►
Description
Speaker: Nate McCall, Co-Founder
Company: The Last Pickle
Adding a new technology to your development process can be challenging, and the distributed nature of Apache Cassandra can make it daunting. However the drivers, utilities and tooling now available for Apache Cassandra make this process as familiar as possible to developers, with a few minor caveats. After all, it is still a distributed system.
In this presentation, we will do several quick iterations through a simple Java project, demonstrating the following:
• Creating and modifying a data model
• Writing some code working with this model
• Using your local environment for single and multi-node cluster tests
• Integration testing with Jenkins
• Sending it off to production
New and existing users will leave this presentation with the necessary knowledge to make their next Apache Cassandra-based project a success.
B
So
there
are
just
some
things
that,
like
don't
line
up
like
you
expected
there's
failures
and
there's
situations
you
can
think
about.
You
just
wouldn't
have
considered
a
traditional
development
development.
Some
of
those
are
and
talk
about
the
detail.
Of
course,
this
presentation
this
is
an
hour
long.
Some
of
these
other
ones
have
been
30
minutes
long,
so
you
can
say
we're
getting
into
stuff.
If
you
see
something,
that's
interesting
and
asking,
I
have
the
whiteboard.
A
B
That's
what
document,
whatever
you
guys,
can
see,
so
we
can
just
dive
into
this
stuff,
as
it
comes
up
so
first
thing
with
any
sort
of
christian
project.
To
think
about
is
you
know
I
was
talking
to
very
much
about
this.
Is
that
when
you're
starting
the
price
point,
you
want
to
write
your
data,
how
you're
going
to
read
and
that.
A
B
There's
a
number
of
good
presentations
out
there.
Thank
you
by
packing
the
trash
and
virtually
something
mom's
done.
Some
really
good
presentations,
the
best
18
months,
two
years
earlier
today,
you
know:
do
some
research,
it's
really
hard!
It's
it's!
Not!
If
you're
coming
from
a
you
know
coming
from
a
relational
database
world,
it's
different
and
you
really
need
to
sit
down
and
think
about
that,
and
it's
all
like
for
my.
B
B
But
that
said,
data
allowing
and
using
cql
query
language
on
top
there,
but
it's
got
some
really
good
tool.
Support
built
in
now.
The
tracing
the
trace
discovery
is
really
easy
and
it's
a
great
documentation.
It's
actually
maintained
in
solution
which
is
really
cool.
Some
documentation
of
stays
instead
of
the
project
data
stacks
makes
a
really
good
tool
called
data
center.
Now
a
good
example:
two
minute
video
is
a
quick
watch
up
on
the
website
about
what
you
can
do
with
it.
B
B
The
writing
code
section
and
again
it's
not
all
damn
date
models,
but
writing
code
is
a
beast
into
itself,
with
no
things
to
consider.
First,
that
they're,
this
has
been
tried
a
number
of
times
over
the
years.
I've
traveled
twice
myself,
it's
already
a
big
enough.
It
means
match
for
the
database.
You
know
maybe
considering
there's
a
dc
one
built
into
the
job
driver.
That
is
okay,
but
I
I
I
don't
recommend
you
to
do
a
simple
thread
operations.
B
Really
good
test
coverage.
I
use
this
throughout
my
presentations
like
points
with
some
examples.
The
driver
itself
like
how
these
are
done
and
how
you
might
want
to
consider
doing
stuff
on
your
own
projects.
Of
course
you
take
it
apart.
What's
going
on
it's
an
excellent
example
of
how
to
deal
with
netting,
you
know
pipelines
and
factories,
but
if
you're
doing
anything
else,
then
it's
a
good
resource
to
look
at
and
and
dave's.
B
A
B
Documentation
and
governance
of
detail,
but
one
cluster
object
for
each
physical
cluster
one
session
per
app
for
key
space,
prepare
statements
if
you're
going
to
execute
more
than
once
otherwise
you're
going
to
burn
a
lot
of
that.
I
o
and
these
batches
to
use
network
value,
not
necessarily
for
efficiency.
B
Configuration
itself
is
similar
to
other
database
drivers
with
some
caveats
of
detail,
but
the
major
difference
is
as
well
so
you're
not
just
dealing
with
a
single
master
or
single
master
state.
So
really
there's
policies
and
connections
and
the
policies
have
to
do
with
like
load
balancing
what
kind
of
the
balancing
algorithm
you
want
to
use.
What
type
of
connection
connection
hangs
one
set
up
and
compression.
B
B
A
A
B
B
You
know
these
things
through
the
student
system.
You
need
to
have
sort
of
intelligence
across
your
architecture,
developments,
new
usages
and
the
logging
infrastructure
we're
doing
and
supporting
these
things
together.
When
you're
trying
to
hunt
down
a
problem
because
you
distribute
the
system,
the
more
data
you
have,
the
easier
it's
going
to
be
trackings
down,
particularly
and
building
up
historicals.
B
B
It's
just
as
data
volumes
grow,
your
systems
get
bigger,
employ
more
hard
works.
You
need
to
have
stuff
so
start
now
put
in
your
own
code.
Get
it
all
under
that.
Whatever
right
ryan
is
an
excellent
product,
open
source,
it's
just
a
real-time
streaming
api,
it's
really
lightweight
it's
3000
lines
of
code
or
something
like
that.
B
B
Does
anybody
do
anything
like
this,
like
a
like
travel
monitoring
on
your
systems
when
you're
doing
development
you're
an
officer
outside
of
doing
this?
That's
very
lightweight,
not
quite
anyways
around
the
light,
but
enough.
That's
another
good
thing
to
have
in
the
background
there,
because
once
you
get
comfortable
looking
at
this
from
a
development
perspective,
it
becomes
that
much
easier
chasing
these
things
down
in
production
and,
like
I
can't
say
that
enough
that
you
know
becoming
comfortable
with
the
tools
and
how
the
monitoring
into
them
the
blog
file
messages
are
just
like.
B
B
One
that
we're
doing
here
so
on
cql
challenge,
simple
city
tracing
off,
and
so
you
turn
maybe
tracing
for
that
session.
This.
This
is
a
example.
I
took
out
a
production
environment
commissioned
a
client
once
it's
been
dominated,
but
and
we
can
see
the
actual
output
maturity
session
statement,
preparing
community
data
requests.
B
This
is
all
different
conversations
going
on
by
gossip
and
how
the
crew's
gonna
be
broken
down,
and
you
start
to
see
on
board
here
notes
of
how
they
actually
are
hunting
the
data
on
a
disk
and
what's
coming
back
and
interesting
things
like
key
cache.
That
rate
the
number
two
stoves
guesses
tables,
where
the
node
is
actually
looking
for
data
on
the
disk-
and
you
see
this
here.
This
is
a
tightening
column
and
you
go
from
you
know:
3605
microseconds
to
928
milliseconds,
like
almost
a
second
and
then.
B
And
this
particular
road
is
an
outlier
with
it,
and
so
you're
gonna
operate,
get
rid
of
the
volume
on
the
system,
so
the
tracing
just
becomes
extremely
valuable
and
just
because
you
can
kind
of
try
this
queries
out.
So
you
know
without
your
question
on
your
glad
data
set
and
keep
finding
problems
before
they
come
around
the
bike.
B
B
A
B
A
B
And
speaking
of
which
no
tool
for
developers,
you
could
call
us
some
sort
of
the
holy
trinity.
You
know
this
is
geostats.
You
have
histograms
and
proxy
histograms
c
stats
statistics
about
size
and
performance.
It's
the
single
most
useful
command.
You
know
nothing
else
for
this.
You
know
you
know
two
things
grams
as
well.
Well,
you
call
the
partition
count
and
the
size
versus
distribution.
B
B
B
B
I
have
a
couple
of
options
listed
here
and
the
first
one.
Unfortunately,
I
see
this
a
lot
is
that
if
you
use
a
system
like
cassandra
and
using
like
a
shared
instance
somewhere
or
like
a
shared
cluster,
if
you
do
this
like
any
resource
or
you
know,
there's
development
server
in
the
other
room
and
everyone's
using
that
right,
you're
doing
it
wrong,
because
there's
no
cause
for
doing
that
for
every
single
project
or
the
university
costs
for
the
wealth.
B
Much
better
is
using
something
like
sysgen,
which
is
a
python
wrapper
around
the
cassandra
jars
itself
will
actually
download
the
container
manage
download
process
for
use.
You
can
specify
versions,
run
some
like
run
time,
multiple
versions
downloaded
on
your
machine
and
choose
between
them.
We
set
up
clusters
and
data
centers
on
your
local
laptop.
Just
by
focusing
general
processes,
you
can
take
conditional
notes
up
and
down.
So
you
can
see
what
happens.
It's
a
really
useful
tool.
The
distributed
tests
in
the
source
tree
are
built
around
without
sccm.
B
It's
just
used
by
data
directly
on
the
swiss
project,
containing
my
developers
downloaded
hack
on
it.
Whatever
you
want,
I
was
thought
of
a
big
fan
banner
because
icelandic
controllable
environment,
the
way
that
mobile
vendor
machines
are
set
up,
you
can
create
private
networks.
To
have
these
things
be
paid.
Ccl
is
great.
It
packs,
like
the
logos
drivers,
you
have
to
go
to
manage
like
one
two,
seven
over
227
those
three.
B
A
B
Have
a
defined
top-down
server
account
network
referred
that
key
address,
so
I
say
my
first
node's
going
to
be
192
once
we
save
2.10,
we
go
down
there
and
this
is
just
ruby,
which
is
a
configuration
mechanism,
some
tactical
sugar
to
like
tacon
10,
11
12
in
the
back
half
the
back
end
calculate
the
tokens
I'm
going
to
use
for
the
sander
and
drop
those
into
the
listing
address.
This
is
just
a
configuration
of
these
pumps,
so
you
can
see
the
variation
in
that.
Actually,
I
spent
a
lot
of
time
with
this.
B
B
B
B
B
B
B
In
general,
testing
on
distributed
systems,
you
want
to
test
your
failure
scenarios
explicitly
what
happens
when
I
try
to
read
a
quorum
and
there's
not
you
know
it
doesn't.
What
replication
factor
is
necessarily
is
that
you
want
to
know
what
your
application
behaves
like
and
what
the
successions
look
like,
not
only
down
here
directly
but.
B
B
B
Integrations
talked
about
already,
but
just
in
conceptually
it's
through
different
types,
local
and
process
work
process
and
better
concealer.
It's
kind
of
interesting
there's
a
project
out
there
called
the
sander
unit.
That
does
this.
It's
just
basically
launches
in
memory
assume
that
the
disability
students
have
amazing
stuff.
B
B
B
Infrastructure
to
ccm
to
actually
say
I
want
to
create
a
three
node
cluster
for
this
you
know,
setup
go
and
ccm
manages
the
cluster
7
distribution
and
everything
within
the
context
of
the
test.
So
if
you
point
out
something
like
jenkins
as
long
as
you
have
an
icon
installed,
that's
an
excellent
way
to
isolate
that
and,
unfortunately
it's
something
that
it's
not
available
on
the
jar,
because
it's
not
put
in
there's
proper
test
classes.
You
kind
of
have
like
to
cut
it
out,
but
it's
stand
alone
alone.
Actually.
B
Process
talking
to
the
banner
and
cli
to
do
that
either
ways
and
james
but
travis
cis
need
to
measure
cluster
that
way.
Mason
says
things
come
in
parallel.
They
can
be
in
isolated
environments,
multiple
chains
going
through
rescue,
which
is
ideal
and
the
difference
between
two
is
really
generic
choice
and
integrations.
You
know
I
I
really
like
the
idea
of
fingering
the
potential
of
docker
as
well
as
configuration
stuff
is
straightforward.
A
B
B
That's
underpowered
and
something
under
duress
behaved
very
similarly,
so
throwing
these
really
small
configurations
at
these
things
and
just
pounding
like
you
can
from
a
laptop.
It
actually
gives
you
meaningful
data,
because
that's
what's
going
to
behave
like
when
your
big
hardware
on
production
starts
to
have
problems
as
they're
all
set
up
in
terms
of
like
where
the
things
start
to
break.
B
B
B
B
So,
with
anything,
that's
open
source,
the
more
attention
issues
get
more
people
vote
on
from
our
people
work
on
them.
They
should
be
used
to
submit
a
patch.
It
sounds
like
something's
interesting.
These
types
of
sort
of
outside
issues
are
an
excellent
way
to
work
on
the
system
interested
in
internals
general
preston.
The
primary
testing
goal
should
have
initiative.
Is
that
don't
let
the
cluster
behave?