►
Description
Learn how running a storage platform in Kubernetes pods is a game-changer not just for storage administrators but for application developers as well. This ground-breaking technology runs containerized Gluster (a mature distributed open source storage platform) inside Kubernetes and on Red Hat’s OpenShift Container Platform as a completely integrated solution, aggregating storage form local hosts (Kubernetes worker nodes) and serving it out to application containers. This solution integrates a fully-featured enterprise-grade storage platform with a wide variety of data services including snapshots, geo-replication, tiering, cloning, encryption; runs storage and compute containers on the same set of nodes, all provisioned, scaled and upgraded using Kubernetes. The Briefing will feature a demo and a detailed roadmap for this solution.
Guest Speakers:
Sayandeb Saha (Red Hat) Head of Product, Container Storage, Red Hat Gluster Storage & Storage Management, Red Hat – Sayan is responsible for product strategy and leads a team of product managers for Red Hat Gluster Storage, Red Hat’s container native storage solution with Red Hat OpenShift, hyper-converged virt+storage solution with Red Hat Virtualization/KVM, unified storage management platform for Ceph & Gluster and is the technical Product Manager for CephFS.
Michael Adam (Red Hat) is an enthusiastic and experienced open source software developer, interested in all things about storage and containers. One of the main developers of Samba since more than a decade, Michael is today an engineering manager at Red Hat, leading two worldwide teams: The Samba team for Gluster storage, and the team that develops the Container Native Storage solution, which brings distributed and dynamic persistent storage with Gluster to Kubernetes.
A
A
Well,
hello
and
welcome
again
to
another
openshift
Commons
briefing,
this
time
we're
getting
a
tech
preview
of
cluster
FS
and
OpenShift.
So
I've
invited
two
of
the
folks
from
the
Red
Hat
storage
group,
I
in
Saha
and
Michael
Adam,
to
give
us
an
update
and
an
overview
of
Buster
and
open
shipped
and
I'm
gonna.
Let
them
introduce
themselves
and
kick
it
off,
because
I
hear
they
have
a
lot
of
content.
So
let's
get
going
Michael
Sam.
C
B
Basically,
you
know
providing
a
solution
where
you
can
run
storage
containers
side
by
side
with
application
containers.
This
is
actually
generally
available
today
in
form
of
a
solution
from
Red
Hat
so
and
I
have
other
responsibilities
in
Red
Hat
regarding
storage
product
I'll
not
go
into
those
details.
Michael
go
ahead.
Thank.
C
You
so
yes,
I'm
part
of
the
open
of
the
starch
ingenuity
and
I'm
working
I'm
running
the
team
that
actually
works
on
that
so-called
container
native
storage
solution,
which
is
the
solution
that
ties
together,
cluster
and
openshift,
which
we
are
going
to
talk
about
today
and
yeah.
Let's
proceed
and
we'll
explain
what
this
all
is
about.
C
Saying
if
you
could
advance
this,
my
excuse
right?
Yes,
cliche!
Yes,
so
this
is
the
rough
agenda
or
what
we're
going
to
cover
I'm
going
to
talk
about
what
actually
is
Gluster
I'm
going
to
talk
about
briefly
about
the
starches
cases
in
shift,
then
I'm
going
to
describe
the
weight,
what
we
call
container
native
storage
and
what
it
actually
is
and
then
I'm
going
into
the
details,
how
it
all
works.
C
Finally,
there's
the
roadmap
thing:
if
we
have
time
and
slightly
how
it
all
works
sections,
we
have
two
demos,
but
if
we
don't
have
time
they're
recorded,
they
are
available
on
YouTube.
Even
so,
if
you
are,
if
you
don't
have
the
time,
then
we're
gonna
stick.
Those
and
yes,
of
course,
questions
are
encouraged
whenever
something
is
unclear
or
people
on
more
details.
So
let's
go
on
next,
like
this,
so
no
gone.
What
cluster?
What
actually
is
Gluster
next
slide?
Please!
Oh!
C
C
It
is
also
in
all
its
simplicity,
very
flexible
offers.
A
lot
of
additional
nice
features,
and
it's
very
flexible
in
this
configuration
the
how
what
does
it
offer
about
way
of
access
methods
so
they're?
The
Native
Client
is
a
fuse
based
a
POSIX
file
system
mount
that
is
available
on
Linux
or
UNIX
clients.
C
We
have
NAS
heads
like
NFS
SMB,
you
so
Samba,
offering
the
SMB
protocol
for
typically
Windows
clients
and
a
built-in
NFS
server
and
a
new
extra
NFS,
fellow
with
the
Ganesha
nfsv4
and
your
system.
So
this
is
sitting
on
top
of
Gloucester
basically,
but
you
also
as
new
features,
have
about
to
have
a
very
release:
an
ice
cozy
a
front
end
based
on
a
big
file
sitting
inside
to
us
to
storage,
and
we
also
have
a
object
accessed
by
so
called
Gloucester
swift
front-end,
which
also
features
a
plugin
for
offering
Amazon
s3
object
access.
C
So,
as
you
see,
file
is
the
native
access
likely
up
there
offering
X
cannot
file
system,
but
block
and
object
access
is
added.
On
top
of
that.
So
that's
roughly
the
the
overview
of
what
cluster
is
and
what
Gloucester
offers
and
the
next
slide
of
a
few
more
details
about
the
community.
Stoke
cluster
is
an
open
source
project.
You
can
visit
its
home
on
glass
to
the
org
you
can.
This
is
the
github
space,
with
a
lot
of
repositories
with
a
code
and
and
stuff
surrounding
it
on
get
upload
comp,
slash
cluster
and
it's?
C
What
is
not
written
here
record
ships
nest
this
as
part
of
the
record
Colossus
storage
product
rhg
is,
but
this
is
the
upstream,
the
community
facing
thing
some
more
technical
details
about
cluster
on
the
next
slide.
Thank
you.
So
the
basic
concept
of
a
volume
is
in
semester,
so
the
volume
is
the
unit
of
start
that
straw
furs
I
mean
these
volumes
are
compelled
composed
of
local
file
system
directories.
So
it's
really
very
simple:
you
have
local
file
system
like
typically
XFS,
and
this
directories.
Composing
volume
are
called
bricks
these.
C
What
these
volumes
can
be
composing
of
bricks
in
different
types,
their
volume
types
typically
different,
the
characteristic
of
durability.
We
have
replicate
and
distribute
those
can
be
combined
to
have
a
volume
that
is
distributed.
Distributing
it
starts
over
various
local
replica
sets.
Rocky
cassette
is
a
group
of
bricks,
typically
located
on
different
notes
where
you
store
something
it
is
replicated
to
all.
Let's,
for
instance,
say
your
three
replicas,
then
it
when
storing
a
file
will
replicate
the
contents
to
three
bricks.
C
Another
characteristic
is
the
dispersed
type
of
volume,
which
is
an
erasure,
codex
way
of
doing
volumes,
and
there
are
more,
but
these
are
for
our
purposes,
the
most
important
ones,
I
alluded
to
the
flexibility
and
feature
rich
Ness.
This
is
mostly
due
to
the
architecture
of
a
stack
of
translators,
so
these
translators
are
basically
pluggable
modules
that
can
be
stacked
and
volume
definition
consists
of
a
list
of
these
translators.
These
translators
add
some
features
like
I
mean
some
more
interesting.
High-Level
features
are
encrypted
volumes.
There's
a
snapshot
feature.
C
The
last
reason
has
this
so-called
user
serviceable
snapshots,
where
the
snapshots
are
visible
inside
the
actual
mounted
directory,
and
you
can
very
conveniently
access
your
old
versions
of
the
files.
We
have
something
called
GA
replication.
As
I
said,
cluster
is
a
clustered
file
system,
but
what,
if
you
have
multiple
clusters
on
different
sites,
where
the
link
that
is
it
has
too
high
latency
in
order
to
form
a
single
cluster,
then
there
is
something
called
your
application
where
data
is
replicated
between
these
clusters
in
is
in
Chronos
fashion
quota.
Is
there
etc,
etc?
C
Many
of
these
things
often
also
not
POSIX,
file
system
features
like
POSIX,
echolls,
etc,
etc.
So
access
control,
all
these
things
are
available
through
translators
and
also
additional
demons
demons
that
around
so
this
is
not
as
a
monolithic
software.
It
is
for
the
whole
glassware
system
in
consists
of
many
demons
that
are
running
on
each
of
the
storage
nodes.
We
have
a
cluster
D,
which
is
kind
of
a
central
cluster
membership
and
volume
management
demon.
We
have
bricks
bricks
targets
for
each
brick.
We
have
demons
for
coltrane
snapshots,
etc.
C
So
this
is
also
very
flexible
and
very
robust.
In
the
sense,
when
something
bad
happens,
a
disk
dies
or
something
then
there's
you
can
repair
a
volume.
You
can't
react
new
disks,
there's
something
called
cell
field
which
will
lead
to
automatic
repair,
repairing
of
a
failing
disk
when
you
add
a
substitute,
etc,
etc.
So
this
is
very
feature-rich.
You
can
visit
cluster
org
and
navigate
to
the
documentation
section
there
to
read
more
about
the
details
and
with
that,
of
course,
any
times
when
there
are
more
questions,
we
can
come.
How
come
to
the
story?
C
C
So
what
are
the
use
cases
for
storage
in
OpenShift?
There
are
basically
in
two
main
use
cases.
There
are
application,
containers
and
there's
the
registry
so
for
both
poems
I
have
a
slide,
explain
a
little
bit
what
this
means.
So
persistent
storage
for
application
containers
is
next
next
lap,
yes
Thanks,
so
the
containers
in
itself
have
the
paradigm
of
being
stateless,
ephemeral
animals.
Basically,
so
they
can
be
brought
up
and
down
in
a
stir
on
different
nodes,
and
this
will
actually
be
just
instances
of
the
same
application
container.
C
And
what,
if
I
brought
down
up
again
any
data
they
start
on
the
local
disk?
Have
it
has
been
removed
its
reserved
to
the
urgent
state,
so
the
apps
that
usually
they
want
to
store
their
data
somewhere?
Persistently,
though
the
state
of
the
application,
the
users
data,
the
configuration
of
the
application,
all
that
needs
to
be
stored
and
persistently,
and
for
that
something
like
luster,
which
is
highly
variable,
robust
scalable,
seems
like
a
natural
fit.
So
this
is
the
first
use
case
that
can
here's
a
question.
C
B
I'll
take
that
question,
so
that
is
yes,
I
mean
the
way
to
look
at
it
is.
We
are
chipping
away
at
the
problem
one
by
one.
So
we
understand
that
open
shipped
has
infrastructure
wit,
need
storage
infrastructure
includes
things
like
registry
metrics,
with
Cassandra
and
ocular,
and
logging
with
elasticsearch.
Yes,
those
need
storage.
We
we
are,
will
do
the
registry
first
and
then
we'll
look
at
Cassandra
and
and
elasticsearch
in
a
future
release
to
see
what
we
can
do
there.
Both
Cassandra
and
elasticsearch,
recommend
that
local
storage
be
used
with
SSDs.
B
C
Raaah
except
you
so
that
seems
to
answer
the
question.
There
was
two
things
wrong:
Alex
on
the
chat
so
right.
The
second
natural
use
case
would
be.
The
registry
registry
is
also
a
place
where
data
is
stored,
persistently
for
access
by
by
many
processes
and
the
authorship
cluster,
namely
the
container
images.
It's
the
repository
containing
images,
all
50
uses
NFS
as
a
registry
storage
back-end
by
default,
and
this
seems
to
be
a
very
good
fit
for
cluster
as
well,
because
how
is
NFS
usually
done?
C
C
There
is
another
question
here
from
jonathan:
are
you
running
NFS
on
top
of
bluster,
transparent,
openshift
or
does
oven?
15
know
about
clusters
as
follows?
This
is
a
good
question.
I
will
cover
the
details
in
later
slides.
We
are
not
running
on
top
of
NFS,
you
are
doing
it
natively,
and
this
on
somebody
comments
they're
using
cinder.
Instead
of
NFS,
yes,.
C
C
And
if
you
have,
of
course,
an
acoustic
environment
is
in
there,
maybe
a
got
a
natural
fit
for
the
registry,
of
course
as
well.
That's
right
so
we'll
cover
a
little
bit
more
about
the
registry,
but
since
registry
is
currently
in
under
development-
and
we
are
talking
about
what
we've
done
in
the
past
first
and
what
we
currently
ship
in
and
then
we
go
into
the
roadmap.
So
this
is
what
I
call
two
awards
in
the
container
native
storage.
C
C
C
A
C
Thank
you
so
that
in
this
case
the
glass
of
storage
cluster
runs
outside
of
the
object
cluster
somewhere
on
bare
metal
or
whatever.
Since
it's
a
software-defined
storage
system,
it
basically
can
run
anywhere,
can
be
in
a
virtual
environment
or
very
metal,
and
the
open
containers
make
use
of
the
volumes
that
are
provided
by
that
external
cluster
storage.
So
there
is
some
link
from
these
containers
to
the
outside
storage
world
and
you
have
all
the
flexibility,
but
it's
like
a
general
set,
a
separate
tab,
and
that
was
step
number
one.
C
Called
that
yes
container
rice
storage,
thank
you.
So
we
started
to
containerize
blaster
into
a
cell,
putting
the
glass
to
starts
node
into
containers.
That
was
not
done
before.
As
I
explained
earlier.
Cluster
itself
is
a
relatively
complicated
set
of
diamonds.
They
need
a
lot
of
special
privileges
since
they
and
the
act
on
local
storage
devices
etc,
relatively
low
level
system
demons,
and
so
it
was
not
entirely
trivial
to
containerize
those,
but
it
was
done
and
so
in
the
second
intermediate
step
called
container
s.
C
C
This
is
container
native
storage
in
this
setup.
Glasser
runs
inside
containers
in
the
overshift
cluster
itself.
That
means
the
containers
that
were
previously
somewhere
outside
and
somewhat
only
cost
set
up,
we're
now
running
inside
out
drift
and
managed
upshift,
basically,
and
so
it
is
using
local
disks
from
the
hosts
that
are
consumed
inside
the
oven
and
set
the
class
for
storage
containers
and
the
cluster
is
really
running.
C
As
we
say,
hyper-converged
inside
the
object,
node
and
the
application
parts
that
are
consuming
those
glassful
volumes,
they
can
either
run
on
separate
object
nodes
or
they
can
run
on
the
same
notes
they
can,
but
must
don't
have
to
be
co-located.
This
is
also
demonstrated
in
this
in
this
picture.
Here
you
see
there
is
an
object,
node
one
which
only
runs
application
containers,
Abu,
Chef,
Note,
2,
&
3.
They
run
both
storage
and
application,
containers
and
open
v.
Note
4
exclusively
runs
starch
containers,
so
all
three
combinations
are
possible.
C
Then
there
are
possible
on
the
same
cluster.
So
it's
entirely
up
to
the
administrator.
How
to
set
this
all
up,
sometimes
resource
wise.
It
may
make
sense
to
have
it
entirely
separate.
Thank
you
yes.
So
this
is
the
general
what
we
are
doing,
it's
fine.
We
can
move
on
and
explain
in
a
little
more
detail
how
this
all
works,
because
there
are
many
nitty-gritty
details
about
that.
C
C
We
talked
about
cluster
of
las
profess
clusters
that
are
providing
the
storage,
but
that's
basically
not
all.
So
how
does
often
shift
actually
talk
to
cluster?
How
does
it
make
the
connection?
It
does
so
by
some
components
that
we
headed
to
Oakland
shift
or
to
kubernetes,
which
is
the
one
inside
the
dynamic
or
it
altered
guster
vest--
provisioner,
which
is
piece
that
creates
the
so
called
persistent
volumes
to
be
used
by
the
appreciation
containers
and
we
have
the
mount
plugin.
C
That
is,
that
responsible
for
actually
doing
the
mounting
so
that
the
application
container
can
use
the
starch
that
the
provision
of
created
for
Astra
s
as
I
said.
Yes,
one
or
more
clusters
clusters
running
inside
of
a
shift
in
s
containers
but
open
chef
does
not
directly
talk
to
them,
but
we
have
in
the
middle
between
the
two.
We
have
a
service
interface
that
is
called
a
Katie.
C
The
fourth
component
that
is
also
in
the
mix,
is
the
deployment.
So
this
will
be
the
last
part
of
the
overview
of
what
we
currently
have
I
mean
the
first
three
components
describe
an
installed
system
and
how
this
all
works.
The
last
thing
is:
how
do
we
install
something
like
that,
and
so
there
is
a
deployment
tool
which
is
basically
one
command,
and
it
will
install
everything
for
you.
C
So
how
does
persistence
start
in
alkyl
shift
today
work?
This
is
just
a
brief
recap,
assuming
that
most
of
the
audience
will
be
familiar
with
that.
So
when
I'm
talking
about
a
part,
this
is
the
group
of
one
or
more
containers
a
foreign
entity.
Basically,
the
smallest
entity
of
applications
that
we
have
a
persistent
volume
is
the
concept
that
provides
persistent
storage
and
can
be
monitored
by
an
application
part.
A
provisional,
as
already
mentioned,
provides
persistent
volumes
upon
request
and
the
plug
at
the
mount.
C
Plugin
is
the
mechanism
that
mounts
the
persistent
volumes
and
it
is
referenced
in
the
PV
server
pity
the
person
volume
carries
the
knowledge
how
it
can
be
mounted.
A
persistent
volume
claim
or
PPC
is
the
mechanism
by
which
a
user
can
request
a
persistent
volume
to
be
created.
It
is
also
the
link
by
which
the
user
can
later
use
that
volume
in
an
application
container
to
be
mounted.
C
There
are
various
access
types
for
volumes,
I
have
listed
the
acronyms
and
all
sort
of
Spelling's
out,
so
there
are
double
yellow
with
white
ones.
Are
Delta
X
required
many
ro
X
with
one
only
minima,
so
this
is
very
important,
so
the
typical
thing
we
are
currently
targeting
for
or
concentrating
on
is
the
rwx
vitamin
e.
So
there
is
one
volume
and
a
lot
of
application.
Parts
on
multiple
nodes
can
access
that
volume
simultaneously.
C
Imagine
a
application
and
you
have
the
data
with
a
persistent
to
store
data
for
the
web
sites
and
the
web
server
heads
these
parts.
They
are
accessing
the
same
data
and
this
can
scale
out.
You
can
spawn
more
of
them
of
these
web
server
parts.
They
all
exist.
The
same
day.
I
can
show
it
out
to
the
clients
for
required
once
I
have
some
more
details,
it's
also
supported,
but
there
is
a
some
special
thing
on
the
roadmap
and,
yes,
the
of
it
provisioning,
there's
actually
static
and
dynamic
provision.
C
All
right
so
now,
in
order
to
be
able
to
appreciate
the
dynamic
provisioning
here
is
what
static
provision
is.
This
was
the
only
way
to
provision
persistent
volumes
before
OpenShift,
3
or
4,
though
the
admin
had
to
pre
create
a
set
of
persistent
volumes
and
the
static
provisioner
gets
a
request
for
certain
size
and
then
chooses
from
the
existing
persistent
volumes
based
on
the
type
and
size
for
from
the
claim,
and
it
just
chooses
it
tries
to
best
approximate
from
above
the
size
requesters.
C
So
that
means
if
the
admin
has
pre
created
like
a
couple
of
100
gigabyte
volumes,
and
then
it
gets
requests
for
one
gigabyte
volumes.
A
lot
of
space
is
wasted,
and
so
this
is,
it
has
more
work
that
has
to
be
done
and
it
is
likely
wasting
space
so
on
the
next
slides
I
have
just
a
few
like
screenshots
to
see.
Ok,
what
is
the
workflow
of
commands
for
static
provisioning?
C
First,
you
have
to
create
some
endpoints
and
service.
I,
don't
go
into
the
details
here
for
application
parts
to
be
able
to
reach
your
storage
provider
gospel
in
this
case
and
then
in
the
case
of
cluster,
you
I
use
the
heck,
Hettie
CLI,
the
activity,
command-line
interface
in
order
to
pre,
create,
for
instance,
a
100
gigabyte,
persistent
volumes.
These
would
be
a
lot
of
those,
and
on
top
of
that,
we
need
to
create
open
fifth
level.
C
Persistent
volumes
with
OSI
created
with
some
kind
of
JSON
file,
describing
the
persistent
volume
here
and
an
application
part
could
them
as
or
a
user
could
create
a
persistent
vol
claim
so
from
here.
It's
easy.
It
works
if
the
persistent
volume
and
could
use
it
in
an
applique.
But
the
tedious
part
is
the
manual
3
creation.
So
you
will
go
to
the
storage
back-end
and
create
the
volumes,
and
it
would
create
a
lot
of
these
persistent
volumes
manually.
C
So
here
is
a
question
as
I
understood
in
the
past.
You
will
need
one
cluster
employed
purposes
and
volume.
Is
this
still
the
case?
Yes,
this
is
also
in
the
in
the
later
in
the
dynamic
provision.
Right
is.
This
is
very
much
true.
We
will
see
that
a
bit
later
so
for
the
static
Karishma.
Here's
an
example
how
the
PBC
would
look
like.
We
see
that
yeah.
C
C
So
it
was
added
in
3.4
and
how
this
works
in
general
is
a
start.
Class
concept
is
added
and
the
administrator
creates
the
start
class.
This
describes
the
storage,
basically
in
a
high
level
and
it
references
a
dynamic
provisioner
you'll
see
that
an
example
of
the
next
slides,
and
so
the
difference
now
is
the
first
important
claim
that
is
done
by
the
user.
It
references
the
start
class
and
the
provisioner
from
that
storage
class
is
automatically
called.
It
creates
the
persistent
volume
of
requested
size
and
type
etc
dynamically
on
demand.
C
So
there's
no
pre
creation
that
whole
thing
is
going
away
and
the
user
can
just
mount
the
PBC,
FEV
or
Brit
reference.
The
PVC
after
it's
bound
and
an
application
for
just
like
before
the
main
part
is
not
pre
creation
required
and
the
the
requested
size
will
be
fulfilled
exactly
and
not
approximated,
maybe
poorly.
So
please
proceed.
C
Though,
how
does
that
work
with
a
dynamic
provisioner,
though
I
only
have
it
in
the
flow
of
bullets
from
left
to
right
and
back
again,
apologies
for
not
having
a
real
diagram,
but
this
is
how
it
goes.
As
I
said,
the
persistent
volume
claim
references
there,
yeah,
basically
the
starch
class
and
implicitly
there
by
the
glass
surface
provisioner.
It
is
called
the
glossners
prisoner
extracts.
The
details
of
the
request
from
the
persistent
pudding
claim
it
tells
it
goes
out
to
hit
Kevin,
tells
it
to
create
a
certain
volume
of
a
given
size.
C
On
top
a
getting
looks
for
a
cluster
cluster
that
can
satisfy
the
request,
though,
it
looks
for
a
cluster
that
has
enough
storage
available
if
it
finds
one.
So
that
can
be
multiple
of
them.
Then
a
KD
tells
the
cluster
instance
to
create
the
volume
once
that
is
successfully
done,
Gloucester
reports
back
and
Acadie,
hence
the
volume
that
was
created
back
to
the
provisioner.
The
provisioner
creates
a
persistent
volume
of
that
and
puts
the
glassful
volume
details
into
it
so
how
to
reach
it,
etc.
C
It
also
referents
the
Dasara
festival
and
to
be
used
for
mounting
that
and
the
provision
of
them
we
transfer
persistent
volume
to
the
caller
the
upon
success.
The
process
boarding
claim
is
bound
to
this
newly
created
persistent
volume
and
then
can
then
be
used
in
a
product
by
the
user.
As
before,
though,
this
is
how
it
works.
One
left
to
right
PVC
to
the
provisions
of
kubernetes
to
a
Katie
and
Katie,
finds
a
cluster
cluster.
That
is
fitting.
It
calls
that
creates
volume
there
and
goes
back
all
the
way.
C
Thanks.
Thank
you.
So
here
is
an
example
for
storage
class,
and
you
see
the
professional
:
cuban
Edisto
io
/
cluster
fest.
This
is
the
reference
that
tell
them
that
this
is
to
be
used
so
and
also
in
the
parameters.
There
are
the
details:
how
reach
the
Giachetti
in
this
case?
Oh
this
is
basically
even
then
that's
it
Eric
Eric,
this
shouldn't
be
one
around
a
27001.
It
should
be
a
real
resolve,
the
name
that
leads
to
the
genetic
part
wherever
it
is
running
currency
because
it
can
move
in
the
cluster
though,
but
in
general.
C
This
is
the
main
point
here
is
we
have
the
provisional
named
in
the
starch
class
and
we
have
how
to
reach
the
Academy
inside
the
promoters
and
uni
notice,
the
name
of
the
starch
class.
In
this
case
it's
called
cluster
container,
and
you
can
give
any
name
that
should
just
be
unique
and
so
on.
The
next
slide
we'll
see
how
this
is
used
by
Ramon
in
a
PVC
for
system
volume
claim.
C
If
you
reference
it
here,
this
will
laps
OC
create
when
you
run
it
with
this
claim,
it
will
let
it
know
that
it
has
to
go
out
with
a
glass
surface
provisioner
with
data
down
here,
for
instance,
we
have
access
mode
required
once
we
have
starch
for
gigabyte
and
it
will
try
and
create
exactly
a
volume
of
this
class
next
slide.
Please
another
example
of
how
this
thing
is
used
in
an
application
definition.
C
C
C
Whether
it's
fully
supported
at
least
in
like
granite
product
is
another
thing,
but
technically
it
is
there.
Yes,
a
Cathy
CLI
has
a
lot
of
these
things.
F
thing
to
my
knowledge
is
not
able
to
trigger
that
with
with
OC
commands
currently,
but
you
can
still
work
on
the
volume
with
the
heck
Etica
mater.
Okay.
C
So
again,
do
not
be
able
to
compare
to
the
previous
aesthetic
provisioner
I
have
here
the
flow
of
commands
for
the
dynamic
provisioner.
Of
course
you
first
have
to
create
the
storage
class
and
then
also
you
need
the
secret
here
for
excessive
naked,
but
then
all
the
intermediate
steps
they
are
missing,
like
you,
don't
need
to
recreate
the
heck
and
evolve
of
the
Caddy
and
persistent
volumes
with
Ozzy
create,
but
then
Duran,
just
as
before
the
class.
C
They
also
create
on
the
process
system
volume
claim,
but
this
time
referencing
our
storage
class,
and
then
it
works
exactly
the
same,
we're
just
from
there
and
I.
Don't
have
all
the
screenshots
here
for
that
now,
but
you'll
see
that
this
dynamic
provision
would
always
for
each
persistent
loading
claim
a
persistent
one
that
is
successfully
created.
It
will
create
an
funding
service,
because
this
will
this
is
important.
So
even
though,
maybe
each
time
the
same
cluster
cluster
is
referenced.
C
C
Right
Cheyenne
comments
that
commands
does
not
support
recess
operation
yet,
hence
you
have
to
use
the
a
kitty
in
a
future
version.
This
will
likely
be
integrated.
So
what
does
the
mount
plugin
do?
The
openshift
host
has
glass
refers
client
installed,
though
it
is
not
imposing
any
requests
or
any
requirement
on
the
application
parts.
The
host
mounts
the
class
refers
volume
and
then
bang
lands
it
into
the
application
container.
C
So
for
the
container,
it
will
just
see
some
directory,
which
presents
the
filesystem,
the
POSIX
filesystem
to
him,
and
all
the
magic
is
happening
on
the
host.
This
is
how
the
Gloucester
has
mount.
Plugin
works
in
simple
words,
all
right.
So
next
slide.
Please,
let
me
see
we
are
20
minutes
at
the
end
of
the
I
think
we
should
move
on
and
if
we
have
time
at
the
end,
you
will
have
a
demo
I
think
we
have
probably
sufficiently
explained
how
it
works
in
principle
and
the
Schlachter
slides,
which
can
be
shared
later
or
contain.
A
C
And
it
can
be
of
various
forms.
I
mean
is
the
concept
of
of
open
character,
community
secrets
here,
so
it
can
be
password
or
something
else.
Okay,
okay,
then,
let's
move
on
and
defer
the
demos
to
the
end
of
the
list.
If
you
have
time
so
a
little
bit
more
detail
about
a
kiddy
I've
mentioned
a
little
bit
of
it.
So
it
is
not
high
level
service
interface
for
managing
likes,
a
black
speck
of
dust
volumes
in
the
sense
it
can
create
it.
It
can
expand
it
and
it
can
delete
the
volumes.
C
It
is
also
providing
this
restful
api
and
the
commander
interface
to
access
these
features
so
kubernetes.
Well,
when
creating
a
persistent
volume,
the
cluster
does
provision,
it
will
talk
to
Katie
by
our
restful
api
and
for
any
additional
stuff
that
is
not
exposed
by
elect
by
kubernetes
or
purchased.
You
can
use
the
ATC
live
from
some,
for
instance,
if
on
a
master
know
what
you
install
the
hey,
Katie
CLI
package,
then
you
will
be
able
to
use
that
the
main
point
is
it
manages
one
or
several
vessel
clusters
yeah
what
it
can
do.
C
One
of
the
main
points
is
that
really
highest
nitty-gritty
details
of
organ
creation
and
cluster
selection
from
the
caller,
so
usually,
you
have
to
specify
each
and
every
single
brick
a
discs
at
etc
and
the
body
type
etc.
So
the
creation
cluster
is
some
amount
of
command
line
and
foreign
caddy
address,
say:
hey
I
need
a
5
gigabyte
volume
by
default,
it
uses
replica
3
volumes
and
it
could
also
upon
request
to
other
types
of
volumes.
Currently
kubernetes
always
requests
replica
3
in
the
future.
C
We
are
going
to
offer
other
types
as
well
exposed
through
the
storage
class,
but
but
this
is
how
it
currently
works,
as
I
said
just
text,
size
and
durability
type,
and
it
finds
a
cluster
with
enough
disk
space
to
satisfy
the
requests
goes
there
creates
it
and
yeah
in
order
to
do
all
that
it
has
its
own
state
startling
database.
That
is
one
thing
to
be
aware
of
this
database
maintain
said
about
the
nodes
and
all
the
clusters,
and
also
about
the
volumes
and
the
disks,
though
this
is
very
important
to
bear
in
mind.
C
I
also
put
in
the
link
here
and
as
the
other
components.
This
is
all
open
source
software,
here's
its
home
on
github.com,
Stesha,
caddy
speciality.
By
the
way,
hey
kitty,
you
may
wonder
what
what
the
heck!
That
name
means.
It
is
the
word
for
one
in
the
native
Puerto
Rican
language,
because
the
original
author,
author
of
a
list
of
error,
was
originating
from
and
patrika.
So
this
is
the
origin
like
a
fun
fact
about
this
software.
C
Let
me
see
there
is
a
question,
but
you
already
answered
what
I
think
I
understand
is
that
you
might
find
a
Gloucester
plaster
all
sort
of
opposite.
Then
the
hosts
will
net
remote
volumes
locally.
Is
that
recommended
approach
here
that
you
can?
You
can
do
it?
You
can
running
outside
and
you
can
even
run
it
inside.
In
both
cases,
the
Caddy
will
run
as
a
container
right,
so
the
hickety
container
is
much
less
special
than
the
Gloucester
containers.
I
told
you
earlier,
the
Gloucester
here
was
not
trivial
to
container
eyes.
There's
a
lot
of
things.
C
It's
even
running
today
as
the
privileged
container
accessing
the
disk
and
network
from
the
host
etc.
Hey
kitty
is
much
less
special.
It
is
a
single
container
cluster.
It
is
not
highly
available
in
it
can
move
in
the
cluster
communities
of
Geritol
place
it
for
you.
The
main
point
is
that
its
database
needs
to
be
persisted
in
order
that
the
state
is
not
for
own
lost,
so
this
is
currently
stored
in
cluster
volume
and
the
next
releases.
There
will
be
another
storage
place,
but
currently
this
is
how
we
doing
it.
C
There
is
this
kind
of
chicken
and
egg
thing,
Mary
already
create
a
cluster
volume
and
then
put
our
database
inside
that,
and
the
database
is
also
the
reason.
Well,
you
have
to
be
very
careful.
You
should
only
use
a
kitty
to
modify
your
volumes
and
you
should
not
fill
with
cluster
directly
for
questions
to
run
inside.
Your
hosts
need
big
drives,
is
one
coming
for
Jonathan
shine
says
yes
and
no.
You
could
load
up
some
of
the
hosts
with
big
drives
and
run
cluster
container.
C
So
here
is
the
warning
sign
again
right:
don't
mess
with
the
ball
is
manually
manually,
meaning
using
going
into
the
cluster
containers
or
cluster
nodes
and
running
the
cluster
commands
to
modify
the
volumes
that
should
really
be
not
not
be
done.
Use
a
kitty
or
in
catacomb
state
may
get
disturbed.
So
this
warning
sign.
So
we
are
wait,
working
on
future
being
able
to
remove
that
warning
sign
to
to
more
tightly
integrated
Kitty.
The
Kitty's
stayed
with
cluster
itself,
but
currently
this
is
order
this.
So
the
warning
needs
to
be
done.
C
Though
we
can
move
on
please
so
the
next
point
is
now
that
we've
described
how
this
all
works,
how
to
get
to
such
a
system,
be
it
with
glass
for
outside
or
glass
straight
inside
and
shift.
So
for
these
purposes
we
added
a
tool.
The
upstream
version
of
this,
like
the
option
project
on
github,
is
under
the
cluster
organization.
Currently
it's
called
cluster
kubernetes
and
the
tool
that
does
the
deployment
for
us
is
called
in.
C
Gk
deployed
like
cluster
coolant
is
deployed,
but
it's
also
capable
of
running
in
an
open
shift,
we're
all
using
OC
commands
instead
of
Q
control
and
in
the
redhead
products
we
ship
it
as
a
CNS
deploy
CNS
being
the
acronym
of
container
native
storage,
which
we
kind
of
hook
up
on
the
openshift
releases.
So
you
see
that
I'm
talking
about
CNS,
3
or
4,
which
is
more
or
less
unum
for
open
shift
before
this
respect.
So
that's
not
to
do
right.
I
mean
we
need
to
install
the
cluster
cluster
or
more
of
them.
C
C
So
this
is
what
had
what's
happening.
This
tool
takes
a
topology
file,
so
called
pathology
far
as
an
input.
The
topology
file
has
the
details
about
the
note.
Ip
addresses
the
names
of
the
disk
devices
in
each
node,
etc,
etc.
It
takes
that
and
it
creates
an
egghead
apart.
It
creates,
if
told
so,
it
is
optional.
C
It
also
creates
the
glass
broken
apart
if
it
does
that
it
chooses
a
diamond
set
to
set
up
the
constrictor,
the
cluster
clusters
and
then
it
creates
and
deploys
the
Hickey
teapot
and
then
has
her
Ketty
loaded
topology
file
and
do
all
the
necessary
steps
of
the
setting
up
of
the
actual
cluster.
So
the
deployment
of
the
classroom
nodes
is
just
the
notes.
The
software's
there,
the
disk,
are
there
and
then
hey
kid.
It
goes
and
joins
these
Glastron
notes
into
clusters.
It
was
also
called
trusted.
C
There
are
a
couple
of
prerequisites
to
fulfill
or
regarding
file
over
walls
and
stuff,
like
that.
I
have
not
put
all
the
nitty-gritty
details
in
here,
but
the
deployment
tool
will
tell
you
about
them
in
the
future.
The
next
version
will
even
be
able
to
check
these
prerequisites
and
also
potentially
satisfy
some
of
them
automatically.
C
C
C
Maybe
let's
continue
with
a
few
slides,
because
time
is
running
short
and
maybe
I
can
read
some
selected
questions
later
on,
because
the
discussion
seems
to
be
framed
right.
Some
earlier
points,
look
like
the
demo
of
the
dynamic
provisioning.
Let's
defer
the
deployment
demo
to
the
end.
If
there's
more
time
left
in
advance,
we.
A
C
Continue,
please
I've
told
you
what
is
available
in
in
OpenShift
3.4,
but
jets.
Gustov
has
a
provision
and
mount
plugin
satisfying
persistent
volume
claim
and
dynamically,
creating
their
recording
volumes.
It
is
possible
to
use
that
for
all
types
of
access
and
you
can,
if
you
know
how
to
do
it.
In
theory,
also
videos,
glass
device
is
a
registered
back-end,
but
in
3.5
live
for
the
first
time
documented
officially
and
have
the
reference
process
for
doing
that.
C
We
are
also
working
towards
improved
what
we
call
day-to-day
maintenance
operations
like
shrinking
the
volume
by
removing
discs,
replacing
a
disk
with
a
bigger
disk
or
different
disks,
and
without
a
downtime,
and
this
is
what
we
were
working
towards
in
the
very
very
soon
release
of
three
up
five
for
3.6.
There
are
some
bigger
change
can
know,
namely,
as
I
said
all
access.
The
types
are
access
modes
are
supported,
including
with
white
ones,
so
the
grid
white,
one
is
access.
C
Typically,
this
is
in
many
cases,
though,
used
as
access
from
a
single
container
like
exclusive
XS
m,
and
this
these
applications
requesting
exclusive
access
for
some
persistent
volume.
They
usually
expect
a
certain
performance
guarantees
like,
for
instance,
the
things
alluded
to
earlier
in
some
question,
like
logging
and
so
on.
Shannon
mentioned
the
Cassandra,
etc.
All
stuff
like
that
and
I
mean
just
plain
databases
where
there
are
some
application
is
exclusively
accessing
this.
They
usually
have
very
high
performance
demands
and
the
cluster
of
s
plug
alkalosis
s
pilot
mount.
C
This
is,
of
course,
never
as
fast
as
a
local
file
system
because
of
many
of
the
operations,
especially
the
metadata
operations
have
to
be
coordinated
in
the
whole
cluster,
and
this
is
much
more
expensive
than
just
going
to
a
local
disc,
so
creating
firewalls,
locking
parts
of
the
file
and
stuff
like
that.
All
this
is
very
expensive
and
the
text
time
and
if
some
application
has
needs
for
some
latency
here
or
they
may
not
people
so
them
you
can
always
try.
It's
not
it's
a
portrait,
but
it
may
not
be
satisfactory.
C
So
the
interesting
point
is
that
this
will
also
be
hosted
by
cluster
and
it
will
also
be
able
to
be
dynamically
provisioned,
so
we
will
have
dynamically
provisioned
volumes
which,
in
the
backend
are
implemented
as
some
kind
of
block
transport,
but
this
is
completely
transparent
for
application.
The
application
will
request
required
once
volume
from
a
certain
provision
er-
and
it
look
at
this-
and
this
is
going
to
be.
This-
is
planned
for
the
summer
release
of
3.6
and
for.
B
C
Versions,
2.7
and
later
we
have
some
more
things
on
the
roadmap
here,
one
thing
is
object.
Access
I
mentioned
that
s3
based
object.
It
says,
on
top
of
glossary
volumes,
firewall
use
in
its
gains
is,
is
available
and
we
are
planning
to
implement
that
so
that
applications
who
don't
want
to
use
file
mounds,
but
won't
you
have
some
object.
Storage
for
some
put
and
get
operations
or
even
and
is
the
second
bullet
point
says
even
the
registry
may
be
a
perfect
fit
for
some
object.
Interface,
improve
packet
support
for
the
registry.
C
It's
the
second
bullet
point.
So
these
are
the
very
high
level
objectives
for
the
future,
and
so
3.7
is
roughly
planned
for
autumn.
So
from
autumn.
Until
later,
you
can
expect
those
features.
Then
generally
I
mean
the
improved
day-to-day
maintenance
operations.
They
are
going
to
continue
over
the
next
couple
of
releases,
the
feature
separately,
it
much
more
complete
and
we
are
working
on
the
scalability.
So
one
thing
that
has
not
been
mentioned
explicitly
yet
is
scalability
is
not
yet
optimum
of
glasper.
C
It
has
a
certain
amount
of
memory
requirements
and
there
is
currently
a
limit
of
roughly
a
hundred
hundred
fifty
volumes
per
cluster
cluster
at
maximum
in
the
next
release
of
2.5.
This
will
already
be
improved
later.
It
will
be
improved
very
greatly,
so
that
will
be
possible
to
host
many
hundreds
to
maybe
even
thousand
or
a
few
thousand
volumes
on
a
single
cluster
cluster.
C
So
these
are
the
limitations,
but
we
are
working
on
improving
their
SSN.
2.5
will
already
be
improved
to
some
extent
due
to
300
volumes
likely
will
be
possible
and
2.6
timeline
will
have
probably
a
factor
of
five
to
ten
that
we
can
add
to
the
number
of
volumes
that
can
host.
So
this
is
roughly
the
roadmap
for
the
next
year
or
so,
and
with
that
I
think
we
can
conclude
the
presentation
and
move
on
to
last
slide,
which
is
basically
at
some
pointers.
C
You
can
also
see
the
the
links
in
the
previous
slides,
where
I
put
out
the
communities
of
of
cluster
of
a
KD
of
cluster
kubernetes.
All
those
have
github
presences
and
many
of
the
things
are
present
in
slack
or
on
IRC
on
freenode,
etc.
So
you
have
both
the
upstream
and
the
downstream
contacts
are
basically.
A
Really
good-
and
it
is
a
was
a
ton
of
information
and
I
know
we
didn't
get
to
the
demos
and
there's
a
few
people
commenting
here.
Can
we
bring
you
guys
back
to
do
more
demos,
another
time
because
they'd
like
to
see
it
in
action?
Yeah?
Really
it
people
just
thanks
very
much
for
coming
and
doing
this
today.
We
only
to
keep
these
things
to
an
hour
sure
when
three-point-five
comes
out.
A
B
Think
what
what
we
could
do
is,
after
3.5
comes
out,
we'll
actually
be
using
container
daily
storage,
which
is
cluster
running
inside
open
ship
back
open
ship
registry.
We
can
probably
do
a
demo
for
that,
plus
the
demos
that
we
missed
today,
so
you
know
so,
and
so
that
would
be
a
nice
session
to
do.
You
know
in
April
mid
April
or
May
whenever
we
can
schedule
this.
That.