Utilizing arrays: modeling, indexing, and querying

{ "Utilizing Arrays" : ["Modeling", "Querying",

"Indexing"] }

Keshav MurthyDirector, Couchbase R&D

Agenda

• Introduction to Arrays• Data Modeling with Arrays• Query Performance With Arrays

• Array Indexing• Fun With Arrays

• Query Performance• Tag Search• String Search

Introduction To Arrays

Every N1QL query returns Arrayscbq> select distinct type from `travel-sample`;{… "results": [ { "type": "route“ }, { "type": "airport” }, { "type": "hotel" }, { "type": "airline” }, { "type": "landmark” } ] , "status": "success", "metrics": { "elapsedTime": "840.518052ms", "executionTime": "840.478414ms", "resultCount": 5, "resultSize": 202 }}

Results from every query is an array.

cbq> SELECT * FROM `travel-sample`WHERE type = 'airport' and faa = 'BLR';{ "results": [], "metrics": { "elapsedTime": "9.606755ms", "executionTime": "9.548749ms", "resultCount": 0, "resultSize": 0 }}

Introduction to Arrays

• An arrangement of quantities or symbols in rows and columns; a matrix

• An indexed set of related elements

JSON Arrays

{ "Name" : "Jane Smith", "DOB" : "1990-01-30", "hobbies" : ["lego", "piano", "badminton", "robotics"], "scores" : [3.4, 2.9, 9.2, 4.1], "legos" : [ true, 9292, "fighter 2", { "name" : "Millenium Falcon", "type" : "Starwars" } ]}

• Arrays in JSON can contain simply values, or any combination of JSON types within the same array.

• No type or structure enforcement within the array.

JSON Arrays

{ "Name": "Jane Smith", "DOB" : "1990-01-30", "phones" : [ "+1 510-523-3529", "+1 650-392-4923" ], "Billing": [ { "type": "visa", "cardnum": "5827-2842-2847-3909", "expiry": "2019-03" }, { "type": "master", "cardnum": "6274-2542-5847-3949", "expiry": "2018-12" } ]}

Billing has two credit card entries, stored as an ARRAY

Two phone number entries

JSON Arrays : Syntax Diagram

Data Modeling with Arrays

Properties of Real-World Data

• Rich structure• Attributes, Sub-structure

• Relationships• To other data

• Value evolution• Data is updated

• Structure evolution• Data is reshaped

Customer

Billing

Connections

Purchases

Jane Smith

Jan-30-1990

Modeling Data in Relational World

Billing

ConnectionsPurchases

Contacts

Customer

• Rich structure• Normalize & JOIN Queries

• Relationships• JOINS and Constraints

• Value evolution• INSERT, UPDATE, DELETE

• Structure evolution• ALTER TABLE• Application Downtime• Application Migration• Application Versioning

Using JSON For Real World Data

CustomerID Name DOB

CBL2015 Jane Smith 1990-01-30

Table: Customer{ "Name" : "Jane Smith", "DOB" : "1990-01-30"}

• The primary (CustomerID) becomes the DocumentKey

• Column name-Column value become KEY-VALUE pair.

{ "Name" : { "fname": "Jane", "lname": "Smith" } "DOB" : "1990-01-30"}

Customer DocumentKey: CBL2015

Using JSON to Store Data

CustomerID Name DOB

Table: Customer

{ "Name" : "Jane Smith", "DOB" : "1990-01-30", "Billing" : [ { "type" : "visa", "cardnum" : "5827-2842-2847-3909", "expiry" : "2019-03" } ]}

CustomerID Type Cardnum Expiry

CBL2015 visa 5827… 2019-03

Table: Billing

• Rich Structure & Relationships• Billing information is stored as a sub-document• There could be more than a single credit card. So, use an array.

CustomerID Name DOB

Table: Customer

{ "Name" : "Jane Smith", "DOB" : "1990-01-30", "Billing" : [ { "type" : "visa", "cardnum" : "5827-2842-2847-3909", "expiry" : "2019-03" }, { "type" : "master", "cardnum" : "6274-2542-5847-3949", "expiry" : "2018-12" } ]}

CustomerID Type Cardnum Expiry

CBL2015 visa 5827… 2019-03

CBL2015 master 6274… 2018-12

Table: Billing

Value evolution Simply add additional array element or

update a value.

CustomerID ConnId NameCBL2015 XYZ987 Joe SmithCBL2015 SKR007 Sam SmithCBL2015 RGV492 Rav Smith

Table: Connections{ "Name" : "Jane Smith", "DOB" : "1990-01-30", "Billing" : [ { "type" : "visa", "cardnum" : "5827-2842-2847-3909", "expiry" : "2019-03" }, { "type" : "master", "cardnum" : "6274-2542-5847-3949", "expiry" : "2018-12" } ], "Connections" : [ { "ConnId" : "XYZ987", "Name" : "Joe Smith" }, { "ConnId" : "SKR007", "Name" : "Sam Smith" }, { "ConnId" : "RGV491", "Name" : "Rav Smith" }}

Structure evolution Simply add new key-value pairs No downtime to add new KV pairs Applications can validate data Structure evolution over time.

Relations via Reference

Using JSON to Store Data{ "Name" : "Jane Smith", "DOB" : "1990-01-30", "Billing" : [ { "type" : "visa", "cardnum" : "5827-2842-2847-3909", "expiry" : "2019-03" }, { "type" : "master", "cardnum" : "6274-2842-2847-3909", "expiry" : "2019-03" } ], "Connections" : [ { "ConnId" : "XYZ987", "Name" : "Joe Smith" }, { "ConnId" : "SKR007", "Name" : "Sam Smith" }, { "ConnId" : "RGV491", "Name" : "Rav Smith" } ], "Purchases" : [ { "id":12, item: "mac", "amt": 2823.52 } { "id":19, item: "ipad2", "amt": 623.52 } ]}

CustomerID Name DOB

CustomerID

Type Cardnum Expiry

CBL2015 visa 5827… 2019-03

CBL2015 master

6274… 2018-12

CustomerID ConnId Name

CBL2015 XYZ987 Joe Smith

CBL2015 SKR007 Sam Smith

CBL2015 RGV492 Rav Smith

CustomerID item amt

CBL2015 mac 2823.52

CBL2015 ipad2 623.52

CustomerID ConnId Name

CBL2015 XYZ987 Joe Smith

CBL2015 SKR007 Sam Smith

Contacts

CustomerBilling

ConnectionsPurchases

Models for Representing Data

Data Concern Relational Model JSON Document Model (NoSQL)

Rich Structure Multiple flat tables Constant assembly / disassembly

Documents No assembly required!

Relationships Represented Queried (SQL)

Represented N1QL, MongoDB, CQL

Value Evolution Data can be updated Data can be updated

Structure Evolution Uniform and rigid Manual change (disruptive)

Flexible Dynamic change

Querying Arrays

• Array Access• Expressions• Functions• Aggregates

• Statements• Array Clauses

Array Access: Expressions, Functions and Aggregates.

• EXPRESSIONS• ARRAY• ANY• EVERY• IN• WITHIN• Construct [elem]• Slice

array[start:end]• Selection

array[#pos]

• FUNCTIONS• ISARRAY• TYPE• ARRAY_APPEND• ARRAY_CONCAT• ARRAY_CONTAINS• ARRAY_DISTINCT• ARRAY_IFNULL• ARRAY_FLATTEN• ARRAY_INSERT• ARRAY_INTERSECT• ARRAY_LENGTH• ARRAY_POSITION

• AGGREGATES• ARRAY_AVG• ARRAY_COUNT• ARRAY_MIN• ARRAY_MAX

• FUNCTIONS• ARRAY_PREPEND• ARRAY_PUT• ARRAY_RANGE• ARRAY_REMOVE• ARRAY_REPEAT• ARRAY_REPLACE• ARRAY_REVERSE• ARRAY_SORT• ARRAY_STAR• ARRAY_SUM

Array access{ "Name": "Jane Smith", "DOB" : "1990-01-30", "phones" : [ "+1 510-523-3529", "+1 650-392-4923" ], "Billing": [ { "type": "visa", "cardnum": "5827-2842-2847-3909", "expiry": "2019-03" }, { "type": "master", "cardnum": "6274-2542-5847-3949", "expiry": "2018-12" } ]}

SELECT phones from t;[ { "phones": [ "+1 510-523-3529", "+1 650-392-4923" ] }]

SELECT phones[1] from t;[ { "$1": "+1 650-392-4923" }]SELECT phones[0:1] from t;[ { "$1": [ "+1 510-523-3529" ] }]

Array access: Expressions and functions{ "Name": "Jane Smith", "DOB" : "1990-01-30", "phones" : [ "+1 510-523-3529", "+1 650-392-4923" ], "Billing": [ { "type": "visa", "cardnum": "5827-2842-2847-3909", "expiry": "2019-03" }, { "type": "master", "cardnum": "6274-2542-5847-3949", "expiry": "2018-12" } ]}

SELECT Billing[0].cardnum from t;[ { "cardnum": "5827-2842-2847-3909" }]SELECT Billing[*].cardnum from t;[ { "cardnum": [ "5827-2842-2847-3909", "6274-2542-5847-3949" ] }]SELECT ISARRAY(Name) name, ISARRAY(phones) phones from t;[ { "name": false, "phones": true }]

Array access : Functions{ "Name": "Jane Smith", "DOB" : "1990-01-30", "phones" : [ "+1 510-523-3529", "+1 650-392-4923" ], "Billing": [ { "type": "visa", "cardnum": "5827-2842-2847-3909", "expiry": "2019-03" }, { "type": "master", "cardnum": "6274-2542-5847-3949", "expiry": "2018-12" } ]}

SELECT ARRAY_CONCAT(phones, ["+1 408-284-2921"]) from t;[ { "$1": [ "+1 510-523-3529", "+1 650-392-4923", "+1 408-284-2921" ] }]

SELECT ARRAY_COUNT(Billing) billing, ARRAY_COUNT(phones) phones from t;[ { "billing": 2, "phones": 2 }]

Array access : Functions

SELECT phones, ARRAY_REVERSE(phones) reverse from t;

{ "phones": [ "+1 510-523-3529", "+1 650-392-4923" ], "reverse": [ "+1 650-392-4923", "+1 510-523-3529" ] }]

SELECT phones, ARRAY_INSERT(phones, 0, "+1 415-439-4923") newlist from t;[ { "billing": 2, "phones": 2 }]

SELECT phones, ARRAY_INSERT(phones, 0, "+1 415-439-4923") newlist from t;[ { "newlist": [ "+1 415-439-4923", "+1 510-523-3529", "+1 650-392-4923" ], "phones": [ "+1 510-523-3529", "+1 650-392-4923" ] }]

Array access : Aggregates

SELECT ARRAY_MIN(Billing) AS minbill FROM t;[ { "minbill": { "cardnum": "5827-2842-2847-3909", "expiry": "2019-03", "type": "visa" } }]

SELECT name, ARRAY_AVG(reviews[*].ratings[*].Overall) AS avghotelrating FROM `travel-sample` WHERE type = 'hotel' ORDER BY avghotelrating descLIMIT 3;

[ { "avghotelrating": 5, "name": "Culloden House Hotel" }, { "avghotelrating": 5, "name": "The Bulls Head" }, { "avghotelrating": 5, "name": "La Pradella" }]

SELECT: ARRAY & FIRST Expression

ARRAY: The ARRAY operator lets you map and filter the elements or attributes of a collection, object, or objects. It evaluates to an array of the operand expression, that satisfies the WHEN clause, if provided.

SELECT ARRAY [name, r.ratings.`Value`] FOR r IN reviews WHEN r.ratings.`Value` = 4 END FROM `travel-sample` WHERE type = 'hotel'

SELECT FIRST [name, r.ratings.`Value`] FOR r IN reviews WHEN r.ratings.`Value` = 4 END FROM `travel-sample` WHERE type = 'hotel'

FIRST: The FIRST operator enables you to map and filter the elements or attributes of a collection, object, or objects. It evaluates to a single element based on the operand expression that satisfies the WHEN clause, if provided.

Statements

• INSERT• INSERT documents with arrays• INSERT multiple documents with arrays• INSERT result of documents from SELECT

• UPDATE• UPDATE specific elements and objects within an array

• DELETE• DELETE documents based on values within one or more arrays

• MERGE• MERGE documents to INSERT, UPDATE or DELETE documents.

• SELECT• Fetch documents given an array of keys• JOIN based on array of keys• Predicates (filters) on arrays• Array expressions, functions and aggregates• UNNEST, NEST operations

Statements:INSERT

INSERT INTO customer VALUES ("KEY01", { "cid": "ABC01", "orders": ["LG012", "LG482", "LG134"] });

INSERT INTO customer VALUES (("KEY01", { "cid": "XYC21", "orders": ["LG92", "LG859"] }),

VALUES (("KEY04", { "cid": "PQR49", "orders": ["LG47", "LG09", "LG134"] }), VALUES (("KEY09", { "cid": "KDL29", "orders": ["LG082"] });

INSERT INTO customer ( KEY uuid(), value c ) SELECT mycustomers AS c FROM newcustomers AS n WHERE n.type = "premium";

Statements: DELETE

DELETE FROM customerWHERE orders = ["LG012", "LG482", "LG134"];

DELETE FROM customerWHERE ANY o IN orders SATISFIES o = "LG012" END;

DELETE FROM customerWHERE ANY o IN orders SATISFIES o = "LG012" ENDRETURNING meta().id, *;

Statements:UPDATE

UPDATE customer USE KEYS ["KEY091"] SET orders = ["LG012", "LG482", "LG134"];

UPDATE customer USE KEYS ["KEY091"] SET orders = ARRAY_REMOVE(orders, "LG012") ;

UPDATE customer USE KEYS ["KEY091"] SET orders = ARRAY_APPEND(orders, "LG892") ;

Statements : SELECT

• SELECT• Array predicates• NEST, UNNEST• Fetch documents given an array of keys• JOIN based on array of keys

SELECT statementARRAY PREDICATES

SELECT: Array predicates

• ANY• EVERY• SATISFIES• IN• WITH• WHEN

SELECT: Array predicates• Arrays and Objects: Arrays are

compared element-wise. Objects are first compared by length; objects of equal length are compared pairwise, with the pairs sorted by name.

• IN clause: Use this when you want to evaluate based on specific field.

• WITHIN clause: Use this when you don’t know which field contains the value you’re looking for. The WITHIN operator evaluates to TRUE if the right-side value contains the left-side value as a child or descendant. The NOT WITHIN operator evaluates to TRUE if the right-side value does not contain the left-side value as a child or descendant.

SELECT * FROM `travel-sample` WHERE type = 'hotel’ AND ANY r IN reviews SATISFIES r.ratings.`Value` >= 3 END;SELECT * FROM `travel-sample` WHERE type = 'hotel’ AND ANY r WITHIN reviews SATISFIES r LIKE '%Ozella%' END;

• EVERY: EVERY is a range predicate that tests a Boolean condition over the elements or attributes of a collection, object, or objects. It uses the IN and WITHIN operators to range through the collection.

SELECT * FROM `travel-sample` WHERE type = 'hotel’ AND EVERY r IN reviews SATISFIES r.ratings.Cleanliness >= 4 END;

SELECT: Array predicates• ARRAY_CONTAINS

• Returns true if the array contains value.

SELECT name, t.public_likes FROM `travel-sample` t WHERE type="hotel" AND ARRAY_CONTAINS(t.public_likes, "Vallie Ryan") = true;[ { "name": "Medway Youth Hostel", "public_likes": [ "Julius Tromp I", "Corrine Hilll", "Jaeden McKenzie", "Vallie Ryan", "Brian Kilback", "Lilian McLaughlin", "Ms. Moses Feeney", "Elnora Trantow" ] }]

Array Expressions, Functions and Aggregates.

• EXPRESSIONS• ARRAY• ANY• EVERY• IN• WITHIN• Construct [elem]• Slice

array[start:end]• Selection

array[#pos]

• FUNCTIONS• ISARRAY• TYPE• ARRAY_APPEND• ARRAY_CONCAT• ARRAY_CONTAINS• ARRAY_DISTINCT• ARRAY_IFNULL• ARRAY_FLATTEN• ARRAY_INSERT• ARRAY_INTERSECT• ARRAY_LENGTH• ARRAY_POSITION

• AGGREGATES• ARRAY_AVG• ARRAY_COUNT• ARRAY_MIN• ARRAY_MAX• ARRAY_SUM

• FUNCTIONS• ARRAY_PREPEND• ARRAY_PUT• ARRAY_RANGE• ARRAY_REMOVE• ARRAY_REPEAT• ARRAY_REPLACE• ARRAY_REVERSE• ARRAY_SORT• ARRAY_STAR

SELECT: ARRAY & FIRST Expression

ARRAY: The ARRAY operator lets you map and filter the elements or attributes of a collection, object, or objects. It evaluates to an array of the operand expression, that satisfies the WHEN clause, if provided.

SELECT ARRAY [name, r.ratings.`Value`] FOR r IN reviews WHEN r.ratings.`Value` = 4 END FROM `travel-sample` WHERE type = 'hotel'

SELECT FIRST [name, r.ratings.`Value`] FOR r IN reviews WHEN r.ratings.`Value` = 4 END FROM `travel-sample` WHERE type = 'hotel'

FIRST: The FIRST operator enables you to map and filter the elements or attributes of a collection, object, or objects. It evaluates to a single element based on the operand expression that satisfies the WHEN clause, if provided.

SELECT statementUNNEST and NEST

Querying Arrays: UNNEST

• UNNEST : If a document or object contains an array, UNNEST performs a join of the nested array with its parent document. Each resulting joined object becomes an input to the query. UNNEST, JOINs can be chained.

SELECT r.author, COUNT(r.author) AS authcountFROM `travel-sample` t UNNEST reviews rWHERE t.type="hotel" GROUP BY r.authorORDER BY COUNT(r.author) DESCLIMIT 5;[ { "authcount": 2, "author": "Anita Baumbach" }, { "authcount": 2, "author": "Uriah Gutmann" }, { "authcount": 2, "author": "Ashlee Champlin" }, { "authcount": 2, "author": "Cassie O'Hara" }, { "authcount": 1, "author": "Zoe Kshlerin" }]

Querying Arrays: NEST

• NEST is the inverse of UNNEST.• Nesting is conceptually the inverse

of unnesting. Nesting performs a join across two keyspaces. But instead of producing a cross-product of the left and right inputs, a single result is produced for each left input, while the corresponding right inputs are collected into an array and nested as a single array-valued field in the result object.

SELECT * FROM `travel-sample` routeNEST `travel-sample` airline ON KEYS route.airlineid WHERE route.type = ‘airline' LIMIT 1;

[ { "airline": [ { "callsign": "AIRFRANS", "country": "France", "iata": "AF", "icao": "AFR", "id": 137, "name": "Air France", "type": "airline" } ], "route": { "airline": "AF", "airlineid": "airline_137", "destinationairport": "MRS", "distance": 2881.617376098415, "equipment": "320", "id": 10000, "schedule": [ { "day": 0, "flight": "AF198", "utc": "10:13:00" }, { "day": 0, "flight": "AF547", "utc": "19:14:00" }, { "day": 0, "flight": "AF943", "utc": "01:31:00" }, { "day": 1, "flight": "AF356", "utc": "12:40:00" }, { "day": 1, "flight": "AF480", "utc": "08:58:00" }, { "day": 1, "flight": "AF250", "utc": "12:59:00" }, { "day": 1, "flight": "AF130", "utc": "04:45:00" }, { "day": 2,

Query Performance with Arrays

Array Indexing

• Before 4.5, creating index on array attribute would index the entire array as a single scalar value.

CREATE INDEX i1 ON `travel-sample`(schedule);

"schedule": [ { "day" : 0, "flight" : "AI111", "utc" : "1:11:11"} }, { "day": 1, "flight": "AF552", "utc": "14:41:00" }, { "day": 2, "flight": "AF166",

"utc": "08:59:00" }, …]

Array Indexing - motivation

[ { "day" : 0, "special_flights" : [ { "flight" : "AI111", "utc" : ”1:11:11"}, { "flight" : "AI222", "utc" : ”2:22:22" } ] }, { "day": 1, "flight": "AF552", "utc": "14:41:00” }, …]

"London":[ "London", "Tokyo", "NewYork", …]

Why array indexing?• When NoSQL databases asked customers to denormalize, they

put the child table info into arrays in parent tables. • E.g. Each customer doc had all phone numbers, contacts, orders in arrays.

• Not easy to query - need to specify full array value in where predicates. • Ex: list of users who purchased a product – Unknown values & large list

• Was not possible to index part of the array with objects• Bloated index size (indexes whole array value)• Example: Index just the day field in array of flights in schedule.

• Performance Limitation • ANY…IN or WITHIN array

• Ease of querying - Must specify full array value in WHERE-clause• Manageable for Known or handful of values • Difficult for Unknown or Large list of values.

Who wants array indexing?

• Find my crew based on the airline.WHERE ANY p IN ods.pilot satisfies p.filen = ”XYZ1012" END ;

• Find my customer based on one of the emails on the customerWHERE ANY a IN u.telecom SATISFIES a.system = ‘email’ AND a.value = ‘a@b.com’ END ;

• Find service qualification based on arrays of arrays.WHERE ANY c_0 IN `item`.`blackoutserviceblocklist` SATISFIES

ANY c_1 IN c_0.`blackoutserviceblock`.`ppvservicelist` SATISFIES c_1.`ppvservice`.`eventcode` = "E001"

END END ;

What is Array Indexing?• Enables visibility into the array structure

schedule =

• Subset of array elements can be indexed & searched efficiently

[ { "day" : 0, "special_flights" : [ { "flight" : "AI111" , "utc" : "1:11:11"}, { "flight" : "AI932" , "utc" : "2:22:22"} ] }, { "day": 1, "flight": "AF552", "utc": "14:41:00" }, …]

How Array Indexing Helps?

• Index only required elements or attributes in the the array

• Efficient on Index storage & search time

• Benefits are lot more significant for nested arrays/objects

How Array Indexing Helps -- Example

"schedule”:[ { "day" : 0, "special_flights" : [ { "flight" : "AI111", "utc" : "1:11:11"}, { "flight" : "AI222", "utc" : "2:22:22"} ] }, { "day": 1, "flight": "AF552", "utc": "14:41:00" }, { "day": 2, "flight": "AF166",

"utc": "08:59:00" }, …]

Array Index in Couchbase

Create Array Index

• No syntax changes to DML statements• Supports all DML statements with a WHERE-clause• SELECT, UPDATE, DELETE, etc.

• Array index support only for GSI indexes. • Supports both standard secondary and memory optimized index.

CREATE INDEX isched ON `travel-sample`(DISTINCT ARRAY v.flight FOR v IN schedule END) WHERE type = "route";

Array Index syntax

CREATE INDEX isched ON `travel-sample`

(ALL ARRAY p FOR p IN public_likes END)

WHERE type = "hotel" ;

"Julius Smith", [DocID] "Corrine Hill", [DocID] "Jaeden McKenzie", [DocID] "Vallie Ryan", [DocID] "Brian Kilback", [DocID] "Lilian McLaughlin", [DocID] "Ms. Moses Feeney", [DocID] "Elnora Trantow”, [DocID]

"public_likes": [ "Julius Smith", "Corrine Hill", "Jaeden McKenzie", "Vallie Ryan", "Brian Kilback", "Lilian McLaughlin", "Ms. Moses Feeney", "Elnora Trantow" ]

Example - Indexing individual attributes/elements

• "Find the total number of flights scheduled on 3rd day" CREATE INDEX isched ON `travel-sample`(DISTINCT ARRAY v.day FOR v IN schedule END) WHERE type = "route” ;

SELECT count(*) FROM `travel-sample` WHERE type = "route" AND ANY v IN schedule SATISFIES v.day = 3 END;

Example - Indexing individual attributes/elements explain SELECT count(1) FROM `travel-sample` WHERE type = "route" AND ANY v IN schedule SATISFIES v.day = 3 END; { "#operator": "DistinctScan", "scan": { "#operator": "IndexScan", "index": "isched", "index_id": "2b24c681fa54d83f", "keyspace": "travel-sample", "namespace": "default", "spans": [ { "Range": { "High": [ "3" ], "Inclusion": 3, "Low": [ "3" ]

Example - Index with Array Elements and Other Attributes

• "Find all scheduled flights with hops, and group by number of stops"

CREATE INDEX iflight_stops ON `travel-sample` ( stops, DISTINCT ARRAY v.flight FOR v IN schedule END ) WHERE type = "route" ;

SELECT * FROM `travel-sample`WHERE type = "route" AND ANY v IN schedule SATISFIES v.flight LIKE 'AA%' END AND stops >= 0;

Example - Indexing Nested Arrays

"schedule" : [ {"day" : 0, "special_flights" : [

{"flight" : "AI111", "utc" : "1:11:11"},

{"flight" : "AI222", "utc" : "2:22:22" } ]},{"day": 1,

"flight": "AF552", "utc": "14:41:00"

} … ]

Example - Indexing Nested Arrays

• "Find the total number of special flights scheduled"

CREATE INDEX inested ON `travel-sample` (DISTINCT ARRAY(DISTINCT ARRAY y.flight FOR y IN x.special_flights END) FOR x IN schedule END) WHERE type = "route" ;

SELECT count(*) FROM `travel-sample` WHERE type ="route" ANDANY x IN schedule SATISFIES (ANY y IN x.special_flights SATISFIES y.flight IS NOT NULL END) END ;

"schedule”:[ { "day" : 0, "special_flights" : [ { "flight" : "AI111", "utc":"1:11:11"}, { "flight" : "AI222", "utc":"2:22:22"} ] }, { "day": 1, "flight": "AF552", "utc": "14:41:00" }, { "day": 2, "flight": "AF166",

"utc": "08:59:00" }, …]

Example – UNNEST

• N1QL array indexing supports both collection predicates • ANY• ANY AND EVERY

• Exploited UNNEST

CREATE INDEX idx_crew ON flight(DISTINCT ARRAY c FOR c IN public_likes END);

SELECT * FROM flight UNNEST crew_ids AS c WHERE c = "Joe Smith" ;

Restrictions in 4.5

Variable names and index keys, such as v & v.dayused in CREATE INDEX and subsequent SELECT statements must be same.

CREATE INDEX isched ON `travel-sample`(DISTINCT ARRAY v.day FOR v IN schedule END) WHERE type = "route" ;

SELECT count(*) FROM `travel-sample` WHERE type = "route" AND ANY v IN schedule SATISFIES v.day = 3 END;

Restrictions in 4.5

• Supported operators: DISTINCT ARRAY ALL ARRAY ARRAY ANY ANY AND EVERY IN, WITHIN UNNEST

• NOT supported operators: EVERY

Fun with Arrays

SELECT: Fetch Documents

SELECT * FROM customer USE KEYS ["KEY01"] ;

SELECT * FROM customer USE KEYS [ "CUST:09", "CUST:29", "CUST:234", "CUST:852", "CUST:258"] ;

SELECT status, COUNT(status)FROM customer c USE KEYS [ "CUST:09", "CUST:29", "CUST:234", "CUST:852", "CUST:258" ] WHERE c.region = 'US’GROUP BY status;

SELECT product, COUNT(product)FROM customer c USE KEYS [ "CUST:09", "CUST:29", "CUST:234", "CUST:852", "CUST:258" ] INNER JOIN locations ON KEYS c.lid WHERE c.region = 'US’GROUP BY product;

SELECT: JOIN

SELECT COUNT(1)FROM `beer-sample` beer INNER JOIN `beer-sample` brewery ON KEYS beer.brewery_idWHERE state = ‘CA’

• JOIN operation combines documents from two key spaces

• JOIN criteria is based on ON KEYS clause

• The outer table uses the index scan, if possible

• The fetch of the inner table (brewery) document-by-document

• Couchbase 4.6 improves this by fetching in batches.

SELECT: JOIN

SELECT COUNT(1)FROM ( SELECT RAW META().id FROM `beer-sample` beer WHERE state = ‘CA’) as blistINNER JOIN `beer-sample` brewery ON KEYS blist;

SELECT COUNT(1)FROM ( SELECT ARRAY_AGG(META().id) karray FROM `beer-sample` beer WHERE state = ‘CA’) as bINNER JOIN `beer-sample` brewery ON KEYS b.karray;

• Why not get all of the required document IDs from the index scan then do a big bulk get on the outer table?

• Two ways to do it.a) Use the array aggregate (ARRAY_AGG()) to create the listb) Use RAW to create the the array and then use that to JOIN.

Data.gov : New York Names{ "meta": { "view": { "id": "25th-nujf", "name": "Most Popular Baby Names by Sex and Mother's Ethnic Group, New York City", "category": "Health", "createdAt": 1382724894, "description": "The most popular baby names by sex and mother's ethnicity in New York City.", "displayType": "table", … "columns": [{ "id": -1, "name": "sid", "dataTypeName": "meta_data", "fieldName": ":sid", "position": 0, "renderTypeName": "meta_data", "format": {} }, { "id": -1, "name": "id", "dataTypeName": "meta_data", "fieldName": ":id", "position": 0, "renderTypeName": "meta_data", "format": {} } ... ] "data": [ [1, "EB6FAA1B-EE35-4D55-B07B-8E663565CCDF", 1, 1386853125, "399231", 1386853125, "399231", "{\n}", "2011", "FEMALE", "HISPANIC", "GERALDINE", "13", "75"], [2, "2DBBA431-D26F-40A1-9375-AF7C16FF2987", 2, 1386853125, "399231", 1386853125, "399231", "{\n}", "2011", "FEMALE", "HISPANIC", "GIA", "21", "67"], [3, "54318692-0577-4B21-80C8-9CAEFCEDA8BA", 3, 1386853125, "399231", 1386853125, "399231", "{\n}", "2011", "FEMALE", "HISPANIC", "GIANNA", "49", "42"] ... ] }

Data.gov : New York NamesINSERT INTO nynames (KEY UUID(), VALUE kname)SELECT {":sid":d[0],

":id":d[1], ":position":d[2],":created_at":d[3], ":created_meta":d[4],":updated_at":d[5],":updated_meta":d[6],":meta":d[7],"brth_yr":d[8],"brth_yr":d[9],"ethcty":d[10],"nm":d[11], "cnt":d[12],"rnk":d[13]} kname

FROM (SELECT d FROM datagov UNNEST data d) as u1;

Data.gov : New York NamesINSERT INTO nynames

( KEY UUID(), value o )

SELECT o FROM (

SELECT meta.`view`.columns[*].fieldName f, data FROM datagov) d UNNEST data d1

LET o = OBJECT p:d1[ARRAY_POSITION(d.f, p)] FOR p IN d.f END ;

SPLIT & CONQUOR:

SELECT name FROM `travel-sam5ple` WHERE type = 'hotel' LIMIT 5;

[ { "name": "Medway Youth Hostel" }, { "name": "The Balmoral Guesthouse" }, { "name": "The Robins" }, { "name": "Le Clos Fleuri" }, { "name": "Glasgow Grand Central" }]

• Problem: Search for a word within a string

SPLIT & CONQUER:

select name from `travel-sample` where type = 'hotel' and lower(name) LIKE '%grand%';

[ { "name": "Glasgow Grand Central" }, { "name": "Horton Grand Hotel" }, { "name": "Manchester Grand Hyatt" }, { "name": "Grande Colonial Hotel" }, { "name": "Grand Hotel Serre Chevalier" }, { "name": "The Sheraton Grand Hotel" }]

• Use the LIKE predicate• Runs in about 81 milliseconds to

search 917 documents

SPLIT & CONQUER:

CREATE INDEX idxtravelname ON `travel-sample` (DISTINCT ARRAY wrd FOR wrd IN SPLIT(LOWER(name)) END) where type = 'hotel';

SELECT name FROM `travel-sample` WHERE ANY wrd IN SPLIT(LOWER(name)) satisfies wrd = 'grand' END AND type = 'hotel';[ { "name": "The Sheraton Grand Hotel" }, { "name": "Horton Grand Hotel" }, { "name": "Grand Hotel Serre Chevalier" }, { "name": "Glasgow Grand Central" }, { "name": "Manchester Grand Hyatt" }]~

• Convert into LOWER case• Split the name into words.• SPLIT() returns a ARRAY of these

words.• Create the INDEX on this array.• Query using the Array predicate.• Query runs in 10 ms.• Benefits grow with number of

Bucket: article

{ { "tags": "JSON,N1QL,COUCHBASE,BIGDATA,NAME,data.gov,SQL", "title": "What's in a New York Name? Unlock data.gov Using N1QL " }, { "tags": "TWITTER,NOSQL,SQL,QUERIES,ANALYSIS,HASHTAGS,JSON,COUCHBASE,ANALYTICS,INDEX", "title": "SQL on Twitter: Analysis Made Easy Using N1QL" }, { "tags": "CONCURRENCY,MONGODB,COUCHBASE,INDEX,READ,WRITE,PERFORMANCE,SNAPSHOT,CONSISTENCY", "title": "Concurrency Behavior: MongoDB vs. Couchbase" }, { "tags": "COUCHBASE,N1QL,JOIN,PERFORMANCE,INDEX,DATA MODEL,FLEXIBLE,SCHEMA", "title": "JOIN Faster With Couchbase Index JOINs" }, { "tags": "NOSQL,NOSQL,BENCHMARK,SQL,JSON,COUCHBASE,MONGODB,YCSB,PERFORMANCE,QUERY,INDEX", "title": "How Couchbase Won YCSB" }}

Questions:Find all the articles with N1QL in their titleFind all the articles with COUCHBASE in their tags

{ { "tags": "JSON,N1QL,COUCHBASE,BIGDATA,NAME,data.gov,SQL", "title": "What's in a New York Name? Unlock data.gov Using N1QL " }, { "tags": "TWITTER,NOSQL,SQL,QUERIES,ANALYSIS,HASHTAGS,JSON,COUCHBASE,ANALYTICS,INDEX", "title": "SQL on Twitter: Analysis Made Easy Using N1QL" }, { "tags": "CONCURRENCY,MONGODB,COUCHBASE,INDEX,READ,WRITE,PERFORMANCE,SNAPSHOT,CONSISTENCY", "title": "Concurrency Behavior: MongoDB vs. Couchbase" }, { "tags": "COUCHBASE,N1QL,JOIN,PERFORMANCE,INDEX,DATA MODEL,FLEXIBLE,SCHEMA", "title": "JOIN Faster With Couchbase Index JOINs" }, { "tags": "NOSQL,NOSQL,BENCHMARK,SQL,JSON,COUCHBASE,MONGODB,YCSB,PERFORMANCE,QUERY,INDEX", "title": "How Couchbase Won YCSB" }}

Basic Framework

"tags": "JSON,N1QL,COUCHBASE,BIGDATA,NAME,data.gov,SQL"

["JSON", "N1QL", "COUCHBASE", "BIGDATA", "NAME", "data.gov", "SQL" ]

SPLIT() into an array Array

Distinct array wrd for wrd in split(tags,”,”) end

Index this array N1QL Query Service

SELECT *FROM articlesWHERE ANY wrd IN SPLIT(tags, ",") satisfies wrd = "COUCHBASE” END

Basic Framework

"title": "What's in a New York Name? Unlock data.gov Using N1QL "

["What's", "in", "a", "New", "York", "Name?", "Unlock", "data.gov", "Using", "N1QL" ]

SPLIT() into an array Array

??? N1QL Query Service

New Function: TOKENS() in Couchbase 4.6 – OUT in DP now. TOKENS(expression [, parameter])

expression : JSON expression parameter : options {"names":true} Include the key names in the JSON “key”:value pair. {"case":"lower"} Return the values in upper/lower case

{"specials":true} Recognize special characters like @, - to form tokens.

"tags": "JSON,N1QL,COUCHBASE,BIGDATA,NAME,data.gov,SQL", "tagsarray": [ "data", "gov", "bigdata", "n1ql", "couchbase", "sql", "json", "name" ],

select title, tags, tokens(tags, {"case":"lower"}) tagsarray, tokens(title) titlearray from articles limit 1;

"title": "What's in a New York Name? Unlock data.gov Using N1QL ", "titlearray": [ "s", "Unlock", "data", "N1QL", "gov", "in", "Using", "New", "What", "York", "a", "Name" ]

Using TOKENS() – Index on title, lower case create index ititlesearch on articles(distinct array wrd for wrd in tokens(title, {"case":"lower"}) end);explain select title from articles where any wrd in tokens(title, {"case":"lower"}) satisfies wrd = 'n1ql' end; { "#operator": "DistinctScan", "scan": { "#operator": "IndexScan", "index": "ititlesearch", "index_id": "7a162af1199565b5", "keyspace": "articles", "namespace": "default", "spans": [ { "Range": { "High": [ "\"n1ql\"" ], "Inclusion": 3, "Low": [ "\"n1ql\"" ] } } ], "using": "gsi" } },

Using TOKENS() Index on the WHOLE document create index ititlesearch2 on articles (distinct array wrd for wrd in tokens(articles, { "case":"lower" , "names":true }) end);

explain select title from articles where any wrd in tokens(articles, {"case":"lower", "names":true }) satisfies wrd = ’title' end; "#operator": "DistinctScan", "scan": { "#operator": "IndexScan", "index": "ititlesearch2", "index_id": "c60792ca9f957cfd", "keyspace": "articles", "namespace": "default", "spans": [ { "Range": { "High": [ "\"title\"" ], "Inclusion": 3, "Low": [ "\“title\"" ] } } ], "using": "gsi" } },

Keshav MurthyDirector, Couchbase Engineeringkeshav@couchase.com

Thank You!

Goal of N1QL

Give developers and enterprises an expressive, powerful, and complete language for querying, transforming, and manipulating JSON data.

Array Indexing – How array is expanded in GSI

Sl. Create Index Expression Key versions generated by Projector

Index Entries in GSI storage

1. age [K1] [K1]docid

2. age, name, children [K1, K2, [c1, c2, c3]] [K1, K2, [c1, c2, c3]]docid

3. ALL ARRAY c FOR c IN cities END [[K11, K12, K13]] [ K11]docid[ K12]docid[ K13]docid

4. ALL ARRAY c FOR c IN cities END, age [[K11, K12, K13], K2] [ K11, K2]docid[ K12, K2]docid[ K13, K2]docid

4.1 age, ALL ARRAY c FOR c IN cities END, name [K1, [ K21, K22, K23,], K3] [K1, K21, K3]docid[K1, K22, K3]docid[K1, K23, K3]docid

Array Indexing – How array is expanded in GSI

Sl. Create Index Expression Key versions generated by Projector

Index Entries in GSI storage

5. ALL ARRAY c FOR c IN cities END, children

[[K11, K12, K13], [c1, c2, c3]] [ K11, [c1, c2, c3]]docid[ K12, [c1, c2, c3]]docid[ K13, [c1, c2, c3]]docid

6. ALL ARRAY (ALL ARRAY y FOR y IN c END) FOR c IN cities END

[ [K1, K2, K3, K4, K5]]

[K1]docid[K2]docid[K3]docid[K4]docid[K5]docid

Array Indexing Performance in ForestDB (3.6K sets)

Metrics KPI Measured comments

Array Q2(stale=ok) 13000 15140 Single doc match & fetch

Array Q2(stale=false) 700 9420 Same with consistency

Array Q3(stale=ok) 1100 1435 100 doc match and fetch.

Array Indexing Performance in MOI with 30K sets

Metrics KPI Measured comments

Array Q2(stale=ok) 13000 15251 Single doc match & fetch

Array Q3(stale=ok) 1100 1371 100 doc match and fetch.

UNITED – POC on 4.0

Response times in milliseconds1 Thread 10 Thread 50 Thread

Q1 13 35.1 197.84Q2 28 66.8 285.32Q3 - 7d 160 606 2960.2Q3 - 28d 1725 8240.3 41439.86

1 Thread 5 Threads Q1 1500 31000Q2 Timed out.Q3 23000 90000

MongoDB Query

Couchbase Query

Response times in milliseconds

UNITED -- POC• Query 2 – Get the selected flight using the document key. For

each crew member (pilot and flight attendant) found in the flight details.

• Fetch the previous flight assigned to the crew member• Fetch the next flight assigned to the crew member

select ods.GMT_EST_DEP_DTM,ods.PRFL_ACT_GMT_DEP_DTM,ods.PRFL_SCHED_GMT_DEP_DTM,ods.GMT_EST_ARR_DTM,ods.PRFL_ACT_GMT_ARR_DTM,ods.PRFL_SCHED_GMT_ARR_DTM,ods.FLT_LCL_ORIG_DT,ods.PRFL_FLT_NBR,ods.PRFL_TAIL_NBR,PILOT.PRPS_RSV_IND from ods unnest ods.PILOT where ods.TYPE='CREW_ON_FLIGHT' and ((ods.PRFL_ACT_GMT_DEP_DTM is not missing and ods.PRFL_ACT_GMT_DEP_DTM > "2015-07-15T02:45:00Z") OR (ods.PRFL_ACT_GMT_DEP_DTM is missing and ods.GMT_EST_DEP_DTM is not null and ods.GMT_EST_DEP_DTM > "2015-07-15T02:45:00Z")) and any p in ods.PILOT satisfies p.FILEN = "U110679" end order by ods.GMT_EST_DEP_DTM limit 1

UNITED – POC Queries on 4.5• 422,137 documents.• Query2: BEFORE array indexing • Primary index scan• 38.91 seconds.

create index idx_odspilot on ods(DISTINCT ARRAY p.FILEN in p in PILOT END);

• Query2: AFTER array indexing • Array index scan [DistinctScan]• 8.51 millisecond

•Improvement of 4572 TIMES

Array Indexing – Size and numbers• There is no limit on number of elements in the array.

• Total size of array index key should not exceed setting max_array_seckey_size (Default = 10K)

CREATE INDEX i1 on default(ALL flights, airlineid) . Lets say a given document is:

{ "flights": ["AF552", "AF166", "AF268", "AF422"], "airlineid": "airline_137"}

The indexable array keys for the document are: [ ["AF552", "airline_137"], ["AF166", "airline_137"], ["AF268", "airline_137"], ["AF422", "airline_137"] ]

Sum of lengths above items should be < max_array_seckey_size. Setting can be increased but not decreased.

Statements : MERGE

BIG MERGE statement – Use travel-sampleexplain merge into b1 using b2 on key "11" when matched then update set b1.o3=1;

merge into b1 using (select id from b2 where x < 10) as b3 on key b3.id when matched then update set b1.o4=1;

merge into `travel-sample` using default on key "2" when matched then update set `travel-sample`.name="aaa";MERGE into WAREHOUSE using `beer-sample` ON KEY to_string("yakima_brewing_and_malting_grant_s_ales-deep_powder_winter_ale²) when matched then delete;

Utilizing arrays: modeling, indexing, and querying

Software

Distributed Indexing and Querying in Sensor Networks using Statistical Models

Hop Doub lin g Label Indexing for Point-to-Point Distance Querying on Scale-Free Networks

Working with 2D arrays - Amazon S3 · 2016-12-20 · Introduction to Data Visualization with Python Reminder: NumPy arrays Homogeneous in type Calculations all at once Indexing with

Compressed suffix arrays and suffix trees with applications to text indexing and string matching

Arrays Manipulation and 2D Plots - UC Santa Cruzdmk/old-matlab-mini-course/notes... · Arrays Manipulation and 2D Plots Generating arrays Array indexing Joining, growing and splitting

Couchbase Server 2.0 - Indexing and Querying - Deep dive

Web Search COMP90049 Knowledge Technologies Web Search ... · Basics Challenges Parsing Page analysis Tokenisation Stemming Zoning Indexing Concepts Inverted ﬁles Querying Boolean

Querying JSON with Oracle Database 12c Release 2 · Storing, Indexing and Querying JSON data in Oracle Database 12c Release 2 JSON is stored in the database using standard VARCHAR2,

Indexing and querying 1_CouchbaseSF_2013

Fast and practical indexing and querying of very large graphs Silke Triβl, Ulf Leser Humboldt-Universitat zu Berlin Presenter: Liwen Sun (Stephen) SIGMOD’07

GiS: Fast Indexing and Querying of Graph Structuresr.web.umkc.edu/raopr/TR-DB-2009-01.pdf · We propose a new way of indexing a large database of graphs and processing ... Exact subgraph

Indexing. The goal of indexing is to speed up online querying – Retrieval needs to be performed in milliseconds – Without an index, retrieval would require

Indexing and Querying XML Data for Regular Path Expressions

Spatio-temporal Indexing in Non-relational Distributed ... · indexing structure built on top of a column-family oriented distributed data store that enables efﬁcient storage, querying,

Couchbase 20-indexing querying-deep-dive-12012012

Working with Datanens230.stanford.edu/slides/week2.pdf · Week 2: Working with Data. Demo: Manipulating MRI Data. Data Types in MATLAB Numeric arrays Indexing Arrays Logicals and

Information Retrieval - Universidade NOVA de Lisboactp.di.fct.unl.pt/~jmag/ir/materials/Laboratory_guides.pdf · Information Retrieval, i.e., text representation, indexing, querying,

The Evolution of LiDAR for 3D Data Storage, Indexing and ... · PDF fileDIT Computing Research Seminar Dr. Bianca Schoen-Phelan LiDAR Storage, Indexing and Querying Dr. Bianca Schoen-Phelan

Fast querying indexing for performance (4)

What do we do in Saturdays.AI?...• Things to learn: Numpy basics, most frequent operations in Numpy,such as working with N-dimensional arrays, indexing, indexing using integer arrays,