►
From YouTube: C* Summit 2013: Taking Bytes from Cassandra Clients
Description
Speaker: Renato Javier and Lewis John McGibbney
Slides: http://www.slideshare.net/planetcassandra/c-summit-2013-taking-bytes-from-cassandra-clients-by-renato
Since early 2012 Gora has been proudly participating as an honorary Incubator post-grad within the ASF. This presentation provides Renato and Lewis' perspective on a phenomenon they refer to as the "big datastore client wars", which is a real life challenge they've discovered whilst attempting to integrate several big data backends (Accumulo, Cassandra, HBase, MySQL, HSQLDB, Amazon's DynamoDB, MongoDB) under one common persistence layer and, in the process, obtain optimal results over Gora operations. They emphasize their approach to addressing this problem by discussing a pluggable Cassandra client infrastructure (Hector-client, Datastax java driver, intravert-ug, etc) adapted specifically for the gora-cassandra module.
A
Okay,
so
we
are
going
to
start
now.
We
are
giving
this
presentation
that
is
called
taking
bites
from
cassandra,
clients
and
apache
gora
perspective.
So
my
name
is
renato:
he's
luis,
we
are
going
to
be
giving
the
talk
together.
Hope,
I'm
hoping
that
doesn't
annoy
you,
people
that
much
so
about
us.
We
are
committers
on
this
apache
parade
that
is
called
gora
in
just
a
minute.
We
will
explain
you
what
what
is
all
this
about.
We
actually
met
on
an
apache
conference
a
couple
years
ago,
and
then
we
just
started
working
together.
A
It's
really
crazy
because
he's
scottish
I'm
from
peru.
We
have
really
different
cultures,
but
we've
we
found
on
open
source
and
code,
a
really
good
friendship.
So
I
we've
been
working
on
on
graph
that
we
like
to
think
that
it's
an
orm
and
all
the
relational
mapping
framework
for
for
nosql.
We
want
to
think
that
we
are
the
the
framework
for
orm
for
nosql
data
storage.
So
we're
going
to
talk
to
you
about
it,
how
we
are
using
it
with
cassandra.
A
We
have
some
implementations
that
uses
dynamodb
from
amazon
and
we
are
really
happy
to
be
here.
So
just
a
little
thing
about
about
me
that
wasn't,
I
was
doing
paragliding
in
rio
de
janeiro.
I
studied
my
master's
there.
It's
just
awesome,
really
generous
school
and
well.
I
am
a
database
database
consultant.
I
think,
like
most
of
us
here,
I'm
a
big
data
and
open
source
enthusiast
and
I
do
a
lot
of
coding.
I
do
some
coding
for
the
apache
software
foundation.
So
now
now
it's
luis
up.
B
Hello
first
things:
first,
I'm
from
scotland
and,
as
renato
said,
we
have
quite
different
different
cultures
and
the
way
we
speak
today,
I've
had
numerous
conversations
with
people
and
I've
had
conversations
where
an
example.
I've
been
talking
about
cascading
and
somebody
thought
I
was
talking
about
cassandra
and
I
spoke
to
them
for
about
five
minutes
on
cascading
and
they
were
convinced
they
were
talking
about
cassandra.
So
if
anything
comes
across
the
wrong
way,
I
apologize
in
advance
anyway.
B
B
It's
got
a
lot
worse.
Actually
so
just
a
quick
introduction
to
gora
before
we
got
into
the
cassandra
stuff
simply
put
gore
is
an
unmemory
data
model
and
persistence
framework
for
big
data
and
what
we
need
mean
by
that
there's.
A
number
of
slides
that
we'll
just
go
through
to
you,
know,
try
and
develop
on
that.
But
you
know
where
it's
in
its
niche
area
is
the
main
use
case
for
gold
originally
was
to
to
support
map
bridges
over
the
data
that
was
in
the
data
stores
a
number
of
locations.
B
The
point
I
want
to
get
across
here
is
that
if
people
that's
been
familiarized
with
with
frameworks
like
a
hybrid
etc
will
have
seen,
you
know
the
generation
of
of
java
beans
or
whatever
and-
and
you
know,
mapping
this
stuff
gora.
However,
I
like
to
think
of,
as
is
much
more
than
just
orem
and
well
I
mean
it's
not
relational
for
starters,
and
you
know
simply
because
of
the
fact
that
it
supports
hadoop
map
reduce
out
of
the
box,
there's
much
more.
That
can
be
done
with
it.
B
B
That
was
really
supported
in
very,
very
early
versions
of
gora
and
data
access
as
well,
and
we
want
to
be
able
to
access
data,
that's
put
into
data
stores
and
to
be
able
to
do
queries
there,
whatever,
regardless
of
its
location,
and
we're
constantly
updating,
what's
actually
angora.
So
we're
looking
forward
to
you
know
moving
on
with
it
and
as
I've
mentioned,
you
know
the
hap,
the
map
reduced
stuff.
B
B
So
very
briefly,
just
before
we
go
on
to
you
know
talking
about
the
main
point
of
this
presentation.
We
just
go
through.
You
know,
what's
what's
really
in
the
pipeline,
for
coda
moving
forward,
that's
of
certain
relevance
to
the
big
data
space
and
the
answer
is
to
be
quite
concise
about.
It
is
just
a
lot
going
on.
A
So,
for
example,
what
we're
planning
for
this
year
is
we
are
playing
integration
with
cascading.
That's
why
he
was
asking
some
developers
of
cascading.
How
would
we
be
integrating
our
project
that
has
got
with
cascading?
We
also
are
thinking
on
integrating
for
with
apache
hama.
I
don't
know
how
many
of
you
are
familiar
with
it.
It's
it's
just
a
bsp.
A
A
We
know
that
it
might
not
be
the
most
popular
data
store
out
there,
but
we
know
there
are
people
using
it,
so
we
are
planning
to
integrate
it
as
well.
Gore
is
actually
being
used.
Its
main
use
cases
is
apache
notch,
which
is
a
web
crawler,
and
we
provide
notch
the
ability
to
store
data
wherever
they
want
right.
So
we
are
now
trying
to
provide
the
same
thing
for
apache.
Gf
graph
is
another
bsp,
a
synchronous
processing
framework
that
is
based
on
the
paragraph
paper
that
google
released
a
couple
of
yeah
this
year.
A
B
This
is
something
that
came
about
and
this
particular
project
is
planned
within
odt,
which
is
a
scientific
project
with
an
apache
and
it's
described
as
a
metadata
from
middleware
and
one
one
aspect.
One
a
module
within
odt
is
a
the
catalog
in
an
archive
system,
metadata
cataloging
and
basically,
what
we
want
to
be
able
to
do
is
right.
Now
they
build
leucine,
they
sort
of
about
losing
indexes
for
their
metadata
metadata,
cataloging
purposes,
and-
and
we
really
want
to
try
and
open
up
use
gora
to
open
up.
B
So
that's
a
step
in
that
direction
and,
finally,
this
one's
kind
of
like
I've
just
named
it
the
vacant
slide.
But
it's
not
really
it's
like.
Maybe
gordo
can
you
know
fit
your
use
case.
Obviously
people
can
design
systems
whatever
way
they
want,
but
you
know
the
the
benefit
that
gora
gives
you
is.
It
gives
you
a
hell
of
a
lot
of
flexibility
in
the
way
that
you
want
to
store
and
access
your
data.
There's
no
doubt
about
that.
B
So
you
know,
why
are
we?
Why
are
we
here
telling
you
that?
Oh
you
know,
this
is
what's
happening
with
gora
and
the
the
justification
is
really
quite
simple.
Is
that
it
opens
up
all
these
other
projects
that
I'm
talking
about
to
also
use
cassandra,
so
you
know
what
we'll
we're
kind
of
taking
with
and
all
these
other
projects
that
otherwise
wouldn't
have
the
ability
to
do
so,
we're
taking
the
ability
to
them
that
you
know
they
can
push
the
data
towards
cassandra
and
they
can
access
it
from
cassandra.
B
So
introduction
dusted
and
I'm
going
to
go
on
and
just
describe
what's
going
on
in
the
presentation
and
what
we
refer
to
as
kind
of
thought,
especially
for
cassandra
and
the
kind
of
the
client
they
are
star
wars.
You
know,
there's
a
lot
of
there's
a
lot
of
vendors
out
there
that
have
got
stuff
some
commercials,
some
otherwise,
and
although
it
seems
to
be
that
the
clients
are,
you
know,
open
source,
it's
not
exactly
a
trivial
decision
when
you
want
to
go
in
and
select
the
client.
B
However,
when
you
look
at
the
nosql
space
generally
speaking,
clients
are
not
something
which
is
focused
on
tremendously
and
which
is
for
us
when
we,
when
that
was
the
outlook
that
we
got
was
I
mean
it's
quite
it's
quite
surprising
that
people
don't
focus
more
on.
You
know
the
client
part
of
the
stack
and
what
were
the
people
that
we've
been
talking
to
in
the
use
cases
that
we've
been
hearing
about.
B
So
our
motivation
really
for
for
looking
into
that
and
for
for
you
know,
trying
out
different
clients
and
we
natively
supported
hector
within
gora.
Renato
is
going
to
talk
a
good
more.
You
know
a
bit
about,
you,
know
the
go
to
architecture
and
and
what
was
going
on,
but
we
started
getting.
You
know
curious
about
maybe
implementing
a
plugable
architecture,
client
architecture
for
the
gora
cassandra
module
and
there's
some
inquisitive
thoughts
in
there,
which
we
get
through
and
some
technical
thoughts
on
there
as
well.
B
There's
a
testing
suite
for
gora
ci,
sorry
for
gora
called
gora
ci,
which
I've
got
time
would
like
to
talk
about
as
well.
You
know
that's
been
developed
and
contributed,
but
with
regards
to
using
it
with
cassandra
and
we've
not
really
had
an
opportunity
yet
and
then
summary
and
and
really
some
discussion,
so
I'm
hoping
that
that
we
can
try
and
invoke
some
reasonable
discussion
towards
the
end.
B
And
you
know
this
is
a
part
of
the
the
core
api
and
for
each
data
store
implementation.
We
have
in
gora
gura
cassandra
being
no
exception,
and
you
know
we
can.
We
can,
you
know,
do
these
operations,
and
this
is
the
beauty
one
of
the
beauties
about
gora.
Is
that,
depending
on
which
data
model
you
have,
and
we
like
to
push
most
of
the
functionality
out,
define
the
core
api
and
push
most
of
the
functionality
out.
B
However,
the
questions
we
had
you
know
if
we're
using
clients-
technically
all
we're
doing,
is
kind
of
wrapping
the
code
packaging
it
and
using
it
with
an
agora
api,
and
you
know
what
kind
of
overhead
is
this
actually
costing
us?
That
was
one.
You
know
that's
kind
of
the
justification
behind
the
technical
side
of
things
from
the
inquisitive
side
of
things
we
really
wanted
to
know
about
which
clients
they
actually
use,
there's
plenty
of
them
out
there.
B
Everybody
knows
this
and
cassandra
kind
of
uniquely
here
there
seems
to
be
a
hellish
lot
more
clients
out
there
for
cassandra
than
than
than
other
stores
and
there's
no
real.
You
know
guidance
out
there
to
to
you
know
to
push
you
in
the
direction
of
what
one
suits
your
use
case.
You've
kind
of
just
got
to
go
out
there
and
make
the
decision
yourself
so
and
and
implementing
plugable
architecture,
and
we
want
to
do
it
in
such
a
way
that
you
know
we
can
address
these
concerns.
B
B
B
An
observation
you
know:
no
sql
is
newish.
The
clients
that
surround
cassandra
are
newish
as
well.
Gore
is
even
newer.
The
tag
cloud
there
is
taken
from
the
cassandra
wiki
page,
and
you
know
it's
obvious
that
it's
not
only
the
java
drivers
that
are,
you
know
numerous
it's
for
other
languages
as
well.
We
don't
like
what
I
mean
we're
focused
in
java,
but
I
mean
it
just
goes
that
I'm
going
to
go
on
and
talk
actually
about.
B
B
That's
got
a
book
coming
out
on
nosql
and
the
book
aims
to
you
know
present
some
use
cases,
problems
essentially
and
describe
you
know
what
storage,
which
no
sql
solution
should
be
suggested
for
that
particular
problem
and
after
having
a
conversation
with
a
gentleman
for
an
hour
two
hours.
However
long
it
was,
I
just
dropped
and
because
I
was
working
on
gore,
I
dropped
in
so
what's
in
the
book
covering
clients
and
there's
nothing.
B
So
that
was
really
the
the
kind
of
that's
that's.
That's
where
our
curiosity
stemmed
about
you
know.
We
really
need
to
try
and
get
this
working
better
and
go
around.
We
need
to
try
and
find
out
more
about
how
java
drivers
are
actually
working
and
how
we
can
try
and
use
more
of
them
provide
more
of
a
flexible
model
for
others
to
use
them
and
so
story
over.
Let's.
C
A
So,
as
we've
been
talking
here
for
for
a
while
already,
we've
noticed
that
using
decline
or
choosing
a
specific
client
for
cassandra
is
not
actually
a
trivial
decision.
We
know
there
are
several
of
them
outside,
but
we
really
don't
know
which
one
is
the
best
one
right,
even
though
I
think
there
was
probably
most
of
you
are
on
the
cassandra
users
list.
There
was
an
email
a
couple
of
months
ago
asking
exactly
this
question
so
which
one
is
better
one.
A
You
could
just
see
people
talking
about
it
and
say:
oh
the
the
java
driver
from
datastax
is
the
best
one.
How
you
can
do
multiplex
requests
at
the
same
time,
and
then
there
was
the
people
from
asian
accent,
exactly
the
same
thing
and
some
other
ones
defending
hector,
but
for
us,
for
example,
that
we
are
users
of
cassandra
and
the
main
idea.
What
we
want
is
just
we
just
want
to
take
our
data
out
of
cassandra.
We
just
want
to
take
our
data
out.
A
We
just
want
to
put
our
data
in
in
the
most
efficient
way.
So,
as
we've
been
talking
here,
what
gora
aims
is
just
to
be
able
to
provide
users
the
ability
to
to
take
your
data
in
to
just
to
take
your
data
out
in
the
simplest
way,
so
we
define
four
four
main
operations
for
for
all
our
data
stores
put
gets
and
deletes
and
we
have
a
scans.
I
think
so
what
we
are
aiming.
We
is
just
to
provide
those
operations
for
any
of
the
data
stores.
A
So
one
of
the
most
important
thing
on
our
plugable
architecture
is
just
to
provide
by
a
simple
opera
by
simple
properties:
let
the
user
decide
which
one
of
those
clients
to
use.
So
we
can
just
go
there
and
choose
which
client
to
use
and
then
we'll
be
switching
our
whole
application
to
use
a
different
client
without
actually
having
to
rewrite
our
whole
application.
So
that's
our
main
idea.
A
So
we've
done
this.
We
have
a
gora
cassandra
client,
which
is
actually
in
charge
of
telling
an
ecto
client
to
do
to
do
the
operations.
We
want
right
to
read
to
write,
to
delete
or
yeah
to
add
some
columns
or
whatever,
but
there
are
a
lot
of
implementation
inside
of
of
these
ones
that
we
as
users
or
as
developers,
we
probably
don't,
really
want
to
get
too
deep
right.
We
just
want
to
get
our
data
and
get
our
data
out
and
just
keep
on
moving
keep
on
developing.
A
So
so
we've
talked
about
about
goa,
so
the
main
properties.
We
have
three
main
files
that
we
we
need.
One
is
the
goa
properties
file
and
on
the
go
up
properties
file,
we
can
decide
which
data
store
we
want
to
use.
So
we
I
could
decide
cassandra
data
store.
I
could
use
hbase
dynamo.
I
could
use
just
any
one
and
there's
also
two
more
files
that
I
need
to
decide
to
to
use,
and
one
of
them
is
the
cassandra
mapping
one
on
the
cassandra
mapping
file.
A
We
decide
which
parts
of
our
data
is
going
to
be
used
in
which
a
column
or
super
column
inside
cassandra.
So
we
use
this
based
on
our.
We
have
to
define
our
schema
right.
Our
schema
is
the
most
important
part.
Could
you
go
to
the
schema,
so
our
schema
is
just
defined
on
on
an
average
on
an
average
schema,
our
average
schema
is
just
json,
so
we
can
just
define
our
schemas
json.
So
after
we
define
our
schemas
as
a
json
file,
as
an
average
schema,
we
can
just.
A
We
have
a
compiler
that
compiles
this
code
into
data
bins
and
then
we
use
these
data
bins
to
persist
on
any
data
store.
So
the
only
three
files
we
need
to
actually
run
on
any
on
any
data
store.
We
provide
it's
only
these
three
files,
so
we
only
need
three
files
and
then
we
can
have.
We
can
have
the
ability
to
write
into
hbase
dynamo,
cassandra
mysql
and
now
we
are
having
all
these
new
data
stores
plug
into
our
project.
A
So
the
idea,
of
course,
like
I'm,
I
know
there
are
some
limitations
as
in
the
configuration
on
the
special
configuration
of
each
data
store.
But
if,
for
example,
for
me,
I
am
a
developer,
it's
easy.
I
mean
it's
easy.
I
just
want
to
try
how
my
schema
is
going
to
how
it's
going
to
work
on
different
data
stores.
I
can
just
put
it
there
and
change
it
and
just
start
pointing
to
a
different
cluster
and
start
writing
my
data
to
dynamo
to
hbase
to
cassandra
to
anything.
A
So
our
main
motivation
here
was
to
decide
on
which
client
to
use,
so
I
don't
want
to
be
having
to
rewrite
my
application.
Every
time
I
have.
I
want
to
change
the
clients.
We
know
they
are
serving
clients
out
there,
so
we've
decided
to
add
a
new
property
that
is,
that
is
going
to
help
us
decide
which
one,
and
so
how
serialization
is
being
worked
on
on
goya
a
couple
of
releases
ago,
we
used
to
serialize
everything
into
cassandra
the
way
the
way
they
were
right.
A
So
if
I
was
storing
strings
so
I
would
be
serializing
you
utf-8
objects
inside
or
numbers.
I
would
be
doing
the
same
thing
nowadays.
What
we
are
doing,
we
are
serializing
everything
else
bytes,
so
we
are
using
just
bytes
comparator
just
to
get
it
in.
We
are.
We
are
planning
to
do
some,
some
benchmarking,
but
sometimes
the
time
is
just
limiting
right.
We
are
not
being
able
to
do
a
lot
of
benchmarking,
but
that's
one
thing
we
have
in
mind
and
we're
planning
to
do,
and
we
are
also
planning
to.
A
We
have
a
lot
of
dependencies
on
our
average
realization
format
that
hadoop
uses,
but
we
are
planning
also
to
support
another
one
that
is
twitter
sparket.
That
was
just
released
a
couple
months
ago,
so
we
are
planning
to
just
expand
ourselves
to
other
different
serialization
models,
and
we
are
also
taking
into
consideration,
for
example,
linkedin's
voldemort,
which
they
have
a
serialization
layer
to
decide
how
they
are
going
to
yeah
how
they're
going
to
serialize.
A
So
the
main
definition
I
think
I've
talked
about
I've
already
talked
about
it
is
really
you
don't
have
to
worry
about
modeling
your
data
right.
Sometimes
it's
hard
to
move
from
relational
database
modeling
to
to
cassandra
right
for
for
most
people,
at
least
for
me.
It
was
at
the
beginning.
I
didn't
quite
understand
it
easily,
but
sometimes,
if
you
just
have
to
define
your
data,
your
xml,
your
schema,
you
only
define
your
schema
and
then
you
are
ready
to
go,
and
then
you
can
procedure
your
data
where,
wherever
you
want,
please
go
ahead.
A
So
that's
what
we
are,
what
we
we've
done
now
right
now
we
are
supporting
ecto,
client
and
asean
x.
We
are
also
looking
for
contributions,
code
contributions,
of
course,
into
the
java.
The
java
driver.
They
are
introvert
their
pillows.
There
are
several
clients
that
we
would
actually
would
like
to
support
for
users
to
use
right,
so
just
go
ahead.
A
So
as
soon
as
we
we
we've
already
done
this.
So
you
have.
We
have
another
property
that
is
called,
they
go
arcasana
client
type
and
then
you
can
decide
which
client
to
use
ector,
stnx
phillips
data
stacks
one
and
for
the
cassandra
mapping.
You
won't
do
anything,
you
will
just
you,
you
can
do,
go
and
have
a
banana.
So
everything
will
just
stay
the
same
and
you
won't
do
anything
else.
So
you
only
have
to
decide
which
client
to
use
and
about
your
schema.
You
won't
do
anything
as
well.
A
Yes,
they
hide
some
some
things
for
us
as
as
developers
right
some
of
them.
They
give
us
the
the
error
handling,
for
example,
in
some
of
them
they
send
dance,
they
send
us
an
exception,
some
others
they
just
send
us
a
boolean
like
I
know,
for
example,
we
were
working
on
creating
schemas
or
deleting
schemas.
Some
of
them
would
send
us
an
exception
if
they
couldn't
find
it
some
others
would.
They
would
just
send
us
a
boolean
if
it
was
or
it
was.
A
Oh,
it
was
not
deleted
so
to
handle
all
these
diff
different
features
that
each
client
has.
It
is
problematic
and,
as
I
I've
already
told
you,
nobody
is
wanting
to
to
rewrite
their
application
every
time
we
want
to
try
a
different
client.
So
this
is
the
main
focus
right
that
we
can
provide.
That's
for
for
you,
so
every
client
is
different.
They
all
have
their
own
characteristics,
but-
and
of
course
this
creates
problems,
but
these
problems
are
trying
to
being
solved
by
gora
by
the
apache
project.
A
So
what's
the
positive
outcome
of
all
this
is
that
serialization
is
still
being
done
by
each
client.
So
each
kind
is
supposed
to
be
good
enough
to
not
lose
the
data.
We
have
the
gora
ci
model.
That
is
a
continuous
ingestion
test.
That
luis
is
going
to
talk
about
it
in
just
in
a
sec,
and
the
concurrencies
also
has
to
be
handled
by
each
client.
Gora,
of
course,
takes
care
of
it
or
part
of
the
concurrency
model,
but
yeah
it's
the
client's
main
main
task,
so
data
modeling.
A
We
can
just
focus
on
modeling
our
data,
just
putting
our
schema
on
on
what
we
want.
We
don't
want.
We
don't
have
to
worry
about
the
client's
error,
handling
and
all
that
stuff,
because
he's
already,
if
somebody
else
would
take
care
of
it
so
and
and
the
freedom
from
clients,
that's
what
we
are
trying
to
put,
because
as
we're
saying
it
sometimes
is
just
difficult
to
choose
one.
A
A
That
is,
we
call
it
a
query
optimizer,
so
we
can
actually
decide
what
type
of
query
we
want
to
use
within
graph
if
we
are
doing
ranges
like
squares
or
we
are
using
composite
columns
or
composite
queries.
So
we
are
trying
to
put
that
in
so
goa
will
actually
take
care
of
it
and
we
will
only
do
gets.
We
will
only
do
puts
we
will.
A
We
will
only
do
deletes
and
that's
it
I
mean
it's
going
to
be
drop
that
simple
and
that's
the
that's
the
that's
the
idea
of
it,
and
it's
really
easy
to
extend
it.
The
api
that
this
has
already
been
set
up.
It's
really
easy.
We
only
have
to
extend
a
couple
of
classes
and
that's
it.
We
can
have
a
different
client.
We
can
have
our
applications
up
and
running
in
minutes
literally.
B
B
If
you
want
to
the
point
about
this,
one
is
that
you
know
renato
myself
haven't
had
you
know,
access
to
the
the
infrastructure
to
run
this
kind
of
stuff
and
the
way
that
we
wanted
to
for
quite
some
time,
we've
been
tied
up
with
other
things
as
well.
Yes,
but
you
know
it
comes
down
to
the
fact
that
you
know
we've
not
had
you
know
the
infrastructure
to
kind
of
run
this
stuff.
However,
basically
the
the
the
testing
suite
enables
you
to
generate,
and
you
know
a
large
a
linked
list.
B
However,
for
for
our
use
case,
this
is
this
code's
been
sitting
there
and
and
we've
not
really
the
ability
to
try
and
go
on
and
do
the
things
that
we
essentially
want
to
do
with
it.
And
so
I
suppose,
if,
if
you
do
pull
gora
and
you
do
have
access
to,
you
know
the
infrastructure
that
we're
talking
about
to
try
and
run
this
stuff.
And
you
know,
please
drop
us
a
line
on
the
development
list,
all
the
users
list,
and
we
would
really
really
love
to
hear
your
results.
B
So
a
review
of
of
you
know
what
we've
been
over
here.
Really
I
mean
this
is
a
great
exposure
for
gora
and
renato
spoke
last
year
at
apache
con
over
in
germany,
but
I
mean
we've
not
really
been
able
to
speak
as
much
about
gold
as
we'd
wanted
to.
However,
in
review
there
are
many
many
clients
out
there
and
we've
learned
a
lot
about
the
clients
eastern
acts
hector-
and
you
know
by
the
end
of
this
week,
with
the
help
of
michelle,
we're
hoping
that
he's
gonna.
B
We
can
work
in
tandem
and
try
and
get
data
stacks
java
driver
working
by
the
end
of
the
conference.
If
we
could
achieve
that
goal,
that
would
be
excellent.
We
are,
we
were
sorry
more
to
the
point
shocked
pretty
much
when
we
started
out
this.
That
was
what
led
to
the
initial
motivation
to
try
and
you
know,
implement
a
plugable
client
architecture
for
gloria
cassandra,
and
you
know
that
the
client,
the
choice
of
clients
shouldn't,
be
something
that's
kind
of
taken
trivially.
B
It
should
be
thought
about,
and
you
should
understand
exactly
what's
going
on
when
you
think
about
implementing
it
within
your
stack
when
I
was
at
apache
con
in
portland
earlier
on
this
year
and
one
of
the
keynotes
there
was
a
gentleman
talking
about.
You
know
trying
to
be
a
generalist
and
understanding
exactly
what's
in
the
stack.
So
when
something
goes
wrong,
you
know
you
can
jump
up
and
you
can
jump
back
down
and
you
know
it's
totally
to
your
advantage
to
be
able
to
do
that
kind
of
stuff.
B
The
plugable
client
architecture
really
came
to
the
surface
only
with
the
cassandra
module
and,
from
my
point
of
view,
that
says
quite
a
lot
more
about
cassandra
than
it
does
about
gora
and
how
vibrant
it
is,
how
many
people
are
working
in
different
aspects
of
cassandra
and
how
many
kind
of
sub
communities,
micro
communities
there
are
on
the
client
side
of
stuff
and
and
really
how
that's
driving
the
product
forward
in
many
many
different
ways
and
we
fit
maybe
above
that
level.
But
I
mean
we're
still.
B
B
In
conclusion,
sorry
plugable
client
code
is
currently
sitting
on
the
natto's
get
github
account
we're
going
to
be
working
on
it
this
week,
and
really
this
is
kind
of
call
to
arms.
I
mean
if
you,
if
you're
working
on
clients,
you're
interested
in
clients-
and
you
maybe
maintain
one
or
you
know-
development
team
that
doesn't
maintain
one
over
and
above
hector
and
esther
max
we'd,
really
love
to
hear
from
you
get
to
know
the
api
see.
B
If
we
can
try
and
you
know,
implement
it
within
the
operations
that
we
currently
support,
with
angora
and
and
also
as
well-
and
you
know,
we've
got
a
real
desire
to
try
and
pull
the
pros
from
all
these
clients
and
make
these
pros
configurable
very,
very
easily,
with
angora
properties.
Just
you
know,
set
some
trivial
properties
that
you
know,
for
example,
consistency
within
hectare.
You
put
it
into
the
properties
file.
You
change
one
string
and
you
know
you've
got
different
consistency
going
on.
This
is
stuff
like
that.
B
D
B
I'm
hoping
to
have
better
answers
for
you
tomorrow,
the
next
day
whenever,
when
we
get
a
better
idea,
I
mean
we're
trying
to
solidify
the
api
right
now
and
and
work
with
different
clients
and
see.
What's
going
on
with
regards
to
actually
talking
about
stuff
like
latency
and
reads
rights,
you
know
how
the
performance
works.
We
don't
have
that
just
now.
One
thing
that
the
that
I
should
maybe
probably
say
here
is
that
you
know
the
purpose
of
us
doing.
B
This
is
not
to
specifically
attack
any
one
client
or
to
take
any
communities
and
say
you
know
this
is
better
than
that,
or
this
is
better
than
that.
Because
I
mean
that's
not
that's
not
the
purpose
of
this.
The
purpose
is
to
give
users,
you
know
the
option
to
pick
whatever
they
want,
but
I'm
sorry
with
regards
to
that
kind
of
stuff.
We
can't
we
don't
have
we
don't
we
haven't
run,
we
basically
don't
have
the
numbers
just
now.
That's.
D
A
So,
just
about
your
your
first
question
is
that's
really
a
hard
question
to
ask,
and
I
would
ask
you
another
question
which
client
is
better
as
tnx
java
driver
there's,
no
there's
no
benchmark
that
we
can
actually
go
and
run
for
each
one
of
them.
So
that's
one
that
was
one
of
our
motivations
right.
We
don't
know
which
one
is
the
best
one
and
we
would
like
the
help
of
other
people
to
help
us
start
designing
tests
to
run
to
run
this.
The
test
we
run
is
the
the
graciemo.
A
A
A
The
idea
is
that
we
write
the
25
million
and
then
we
have
to
start
deleting
some
some
clients
that
are
doing
the
reading,
and
then
we
and
then
we
run
another
job,
another
hadoop
job,
to
verify
if
we've
actually
lost
any
data
on
this
process,
so
these
nodes
are
just
anything
you
can
just
do
you
can
do
whatever
you
want
yeah.
We
actually
run
it
with
notch
with
apache
notch.
So
most
of
it
is
just
web
pages.
D
B
E
So
one
thing
I'm
seeing
from
this
week
is
that
we
might
need
to
do
a
painful
client
switch,
so
I
would
be
interested
in
gora,
but
I
guess
my
question
is:
if
you're
back
to
like
get
put
delete,
are
you
back
to
using
cassandra
at
the
lowest
level?
What
is
there
any
support
for
cql
and
then
the
other
question
is
a
little
different,
but
one
of
the
ways
we're
using
cassandra
is
we
do
have
this
concept
of
the
domain
object,
so
we'd
be
a
great
match
for
an
orm.
B
We
we
we
define
this
stuff
here
and
we
can
override
that,
but
we
can
add
a
lot
more
to
it
as
well.
We
we
want
to
push
that
out
and
and
use
the
expressiveness
of
the
client
or,
or
you
know
whatever
it
is-
that
you're
interacting
with
cassandra
with
so
with
regards
to
us
hitting
directly
into
cassandra.
We
sit
on
top
of
clients
and
we
want
to
you
know.
Well
that's
what
we're
doing
now.
It
was
hector,
and
you
know
obviously
want
to
extend
that.
So,
with
regards
to
that,
I
mean
it's.
B
A
Yeah,
that's
one
of
the
things
we
we're
probably
going
to
have
to
put
in
into
into
gora.
This
is
because,
like
every
client
is
different,
I
mean
I
don't
think
hector
I
don't
well.
I
don't
know,
I'm
not
sure
if
they
so
yeah.
Probably
we
will
put
secure
like
queries
right
now.
We
are
only
supporting
these
abstractions,
so
they
can
be
consistent
across
all
the
data
stores.
We
are
supporting
like
security
as
specific
for
cassandra,
so
in
each
base
we
we
don't
have
that
or
in
dynamo.
A
We
don't
have
it
so
yeah,
but
of
course
that
would
be
an
interesting
thing
to
do
our
abstraction
it
would.
It
would
allow
it
pretty
fairly
simple.
I
I
think
it's
like
we
just
haven't
gotten
around
to
it.
Sometimes
it's
really.
Our
main
english
is
kind
of
it's
kind
of
hard
to
follow,
because
I
mean
we
are.
We
are
doing
most
of
the
cassandra
stuff
on
the
dynamo
stuff,
but
there's
people
playing
around
with
there's
people
around
playing
with
hbase,
and
we
have
difficult
configurations-
and
you
know
yes,
you
guys
can
can't
know
it
all.
A
Sure
you
could
just
have
it.
You
only
have
to
map
it
differently.
Do
you
want
to
go
back
the
xml
xml
file,
so
we
have
the
xml
file.
I
we've
told
you
that
you
need
three
different
files
right.
So
the
first
one
go
back,
it's
this
one
and
that's
a
mapping
file.
So
on
this
mapping
file
you
decide
which
key
space
is
going
to
be
in
which
cluster
and
well
what's
the
whole.
What's
the
host,
so
you
could
actually
define
here
which
fields
you
would
like
to
map
into
which
column
family
or
super
column.
B
B
B
On
top
is
your
javascript
schema
and,
bearing
in
mind
that
to
write
the
jazz
song
I
mean,
and
depending
on
how
you
want
to
persist
the
data
you
need
to
have
a
reasonably
sound
understanding
of
you
know
how
the
gystron
schema
goes
together,
and
you
know
various
nesting
and
whatever
you
want
to
really
consider,
and
so,
as
renato
said,
it
would
be
good.
You
know
if
we
could
walk
towards
you
know,
refining
the
the
the
configuration
part
of
that's
one
of
the
aims.
B
C
A
No,
we
actually
don't
have
any
dynamic
yeah,
we
are
not
doing
dynamically
and
one
problem
of
that
is.
We
are
using
the
average
schema.
The
json
schema
to
create
these
data
bins,
so
we
will
actually
have
to
recreate
our
database.
Every
time
you
decide
to
change
the
schema
yeah,
so
we
don't
have
that
so.
C
A
A
No
yeah,
that's
one
of
the
things
we
actually
have
laying
down
on
on
jira.
We
have
one
patch
that
is
called
the
query,
optimizer
that
we're
trying
to
push
that
in
so
gore
will
actually
be
able
to
decide
what
opportunities
to
to
to
do
to
apply
it.
That's
kind
of
a
hard
part
to
do,
because
all
data
started
they're
different
and
they
have
different
optimizations.
So
we
can
we
yeah.
We
have
to
decide
which
optimization
is
better
in
which
data
store
right.
So
that's
kind
of
a
hard
thing
to
do.
B
I
mean
in
all
honesty,
renato
and
myself
came
on
to
start
working
with
cassandra
stuff
inheriting,
you
know
a
project,
well
a
part
of
the
project
that
that
really
wasn't
too
healthy.
It
wasn't
used
too
much
and
I
mean
it's
come
on
a
lot.
It's
come
on
a
lot.
I've
enjoyed
working
on
it,
so
you
know
this
is
just
another
move
to
try
and
improve
it.
B
Even
still,
you
know
what
keeping
up
with
with
until
recently
the
revisions
of
hector
are
going
to
be
running
with
the
the
client
and
the
the
most
up-to-date
clients
now
and
the
same
with
the
cassandra.
You
know
the
the
dependency
on
any
of
the
cassandra
stuff
that
we're
using
directly
and
if
we
are
using
any
of
it
so
yeah.
B
Like
that
actually
has
been
coming
up,
I
think,
if
I
understand
the
problem
more
recently
and
there's
a
couple
of
issues
that
we
addressed
correct
me
if
I'm
wrong
and
the
I
don't
know,
don't
get
me
wrong.
I
don't
think
the
use
case
was
exactly
like
that,
but
where
synchronization
was
not
done
very
well
with
angora
the
way
that
we
were
using
hector
and
that
stuff's
been
rooted
out
now-
and
that
was
just
that
was
actually
brought
to
the
surface
by
the
use
of
much
highly
concurrent
environment.
B
You,
you
wouldn't
probably
need
to
go
in
and
see
what's
going
on,
but
I
mean
more
of
that
stuff's
been
coming
up
as
more
people
are
using
the
code
and
they
may
not
the
main
direction.
We've
been
trying
to
go
forward
and
packaging
the
relations
and
getting
out
and
get
more
people
using
it.
So
we've
been
trying
to
root
out
stuff
like
that.
A
A
You
could
actually
try
it,
but
the
thing
is
that
being
an
apache
project,
sometimes
we
don't
have
the
how
to
say
that
yeah,
we
don't
have
the
access
to
that
to
that
many
resources
or
we
don't
have
the
those
really
big
use
cases
right.
So
you
are
welcome
to
drop
us
a
line
on
the
main
list
and
we'll
be
hoping
to
to
help
you
getting
all
those,
and
one
of
the
problems
is
also
the
concurrency
control.