
rc
as
early
as
next
week,
and
so
this
has,
you
know
all
the
features
that
we
talked
so
much
about
the
ability
for
harper
to
act
as
a
proxy
cache
p2p
integration
with
rupert
kraken
and
alibaba
dragonfly
non-blocking
garbage
collection,
as
well
as
the
work
we
did
with
thai
cloud
to
allow
to
allow
harbor
to
be
able
to
host
machine
learning
workloads.
A
So
I'm
not
sure
if
that
this
was
ever
covered
in
detail,
but
there
is
a
proposal
here
that
you
can
learn
about
another
community.
There's
a
pr
open
called
a
proposal
for
enhanced
default
processor
to
handle
user-defined
artifact
like
machine
learning,
artifacts,
and
so
basically
this
is
the
work
that
we
worked
with.
A
A
A
B
A
Yeah,
so,
basically
for
typical
machine
learning,
workloads
on
kubernetes,
there's,
there's
a
bunch
of
there's
an
inference.
A
There's
a
model
server,
that's
used
to
consume
the
different
data
models
with
with
varying
parameters,
and
it
does
some
training
and
it
gives
you
some
results,
I'm
not
an
expert
in
this
area,
but
this
was
just
some
background
that
they
had
provided
us,
I'm
just
kind
of
regurgitating
back
and
essentially
they
had
first
tried
to
package
the
server
itself,
along
with
the
images
or
along
with
the
data
models,
the
training
models
into
a
single
container
image
and
push
that
to
you
know,
docker
registry,
but
the
way
it
works
is
the
model.
A
Server
itself
is
quite
huge.
It's
like
a
couple
gigabytes
in
size
and
the
way
it
interacts,
I
think,
typically,
how
it
works
is
the
the
models
are
maintained
by
the
algorithm
team,
but
the
model
inference
server
is
maintained
by
a
separate
team,
so
the
cost
of
deploying
it
in
such
a
way
as
a
container
as
a
single
container
image
is
quite
high
and
it
doesn't
really
make
sense.
A
So
they've
sort
of
looked
into
just
deploying
the
models
on
the
data
models
themselves
as
control
images,
and
there
really
wasn't
a
content
management
system
or
a
registry
type
of
product
geared
towards
machine
learning.
Data
models
on
the
market.
So
thai
cloud
was
a
company
that
had
worked
with
harvard
pretty
early
on.
They
actually
have
a
productized
version
of
harvard
and
then
they've
done
their
own
sku
running
on
their
on
their
sas
and
so
and
they're.
Also
one
of
the
earlier
practitioners,
or
one.
B
A
Founders
of
kubeflow,
actually,
which
is
a
project
for
an
ai
project,
ai
machine
learning,
deep
learning,
project
on
kubernetes,
and
so
you
know
they
they're
privy
to
all
the
different
aspects
of
all
the
different
advantages
of
what
harvard
offers,
as
a
registry
type
of
product
to
maintain.
A
Artifacts
such
as
you
know,
policy
management
and
role-based
access
control
and
like
tag
tag
retention,
something
they
can
leverage
as
well,
and
so
they
reached
out
to
us-
and
we
worked
together
over
the
course
of
the
previous
release
to
come
up
with
a
way
so
that
these
machine
learning
artifacts
can
be
onboarded
pushed
to
harbor
and
managed
through
harbor.
A
A
Called
ormb
has
its
own
like
push
and
pull
commands
just
like
docker
pull
docker
docker
push
docker
pool
to
interact
with
oci
registries,
and
you
know
the
work
that
we
did
for
for
the
2.0
release
with
harvard
becoming
an
osec
compliant
registry.
That
really
came
to
fruition.
With
with
this
collaboration
and
so
now,
they're
able
to
we,
we
essentially
customized
a
configuration
file
which
is
needed
for
every
oc
image
to
append
like
a
hardware
name
space,
and
then
they
can
put
whatever
information.
A
They
want
like
metadata,
that
they
want
captured
on
the
harvard
ui
in
that
namespace
and
then
just
push
that,
along
with
the
image
onto
harbor.
So
the
proposal
is
almost
merged
and
the
work
is
we
actually
had
to
delete
the
2.1
release
because
we
had
to
work
with
icloud
to
get
the
code
in.
So
I
think
either
on
the
next
one
or
the
next
next
meeting
I'll
be
able
to
demo
what
that
looks
like
or
talk
in
more
detail
about.
You
know
what
kind
of
things
they're
hosting
what
kind
of
problems?
A
That's
that
they're
actually
solving
with
with
this
collaboration.
So
but
it's
it's
pretty
exciting
that
harvard's
being
used
for
you,
know
machine
learning,
type
of
scenarios.
A
So
and
hopefully
you
know
we'll
reach
out
to
more
more
folks
in
you
know,
machine
learning,
ai
to
see
you
know
if
this
helps
and
what
are
some
other
requirements
for
hosting
and
managing
their
artifacts.
So.
B
A
Yeah,
so
that's
all
I
had
for
today.
I
don't
know
if
jonas
had
any
other
updates
or
things
to
share
about
harper
that
online.
B
Yeah,
I
think
we
so
the
discussion
that
was
started
this
morning
by
boris
there.
The
state-of-the-art
registry
is
something
that
we've
talked
about
essentially
since
the
last
year
when
they
started
replying
in
answering
questions
and
stuff.
Like
that,
I
I
know
we
don't
have
an
answer
right
now
and
we're
kind
of
looking
to
see
what
what
will
happen
with
the
docker
registry
itself.
B
A
Yeah
I
mean
from
what
we
heard
what
we've
heard
it
doesn't
seem
very
likely.
I
think
the
I've
heard
that
they're
trying
to
donate
to
cncf
trying
to
bring
it
back,
because
there's
still
quite
a
few
companies,
that's
using
it.
You
know
open
source
projects
like
harbor.
A
I
think
we've
had
more
response
over
the
last
couple
months
compared
to
like
six
or
12
months
ago,
when
they
were
just
you
know,
going
through
their
some
personal
things,
personal
things
for
the
project
and
we
talked
to.
I
think
they
have
one
maintainer
who's
like
an
active
maintainer
that
that
we
talked
to
fairly
regularly
with
some
requirements
around
garbage
collection.
A
You
know
that
we
required
some
some
some
prs
to
be
merged,
as
well
as
the
oci
conformance
testing
that
we're
doing
or
working
pretty
closely
with
the
docker
registration
with
the
docker
distribution,
guys
as
well
in
terms
of
a
plan.
I
think
you
know
we
had
talked
about
if
they
were
to
donate
it,
and
then
we
would
be
interested
in
having
someone
become
a
maintainer
or
an
active
collaborator
of
the
project.
A
In
some
way
to
you
know
not
only
drive
the
harvard's
agenda
but
to
to
you
know,
listen
to
what
the
community
is,
what
the
community
needs
for
the
dog
distribution
and
to
sort
of
be
a
maintainer
of
the
doc
distribution
project
as
a
whole,
and
I
I
guess
you
know,
maintaining
our
own
version
of
distribution
is
always
a
possibility.
I
think
it's
more
costly,
but
they're.
You
know
because
they
they
release
so
infrequently
we
did
have
to.
A
It,
did
force
us
to
sort
of
deep
dive
the
code
and
understand
the
doc
distribution
project
a
lot
better,
so
we're
pretty
comfortable
with
you
know
what
it
can
do
and
what
we
need
to
be
opened
up.
If
you
know
if
it
were
up
to
us
or
if
it
were
under
control,
so
I
think
when
the
time
comes,
and
what
we're
seeing,
certainly
paying
close
attention
to
what's
happening
with
the
doctor
speech
project.
A
B
Yeah,
I'm
just
taking
a
look
at
the
the
distribution
repo
right
now,
there's
not
a
lot
of
activity
in
the
registry
portion
of
it.
There
was.
There
are
two
commits
essentially
since
march.
A
That's
that's
true.
Yeah,
it's
been
relegated
to
the
role
of
a
storage
provider,
but
you
know
we
tried
to
decouple
it
from
the
couple
from
a
little
more,
even
or
to
move
away
from
it,
and
we
still
need
that
piece.
A
But
yeah.
We
have
a
separate
database,
that's
being
used
as
the
like
single
source
of
truth
to
track
all
the
changes
right
and
that's
the
basis
for
you
know
all
the
tagging,
maintaining
the
relationships
between
the
doctrine,
oci
indexes
or
manifest
lists,
and
the
individual
images
within
as
well
as
you
know,
figuring
out
what
image,
what
what
tags
to
delete
or
to
retain
when
running
garbage
collection,
so
we're
putting
more
and
more
emphasis
on
our
end.
A
But
distribution
is
still
it's
still
a
big
piece
in
harvard
right
now,
like
it's
still
the
data
layer
or
the
storage
layer,
and
we
obviously
want
to
work
with
the
docker
guys.
You
know.
Hopefully
we
don't
want
to
have
to
do
it
ourselves,
if
possible,
so
trying
to
engage.
C
A
Our
distribution
team
as
much
as
possible,
like
I
said
you
know,
I
haven't
chat,
to
commits
in
the
last
couple
months,
but
at
least
before,
like
kobe,
we
had
some
pretty
healthy
discussions
and
you
know
they
helped
us
with
the
planning
phase
of
garbage
collection
and
yeah.
A
B
Yeah
yeah
I
mean
there
are.
There
are
more
commits
to
the
distribution
repo
itself.
Since
march,
I
was
just
looking
at
the
registry
portion,
but
still
not
a
lot
less
than
10..
Essentially
I.
A
Can
dig
up
one
of
our
engineers
me
or
one
of
our
maintainers
meet
with
derek
who
is
the
the
sole
maintainer
of
distribution
pretty
regularly?
I
think
there
are
plans
to
release
pretty
soon.
I
can
find
out
an
update
on
the
next
meeting.
So
let
me
make
a
note
here.
B
C
A
Yeah,
I
mean
it's
definitely
it's.
You
know
something
that
we
are
aware
of,
brings
a
little
uncertainty,
but
that
there
really
wasn't
anything
better
to
use
and
there's
nothing
better
to
use
now
and
so
or
sort
of
don't
really
have
a
really
good
solution
for
other
than
to
try
to
engage
with
them
and
try
to
keep
it
alive.
Breathe
some
breathe
some
life
back
into
it.
B
Yeah
yeah,
I
think
that
I
think
that's
fine
and
yeah.
I
definitely
don't
want
us
to
create
a
new
data
layer
just
within
harvard
I
would
rather
see
if
we
can
use
something
else,
we're
already
using
something
that
works.
So,
let's
keep
doing
that
until
it
breaks.
A
Right
and
I've
heard
that
justin
cormack,
the
I'm
not
sure
well
he's,
I'm
not
sure
if
he's
he's
involved
with
the
distribution
project
itself,
but
you
know
he's
on
the
tnc
ncftob
and
fairly
involved
with
the
docker
side,
but
I
think
we've
heard
that
he
wanted
to
possibly
donate
it
to
cncf.
A
You
know
and
reached
out
to
us
to
see
if
neil
was
interested
in
becoming
a
maintainer
to
help
keep
the
project
alive
and
to
drive
the
project
because
we're
so
in
tune
with.
What's
going
on
with
the
project
itself,
we
leveraged
it
so
much
so,
and
that's
definitely
something
more
interesting
if
there
were
ever
an
option
on
the
table.
So.
C
B
Tweeting
that
out,
but
other
than
that,
nothing
really
right.
Now:
okay,.