►
Description
Meeting of Kubernetes Storage Special-Interest-Group (SIG) Volume Snapshot Workgroup - 30 July 2018
Meeting Notes/Agenda: https://docs.google.com/document/d/1qdfvAj5O-tTAZzqJyz3B-yczLLxOiQd-XKpJmTEMazs/edit#heading=h.qbngbmepbnss
Find out more about the Storage SIG here: https://github.com/kubernetes/community/tree/master/sig-storage
Moderator: Jing Xu (Google)
A
A
From
last
time,
we
kind
of
decide
the
bottom
snapshot
status
struct,
but
someone
mentioned
the
rights
or
user
side
how
to
monitor
whether
this
snapshot
is
that
complete
already
so
that
the
user
know
when
they
can
start
using
it
to
prove
it
morning.
Last
time
we
mentioned
we
use
chem
stamp
for
available
at
to
represent
this,
and
the
window
can
stand
for
available
at
is
that
we
say
it
is
the
whole
thing
complete,
but
everything
about
is
and
also
checking
our
controller.
A
A
So,
let's
have
a
quick
overview:
what
happened
that
between
snapshot
controller
and
the
CSI
modem
driver
during
this
whole
process?
So
at
the
very
beginning,
when
users
start
like
create
snapshots
object,
smart
controller
will
send
a
request.
A
Chris
natural
request
to
CSI
driver
and
co-driver
bill
go
ahead
to
karlik.
A
A
First,
because
snapshot
is
continent
so
any
to
update
the
status
to
set
the
created
at
as
the
temp
stamp
of
snapshot,
creation
temps
then
in
case
of
for
some
modern
plugins,
there's
no
uploading
face.
So
probably
it's
also
return
like
actually
writing
status
from
CSI
response.
Meta
means.
You
also
need
to
send
a
variable
act
equal
to
new
like
times
and
t1,
which
means
great,
add
and
no
way
to
act.
A
They
are
actually
the
same
time
and
the
controller
will
also
go
ahead
to
create
volume,
snapshot,
content,
object
and
the
storage
to
API
server
and
then
find
those
two
API
objects
finds
just
means.
You
specify
the
bottom
snapshots,
content
name
equal
to
the
name.
You
created
it
just
now,
and
this
is
not
atomic
operation
that
you
need
to
perform.
Each
operation
sequentially.
A
After
this
update,
create
points
finishes
for
some
modern
plugging.
If
there
is
uploading
phases,
the
controller
will
keep
checking
periodically,
checking
the
snapshot
status.
It
will
cause
yes
I,
let's
say
list
them
shot,
API
cause
and
the
CSA
driver
return
response
when
the
snapshots
uploading
finishes
so
eventually
it
will
return
for
ready
status,
and
that
is
when
controller
will
set
available
at
a
different
time
stand,
and
then
that
is
the
moments
we
can
say.
This
snapshot
is
complete.
A
These
completely
ready
to
be
used
so
I'm
thinking,
we
can
add
a
boolean
value
to
represent
this
whole
thing.
The
whole
process,
complete
and
central,
is
completely
ready
to
be
used.
Another
word
is
that
means,
when
you
mark
this
natural,
complete
there's,
a
few
conditions
must
be
satisfied.
First,
cricked
available
ads
should
be
already
set,
so
those
are
two
timestamps
basically
telling
you
when
the
snapshot
is
created
when
a
snapshot
finish
uploading
and
also
the
bottom
snapshot.
A
That
kind
of
timeline,
workflow,
okay,
steering
static
provisioning.
That
means
there
is
a
snapshot
already
available
from
the
provider
or
some
modern
provider
and
user
just
want
to
specify
by
themselves
the
volume
step
charge,
content
where
they
create
find
themselves
manually
and
then
point
to
the
functions
name
to
that
button.
Name
to
that
object
and
the
controller
should
be
able
to
handle
this
illusion
to
to
bind
them
together
so
to
create
add
and
available
may
not
be
available.
A
So
I
think
we
can
set
a
writable
adds
by
checking
the
snapshot
in
status.
The
controller
can
still
go
ahead
to
check
in
with
Natron
status
and
if
an
AC
is
already
ready,
then
that's
available
at
equal
to
some
time
but
quick.
Yet
it's
my
return.
The
correct
time
stamp
or
not
so
consider
static,
find
mean
part,
so
available,
use
available,
ads
or
kudex.
It's
also
not
very
accurate.
A
To
represent
this
next
shot
is
complete,
so
here
now
proposing
to
add
a
new
fully
Mario
or
we
can
call
it
I'm,
not
sure
the
name
that
it
represents.
This
status
of
the
snapshot
is
complete
so
for
TV
BBC.
If
thank
you,
you
see,
we
use
tick.
Respond
to
represent
a
face
that
this,
the
PVC
is
ready
to
be
used.
We
can
also
use
bond,
but
actually,
for
our
case
we
mean
so
snapshot
is
available
right.
It's
finish,
uploading.
If
there
is
a
protein-based
and
also
it
finished
binding,
we
can
also
call
it
complete.
A
A
A
A
C
A
When
does
status,
I
say
the
pump
or
complete
equal
to
qu?
We
no
longer
holding
this
snapshot
only
when
after
a
shot
is
created,
but
is
the
way
to
go,
ed
is
not
set
yet.
So
we
need
to
know
when
the
snapshot
is
complete,
finish
uploading.
So
during
that
period
time
we
need
to
call
this
a
snapshot.
Well,
we
can
kind
of
reduce
the
overhead
by
using
I,
say
back
off
exponential
back-off.
B
A
D
A
Observable
at
is
that
right,
so
there's
no
need
to
keep
checking
unless
okay
verify
that
yeah
sure,
unless
some
user
preferred
I
could
say
they
want.
You
considered
shadows
in
case
some.
They
manually
delete
into
snapshot.
That's
currently
in
your
design.
We
say
after
checking
is
already
finished,
uploading
and
started
with
wet.
We
no
longer
check
anymore.
A
B
A
E
A
Because
remember
PvP
BC,
we
have
two
objects
and
finds
them
news.
He
means
he
has
a
pointer
to
a
PDA,
so
have
a
pointer
back
to
BBC.
Here
is
the
same
thing:
the
warning
snapshot,
wise
means
it
has
a
pointer
that
is
that's
the
pointer
to
the
falling
snapshots,
content
and
content
object.
Also,
of
course,
have
a
pointer
back
to
the
volume
snapshot.
F
A
Yeah
I
recommend
it,
as
we
mentioned
before.
This
must
begin
as
very
necessary
right,
because
the
issue
of
like
updating
in
case
you
have
new
faces.
So
here
we
just
at
least
for
current
days.
I
will
only
need
a
boolean
value
to
represent
everything,
finish
everything
done
and
is
ready
to
be
used.
So
it's
very
simple
and
clear
to
you.
You
just
use
a
putting
value,
there's
no
need
to
have
like
different
places
like
uploading
face,
or
so
we
want
to
avoid
doing
that.
Like
state
machine
model,
yeah.
F
A
A
B
Play
devil's
advocate
on
that
I
think
I
disagreed
with
our
tool
on
last
time,
but
now
thinking
it
over
you
mentioned
was
the
user
might
want
to
use
this
information
to
decide
when
they
can
start
their
application
again?
Did
that
process
be
controlled
by
kubernetes,
meaning
we're
the
ones
that
are
going
to
signal
to
the
application?
At
some
point.
B
A
B
Not
sure
what
the
default
configuration
is
for
our
back
for
most
clusters,
but
I,
imagine
it
should
be
that
the
end
user
should
be
able
to
read
it,
but
not
write
it.
If
that's
the
case,
then
it
would
be
sufficient
to
put
it
inside
snapshot
content
object,
but
if
the
default
is
that
the
user
has
absolutely
no
access
to
the
non
namespaced
object,
then
it
makes
sense
to
put
it
on
the
namespace
object.
D
Previously
timestamp
one
I
mean
to
status
relying
in
the
loading
snapshot,
wanyan
one
essential
content,
but
then
we
decide
to
come.
G
F
B
G
A
A
A
I'll
update
it
again
right
now,
I,
just
yeah,
seeing
how
me
for
updates,
puts
pond
putting
weather
there
and
we'll
also
probably
add
a
little
bit
more
information,
including
the
timeline.
So
I'm
not
sure
this
is
too
much
or
not
so
busy
showing
when
those
crazy
odd
available
at
is
that
and
what
happened
between
snapshot,
controller
and
the
volume
driver
and
when
the
the
pond
will
be
set,
it
means
sensuality
is
complete
and
also
we
have
the
ping.
The
PR
for.
A
The
controller
and
the
in
external
step
shorter.
So
if
anyone
interesting
like
how
controller
design
please
review
it,
I
think
today
will
place
our
like
decision.
We
also
have
a
new
version
to
add
the
the
pond.
Putting
Mario
and
I
will
also
submit
a
GC
CSI
driver
problem
is
couple
days
and
I'm
thinking.
We
can
start
testing
our
controller
and
if
anyone
interesting
it,
we
can
create
their
own
CSI
volume,
plugins
and
test
it
and
we'll
start
adding
like
more
tests.
A
E
A
A
A
We
cannot
mark
step
shots
like
complete
in
our
cell
phone
to
children,
and
the
static
controller
will
continue
to
sink.
There
is
like
the
article
sink
loop
for
each
snapshot,
object
and
next
time
in
see
this
snapshot
object.
It
might
steal
credit
as
already
sex,
but
there
is
no
bottom
central
content
object
and
it
will
child
game.
So
basically,
the
controller
will
keep
trying
to
create
this
volume
surcharge
contents
over
and
over
again,
if
your
heart's
good
it
until.
E
A
A
E
A
A
A
A
It's
like,
however,
yeah
it
has
a
sink,
so
I
say
every
one
minute.
It
always
think
the
volume
snapshots
object
if
the
finals,
that
is
it's
not
fun
yet
and
we'll
try
to
kind
of
recover
from
any
like
failed
points.
For
example,
you'll
fail
to
create
content
before
so
it
should
start
creating
natural
content
again
yeah
and.
A
Because
of
this
I,
don't
think
we
need
to
let
go
how
to
delete
the
snapshots.
The
other
line
is
go
snapshot,
and
so,
if
just
because
of
API
server
is
temporary
done
and
you
actually
already
creates
natural
successfully,
that's
you
failed,
create
content.
One
time
like
at
that
moment,
I,
don't
think
we
need
to
delete
the
physical
snapshot,
because
the
controller
will
try
to
recover
and
create
content.
Next
time.
G
F
A
F
A
C
I
suspect
it's
I
have
to
put
in
to
CSI
first
is
to
do
with
natural
groups,
and
so
currently
with
the
control.
There's
no
guarantee
or
no
way
of
say.
If
the
pod
has
multiple
TVs,
there's
no
way
of
gain
at
the
consistent
point
in
time,
snapshot
of
those
logical
TVs
and
so
having
the
ability
to
snapchat
a
group
of
volumes
at
the
same
time
and
guarantee
that
that
was
at
the
same
time
would
be
a
good
feature.
A
A
I
think
we
briefly
like
this.
Just
that's
definitely
where
history
as
a
future,
but
it
has
a
lot
of
like
great
changes,
have
that
kind
of
guarantee,
I
think
that
shots
consistently
or
group.
But
if
your
dice,
a
warning
ronnie
has
the
capability
of
doing
that
and
we
might
be
able
to
like
add
as
a
feature
and
for
us
we
cannot
in
such
a
guarantee
like
we
can
take
consistent
group
of
snapshots.
Only
plugin
has.
H
H
The
problem
is,
is
we've
talked
about
that
before
in
I?
Agree,
it's
a
good
capability,
but
the
trick
is,
is
it's
it's
vendor
dependent?
First
of
all,
so
not
everybody
can
do
it
and
I
think
the
the
first
step
would
be
to
get
a
same
snapshots,
implementation
in
and
then
figure
out.
You
know
how
we
want
to
expose
vendor
capabilities
like
taking
the
screen
snapshots
and
things
like
that.
A
I
A
Yes,
right
now,
I
think
we
pretty
much
said,
along
with
add
a
data
source
API
which
they
destruct
in
PVC.
So
this
data
source
reference
is
had
local
object
references.
So
it's
only
contains
the
name
of
the
object
and
the
kind
of
the
object
so
for
snapshot.
The
kind
would
be
you
know,
volumes
then
shot
and
the
name
would
be
your
volume
snapshots,
the
object
name
and
we
will
work
on
a
external
provisioner
and
to
be
able
to
recognize
this
a
source.
A
For
other
type
of
will
I
say
operation
a
clone,
you
can
also
use
the
same
data
source
object,
the
destruct
and
the
kind
will
be
another
PVC,
for
example,
and
the
name
is
a
PVC
name
right
now
we
only
like
allow
snapshots
and
all
the
other
type
we
can
validate
and
fail
the
operation
and
until
I
say
another
clone
function
available.
We
can
open
it
up.
Yep.
I
A
So
the
main
thing
is:
we
have
to
go
through
the
API
review
process
so
force
natural,
although
it
is
out
of
tree,
but
we
still
it
is
there
part
of
the
coup
Nettie's.
No
teachers,
though,
is
you
have
to
go
through
the
API
review
and
I
think
the
API
review
group,
someone
start,
maybe
our
API
and
this
destruct
also
added
to
the
core
the
object.
So
we
also
need
to
get
approval,
but
I
think
it's
good
and
I
should
be
okay.
It's
promising
yeah.
A
And
hopefully
I
will
ping
the
API
a
review
group
this
week
and
they
give
us
feedback
and
see
how
it
goes
and
our
controller
code
should
be
ready
for
testing
and
if
so
you
are
using.
Gc
I
will
have
that.
You
see
Trevor
available
for
testing
and
let
me
know-
and
this
you
know
like
if
you
have
other
driver,
you
want.
D
A
A
A
B
G
D
D
D
B
A
B
Yeah,
basically,
the
idea
that
the
snapshot
itself
is
not
equally
available
to
all
volumes.
We
need
to
figure
out
it.
We
have
to
press
that
I
think
it'll
be
generally
very
similar
to
what
we
did
with
volumes,
but
with
volumes.
The
affinity
is
with
nodes,
and
in
this
case
we
need
to
figure
out,
if
that's
still
the
case
with
snapshots
or
not
well,.
I
B
A
You
so
that's
great
and
like
like,
as
I
said
for
this
week,
we
should
have
the
controller
and
some
host
the
pass
and
the
GC
CSI
driver
available,
and
that's
now
like
but
see
I
said,
driver
you're,
more
interested
and
work.
We
can
start
working
on
that
too,
and
hopefully
we
can
have
this
alpha
available
as
soon
as
possible.