agile document models & data structures

©2016 Couchbase Inc. 1

Agile Document Models & Data Structures

©2016 Couchbase Inc. 2©2016 Couchbase Inc.

Speaking Your Language

• Topics for today:• Data structures - tie into native language collection interfaces• Sub-document - lower level access with focused power• Data modeling with Couchbase• Session: “Picking the right API for the right job”

• SDK Goal: complex data access made easy• More than just a document storage/retrieval system• Tight SDK integration is key• Consistent, transparent developer experience across languages


Data Structures API


Couchbase SDK Data Structures API

• Target SDK release along with 4.6• Builds on awesomeness of sub-document API• Simplified access without touching whole document• Make JSON data types transparent• Native integration of Map, List, Set, Queues…

• Java Collections Framework• .NET System.Collections• Python, Node.js, Go


Typical Document Data Access

JSONDocCB JSON ObjectSDK Collections

FrameworkApp


Simplified Data Structure Access

JSONDocCB Collections

FrameworkSDK DS App

“user1”: {“name”:... , “address”:.. ,

“favs”: [...]},“user2”:{“name” ,

“address” ...,”favs”: [...]},

for (String f : favs) {}“user1”: {“name”:... ,

“address”:.. ,“favs”: [...]},

“user2”:{“name” , “address” ...,

”favs”: [...]},


Targeted Collection Updates

Item From CollectionAppSub-doc

Update CB

MapAdd(“user1”,”favs”, “newfav”)“user1”: {“name”:... , “address”:.. ,

“favs”: [...]},“user2”:{“name” , “address” ...,

”favs”: [...]},


The Four Data Structures…

Structure JSON Type JSON ExampleLists- Append, prepend, insert- Size/count

JSON Array: [… , ... ] [ 1, 2, “abc” ]

Maps- Add/remove by key- Size/count

JSON Object: { “key”: “value”}

{ “name”: “value” }

Sets- Specialized add/remove- Unique values- Size/count

JSON Array: [ … , ... ] [ 1, 3, 6, 8 ]

Queue- First in – first out- Pop – retrieve/remove- Size/count

JSON Array: [… , ... ] [ “task1”, “task2”, “task3” ] remove 1...[ “task2”, “task3”, “task4” ]


Consistent Access Across Languages

FunctionsLists ListGet ListPush ListShift ListDelete ListSet ListSize

>namesList = bucket.ListGet(“key”)>print namesList[‘name1’,’name2’,’name3’]

Maps MapGet MapRemove MapSize MapSetSets SetAdd SetExists SetSize SetRemoveQueue QueuePush QueuePop QueueSize QueueRemove

• Idiomatic -vs- functional• Java Collections Framework• .NET System.Collections• As well as functional approach

* Experimental features alert: may add/remove to this list – feedback welcome!


Consistent Access Across Languages

Collections ApproachLists List<String> namesList = new CouchbaseArrayList<String>("key", bucket);

for (String name : namesList) { … }

Maps var namesDict = new CouchbaseDictionary<string, Poco>(_bucket, “key”); namesDict.Add(“newkey1”, new Poco { Name = “poco1” });

Sets var namesSet = new CouchbaseSet<Poco>(_bucket, "pocos");namesSet.Add(new Poco { Key = "poco1", Name = "Poco-pica" });namesSet.Remove(new Poco {Key = "poco1", Name = "Poco-pica"});foreach(var poco in namesSet){ … }

Queue var namesQueue = new CouchbaseQueue<Poco>(_bucket, key);namesQueue.Enqueue(new Poco { Name = "pcoco1" });var item = namesQueue.Peek();

• Support for advanced capabilities of collection frameworks


Sub-Document API


Sub-Document API

“The sub-document API enables you to access parts of JSON documents (sub-documents) efficiently without requiring the transfer of the entire document over the network. This improves performance and brings better efficiency to the network IO path, especially when working with large JSON documents.”

• First released in 4.5, support cross SDK• Efficient document lookup, insert & update• Powerful lower level control, focusing on particular elements• Keep work on server• Two methods available – lookup and mutate/change


Digging Below Data Structures

Data Structures API Sub-Document API

MapGet(key, mapkey) LookupIn(key).get(mapkey)

MapRemove(key, mapkey) MutateIn(key).remove(mapkey)

MapSet(key, mapkey, value, createMap)

MutateIn(key).(mapkey, value, create_doc=createMap)


Sub-Document APIOperations

LookupIn LookupIn(key, operation(path))

Get Exists Execute

MutateIn MutateIn(key, operation(path, value))

Counter Insert Remove Replace Upsert ExecutearrayAddunique arrayAppend arrayInsert arrayPrepend

Chaining Operations

MutateIn(key, operation(path, value), operation(path, value), operation(path, value))

Returns SubdocResult<rc=0x0, key='map1', cas=0x14b6458980042, specs=(Spec<GET, 'subkey1'>, Spec<EXISTS, 'subkey1'>), results=[(0, u'subvalue1'), (0, None)]>


Sample Sub-Document Lookup

LookupIn(key, operation(path))

LookupIn(‘copilotmark’).get(‘phones.number').execute();

LookupIn(‘copilotmark’).exists(‘phones’).get(‘phones.number').get(‘gender’).execute();

SubdocResult<rc=0x0, key=’copilotmark', cas=0x14b6458980042, specs=(Spec<EXISTS, ‘phones’>, <GET, ’phones.number'>, <GET, ‘gender’), results=[(0, None,), (0, ’212-771-1834’), (0, u’male')]>


Sample Sub-Document Change

MutateIn(key, path, value)MutateIn(‘copilotmark’) .replace(‘phones.number’,

‘212-787-2212’) .upsert(‘nickname’, ‘Freddie’) .execute()


Data Modeling for Couchbase Server

©2016 Couchbase Inc.©2016 Couchbase Inc.

What is Data Modeling?

18

• A data model is a conceptual representation of the data structures that are required by a database

• The data structures include the data objects, the associations between data objects, and the rules which govern operations on the objects.


Data Modeling Approaches

19

NoSQLRelaxed Normalizationschema implied by structurefields may be empty, duplicate, or missing

RelationalRequired Normalization

schema enforced by dbsame fields in all records

• Minimize data inconsistencies (one item = one location)• Reduced update cost (no duplicated data)• Preserve storage resources

• Optimized to planned/actual access patterns• Flexibly with software architecture• Supports clustered architecture• Reduced server overhead


Modeling Couchbase Documents

20

• Couchbase Server is a document database • Data is stored in JSON documents, not in tables

• Relational databases rely on an explicit pre-defined schema to describe the structure of data

• JSON documents are self-describing


What and Why JSON?

21

• What is JSON?– Schema flexibility– Lightweight data interchange format– Based on JavaScript– Programming language independent– Field names must be unique

• Why JSON?– Less verbose– Can represent Objects and Arrays

(including nested documents)

There is NO IMPEDENCE MISMATCH between a JSON Document and a Java Object


JSON Design Choices

22

• Couchbase Server neither enforces nor validates for any particular document structure

• Choices that impact JSON document design:– Single Root Attributes– Objects vs. Arrays– Array Element Types– Timestamp Formats– Property Names– Empty and Null Property Values– JSON Schema


Root Attributes vs. Embedded Attributes

23

• The choice of having a single root attribute or the “type” attribute embedded.



24

• Accessing the document with a root attribute

SELECT track.* FROM couchmusic



25

• Accessing the document with the “type” attribute

SELECT * FROM couchmusic WHEREtype=‘track’


Objects vs. Arrays

26

• The choice of having an object type, or an array type


Objects vs. Arrays

27

• How would the object look like?

class UserProfile{Phone phones;

}

class Phone{String cell;String landline;

}


Objects vs. Arrays

28

• How would the object look like?

class UserProfile{List<Phone> phones;

}class Phone{

String number;String type;

}


Array Element Types

Array of strings

Array of objects

29

• Array elements can be simple types, objects or arrays:


Array Element Types

Array of strings

30


class Playlist{

List<String> tracks;}

...

String trackId = tracks.get(1);

JsonDocument trackDocument =

bucket.get(trackId)

Multiple get() calls to retrieve the document. Worth it?


Array Element Types

31


class Playlist{

List<Track> tracks;}

...

myPlaylist.getTracks()

.get(1).getArtistName();

Limited Denormalization: commonly needed data (e.g., title) in local object, detail available in referenced foreign document


Timestamp Formats

Array of time components

String (ISO 8601)

Number (Unix style)(Epoch)

• Working and dealing with timestamps has been challenging ever since• When storing timestamps, you have at least 3 options:

16


Observed Practices with Timestamp Formats

• Storing as Epoch will help you to easily sort the documents• If you wanted the documents to be sorted in the order of their “last update” time• SELECT * FROM couchmusic WHERE type = ‘track’

ORDER BY updates DESC

• Storing date as array format helps• To grouping

16


Taking Advantage of Storing Date as an Array

• Group options can be specified to control the execution of the view• The group and group_level options are only useful

when a Reduce function has been defined in thecorresponding View

• The group_level option, used when the key is an Array, determines how many elements of the key are used when aggregating the results.

16


Example of View group_level = 1

Key Value

[2014] 36

[2015] 20Execute Reduce

Key Value

[2014,11,29,18,49,36]

3

[2014,12,03,20,11,26]

5

[2014,12,03,23,37,21]

2

[2014,12,06,10,12,19]

8

[2014,12,09,05,01,26]

3

[2014,12,18,01,04,30]

11

[2014,12,26,18,34,44]

4

[2015,01,03,16,48,32]

7

[2015,01,03,20,20,06]

5

[2015,01,15,08,17,28]

8

Copyright © 2015 Couchbase, Inc. 35

• For the data below with Reduce function defined as _sum and group_level = 1



Key Value

[2014,11] 3

[2014,12] 33

[2015,01] 20

Key Value

[2014,11,29,18,49,36]

3

[2014,12,03,20,11,26]

5

[2014,12,03,23,37,21]

2

[2014,12,06,10,12,19]

8

[2014,12,09,05,01,26]

3

[2014,12,18,01,04,30]

11

[2014,12,26,18,34,44]

4

[2015,01,03,16,48,32]

7

[2015,01,03,20,20,06]

5

[2015,01,15,08,17,28]

8



Execute Reduce



Key Value

[2014,11,29,18,49,36]

3

[2014,12,03,20,11,26]

5

[2014,12,03,23,37,21]

2

[2014,12,06,10,12,19]

8

[2014,12,09,05,01,26]

3

[2014,12,18,01,04,30]

11

[2014,12,26,18,34,44]

4

[2015,01,03,16,48,32]

7

[2015,01,03,20,20,06]

5

[2015,01,15,08,17,28]

8

Key Value

[2014,11,29] 3

[2014,12,03] 7

[2015,12,06] 8

[2015,12,09] 3

[2015,12,18] 11

[2015,12,26] 4

[2014,01,03] 12

[2014,01,15] 8



Execute Reduce


Empty and Null Property Values

38

• Keep in mind that JSON supports optional properties• If a property has a null value, consider dropping it from the JSON, unless

there's a good reason not to• N1QL makes it easy to test for missing or null property values• Be sure your application code handles the case where a property value is

missing

SELECT * FROM couchmusic1 WHERE userprofile.address IS NULL;

SELECT * FROM couchmusic1 WHERE userprofile.gender IS MISSING;


Empty, Null and Missing Property Values

39

{ countryCode: “UK”, currencyCode: “GBP”, region: “Europe”}

{ countryCode: “UK”, currencyCode: “GBP”, region: “”}

WHERE region IS NOT MISSING, IS NOT NULL, IS VALUED

WHERE region IS NOT MISSING, IS NOT NULL, IS NOT VALUED

{ countryCode: “UK”, currencyCode: “GBP” }

{ countryCode: “UK”, currencyCode: “GBP”, region: null}

WHERE region IS MISSING WHERE region IS NULL


JSON Schema

40

• Couchbase Server pays absolutely no attention to the shape of your JSON documents so long as they are well-formed

• There are times when it is useful to validate that a JSON document conforms to some expected shape

• JSON Schema is a JSON-based format for defining the structure of JSON data• There are implementations for most popular programming languages• Learn more here: http://json-schema.org

http://json-schema.org/


Example of JSON Schema

41


Example of JSON Schema – Type Specification

Available type specifications include:• array• boolean• integer• number• object• string• enum

42


Type specific validations include:• minimum• maximum• minLength• maxLength• format• pattern

43

Example of JSON Schema – Type Specific Validation


Example of JSON Schema – Required Properties

Required properties can be specified for each object

44


Example of JSON Schema – Additional Properties

Additional properties can be disabled

45


Data Nesting (aka Denormalization)

46

• As you know, relational database design promotes separating data using normalization, which doesn’t scale

• For NoSQL systems, we often avoid normalization so that we can scale• Nesting allows related objects to be organized into a hierarchical tree

structure where you can have multiple levels of grouping• Rule of thumb is to nest no more than 3 levels deep unless there is a very

good reason to do so• You will often want to include a timestamp in the nested data


Example #1 of Data Nesting

• Playlist with owner attribute containing username of corresponding userprofile

Document Key: copilotmarks61569



• Playlist with owner attribute containing a subset of the corresponding userprofile

* Note the inclusion of the updated attribute



• Playlist with tracks attribute containing an array of track IDs



• Playlist with tracks attribute containing an array of track objects

* Note the inclusion of the updated attribute


Key Design


Choices with JSON Key Design

52

• A key formed of attributes that exist in the real world:– Phone numbers– Usernames– Social security numbers– Account numbers– SKU, UPC or QR codes– Device IDs

• Often the first choice for document keys• Be careful when working with any personally identifiable information (PII),

sensitive personal information (SPI) or protected health information (PHI)


Surrogate Keys

53

• We often use surrogate keys when no obvious natural key exist• They are not derived from application data• They can be generated values

– 3305311F4A0FAAFEABD001D324906748B18FB24A (SHA-1)– 003C6F65-641A-4CGA-8E5E-41C947086CAE (UUID)

• They can be sequential numbers (often implemented using the Counter feature of Couchbase Server)– 456789, 456790, 456791, …


Key Value Patterns

• Common practice for users of Couchbase Server to follow patterns for formatting key values by using symbols such as single or double colons

• DocType::ID– userprofile::fredsmith79– playlist::003c6f65-641a-4c9a-8e5e-41c947086cae

• AppName::DocType::ID– couchmusic::userprofile::fredsmith79

54

Enables Multi-Tenency

– pizza::user::101– Pizza::user::102

– burger::user::101– burger::user::102


Lookup Key Pattern

55

• The purpose of the Lookup Key Pattern is to allow multiple ways to reach the same data, essentially a secondary index

• For example, we want to lookup a Userprofile by their email address instead of their ID

• To accomplish this, we create another small document that refers to the Userprofile document we are interested in

• Implementing this pattern is straightforward, just create an additional document containing a single property that stores the key to the primary document

• With the introduction of N1QL, this pattern will be less commonly used


Lookup Key Pattern

56

userprofile::copilotmarks61569 [email protected]

JSON

String

[email protected]

• Lookup document can be JsonDocument or StringDocument


Trade-offs in Data Modeling


Making Tough Choices

58

• We must also make trade-offs in data modeling:– Document size– Atomicity– Complexity– Speed


Document Size

59

• Couchbase Server supports documents up to 20 Mb• Larger documents take more disk space, more time to transfer across the

network and more time to serialize/deserialize• If you are dealing with documents that are potentially large (greater than 1

Mb), you must test thoroughly to find out if speed of access is adequate as you scale. If not, you will need to break up the document into smaller ones.

• You may need to limit the number of dependent child objects you embed


Atomicity

60

• Atomicity in Couchbase Server is at the document level• Couchbase Server does not support transactions• They can be simulated if you are willing to write and maintain additional

code to implement them (generally not recommended)• If you absolutely need changes to be atomic, they will have to be part of the

same document• The maximum document size for Couchbase Server may limit how much

data you can store in a single document


Complexity

61

• Complexity affects every area of software systems including data modeling• The complexity of queries (N1QL)• The complexity of code for updating multiple copies of the same data


Speed

62

• As it relates to data modeling, speed of access is critical• When using N1QL to access data, keep in mind that query by document key

is fastest and query by secondary index is usually much slower• If implementing an interactive use case, you will want to avoid using JOINs• You can use data duplication to improve the speed of accessing related data

and thus trade improved speed for greater complexity and larger document size

• Keep in mind that Couchbase Views can be used when up to the second accuracy is not required


Remember

63

SDK get() is faster than (get by key)N1QL with MOI is faster thanN1QL with GSI is faster than

Model you document key, such that you document can be retrieved with the key, if possible, than a N1QL query


Embed vs. Refer

64

• All of the previous trade-offs are usually rolled into a single decision – whether to embed or refer

• When to embed:– Reads greatly outnumber writes– You're comfortable with the slim risk of inconsistent data across the multiple

copies– You're optimizing for speed of access

• When to refer:– Consistency of the data is a priority– You want to ensure your cache is used efficiently– The embedded version would be too large or complex


Next Steps

• Flexible data access is key to solutions using document stores• Join us for discussion on Forums or discuss with our experts here

• https://forums.couchbase.com• https://developer.couchbase.com/server


Get Trained on Couchbase http://training.couchbase.com

CS300: Couchbase NoSQL Server AdministrationCD220: Developing Couchbase NoSQL Applications

CD210: Couchbase NoSQL Data Modeling, Querying, and Tuning Using N1QL

CD257: Developing Couchbase Mobile NoSQL Applications


Tyler MitchellSenior Product Manager, [email protected] @1tylermitchell

Clarence J M Tauro, Ph.D.Senior Instructor

[email protected]@javapsyche


Thank You!

agile document models & data structures

Software