KharkivJS 2017: Коллаборативные системы и CRDT

Коллаборативные системы
Collaborative systems and Conﬂict-Free Replicated Data Types
Максим Климишин
Head of Software Architecture at Takeoﬀ Technologies
29 октября 2017 г.
Максим Климишин Коллаборативные системы 1 / 48

Работа
Head of Software Architecture, Takeoﬀ Technologies, 2017
CTO @ ZAKAZ.UA and CartFresh, 2012
Team Lead @ oDesk (now Upwork), 2010
Project Coordinator @ 42 Coﬀee Cups, 2009

Организация
co-organizer of PapersWeLove Kyiv
co-founder of KyivJS
co-founder of LvivJS
co-organizer of PyCon Ukraine
co-organizer of Hotcode
judge at UA Web Challenge

Takeoﬀ Technologies
eGrocery: Grocery Delivery startup with robotic warehouse

Distributed systems & Groupware systems
Онлайн игры
Календарь, контакты, напоминалки
Софт для конференций
Редактирование онлнай несколькими людьми
Групповые чаты

Распределенные приложения
Использование на разных устройствах в разное время
Одновременное использование на разных устройствах
Работа офлайн

Data and Consistency

Serializability
serial schedule i.e. sequential with no operations overlap in time
Solution of a Problem in Concurrent Programming Control (E.W. Dijkstra,
1965)
A Quorum-based Commit Protocol (Dale Skeen, 1982)

Concurrency fundamentals
divergence или расхождение: ∪n
i=1o1
i ̸= ∪m
j=1o2
j
causality-violations или нарушение порядка операций:
o1 → o3
intention-violations или нарушение намерений: o1||o2

Что приходит в голову
Locking
Transactions
Tentative Transactions
Single Active Participant
Dependency Detection
Reversible Execution

Good luck with that
The Ultimate Guide to Becoming Your Best Self

Consistency Models
Strong Consistency
Weak Consistency
Read Your Writes
Eventual Consistency
Strong Eventual Consistency

Replication Types
Pessimistic
Optimistic or Eventual Consistency

Concurrency Control in Distributed Systems
1989 by Ellis and Gibbs/GROVE editor
Операции должны быть коммутативными, чтобы удовлетворить
convergence property:
op1 ⊙ op2 = op2 ⊙ op1

Кто использует OT?
Wave took 2 years to write and if we rewrote it today, it
would take almost as long to write a second time
on ShareJS

Proving correctness of transformation functions in
collaborative editing systems
2006 Gerald Oster et al
Доказал что все алгоритмы некорректные (с
контрпримерами), упс
Предложил TTF (Tombstone Transformation Functions)

Operation Transformations on Google Scholar
1989-2006

OP DOH

Conﬂict-Free Replicated Data Types
CRDTs

Ограничения сети
Задержки
Drop пакетов
Порядок
Дубликаты

Conﬂict-free replicated data types (CRDT)
2011 Shapiro et al
коммутативность – ∀x, y : x ⊙ y = y ⊙ x
ассоциативность – x ⊙ (y ⊙ z) = (x ⊙ y) ⊙ z
идемпотентность – x ⊙ x = x

Strong Eventual Consistency
C = [c1, ..., cn], ∀i, j : ci = cj ⇒ si ≡ sj

Счетчик
1 + 1

+ идемпотентность нарушена
1 + 1 ̸= 1

∪ – ништяк с сетами
1 ∪ 1 = 1

Прочие свойства в семантике множеств
(1 ∪ 2) ∪ 3 = 1 ∪ (2 ∪ 3)
1 ∪ 2 = 2 ∪ 1

CRDT
ci
CRDT State
(convergent)
Conﬂict
Resolution
Semantics
Application
access API
Si

CRDT: GCounter
Growth-only counter: Interface
export class GCounter {
constructor(counters) { this.counters = counters; }
increment(id) { ... }
query() { ... }
merge(counter) { ... }
}

CRDT: GCounter payload
Counter State on init
let counters = {
1: 0
2: 0
...
N: 0}

CRDT: GCounter increment op
increment(id) {
return new GCounter(Object.assign({
[id]: (this.counters[id] || 0) + 1}))}
}

CRDT: GCounter query op
query() {
return Object.values(this.counters)
.reduce((a, b) => a + b, 0);}
}

CRDT: GCounter merge op
merge(counter) {
let sites = Object.keys(counter.counters)
.concat(Object.keys(this.counters));
return new GCounter(sites.reduce(
(merged, site) => Object.assign(
merged, {[site]: Math.max(
merged[site] || 0,
counter.counters[site] || 0)}),
this.counters));
}
}

CRDT: GCounter sum it up
constructor(counters) { this.counters = counters; }
increment(id) {
return new GCounter(Object.assign({
[id]: (this.counters[id] || 0) + 1}))}
query() {
return Object.values(this.counters)
.reduce((a, b) => a + b, 0); }
merge(counter) {
let sites = Object.keys(counter.counters)
.concat(Object.keys(this.counters));
return new GCounter(sites.reduce(
(merged, site) => Object.assign(
merged, {[site]: Math.max(
merged[site] || 0,
counter.counters[site] || 0)}),
this.counters));
}
}

CRDT: GCounter-test
Increment only counter
linenos function examples_GCounter() {
linenos let log = (op, c1, c2) => console.log(
linenos "[OP] " + op + ": c1=" + c1.query() + ",
linenos c2=" + c2.query());
linenos let counter1 = new GCounter({});
linenos let counter2 = new GCounter({});
linenos
linenos log("counter1", counter1, counter2);
linenos counter1 = counter1.increment(0);
linenos log("counter1++", counter1, counter2);
linenos counter2 = counter2.merge(counter1);
linenos log("counter1 U counter2", counter1, counter2);
linenos }

CRDT: GCounter example output
linenos node_modules/.bin/babel-node --presets es2016,es2015 main.js
linenos [OP] counter1: c1=0, c2=0
linenos [OP] counter1++: c1=1, c2=0
linenos [OP] counter1 U counter2: c1=1, c2=1

CRDT: PNCounter
Positive-Negative Counter
class PNCounter {
constructor(p, n) { this.n = n; this.p = p}
increment() { return new PNCounter(this.p.increment(), this.n); }
decrement() { return new PNCounter(this.p, this.n.increment()); }
query() { return this.p.query() - this.n.query(); }
merge(pncounter) {
return new PNCounter(
this.p.merge(pncounter.p),
this.n.merge(pncounter.n)); }
}

CRDT: MVRegister
Multi-Value Register
export class MVRegister {
constructor(id, register) { this.id = id; this.register = register; }
set(value) {
return new MVRegister(
this.id,
Object.assign(this.register, {[this.id]: value}))}
query() {return Object.values(this.register); }
merge(register) {
let state = this.register[this.id] === undefined ?
{} : {[this.id]: this.register[this.id]};
return new MVRegister(this.id,
Object.assign(register.register, state))
}
}

CRDT: MVRegister Test
const repr = (arr) => arr.length === 0 ? '[]' :
'["' + arr.join('", "') + '"]';
function example_mvregister() {
let log = (op, r1, r2) => console.log("[OP] " + op +
": r1=" + repr(r1.query()) +
", r2=" + repr(r2.query()));
let r1 = new MVRegister(1, {});
let r2 = new MVRegister(2, {});
log("register1", r1, r2);
r1 = r1.set("key1")
log("r1.set(key1)", r1, r2);
r2 = r2.merge(r1);
log("r1 U r2", r1, r2);
r1.set("v1");
log("r1.set(v1)", r1, r2)
r2.set("v2")
log("[CONCURRENT] r2.set(v2)", r1, r2)
r2 = r2.merge(r1);
log("[MERGED]", r1, r2);
}

CRDT: MVRegister Output
[OP] register1: r1=[], r2=[]
[OP] r1.set(key1): r1=["key1"], r2=[]
[OP] r1 U r2: r1=["key1"], r2=["key1"]
[OP] r1.set(v1): r1=["v1"], r2=["key1"]
[OP] [CONCURRENT] r2.set(v2): r1=["v1"], r2=["key1", "v2"]
[OP] [MERGED]: r1=["v1", "v2"], r2=["v1", "v2"]

JSON CRDT
A Conﬂict-Free Replicated JSON Datatype (August 15, 2017)

JSON CRDT
Concurrent mutation on same key with diﬀerent types

CRDT Shopping Cart
2017: redesign of Shopping Cart API

CRDTs: CvRDT, CmRDT
state-based and op-based replication
Рис.: State-based
Convergent Replicated Data
Type or CvRDT
Рис.: Op-based Commutative
Replicated Data Type or
CmRDT

CRDT: Types
Registers, Counters, Sets
Register: LWW or Multi-Value (Dynamo or Couchdb-like)
Counter (growth-only) and Counter w/decrementing
G-Set – growth-only set
2P-Set – remove only once set (G-Set + Tombstones set)
LWW-Element-Set – vector clocks
OR-Set – unique-tagged elements and list of tags within
Tombstones set
WOOT, LOGOOT, Treedoc, RGA, LSEQ for ordered lists

Инструменты
Y-js – framework for oﬄine-ﬁrst p2p shared editing on structured
data
Riak 2.0: Counters, Flags, Sets, Registers, Maps
Roshi by Soundcloud
ShareDB (and fall of Google Wave)
Swarm (and forever-in-pre-alpha tool)
OT.js (Operational Transformation for JS)
CRDT – github.com/dominictarr/crdt
replikativ.io – p2p distributed system framework
GUN framework: p2p distributed framework

Cвой CRDT
+ Мало кода
+ Относительно легко дополнять
+ Предсказуемо работает
+ Надежно при любом качестве связи
+ Работа оффлайн
- Сложная ментальная модель
- Семантическое разрешение конфликтов
- Сложно оптимально подобрать параметры GC
- Реально сложно работать с большим количеством данных
(δ-mutation)

У кого в проде?
Facebook
TomTom
League of Legends
SoundCloud
Bet265
RIAK Distributed Database

Thanks
@maxmaxmaxmax

KharkivJS 2017: Коллаборативные системы и CRDT

More Related Content

What's hot

More from Max Klymyshyn

KharkivJS 2017: Коллаборативные системы и CRDT