►
From YouTube: This Week In Cassandra: 3.0 in the Wild 5/13/2016
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
You
know
it
would
be
really
sweet
if
I
had
like
a
guitar
solo
to
start
out
every
one
of
these
things
all
right
here.
We
are
this
week
in
Cassandra.
What's
today,
it's
May
thirteenth
come
into
the
world
with
the
Cassandra
news
we
also.
So
what
do
we
got
here?
I'm,
John,
Haddad,
right,
technical
evangelist
for
datastax
we've
got
Luke
Tillman.
Also
technical
evangelist
for
datastax
lovely
to
see
you
Luke
and
Julian
are
gonna.
He's
the
vp
of
engineering
frog
to
cloud
I'm.
Sorry,
I
can't
actually
pronounce
your
last
name
correctly.
It.
A
I
can't
pronounce
everybody
is
the
last
name
thanks
for
thanks
for
coming
on
Julian.
Thank
you
guys.
So
today
we're
going
to
be
talking
about.
We
got
some
Cassandra
news
and
we're
going
to
be
talking
about
Julian's
experiences
with
putting
Cassandra
30
into
production,
so
new
new
ish
kind
of
release.
A
You
know
getting
getting
ahead
of
the
ball
having
a
good
time
in
the
open
source
world,
but
first,
let's
talk
about
materialized
views,
so
Jonathan
Ellis
need
for
patchy,
Cassandra,
all-around,
good
guy,
just
posted
a
blog
post
and
datastax
blog
about
materialized
views
on
their
performance.
Look
you
had
an
opportunity
to
let
this
over.
Would
you
think
so.
B
It's
a
it's
cool
to
actually
see
some
performance
numbers.
You
know
because
I
know
a
lot
of
people
you
know
are
interested
in
materialized
views,
I.
Think
just
from
a
like
usability
standpoint,
you
know
not
having
to
do
all
the
manual
denormalization
and
manage
that
in
your
code,
anymore
is
pretty
attractive,
I
think
for
for
a
lot
of
developers,
but
it's
it's
also
interesting
to
see
the
you
know
the
performance,
numbers
and
and
kind
of
gauge.
B
You
know
not
only
what
kind
of
the
impact
does
this
have
on
your
rights,
but
then
you
know
it
was
I
was
actually
kind
of
shocked
or
surprised.
You
know
how
you
know
some
of
the
reed
perform
numbers
that
he
showed
in
the
in
the
blog,
so
cool
to
actually
have
some
numbers
and
a
tool.
You
know,
that's
that's
an
open
source
that
you
can
go
and
test
for
yourself.
If
you
want
to
try,
you
know
see
what
your
own
performance
is
like.
This.
A
Think
the
number
one
thing
that
people
kind
of
get
hung
up
on
is
you
know
secondary
indexes,
and
the
reality
is
that
most
of
the
time
you're
going
to
actually
going
to
want
to
use
materialized
views
if
you're,
especially
if
you're
selecting
from
a
single
partition-
and
you
know
it's
cool-
to
see
that,
though,
that
performance
differentiation
as
the
closer
it
gets
bigger
you
can
see,
it
realized
view
performance,
get
better
and
you
know
secondary
indexes
kind
of
level
off.
You
know
yep
feeling.
B
C
So
basically,
I
think,
as
you
said,
we're
interested
in
the
usability
part
of
the
materialized
view
to
you
know,
have
a
better
or
cleaner
that
I'm
data
model
and,
as
you
said,
we
were
like
no
waiting
for
performance
performance
benchmarks.
You
try
to
basically
like
put
that
in
place,
so
we
actually
have
planned
in
internally
with
some
of
our
table
layout
to
migrate
and
basically
do
some
testing.
So,
as
you
said,
awesome
that
we
start
having
members
showing
up
and
the
tuning
to
actually
measure
for
ourselves
the
actual
drama
performances.
C
A
B
So
John
John
John
wrote
it.
You
wrote
a
blog
post
this
week,
surprise
yeah,
so
working
relationally.
You
actually
are
on
this
list
twice
this
list
of
blog
posts,
if
you're,
if
you're,
following
along
in
the
blog
post
on
planet,
Cassandra
and
so
John
and
Danny,
traphagen
ridah
wrote
a
post
together
and
then
John
wrote
one
himself
about
working
relationally
with
Cassandra,
and
so
this
was
kind
of
about
using
spark
and
Cassandra.
You
know,
but
what
that's
SQL
spark
SQL.
So
why
did
you
even
like
why
write
it?
A
I
I
will,
let's,
let's
touch
on
one
thing,
so
the
the
previous
post,
the
first
one
with
Danny,
that's
a
kind
of
a
preview
as
to
what
this
kind
of
stuff
that
we're
going
to
be
talking
about.
I
was
calling
which
is
going
to
be
next
week.
So,
if
you're
watching
this
and
you're
going
to
be
an
oz
con
and
you're
interested
in
learning
about
standard
spark,
we
sell
a
few
spaces
left
in
our
three
hour.
Tutorials
can
be
hands-on.
You
get
a
vm,
you
can
play
with
stuff.
A
A
We
hang
out
and
take
my
shift
shots,
which
will
be
amazing,
the
the
second
part
with
with
the
sparks
equals.
So
the
reason
why
I
wrote
this
post
is
there's
a
lot
of
really
good
documentation
in
the
spark
world
and
then
there's
some
stuff,
that's
poorly
documented
and
unfortunately,
the
spark
like
using
this
spark,
sequel
and
like
the
hive
like
side
of
things
is
not
documented
well
at
all,
and
I
like
people
would
like.
Datastax
has
done
a
lot
of
work
to
kind
of
make
this
stuff.
A
You
know
work
well
and
straight
in
a
straightforward
manner,
but
outside
of
that
there's
like
no
documentation,
so
I
took
a
look
at
what
was
going
on
and
poked
around
and
made
it
so
that
you
know
you
can
take
the
connector
and
you
can
fire
up
the
sparks
equal
shell
and
basically
create
tables
and
sparks
equal,
that
map's,
your
casino
tables,
and
then
you
can
do
like
joins
and
aggregations.
And
you
know
some
of
the
more
like
hive
style.
B
A
lot
of
people
probably
realize
you
can
SQL
against
your
Cassandra
cluster,
or
you
know
even
easier
if
you're
using
datastax
enterprise,
but
because
we
don't
have
to
do
all
the
setup
but
yeah
and
you
can
absolutely
use
you
spark
with
Cassandra.
So
one
thing
I
noticed
in
the
blog
post.
That
I
definitely
wanted
to
ask
you
about
so
you're
talking
about
sort
of
exploring
this.
This
movie
lens
data
set
and
as
set
up
you
do
this
this
step
where
you
do
a
pip
install
and
then
and
then
you
do
a
CM
install.
B
A
So
I
kind
of
yeah,
I
sort
of
like
glossed
over
that
and
it's
like
a
pretty
big
point
and
I
had
a
previous
blog
post,
which
kind
of
showed
it
off
a
little
bit,
but
the
Cassandra
dataset
manager
is,
if
you're,
looking
to
learn
Cassandra
right
now.
What
normally
happens
is
you
download,
Cassandra
and
then,
like?
A
What
we
need
is
a
way
to
install
sample
data,
sets
for
the
purposes
of
learning
and
so
I
built
this
tool.
Cassandra
data
set
manager
which
effectively
treats
Cassandra
data
kind
of
like
how
you
would
treat
like
packages
in
debian
or
in
Red
Hat.
So
you
can
just
do
CDM
install
and
you
give
it
the
name
of
the
data
set
and
it
will
just
load
your
Cassandra
cluster
for
you
and
it's
it's
just
really
nice
to
be
able
to
use
a
tool
like
this
in
blog
posts,
so
I
don't
have
to
come
up
with.
A
You
know:
okay,
we're
going
to
install
this
data
model
and
here's
some
sample
data
and
blah
blah
blah
blah
blah.
This
is
nicer
to
go
we'll
just
install
this
and
you
get
like
you
know,
hundreds
of
thousands
of
Records,
as
opposed
to
like
four,
which
is
what
normally
happens
when
you're
talking
about
data
set.
So
how.
B
Is
so
how
are
packaged
like
so
you
did
CDM
installed
movie
lens
small.
My
car
packages
actually
manage.
Do
you
have
a
registry
set
up
or
what
are
you
doing
under
the
covers
there?
Yeah.
A
It's
it's
kind
of
a
little
harry's.
Well,
I,
don't
know
about
here:
let's
go
with
fun,
so
what
what
I
do
is
each
each
data
set
is
really
get
repo
and
the
CDM
maintains
its
own
list
of
data
sets.
So
you
can
just
like
a
apt
repo.
You
could
have
apt
CDM
update
and
it
will
fetch
the
latest
list
of
repose
and
those
will
just
basically
have
descriptions
and
text
information
about
them,
along
with
a
reference
to
a
git
repo
that
you
can
download.
A
Are
curated
small
batch
data
and
the
idea
here,
though,
is
that
you,
if
we
can
get
enough
data
sets
in
like,
for
instance,
movie
lens,
we
can
start
to
build
machine
learning,
tutorials,
right
and
they're
they're
nice,
because
they're
reproducible
you
can
just
download,
you
know
you
can
download
the
you
can
stall.
The
data
set
I
think
you
can
download
some.
A
You
know
Jupiter
notebook
or
some
sample
code,
or
maybe
you
can
download
a
net
application
which
does
it
one
way,
and
you
know
you
can
kind
of
follow
along
and
when
it
comes
to
learning
like
having
20
tutorials
that
all
use
the
same
set
of
data
models
or
the
same
yeah.
The
same
set
of
data
models
gives
you
kind
of
a
lot
of
flexibility
there.
So
I
know
that,
like
Danny
on
Co
speaking
with
oz,
Khan
is
going
to
be
writing.
Some
data
sets
to
do
like
look.
A
A
lot
of
medical
research
like
there's
like
an
open
diabetes,
related
data
set
and
she's
going
to
pull
that
one
in
deceit
again
things
like
cancer
rates,
and
you
know
various
economic
data
sets
like
we
can
have
some
really
really
cool
things
and
then
Patrick
is
working
on
getting
killer
like
we
were
messing
around
with
killer
weather
and
we've
kind
of
talked
about
killer
video
getting
support
fat
in
there.
A
B
C
We
do
the
analytics
directly
of
the
customer
database
and
we
have
basically
G
cluster
on
top
wait
to
perform
or
dis
with
api's
we've
been
doing
research
at
the
moment
to
integrate
spark
and
for
one
reason,
is
that
so
weird
cloud
provider.
Okay,
so
we
have
six
data
centers
and
we
would
like
to
start
analyzing
and
streaming
the
actual
network,
that's
basically
coming
in
and
out
of
our
data
centers.
C
So
as
you
can
imagine,
this
desire,
like
Freddie
heavy
and
large
data
like
constantly
like
bringing
an
out
and
for
that
we
will
be
using
a
Kafka
and
inspired
so
we've
been
like
walking
on
it
and
doing
some
R&D
is
not
ejecting
production,
but
it
is
difficult.
We
are
looking
forward
to
to
start
deploying
for
some
initial
use
cases
around
the
actual
network
analysis
by
the
end
of
the
year,
cool.
A
So
that's
that's
kind
of
a
good
background
as
to
what
you're
doing
I
think
this
is
kind
of
where
we
shift
gears.
Now
you
you
were
talking
before
about
how
your
migration
to
30,
how
did
that?
How
did
that
go
like
what
did
that
look
like
for
you
guys?
Alright.
C
So
I
mean
we've
been
running,
casts
on
Raw
in
production
since
02
dot
0,
so
we
did
20
in
2013
and
shoot
up
one
day
after
and
this
year
we
did
Trudeau
to
and
sweet
home,
and
let
me
if
I
knew
why
we
did
cheat
at
you
first
so
to
do
too.
If
you
want
one
good
reason
to
actually
go
to
to
the
to
be
for
300.
That
would
be
because
of
the
drivers.
C
That's
how
we
did
it,
so
it's
been
pretty
easy
going
to
21
to
22
on
Miss
Lee
super
super
easy
with
the
driver.
Everything
was
working,
fine
and
a
new
position
to
the
20
to
21
migration
that
require
some.
You
know
different
tuning
and
we're
a
bit
of
hardware
bump
when
the
happen
at
the
time
he's
been
mostly
a
flawless
now
to
do
hard
and
play
with
some
stuff
as
well.
C
C
C
I
mean
for
me
just
just
that
trust
the
ability
to
run.
You
know
the
new
driver
against
22.
That's
really
a
good
my
opinion,
a
good
step,
pretty
easy
one
to
make
sure
that
everything's,
okay
and
who
you're
getting
the
bootstrap
resume,
which
is
honestly
a
super
handy
for
for
us
and
I,
worked
really
well
for
you.
Yes,
yes,
that
was
at
that
that
that
was
easy,
almost
no
changes.
As
far
as
configuration
like
same
hardware.
No,
we
don't
I've
been
monitoring
that
for
like
two
three
four
weeks
at
a
time,
but
everything
was
okay.
C
A
A
C
What
we
have
in
between
one
terabyte,
which
each
otha
bite
on
the
node,
so
you
can
imagine
I
mean
you
know
it
can
take
a
while
right
yep.
But
usually
we
don't
need,
like
you
know,
a
lot
of
elasticity.
We
have,
you
know,
put
in
every
data,
centers,
no
settlement
of
resources,
and
you
know
I'm
monitoring.
We
can
basically
anticipate
the
walls
as
it
comes
because
we
know
the
amount
of
vm
that
are
getting
put
strapped
on
our
crow
at
that
at
that
point
yeah.
C
So,
after
what
we
did
is
that
once
the
application
was
migrated
running
up
the
new
Java
driver
against
you
tune,
we
started
migrating
to
30
and
it's
been
mostly
okay,
but
we've
been
hitting
couple
of
roadblocks.
So
let's
talk
about
the
good
thing
first
about
300
and
was
something
that
we've
seen
is
the
space
the
disk
space
used,
compare
and
between
22
and
30
with
the
new
storage
engine.
C
For
us,
it's
been
quite
interesting
because
we
could
see
up
to,
I
would
say,
twenty
percent
just
saving
on
disc
after
the
nerds
were
basically
am
I
grouchy
with
the
new
SS
tables.
Yes,
so
yeah.
That
was
really
nice.
We
were
actually
quite
surprised
about
it,
and
and,
as
ben
has
been
really
good
cool.
C
What
else
can
I
say?
So?
Yes,
we
got
like
you
know
my
no
issues,
maybe
a
little
bit
more
on
the
actual
assess
table
itself,
a
couple
of
Corrections.
Why
am
I
writing
it,
but
honestly,
considering
the
the
size
of
the
refactoring
on
storage,
Laird,
I
was
really
I
was
really
anticipating
way
more
problems
than
than
that,
so
overall,
great,
a
great
upgrade
and
and
one
huge
benefit
that
you
get
as
well
with
302
is
the
hints,
so
the
hints
so
there's
two
things
about
the
hint.
C
So
first
is
not
in
the
SS
tables
anymore,
which
is
which
is
never
been
like,
really
working
right
and
I
guess.
This
is
one
of
the
reason
why
it's
been
put
back
on.
You
know
on
on
the
file
system,
right
so
before
I
used
to
refer
to
what
you
have
to
you
know
to
watch
the
jeans
trunk,
a
blue
jeans
that
were
just
like
stuck
there.
You
had
compaction
going
on
on
the
hints,
etc,
and
we
straight
out
who
we
don't
have
that
anymore.
It's
mostly
Italy
perfect,
you
just
a
couple
of
hints.
C
C
Well,
yes,
oh
the
hinge.
The
program
is
double
and
really
performance
problem
as
such.
That
was
just
that
I
had
to
monitor
them
and
make
sure
to
take
care
of
him.
So
I
was
using
actually
to
splurge
fighters
at
the
time
and
they
had
like
a
bunch
of
tuning
to
do
that.
That
was
a
pretty
handy,
but
no
no
I
have
to
do
that.
I
have.
I
have
actually
remove
that
sound
monitoring
the
hints,
of
course,
making
sure
that
everything
is
ok,
but
I
can
tell
you
after
I
know
several
weeks
having
300
in
production.
C
That's
really
is
a
huge
huge
improvement.
You're
seeing
as
well
is
that
you
can
now
a
disable
or
enable
on
a
pair
that
essential
basis,
the
hinge
delivery
in
streator
0,
which
is
like
really
awesome,
because
you
know,
depending
on
the
latin
see
you
want
to
know
tune.
Actually
you
know
we
have
data
centers
in
singapore
and
we
have
them
as
well.
In
a
you
know,
Central
America
and
East
Side
in
the
US,
so
it's
been
really
handy
for
us
as
well.
Awesome
yeah,.
A
Those
are,
those
are
definitely
a
big
upgrade.
I
was,
I
was
very
happy
to
see
that
I
mean
there's
a
lot
of
people,
that
you
know
they
talk
about
the
problems
with
hints
and
how
they
can
actually
result
in
a
poorly
performing
cluster
like
performing
even
worse,
like
compaction
problems
that
result
from
it.
So,
oh
yeah.
C
I
think
that
does
defeat
at
was
defeat.
Look
as
if
you
do
money
toward
the
hints
with
a
21
or
22
cluster.
Definitely
you'll
have
basically
the
compaction
threads
being
busy
stuck
on
compacting
hints,
that's
basically
what
happening
so
you
can
remediate
like
to
eat
once
you
know
this
is
the
case.
You
know
you
can
basically
like
take
care
of
it
very
easily,
but
the
fact
that
we
don't
have
to
do
that
anymore
with
300
super
ugur.
You
do
it
for
us,
nice.
C
And
yeah-
and
I
want
to
mention
something
as
well
like
you
guys,
been
doing
an
awesome
job
with
the
documentation.
You
should
check
the
right
documentation
section
that
you,
a
bright
guy
died.
Remember
how
you
guys,
calling
it
nowadays
but
really
complete,
I
mean
everything's
in
there.
As
you
know,
the
tips,
the
good
practice,
the
basic
that
you
have
to
not
do
when
you
do
it,
when
you
do
a
migration
so
and
be
so,
people
are
basically
I,
don't
be
afraid
of
a
grating.
It's
mostly
it's
mostly
doing
doing
doing
right.
Awesome.
A
All
right,
well,
I,
think
we're
about
out
of
time.
So
Luke
is
anything
you
want
to
say
before
we
sign
off
no.