►
From YouTube: CNCF SIG Storage 2020-09-23
Description
CNCF SIG Storage 2020-09-23
A
C
C
A
I'm
gonna
give
everyone
two
more
minutes
to
join,
doesn't
look
like.
We
have
very
many
people.
A
A
A
A
The
first
one
is
the
dataset
lifecycle
framework
yeah.
I
believe
you're
you're
joining
to
talk
about
that.
The
project
went
through
a
sandbox
review
on
the
toc
yesterday
and
they
had
some
questions
about
how
it
differentiates
from
the
cozy
cap.
So
I
wanted
to
to
start
just
with
that
discussion.
If
we
could
please.
D
Okay,
so
to
that
context,
I
I
shared
the
frequently
asked
questions
I
just
compiled
it
today,
because
the
documentation
is
still
lagging
and
it's
a
bit
difficult
to
digest.
What
is
exactly
the
framework
trying
to
accomplish,
and
the
comments
were
fair
right
about,
you
know
not
being
very
clear
on
what
is
adding
to
the
ecosystem
and
how
does
it
compare
with
the
cosy
or
with
the
other
csi?
So
if
you
have
so,
I
can,
I
can
just
share
the
screen
for
the
wiki
or
you
can
just
yeah
that'd.
D
Okay,
can
you
see
my
screen?
Yes,
okay,
so
yes,
so
I'll,
I
will
just
go
through
everything,
because
it's
just
a
few
questions
right.
So
what
does
the
framework
do
exactly
right?
So
it
brings
one
new
custom
resource
definition
right.
The
data
set,
so
basically
it's
a
pointer
to
an
existing
data
source
and
the
current
implementation
is
both
s3
and
nfs,
and
so
yeah.
Is
that
just
a
crd?
No,
so,
basically,
what
we're
doing
is
that
we're
trying
to
map
not
we're
trying
to
map.
D
We
were
mapping
every
data
set
that
you
create
with
one
pvc
that
the
users
can
directly
mount
into
their
pods
right
and
the
logic
is
implemented
as
a
normal
kubernetes
operator.
Now,
what
is
the
motivation
right
for
this
work?
Why
why
we
started
looking
at
that
right?
So
when
container
storage
interface
was
introduced,
right,
more
and
more
storage
providers
were
available
on
kubernetes
environments
right.
It's
just
that.
D
From
our
perspective,
we
feel
that
the
barrier
for
non-experienced
non-power
users
of
kubernetes
right-
it's
the
buyer-
is
a
bit
high
to
leverage
the
available
csi
plugins
right
and
gain
access
to
the
remote
data
and
sources
on
their
workloads.
D
So
that's
what
we're
trying
to
enhance
we're
trying
to
enhance
the
user
experience
mostly
of
data
access
in
kubernetes,
so
we're
bringing
the
higher
level
of
an
abstraction
the
data
set,
and
we
take
care
of
all
the
necessary
work
about
invoking
the
appropriate
csi
plugin
configuring
provisioning
and
giving
in
the
end
the
pvc
for
the
end
users
to
use
right,
of
course,
we're
not
looking
to
replace
csi
right.
D
If
you
go
through
our
framework
right,
we
have.
We
have
implementations
of
csi
plugins
that
are
standalone
right,
so
you
can
take
this
part
of
the
code
and
use
it.
So
we
have
csi
s3
and
the
csi
nfs
implementations
that
are
actually
open
source
and
I
will
talk
about
what
we
did
there.
So
our
aspiration
right
is
to
be
like
a
meta
framework
right
for
csi
plugins,
and
the
comparison
that
I
like
to
make
is
like
the
same
way:
cooper
flow.
You
know
makes
machine
learning
frameworks
accessible
in
the
same
fashion.
D
We
just
want
to
make
csi
plugins
and
pvcs
accessible
to
kubernetes
environments
so
to
the
cosy
proposal.
Right
so,
of
course,
we're
not
competing
with
the
cosy
proposal
and
we're
not
currently
right
aiming
to
have
this
functionality.
D
You
know
rewritten
as
part
of
our
framework
right.
So
when
we
started
the
project
almost
a
year
ago
right,
the
only
csr
plugin
that
we
were
aware
of
was
the
csi
s3
that
I'm
pointing
in
here,
and
actually
we
have
four
and
we
maintain,
because
there
were
some
dependencies
with
the
side,
cars
and
all
this
stuff.
We
are
updating
this
as
as
is
in
our
in
our
repo
right
so
now
in
the
future.
Right
when
cause
the
interface
becomes
part
of
kubernetes,
of
course,
we'll
stop.
D
Maintaining
our
forked
version
of
the
csi
s3
right
and
directly
invoke
cosy
for
creating
a
pvc
for
buckets
in
cloud
object.
Store.
Cozy
also
aims
to
manage
the
full
lifecycle
of
a
bucket
provisioning
configuring
access,
which
we
don't,
which
is
actually
beyond
our
scope
right
and
the
buckets
and
the
s3
is
just
part
of
what
we
want
to
support
as
type
of
data
sets
right
and
an
additional
some
additional
benefits
that
are
on
the
roadmap,
and
we
have
some
initial
implementations.
D
We
feel
that
it's
after
you
introduce
that
concept
of
a
data
set
as
a
higher
level
abstraction.
Then
you
can
build
also
high
level
orchestration
right,
so
I
think
we
can
achieve
improvements
in
terms
of
performance,
so
we
have,
you
can
have
we
try
to
present
a
plugable
casting
interface
and
we
have
an
example
implementation
of
how
this
would
work,
so
it
will
be
completely
transparent
to
the
user
and
they
can
provision
cases
depending
on
the
type
of
data
sets
without
the
user
explicitly.
D
You
know
specifying
cars
or
configuring
the
cast
on
their
own
and
also
we
feel
there
might
be
interest
on
the
security
aspect,
because
imagine
that
we
could
have
a
common
layer
of
access
management
for
credentials
of
the
different
type
of
data
sources.
Whether
you
have
you
know
your
normal
stream
credentials
like
secret
access,
key
and
the
access
key
api
and
the
same
fashion,
you
can
have
username
password
all
part
of
the
same,
a
access
management
layer
and
we
believe
that
we
there
are
there
are.
There
is
some
interest.
D
So
I
would
like
to
point
out
and
give
a
shout
out
to
the
people
who
have
embraced
the
framework.
Even
in
this
very
early
stage.
The
european
bioinformatics
institute
and
david
specifically
are
running
a
boc
with
dlf
and
cubaflow
under
cloud
infrastructure,
so
basically
they're
using
pipelines
and
s3
and
the
1000
genomes
data
are
on
s3
packets
and
they're,
rewriting
their
pipelines
to
with
the
data
set
convention,
because
the
s3
credentials
are
before
the
left
were
repeated,
and
you
know
plug
they're
everywhere,
like
environmental
variables.
D
Instead
with
a
frame
with
a
frame,
they
feel
it's
more
convenient
for
the
user
to
digest
and
write
their
pipelines.
D
The
reason
there
is
from
the
open
data
hub
and
you
can
see
a
relevant
issue-
that
there
is
interested
in
integrating
dlf
directly
into
open
data
hub
and,
of
course,
there's
pathetic
proposal,
which
is
actually
very
close
to
the
data
dataset
specifications
that
we
are
supporting
in
our
code
and
if
you
look
in
their
in
their
code,
dlf
is
actually
forked
and
it's
being
under
evaluation.
Whether
you
know
it
can
support
the
implementation
right.
D
So
we
make
a
post
right
now
and
you
know
take
up
any
questions
or
comments
that
you
might
have.
A
I
just
had
one
question,
so
I
thought
the
purpose
of
using
the
pvc,
though,
was
to
leverage
the
way
that
we
mount
volumes
to
to
keep
that
inherent
functionality
for
the
data
sets
I'm
switching
to
cozy.
It's
not
going
to
be
exactly
the
same,
but
that
you,
regardless
you're
still
the
idea
is
still
you're
wanting
to
basically
provide
an
easy,
accessible
way
to
a
specific
data
set.
So
it's
almost
like
pre-populating
a
bucket
basically
and
allowing
a
pod
to
point
directly.
D
I
I
need
to
study
a
bit
more,
the
cozy
proposal,
but
I
I
thought
also
it
was.
I
wasn't
sure
that
it
was
part
of
you
know
creating
a
pvc,
but
you
know
if,
if
it's
more
integrated
the
more
it's
integrated
to
the
kubernetes
environment,
the
better
for
us
right,
so
we
won't
have.
We
don't
have
to
do
any
new
types
of
orchestration
on
our
own
right.
The
pvc.
So
imagine
that
there's
a
one-to-one
mapping
right.
One
data
set
one
pvc.
D
If
we
are
thinking
the
data
set
to
be
like
a
user-facing
thing
right,
so
the
user
is
aware
of
data
sets
and
in
their
posts
they
just
use
a
pvc
them
a
configured
pvc
from
from
the
framework
right
and
also
we're
envisioning
scenarios
where
there
would
be
another
another
provider,
let's
say
creating
the
data
set
pointers,
let's
say
so:
it's
a
data
set
object.
There
would
be
another
persona
creating
in
the
cluster.
The
data
sets
and
would
be
the
other
persona,
a
simple
user
single.
D
You
know
a
user
who
just
wants
to
launch
pods
and
loans
workloads
that
will
mount
in
their
pods
without,
without
you
know,
configuring
finding
pvcs
and
all
this
stuff.
We
are
adding
an
admission
controller
to
inject
those
pvcs
in
in
the
pods
with
labels.
But
this
is
an
additional
feature
it
it
will.
D
The
pvc
would
work
on
its
own
as
a
normal
pvc
in
kubernetes
right
and
that
that's
our
goal
not
to
bring
not
to
replace
pvcs
not
to
replace
csi,
but
instead
have
a
more
user-friendly
interface,
and
maybe
you
know
even
configuring
it
for
for
the
users
who
don't
want
to
be
bothered
with
you
know,
provisioning
and
creating
the
configuration
for
their
pvcs.
A
Okay,
I
this
is
very
useful,
so
thank
you
for
putting
together
this
faq.
I
think
that
will
help
clarify
some
of
the
questions
that
the
toc
had.
I'm
not
sure,
and
it
doesn't
look
like
amy's
on
what
the
review
cycle
is.
Maybe
it'll
be
pushed
out
until
october
again,
when
they
can
get
back
to
it,
but
I'll
find
out
and
I'll
I'll.
Let
you
know.
D
So
the
the
other
thing
I
would
say
is
you
know
we
were
striving
to
get
more
feedback.
So
what
I'm
trying
to
understand?
The
new
process
is
a
bit
not
clear
to
me,
because
I
was
starting
to
put
on
a
pull
request
and
it
was
a
google
form,
and
now
it
was
a
bit
kind
of
hard
to
keep
track
on
where
the
review
was
happening,
and
I
didn't
get
a
chance
to
explain
a
bit
more
well,
I
understand
it's.
D
It
should
be
part
of
the
framework
and
the
documentation,
but
yeah
anything
anything
that
I
can
answer
or
give
a
better
explanation
or
demo.
I
would
be
happy
to.
A
Right,
so
it's
perfectly
fine
that
you
find
the
new
process
a
little
confusing
because
it's
only
been
since
august
and
I
don't
think
it's
ironed
out
yet
so
the
new
process.
The
idea
was
that
they
were
going
to
try
to
simplify
the
way
that
projects
got
into
sandbox
and
not
require
a
presentation.
A
The
idea
was
that
the
toc
could
look
over
the
questionnaire
that
was
filled
out
and
then,
if
they
had
additional
information
they
would
kick
it
back
to
the
sig
as
they
did
here,
and
then
we
would
provide
that
information.
So
yeah
you're
not
required
to
do
a
presentation.
You
know
you
you
already
prevent
you
know
you
already
came
on
the
sig.
We
recorded
that
we
provided
that
information.
That
communication
is
still
being
ironed
out.
So
I
think
you've
provided
everything
you
need,
if
not
I'll,
reach
back
out
to
you
and
let
you
know.
A
Yeah
thanks
I'll
I'll,
send
this
over
to
the
toc
find
out
when
their
next
meeting
is
and
I'll
get
back
to
you,
then
thank
you
for
this
all
right.
Moving
down
the
list
kieran,
I
believe
you
had
some
updates
on
the
licensing
questions
we
had
from
the
last
time.
We
talked
about
open
ebs.
C
Yes
siren,
I
have
updated
the
agenda
doc
with
the
information
I'm
trying
to
put
together.
The
action
item
was
to
list
all
the
open,
ebs
top
level
repositories
and
what
are
the
dependencies
those
repositories
have
on
other
projects.
I
can
probably
share
my
screen
and
quickly
walk
through
that.
That's
okay,.
B
C
C
So
what
I
hope
you
can
see
my
screen
now,
yeah
perfect,
along
with
the
c
store,
I
kind
of
updated
for
the
remaining
repositories.
Also,
so
maybe
I
you
know
since
c
store
was
the
first
one
that
had
a
lot
of
open
questions,
I'll
cover
that
and
then
we
can
go
back
to
the
other
ones,
all
right.
C
So
the
main
concern
with
the
c
store
is
it's
a
it
kind
of
depends
on
the
zfs
and
zfs
itself
is
a
cdl
licensed
project.
So
what
openvps
did
was
for
the
openvps
dfs
project,
which
is
actually
a
kernel
cfs.
C
You
know
you
can
kind
of
build
kernel
modules
with
that
and
ported
that
to
work
for
running
it
in
user
space,
so
the
open
vpc
store,
which
is
a
fork
of
open,
eps
dfs
cfs,
is
a
modifications
for
making
it
work
on
user
space
and
then
the
actual
functionality
of
how
open
ebs
uses
that
framework
or
uses
the
cfs
for
storing
the
data
and
the
replication
and
high
availability
features
that
were
added
on
top
of
it.
They
are
all
part
of
the
open,
eb
slip,
c
store
and
open.
Eps
is
gtt
reports.
C
C
Now
the
you
know,
prior
to
the
last
call
the
way
the
code
was
being
built
was
the
openvc
store
was
actually
pulling
in
the
changes
from
libsy
store
and
building
that
binary.
I
think
that
was
kind
of
highlighted
as
an
issue.
So
now
we
turned
around
that,
so
lip
c
store
is
the
one
that
actually
contains
the
main
call.
If
you
will
that
instantiate
the
binary
and
it
uses
the
open
bbc
store
as
a
library
right
just
like
any
other
project,
we
use
any
other
dependencies.
C
So
that's
where
it
is
at
now
still
keeping
it
open
for
discussion
and
trying
to
understand
if
this
is
okay
or
anything
further
needs
to
be
done
and
leave
it
at
that.
A
So
yeah,
I
think
this
helps
clarify
how
you
know
how
you're
using
it,
how
it's
being
built
differently.
I
think
it
addresses
the
questions.
I
would
need
to
probably
run
this
by
alex
and
quentin.
I
think
we
have
to
have
consensus
from
the
leads
before
we
move
forward
and
then
it
would,
of
course,
still
go
through
the
the
due
diligence
as
it
was
normally
so.
But
before
that,
does
anyone
have
any
questions
around
this
or
concerns
that
they
want
to
bring
up.
A
Off
the
cuff,
it
seems
to
satisfy
the
concerns
we
had,
but
let
me
talk
with
alex
and
clinton
and
get
back
to
you
karen
does
that
work.
C
Last
time
when
we
had
the
presentation,
I
think
this
was
the
first
thing
that
we
had
to
get
across
before
assigning
the
reviewer.
I
I
don't
think
anybody
is
assigned
yet.
Okay.
A
A
Okay,
the
last
item
we
had
was
provega.
We
need
so
justin
cormac
from
the
toc
offered
yesterday,
the
day
before
to
run
the
due
diligence,
but
we
also
need
a
tech
lead
from
the
sig,
so
is.
Is
there
any
volunteers.
A
B
A
B
Actually,
it
would
be-
maybe
nice,
maybe
maybe
at
the
next
meeting,
if
we
could
go
through
the
due
diligence
such
process,
because
I
myself
do
not
know
it
very
well.
I
don't
know
if
many
people
do.
I
don't
know
if
that
will
help.
B
E
Aaron
this
is
tom.
I
I
would
like
to
draft
that
process
and
kind
of
see
it
in
action.
So
the
provega
is
an
interesting
technology.
If,
if
there's
a
chance
for
a
newbie
to
kind
of
watch
along
and
see
it
see
it
go
through
the
paces
of
coming
in
that
would
be
great.
A
Yeah
that'd
be
awesome,
I
I
don't
think
we
have
a
good
process
today.
I
think
in
the
past
it's
just
been
whoever
has
the
time
and
we
go
through
it
and
we
just
work
together
and
it's
it's
not
necessarily
repeatable
or
comprehensive.
So
absolutely
if
you
have
time
to
volunteer
and
want
to
do
that,
that
would
be
awesome
tom.
Thank
you.
A
All
right,
that's
all
I
had
for
the
agenda
today.
Does
anyone
else
have
anything
that
they
want
to
talk
about
or
bring
up.