designing for concurrency with riak

Slide 1

Slide 1 text

designing for concurrency with riak nosql matters mathias meyer, @roidrage

Slide 2

Slide 2 text

No content

Slide 3

Slide 3 text

http://riakhandbook.com

Slide 4

Slide 4 text

design for concurrency?

Slide 5

Slide 5 text

design data for concurrency

Slide 6

Slide 6 text

data starts out simple

Slide 7

Slide 7 text

ID Username Email 1 roidrage [email protected] 2 thomas [email protected] 3 karen [email protected]

Slide 8

Slide 8 text

single source of truth

Slide 9

Slide 9 text

always consistent

Slide 10

Slide 10 text

mostly consistent

Slide 11

Slide 11 text

monotonic

Slide 12

Slide 12 text

increase number of sources

Slide 13

Slide 13 text

replication

Slide 14

Slide 14 text

ID Username Email 1 roidrage [email protected] 2 thomas [email protected] 3 karen [email protected] ID Username Email 1 roidrage [email protected] 2 thomas [email protected] 3 karen [email protected]

Slide 15

Slide 15 text

Slide 16

Slide 16 text

ID Username Email 1 roidrage [email protected] 2 thomas [email protected] 3 karen [email protected] ID Username Email 1 roidrage [email protected] 2 thomas [email protected] 3 karen [email protected] ID Username Email 1 roidrage [email protected] 2 thomas [email protected] 3 karen [email protected] ID Username Email 1 roidrage [email protected] 2 thomas [email protected] 3 karen [email protected] ID Username Email 1 roidrage [email protected] 2 thomas [email protected] 3 karen [email protected] ID Username Email 1 roidrage [email protected] 2 thomas [email protected] 3 karen [email protected]

Slide 17

Slide 17 text

eventual consistency* * if no new updates are made to the object, eventually all accesses will return the last updated value. werner vogels, 2008, http://queue.acm.org/detail.cfm?id=1466448

Slide 18

Slide 18 text

multiple clients

Slide 19

Slide 19 text

Slide 20

Slide 20 text

conflicting writes

Slide 21

Slide 21 text

siblings

Slide 22

Slide 22 text

data diverges

Slide 23

Slide 23 text

the challenge

Slide 24

Slide 24 text

determine the winner

Slide 25

Slide 25 text

determine order

Slide 26

Slide 26 text

designing data for concurrency

Slide 27

Slide 27 text

designing data for non-monotonic writes

Slide 28

Slide 28 text

no atomicity in riak

Slide 29

Slide 29 text

no coordination

Slide 30

Slide 30 text

all state is in the data

Slide 31

Slide 31 text

(eventual) consistency and logical monoticity * hellerstein: the declarative imperative: experiences and conjectures in distributed logic (2010)

Slide 32

Slide 32 text

designing data with conflicts in mind

Slide 33

Slide 33 text

write now, converge later

Slide 34

Slide 34 text

rethink the data structures

Slide 35

Slide 35 text

ID Username Email 1 roidrage [email protected] { "id": 1, "username": "roidrage", "email": "[email protected]" }

Slide 36

Slide 36 text

track updates

Slide 37

Slide 37 text

{ "id": 1, "username": "roidrage", "email": "[email protected]" "changes": [ { "client": "client-‐1", "timestamp": 1337001337, "updates": [ "firstname": "Mathias", "lastname": "Meyer" ] } ] }

Slide 38

Slide 38 text

{ "id": 1, "username": "roidrage", "email": "[email protected]" "changes": [ { "client": "client-‐2", "timestamp": 1337001337, "updates": [ "email": "[email protected]" ] } ] }

Slide 39

Slide 39 text

apply all updates ordered by time

Slide 40

Slide 40 text

what about removing data?

Slide 41

Slide 41 text

{ "id": 1, "username": "roidrage", "email": "[email protected]" "changes": [{ "client": "client-‐1", "timestamp": 1337001337, "updates": [ { "_op": "delete", "attribute": "email" } ] }] }

Slide 42

Slide 42 text

{ "id": 1, "username": "roidrage", "email": "[email protected]" "changes": [{ "client": "client-‐2", "timestamp": 1337001337, "updates": [ { "_op": "add", "attribute": "email", "value": "[email protected]" } ] }] }

Slide 43

Slide 43 text

keep a changelog

Slide 44

Slide 44 text

client converges data

Slide 45

Slide 45 text

time as a means of ordering* * leslie lamport, et. al.: time, clocks and the ordering of events in a distributed system (1977)

Slide 46

Slide 46 text

time is not a guarantee for uniqueness

Slide 47

Slide 47 text

vector clocks?

Slide 48

Slide 48 text

{ "id": 1, "username": "roidrage", "email": "[email protected]" "changes": [{ "id": "ca0cb932-‐a74e-‐11e1-‐9ce4-‐1093e90b5d80", "timestamp": 1337001337, "updates": [ { "_op": "delete", "attribute": "email" } ] ] }

Slide 49

Slide 49 text

timelines* * riak at yammer: http://basho.com/blog/technical/2011/03/28/Riak-and-Scala-at-Yammer/

Slide 50

Slide 50 text

time-ordered series of events

Slide 51

Slide 51 text

kept per user

Slide 52

Slide 52 text

{ "events": [ { "id": "ca0cb932-‐a74e-‐11e1-‐9ce4-‐1093e90b5d80", "timestamp": 1337001337, "event": { "type": "push", "repository": "rails/rails", "sha1": "0ea43bf" } }, { "id": "e018f024-‐a74e-‐11e1-‐9feb-‐1093e90b5d80", "timestamp": 1337001337, "event": { "type": "pull_request", "repository": "rails/rails", "sha1": "84efda0" } } ] }

Slide 53

Slide 53 text

clients dedup, sort and truncate

Slide 54

Slide 54 text

observation: clients manage the data

Slide 55

Slide 55 text

sets, counters, graphs

Slide 56

Slide 56 text

monotonic data structures

Slide 57

Slide 57 text

sets

Slide 58

Slide 58 text

an unordered bag of unique items

Slide 59

Slide 59 text

simplest thing that could possibly work...in riak

Slide 60

Slide 60 text

secondary indexes

Slide 61

Slide 61 text

X-‐Riak-‐Index-‐tags_bin: nosql, cloud, infrastructure { "id": 1, "username": "roidrage", "email": "[email protected]" }

Slide 62

Slide 62 text

always unique

Slide 63

Slide 63 text

useful for simple things

Slide 64

Slide 64 text

useful for object associations

Slide 65

Slide 65 text

add-only

Slide 66

Slide 66 text

set: time-ordered list of operations

Slide 67

Slide 67 text

{ "set": [ { "id": "e018f024-‐a74e-‐11e1-‐9feb-‐1093e90b5d80", "timestamp": 1337001337, "op": "add", "value": "roidrage" } ] }

Slide 68

Slide 68 text

Slide 69

Slide 69 text

{ "set": [ { "id": "e018f024-‐a74e-‐11e1-‐9feb-‐1093e90b5d80", "timestamp": 1337001337, "op": "add", "value": "roidrage" }, { "id": "56707cee-‐a757-‐11e1-‐8e1b-‐1093e90b5d80", "timestamp": 1337001339, "op": "add", "value": "josh" }, { "id": "a525f16c-‐a968-‐11e1-‐8b07-‐1093e90b5d80", "timestamp": 1337001343, "op": "remove", "value": "josh" } ] }

Slide 70

Slide 70 text

slightly inefficient

Slide 71

Slide 71 text

2-phase set* * https://github.com/aphyr/meangirls

Slide 72

Slide 72 text

{ "set": { "adds": ["roidrage", "josh"], "removes": ["josh"] } }

Slide 73

Slide 73 text

counters

Slide 74

Slide 74 text

increment, decrement

Slide 75

Slide 75 text

{ "counter": [ { "id": "e018f024-‐a74e-‐11e1-‐9feb-‐1093e90b5d80", "timestamp": 1337001337, "op": "incr", "value": 4 } ], }

Slide 76

Slide 76 text

g-counters* *a comprehensive study of convergent and commutative replicated data types http://hal.inria.fr/docs/00/55/55/88/PDF/techreport.pdf

Slide 77

Slide 77 text

{ "elements": { "client-‐1": 1, "client-‐2": 3, "client-‐3": 5 } } value = 1 + 3 + 5 = 9

Slide 78

Slide 78 text

counters are easy when you increment only

Slide 79

Slide 79 text

convergent replicated data types *shapiro et. al.: a comprehensive study of convergent and commutative replicated data types http://hal.inria.fr/docs/00/55/55/88/PDF/techreport.pdf

Slide 80

Slide 80 text

statebox for erlang* * https://github.com/mochi/statebox

Slide 81

Slide 81 text

knockbox for clojure* * https://github.com/reiddraper/knockbox

Slide 82

Slide 82 text

data represents state

Slide 83

Slide 83 text

state-based means growth

Slide 84

Slide 84 text

data increases with lots of updates

Slide 85

Slide 85 text

dealing with growth

Slide 86

Slide 86 text

truncate

Slide 87

Slide 87 text

roll up, discard

Slide 88

Slide 88 text

{ "counter": [{ "id": "458f5936-‐a752-‐11e1-‐a876-‐1093e90b5d80", "timestamp": 1337001347, "op": "inc", "value": 1 }], "value": 2 }

Slide 89

Slide 89 text

garbage collection

Slide 90

Slide 90 text

not easy with riak

Slide 91

Slide 91 text

not easy with stateful data

Slide 92

Slide 92 text

garbage collection requires coordination

Slide 93

Slide 93 text

network partitions cause stale data

Slide 94

Slide 94 text

the solution?

Slide 95

Slide 95 text

trade off data size vs. consistency

Slide 96

Slide 96 text

commutative replicated data types* *shapiro et. al.: a comprehensive study of convergent and commutative replicated data types http://hal.inria.fr/docs/00/55/55/88/PDF/techreport.pdf