►
From YouTube: Cassandra Day Chicago 2015: Cassandra: Consistency & Tolerance (a morality guide for databases)
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
She'll
start
everybody
hear
me
so
based
on
Tim's
advice
in
the
last
session,
how
many
of
you
are
actually
considering
going
out
reading
the
paxos
paper
she'll?
Do
it
there's
even
a
paper
called
paxos
made
easy,
which
is
actually
harder
to
read:
yeah
I
wouldn't
office.
Actually
it's
one
of
the
rare
cases
where
some
of
the
implementations
in
code
are
easier
to
understand
that
the
actual
written
English
it's
pretty.
B
A
So
I'm
actually
going
to
rehash
a
little
bit
of
that
pack
so
stuff
a
little
bit
later
on
pursuing
that
I
speak
at
my
usual
pace,
but
first
I
have
some
player
things
to
cover
like,
but
importantly,
I
realized
I'm.
Well,
the
last
talks
today
I
am
between
you
and
alcohol
and
I.
Take
that
responsibility
seriously.
A
It's
30
minutes
is
most
the
last
one.
So
hopefully
we
can
get
through
this
enjoy
the
time
and
then
go
hang
out
at
the
bar
and
if
you
can
do
questions
there,
because
I
don't
think
a
lot
of
time
give
the
course
of
talk.
Okay,
so
many
of
you
may
not
think
of
databases,
it's
having
a
ball
code,
but
Cassandra
very
much
does
and
I'm
hoping
to
be
able
to
help.
You
understand
that
today,
because
the
Sanders
wall
code
is
the
reason
I
love
her.
A
My
wife
calls
to
send
her
the
other
woman
and
Cassandra's
moral
code
in
his
Lynch.
Where
were
they
image
problem
that
we
need
to
talk
about
so
we'll
come
back
to
the
image
problem
later.
First
I
want
to
tell
you
the
story
that
led
me
to
fully
appreciate
Cassandra's
moral
high
ground
with
the
database
world.
First,
a
little
bit
about
myself.
My
name
is
matt
kennedy,
I'm
an
architect
datastax.
A
A
And
Zach
Galifianakis,
as
you
can
tell
by
the
way
you
may
have
discerned
from
other,
tops
today,
that
those
machines
are
useless
right.
This
is
the
one
good
thing
you
can
do
with
a
hard
right.
These
days
you
take
the
top
off.
They
have
a
really
polished
flatter
surface,
and
you
can
see
your
say
it's
good
for
doing
the
backyard
air
or
it's
hard
to
see,
get
one.
What
car
drive
in
front,
let
her
drive
it
back
and
you
can
make
sure
you're
not
sticking
up
in
the
back.
A
A
A
A
A
Somewhat
less
known
for
a
very
clever
group
of
squirrels
show
hands
how
many
of
you
have
heard
of
Georgetown
squirrels
good
as
I
just
made
the
story
up
so
Georgetown
is
faced
with
some
very
harsh
winters
I
feel
like
we
charlatan,
sadies
and
Chicago,
but
for
the
mid-atlantic
the
sheer
presence
of
snow
is
is
a
bad
sign
right
and
for
a
long
time
the
squirrels
on
campus
work
together
and
they
store
their
nuts
in
a
tree
on
campus
for
the
winter.
So
they
all
have
my
food,
and
this
works
out
for
a
really
long
time.
A
Some
squirrels
from
American
University
make
their
way
over
to
campus
and
the
today
borrow
from
the
food
supply,
and
this,
of
course
leads
the
squirrels
to
rethink
their
strategy.
So,
as
the
spring
thaw
happens,
one
scroll
says
a
gut
idea.
What
we're
going
to
do
is
we're
going
to
keep
a
ledger
now.
I
will
point
out
this
white
that
they
still
have
a
bit
of
an
authentication
problem
since
all
squirrels
but
I
think
that
is
an
exercise
to
the
reader.
A
A
So
they
go
with
this
system
where
they
had
a
ledger
to
their
a
card
collection,
and
this
ledger
serves
them.
Well,
it
allows
them
to,
you
know,
have
some
concept
of
acorns
in
MA
porns
out
who's
supposed
to
get
a
horns
and
they
make
sure
that
there's
you
know
kind
of
a
fair
distribution
for
the
eighth
for
the
squirrel
on
campus.
A
Of
course,
they
account
for
the
fact
that
there's
gotta
be
a
squirrel
running
the
ledger
and
your
squirrels
that
go
and
fetch
acorns
and
they're
squirrels
attendee
a
porn
store
and
got
that
all
worked
out
right,
and
so
this
works
really
well
until
one
spring
tree
gets
cut
down,
and
this
throws
their
world
into
disarray.
Does
the
nut
trees
at
all
right?
A
So
what
do
they
do?
Well,
so
they
observe
that
they've
actually
got
multiple.
Nodes
of
us
are
multiple
trees
to
to
use
for
their
problem
right,
and
so
this
one
squirrel
says
Philip
who
stabbed
his
chicken
friend,
and
he
had
this
thing
kept
saying
about
put
all
your
eggs
in
one
basket
and
I.
Think
that
for
us
basket,
is
a
treat
I
think
we
can
draw
this
out
so
that,
if
we
put
all
of
our
nuts
in
different
trees
will
be
safer
or
food
supply.
A
A
That
has
done
certain
positive
of
the
trees.
They'll
still
use
the
letter,
but
a
rubber
squirrel
will
take
a
copy
of
the
ledger
and
copy
that
to
other
trees,
and
what
is
that
it
may
be
if
you
take
some
nuts
from
that
deposit
with
them
and
deposits
them
around
the
trees,
they
good
enough
distribution.
A
A
A
The
more
important
thing
was
to
not
waste
time
waiting
for
these
acknowledgments
that
might
be
delayed
for
an
indefinite
amount
of
time
and
get
back
out
there
and
get
more
nuts
so
that
wise
old
squirrel
asked
well,
which
is
right,
and
the
squirrels
debated
this
further
and
said
on
one
hand,
depending
on
the
conditions
on
the
lawn.
We
may
be
waiting
a
very
long
time
for
this
receive
acknowledgement
process
to
complete.
On
the
other
hand,
what?
If
the
legends
of
who
has
positive
what
aren't
an
agreement?
A
So
they
debated
this
for
a
while
and
they
were
clear
out
of
correct
answer
was
so
we
look
at
what
they're
debating
they're
saying?
Okay,
so
the
squirrel
deposits
the
nut,
and
then
they
have
to
make
this
decision.
The
nuts
go
out
and
the
ledger
receipts
go
out
to
the
other
trees
and
then
they
come
back,
and
so
what
is
this
squirrel
get
to
leave?
What,
if
those
return
arrows
take
a
really
long
time
or
the
arrows
going
out
take
a
long
time.
A
The
distance
between
the
two
campuses
is
to
vexed
to
insist
that
all
squirrels
wait
for
complete
proceeded
knowledge,
so
the
correct
thing
to
do
is:
leave
it
up
to
each
individual
squirrel
as
to
whether
or
not
they
want
to
wait
for
the
full
cross
campus
receipt.
Each
squirrel
is
the
best
judge
how
important
their
own
work
time
is
so
each
squirrel
inside
for
themselves.
B
A
Particular
thing
we
have
the
lower
our
probability
of
losing
it,
but
in
order
to
do
that,
we
have
to
distribute
our
stuff
and
distributing
our
stuff
sub
comes
with
some
additional
challenges.
We
need
to
accommodate
unreliable
work
suit.
We
need
to
have
different
strategies
for
different
path
costs
and
we
need
to
plan
for
outright
failures,
but
most
of
all,
we
need
to
cooperate
so
before.
A
Second,
half
of
the
talk
I
would
like
to
point
Ellen:
don't
underestimate
the
role
of
a
Jesuit
education
so
before
we
get
into
the
rest
of
talk,
I
would
like
to
point
out
that,
while
I
was
doing
this,
I
was
obviously
going
to
google
in
the
church
and
if
you
go
to
google
image,
search
and
image
search
and
type
swirls,
there
are
some
fairly
obvious
completions.
Do
you
maybe
clipart,
but.
A
I'm
going
to
anthropomorphize
for
a
minute,
cassandra
is
widely
rumored
to
be
eventually
consistent
as
if
she
was
some
slacker
college
student
that
gets
around
to
bring
about
a
consistency
when
it's
convenient
to
her.
Eventually,
the
laundry
will
get
done,
and
eventually
the
term
paper
will
get
ready.
Eventually,
the
mobile
phone
bill
will
get
paid.
This
is
an
unfortunate,
an
inaccurate
impression.
In
fact,
cassandra
is
nothing
less
than
the
world's
greatest
practices,
and,
while
these
rumors
/
being
eventually
consistent,
are
true,
that
doesn't
quite
mean
what
everybody
thinks
in
teams.
Cassandra
is
no
slacker.
A
A
So
the
first
problem
with
labeling
Cassandra,
eventually
consistent
database,
is
that
it
excludes
other
classes
of
consistency
of
which
Cassandra
is
capable
we'll
get
to
those
later.
Also,
setting
aside
discussions
of
the
word
consistency
for
the
moment,
let's
probe
it
deeper
on
word
eventual
in
the
context
of
distributed
computer
systems.
Eventual
is
a
term
that
describes
some
aspect
of
the
timing
or
synchronization.
A
Characteristics
of
system
is
fairly
innocuous
bit
of
jargon,
but
in
the
real
world,
where
we
spend
the
rest
of
our
lives,
it
connotes
laziness,
passivity
and
a
complete
disregard
for
anything
that
needs
to
happen
on
time
like
or
not.
These
negative
connotations
are
attached
to
the
term,
even
when
we
should
be
thinking
like
logic,
bourbon
engineers,
this
behavior
can
lead
to
emphasize
conclusions
about
sort
of
problems
to
which
Cassandra
is
suited.
Worse,
inadequate
research
can
lead
decision-makers
to
strongly
consistent
systems
simply
because
they
sound
more
rigorous,
so
Cassandra's.
A
Eventual
consistency
is
sometimes
regarded
as
this
liability
rather
than
capability.
Yet
Cassandra's
ability
to
provide
active,
active,
multi-regional
data
centers
is
always
highlighted
as
a
key
strengths
and
differentiator.
As
this
Cassandra's
ability
to
ingest
data
at
very
high
right
rates,
regardless
an
active
or
failures
both
of
these
capabilities
work
as
well
as
they
do
as
a
direct
result
of
Cassandras,
eventually
consistent
replication.
A
Consistency
in
the
context
of
an
asset
transaction
has
to
do
with
the
database
state
at
the
beginning
and
end
of
the
transaction
and
whether
or
not
integrity,
constraints
which
and
maintained
meaning
here
is
a
little
bit
different.
The
consistency,
meaning
in
Sandra,
comes
from
the
distributed
computing
world
to
step.
A
Consistency
in
the
context
of
kazan
refers
to
different
nodes,
having
a
coherent
view
of
the
same
data.
For
example.
Suppose
I
have
my
email
address
to
a
table
any
standard
database.
In
this
example,
the
table
is
the
key
space
that
has
a
replication
factor
of
three.
So
three
nodes
will
be
mean
to
be
made.
A
Aware
is
change
suppose
that
one
of
those
notes
is
rebooting
because
we
just
updated
currently
get
the
latest
security
updates
my
application
and
choose
to
proceed
with
writing
my
email
addresses
by
the
stone
angel
when
the
note
comes
back
comes
back
up.
Its
data
won't
be
entirely
consistent
with
the
data
on
the
other
nose,
but
that's
okay
Cassandra.
Will
it
eventually
fix
it?
The
important
thing
is
that,
if
it
needs
to
our
application
can
always
insist
on
seeing
a
consistent
version
of
the
record.
A
So
that's
an
important
point:
the
notion
of
an
application
assisting
to
a
database
that
has
preferences
on
how
its
data
example
is
often
very
foreign
concept
to
newcomers
to
Cassandra.
But
it's
a
powerful
differentiator
from
other
databases.
Cassandra
listens
to
the
needs
of
its
applications,
not
just
at
the
database
instance
or
session
level,
but
the
granularity
of
each
and
every
operation
by
taking
advantage
of
this
facility
operations
can
tune
their
consistency
demands
according
to
logical
tolerances
of
the
application.
It
can
achieve
immediate
consistency
for
write
and
read
operations
without
sacrificing
high
availability.
A
So
immediate
consistency
of
data
is
completely
achievable
in
Cassandra,
without
sacrificing
high
availability.
If
eventual
consistency
is
that
appropriate
for
some
or
all
of
the
application
operations
can
simply
insisting
Sandra
that
those
operations
Ian
natively
consistent.
So
in
order
to
understand
how
this
works,
we
need
to
look
at
some
new
concepts,
so
we
have
the
replication
factor,
which
is
the
number
of
copies
of
the
elements
of
the
data
in
the
natives
right.
The
replication
factor
is
set
on
the
database
side.
A
To
be
careful
about
that,
because
it's
easy
to
do
a
consistency,
level
and
replication
factor
work
together
to
help
to
help
achieve
what
we
need
to
from
our
application,
our
application
assistance
he
needs.
So,
let's
take
a
detailed
look
at
a
quorum
right
memory,
because
this
is
important
understand.
So
the
client
makes
a
corner
right
request.
Okay
and
Sandra
receives
that
request
that
poor
neighborhood
and
sends
that
off
in
this
case,
just
to
change
things
up,
we're
looking
at
replication
factor
of
five.
A
In
the
six-month
cluster,
ok,
so
the
coordinator
node
sends
off
the
five
raffles
and
unfortunately,
two
of
those
nodes
are
down
right,
so
those
can't
complete,
but
we
complete
the
request
on
three,
which
happens
to
be
a
quorum
of
five,
and
now
we
can
return
it
okay
on
the
right.
So
what
happens
when
we
go
to
try
to
read
that
data
that
we
see?
We've
got
three
nodes
that
have
the
day
right,
and
our
film
knows
now:
I'm
back
online.
A
A
Coordinator
and
it
sends
out
three
requests
to
data
to
notice
that
should
have
the
data
and
it
doesn't
send
it
to
the
notes
that
those
are
down.
So
we
connect
answers
and
the
first
one
responds
with
the
data
and
a
time
stamp
on
the
data,
and
then
we
get
bastard
back
from
these
two
knows:
don't
have
data,
but
that's
okay.
Corrine
doesn't
mean
that
we
have
to
have
quorum
of
knows
that
agreed
on
the
same
data.
B
A
B
A
So
fast,
so
what
about
our
replicas
from
those
last
things
from
that
last
full
that
didn't
acknowledge
the
former
I
request.
We
still
need
a
way
to
make
those
notes
consistent
with
the
other
notes.
You
know
eventually
and
there's
also
the
case
where
we
don't
care
about
medium
consistency.
So
this
is
true.
If
you
have
a
particularly
high
right
rate
workload-
and
perhaps
you
are
writing
data
recorded
never.
A
So
the
way
we
keep
Cassandra
internally
consistent
is
with
mechanisms
like
into
the
handoff
rate,
repair
and
anti
interview.
Repair
repair
is
a
horrible
horrible
naming
convention
for
what
this
is.
It's
very
efficient
and
we're
talking
about
chaining,
because
repair
makes
it
sound
like
a
my
no
traction.
I
need
to
run
repair
line,
but
really
prepare
is
a
health
operation
right.
We
have
to
run
repair
on
the
cluster,
because
what
we're
doing
is
we're
taking
care
he's
brief
windows,
a
period
where,
maybe
not
all
the
nodes,
put
acknowledge
all
rights.
A
We're
making
the
state
internally
consistent
so
that
it
is
more
efficient
in
the
future
for
client
knows
to
see
consistent
the
data
when
they
need
to
so
we
just
looked
at
axis
and
Sam's
talk
bubble
of
your
games
that
they
were
purchasing
really
State's
repeated
linearize
herbal
consistency
is
a
new
category
for
Cassandra.
We're
used
to
thinking
in
terms
of
eventual
consistency
and
tunable
consistency,
but
paxos
adds
a
new
aspect
to
consistency,
which
is
this
notion
that
some
things
should
happen
before
this
other
thing.
A
So
this
is
variously
known
as
parents
that
check
and
said
lightweight
transactions.
An
important
note
here
is
it:
if
you
are
using
lightweight
transactions
to
access
a
table
and
do
particular
types
of
right
operations,
you
need
to
use
lightweight
transactions
for
all
of
those
kinds
of
rights,
because
if
you
go
and
do
a
regular
write
operation,
you
can
squash
any
in-flight
likely
transactions
that
you
expect
it
to
be
serializable.
A
Furthermore,
that
consistency
level
that
we
talked
about
serial
consistency
level
has
a
multi-day
center
version.
A
local
data
center
version
to
be
extremely
careful
with
this,
because,
if
you
insist
on
local
serializability
and
there
are
operations
going
on
in
two
different
data-
centers
on
potentially
the
same
dinner,
if
we're
only
using
local
serializability,
those
can
also
squash
each
other.
Okay.
So.
A
Don't
bother
understanding
the
pathos
paper
understand
how
to
use
Paxos
into
sandra.
It
will
be
time
much
better
sighs
if
you
want
to
read
up
on
the
ticket
of
57
97
is
the
0
to
refer
to.
So,
in
short,
this
is
allowing
us
to
say
click
x
in
the
database
indexes
not
already
in
the
database
right,
so
we
can
say,
set
foo
equals
bar
if
foo
equals
bass
or
Sifu
equals
bar
louie
equals
faz.
A
A
So
this
is
a
game
where
players
create
monsters
that
compete
to
see
who
can
mash
potatoes
the
fastest.
The
game
allows
players
to
create
monsters
with
different
attributes
to
help
them
mash
faster.
Each
monster
has
a
name
that
needs
to
be
unique,
so
it
goes
without
saying.
This
is
the
most
popular
games
in
the
world,
so
it
absolutely
has
to
use
the
standards
its
database
now
suppose
two
players
Allison
Bob,
want
to
create
a
monster
named
stinky.
A
They
create
their
monsters,
as
they
say
as
they
see
fit,
and
when
the
monster
is
ready
to
be
safe
to
Center,
they
have
to
provide
a
name
for
the
monster
if
they
both
choose
stink.
Even
without
serializable.
Consistency
can
have
a
problem
tired
at
the
same
a
2
point,
0
allicin
off
Monster
Mash
clients
would
perform
a
quick
check
with
the
sender
to
make
sure
the
name
is
obtained.
That's
what
we
see
here,
stop
taking
and.
B
A
So
we
say
does
think
he
exists
and
stinky
does
not
exist.
So
now
we're
going
to
do
a
lightweight
transaction,
insert,
ok,
we're
going
to
say,
create
stinky,
Oh
stinky
does
not
exist
and
he's
both
coming
relatively
the
same
time
that,
thanks
to
lightweight
transactions,
Bob
gets
a
denial
right.
This
is
important.
The
application
is
the
thing
that
is
enforcing,
that
every
user
has
good
experience,
and
so
now
Alice
got
there.
First
she
gets
the
name
or
monster
stinky
and
Bob
doesn't,
and
things
are
now
working
the
way
it
should
so
this.
A
This
kind
of
complete
set
of
consistency,
capabilities
of
the
database
to
animal
consistency,
linearize
will
consistency.
The
important
thing
is
is
that
cassandra
is
honest
with
the
applications
and
the
applications
are
honest
with
the
center.
They
get
to
tell
each
other
what
their
needs
are
and
have
those
needs
met.
So
you
hear
us
a
lot.
A
Cassandra
is
the
most
honest
database
out
there,
and
you
know
these
are
kind
of
some
of
the
reasons
why
so
in
the
Sandra,
we
sometimes
hear
you
know
mutterings
in
the
marketplace.
Well,
what
do
we
do
with
sander
eventual
consistency?
I
can't
have
a
database,
that's
eventually
consistent,
but
really
there's
also
some
jealousy
of
Cassandras
other
features
that
really
require
consistent
consistency
of
work.
So
it's
important
to
note
this.
B
A
Say
sailors
and
keep
our
consistency
as
a
key
value,
get
it
to
those
people
always
on
a
big
data
demands
multi-region
DC's,
just
like
our
squirrels
demanded
multi-campus
enough
storage
to
keep
their
food
supply.
Safe,
multi-region
DC's
must
face
the
reality
of
network
partitions
and
they
also
have
to
be
sensitive
to
the
fact
that
clients,
local
to
DC's
don't
want
to
pay
Geographic
penalties
to
perform
their
operations.