►
From YouTube: GumGum: Multi Region Cassandra in AWS
Description
Speaker: Mario Lazaro, Big Data - Software Engineer
GumGum relies heavily on Cassandra for storing different kinds of metadata. Currently GumGum reaches 1 billion unique visitors per month using 3 Cassandra datacenters in Amazon Web Services spread across the globe.
This presentation will detail how we scaled out from one local Cassandra datacenter to a multi-datacenter Cassandra cluster and all the problems we encountered and choices we made while implementing it.
How did we architect multi-region Cassandra in AWS? What were our experiences in implementing multi-datacenter Cassandra? How did we achieve low latency with multi-region Cassandra and the Datastax Driver? What are the different Cassandra use cases at GumGum? How did we integrate our Cassandra with Spark?
A
Thank
you
so
welcome
everyone
to
the
region,
Cassandra
in
AWS
I
have
a
lot
of
content
to
go
through
and
only
40
minutes.
So
this
is
that
you
knew
it.
First
of
all,
for
my
my
name
is
Mario
terminal,
a
thorough
I
work
as
a
big
date
engineer
at
undone,
I
was
born
and
raised
in
Spain.
That's
my
weird
accent
come
from
and
I
joined
gong-gong
18
months
ago,
and
I
have
about
a
year
and
a
half
experience
working
with
a
son
during
production.
B
A
Do
you
focus
your
site?
We
have
other
product
like
in
screen
and
standard
veneers
native
ads
for
charter
yachts.
Now
let
me
give
you
some
numbers
that
we
are
really
proud
of,
like,
for
example,
where
the
pit
as
platform
in
the
entire
US,
we
were
with
more
than
2,000
premium
publishers,
sort
of
them
really
well
known
in
the
domains
and
just
to
give
you
an
idea,
we
reach
more
than
1
billion
global,
unique
visitors,
and
this
number
keeps
growing
and
growing
on
the
numbers
that
you
see
here.
They.
A
A
A
Why
we
decided
to
studies.
Cassandra
multi-region
in
AWS
will
go
through
the
big
challenges
that
we
face.
Their
testing
that
we
did
before
putting
in
production
modus
operandi.
Oh
we
did
is
in
production
the
real
work
sometimes
now
that
we
have
multiple
data
centers.
How
to
do
maintenance
deaths
is
depend
when
your
husband's
super
simple
on.
Finally
questions
answer,
so,
let's
go
back
to
much
2015
well
time.
A
Noggin
was
using
just
one
Cassandra
data
center,
one
cluster
with
one
data
center,
the
one
that
you
see
here:
25
classic
Knox,
cluster
and
holiday
in
one
region,
one
crap,
hasta,
teen,
AWS,
a
dollar
system
for
amazon
web
services.
We
use
amazon
and
source
cloud
provider
and
then
we
were
using
the
best
human
2.0.
A
was
the
latest
one.
When
we
build
it,
one
of
the
latest
one
I
mean
on
the
client
side.
We
were
using.
The
data
stack,
secure,
driver,
the
Java
better.
A
We
we
were
using
a
Cassandra.
We
still
do
present
at
the
store
inside
metadata,
which
includes
visitors.
Data
in
it
is
some
babies
later,
for
example,
things
that
we
have
been
able
to
recognize
in
an
image
using
our
patented
image
recognition
system
like,
for
example,
or
in
this
image.
There
is
a
girl
and
she's
blonde,
and
this
keywords
are.
We
are
dead
inside
all
kind
of
things
that
you
can
imagine
another
formats
to
keep
track
of
number
of
views,
number
of
clicks
number
of
everything
and
for.
B
A
Other
here
we
have
two
use
cases:
let's
go
with
the
first
one.
Third,
one
will
be
the
real
kind
use
cases
for
real-time
data
in
order
to
serve
as
we
need
data
that
we
sing
cassandra
and
we
started
literally
like
billions
of
rows
in
different
tables
case
basis.
So
it
means
that
per
day
we
create
billions
of
rows,
and
this
presents
are
any
rate
below
outside
around
sixty
percent
rates,
for
the
person
who
writes
and
something
very
characteristic
that
will
use
ptl's
everywhere.
Gtl
stands
for,
thankfully,
so
this
is
really
amazing.
A
We'd
only
to
be
data
as
I'm
just
gonna
do
it
by
itself,
but
you
really
need
to
understand
the
impact
of
using
the
deals
everywhere.
You're
going
to
eat
constance.
You
need
to
you
need
to
know
that
your
pump
action
strategy
is
gonna,
be
complete
ones.
Otherwise,
you
can
interact
with
thousands
of
some
stones
that
you
read
every
time.
You
request
something
and
your
baggage
is
going
to
increase
and
increase
crazy.
We
do
canyon,
critical
use
of
contacts
we
used
to.
Then
it
was
impossible
with
20
version
of
Cassandra.
A
Now
we
are
using
2.1,
we
use
contacts
again
and
with
a
high
percentage
of
this
heavyweight
world
is
because
we
do.
Our
ideas
are
the
reason
for
real-time
getting
us
for
these
people
not
familiar
with
the
other
Tyson
content.
Basically,
you
can
think
about
this
as
like
in
the
stock
market,
so
an
impression
good.
For
me,
one
of
our
body
cells
and
this
guy
wants
to
display
an
athlete.
A
Let's
say
this
place,
so
someone
in
this
case,
as
we
will
conduct
an
axiom,
will
go
to
the
stock
market,
went
up
an
axiom
and
different
advertisers
and
they
will
be
on
that
place
because
they
want
to
display
another
and,
after
that
one
of
them
will
win
and
we'll
be
able
to
this
Lydia.
So,
as
you
can
imagine,
this
needs
to
happen
lacking
nothing.
We
are
talking
about
us
total
execution,
10
or
15
seconds
here
and
in
order
to
see
itself
our
video
ads,
we
need
to
contact
person.
I
get
some
data.
A
Talking
about
the
analytics
use
case,
basically,
we
run
daily
et
al
jobs
to
strap
join
data
from
Cassandra
and
backing
match,
2015
you're
using
Hadoop
MapReduce
Europe
to
do
that
and
press
for
sauna
hot
worries.
So
now
that
we
know
the
status
in
march,
this
talk
about
the
international
expansion.
In
month
2015
we
decided
to
open
a
new
data
center
amazon
in
ways
just
because
we
saw
as
I
say
we
keep
growing
and
growing.
We
saw
a
really
big
business
opportunity
there.
A
So
we
say:
ok,
the
star,
wonder
the
steps
have
been
us
from
Europe
to
you.
That's
why
that's
when
this
international
ransom,
it
happened,
Afghan,
town
and
at
the
very
beginning,
we
we
just
did
some
brainstorming.
What
we
need
to
do,
they're
doing
it
to
put
casts
on
another,
it's
really
necessary.
Maybe
we
just
need
to
put
yes.
Everything
is
that
the
data,
yet
we
are
talking
about
Cassandra
under
10
enough
homes.
We
just
need
to
go
to
Virginia
us
is
get
the
data
boba.
A
This
is
not
gonna
work
right,
our
TV
total
execution
times
within
a
second
no
possible,
so
children
starving.
We
decided
that
we're
going
to
open
is
at
the
center
with
Cassandra
there,
and
we
are
going
to
use
the
AWS
will
12
ready
flowers
like
the
next
version
of
the
classic
club.
It
basically
gives
you
more
control
over.
Your
networking
is
more
secure
to
the
next
version,
although
we
know
that
this
was
gonna
presents
and
collectively
challenge
is
not
easy
for
us
survey
also.
A
To
replicating
their
data
with
the
same
number
of
replicas,
so
everything
that
we
have
in
asthma
and
your
sis
with
Gina,
it's
gonna,
be
in
a
waste,
a
iron,
and
this
is
the
first
step
with
Sally
to
the
specs
on
Cassandra
test
at
the
centers
in
your
silly
us,
and
therefore
this
Cassandra
material
works
in
AWS
Branson
capacity
performance
test,
because
we
want
to
yes
build
the
new
Cassandra
cluster
that
we're
gonna
use
for
the
entire
year.
We
expect
three
times
more
traffic
in
q4
and
we
don't
worry
kiefer
to
go
there
after
scalability
issues.
A
Add
more
nose
remove
known,
we
just
want
to
run
soon.
So
this
is
look
simple
right
leg.
It's
doesn't
age
data
centers.
It
works
both
through
the
documentation
to
go
to
be
true.
How
many
times
you
have
won't.
You
have
returned
a
bachelor
mutation.
Let's
say
eight
points.
He
says
you
follow
this
guy
this
to
buy
points
a
and
then
at
the
end,
you're
shocked
war
is
going
to
be
up
and
running.
Writing
it
never
happens.
So,
let's
talk
about
the
big
challenges
that
we
face
their
third
one
problems
between
cassandra.
A
Is
it
too
classy
BBC
and
it
does
s
taxi
a
bad
driver?
First
of
all,
I
encourage
you
to
use
the
easy
to
move
the
originals
niche.
You
are
in
amazon
and
you
are
going
to
have
multiple
data.
Centers
use
this
one
and-
and
you
cannot
go
with
the
situation
it-
that
one
use
private
up.
Is
this
one
use
public
at
ease?
But
what
happens
here
easy
two
instances:
don't
have
a
interface
with
a
public
IP
address,
which
means
that
note
span
opponent
between
each
other
in
the
city.
A
They
are
innocent
region
with
public,
at
least
it's
not
possible.
So
how
to
fix
this
whole
deal
with
these
well
you're,
going
to
need
to
understand
who
a
center
notes
interact
with
each
other,
which
ones
they
use,
and
you
are
going
to
need
to
play
with
the
security
groups,
their
trash,
those
IDs
for
notching,
the
same
region,
the
public
at
ease
of
these
guys
and
just
play
you
need
to
transfer
in
there.
This
is
an
example
so
I,
I
just
started
one
Cassandra
test
at
the
center.
Only
one
with
easy
to
moody,
Rihanna.
A
B
A
My
class
was
up
and
running
what
was
happening
with
might
know
they
were
down,
leaving
the
word
perfect,
you
know
that's
because
of
the
Public
ethics
issue,
certainly
Reagan
to
really
feel
it
will
use
public
IDs,
it's
okay.
What
we
decided
to
do
is
just
trust,
those
IPS
from
one
area
to
the
other
region
and
everything
with
scalar.
We
just
need
to
add
a
new
IP
there.
I
think
you
can
do
a
company's
do
is
do
so.
What
are
what
we
hear,
but
it
works.
A
Fine,
yes,
public
in
case
you
can
put
this
Australian
that
it
just
works
and
finally,
from
the
client
side.
Well,
your
application.
In
the
same
radio,
when
is
going
to
contact
us
online
this
invasion,
he
needs
to
use
private
eye
peas.
You
cannot
expect
at
least
because
of
the
same
reason
I
said
before
right
how
this
works.
Well,
at
the
beginning,
you
will
astound
you
and
your
client
and
this
one
will
try
to
connect
to
Cassandra.
We
have
hosted
you
be
you
get
to
him,
and
this
one
is
probably
is
gonna,
be
a
private
IP.
A
So
it's
going
to
go
to
Cassandra
one
person
or
no
and
this
person
a
node
is
gonna
return,
the
least
entirely
sub
notes
in
that
cluster.
But
this
one
is
going
to
be
all
public
ip's.
We
cannot
use
public
athlete
right
so
totally
with
this,
you
need
to
implement
the
address
translator
interface
that
comes
with
Cassandra
datastax.
That's,
basically,
what
is
going
to
do
is
translate
these
IPS
from
ballet
puppies
to
private.
That
means
on
the
fly
and
then
your
client
is
going
to
be
able
to
connect
to
all
the
customers.
A
Other
big
challenge
here
and
talking
about
the
data
stabs
gave
a
regular
load
balancing
it's
amazing.
Basically,
in
the
stacks,
you
have
a
client.
It
allows
you
to
specify
hope
you
want
your
plug-in
to
connect
to
your
Cassandra
knows,
which
now
should
I
use
right.
There
are
multiple
choices
and
when
you
go
with
you
region,
you
check
them,
click
bunch
of
them
and
what
we
decided
is
to
use
a
DC
aware,
which
is
the
descent,
are
aware
that
tells
Mike
Lane
an
egg
with
this.
Is
your
local
data
center
use
this
one?
A
B
A
A
Well,
my
image,
you
have
your
web
app
in
three
different
availability
zones,
sign
of
the
in
180
that
in
one
beats
another
new
one
see
and
also
you
have
a
cassava
that
the
center
is
finding
three
availability
zones
with
repeaters,
so
sometimes
will
be
180
notes
for
me
noticing,
1c
and
because
we
are
using
network
topology
strategy,
my
replica,
whatever
you
guys
want
to
be
in
one,
be
next
a
18
next
in
one
be
next
in
watching.
So
a
request
comes
from.
The
internet
goes
through
the
load
balancer
and
hits
one
server
that
will
happen.
A
That
needs
to
contact
this
person's
a
data
center
and
because
we
are
using
also
talking
about
knee-
and
it
knows
that
my
data
is
one
of
these
nodes,
but
if
we
haven't
about
it
so
now
we're
gonna
choose
this
guy
will
try
to
get
the
data
from
the
node.
In
one
be
so
we
have
already
expended
workbooks
that
the
level
is
gonna
increase
right.
This
is
not
another
steps.
We
are
reading
it.
What
we
call
the
wrap.
A
A
So
next
challenge
we
talk
about
ourselves,
the
defend
at
the
center.
We
decided
to
create
an
extra
data
center
just
for
analytics,
because
we
keep
running
more
and
more
easier
jobs
every
day
and
we
don't
want
them
to
impact
the
real
tangy
tart
this
right.
So,
in
order
to
do
this,
we
decided
to
put
a
spell
on
top
of
cassandra
rose
for
this
analysis
class
data
center.
B
A
A
B
A
Now,
let's
go
next
challenge,
we
decided
to
also
you
know
we
are
doing.
We
are
going
with
your
ad
on
starting
new
data
centers.
This
was
to
upgrade
from
20
a
2211.
Why?
Well,
because
contrast
implementation
in
3,
2.1,
batsman
casilla,
is
ladies,
his
body.
Did
they
don't
work
silly
like
crazy,
and
we
know
that
in
2.1
they
they
rebuild
them
from
scratch
in
the
way
they
work
is
completely
different.
A
So
we
said:
okay,
let's
try
them
we'll
talk
about
this
a
bit
later
and
next
challenge
to
choose
for
not
to
choose
which
one
knows,
as
I
said
before
we
were
using
classic
knows,
but
just
because
classic
notes
they
were
fine
with
Hadoop.
If
you
use
will
to
a
notes-
and
let's
say
you
have
many
notes
like
50
notes
and
you
use
256
token,
trainees
basically
means
that
Hadoop
is
not
gonna
work.
You
are
going
to
run
into.
A
We
are
issues
like,
for
example,
too
many
open
files
connection
issue
is
crazy,
so
previous
one
was
using
MapReduce
jobs
and
it
adds
some
complexity
for
a
dinner
moving,
because
now
you
need
to
manage
the
top
and
ranges
yourself
right.
That's
not
good
in
the
new
one,
as
you
can
imagine,
we
decided
to
use
build
robots.
Why?
Because
we
are
going
to
3
fo
50,
we
are
going
to
just
five
and
we
know,
despite
works
fine
with
volcanoes,
it's.
A
And
also
it's
easy
to
watch
remove
new
nodes
as
traffic
increase
some
bad
side,
these
repairs
they
take
forever.
You
have
let
say:
300,000
ranges.
It's
ok
and
now,
let's
go
with
the
testing
offense
before
jumping
into
testing
I,
encourage
you
to
us
an
ultimatum
protector
because
they
still
advised
that
creating
a
modified
in
any
Cassandra
nose
and
these
tasks
like
sub-,
passes
and
consuming.
A
So
what
are
amazing?
The
box
beam
width
is
like
create
a
fully
automated
process
for
creating
modifying
the
strain.
Do
everything
you
want
for
can
with
Cassandra
classes
using
uncivil
so
myself
that
I
know
that
the
Box
guy
I
was
able
to
start
a
a
Sandra
cluster
with,
let's
say,
five
data
centers
in
different
regions
of
the
world
by
just
changing
some
settings
here
in
this
template,
I'm
just
clicking
one
button
after
20
minutes,
I
was
able
to
start
playing
with
this
test
data
center,
and
now,
let's
jump
into
the
testing
platform
on
face.
A
As
we
did
there,
we
run
some
performance
tests
using
the
Cassandra
2.1
stress
tool,
which
is
it's
amazing,
allows
you
to
recreate
your
scheme.
As
you
can
see
here,
we
regulate
one
of
the
tables
that
will
have
digital
one
same
number
of
columns
that
we
have
following
the
same
distribution
for
the
size
and
this
many
requests
and
writes.
Arenanet
worry.
This
is
extremely
useful
and
super
easy
to
use
and
you're
not
gonna
play
with
real
data
by
but
close
to
cope
with
span
of
this
Cassandra
cluster
with
navarasa.
A
We
thought
we
were
going
to
need
and
we
just
try
to
find
instead
right,
Ronen
stress
too.
This
is
ok,
we
need
more
places.
Minaj
grammar
test
was
happening
there.
Also,
while
doing
this
performance
test,
we
wanted
to
test
this
replication
oversees
the
Latin
see
how
this
is
going
to
work
right
because
he's
like
Maggie
looks
like
money
for
us.
So
what
I
can
tell
you?
Is
this
replica
sword,
simple
nervously?
It
doesn't
impact
Latin
see
it's
just
amazing.
A
Another
doesn't
that
with
12
replicas
only
want
a
certain
of
will.
Contact
will
send
this
to
the
other
data
center.
Also
in
your
client,
you
need
to
use
local
consistency
levels
because
you
don't
want
clients
in
let's
say
Virginia
conduct
Cassandra
to
get
the
data
and
wanted
the
Cassandra,
the
boys
hockey
nya,
and
also
we
were
like
okay.
This
works
replication
of
a
cig
works,
but
when
I'm
going
to
be
able
to
read
something
like
two
retaining
island
from
us
east
right,
so
we're
about
to
run
some
tests,
then.
B
A
Then
we
show
this
amazing
blog
again
Netflix
and
that
they
were
worried
about
this.
They
run
st.
Asaph.
We
were
about
to
run
using
amazon,
so
they
started
one
person.
Data
center
in
USC's
know
that
that
the
center
in
u.s.
west-
and
they
did
1
million
rights
with
consistency
level,
one
here
in
the
thing
and
you
assist,
and
then
they
prove
that
they
were
able
to
read
through
1,000,000
rates
after
500
milliseconds
with
consistency
level,
one
with
zero
data
loss.
It
means
that
in
500
milliseconds
all
the
data
was
aggregated
and
this
works
for
us.
A
This
is
more
than
enough
more
than
we
need,
so
we
decided
to
continue
our
instance
hide
with
threshold
kind
of
with
tests,
all
kind
of
instances
that
amazon
provides
you
and
we
decided
to
go
with
our
three
to
attract
machines.
They
have
60
gigabytes
program
and
course
160
gigabytes
of
the
criminal.
This
is
the
storage
that
we
use
for
the
common
dogs
and
safe
passage
and
a
great
0
/
4
s's
of
the
EBS
volumes
that
we
use
for
the
data.
This
is
the
best
juice
case.
Is
the
performance
on
demand?
Use
cases?
A
A
Is
being
right,
but
we
decide
to
you,
know:
just
go
without
a
32
inch
legs
and
if
you
need
more
machine
will
be
left
they
played
and
we
talked
about.
Well,
we
test
this
a
great
hope
to
do
disagree
from
208
to
2.15
and
what
wishes
that
in
both
versions,
can
go
a
bit
in
the
same
data
center
and
is
that
you
can
roll
in
a
plate.
A
simple
answer
is
still
a
simple,
so
this
was.
We
saw
that
this
was
not.
A
It
was
not
going
to
be
an
issue
while
doing
this
in
production,
and
we
say
to
test
these
new
features
that
are
in
2.1
right,
like
we
want
you
to
use
like,
for
example,
day
that
data
here
compaction
strategy,
and
we
have
time
series
data
we
tested
it.
We
will
love
it.
So
we
say
that
involution
at
some
point
we
will
use
it
incremental
repairs.
We
know
that
repairs
are
a
headache
amor,
if
you
use
with
your
notes.
This
helps
your
usual
tries
still
in
simple
and
context
they
work.
A
B
A
A
Any
ones
that
we
want
with
all
the
settings
that
we
want
and
at
the
end
we
are
going
to
get
rid
of
the
old
one
with
zero
downtime
first
step
here
upgrade
the
old
cluster
snitch
from
a
situation
it
too
easy
to
multi
region
is
neat.
You
can
do
this
in
a
Berlin
upgrade
is
still
easy.
It
worked.
Listen
to
make
sure
that
you
play
with
the
security
group
jutros
dos
and
tres.
A
This
is
also
a
great
acclaim
is
200
the
translators
that
we
talked
about,
and
it's
just
really
easy
to
do,
but
we
found
this
crap
while
doing
this
in
production,
and
basically
you
need
to
make
sure
that
your
clients
don't
loose
connection
to
upgrade.
Cassandra
knows
we
open
a
ticket,
I
think
is
going
to
be
fixed,
excellent
kiss.
This
is
crazy
because
you
do
the
role
in
a
break,
and
you
one
next
one
project,
your
kitchen
consistency
later
one
and
you
don't
see
anything,
that's
always
telling
you
that
everything
is
fine.
A
When
your
clients
are
not
able
to
reconnect
to
discuss
and
ran
off
to
the
untrained
one,
it's
like
it
doesn't
work
they
can.
There
is
a
back
there,
so
what
that
is
like
you
do
you
continue
to
link
the
role
in
a
play
and,
at
some
point
everything
press
underneath
I
mean
my
moving
production
rider
like
oh,
my
god,
sobbing
here.
Yes,
you
discard
your
clients
and
everything
works,
so
make
sure
that
this
doesn't
happen
to
you
second
step,
I'm
grady
all
data
center
from
2082
2015
things
that
you
need
to.
B
A
I
say
rolling,
upgrade
and
just
run
the
tool
updates
tables.
Yes,
because
2.15
nodes,
you're
gonna,
be
able
to
reduce
stables
in
an
old
format,
but
when
it
comes
to
string
in
detail
and,
let's
say
a
repair
and
they
need
to
stream
data,
they
are
not
going
to
be
able
to
do
that.
They
are
not
going
to
be
able
to
read
these
new
blue
filters
and
metadata
for
this
is
stable
and
we
are
not
crazy.
A
You
can
think
that
why
you
are
doing
this
in
your
class
and
if
you
are
going
to
get
rid
of
this
one
right,
what
is
because
it's
not
possible
to
reveal
to
get
the
old
data
and
to
a
coupon
100
of
prana
2.0
the
center.
You
will
try
to
do
this.
A
guy
give
me
the
alldata
I
needed
different
versions.
Boom
exception
is
obsessing.
A
It's
not
gonna
work
on
thur,
step,
we're
going
to
start
and
you
waste
a
new
US
data
centers
within
the
same
cluster.
At
the
beginning,
we
are
going
to
put
replication
factor
f
0
for
the
new
guys.
We
just
want
them
to
a
start
and
make
sure
that
everything
works.
Fine.
There
no
data
is
going
to
go
there.
No
new
data,
then
we're
gonna
play
with
the
DC
surface
as
a
hard
to
differentiate
between
new
Virginia
data,
centers
and
the
old
one.
Otherwise
they
will
join
Shane
green,
a
new
guinea.
C
B
A
Cassandra
know
that
they
exist.
That's
all
after
that.
We
are
going
to
change
the
replication
factor,
2
3
4
their
new
real-time
data,
centers
and
14.
Underneath
we
just
need
one
replica
there.
It
means
that
they're
gonna
start
receiving
new
data,
we're
gonna
start
educating
new
data
or
repair
theta,
and
after
that
we
are
going
to
do
a
note
to
rebuild
from
your
data
center.
To
get
me
all
data,
for
this
looks
like
well.
B
A
The
beginning,
we
have
four
clients
on
all
glass
there
with
replication
factor:
three,
that's
all
we
have.
We
are
gonna
and
these
new
data
centers
one
in
USC's,
no
one
in
Ireland
to
the
notes
and
another
one
in
Josie's
for
analytics.
Then
we
are
going
to
change
the
replication
butter
from
0
0
2
331.
It
means
that
women
are
replicating
new
data
here
here
and
also
here,
but
just
one
Radhika
for
I
analytics
and
after
that
we
are
going
to
start
rebuilding
them.
B
A
Get
your
data,
so
you
can
trigger
multiple
reveals
at
the
same
time.
It's
okay!
Yes,
make
sure
that
you
don't
saturate
your
network.
You
don't
want
that
to
happen.
Believe
me,
and
this
process
takes
time.
We
are
we're
transferring
like
terabytes
of
data.
That
is
ok,
so
we
reveal
them
all
of
them.
You
waste
all
of
the
non
same
time.
We
don't
get
about
high
CPU
yeah,
like
grand
notes,
because
remember
or
planes
are
talking
to
the
old
passenger
class
said
we
don't
get
about
legacy
here.
A
Yet
after
that
we
are
going
to
start
using
this
new
data
centers.
So
it
means
that
our
clients
are
gonna.
Change
on
clients
in
Europe
are
gonna
start
using
the
Cassandra
data
center
in
Europe
in
Ireland
and
clients
in
the
rest
of
the
world.
I'm
gonna
start
using
us
is
real
that
integrate
the
data
after
that
we're
gonna
change.
The
replication
first
out
of
the
old
cluster
20
just
make
sure
that
we
don't
replicate
data
that
we
are
going
to
get
rid
of
that
light
run,
and
we
don't
understand
the
Commission
minute.
A
I
recommend
you
to
do
this
and
not
go
to
your
control
and
just
terminate
your
instances,
because
you
don't
want
winter
in
a
bust,
go
see
the
stake.
That's
really
bad
takes
like
days
to
clean
and
also
say
because
now
a
replication
factor
is
zero
in
your
class
ten.
So
we
are
not
going
to
stream
real
data
to
the
rest
of
the
notes
of
these
data
center.
A
A
We
will
this
piece
observer
reduced
uncibal
to
trigger
the
maintenance
operations
like
cleanups
repairs,
compulsions,
and
if
you
saw
Cassandra
maintenance
k
space.
We
put
the
data
then
like
we
have
repaired
this
node
at
this
time,
and
we
have
done
these
things
there.
It
has
take
this
much
time
so
then
ansible
know
that.
Okay,
I
read
from
Cassandra,
I
get
the
data
and
then
I
know
that
my
next
note
that
I
need
to
do
a
repair.
A
Other
tapes
I
say
that
now
we
are
using
a
spell
right
now
do
well
things
that
we
have
learned
is
that
the
number
of
workers
can.
We
are
both
the
number
of
total
Cassandra
notes
in
urinal.
It
is
cluster,
it's
ok,
you
don't
care
about
blood
and
see
them.
Ok,
each
worker!
For
us
these
works
use,
one
for
number,
of
course,
of
each
instance
and
one
third
of
total
valuable
run
of
each
instance.
A
Remember
that
we
have
a
spell
on
top
of
Cassandra
things
that
we
have
learned
about
the
spoon
while
using
the
ascenders
para
conectar.
Is
that
you
don't
want
to
use
reduced
wife.
Believe
me,
I
do
not
want
to
start
like
sucking
data
terabytes
and
instead
uses
fan
by
simple
it
just
work
with
partition.
So
you're
not
going
to
stop
all
data,
don't
with
us
another
table.
We
love
this
one.
It's
like
okay,
I
need
to
get
the
day.
I
have
these
terabytes
of
data
coming
from.
A
Let's
say,
elastic,
sent
and
I
need
to
get
information
that
is
in
Cassandra
fun,
but
only
for
these
cuts
right
young
with
Cassandra
table,
is
as
19
and
he's
going
to
contact
us
under
two
bit.
Yes,
the
data
that
you
need,
it's
really
simple,
and
then
there
are
bunch
of
settings
that
I
recommend
you
to
go
through
the
commendation
ready-mix.
A
What
that
you
understand,
then
and
play
with
them,
because
you
can
improve
performance
by
ago,
like,
for
example,
here
are
two
example:
the
batch
size
you
can
play
with
that
again,
you
don't
get
about
that
is
in
here,
but,
for
example,
the
number
of
concurrent
lines,
it's
okay
and
something
really
important
is
that
remember.
We
are
using
easy
to
enter
in
a
snit
here,
so
you
are
going
to
do
the
translator
team
also
for
yours.
A
Perhaps
it
means
that
you
just
need
to
implement
a
special
connection,
factory
set
it
in
the
bathroom
and
then
a
spam
will
be
able
to
connect
to
all
your
Cassandra
notes.
Otherwise,
your
span
yolks
are
gonna,
run,
probably
but
they're
gonna,
reheat
1%
Renault,
and
that's
something
that
you
don't
want
and
since
Cassandra
in
you
West
Island.
This
is
the
status
of
a
cluster.
We
have
one
in
Ireland.
We
have
us
East
data
center,
more
notes
there,
the
analytic
one
and
because
we
went
through
this
pain-
and
we
know
that
this
process
works.
B
A
Month
to
complete,
we
open
the
new
data
center
with
Cassandra
they're,
letting
guess
one
week,
so
we
obey
us
waste
at
the
center,
with
Cassandra
left
two
steps
having
us
worst
traffic
from
news
with,
and
that's
all
our
time
for
questions
and
answers
and
Anthony
that
Gundam
is
hiding.
You
just
need
to
go
to
this
page.
You
will
see.
We
have
no
the
engineer,
open
positions
with
it's
an
amazing
place
to
work.
Believe
me
and
well.
I
hope
you
enjoy.
E
B
A
A
That's
up
like
you,
don't
want
that
to
happen.
You
can
end
up
with
housing,
offices
tables
or
a
table
not
getting
compactly.
That's
all
that's
why
I
say
that
we
tested
it
and
we
love
it,
but
still
a
singer
kim
production
still
in
our
test
data
centers,
because
we
know
that
there
are
other
stack
seasonings
and
improvements
there.
We
just
want
to
wait
and
then
we
will
take
a
bit
but
the
content
itself.
It's
super
simple
and
when
testing
it
was
like
perfect
lethargy
was
way
lower
than
using
like
sitesteer
compulsion
is
13,
for
example.
A
A
D
D
A
A
Increase
like
Amazon
goes
crazy.
Okay,
we're
gonna,
be
delayed
until
traffic
toilet
right
or
now
to
us
wish,
like
width.
Is
that
like
I,
don't
know
if
you
guys
use
Amazon,
but
this
week
they
did
some
funny
games
with
dynamos
like
cool
things.
Thank
you.
Everything
was
down.
It
was
like
23
when
they
come
mess
and
we
really
like
things.
I
do
its
traffic
to
only
do
is
question.
B
C
A
With
that
yeah,
okay,
so
the
question:
ever
it's
a
what
what
happens?
You
would
have
some
personal
machine.
You
know
that
they
don't
want
they
later
to
leave
Europe
right
well,
for
the
dissenting
you
assist,
we
don't
have
that.
We
are
we're
lucky
right,
so
not
really
I
mean
they.
They
don't
think
of
an
aspect.
We
don't
have
that
security.
We.
B
A
Yes,
so
this
uncivil
class
Cassandra
a
space
plus
a
miracle
is
gonna:
do
it
right
itself
in
order
the
in
10
days
and
need
to
repair
discussion
dr.
step?
I,
don't
want
to
resurrect
a
delete
theta
by
theta,
so
you
know
that
okay
I
need
to
trigger
more
on
this
time,
like,
for
example,
when
we
are
here
in
Europe,
is
9,
so
he
knows
that.
Okay,
now
I
can
play
with
euro
then
night
time
here
I
can
play
with
your
system
its
handling.
All
these
games.
A
I
know
there
are
no
venture
to
the
chief
all
announced
here
and
the
cash
understand
me.
A
netflix
know
it's
fortified.
They
they
have
done
this.
It's
our
venture
cool,
a
taco
bell,
sauce
it,
but
the
difference
is
same
content,
but
they
have
you
seen
my
sequel
toaster.
This
video's
Cassandra
I
mean
why
not
right
I
mean
this
is
cassandra.
Maintenance
operation
from
is
Eustace
under,
although
that
one
has
20
more
features,
fresh
idea
and
stuff.
E
A
The
question
is
why
evs
and
this
place
here?
Well:
it's
because
this
is
these
stringy
expensive!
That's
why
I
say
like
if
you
want
pure
recently
upendra,
you
should
go
with
a
cho
and
that's
that's
the
best
option.
Then
you
have
like
eight
hundred
gigabytes
offices,
the
fishiness
we
get
referred
to
her
more
know,
and
that
way
we
know
that
these
are
trained
to
attract
instances.
They
have
found
160
gigabytes
of
the
final
study
we
use
for
the
coming
docs
and
safe
passage
for
today's
ICS
EBS,
which
works
fine
for
us
SSD
yeah.