►
From YouTube: Intro to Longhorn: Open Source Cloud-Native Distributed Block Storage Built On and For... Sheng Yang
Description
Join us for Kubernetes Forums Seoul, Sydney, Bengaluru and Delhi - learn more at kubecon.io
Don't miss KubeCon + CloudNativeCon 2020 events in Amsterdam March 30 - April 2, Shanghai July 28-30 and Boston November 17-20! Learn more at kubecon.io. The conference features presentations from developers and end users of Kubernetes, Prometheus, Envoy, and all of the other CNCF-hosted projects
Intro to Longhorn: Open Source Cloud-Native Distributed Block Storage Built On and For K8s - Sheng Yang, Rancher Labs
Longhorn is an Open Source Cloud-Native distributed block storage built on and for Kubernetes. It provides persistent storage support for any Kubernetes cluster with one-click installation. It also supports advanced features like the built-in incremental backup and across cluster disaster recovery of the data. Longhorn leverages Kubernetes to build an easy to use, reliable and powerful storage experience for the users. Join us to learn about the architecture of Longhorn, how it was built on Kubernetes, how to use it with your stateful applications, and see a live demo of the latest release!
https://sched.co/Wepi
A
A
A
One
thing
about
long
quark:
we
want
to
emphasize
this.
We
want
to
easy
to
be
easy
to
use
easy
to
be
outraged
and
have
a
great
user
experience
compared
to
and
compared
to
other
storage
software's.
We
feel
like
many
of
them
are
really
unnecessarily
complex
and
also
because
of
their
complex
architectures.
There's
many
technical
difficulties
and
the
difficulty
in
operations
so
compare
Longhorn
to
the
other
legacy
story.
Software's.
A
So
let's
talk
a
bit
a
little
bit
about
the
history
of
long
form.
In
fact,
we
started
project
worried
early
and
is,
if
you
remember,
the
rancher
labs
start
at
September
2014
and
which
is
also
the
start.
First
start
at
time
for
long
form,
the
first
iteration
we
don't
really
sing
out
of
box.
We
think
okay
storage
have
to
be
complex.
You
have
to
be
very
high-performance
and
they
probably
need
to
be
writing
in
C++.
A
So
that's
basically
what
we
go
with
the
first
iteration,
but
since
the
second
iteration
we
we
realize
that
we
can
do
since
it's
much
a
simple
way
by
the
steel
over
the
sting
stability
and
as
still
over
a
very
powerful
experience.
So
on
the
second
iteration,
we
decided
to
write
a
long-running
goal
and
order,
and
also
in
order
to
track
more
traction
on
the
community.
We
decide
just
building
the
from
scratch
and
make
it
open
source
with
Apache
2
license.
A
But
at
the
time
no
more
was
building
with
the
doctrine
in
mind,
as
at
the
time,
kubernetes
is
basically
competing
with
mesosphere
tortoise
worm
and
rancher
kettle.
So,
but
a
few
years
later,
around
the
end
of
2017
and
2018,
we
saw
that
kubernetes
is
basically
taking
off
and
at
the
September
2017
Rancher
labs,
as
has
published
Rancher
mm
OH,
which
is
the
totally
different
product
from
the
branch
of
106
which
you
use
in
the
Rancher
cattle
and
the
rancher.
A
That
is,
the
multiple
class
kubernetes
class,
the
management
platform
for,
of
course,
kubernetes
and
things
rancher
pivoted
and
the
rancher
laughs
other
project.
We
also
see
that
it's
really
necessary
for
us
to
change
so
a
master,
the
iteration.
We
are
still
writing
a
goal,
but
we
start
in
in
Longhorn
repealed
the
local
control
playing
I'm
over
kubernetes.
A
So
since
then
we
have
a
child
former
releases.
We
have
about
a
nine
hundred
stars
in
key
kept
now
and
the
right
side.
You
can
see
that
the
trained
of
our
kid
how
the
stars
is
starting,
pretty
slow
at
the
bottom
and
we
just
before
we
join
the
CN
CF,
the
attraction
getting
up
and
after
we
joins
the
n
CF.
The
things
really
a
little
bit
of
looks
like
a
taking
off
I
hope.
That's
the
train.
A
So
in
the
latest
release
staring
paradox:
eventhough
we
have,
we
still
offering
the
enterprise-grade
distributed,
block
story,
software
for
kubernetes,
and
we
support
warning
snapshot
on
a
backup
and
restore
the
different
between
a
snapshot
and
a
backup
is
that
snapshot
you
take
will
be
kept
inside
your
kubernetes
cluster,
but
you
can
backup
your
snapshot
into
the
third-party
like
the
s3
or
NFS
server.
So
even
your
whole
cluster
is
somehow
suffers
some
disasters.
A
You
still
have
your
data
kept
safe,
so
we
support
storage
tag,
feature
for
node
and
a
disk
selection,
so
you
can
select
which
disk
which,
for
example,
some
with
SSD
and
some
a
spinning
disk.
If
you
want
more
IO
ops,
you
can
put
your
long
home
volume
based
on
the
SSD
disk
or
you
can
select
to
say
using
spin.
It
is
for
a
slower
workload.
A
Many
users
told
me
the
first
thing
we
call
them
eyes
on.
Longhorn
is
because
that
the
intuitive
UI,
so
by
default,
Longhorn
install,
is
UI
and
it's
open
source
is
of
course
open
source
and
we
are
we're
finding
that
UI
really
helps
user
to
understand
what
is
going
on
in
their
system,
and
many
users
also
give
us
feedbacks
and
how
to
improve
the
usability
on
the
UI
longhorn.
As
any
other
cloud
native
software,
we
support
one-click
installation,
you
can
install
own
horn,
use
one
map,
llamo,
oh
hell,
min
store,
or
you
can
just
elector.
A
A
A
So
for
now,
I'm
going
to
select
storage
project,
which
is
basically
like
just
Rancher
concepts,
saying
that
is
the
name
space
group
and
I'm
going
to
select
app
apps
and
lunch
and
I
will
find
long
phone
here
and
in
the
backend
is
basically
it's
essentially
a
helm
charge
and
I
make
sure
that
version
is
at
0,
7.0
and
I.
Don't
need
to
set
for
anything
I,
just
click
Launch.
A
So
now
we
can
go
back
to
take
a
look
at
the
workload
and
you
can
see
that
the
local
manager,
UI
and
private
employer
there's
in
the
update
in
the
deploying
process
and
one
by
one
they're
starting,
and
you
can
see
that
this
one
another,
the
moose
app
called
engine
image
and
some
instance
manager
are
starting
up.
Those
are
old
lankans,
components
and,
finally,
the
driver
deployer
will
see
that
local
manager
is
available
and
the
start
and
deploy
in
the
CSS
driver
so
long
for
Cameron
is
in
the
latest.
A
A
We
are
going
to
take
a
look
at
the
loan
for
you
I'm,
so
the
first
page,
you
see
that
it's
a
dashboard
and
the
life
side
and
the
left
side
is
the
warning
count
which
we
don't
have
any
right
now
to
me,
the
side
that
you
can
see
the
real-time
storage
usage
in
your
system
and
your
hope
across
your
whole
cluster
and
the
right
side.
You
say
that
was
the
note
status
in
this
cluster.
A
So,
in
the
note
page,
this
is
the
place
you
can
customize.
How
do
you
like?
Don't
want
to
use
your
storage,
or
do
you
want
your
Longhorn
to
only
use
a
part
of
storage
in
one
disk
or
a
sari
in
the
OneNote,
or
you
can
add
more
disk
into
here.
So,
for
example,
you
can
see
that
because
I
did
no
configuration
by
default
Lauren's
using
the
wildly
branch
of
local
directory
I'm,
your
I'm,
your
host
and
your
node-
to
provide
a
storage
for
long
form.
A
Of
course,
this
is
a
changeable
and
we're
also
in
the
progress
of
change.
This
to
secretary
to
the
wildly
branch
body
belong
form
because
long
for
now
is
part
of
CN,
CF
and
no
longer
Rancher,
so
click
add
a
node
and
disk.
Now
you
can
choose
that
if
you
want
any
willing
to
be
scheduling
to
be
used
at
the
storage
on
this
node-
and
you
want
to
choose
that
storage
to
be
used
on
this
disk
and
also,
you
can
add
a
new
disk
here
to
add
a
new
disk.
A
So
the
in
the
come
in
the
settings
that
you
choose
this
back
a
target-
and
this
is
the
the
credential-
was
a
store
in
this
Cooper.
Ninety
secret,
its
name
in
s3,
touch
secret
and
once
you
click
safe
long
horse,
you'll
be
to
recognize
the
content
in
basic,
treats
and
connect
to
the
s3
and
the
read
the
list
of
the
word
existing
backup
from
that
backup
store.
So
ok!
So
now
we
kept
the
list
of
the
backup
here
and
that's
a
start
using
old
form.
A
Let's
see,
ok,
so
those
two
mornings
wooden
tiered
moment,
equal
4
and
now
they
are
starting
and
just
rated
and
and
in
the
attaching
and
attach
the
stage
now
on,
were
saying.
Okay,
that
Shukra
newly
creates
the
warning
that
looks
very
healthy,
so
they
changed
attitude
healthy,
saying
in
this
page,
you
can
see
the
information
to
major
information.
You
need,
for
this
volume
likes
what
the
woman's
name,
which
is
include
knighted
cases.
If
you
provision
any
volume
using
storage
costs,
you
normally
will
just
see
that
it's
a
PVC
plus
some
UUID.
A
It's
doesn't
really
make,
doesn't
make
much
sense
for
you
to
associate
a
best
with
the
workload.
But
in
fact,
if
you
look
at
here,
we
can't
attack
the
status
of
the
kubernetes
and
we
can
care
information
from
that.
We
know
that
was
the
pv
name
and
was
the
PVC
name
and
what
namespace
this
warning
being
used
tube
and
which
workload
this
warning
has
been
used
at
you.
So
this
is
mariadb,
and
this
Mirage
IVs
running
as
data
set
and
their
code
name
is
WordPress,
does
married
to
beta
Z
the
test
error.
A
A
But
if
you
want
to
have
a
backup
like
say
just
take
this
volume
outside
cluster
and
make
sure
that
is
a
safe
there
and
you
can
create
backups
for
the
backup,
so
you
can
add
more
label
to
it
and
just
leave
it
empty
for
now
and
the
backup
they
see
in
the
creating
I
believe
in
there
and
talk
about
it
later.
So
we
also
provide
a
feature
to
create
a
recurring
snapshot
and
a
backups.
A
A
Just
set
it
every
minute,
of
course,
this
is
now
practical
in
the
production.
Don't
try
it
yourself,
yeah.
Of
course
you
can
try
the
TVs
just
for
testing
of
a
phone,
so
the
backup
was
set
and
let's
take
a
look
at
its.
What
price
was
really
up
so
because
I
started
from
the
apps,
so
I
can
just
be
right.
Click
on
this
link
and
they
will
bring
me
to
do
a
press
page
yeah.
It
looks
a
little
bit
too
simple,
but
I
was
told.
This
is
to
2019
style.
A
A
Published
and
then
we
do
the
post
now,
the
post
is
the
showing
you
off
and
make
sure
it's
worse.
Yes
right
so,
but
how
do
you
know
that
the
stage
has
been
stored
in
long
form?
In
fact,
is
that
neither
question
can
be
if
okay?
So
this
is
the
one
cluster?
What
if
I
lost
this
cluster,
but
I
do
want
to
have
a
highly
available
service.
A
So
we
do
over
the
disaster
recovery
feature,
which
means
that
I
can
have
a
backup
copy
of
the
of
the
volume
inside
another
car
faster
and
wait
waiting
to
be
activated.
So
in
order
to
do
that,
I
already
have
this
demo
to
class
or
set
up
and
also
in
the
storage
project.
I
have
no
more
already
installed
here,
and
this
is
the
internet
faster.
A
So
here,
if
I
click
here,
because
you
see
non-porn
story
not
clustered
this,
it's
nosing
here,
there's
no
volumes,
but
the
backup
store
has
already
set
up
beforehand.
So
we
should
able
to
see
the
list
after
just
the
backup
we
just
created
there.
So
you
can
see
the
latest
tube
back
up
the
latest
tube
back
up
volume
big
for
ages.
Two
and
three
minutes
ago
that
means
that
they
are
just
traded
in
this
demo
from
the
previous
cluster.
A
So
in
order
to
use
that
they
are
work,
okay,
that's
one
thing:
I
want
to
show
you
that
the
backup
not
only
contain
the
data,
but
also
contain
the
metadata
of
the
application
using
the
volume
at
the
time
of
the
backup.
So,
for
example,
this
one
I
have
can
see
the
label
here,
and
this
volume
is
what
used
by
the
WordPress,
and
you
can
see
the
original
WordPress
information
here,
and
this
helps
us
to
decide.
That
was
this.
A
So,
but
if
that's
recovery,
volume
in
Longhorn
is
just
a
wordham
that
is
keep
stay
active,
but
what
I
mean
to
you
is
just
incrementally
restore
from
what
we
watched
SP
from
the
backup
store.
So
this
one
is
the
word
press
and
nother
one.
We
should
be
them,
of
course,
Mary
DB
and
I'm
going
to
create
this
as
recovery
warning
of
data
as
well.
A
A
All
right,
so
those
two
volume
up
around
me
and
actively
restoring
from
what
they
seem
from
in
the
back
of
store.
So
how
can
proof
that
I
just
go
back
to
this
WordPress
and
I
would
going
because
this
warning
was
created
before
so
now,
I'm
going
to
just
change
this
WordPress
and
and
I
just
add
something
more
to
it.
A
A
Okay,
so
we
just
waiting
right
now
yeah.
If
anyone
has
some
questions,
I
can
answer
a
few
of
them
now
and
also
you
can
see
that
the
last
backup
was
created
about
minutes
ago
and
basically
just
now
I'm
wanting
this
and
make
sure
when
this
number
change
to
a
few
seconds
ago
means
a
new
backup
was
graded.
B
A
A
Yeah,
so
in
fact,
because
the
the
synchronization
is
the
single
replication
synchronized,
it's
very
hard
to
do
because
network
latency
is
really
hard
to
do
the
cross
region
replication.
So
this
is
why
we
have
this.
We
are
using
backup
store
as
a
median
cube
to
deal
with
this
problem
and
also
because
the
long
horn
is
the
block
device
is
read,
write
it
rate
or
at
once
you
can
not
able
to
write
on
a
different
side.
A
So
I
think
this
that
this
current
approach
is
more
simpler
and
easier
to
TTT,
always
alright,
so
I
guess
in
the
meantime
the
backup
should
be
already
done.
So
now,
let's
activate
those
dr
volumes.
Okay,
let's
just
close
those
queue
in
the
previous
clusters
for
now
and
assuming
that.
Well
somehow,
you
lost
the
cluster.
A
A
A
A
Put
the
same
password
enable
persistent
volume
and
you
notice
that
merit
if
it
doesn't
really
give
you
option
to
choose
which
warning
you
want
which
P
we
see.
If
you
want
to
associate
with
that,
because
merit
EB
is
the
staple
set
and
as
long
as
a
stable
set
name
and
it's
determined
and
the
warning
name
is
determined.
The
name
of
that
PVC
is
automatically
determined
like
say
the
pod
name,
Warner
name
Gerald,
Pro,
name
that
you
would
name
one.
So
there's
no
way.
A
A
B
A
A
A
A
So,
in
fact
it's
very
easy.
If
you
have
even
more
parts,
you
just
create
more
engines
and
replicas
and
connect
them
to
that.
You
make
sure
that
tribest
you
have
replicas,
reside
in
a
different
node
and
then
connect
to
them
and
provide
you.
The
warning.
This
approach
have
a
two
advantage
over
the
traditional
way.
The
first
one
is
because
ng
itself
is
like
just
a
simple
RAID
one
controller.
It's
really
easy.
It's
really
easy
to
program
and
also
is
always
resigned
in
the
stang
node
as
the
pot
running.
A
So
we
don't
really
need
you
have
a
controller
and
make
a
lot
of
logical
complex.
In
order
to
do
that.
So
another
thing
is
you
can
see
that
each
the
data
path
of
each
volume,
in
fact,
is
not
really
in
the
action,
which
is
everybody
else,
so
the
data
path
of
the
pod
one
using
a
wooden
one,
it's
a
different
from
the
one
in
tube
from
the
PERT
one
and
three.
A
A
As
another
part
of
the
Longhorn
is
the
Longhorn
manager
and
the
normal
control
playing.
So
in
this
case
we
have
a
kubernetes
cluster
and
they
want
to
create
a
warning
and
the
command
is
crafter
internally
used
in
the
CSI
interface
to
talk
to
the
loan
process.
They're
plugging
and
the
la
marseillaise
are
plugin
going
to
talk
with
the
local
manager
using
the
Longhorn
API,
so
local
manager,
orchestrate
all
the
volumes
and
the
follows
the
kubernetes
controller
pattern,
or
somebody
coded
operator
pattern
in
this
case
loan
for
manager
we're
going
to
create
a
new
warning.
A
Crd
object
inside
the
community
api
server
and
not
a
bunch
of
controller
inside
the
local
manager
are
mortgaging
the
API
servers
response
and
see
that
ok,
there's
a
new
CRT
was
created,
so
I
should
create
a
new
volume.
So
the
other
meant
that
the
manager
inside
long
form,
an
inside
of
a
manager,
are
going
to
create
normal
engine
and
the
replica,
and
then
they
were
former
Emporium.
So
if
they
are
more
required
to
create
more
volumes,
we
just
simply
follow
the
same
same
flow
and
another
important
part
of
long
horn
is
the
low
point.
A
Ui
normal
UI
is
talking
to
the
local
manager,
also
through
the
local
API,
and
is
acting
as
the
complement
of
the
kubernetes,
because
normally
the
kubernetes
they
stop
talking
give
you
the
ability
to
attack,
create
delete
attached,
detachment,
I'm,
not
volume,
and
they
are
snapshot
function
there,
but
they
still
in
alpha,
but
all
other
function
provided
by
Longhorn
are
implemented
through
the
loan,
for
you,
I,
like
snapshot,
backup,
know
the
management,
disk
management
and
old.
All
of
that-
and
we
are
also
planning
to
have
a
longer
CLI
very
soon
before
we
release
the
GA.
A
So
let's
talk
about
a
little
bit
about
our
currently
long-haul
community.
Since
Lauren
is
the
totally
open
source
project.
We
really
want
to
really
encourage
communities.
Engagement
currently
Longhorn
has
over
900
github
stars
and
we
hold
monthly
community
meeting
at
9:00
a.m.
Pacific
time
on
the
second
Friday
morning,
which
I
checked,
which
is
no
other
CNCs
meeting
code
at
the
time,
and
we
have
over
300
users,
which
combined
the
two
snack
channel.
A
Why
is
in
the
select
Austrians
you
see
in
the
CN
CF
channel,
which
is
we
created
after
we
donated
long
current
to
the
CN
CF?
Another
one
is
the
old
one.
We
resign
in
the
rancher
user
and
the
local
storage,
and
we
are
migrating
from
that.
So
the
community
contributions
are
welcome
and
alter.
They
have
different
format
of
contribution.
You
can
fix
that
you
can
help.
You
fix
the
box,
you
can
help
you
find
it
by
some
issues
and
file,
feedbacks
or
feature
request.
A
All
this
is
our
really
wearable
to
ask
many
of
the
feature
I
mentioned
before.
Like
say
the
with
users
feed
the
local
water
name
inside
long
for
you,
I
also
want
to
see
the
workload.
That's
in
fact
it's
the
feedback
from
users,
because
if
we
only
give
them
the
PVC
plus
UID,
they
don't
really
know
how
to
which
workload
or
social
based
is
that.
So
that
is
the
direct
result
from
the
use
of
feedback.
A
So
we
do
have
a
large
target
in
2020.
The
first
thing
is:
we
want
to
reach
1000
1000,
GA
and
including
features
like
volume,
resizing
performance,
metrics,
QoS
Accenture,
the
volume
reciting.
Just
the
many
people
mentioned
that
volume
resizing
is
probably
the
only
the
biggest
issue
that
start
them
for
using
don't
want
more
extensively.
A
So
we
like
to
address
these
sooner
or
later,
and
we
also
want
you
increase
the
community
by
10
times
giving
us
our
base
is
still
first,
more
I
think
is
probably
doable
and
we
want
to
reach
the
incubator
level
in
the
CN
CF,
and
that
requires
the
engagement
into
the
community.
So
we
are
really
trying.
We
are
really
working
on
that.
So
that's
all
for
my
session.
Any
questions.
Oh
thank
you.
B
A
If
you
can
count
those
Kasich,
it's
okay,
so
for
a
loan
pouring,
you
can
install
room
for
using
helm
chart.
That's
it!
That's
code
is
in
the
low
4/11
repo
and
in
fact
the
rancher
app
catalog
is
in
fact,
how
much
a2
and
also
the
one
you
see
that
I
starting
WordPress
using
Rancher
app
Kellogg.
That's
in
fact.
We
also
helped
chart
the
on
in
key
point
that
if
you
want
to
use
the
loan
for
your
storage,
it's
just
that
your
storage
class
to
Longhorn
or
other
storage
cause.
A
B
A
We
don't
really
have
a
official
performance
number
right
now,
I
can
say
I
pretty
confident.
One
fourth
performance
should
be
at
par
or
better
than
Seth.
Of
course,
epic.
In
fact,
this
one
is
coming
from
some
users
their
benchmark,
and
they
found
that
so
we're.
In
fact,
the
most
focus
we
have
right
now
is
on
the
developing
feature
and
stabilize
it
and
of
course
we
take
some
time
to
look
at
the
performance
before
we
reach
GA.
But
it's
not
really
at
our
pressure
at
this
moment.
A
We
want
to
have
more
people
using
it,
because
the
one
thing
about
storage
is
the
stability
is
really
important
and
we
never
want
to
lose
anybody's
data.
So
we
want
to
make
sure
that
will
never
happen
before
we
can
really
start
optimize
that
building
on
that
shouldn't
be
a
really
big
problem,
because
we
don't
really
haven't
look
into
that.
There
should
be
a
lot
of
low-hanging
fruit
there.
A
A
A
Yeah,
so
the
replica
with
the
each
replica
is
identical.
We
buy
people,
we
have
three
of
them,
but
they
contain
the
same
data
and
they
don't
directly
provide
access
to
the
data
usually
have
to
go
through
the
controller,
which
is
going
always
on
the
same
note
as
a
pod
and
provide
a
unit
in
front
and
for
extension
to
the
data
yeah.
A
Okay,
in
fact,
that's
interesting
because
we
used
to
put
engine
replicas
each
in
a
separate
part
and
their
users
with
powerful
machines
quickly
here
to
the
delimitation,
are
one
hundred
ten
parts
per
cooper
net
per
community
snowed,
so
in
the
latest,
in
the
purest
before,
in
the
seven
six
o
release,
we
decided
to
put
the
engine
as
the
process
instead
of
part
into
kubernetes.
So
now
the
engines
are
running.
The
processing
called
instance
manager
part,
and
we
can
have
multiple
engines
in
there.
B
B
A
I,
don't
really
sorry
I,
don't
really
care
the
question
because
the
demon
said
itself
normally
because
there's
a
team
on
stat
right,
so
normally
they
use
the
local
storage.
They
know
when
they
know
that
they're
not
really
migrate
between
the
node.
So
if
you
want
to
use
alone
home
for
demons
that
you
will
probably
heck
own,
you
have
one
volume
for
team
onset,
but
the
team
on
site
doesn't
really
have.
Will
incline
template
ID,
yeah
yeah,
if
you
use
the
staple
set
for
if
you
want
to
scale
your
stay
for
applications.
A
Yeah,
so
if
the
network
partition
happens,
does
the
point
the
choose
is
the
source
of
truth
is
in
the
controller.
If
controller
saying
this
replica
is
dad
is
a
dad
if
controller
down
if
it's
conversing
in
this
Arabic,
a
healthy
is
healthy.
So
all
the
consistence
at
controlled
by
the
controller
and
controller
only
on
runs
and
one
node
as
one
process.
A
That's
okay!
So
when
user
network
partition
happens,
the
part
because,
as
you
are
mentioned
before
the
controller
always
resign,
the
same
node
after
part
so
part
will
always
have
access
to
the
controller.
And
if
you
have,
if
you
get
only
one
node
called
isolated
you,
if
you
have
a
replica
running
on
the
local
node
controller,
we're
still
able
to
accept
you.
That's
replica
I
think
that's!
It's
only
ones
healthy.
A
A
Yeah,
so
it's
up
to
who
it's
after
longer,
to
decide
where
to
place
the
replica.
We
have
a
scheduler
to
basically
look
at
what
that
know
that
you
targeted
was
the
dispute,
have
attack
and
try
to
first
match
that
and
then
try
to
feed
that
okay,
so
you
want
your
20,
G
or
hundred
G
volume,
so
the
replica
will
be
100
G
and,
let's
see
which
piece
can
fit
that
and
we
try
to
put
that
in
on
the
different
note
as
possible.
So
the
basically
it's
a
Laurens
doing
the
scheduling
for
you.
A
A
Not
really
because
you
know
just
before
last
Friday
long
points
are
still
in
alpha
and
now
last
Friday
we'll
release
the
first
beta
release,
and
that's
probably
the
reason
why
you
haven't
heard
about
it
much,
but
we
are
expecting.
We
have
a
lot
of
interest
and
there
are
many
users
I
think
it's
not
really
on
the
warrior:
large-scale
production,
but
their
users
saying
that
they're
running
low
point
production,
probably
in
their
small
website
or
home
environment,
those
kind
of
cases,
and
we
were
expecting
more
to
come.
A
B
A
The
resource
requirement
will
definitely
recommend
that
you
have
at
least
cubic
views
better
with
full
wizard
views,
but
this
basically
depends
on
how
how
many
work,
how
many
wardens
do
you
want
to
create?
So
we
will
have
a
guideline
before
GA
and
that's
also
corresponding
to
what's
the
CPU
utilization,
those
kind
of
stuff
yeah.
We
had
a
performance,
we
definitely
get
a
tight
line
before
the
GA
and
give
you
such
guys
some
ideas.
Yes,.
A
Okay,
that's
the
that's
the
in
fact,
a
very
insurance
question
we
have
probably
spent
a
year
try
to
figure
out.
What's
the
best
way
to
do
it
and
in
the
end
we
choose
to
use
TGT
as
the
backend
and
the
Longhorn
factors,
connection
the
TGT
and
expose
icrc
target,
and
then
long
from
we
are
called
the
open
eyes
gases
running
on
the
host
initiator
to
connect
to
that
target
and
expose
the
block
device.
A
Previously
we
have
tried
an
IBD
and
we
have
tried
the
fuse
which
mounted
and
also
in
fact,
we
have
tried
the
TC
mu,
which
is
the
user
space
Linux
in
kernel
as
casa
target
and
I
have
contributed
a
bunch
of
patches
to
make
it
performance
under
performance
improvement.
But
in
the
end
we
sing
is
that
we
don't
really
it's
hard
for
us
to
require
users
have
to
run
certain
version
of
the
Linux
in
order
to
catch
local
running.
So
we
may
compromise
and
choose
the
Icicle.
B
A
Yeah
so
currently
now
we
we
don't
have
half
extension
yet
and
we
have
abstraction
on
the
different
levels.
For
example,
we
are
open
to
accept
new
package
store.
Drivers
are
dentistry
and
NFS
and
other
than
that
we
don't
really
have
other
extension
plan
at
this
moment.
But
yes,
things.
We
are
open
source.
Everything
is
stopped
subject
to
change.
A
A
B
A
A
Don't
think
I,
don't
believe
many
objects
support
upgrading
seen
in
our
first
stage,
but
we
are
really
trying
to
put
a
lot
of
effort
if
I
make
sure
that
users
have
a
smoothly
transition,
because
we
know
that
the
early
adopter
is
really
important
to
us
and
we
want
to
make
sure
that
we
got
the
feedback
we
make
sure
we
hung
up.
We
fix
every
problem.
We
can
think
so
that
we
want.
That
is
why
we
want
to
make
this
user
experience
great
and
that's
why
we
support
upgrade.
A
B
A
Yes,
yeah
so,
but
in
fact
there
is
a
little
bit
that
we
need
to
look
in
the
details.
What
policy
you
want
to
set
their
users
using
port
security
policies,
but
I
didn't
record
details
this
one
issues
on
that
and
we
have
stopped
it
and
if
you
have
more
information,
we'd
like
to
hear
more
feedbacks
all
right.
So
thank
you.