►
From YouTube: WG Data Protection Biweekly meeting 2020701
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
A
A
B
See
it
cool
thanks,
Shane
I
think
we
so
the
background
of
what
I
want
to
discuss
was.
We
were
looking
at
both
the
execution
hook,
KP
that
merged
sometime
ago,
as
well
as
the
recent
container
notifier
proposals,
and
what
we
wanted
to
do
was
share
some
thoughts
on
you
know
just
our
experiences
based
on
customer
use
cases,
we've
seen
and
kind
of
compare
the
two
approaches
when
we
started
looking
this.
B
My
kind
of
gut
reaction
was
that
I
preferred
the
execution
hook
and
I
think
there's
a
lot
more
nuance
there,
which
we
can
kind
of
dig
into
a
little
bit.
So
we've
put
together
this
document
that
kind
of
outlined
some
more
thoughts,
I
linked
it
in
slide
the
the
main
meeting
dock
as
well,
so
you
should
be
able
to
access
it
from
here.
B
You
know
from
a
high
level,
I
think
the
two
approaches
are
trying
to
achieve
some.
The
same
goals
right
essentially
mean
to
be
able
to
quiesce.
We
S
state
before
taking
a
snapshot,
a
lot
of
experience.
We
have,
if
canister
it's
kind
of
including
this
document.
Typically,
we
quiesce
at
the
application
level.
Part
of
that
is
actually
because
we
don't
have
real
hooks
that
we
can
actually
quiescent
the
fosston
level
right
view.
We
can't
guarantee
that
we
can.
B
A
B
A
Okay,
so
maybe
that's
I
think,
is
it
Nolan?
Well,
she
I
think
Lara
uses
some
of
those
hooks.
B
A
B
A
B
Yeah
I
know,
but
that's
a
great
question
and
I
I
think
the
thing
that
I
really
like
about
the
container
notify
proposal
is
this
would
really
kind
of
guarantee.
You
have
that
privileged
access
to
run
that
FS
freeze
unfreeze,
but
of
course
that
comes
with
all
the
caveats
of
having
that
access.
A
B
So
I
wanted
to
go
through
a
quick
discussion
of
the
things
that
I
really
liked
about
the
execution
hook
KP,
and
so
you
know
essentially
giving
reasons
why
my
hope
is
that
the
KP
is
not
forgotten
or
dead,
and
so
we
can
talk
about
my
discussion.
I'll
go
through
some
customer
use
cases
that
we've
seen
kind
of
in
the
field.
You
know
and
they're
targeted
more
towards
like
the
cancer
model,
but
I
think
a
lot
of
them
won't
apply
here
as
well.
B
You
know,
the
big
kind
of
difference
is
that
we
get
this
extra
level
of
indirection
by
having
a
CR
in
the
X.
We
should
hook
KP
by
the
way,
to
keep
what
most
people
are
familiar
with.
The
execution
hook,
KP
versus
the
container
notifier
right,
I
think
we've
discussed
in
the
Terra
Nova
fire
and
past
few
meetings
here
as
well
right
are
there
any
questions.
I
can
dig
into
on
the
differences
between
the
two
I.
B
Think,
from
a
high
level,
the
execution
hook,
KP
has
a
separate
CR
that
you
define
that
defines
what
the
action
you're
performing,
and
so
you
have
an
object
that
you
know
exists
in
the
cluster
that
you
can
modify.
The
container
notifier
proposal
is
adding
the
command
directly
to
the
pot
or
workload
itself,
and
so
you,
you
know
if
you
want
to
do
an
update
to
that,
you
have
to
modify
the
that
pods
back
itself,
and
so
the
yeah.
B
Know
when
I,
when
I
say
we
I'm
actually
talking
about
kind
of
caftans
parts
and
the
reason
that
it
went
bringing
this
up
is
because
I
wanted
to
start.
A
discussion
to
you
know
essentially
give
our
thoughts
on
the
excretion
hook
proposal
based
on
the
fact
that
I
think
a
lot
of
discussion.
This
group
has
been
leaning
towards
the
Container
notifier
proposal
right,
okay,
so
we're
kind
of
kind
of
hold
a
contrary
position.
I
would
say
in
some
sense:
okay,.
B
B
So
you
can,
you
can
have
our
back
on
the
B
object
itself
right.
Maybe
the
controller
that
you
define
to
implement
the
hook
is
only
in
some
namespace,
and
so
maybe
only
certain
users
can
create
the
that
object
in
that
namespace,
and
so
you
can
kind
of
authorize
her
or
G
authorized
access
to
creating
that
on
a
per
user
basis
right-
and
we
see
that
that's
that's
something
that
cancer
does
as
well.
B
We
said
that
kind
of
all
over
the
place
with
our
customers
right,
there's,
typically
kind
of
admins
and
clusters
with
with
maybe
many
developers
and
the
developers
should
not
have
access
to
run
these
hooks.
In
some
cases
you
know,
there's
obviously
many
different
kind
of
patterns
you
can
set
up,
but
that
one
is
pretty
common.
B
B
Customized
right,
you
know,
with
the
with
the
home,
to
our
author,
for
example,
need
to
provide
hooks.
You
know,
have
values
that
let's
basically
modify
this
annotation
yeah
I,
think
that
would
be
feasible,
but
they
probably
be
cumbersome
to
use
and
get
hurt.
You
know
develop
correctly,
whereas
what
we've
done
for
canisters
we've?
Actually,
we
have
several
of
these
kind
of
CRS
and
define
the
workflows
for
backing
up
applications.
You
know
some
that
use
quiesce
and
on
quiesce
and
some
that
use
logical
backups.
A
B
Right
yeah,
you
know
usually
see
that
the
question
is
who's
modifying
it
and
usually
usually
I.
Let
me
see
developers
kind
of
employing
what
they
need.
Then
you
kind
of
have
like
a
backup
admin
or
cluster
admin
that
would
go
and
define
the
integration
with
their
infrastructure.
That
tells
you
how
to
go
and
back
back
things
up,
you
know
and
with
with
canister.
Maybe
you
can
still
get
permissions
to
what
the
developer
take.
Backups
the
fact
that's
pretty
common.
E
C
A
B
We've
definitely
done
that's
a
great
point.
We've.
You
know
I
strongly
agree
at
that
point.
We've
initially
when
we
started
writing
canister.
We
actually
took
upstream
comforts
and
modified
them
to
include
even
just
the
blueprints
er's
the
CRS
had
to
find
the
hooks
inside
the
home
parts
themselves.
You
know
we
found
a
few
things.
We
found
that
it
was
a
big
pain
to
maintain
right
kind
of
consuming
and
rebasing
on
top
of
upstream
was
always
a
non-trivial
effort.
You
know,
you
know
we
build
some
automation
around
it.
B
B
There's
a
few
other
kind
of
points
to
that
I
think
just
some
of
our
thoughts
we
put
down,
so
we
we
believe
that
it's
a
little
more
difficult
to
modify
core
pod
api's
to
include
new
fields
or
essentially
annotations,
which
you
know
the
guidance
we've
been
getting
this
kind
of
that
it's
similar
to
modifying
core
fields.
I
think
that
a
snapshot
controller
is
a
good
example.
B
You
know,
I
think
it's
just
it's
not
a
huge
concern.
Right,
I
think
we
should
probably
design
we
shouldn't
necessarily
let
the
timeline
influence
the
architecture
too
heavily,
but
do
you
think
it
is
at
least
a
point
where
mentioning
that
it
does
take
longer
to
you?
First,
get
things
even
into
these
core
API,
as
well
as
iterate
on
them
as
well
right,
and
if
the
snapshot
controller
is
a
lesson
I
think
what's
in
there,
that
it's
really
useful
to
be
able
to
iterate,
quickly
and
kind
of
add
new
support
on
there
as
well.
B
B
You
know
the
other
kind
of
minor
point
is
there's
a
lot
of
configuration
that
I
think
you
need
to
be
able
to
run
these
hooks.
So
what
we
found
canisters
people
want
to
run
them
in
different
places,
with
access
to
different
secrets
and
config
maps
and
things
you
are
potentially
limited.
You
know
it's
easier
to
specify
additional
configuration.
If
you
have
a
separate
API
object,
you
know,
and
obviously
you
can.
You
can
continue
to
add
more
and
more
fields
to
the
the
annotations
but
I
think
that's
that
you
just
it's
just
a
hire.
B
B
You
know
like
typically
to
really
back
up
an
application.
You
need
both
the
quiesce
on
quiesce
commands,
as
well
as
obvi
snapshotting,
but
potentially
other
stages
as
well,
for
example,
for
more
advanced
use
cases.
Maybe
you
need
to
you
want
to
discover
some
information
about
the
after
you're
backing
up
figure
out.
What's
the
secondary
execute
specific
database
commands,
you
know,
for
example,
on
freeze,
replication,
etc
scale
down
pods
scale
out
pause.
B
It's
all
those
kind
of
those
kind
of
things,
I,
don't
think
I
think
we
should
focus
on
for
either
proposal
is
building,
is
choosing
the
primitives
that
we
need
to
add
things
right.
So
one
thing
we
mentioned
earlier
was
it'd
be
great.
If
we
could
have
a
primitive
that
we
call
the
quiescent
quiesce
the
volume
quiescent
whoo.
Yes,
yes,
I
think
you
know
in
either
case
like
we
wouldn't
be
able
to
walk
out
of
this
with
everything
we
need
to
be
able
to
take
backups.
A
A
A
You
all
you
are
talking
about
different.
Are
you
talking
about
different
use
case,
not
the
backup,
or
are
you
still
talking
about
backup
here?
When
you
say
one
size
does
not
fit
all,
because
we
are
trying
to
you
when
we
do
continually
fire,
we
are
trying
to
make
it
more.
You
know
a
general
approach.
You
can
kind
of
send
signals
to
the
part.
So
is
that
what
are
you
trying
to
describe
a
different
use
case
or
I'm?
Not
getting
what
you
mean
here?
No.
B
I
I
didn't
make
that
clear
and
I
think
you're
right.
I
did
sneak
a
another
use
case
in
here.
I'll
just
give
you
a
quick
overview
of
kind
of
the
background
of
what
what
I
mean
by
migration.
You
know
when
we
talk
to
our
customers.
We
actually
see
a
different
use
cases
for
backup
and
restore
there's
kind
of
you
can
do
like
in
cluster.
Even
within
the
same
namespace
back
at
the
restore
you
can
do
like
cross,
namespace,
backup
restore
you
can
do
between
different
clusters
form
or
the
dr
use
case.
B
A
B
I
say:
migration:
there
I'm
actually
talking
about
the
far
end
of
that
use
case
right,
where
you're
actually
murdering
a
few
different
clusters
that
have
maybe
have
different
versions
and
what
we
found
for
those
type
of
upgrades
that
customers
actually
preferred
different
workflows.
So
you
know
the
most
recent
one
working
with
wanted
to
kind
of
scale
down
every
application
copy.
The
data
over
which
you
know
it's
kind
of
like
a
you
know
the
way
that
they
would
actually
quiesce
the
volumes
right
by
doing
the
scale
down
and
unmounting
the
the
volumes.
B
A
Right,
but
those
probably
does
not
just
because
I
think
both
proposals-
we
are
talking
about
hardly
wrong
the
command
inside
the
container
right,
but
some
of
those
steps
are
probably
you
know
some
other.
Some
other
comments,
that's
not
necessarily
running
inside
a
container.
Some.
Some
other
type
of
comment
comments
that
you
need
a
wrong
way.
Yeah.
B
No
absolutely
I'm,
and
you
know
I,
think
I
think
it.
What
you've
done
with
both
proposals
is
really
good,
where
you're,
focusing
on
solving
kind
of
you
know
solving
you're,
implementing
a
primitive
right,
it's
something
that
would
be
composed
as
part
of
a
larger
process
to
do
even
backup
or
or
migration
or
these
other
kind
of
use
cases.
B
A
B
You
know,
and
the
other
just
kind
of
tidbit
I
can
share,
is
we.
We
typically
see
people
wanting
to
modify
these
themselves
so
I
have
an
example
in
the
and
some
Docs
that
list
a
bunch
of
blueprints
said:
we've
open-sourced
and
what
these
are
really
kind
of,
like
it:
a
base
for
people
to
modify
themselves,
because
what
we've
seen
with
every
customer
is
that
they
they
really
want
to
do
some
of
their
own
custom
integration.
B
You
know
whether
it's
how
they
consume
configuration
for
their
apps
that
they've
modified,
even
their
their
column,
charts
themselves.
You
know
you
have
to
maybe
modify
the
blueprints
as
well,
which
I
actually
think
that
point
does
kind
of
lend
towards
the
CR
approach,
because
you
can
have
you
can
separate
out
the
quiesce
or
Backup
approach
from
from
the
application
definition
itself.
B
You
know
I
think
we
included
the
serious
because
I
wanted.
You
know
I'd
love
to
dig
into
kind
of
the
problems
with
things
and
so
I'd
love
to
doing
the
problems
with
the
security
concerns.
I
think
they
are
definitely
valid.
You
know
the
the
high
level
it
from
my
understanding
is
basically
that
it's,
we
don't
want
to
give
permission
to
a
controller
to
exact
in
any
pod
right.
You
know
one
I
have
one
controller
running
in
native
space,
which
or
maybe
initially
asking
people
to
deploy
themselves.
I
think
it's
a
pretty
broad
permission
to
me.
B
B
A
A
Teamwork
group,
so
I
think
the
use
cases
that
we
have
been
focusing
on
our
clients
in
unquiet
scene
for
taking
a
backup
and
snapshots,
but
we,
it
will
be
also
good
if
we
can
find
more
use
cases
but
I
think
it's
probably
going
to
be
tough,
since
that's
our
focus,
those
so
I
mean
the
right.
Now
we
still
have
not
already
decided
which
way
to
go,
so
we
are
exploring
the
container
notify
approach,
but
we
did
get
pushback
from
signal
because
they
basically
don't
want
that
anything
complicated
to
the
the
couplet
right.
A
So
if
we
add
this
one,
that's
a
lot
of
complicated
code
changes
there.
So
that's
the
thing
so
I
think
for
for
this
team,
or
for
what
do
we
need
to
have
some
very
strong
use
cases
you
convince
them
that
this
is
solutely,
something
that
we
have
to
do
you
this
way.
I
think
it's
it's
difficult.
I
mean
there
are
so
that
there
are
pros
cons
from
either
side
wave
way,
no
matter
which
approach
we
use
so
I
would
say
that
yeah.
A
We
have
not
really
made
the
decision
on
which
way
to
go
yet,
but
we
will
continue
to
explore
the
container
notify
approach
because
well
the
extreme
cook
one
since
we
already
get
a
cap,
that
is
margin
right.
So
this
is
the
community
fights
this
another
proposal
that
we
just
want
to
see
whether
we
can
persuade
the
signal
settlers
or
otherwise
what
is
is
there
are
only
two
I,
don't
really
see
middle
ground
here,
yeah,
so
I
think
it's
very,
very
important
to
get
feedback
from
you
guys.
A
B
That's
that's
actually,
a
really
good
question
and
I
think
what
you're
asking
is.
Are
there
different?
Should
we
consider
different
approaches
depending
on
the
specific
use
case
we
want
to
add
so?
Is
application
quiescing
different
than
volume?
Quiescing
I?
Think
that's
a
really
good
insight.
You
know
I
think
that
one
world,
maybe
for
example,
that
we
can
solve,
like
the
application,
quiescent
use
case
separately
from
the
volume
quest
in
use
case.
B
You
know,
I
think
what
I
would
love
is
actually
the
ability
to
do
have
an
API
that
is
standardized
for
blowing
quiescing
and
I
honestly
think.
Actually,
the
container
little
fire
is
a
great
proposal
for
that,
and
so
it
may
not
be
an
either/or.
You
know
maybe
a
false
false
comparison.
I
do
feel
that
the
execution
hook
is
potentially
a
better
model
for
application,
quiescing
kind
of
for
the
reasons
I
mentioned
earlier
right.
B
This,
the
level
of
indirection
I,
think
provides
a
lot
of
value
for
for
this
level,
and
you
don't
know,
and
you
don't
need
the
the
escalated
privileges
you
need
for
the
line
dressing
with
the
and
you
a
kind
of
more
stricter
reviews
and
design
process.
You
need
for
winning
container
notifier,
so
so
I
do
definitely
believe.
There's
a
world
where
both
these
things
of
this
four
four
separate
use,
cases
and
I
think
that's
great
right.
These
are
both
kind
of
primitive
Assad
that.
E
Was
that
was
what
I
walked
away
with
cuz
tonight?
I
draw
the
parallel
to
two
NetApp
head
snap,
drivin
snap
manager
and
snap
drive
basically
did
what
the
container
notifier
does,
which
is
sort
of
the
storage
level
quiescing,
and
then
snap
manager
actually
did
the
app
integration,
and
some
customers
wanted
one.
Some
customers
wanted
both,
but
there
were
different
beasts
so
and
and
I
get
your
point,
that's
getting.
What
approved
is
hard
getting
to
approved
is
probably
harder.
Oh.
A
A
If
we
have
a
way
to
run
commanding
the
container
I
think,
then
we
should
be
able
to
do
that
for
both
application,
choirs
and
also
what
in
choirs
I'm,
not
quite
I'm,
not
sure
what
exactly
you're
talking
about
that
you
said
the
net
up.
Has
this
approach
that
it's
more
like
continually
fire
I,
wouldn't
expect
they
change
the
pot
definition
theories.
E
So
I
was,
it
was
a.
It
was
a.
This
was
a
2002
reference
because
I'm
old
now
wooden,
we
added
support
for
saying
what
we
did
is
we
had
a
snap
drive
which
knew
how
to
quiesce
the
the
Lund,
and
then
we
had
a
snap
manager
that
knew
how
to
quiesce,
oracle
or
or
or
sequel,
or
something
like
that
and
the
two
would
work
in
combination
so
snap.
The
snap
manager
would
trigger
a
snap
drive,
but
we
had
some
customers
who
just
said
no
I
just
want
the
snap
driver.
E
E
E
E
E
Consistent,
it
flushes
that
out
I
again
at
the
application
level.
I
may
want
to
say
again
now
that
I've
done
this
back
up.
I
can
truncate
the
log
free
up
space
so
that
the
log
doesn't
grow.
So
there
are
things
that,
even
if
you
get
sort
of
a
storage
consistent
backup
that
doesn't
necessarily
solve
the
application
level
problem
directly
it
it
helps.
It
know
which
was
I,
think
where
tom
was
going,
but
it
doesn't
doesn't
give
me
actual
application.
Backup
I
was.
A
Thinking
they
I
was
I
want
to
think
they
wouldn't
need
another
separate
API.
And
what
did
you
do
that
they
can
just
do
that
class
before
they
take
a
snapshot
inside
their
storage
I'm,
just
saying
I,
don't
think
they
need
a
separate
API
from
kubernetes
level.
That's
guessing
but
of
course
I.
Don't
think
of
it.
I
don't
know
the
whole
story.
So
as
a.
E
C
Yeah,
that's
similar
to
like
performing
FS
free
that
fsync
before
taking
that
part
of
the
volume
Oh,
Tom
and
I
had
another
quick
point
about
this
follow-ups
that
you
had
that.
The
first
of
all,
thanks
for
putting
the
comparisons
out
between
the
two
proposals
and
I
I
have
a
feeling
that
we
probably
might
need
both
proposals
and
here's.
Why
the
issue
with
the
execution
hook
proposal
was
the
implementation
or
the
security
aspect
of
it.
C
However,
if
that
we
can
elevate
alleviate
by
using
the
container
notifier
proposal
and
instead
of
the
user
having
to
update
the
part
spec,
we
can
have
the
execution
hook.
Controller
update
the
part
spec
to
add
the
commands
and
the
cubelet
runs
it,
and
that
way
the
execution
hook
no
longer
has
to
run
commands
in
the
part.
A
B
B
A
C
With
the
with
with
what
I,
with
the
proposal
that
I
just
made
the
execution
hook,
controller
will
not
have
privileges
to
run
commands
inside
containers.
It
will
only
have
privileges
to
update
the
parts
back
with
the
with
adding
the
containers
notifier
and
the
container,
and
the
cubelet
was
still
responsible
for
running
the
command
inside
the
container.
A
D
A
Then
then
we
are
adding
another
thing
that
is,
that
is
an
extra
control
that
is
modified
well,
but
but
I
think
that's
the
the
concern
here
is
actually
not
that
you're
busy
trying
to
have
like
automatically
to
define
the
pod
or
something
it's
almost
like.
You
were
trying
to
do
that
so
because
so
the
so
the
the
concern
for
container
notifier
is
because
we're
trying
to
add
too
many
things
in
it
and
then
signal
they
are
having
concerns
yeah.
So
so
is
this
is.
D
A
C
So,
with
the
cry
proposal
alone,
the
user
will
have
to
go
update
the
parts
back
to
have
took
the
container,
notify
or
two
to
do
its
thing.
But
with
with
what
I
was
saying,
is
the
execution
hook?
Controller
would
instead
of
running
the
commands
inside
the
container,
would
update
the
parts
back
to
add
the
content
of
notifier
block
and
the
cubelet.
A
A
B
I
think
it,
so
let
me
propose
an
alternate
because
I
think
there
is
still
an
issue,
a
security
issue
with
the
execution
controller
itself
that
that
doesn't
solve.
If,
if
you
want
to
go
down
the
approach
of
having
the
execution
hook,
controller,
you
know
invoke
things
through
the
container
notifier
I
would
actually
separate
out
modifying
the
pod
from
the
actual
execution
path
itself.
So,
for
example,
we
already
there's
already
a
mechanism
or
a
mutating
web
hook
which
can
update
pods.
B
B
Have
them
each
chain,
webhook
controller,
go
and
add
those
to
you
know
the
Yasir.
It
selects
and
then
limit
the
surface
area
of
what
the
x
you
should
hook.
Control
can
do
by
only
invoking
you
know,
container
notifiers
on
this
large
subset
of
apps
right.
You
can't
actually
go
and
modify
because
then
anyone
in
cells
can
in
fact
run
the
execution
hook.
Controller
and
you
kind
of
you
know
the
sidestep.
The
issue
with
with
the
broad
permissions
for
the
execution
of
control.
Would
that
make
sense,
association.
C
B
Exactly
because
there's
really
two
security
issues,
I
think
on
the
table,
one
is
the
really
broad
granularity
permissions
of
the
execution
of
controller
I.
Think
you
know,
we
kind
of
agree
is
too
broad
and
the
other
problem
is
the
pushback
from
signo
Don
on
the
container
notifier
I
think
Jing
is
pointing
out
the
the
push
the
security
concerned,
pushback
from
sig
node
I,
think
what
you're
proposing
is
addressing
the
the
granular
nutritions
on
the
execution
hook:
controller
right,
I,.
C
A
There
was
so
there
is
a
1/1
fame
in
the
content,
notifier
that
was
saying
that
we
want
to
have
some
like
well-defined
notify
names.
Just
like
the
example
like
CPU.
Those
are,
you
know
reserved
word
right,
rein,
things
like
that.
So
if
we
have
a
few
like
maybe
choirs
and
then
there
are,
others
are
customized,
but
I
don't
know
if
that
would
be
enough
because
it
seems
like
we
have
so
many
different
use
cases
just
for
quest
and
then
each
one
is
different.
A
So
if
we
have
this
well
we're
no
names,
I
don't
know
if
that
would
help
at
all,
because
I
think
that's
one
thing,
I'm
thinking
signal
would
be
interested
in.
If,
if
we
said
oh,
we
only
need
two.
Let's
say
we
only
need
to
store
the
three
kind
of
notifier.
Why
is
critic
where's?
The
other
one
is
closer
I,
don't
know
like
maybe
shut
up
or
something
right,
so
I'm
just
coming
up
with
something,
but
then
to
acquiesce.
There
are
so
many
different
definitions.
So
that's
why
I'm
like
I,
don't
think
that's
possible.
A
A
Yeah,
so
it's
a
custom
node,
so
we
do
have
a
customized
to
you.
Can
you
can
have
a
customized
community
fire
I
was
just
saying
that
if
just
ideally,
if
it's
possible
for
us
just
say,
okay
I
just
have
one
this
one
or
three
wait.
Three
three
quiet
thing,
then
you
can
have
like
well
no
names
and
then
well
be
easier
to
persuade
them.
Probably
but
I,
don't
think
so.
A
B
Know,
saying
I
think
that's
a
pretty
good
point
if,
if,
for
example
like
we
had,
the
ability
of
you
know
Mycoskie
how
on
if
we
get
the
ability
right
now
to
execute
FS,
freeze
unfreeze
and
that
alone,
that
would
actually
unblock
a
lot
of
use
cases
here,
and
you
can
manage
that
API
specific
me,
rather
than
making
kind
of
general
purpose.
One.
A
A
It
could
be
a
little
tricky
if
you
have
volumes
because
for
each
one
in
each
around
us
and
then
you
need
to
know
the
exact
path
right,
so
I
think
could
become
a
little
tricky,
but
it
is
possible
so
so
time
I
not
sure
your
concern
is
it's
just
it's
difficult
to
make
this
precise
comment
or
because
I
think
you
should
be
possible.
If
you,
you
can
already
run
the
exact
anyway
right
for
your
other
things.
If
you,
if
you
know
what
part
and
what
volume
the
path,
then
you
can't
run
that
I
think
yeah.
C
A
So,
like
a
special
okay,
so
you
do
have
to
have
like
a
special
part:
okay,
yeah,
so
we
we
did
actually
I,
think
that
is
not
here.
No,
we
actually
we
actually
talked
about.
There
is
another
way
which
is
to
do
the
scene.
The
CSI
have
some
some.
Basically
it's
like
a
part
of
your
note.
Plugging
you
run
this
yeah,
but
I.
Try
that
I
I
just
think
this
is
you?
Can
you
can
really
get
into
really
bad
situation?
If
you
don't
do
this
right?
Oh
definitely,.
A
Because
he
could
freeze
the
whole
thing
right.
That
I
was
because
I
was
trying
that
I
screwed
up
planning
my
setup
because
of
that
so
I
think
I
really
feel
hesitant
of
adding
this
as
a
CSI
comment,
we
deep
talk
about
this
one
some
time
ago,
actually
so
that
it
was
like
one
one
thing
that
seems
a
note
plug-in.
You
know
that
that
part
is
already
or
it
has
privilege
it
already
doing
other
things
like
it
has
to
do
month
and
all
of
that
right.
A
A
A
Right,
yes,
that's
yeah,
so
those
we
yes,
those
we
do
have
yeah.
So
there's
a
timeout
for
that.
But
the
problem
is
once
you
run
those
command
if
it's
really
frozen,
there's
no
way
for
me
to
unfreeze
it
right.
So,
like
FS
freeze,
if
you
you
screw
up,
you
know
you.
You
run
that
on
the
route,
then
you
you
know
it's
frizzy
or
homie
am,
and
you
know
certainly
no
way
that
you
just
have
to
reboot
right.
So
there
are
things
like.
F
A
C
A
B
Won
so
I
stood
seconds
and
there's
a
good
common
right.
You
know
what
definitely
make
for
progress
and
I.
Think
having
a
good
support
here
would
definitely
help
partially,
because
it
is
actually
hard.
We
got
things,
and
so
this
will
be
a
dangerous
command
and
it'll
be
more
dangerous.
You
know,
there's
something
very
dangerous
right.
You
wanna
make
it
very
very
safe.
B
The
the
thing
that
the
gene
was
saying
you
know
I
want
to
dig
into
a
little
bit.
If,
if
we're
worried
about
FS,
freeze
and
unfreeze
being
too
broad
right,
it
seems
that
we
need
a
specific
mechanism
which
has
all
the
safety
hook
safety
parameters
around
it,
rather
than
making
a
general
mechanism
that
can
run
FS,
trees
and
Fried's,
as
well
as
many
other
things
right,
because
the
general
mechanism
wouldn't
allow
you
to
build
in
these
like
there
specific
safety
measures.
A
Maybe
but
I
yeah,
so
if
well
so
what
I?
Actually
what
I
was
trying
to
do
was
actually
I.
Think
if
you
make
it
very
special
like
okay,
you
you
do
it
like
a
CSI
matter,
then
that'll
be
very
swish,
but
I
think
that
actually
will
be
dangerous,
that
that
way,
because
it's
more
it's
like
it's
always
running
because
I,
don't
think
I,
don't
think
that
this
is
required
for
all
the
use
cases,
because
I
know
people
holding
back
have
immersed
or
they
don't
really
run
this.
A
They
think
they
just
don't
think
they
don't
really
run.
Life
is
freeze
but
yeah,
but
there
are
all
but
I
do
see
some.
There
are
some
people
saying
they
do
have
the
wrong
list,
maybe
like
maybe
EVs
they
have
some
comment
saying
you
should
run
this
right,
but
not
everybody
so
I,
don't
really
I,
not
sure.
So
maybe
that's
actually
should
come
from
you.
You
guys
that
commercial
vendors
is
there
something
you
know.
How
important
is
this
particular
comment
right
so
so
right
now
we
I
think
we.
A
C
You
want
to
involve
the
application.
In
some
day
we
do
when
I
say
application
here,
the
container
which
is
running
the
run
application
here,
just
not
the
DB
alone
right
that
way
putting
it.
What
we
do
is
we
try
and
do
it
in
two
fold.
Oneness
have
a
kind
of
a
pre/post
hook
before
and
after
snapshotting
kind
of
where
they
can
register
and
run
their
own
application.
C
You
can
draw
an
analogy,
even
in
the
virtual
machine
world,
in
the
virtual
machines,
if
you
run
a
Linux
VM
on
cloud,
and
if
you
want
a
cloud
backup
solution
for
your
virtual
machine
workload,
you
do
a
pre
postscript
kind
of
a
framework
where
they
can
optionally
register
those
or
they
can
skip
that
and
at
least
have
a
FS
freeze
and
thaw
being
done
and
doing
that
time
frame.
You
do
a
volume
snapshotting
right,
the
where
I'm
going
to
is.
C
If
you
only
have
to
support
the
top
five
workloads,
like
my
sequel,
Cassandra,
etc,
you
can
do
a
DB
level,
backup
script.
That's
where
you
don't
do
a
volume
snapshot
in
there.
What
you
do
there
is
basically
take
the
MDF
file
or
a
DB
pack
file,
which
is
provided
by
respective
backup
engines,
but
backup
engines
are
not
the
only
number
of
use
cases
that
are
out
there.
When
we
look
at
container
view.
A
B
Great
point,
or
not
all
applications
are
crashed
in
system,
and
so
you
still
need
word.
We've
even
found
with,
like
the
databases
that
we
still
use
long
snapshots,
because
often
storage
provider
is
much
more
efficient
right.
They
they
integrate.
They
influence
their
own
very
efficient
change,
while
tracking,
for
example,
correct
correct.
B
E
C
And
the
other
direction
I
see
from
backup
printers
and
also
the
customers
of
backup
Enders.
How
they
are
influencing
is
gone
are
the
days
where
the
DB
size
used
to
be
in
GBS
and
maybe
reaching
TVs
I
can
see
applications
where
the
databases
are
in
multiple
terabytes.
A
typical
example
could
be
SAT.
Hana
Hana
TVs
are
in
multiple
TVs
and
their
stream
based
backup
solution.
C
Traditional
backup
solution
will
not
work
and
where
they
are
tending
toward
says
to
I
your
pause
using
FS,
freeze
and
similar
technology
and
do
a
disk
level
snapshotting
and
bring
back
reattach
the
disk
and
make
the
because,
with
with
the
large
DD,
you
cannot
just
do
a
stream
level
backup
solution
sometime
that
will
not
work
at
all.
It
will
take
hours
yeah.
F
A
F
A
C
A
If
that's
actually
that's
why
we
asked
her
Tom,
you
know
look
out
of
this.
What
would
be
helpful
is
to
like
have
more
use
cases
to
say
that
okay,
this
is
really
important.
Really,
we
really
need
this
in
in
couplet.
You
know
so
I
think
we
talked
about
some
of
those
those
today,
but
if
you
have
anything
else,
you
please
maybe
add
you,
this
dog
I
think
that's.
E
A
E
A
A
Go
back
to
all
right,
so
I,
just
we
only
have
two
minutes,
not
just
a
particular
update.
So
last
week,
last
Friday
I
rent
another
meeting
suffering
meeting
to
review
the
volume
group
cat
is
that
you
know
I,
don't
want
to
use
the
our
data
cache
data
protection
meeting
time
for
entirely
just
for
that,
so
the
I
have
the
link
there
to
the
tap.
The
main
thing
that
we
discuss
today
is
what
is
the
right
way
to
create
a
hoarding
group.
There
are
because
there
are
different
vendors.
A
They
support
even
things,
so
we
decided
that
we
don't
want
to
support
an
immutable,
modern
group,
meaning
that
once
they
wanna
go
biz
created,
you
should
still
be
able
to
modify
it,
meaning
adding
new
volumes
to
it
or
removing
bottoms
falling
it.
But
we
don't
want
to
say
once
you
created
the
running
group
that
becomes
immutable.
You
cannot
change
the
membership
of
the
columns
anymore,
so
I
I'm
supposed
to
send
an
email
to
the
maintenance
just
to
see
if
there
was
any
people
who
have
some
strong
use
case
that
you
support
this.
A
If
we
want
to
support
both
it's
a
it's.
A
very
difficult
then,
basically
meaning
that
user,
who
have
to
have
at
least
two
different
type
of
protocol
classes,
to
say
why
is
multiple
the
other
is
interval,
make
the
probability
difficult.
So
yes,
so
that's
the
main
main
thing
that
we
discussed
in
in
that
meeting
yeah.
So
if
you
have
any
comments
about
this,
shout
me,
but
otherwise
I
will
need
to
send
send
out
a
you
know
about
others.
A
So
I
think
so
tom
has
this
document
right.
If
you
have
anything,
you
think
that
helped
that
can
help
us
to
say:
why
are
you
saying
container
notify
is
the
right
approach,
what
it
can
help
us
or
even
if
you
have
some
other
additional
use
cases
that
we
can
reference.
You
know,
please
add
so
anything
that
you
think
can
help
promote
this
case,
because
we're
trying
to
persuade
the
signal
that
it
really
really
need
the
US
right.