►
From YouTube: This Week in Cassandra 2/12/2016
Description
Link to blog post discussed: bit.ly/1PG4x1K
A
All
right,
good
news
here
we
are
this
week
and
Cassandra
number
two.
It
looks
like
we
didn't
give
up
right
away,
pretty
exciting,
so
I'm
John,
Haddad
restoration
right
on
Twitter.
So
today
we're
going
to
be
talking
with
Luke
Tillman.
We
got
Julie
hi,
we
got
Christos
from
netflix.
This
is
awesome
and
I've
already
forgotten
the
last
name,
I'm
totally
terrible
at
pronouncing.
It
collapse,
something
glance
Wilson,
so
yeah
we're
going
to
just
be
talking
about
this
week
in
Cassandra.
A
This
a
company
is
a
blog
post,
that's
up
on
planet
Cassandra,
which
you
can
check
out
now.
So
let's
just
take
a
look
here:
we've
got
some
projects
and
blogs.
We've
got
some
JIRA's
that
we
want
to
talk
about
so
first
up
we
got
three
Cassandra
releases:
bugfix
updates
to
the
2.2
2
1
and
3
dot,
3
pretty
good
stuff.
You
should
definitely
take
a
look
at
the
changelog
for
that.
I
won't
get
into
really
any
detail
on
this
because
it's
bugfix
releases,
but
we
do
want
to
talk
about
some
blog
posts.
A
B
So
my
takeaway
was
basically
he
was.
He
was
kind
of
saying
you
know
getting
started
with.
Cassandra
can
be
hard
make
sure
you
do
your
due
diligence,
and
you
know
if
you
need
to
engage
experts
which
I
think
insta
cluster
certainly
qualifies
as
experts
in
this
area.
You
know,
don't
be
afraid
to
to
reach
out,
and
you
know,
kind
of
kind
of
engage
because
you're
going
to
make
mistakes
essentially
yeah.
A
When
one
thing
that
I
wanted
to
bring
up
I,
don't
know
if
people
know
this
so
instead
cluster
is
actually
a
hosted
Cassandra
and
that's
one
of
the
things
that
you
know
I
hear
a
lot
of
people
asking
questions
about,
like
especially
with
you
know
a
small
teams
not
having
big
operations.
I,
don't
think
a
lot
of
people
even
knew
instakill
existed,
so
that's
kind
of
like
hey
I
can
run
you
know.
I
can
run
a
database
in
Amazon
and
they'll
host
it.
For
me,
I
don't
have
to
worry
about
it.
C
C
A
data
scale
that
I
oh,
but
it
I-
think
it's
just
great
that
there's
this
community
being
built
around
Cassandra,
both
hosting
and
consulting
our
friends
at
the
last
pickle,
do
some
great
work,
helping
people
you
know
get
up
and
running
and
and
over
over
certain
hurdles,
running
Cassandra,
I
know
the
hippie
and
the
Pythian
guys
do
some
good
work
as
well.
So
I'm
just
happy
to
see
this
community
grow
and
it
just
shows
the
viability
of
of
this
project,
and
you
know
the
support
behind
it.
Yeah.
A
C
We
had
a
need,
we
had
a
need,
Cassandra
met
that
need,
and
we
we
just
rolled
with
it.
I
mean
I
mean
when
we
moved
to
the
cloud.
Basically,
everything
was
in
you.
Nothing
out
there
that
met
running
a
large-scale
operation
in
the
cloud,
so
go
on
Cassandra,
even
in
its
very
nascent.
Stay
was
probably
the
most
mature
part
of
our
whole
staff.
C
B
C
D
Cool
and
we'd
like
to
add
something
that
the
need
to
have
a
Cassandra
as
a
service
is
growing
more
and
more
even
for
small
companies.
We
have
seen
here
in
Europe
that
sometimes
you
have
like
very
small
start
up
with
like
10
people
and
when
they
would
strap
with
kasama.
They
try
to
do
with
the
right
way
with
the
data
model,
but
they
don't
have
enough
resources
to
run.
You
know
all
the
infrastructure
of
themselves
so
having
a
service
like
kasama
as
a
database
is
very,
very
useful
for
all
of
their
those
companies.
A
D
And
I
think
we
do
have
here
in
france,
we
have
a
partner
which
also
is
doing
the
same
thing
as
integral,
so
they
are
not
as
well
known,
but
they
are
very
well
known
in
france
and
they
are
basically
offering
the
same
thing
yeah.
I
think
that
in
future
we
will
see
more
and
more
actors
on
this
film
mm-hmm.
A
Alright,
so
the
next
blog
post
that
we
have
walking
down
Apache
Cassandra
logging.
This
is
by
name
a
call
from
the
last
pickle
this
I'll
just
say.
This
is
I,
really
liked
this
post.
Anything
that
touches
operations
is
something
that
I
like
I,
think
we
don't
have
enough
time
looking
at
operations
in
general
or
if
there's
a
lot
of
developers
focus
stuff,
but
you
know
simple
things
like
logging
is,
is
so
important
and
frequently
gets
overlooked.
So
it's
nice
to
see
some
content
around
that.
B
Yeah
I
didn't
even
know
you
could
adjust
logging
levels
with
node
tool.
That
was
news
to
me
like
I
knew
you
could
do
it
through
the
XML
files-
and
you
know
the
other
stuff,
but
that
but
but
I
didn't
know
you
could
do
it
with
no
tool.
I
mean
it
doesn't.
Surprise
me
because
no
tools,
like
the
Swiss
Army
knife
of
with
the
same
drill,
yeah,
definitely
interesting.
Yeah.
C
What
is
your
take
on?
No
tool
being
so
you
mentioned
it's
a
Swiss
Army
knife,
if
some
see
it
as
a
positive
but
I
go
back
to
my
you
know,
unix
my
unix
background,
where
you
know
the
UNIX
ethos
is
a
lot
of
small
programs.
You
can
pipe
together
and
do
something
interesting
and
I
kind
of
shy
away
from
something
where
you
just
add.
You
know
you
make
this
kind
of
big
monolith
of
a
nap
yeah.
A
A
So
I
think,
if
you
take
a
look
at
like
what
hack
like
the
actual
node
tool,
command
line
app,
it's
really
just
a
bunch
of
like
branches
to
go
off
to
like
dedicated
routines
yeah,
so
it
doesn't.
It
doesn't
really
bother
me
that
much
but
I
do
I
would
like
to
see
more,
some
like
on
the
UNIX
idea,
like
being
able
to
take
the
output
of
one
command
and
easily
pipe
it
to
another
command
without
having
like
super
dedicated.
I'll.
A
Put
that's
like
harder
to
parse,
like
you
know,
different
like
column,
family
stats
and
things
like
that.
Like
trying
to
look
looking
at
histograms,
it
would
be
nice
to
get
that
information
out
and,
like
you
know,
either
JSON,
where
you
could
like
easily
bring
it
into
like
Python
programs
and
this
stuff
is.
It
may
actually
have
gone
back
in
there.
A
B
Would
be
interesting
to
ask
some
of
the
commanders
that
have
been
around
for
a
while
with
the
history
of
that
is
like
how
it
kinda
ended
up
the
way
it
did
I.
Imagine
you're
right.
That's
because
it's
a
wrapper
around
a
lot
of
them.
You
know
it's
got
probably
a
lot
of
dependencies
on
the
internal
Java
code.
Ok,
yep,
the.
D
Only
thing
I'm,
afraid
of
is
like
the
we
are
adding
more
and
more
like
comments
with
no
tools,
and
now
you
know
you
have
a
huge
list
of
many
many
things
you
can
add
you
can
you
can
use
I
think
we
may
need
to
organize
those
stuff
cleanly
someday.
You
know
you
have
commands
for
operations
command
for
logging
comment
for
setting
throughputs,
because
right
now,
it's
just
a
bunch
of
things.
That's
not
to
help
a
bra.
It's
like
two
pages
of
commands.
Oh
yeah!
Oh,
where
can
I
go
yeah.
A
A
All
right
what
else
we
got
here:
I
just
totally
lost
my
page,
Apache
Cassandra
for
analytics
and
performance
and
storage
analysis
by
Evan
Chan
of
Temple
jump,
I!
A
C
A
little
bit
yes,
but
probably
not
for
what
you
think
we
are.
We
use
spark
against
raucous
andhra
for
operational
reasons.
Maybe
a
team
wants
to
do
some
batching
on
a
regular
basis,
so
they
they
pick.
You
know
some
downtime
and
not
take
the
cluster
down
just
some
down
turn
of
throughput
on
the
cluster
and
run
some
jobs,
but
we
don't.
We
don't
do
much
of
that,
because
we
we
do
micro
services
and
that's
why
you
know
we
go
out
and
say
we
have
250
clusters,
because
every
app
almost
every
app
has
its
own
cluster.
C
So
so
we
have
this
big
pipeline,
which
we've
documented
called
aegisthus,
which
takes
rs3
backups
d,
dupes
them
transform
them
to
JSON
I,
put
some
in
a
big
HDFS
drive,
and
then
we
run
jobs
off
of
that.
So
it's
used,
but
not
as
much
as
as
I,
don't
even
want
to
say,
as
we
like
it's
just
not
practical
for
us,
gotcha,
okay,.
A
B
Know
you
and
I
I
think
had
the
had
a
similar
reaction,
because
you
just
posted
and
chat
right
before
we
came
on
line.
I
was
when
I
was
reading
and
I
was
thinking
because
he
he
did
Casilla.
He
documented
Cassandra,
two
dot,
one
or
two
dot
to
with
compact
storage
turned
on
and
I
was
thinking
to
myself,
who
I
wonder
what
eighty
ninety
nine
and
the
new
storage
format
and
30
did
up
to
those
numbers.
I
wonder
what
it
would
look
like:
a
more
efficient
storage
format,
yep.
A
One
of
the
things
that
he
mentioned
was
having
just
shorten
the
column
names
because
of
the
overhead
in
SS
tables
from
from
column
names
and
that's
completely
removed
in
30,
which
is
why
I
brought
that
up
the
other
thing
that
I
wanted
to
look
at
and
we'll
touch
on
this.
We
brought
up
the
new
indexes
going
into
Cassandra.
We
talked
about
that
last
week.
A
little
bit
you
know,
I,
would
love
to
see
some
performance
benchmarks
of
the
saasy
indexes
versus
the
existing
Cassandra
indexes.
I.
A
C
Up
great
segment,
gracing,
yeah,
exact
way,
sorry
yeah.
A
I
wrote
this
one
so
I'm
a
little
bit
biased,
I
liked
it,
but
I
talked
a
little
bit
about
the
new
indexes
like
if
you're
at
all,
interested
in
being
out.
It's
like
the
latest.
You
know
cuz
Sam
like
what's
happening
in
the
Cassandra
like
trunk.
This
is
the
this
is
a
post
that
you'd
want
to
read,
really
really
good
work
by
jordan,
west
and
pavel.
You
know
pmc
for
for
Cassandra,
like
really
really
great
stuff.
A
D
I
have
I
have
some
like
index
option
validation
issue
that
can
let
your
index
in
a
store
state,
but
now
it's
fixed
and
I
think
they
have
found
another
issue
with
mem
table
switch
yeah,
it's
like
if,
because
sassy
is
keeping
its
own
data
structure
in
memory,
and
so
if
you
have
a
bunch
of
data
coming
in,
it
can
grow
like
one
gigabyte
to
be
bite
of
beta.
So
now
they
are
fixing
it
on
a
way,
and
I
really
really
hope
that
she
will
be
really
really
performant
a
nice
when
it
gets
much
into
it.
A
A
D
It's
the
most
disruptive
I
mean
them
in
terms
of
usability
for
a
user.
The
most
to
feature
seems
like
chickens
last,
two
years
with
yeah
I
way.
Transaction.
For
me,
lightweight
transaction
was
a
huge
jump.
You
know
in
terms
of
usability
because
he
provide
you
some
some
atomic
construct
you
to
build
like
a
strong,
consistent
operation
and
now
we've
sassy.
We
it
opens
up
all
possibility
for
for
searching
and
clearing
it's
crazy,
hey.
C
A
Cool
alright,
what
else
we
got
here,
Oh
blog
post,
that
went
up
today.
This
one
I
think
is
really
interesting.
You
know
if
you're
thinking
about
30
and
using
the
materialized
views
it's
jake
over
ad
over
with
us
at
datastax,
really
really
good
post
on
the
trade-offs
of
materialized
views.
I'm.
I
love
posts
like
this.
It's
just
huge
all
the
details
that
you
need
to
know
and
its
really
really
well
written.
What
did
you
guys
think
yeah.
B
Even
the
refresher
at
the
beginning,
if
you're
you
know,
if
you're
kind
of
new
to
Cassandra-
and
you
don't
even
know
what
the
materialized
view
stuff
we're
kind
of
talking
about
is
like,
even
though
the
kind
of
quick
refresher
guarantees
and
trade-offs
that
Cassandra
gives,
you
would
be
great
for
somebody
brand
new
to
Cassandra.
He
does
a
really
good
job
of
summing
that
up,
you
know,
even
at
the
beginning
of
the
the
post,
but
yeah
I
I
would
say
you
know
it's
funny.
Do
I
said
you
know.
B
Sazzy
indexes
are
probably
like
the
biggest
step
in
usability
that
we've
seen
in
a
couple
of
years,
like
I,
would
add
materialized
views
to
that
list.
To
you
know,
like
kind
of
making
the
making
the
denormalization
that
we
teach
people
you
know
out
in
the
field
that
you
do
in
your
data
model
kind
of
making
that
baked
into
the
database,
I
think,
is
also
a
huge
step
forward
and
iniesta.
D
Bility,
I
add
a
remark
to
this
from
matter
nice
view.
I
have
been
talking
with
some
end
users
of
Cassandra,
and
so
they
they
were
so
excited
with
mV,
so
they
jump
in
and
what
days
so?
Is
you
now
right
now
we
have
a
limitation
of
one
non
primary
key
column
that
you
can
pick
from
your
base
table
to
add
your
primary
key
column
on
your
view,
and
this
limitation
is
very,
very
a
showstopper
for
them,
because
sometimes
they
are
bucketing.
You
know
they
have
done
some
time
series
and
they
are
using
bucketing.
D
So
it
means
that
a
composite
partition
key.
So
the
fact
that
you
cannot
allow
more
than
one
primary
non
primary
key
column
is
really
really
in
annoying
and
I
I
spend
my
time
explaining
to
them.
Okay,
there
are
reasons
we
can't
do
that
right
now,
because
it
you
know
it
opens
a
lot
of
eight
cases
that
can
break
your
view,
but
as
an
oozer,
they
feel,
if
you
like,
very
frustrated
so
I
told
her
that
there's
a
rider
to
fix
that.
A
Yeah,
it's
definitely
a
complicated
issue.
I
think
that
in
general,
with
Cassandra
one
of
the
things
that
I
notice,
especially
if
you're
talking
to
someone
who's
just
using
like
a
relational
database,
especially
if
it's
on
a
single
machine
and
they're
not
used
to
thinking
about
things
in
a
distributed
fashion,
coming
over
to
Cassandra
there's
a
lot
of
like
hesitancy
like
oh,
how
come
I
can't
just
do
this
people
don't
understand
that
you
know
the
data
centers
suit
like
there
is
latency,
but
networks
fail
like
that.
A
There's
just
a
lot
of
issues
that
you
have
to
think
about
when
you're
working
with
the
distributed
database
that
you
kind
of
just
don't
really
need
to
like
locks,
are
not
cheap,
distributed
world
and
so
I
think
getting
something
like
understanding.
All
the
nuance
of
materialized
views
can
be
really
really
difficult
for
people,
especially
if
they're
they
have
no
experience
with
distributed.
Databases.
C
And
there's
trade-offs
with
materialized
views,
even
in
a
single
machine
or
DBMS
as
well,
and
it
amazes
me,
you
know
back
in
the
day
where
people
didn't
understand
that
either
I
remember
once
working
on
a
db2
database,
and
you
know
they
had
materialized
views
on,
but
they
had
it
set
to
update
every
time.
There's
a
mutate,
Shin
underneath
and
wondered
why
their
database
was
being
thrashed
from
a
load
point
of
view,
so
I
mean
materialized
views,
I,
hope,
people
don't
walk
away.
Thinking
hey!
C
D
What
is
amazing
is
sometimes
when
you
show
them
their
the
trade-off.
They
say.
Oh
that's!
So
it's
not
that
good!
It's
not
that
sexy.
So
I
will
come
back
to
my
you
know,
Manuel
de
normalisation
with
batches
and
I
said
no,
don't
because
if
you
think
about
it,
even
in
your
case,
when
you
are
doing
a
manual
de
normalisation
with
bachchan,
will
be
with
me,
53
people
right.
You
have
a
lot
of
corner
cases.
You
don't
see
and
now
that
we
we
created
nvm
will
show
you
those
common
case.
A
Definitely
all
right!
So,
let's
see
now
moving
along,
we've
got
the
cql
PHP
cql
driver,
1
dot
one
released,
there's
a
blog
post
about
that
supporting
PHP,
7
I
think
we
can
just
let
people
read
that
blog
post,
since
it's
more
of
like
a
changelog
we've
got
Cassie
Q,
which
is
a
queue
built
on
Cassandra
I,
always
tell
people
not
to
do
that.
Oh
oh.
C
C
Wee
wee
wee
sinned,
very
badly
and
in
st
annex
we
released
a
queue
recipe
on
top
of
cassandra
and
unfortunately,
people
started
using
it
here
at
netflix
and
outside
a
netflix.
This
seems
very
interesting.
You
know
not
relying
on
tombstones
and
not
relying
on
delete.
This
is
definitely
something
we're
going
to
test
out,
because
there
is
no
multi
data
center
cues
out
there.
If
this
can
can
you
know
solve
a
problem,
Netflix
has
with
with
cross
data
center
queuing,
then
this
might
be
very
interesting.
Yep.
A
A
We've
got
new
version
of
the
spark
connector
for
Cassandra,
so
if
anyone
is
using,
anyone
is
using
that
definitely
should
check
out
the
Darcy.
One
should
be
have
a
solid
g
GA
release
soon.
Hopefully,
I
love
spark.
So
this
is
just
exciting
to
me.
I'm.
Just
like
your
version
of
the
connector,
we
get
some
events
coming
up.
We
have
Cassandra
de
la
coming
up
on
the
seventeenth.
C
Regulating
netflix
is
always
hiring
and
right
now
I'm
higher,
although
we're
always
hiring
and
we've
got
a
lot
of
positions
open.
This
is
a
position
in
my
team
in
my
organization
and
its
not
directly
a
Cassandra
opportunity,
but
it's
an
automation,
engineer,
opportunity
and
I
invite
you
all
go
through
the
deck.
I
tried
something
different
I
didn't
do
a
job
description.
I
did
a
presentation
of
what
the
job
is,
but
to
sum
it
up,
things
go
wrong.
C
A
C
In
olin
room
yep,
yes,
SD,
annex
st
annex
was
built
on
top
of
the
rift
and
at
the
time
you
know
it
solved
some
of
the
issues
hector
head
and
for
those
who
don't
know
your
Greek
mythology.
Sd
annex
is
the
son
of
Hector
and
I.
Think
Cassandra
summit
2012,
the
the
we
presented
SD
annex
to
the
public
and
we
lovingly
called
it
a
Hector's,
smartass
son,
but
but
yet
no
c
ql.
C
The
secure
a
binary
protocol
is
definitely
moved
way
ahead
of
a
thrift
and
it's
easier
to
use
it's
going
to
be
a
lot
of
refactoring
for
us,
but
we
think
it's
well
worth
it,
and
you
know
we
think
it
was
also
time
to
just
let
the
community
know
that
hey
CQ
Ellis
here
even
ethics
is
going
to
move
off
of
it
in
the
next
18
it
off
of
thrift
in
the
next
18
months
yeah.
So
you
might
want
to
start
thinking
about
it
yourselves
and.
A
C
No,
it
is
not
go
away.
No,
no
we've
gone
down
that
path.
Once
there
was
a
need,
we
filled
it,
but
a
netflix.
Are
we
don't?
We
don't
write
software
for
the
sake
of
writing
software,
we're
not
like
uber
in
that
whoops.
Sorry,
I
really
I
really
believe
that
if
there's
something
out
there
go
and
use
it
contribute
back
to
to
the
project.
If
you
think
you
can
do
it
better
or
add
or
enhance
it
I'm
a
really
believe
a
huge
believer
in
that
and
we
are
actually
going
to
enhance
it
internally.
D
D
It's
so
it's
it's
like
it.
So
I
did
a
test.
I
inserted
a
three
day
like
a
a
wider
with
treat
a
steering
column
I
flush
between
each
insert
and
then
why
do
I
select
limit
one
with
the
tracing
on
I
saw
that
customized
just
hitting
three
SS
table
hey,
so
it
started
like
very
something
very
easy,
but
in
fact
this
Jarrah
is
sitting
for
so
long
because
of
we
have
a
refactoring
in
between
yep.
So
now,
I'm
really
happy
that
it's
getting
in
cool
vessel,
optimization.
D
This
one,
oh,
when
I
was
you
know
showing
you
daf-2
doing
my
meat
tubs
I
created
some
sample
use
cases
and
I
just
realized
that
when
you,
when
you
declare
a
new
year,
a
function
with
parameter
with
which
has
like
a
type
parameters
like
a
list
of
strain
list
of
integer
or
whatever
in
CL
and
when
in
it
comprised
by
the
server
the
server
just
lose
the
generic
type,
it's
just
stupid
and
then
in
your
source
code,
you
need
to
cast
the
type
manual
which
is
absolutely
ugly.
Yes,
now
it.
D
A
D
Can
be
very
handy,
but
because
I
was
thinking
sometime
people
with
us,
sighs
to
your
compaction,
you
can
have
a
huge
table
and
a
lots
of
smaller
one
and
those
huge
guy.
They
will
never
be
thick
by
the
ganga
rhythm.
So
with
this
news
that
you
find
compaction,
you
can
tell
oh
just
pick
those
and
compact
them
together
who
remove
old
autumns
done.
Oh
very
handy,
very,
very
cool.
A
Yeah,
all
right,
that's
half
an
hour
and
I
think
we've
covered
everything.
That's
in
here
I
definitely
recommend.
If
you
want
to
look
into
more
detail
or
any
of
the
stuff
that
we've
been
talking
about,
we've
got
links
to
all
the
releases.
All
the
blog
posts
there's
a
lot
of
good
content.
So
thanks
Christos
for
coming
in
this
is
awesome.
Thanks.