►
Description
Kubernetes Data Protection WG Bi-Weekly Meeting - 20 April 2022
Meeting Notes/Agenda: -
Find out more about the Storage SIG here: https://github.com/kubernetes/community/tree/master/wg-data-protection
Moderator: Xing Yang (VMware)
A
Hello,
everyone
today
is
april
20th
2022..
This
is
the
kubernetes
data
production
orange
group
meeting
today.
I
think
we
will
just
go
over
the
tabs
and
designs
that
we
are
working
on.
Let's
see
the
shung
we'll
do
a
quick
demo
for
sweet
tea,
and
then
we
can
talk
about
the
you
know.
Kubecon.
A
All
right,
I
think,
let's
get
started
the
first
one
is
a
voting.
Populator
is
ben
here
looks
like
penn
has
not
joined
okay,
so
I
think
last
time
he
gave
an
update.
Basically,
all
the
code
emerged,
blog
is
pending
for
volume,
populated
work
and
I
think
that's
probably
still
the
same,
because
the
blog
won't
be
published
until
1.24
is
out,
and
next
one
is
the
body
mode
conversion.
A
So
work
is
still
going
on.
There
are
prs
submitted
in
external
snapshotter
and
external
provisional
repos.
They
have
been
reviewed.
I
think
we
are
on
track
for
that
and
and
then
there's
also
the
there's
a
dog,
pia
that's
still
being
reviewed.
A
It's
just
the
dark
team
have
some
concerns
regarding
how
the
dock
is
written,
because
that's
the
the
entry
entry
dock,
they
found
it's
confusing
when
we
mentioned
external
snapchatter
release.
So
that's
usually
not
in
the
introduction.
We
are
sorting
that
out
and
there's
also
a
blog
pending
and
cozy.
So
this
is
still,
I
think,
half
is
still
being
reviewed.
A
Okay,
so
now
we,
this
doesn't
move
alright
cbt.
A
B
We
can
give
a
quick
update
so
yeah
yesterday
or
the
day
before
a
few
of
us
met
and
yeah.
I
think
we
are
making
some
progress
with
the
crd
controller
yeah.
B
I
was
able
to
do
just
a
quick
demo
using
like
a
test
client
to
write
like
data
into
like
eps
volumes,
yeah
like
a
block
pvc,
and
that
would
set
up
like
the
test
scenarios
from
like
at
least
a
client
inside
how
we
can
like
use
linux
commands
to
to
to
like
yeah,
basically
set
up
that
scenario,
deciding
how
much
data
to
write
which
blocks
right
to
how
many
blocks
to
write
through
and
then
calling
the
aws
direct
ebs
direct
api
to
retrieve
back
like
the
change
blocks.
B
So
all
of
that
would
tie
back
into
like
the
I
guess,
the
entire
end
to
end
for
differential
snapshot
from
you
know
a
client
calling
it
all
the
way
to
get
change
blocks
back
and
then
back
that
back
into
the
custom
resource.
I
think
earlier,
like
a
couple
weeks
ago,
we
also
did
like
a
quick
spike
effort
on
using,
like
aggregated
api
server,
to
pull
back
like
resources
from
a
mock
endpoint.
B
Without
having
to
store
like
custom
resources
into
scd
via
like
the
kubernetes
list,
pagination
mechanism-
so
I
know
like
also
tron-
has
like
a
few
pros
open
with
just
the
grpc
endpoints
defined
and
some
effort
to
try
to
merge
our
stuff
with
the
the
ebs
arms
csi
driver
sean.
Are
you.
C
Yeah
I'm
on
sorry,
I
missed
the
protective
data
protection
working
group
call
it's
pretty
late
for
me,
but
so
I
apologize,
but
I
do
actually
have
a
quick
demo
that
you
know
I
have
the
api
installed
in
a
cluster.
C
I
have
an
ew
and
ebs
driver
implemented
that
it's
actually
using
the
given
grp
service
to
contact
the
aws
apis
to
get
the
change
blocks,
and
then
I
have
the
controller
calling
that
grp
service
and
I
have
like
I
could
just
do
a
quick
like
it
really
is
just
as
simple
as
creating
the
resource,
and
you
can
see
the
change
blocks.
I
think
I
have
a
bug
because
I
don't
think
I
did
the
math
to
get
the
offset
correct
on
the
way
out,
but
that's
fine.
C
We
can
fix
that,
but
I
do
have
a
working
end
to
end
of
the
three
pieces
that
you
would
need
and
the
next
step,
I
think
for
me,
is
to
write
down
some
lessons
learned
and
maybe
some
things
that
I
think
we
can
make
for
a
more
ergonomic
api
from
a
user
perspective
and
then
also
from
the
grpc
perspective
but
yeah.
If
anybody
wants
to
see
it
yeah.
A
Sure
please
go
ahead,
so
I
have
made
you
a
co-host.
C
Oh,
thank
you.
Yes,
I've
honestly
have
never
done
this
with
zoom
before
so.
This
will
be
very
interesting.
Okay,
I
apologize
folks
if
this
is.
C
B
So
yeah,
while
we
are
waiting
for
sean
dave,
has
also
shared
some
like
steepness
of
codes
with
integration
with
the
aws
ebs
direct
api.
Where
yesterday
you
walk
us
through
something
he
has
done
dave
you
wanna,
add
to
that
or
anything
there.
D
No,
this
is
some
work.
We've
done
in
the
past
to
hook
up
ebs
direct
with
astrolabe,
so
I
think
the
next
step
I
started
adding
in
some
cbt
apis
from
the
top
end
and
then
connecting
those
up
to
the
crd
stuff
that
sean's
doing
should
let
us,
actually,
you
know,
try
reading
the
blocks
directly
using
ebs
direct
with
the
crd
list
from
the
with
the
change
block
list
from
the
crd
cool.
B
Thanks
yeah,
so
the
trd
are
over
always
like
the
overall
status
is
like
yeah
between
the
four
of
us,
like
we
have
different
pieces
working
and
now
we
just
need
to
sit
down
together
and
put
our
brains
together
to
make
it
like
into
one
course.
If,
like
demonstratable,
like
slow,
anyways,
sean.
C
Yeah,
okay,
sorry,
I'm
back
and
hopefully
this
works.
I
think
I
need
to
become
a
cohost
again.
I'm
sorry.
B
Did
you
lose
that
co-host
permission
thing
he
said.
C
A
C
Yeah
yeah
yeah,
okay,
so
this
is
just
this
is
the
top
pan.
Pane
is
just
the
controller,
so
this
is
watching
get
change
blocks
api
and
I
have.
C
C
No,
so
it's
actually
using
both.
So
the
controller
is
actually
using
the
grpc
api
to
call
into
the
a
a
aws
driver
that
I've.
C
A
I
was
wondering
because
I
remember
in
in
the
design
that
showed
earlier:
it's
like
a
separator
service,
it's
not
the
controller
service
or
just
wondering
yes,.
C
I
think
eventually
you
would
want
to
move
that
definition
somewhere,
not
in
the
controller
code
like
I
have
it
today,
but
today
it's
in
the
controller
code
and
the
ebs
driver
is
using
that
service
to
serve
that
given
grpc
endpoint
and
then
the
controller
is
using
the
client
that's
generated
to
contact
that
service,
and
so
I
created
a
couple
snapshots
installed
the
api
and
you
just
go
ahead
and
create
that
object,
and
we
should
see
things
happen,
hopefully
cool,
so
you
kind
of
processed
it
right.
A
C
And
this
is
where,
like
I
think,
we
probably
want
to
do
some
work
on
the
api.
The
initial
api
has
like
the
snapshots
as
the
base.
We
probably
want
those
like
the
snap
ids,
rather
like
the
snapshot
handles
from.
C
The
base
right
now
is
snapshot.
I
d,
like
it's,
not
volume
snapshot.
If
I
understood
correctly,
it's
not
like
a
an
object
reference
to
a
volume
snapshot
to
get
that
handle.
It's
like
the
handles
actual.
A
C
Figure
out-
and
I
think
that's
fine
and
I
have
to
write
down
some
of
those
things
but
end
to
end.
I
think
this
would
work.
I
also
have
some
questions.
D
A
D
There's
like
the
ebs
direct
thing,
there's
like
a
token
it's
weird,
but
that
was
what
we're
gonna
put
in
the
context.
C
I
was
not,
I
did
not
need
to
do
any
of
that
to
get
the
adws
driver
to
contact
the
change
block.
All
I
had
to
do
was
add
permissions
to
no.
C
D
I
agree,
however,
in
order
to
actually
read
the
data
using
the
evs
direct
api.
You
need
to
pass.
In
the
token,
I
believe,
along
with
the
offset.
C
D
Well,
because
it
would
be
nice
that
you
don't
have
to
clone
and
attach
the
volume,
if
you're
going
to
get
it
so
so
ebs
direct
right
lets.
You
read
and
read
blocks
directly
from
the
snapshot.
Oh.
A
C
D
No,
so
the
so
in
fong's
design
right
we
have
that
additional
context
field,
and
that
was
something
that
could
be
passed
from
the
change
block
tracking.
So
if
we'd
set
that
with
the
token
then
the
consumer
of
that
can
say,
oh,
I
know
that
I'm
doing
ebs
direct.
So
I
use
the
context.
You
can
also
just
use
the
offset
if
you
clone
and
attach
the
volume,
but
the.
C
A
C
Is
wait?
I'm
sorry.
B
B
So
as
far
as
your
controller's
scope
is
concerned,
put
those
tokens
into
the
context
like
the
context
is
like
it
should
be
part
of
your
got.
It.
C
F
A
F
B
So
so
dave,
I
just
want
to
confirm
my
understanding
so
so
that
token
is
not
about
pagination,
it's
about
actually
getting
the
data,
but
it's
so
easy.
B
A
Are
you
okay,
so
this
is
your
I
think
ivan.
This
is
your
test.
B
Okay,
so
I
think
we
need
to
sean
just
need
to
put
this
into
his
okay
into
the
context
of
viewing.
A
D
A
He
needs
he
should
add
that
token
in
here.
D
Yes,
okay,
I
see,
and
so
it
should
be
opaque
to
kubernetes
right.
It's
just
junk
and
you
don't
have
to
use.
A
Okay,
okay,.
D
D
A
A
D
The
at
the
moment,
the
at
the
moment
you
know
the
at
the
moment
we
don't
have
a
data
path
to
fund
right
the
data
path,
the
the
the
standard.
D
B
Yes,
yeah,
so
the
consumer
would
consume
like
line
54
in
the
status
right,
so
in
there
they
will
have
the
contacts
and
then
whatever,
however,
they
want
to
abstract
or
whatever
they
want
to
do
with
the
context.
It's
up
to
you.
D
D
B
Yeah,
so
sean
yeah
you're
right
he
okay.
He
added
some
questions
in
chat,
but
yes,
I'm
showing
you
all
right.
It's
just
a
pass
through,
like
the
consumer,
will
decide
how
they
want
to
consume
what
is
inside
context
and
how
they
want
to
consume
it
so
but
yeah,
I
think
overall,
we
have
like
it
looks
like
we
have
to
major
different
pieces
together.
Now
we
just
need
to
have
one
meeting
where
we
put
our
brains
together
to
make
it.
B
It's
in
a
branch
I
mean
it's
of
the
pull
request,
so
the
client
side
is
in
this
sub
repo.
So
if
you
click
on
pull
requests
on
there,
so
sean
has
a
okay,
a
work
in
progress
there.
So
this
is
a
client
site.
The
server
side
is
in
the
fork
that,
off
from
the
upstream
ebs
csi
driver
that
sean
has.
A
B
So
I
think
the
one
thing
that
I
just
after
talking
to
foundations,
two
days
yesterday,
like
line
fourteen
and
fifteen
like
where
we
need
to
find
the
snapshot
id
or
the
snapshot,
handles
oh
yeah
on
this
page
like
for.
I
wonder
if
we
can,
just
like
you,
know,
replace
it
with
like
the
the
volume
snapshot
content,
so
that
the
controller
will
go
and
look
for
it
because,
right
now
we
have,
I
think
we
have
to
manually
put
in
the
the
snapshot
handle
like
like
lines,
14
and
15.
G
B
Sean
later
like
like
right
now
like
14
or
15,
it's
like
the
snapshot
handles
right.
I
wonder
if
we
can
yeah
and
wonder
if,
instead
of
snatching
handles
like,
we
can
just
point
it
directly
to
the
volume
snapshot,
content,
the
hassle.
A
A
F
A
B
A
It
should
be
volume
snapshot,
yeah,
one.
A
A
A
D
Well,
no,
we
need
that
to
like
so
the
easter
thing
is
we
should
the
the
recommended
thing
is.
We
should
take
a
snapshot
back.
It
up
then
delete
the
snapshot,
and
then,
when
we
take
the
next
snapshot,
we
can
use
what's
called
the
epoch
from
the
old,
from
the
deleted
snapshot
as
our
base
and
get
a
diff
against
it.
D
A
B
I
think,
like
it's
not
unreasonable
to
make
the
volumes
the
csi.
D
B
D
A
Because
we
also
wasn't,
is
this
a?
Is
this
namespace
or
not
namespace,
that's
another
question
right.
So
if
it's
a,
if
this
is
a
namespace,
then
we
should
not
have
the
snapshot
handle
here.
D
A
D
A
D
A
Better,
both
is
sean
back.
A
A
Yeah
the
handle
should
be
that
should
be
now
name
space.
That's
what
I'm
saying
if
so,
it's
a
little
because
I
was
thinking
this
should
be
when
we
take
snapchat,
that's
namespace,
so
this
one
should
be
namespace,
but
then
but
then
we
don't
have
that
we
can't
have
access
to
the
snatcher
handle
if
we
are
in
namespace.
D
A
Right
but
that,
but
that's
still
oh
you're
saying
you
know
you
could
ask,
but
but
if
that
does
not
exist
anymore,
we
cannot.
We
cannot
provision
it
anymore.
D
D
Oh
yeah,
well:
hey
you're!
The
easter
person
now.
B
I
guess
the
next
converse
the
next
item
in
our
next
sync,
up,
at
least
for
the
the
engineering
team
is
to
like
you
know
what
should
the
controller
interface
with
to
get
the
volume
handle?
That's
sorry,
it's
natural
hands!
What.
A
Is
okay?
It
is
sean
back
or
still
having
a
problem.
A
A
So
the
next
item
we
today
we
we
actually
don't
have
much
next
step
and
basically
just
kubecon.
We,
I
think
we
talked
about
this
briefly
last
time,
so
I
think
we
could
meet
up
at
the
contributor
summit.
So
if
you
are
attending
so
right
now
we
have
a
few
names.
So
if
anyone
else
is
attending,
please
add
your
name
there.
So
we
know
that
you're
coming.
A
All
right
so
basically
we'll
need
to
so.
The
contributor
summary
is
a
monday:
it's
a
it's
like
a
whole
day.
They
will
have
the
room
there,
but
before
going
there,
you
need
to
you
know,
get
this
registration
vaccine
verification
badge
pickup
but
looks
like
it's
at
the
same
location.
F
I
want
a
quick
question
actually
for
the
group.
Have
we
seen
many
people
have
challenges
using
multiple
versions
of
the
snapshotting
api.
F
Using
the
like
the
beta
and
the
ga
or
the
the
alpha
and
the
beta,
for
example,
version
of
the
sound.
A
H
F
C
F
Yeah,
you
know
people
are
running
old
clusters,
so
just
yeah.
H
F
H
G
He
came
just
this
is
eric
from
infinidat
related
to
that
the
snapshot
v1
beta
1
api
and
the
snapshot
v1ga
api
are
the
same
right
or
they're
they're
they're
fully
compatible.
That
was
our
assumption
in
designing
a
new
snapshot.
Sidecar.
A
H
There
will
be
a
new
field,
but
that's
optional.
I
think
the
problem
is
just
people
that
want
to
have
right.
Clients
that
are
compatible
with
a
very
wide
range
of
kubernetes
versions
have
to
suffer
through
this
additional
pain
and
and
the
more
time
that
passes
the
less
people
will
care
about.
The
beta
version
and
v
won't
be
the
only
thing
that
matters.
G
G
All
the
ones
granted
we
we're
a
small
sample
size.
Most
of
our
customers
aren't
extensively
using
snapshots,
but
all
the
customers
that
I
can
think
of
from
our
perspective
are
on
v1,
ga
or
v1
beta1.
G
B
Okay,
sorry,
I
probably
missed
this
like
in
previous
discussions,
but
I
just
wonder
like
what
I
guess
do
we?
What
does
that
migration
path,
look
like
from
one
version
to
another?
Is
it
just
your
typical
cid
version
of
great
type
of
path.
H
A
You
just
need
to
make
sure
you
don't
have
any
invalid
snapshots
there,
because
we
added
the
there's
a
validation
web
hook
that
we
added
some
time
ago
to
prevent
invalid
snapshots.
But
before
that
was
added,
people
you
know
would
be
able
to
create
those.
A
So
so
if
you
so,
if
you
upgrade,
you
still
have
those
you
know
old
snapshot
that
are
invalid,
then
you
can't
remove
it
anymore.
Basically,
but
that's
the
that's
something,
but
probably
that
won't
be.
The
normally
won't
be
the
case.
F
Yeah,
we
have
a
really
wide
range
of
customer
versions.
I
mean
very
old,
even
going
back
to
111,
so
we
we
end
up
we're
kind
of
essentially
shifting
to
using
dynamic
lines
for
everything.
Now
you
like
apis
that
are
changing.
A
1.13,
I
think
it's
c
csi
when
ga
in
windows
13.
I
believe.
A
Okay
yeah,
so
that's
the
yeah.
So
basically
we
will
be
that's
the
basic
two
changes
related
to
one
snapshot.
Apis,
you
might
not
24.
Why
is
we
want
we're
going
to
delete,
remove
the
support
for
we
want
beta
one.
A
The
other
one
is
this
new
alpha
feature
is
we're
adding
this
new
source
volume
mode
in
women's
natural
content.
G
A
All
right,
so
if
there
was
nothing
else
yeah,
I
thought
someone
was
saying
something
so
bad.
I
think
you're
enjoying
late
so
for
voting
voting
populator
we're
basically
just
waiting
for
the
blog
right.
Everything
else
is
done
right.
H
Yes,
yes,
sorry,
I
joined
late,
I
had
a
scheduling,
snafu,
the
the
blog
is
out
for
review
and
the
editors
have
sort
of
disappeared.
I
don't
know
if
it's
related
to
the
delay
in
the
release,
because
they
delayed
the
actual
release
for
a
couple
weeks
due
to
that
compiler
bug.
H
But
I
haven't
seen
any
more
reviews
on
my
blog.