akka - scale 16x · stateful architecture shopping more search tools images videos wut? about...
TRANSCRIPT
Sta
tele
ss A
rchi
tect
ures
A ty
pica
l loa
d ba
lanc
ed A
PI s
erve
r clu
ster
Load
B
alan
cer
AP
IS
ER
VE
R
AP
IS
ER
VE
R
AP
IS
ER
VE
R
DB
Sta
tele
ss R
eque
st F
low
1.H
it lo
ad b
alan
cer
2.H
it A
PI b
ox3.
Look
up s
tate
from
per
sist
ence
laye
r4.
???
5.R
espo
nd6.
Pro
fit
Sta
tele
ss R
eque
st F
low
1.H
it lo
ad b
alan
cer
2.H
it A
PI b
ox3.
Look
up s
tate
from
per
sist
ence
laye
r4.
???
5.R
espo
nd6.
Pro
fit
Sta
tefu
l Req
uest
Flo
w
1.H
it lo
ad b
alan
cer
2.H
it A
PI b
ox3.
Look
up s
tate
from
per
sist
ence
laye
r4.
???
5.R
espo
nd6.
Pro
fit
Sta
tefu
l Arc
hite
ctur
eA
sta
tefu
l AP
I clu
ster
. Eac
h no
de h
as a
sub
set o
f the
app
licat
ion
dom
ain.
AP
IS
ER
VE
R
AP
IS
ER
VE
R
AP
IS
ER
VE
R
1-3
4-6
7-9
DB
Sha
rdin
g A
pplic
atio
ns
●K
eepi
ng d
ata/
logi
c in
a k
now
n pl
ace
●M
embe
rshi
p●
Con
sist
ent h
ashi
ng●
Rou
ting
●C
oord
inat
ion
Wha
t is
Akk
a?
●A
ctor
impl
emen
tatio
n on
the
JVM
●A
ctor
s se
nd a
nd re
ceiv
e m
essa
ges
●A
ctor
s ca
n “b
ecom
e” a
new
sta
te b
etw
een
mes
sage
s●
Act
ors
are
“isla
nds
of s
tate
”●
You
can
not d
irect
ly a
cces
s an
Act
or’s
sta
te●
Man
y gr
eat i
ntro
talk
s/po
sts
avai
labl
e
Akk
a to
the
Res
cue
●Fa
ntas
tic a
bstra
ctio
n of
sta
te●
Per
sist
ent a
ctor
s●
Ric
h m
essa
ge d
eliv
ery
sem
antic
s●
Sim
ple
mes
sage
pas
sing
acr
oss
the
netw
ork
●Fa
ilure
is a
firs
t cla
ss c
itize
n
A L
ittle
Exa
mpl
e
●Te
mpe
ratu
reA
ctor
●A
ll of
the
stat
e is
man
aged
in m
emor
y●
Faul
t tol
eran
t, re
play
able
log
of o
ur c
urre
nt s
tate
http
s://g
ithub
.com
/nic
ky-is
aacs
/SC
ALE
-14x
-dem
o
●R
ecov
erin
g○
No
mes
sage
s fro
m th
e ou
tsid
e w
orld
○R
eadi
ng s
naps
hots
and
jour
nale
d m
essa
ges
○G
ettin
g ba
ck to
last
kno
wn
stat
e
●R
ecei
ving
○R
eady
to re
spon
d to
the
wor
ld○
Sta
te c
hang
es g
o to
dis
k as
ynch
rono
usly
○w
rites
are
all
in m
emor
y, p
erfo
rman
ce++
Fini
te S
tate
Mac
hine
(tha
nks
Akk
a)
How
We
Clu
ster
Tod
ay
●S
tate
ful s
hard
ing
of a
pplic
atio
ns●
Zook
eepe
r bas
ed m
embe
rshi
p●
Con
sist
ently
has
hed
by o
rgan
izat
ion
●S
ingl
e le
ader
Whe
re W
e C
ould
Go
●Le
ader
less
clu
ster
s●
Gos
sip
mem
bers
hip
prot
ocol
●R
eplic
atio
n of
wor
k as
clu
ster
sc
ales
Ext
ra C
redi
t
●U
ber R
ingp
op○
“app
licat
ion-
laye
r sha
rdin
g fo
r Nod
e.js
ap
plic
atio
ns”
●A
kka
Clu
ster
ing
●D
ynam
o pa
per