This document introduces the main features of AsterixDB's data model (ADM) and query language (AQL) by example. The example is a simple scenario involving (synthetic) sample data modeled after data from the social domain. This document describes a set of sample ADM datasets, together with a set of illustrative AQL queries, to introduce you to the "AsterixDB user experience". The complete set of steps required to create and load a handful of sample datasets, along with runnable queries and the expected results for each query, are included.
This document assumes that you are at least vaguely familiar with AsterixDB and why you might want to use it. Most importantly, it assumes you already have a running instance of AsterixDB and that you know how to query it using AsterixDB's basic web interface. For more information on these topics, you should go through the steps in Installing Asterix Using Managix before reading this document and make sure that you have a running AsterixDB instance ready to go. To get your feet wet, you should probably start with a simple local installation of AsterixDB on your favorite machine, accepting all of the default settings that Managix offers. Later you can graduate to trying AsterixDB on a cluster, its real intended home (since it targets Big Data). (Note: With the exception of specifying the correct locations where you put the source data for this example, there should no changes needed in your ADM or AQL statements to run the examples locally and/or to run them on a cluster when you are ready to take that step.)
As you read through this document, you should try each step for yourself on your own AsterixDB instance. Once you have reached the end, you will be fully armed and dangerous, with all the basic AsterixDB knowledge that you'll need to start down the path of modeling, storing, and querying your own semistructured data.
In this section you will learn all about modeling Big Data using ADM, the data model of the AsterixDB BDMS.
The top-level organizing concept in the AsterixDB world is the dataverse. A dataverse---short for "data universe"---is a place (similar to a database in a relational DBMS) in which to create and manage the types, datasets, functions, and other artifacts for a given AsterixDB application. When you start using an AsterixDB instance for the first time, it starts out "empty"; it contains no data other than the AsterixDB system catalogs (which live in a special dataverse called the Metadata dataverse). To store your data in AsterixDB, you will first create a dataverse and then you use it for the datatypes and datasets for managing your own data. A datatype tells AsterixDB what you know (or more accurately, what you want it to know) a priori about one of the kinds of data instances that you want AsterixDB to hold for you. A dataset is a collection of data instances of a datatype, and AsterixDB makes sure that the data instances that you put in it conform to its specified type. Since AsterixDB targets semistructured data, you can use open datatypes and tell it as little or as much as you wish about your data up front; the more you tell it up front, the less information it will have to store repeatedly in the individual data instances that you give it. Instances of open datatypes are permitted to have additional content, beyond what the datatype says, as long as they at least contain the information prescribed by the datatype definition. Open typing allows data to vary from one instance to another and it leaves wiggle room for application evolution in terms of what might need to be stored in the future. If you want to restrict data instances in a dataset to have only what the datatype says, and nothing extra, you can define a closed datatype for that dataset and AsterixDB will keep users from storing objects that have extra data in them. Datatypes are open by default unless you tell AsterixDB otherwise. Let's put these concepts to work
Our little sample scenario involves hypothetical information about users of two popular social networks, Facebook and Twitter, and their messages. We'll start by defining a dataverse called "TinySocial" to hold our datatypes and datasets. The AsterixDB data model (ADM) is essentially a superset of JSON---it's what you get by extending JSON with more data types and additional data modeling constructs borrowed from object databases. The following is how we can create the TinySocial dataverse plus a set of ADM types for modeling Twitter users, their Tweets, Facebook users, their users' employment information, and their messages. (Note: Keep in mind that this is just a tiny and somewhat silly example intended for illustrating some of the key features of AsterixDB. :-))
drop dataverse TinySocial if exists;
create dataverse TinySocial;
use dataverse TinySocial;
create type TwitterUserType as open {
screen-name: string,
lang: string,
friends_count: int64,
statuses_count: int64,
name: string,
followers_count: int64
}
create type TweetMessageType as closed {
tweetid: string,
user: TwitterUserType,
sender-location: point?,
send-time: datetime,
referred-topics: {{ string }},
message-text: string
}
create type EmploymentType as open {
organization-name: string,
start-date: date,
end-date: date?
}
create type FacebookUserType as closed {
id: int64,
alias: string,
name: string,
user-since: datetime,
friend-ids: {{ int64 }},
employment: [EmploymentType]
}
create type FacebookMessageType as closed {
message-id: int64,
author-id: int64,
in-response-to: int64?,
sender-location: point?,
message: string
}
The first three lines above tell AsterixDB to drop the old TinySocial dataverse, if one already exists, and then to create a brand new one and make it the focus of the statements that follow. The first type creation statement creates a datatype for holding information about Twitter users. It is a record type with a mix of integer and string data, very much like a (flat) relational tuple. The indicated fields are all mandatory, but because the type is open, additional fields are welcome. The second statement creates a datatype for Twitter messages; this shows how to specify a closed type. Interestingly (based on one of Twitter's APIs), each Twitter message actually embeds an instance of the sending user's information (current as of when the message was sent), so this is an example of a nested record in ADM. Twitter messages can optionally contain the sender's location, which is modeled via the sender-location field of spatial type point; the question mark following the field type indicates its optionality. An optional field is like a nullable field in SQL---it may be present or missing, but when it's present, its data type will conform to the datatype's specification. The send-time field illustrates the use of a temporal primitive type, datetime. Lastly, the referred-topics field illustrates another way that ADM is richer than the relational model; this field holds a bag (a.k.a. an unordered list) of strings. Since the overall datatype definition for Twitter messages says "closed", the fields that it lists are the only fields that instances of this type will be allowed to contain. The next two create type statements create a record type for holding information about one component of the employment history of a Facebook user and then a record type for holding the user information itself. The Facebook user type highlights a few additional ADM data model features. Its friend-ids field is a bag of integers, presumably the Facebook user ids for this user's friends, and its employment field is an ordered list of employment records. The final create type statement defines a type for handling the content of a Facebook message in our hypothetical social data storage scenario.
Before going on, we need to once again emphasize the idea that AsterixDB is aimed at storing and querying not just Big Data, but Big Semistructured Data. This means that most of the fields listed in the create type statements above could have been omitted without changing anything other than the resulting size of stored data instances on disk. AsterixDB stores its information about the fields defined a priori as separate metadata, whereas the information about other fields that are "just there" in instances of open datatypes is stored with each instance---making for more bits on disk and longer times for operations affected by data size (e.g., dataset scans). The only fields that must be specified a priori are the primary key. (AsterixDB does not yet support auto-generated keys). Indexes could be build on fields, which don't belong to fixed part of datatype's schema, as long as field's type is specified and enforced keyword is provided in the end of index definition. Index fields could also be nested arbitrarily deep in datatype definition.
Now that we have defined our datatypes, we can move on and create datasets to store the actual data. (If we wanted to, we could even have several named datasets based on any one of these datatypes.) We can do this as follows, utilizing the DDL capabilities of AsterixDB.
use dataverse TinySocial;
create dataset FacebookUsers(FacebookUserType)
primary key id;
create dataset FacebookMessages(FacebookMessageType)
primary key message-id;
create dataset TwitterUsers(TwitterUserType)
primary key screen-name;
create dataset TweetMessages(TweetMessageType)
primary key tweetid
hints(cardinality=100);
create index fbUserSinceIdx on FacebookUsers(user-since);
create index fbAuthorIdx on FacebookMessages(author-id) type btree;
create index fbSenderLocIndex on FacebookMessages(sender-location) type rtree;
create index fbMessageIdx on FacebookMessages(message) type keyword;
for $ds in dataset Metadata.Dataset return $ds;
for $ix in dataset Metadata.Index return $ix;
The ADM DDL statements above create four datasets for holding our social data in the TinySocial dataverse: FacebookUsers, FacebookMessages, TwitterUsers, and TweetMessages. The first statement creates the FacebookUsers data set. It specifies that this dataset will store data instances conforming to FacebookUserType and that it has a primary key which is the id field of each instance. The primary key information is used by AsterixDB to uniquely identify instances for the purpose of later lookup and for use in secondary indexes. Each AsterixDB dataset is stored (and indexed) in the form of a B+ tree on primary key; secondary indexes point to their indexed data by primary key. In AsterixDB clusters, the primary key is also used to hash-partition (a.k.a. shard) the dataset across the nodes of the cluster. The next three create dataset statements are similar. The last one illustrates an optional clause for providing useful hints to AsterixDB. In this case, the hint tells AsterixDB that the dataset definer is anticipating that the TweetMessages dataset will contain roughly 100 objects; knowing this can help AsterixDB to more efficiently manage and query this dataset. (AsterixDB does not yet gather and maintain data statistics; it will currently, abitrarily, assume a cardinality of one million objects per dataset in the absence of such an optional definition-time hint.)
The create dataset statements above are followed by four more DDL statements, each of which creates a secondary index on a field of one of the datasets. The first one indexes the FacebookUsers dataset on its user-since field. This index will be a B+ tree index; its type is unspecified and btree is the default type. The other three illustrate how you can explicitly specify the desired type of index. In addition to btree, rtree and inverted keyword indexes are supported by AsterixDB. Indexes can also have composite keys, and more advanced text indexing is available as well (ngram(k), where k is the desired gram length).
The last two statements above show how you can use queries in AQL to examine the AsterixDB system catalogs and tell what artifacts you have created. Just as relational DBMSs use their own tables to store their catalogs, AsterixDB uses its own datasets to persist descriptions of its datasets, datatypes, indexes, and so on. Running the first of the two queries above will list all of your newly created datasets, and it will also show you a full list of all the metadata datasets. (You can then explore from there on your own if you are curious) These last two queries also illustrate one other factoid worth knowing: AsterixDB allows queries to span dataverses by allowing the optional use of fully-qualified dataset names (i.e., dataversename.datasetname) to reference datasets that live in a dataverse other than the one that was named in the most recently executed use dataverse directive.
Okay, so far so good---AsterixDB is now ready for data, so let's give it some data to store Our next task will be to load some sample data into the four datasets that we just defined. Here we will load a tiny set of records, defined in ADM format (a superset of JSON), into each dataset. In the boxes below you can see the actual data instances contained in each of the provided sample files. In order to load this data yourself, you should first store the four corresponding .adm files (whose URLs are indicated on top of each box below) into a filesystem directory accessible to your running AsterixDB instance. Take a few minutes to look carefully at each of the sample data sets. This will give you a better sense of the nature of the data that we are about to load and query. We should note that ADM format is a textual serialization of what AsterixDB will actually store; when persisted in AsterixDB, the data format will be binary and the data in the predefined fields of the data instances will be stored separately from their associated field name and type metadata.
{"screen-name":"NathanGiesen@211","lang":"en","friends_count":18,"statuses_count":473,"name":"Nathan Giesen","followers_count":49416}
{"screen-name":"ColineGeyer@63","lang":"en","friends_count":121,"statuses_count":362,"name":"Coline Geyer","followers_count":17159}
{"screen-name":"NilaMilliron_tw","lang":"en","friends_count":445,"statuses_count":164,"name":"Nila Milliron","followers_count":22649}
{"screen-name":"ChangEwing_573","lang":"en","friends_count":182,"statuses_count":394,"name":"Chang Ewing","followers_count":32136}
{"tweetid":"1","user":{"screen-name":"NathanGiesen@211","lang":"en","friends_count":39339,"statuses_count":473,"name":"Nathan Giesen","followers_count":49416},"sender-location":point("47.44,80.65"),"send-time":datetime("2008-04-26T10:10:00"),"referred-topics":{{"t-mobile","customization"}},"message-text":" love t-mobile its customization is good:)"}
{"tweetid":"2","user":{"screen-name":"ColineGeyer@63","lang":"en","friends_count":121,"statuses_count":362,"name":"Coline Geyer","followers_count":17159},"sender-location":point("32.84,67.14"),"send-time":datetime("2010-05-13T10:10:00"),"referred-topics":{{"verizon","shortcut-menu"}},"message-text":" like verizon its shortcut-menu is awesome:)"}
{"tweetid":"3","user":{"screen-name":"NathanGiesen@211","lang":"en","friends_count":39339,"statuses_count":473,"name":"Nathan Giesen","followers_count":49416},"sender-location":point("29.72,75.8"),"send-time":datetime("2006-11-04T10:10:00"),"referred-topics":{{"motorola","speed"}},"message-text":" like motorola the speed is good:)"}
{"tweetid":"4","user":{"screen-name":"NathanGiesen@211","lang":"en","friends_count":39339,"statuses_count":473,"name":"Nathan Giesen","followers_count":49416},"sender-location":point("39.28,70.48"),"send-time":datetime("2011-12-26T10:10:00"),"referred-topics":{{"sprint","voice-command"}},"message-text":" like sprint the voice-command is mind-blowing:)"}
{"tweetid":"5","user":{"screen-name":"NathanGiesen@211","lang":"en","friends_count":39339,"statuses_count":473,"name":"Nathan Giesen","followers_count":49416},"sender-location":point("40.09,92.69"),"send-time":datetime("2006-08-04T10:10:00"),"referred-topics":{{"motorola","speed"}},"message-text":" can't stand motorola its speed is terrible:("}
{"tweetid":"6","user":{"screen-name":"ColineGeyer@63","lang":"en","friends_count":121,"statuses_count":362,"name":"Coline Geyer","followers_count":17159},"sender-location":point("47.51,83.99"),"send-time":datetime("2010-05-07T10:10:00"),"referred-topics":{{"iphone","voice-clarity"}},"message-text":" like iphone the voice-clarity is good:)"}
{"tweetid":"7","user":{"screen-name":"ChangEwing_573","lang":"en","friends_count":182,"statuses_count":394,"name":"Chang Ewing","followers_count":32136},"sender-location":point("36.21,72.6"),"send-time":datetime("2011-08-25T10:10:00"),"referred-topics":{{"samsung","platform"}},"message-text":" like samsung the platform is good"}
{"tweetid":"8","user":{"screen-name":"NathanGiesen@211","lang":"en","friends_count":39339,"statuses_count":473,"name":"Nathan Giesen","followers_count":49416},"sender-location":point("46.05,93.34"),"send-time":datetime("2005-10-14T10:10:00"),"referred-topics":{{"t-mobile","shortcut-menu"}},"message-text":" like t-mobile the shortcut-menu is awesome:)"}
{"tweetid":"9","user":{"screen-name":"NathanGiesen@211","lang":"en","friends_count":39339,"statuses_count":473,"name":"Nathan Giesen","followers_count":49416},"sender-location":point("36.86,74.62"),"send-time":datetime("2012-07-21T10:10:00"),"referred-topics":{{"verizon","voicemail-service"}},"message-text":" love verizon its voicemail-service is awesome"}
{"tweetid":"10","user":{"screen-name":"ColineGeyer@63","lang":"en","friends_count":121,"statuses_count":362,"name":"Coline Geyer","followers_count":17159},"sender-location":point("29.15,76.53"),"send-time":datetime("2008-01-26T10:10:00"),"referred-topics":{{"verizon","voice-clarity"}},"message-text":" hate verizon its voice-clarity is OMG:("}
{"tweetid":"11","user":{"screen-name":"NilaMilliron_tw","lang":"en","friends_count":445,"statuses_count":164,"name":"Nila Milliron","followers_count":22649},"sender-location":point("37.59,68.42"),"send-time":datetime("2008-03-09T10:10:00"),"referred-topics":{{"iphone","platform"}},"message-text":" can't stand iphone its platform is terrible"}
{"tweetid":"12","user":{"screen-name":"OliJackson_512","lang":"en","friends_count":445,"statuses_count":164,"name":"Oli Jackson","followers_count":22649},"sender-location":point("24.82,94.63"),"send-time":datetime("2010-02-13T10:10:00"),"referred-topics":{{"samsung","voice-command"}},"message-text":" like samsung the voice-command is amazing:)"}
{"id":1,"alias":"Margarita","name":"MargaritaStoddard","user-since":datetime("2012-08-20T10:10:00"),"friend-ids":{{2,3,6,10}},"employment":[{"organization-name":"Codetechno","start-date":date("2006-08-06")}]}
{"id":2,"alias":"Isbel","name":"IsbelDull","user-since":datetime("2011-01-22T10:10:00"),"friend-ids":{{1,4}},"employment":[{"organization-name":"Hexviafind","start-date":date("2010-04-27")}]}
{"id":3,"alias":"Emory","name":"EmoryUnk","user-since":datetime("2012-07-10T10:10:00"),"friend-ids":{{1,5,8,9}},"employment":[{"organization-name":"geomedia","start-date":date("2010-06-17"),"end-date":date("2010-01-26")}]}
{"id":4,"alias":"Nicholas","name":"NicholasStroh","user-since":datetime("2010-12-27T10:10:00"),"friend-ids":{{2}},"employment":[{"organization-name":"Zamcorporation","start-date":date("2010-06-08")}]}
{"id":5,"alias":"Von","name":"VonKemble","user-since":datetime("2010-01-05T10:10:00"),"friend-ids":{{3,6,10}},"employment":[{"organization-name":"Kongreen","start-date":date("2010-11-27")}]}
{"id":6,"alias":"Willis","name":"WillisWynne","user-since":datetime("2005-01-17T10:10:00"),"friend-ids":{{1,3,7}},"employment":[{"organization-name":"jaydax","start-date":date("2009-05-15")}]}
{"id":7,"alias":"Suzanna","name":"SuzannaTillson","user-since":datetime("2012-08-07T10:10:00"),"friend-ids":{{6}},"employment":[{"organization-name":"Labzatron","start-date":date("2011-04-19")}]}
{"id":8,"alias":"Nila","name":"NilaMilliron","user-since":datetime("2008-01-01T10:10:00"),"friend-ids":{{3}},"employment":[{"organization-name":"Plexlane","start-date":date("2010-02-28")}]}
{"id":9,"alias":"Woodrow","name":"WoodrowNehling","user-since":datetime("2005-09-20T10:10:00"),"friend-ids":{{3,10}},"employment":[{"organization-name":"Zuncan","start-date":date("2003-04-22"),"end-date":date("2009-12-13")}]}
{"id":10,"alias":"Bram","name":"BramHatch","user-since":datetime("2010-10-16T10:10:00"),"friend-ids":{{1,5,9}},"employment":[{"organization-name":"physcane","start-date":date("2007-06-05"),"end-date":date("2011-11-05")}]}
{"message-id":1,"author-id":3,"in-response-to":2,"sender-location":point("47.16,77.75"),"message":" love sprint its shortcut-menu is awesome:)"}
{"message-id":2,"author-id":1,"in-response-to":4,"sender-location":point("41.66,80.87"),"message":" dislike iphone its touch-screen is horrible"}
{"message-id":3,"author-id":2,"in-response-to":4,"sender-location":point("48.09,81.01"),"message":" like samsung the plan is amazing"}
{"message-id":4,"author-id":1,"in-response-to":2,"sender-location":point("37.73,97.04"),"message":" can't stand at&t the network is horrible:("}
{"message-id":5,"author-id":6,"in-response-to":2,"sender-location":point("34.7,90.76"),"message":" love sprint the customization is mind-blowing"}
{"message-id":6,"author-id":2,"in-response-to":1,"sender-location":point("31.5,75.56"),"message":" like t-mobile its platform is mind-blowing"}
{"message-id":7,"author-id":5,"in-response-to":15,"sender-location":point("32.91,85.05"),"message":" dislike sprint the speed is horrible"}
{"message-id":8,"author-id":1,"in-response-to":11,"sender-location":point("40.33,80.87"),"message":" like verizon the 3G is awesome:)"}
{"message-id":9,"author-id":3,"in-response-to":12,"sender-location":point("34.45,96.48"),"message":" love verizon its wireless is good"}
{"message-id":10,"author-id":1,"in-response-to":12,"sender-location":point("42.5,70.01"),"message":" can't stand motorola the touch-screen is terrible"}
{"message-id":11,"author-id":1,"in-response-to":1,"sender-location":point("38.97,77.49"),"message":" can't stand at&t its plan is terrible"}
{"message-id":12,"author-id":10,"in-response-to":6,"sender-location":point("42.26,77.76"),"message":" can't stand t-mobile its voicemail-service is OMG:("}
{"message-id":13,"author-id":10,"in-response-to":4,"sender-location":point("42.77,78.92"),"message":" dislike iphone the voice-command is bad:("}
{"message-id":14,"author-id":9,"in-response-to":12,"sender-location":point("41.33,85.28"),"message":" love at&t its 3G is good:)"}
{"message-id":15,"author-id":7,"in-response-to":11,"sender-location":point("44.47,67.11"),"message":" like iphone the voicemail-service is awesome"}
It's loading time! We can use AQL load statements to populate our datasets with the sample records shown above. The following shows how loading can be done for data stored in .adm files in your local filesystem. Note: You MUST replace the <Host Name> and <Absolute File Path> placeholders in each load statement below with valid values based on the host IP address (or host name) for the machine and directory that you have downloaded the provided .adm files to. As you do so, be very, very careful to retain the two slashes in the load statements, i.e., do not delete the two slashes that appear in front of the absolute path to your .adm files. (This will lead to a three-slash character sequence at the start of each load statement's file input path specification.)
use dataverse TinySocial;
load dataset FacebookUsers using localfs
(("path"="<Host Name>://<Absolute File Path>/fbu.adm"),("format"="adm"));
load dataset FacebookMessages using localfs
(("path"="<Host Name>://<Absolute File Path>/fbm.adm"),("format"="adm"));
load dataset TwitterUsers using localfs
(("path"="<Host Name>://<Absolute File Path>/twu.adm"),("format"="adm"));
load dataset TweetMessages using localfs
(("path"="<Host Name>://<Absolute File Path>/twm.adm"),("format"="adm"));
Congratulations! You now have sample social data stored (and indexed) in AsterixDB. (You are part of an elite and adventurous group of individuals. :-)) Now that you have successfully loaded the provided sample data into the datasets that we defined, you can start running queries against them.
The query language for AsterixDB is AQL---the Asterix Query Language. AQL is loosely based on XQuery, the language developed and standardized in the early to mid 2000's by the World Wide Web Consortium (W3C) for querying semistructured data stored in their XML format. We have tossed all of the "XML cruft" out of their language but retained many of its core ideas. We did this because its design was developed over a period of years by a diverse committee of smart and experienced language designers, including "SQL people", "functional programming people", and "XML people", all of whom were focused on how to design a new query language that operates well over semistructured data. (We decided to stand on their shoulders instead of starting from scratch and revisiting many of the same issues.) Note that AQL is not SQL and not based on SQL: In other words, AsterixDB is fully "NoSQL compliant". :-)
In this section we introduce AQL via a set of example queries, along with their expected results, based on the data above, to help you get started. Many of the most important features of AQL are presented in this set of representative queries. You can find more details in the document on the Asterix Data Model (ADM), in the AQL Reference Manual, and a complete list of built-in functions is available in the Asterix Functions document.
AQL is an expression language. Even the expression 1+1 is a valid AQL query that evaluates to 2. (Try it for yourself! Okay, maybe that's not the best use of a 512-node shared-nothing compute cluster.) Most useful AQL queries will be based on the FLWOR (pronounced "flower") expression structure that AQL has borrowed from XQuery ((http://en.wikipedia.org/wiki/FLWOR)). The FLWOR expression syntax supports both the incremental binding (for) of variables to ADM data instances in a dataset (or in the result of any AQL expression, actually) and the full binding (let) of variables to entire intermediate results in a fashion similar to temporary views in the SQL world. FLWOR is an acronym that is short for for-let-where-order by-return, naming five of the most frequently used clauses from the syntax of a full AQL query. AQL also includes group by and limit clauses, as you will see shortly. Roughly speaking, for SQL afficiandos, the for clause in AQL is like the from clause in SQL, the return clause in AQL is like the select clause in SQL (but appears at the end instead of the beginning of a query), the let clause in AQL is like SQL's with clause, and the where and order by clauses in both languages are similar.
Enough talk! Let's go ahead and try writing some queries and see about learning AQL by example.
For our first query, let's find a Facebook user based on his or her user id. Suppose the user we want is the user whose id is 8:
use dataverse TinySocial;
for $user in dataset FacebookUsers
where $user.id = 8
return $user;
The query's for clause binds the variable $user incrementally to the data instances residing in the dataset named FacebookUsers. Its where clause selects only those bindings having a user id of interest, filtering out the rest. The return clause returns the (entire) data instance for each binding that satisfies the predicate. Since this dataset is indexed on user id (its primary key), this query will be done via a quick index lookup.
The expected result for our sample data is as follows:
{ "id": 8, "alias": "Nila", "name": "NilaMilliron", "user-since": datetime("2008-01-01T10:10:00.000Z"), "friend-ids": {{ 3 }}, "employment": [ { "organization-name": "Plexlane", "start-date": date("2010-02-28"), "end-date": null } ] }
AQL, like SQL, supports a variety of different predicates. For example, for our next query, let's find the Facebook users whose ids are in the range between 2 and 4:
use dataverse TinySocial;
for $user in dataset FacebookUsers
where $user.id >= 2 and $user.id <= 4
return $user;
This query's expected result, also evaluable using the primary index on user id, is:
{ "id": 2, "alias": "Isbel", "name": "IsbelDull", "user-since": datetime("2011-01-22T10:10:00.000Z"), "friend-ids": {{ 1, 4 }}, "employment": [ { "organization-name": "Hexviafind", "start-date": date("2010-04-27"), "end-date": null } ] }
{ "id": 3, "alias": "Emory", "name": "EmoryUnk", "user-since": datetime("2012-07-10T10:10:00.000Z"), "friend-ids": {{ 1, 5, 8, 9 }}, "employment": [ { "organization-name": "geomedia", "start-date": date("2010-06-17"), "end-date": date("2010-01-26") } ] }
{ "id": 4, "alias": "Nicholas", "name": "NicholasStroh", "user-since": datetime("2010-12-27T10:10:00.000Z"), "friend-ids": {{ 2 }}, "employment": [ { "organization-name": "Zamcorporation", "start-date": date("2010-06-08"), "end-date": null } ] }
AQL can do range queries on any data type that supports the appropriate set of comparators. As an example, this next query retrieves the Facebook users who joined between July 22, 2010 and July 29, 2012:
use dataverse TinySocial;
for $user in dataset FacebookUsers
where $user.user-since >= datetime('2010-07-22T00:00:00')
and $user.user-since <= datetime('2012-07-29T23:59:59')
return $user;
The expected result for this query, also an indexable query, is as follows:
{ "id": 2, "alias": "Isbel", "name": "IsbelDull", "user-since": datetime("2011-01-22T10:10:00.000Z"), "friend-ids": {{ 1, 4 }}, "employment": [ { "organization-name": "Hexviafind", "start-date": date("2010-04-27"), "end-date": null } ] }
{ "id": 3, "alias": "Emory", "name": "EmoryUnk", "user-since": datetime("2012-07-10T10:10:00.000Z"), "friend-ids": {{ 1, 5, 8, 9 }}, "employment": [ { "organization-name": "geomedia", "start-date": date("2010-06-17"), "end-date": date("2010-01-26") } ] }
{ "id": 4, "alias": "Nicholas", "name": "NicholasStroh", "user-since": datetime("2010-12-27T10:10:00.000Z"), "friend-ids": {{ 2 }}, "employment": [ { "organization-name": "Zamcorporation", "start-date": date("2010-06-08"), "end-date": null } ] }
{ "id": 10, "alias": "Bram", "name": "BramHatch", "user-since": datetime("2010-10-16T10:10:00.000Z"), "friend-ids": {{ 1, 5, 9 }}, "employment": [ { "organization-name": "physcane", "start-date": date("2007-06-05"), "end-date": date("2011-11-05") } ] }
In addition to simply binding variables to data instances and returning them "whole", an AQL query can construct new ADM instances to return based on combinations of its variable bindings. This gives AQL the power to do joins much like those done using multi-table from clauses in SQL. For example, suppose we wanted a list of all Facebook users paired with their associated messages, with the list enumerating the author name and the message text associated with each Facebook message. We could do this as follows in AQL:
use dataverse TinySocial;
for $user in dataset FacebookUsers
for $message in dataset FacebookMessages
where $message.author-id = $user.id
return {
"uname": $user.name,
"message": $message.message
};
The result of this query is a sequence of new ADM instances, one for each author/message pair. Each instance in the result will be an ADM record containing two fields, "uname" and "message", containing the user's name and the message text, respectively, for each author/message pair. (Note that "uname" and "message" are both simple AQL expressions themselves---so in the most general case, even the resulting field names can be computed as part of the query, making AQL a very powerful tool for slicing and dicing semistructured data.)
The expected result of this example AQL join query for our sample data set is:
{ "uname": "MargaritaStoddard", "message": " dislike iphone its touch-screen is horrible" }
{ "uname": "MargaritaStoddard", "message": " can't stand at&t the network is horrible:(" }
{ "uname": "MargaritaStoddard", "message": " like verizon the 3G is awesome:)" }
{ "uname": "MargaritaStoddard", "message": " can't stand motorola the touch-screen is terrible" }
{ "uname": "MargaritaStoddard", "message": " can't stand at&t its plan is terrible" }
{ "uname": "IsbelDull", "message": " like samsung the plan is amazing" }
{ "uname": "IsbelDull", "message": " like t-mobile its platform is mind-blowing" }
{ "uname": "EmoryUnk", "message": " love sprint its shortcut-menu is awesome:)" }
{ "uname": "EmoryUnk", "message": " love verizon its wireless is good" }
{ "uname": "VonKemble", "message": " dislike sprint the speed is horrible" }
{ "uname": "WillisWynne", "message": " love sprint the customization is mind-blowing" }
{ "uname": "SuzannaTillson", "message": " like iphone the voicemail-service is awesome" }
{ "uname": "WoodrowNehling", "message": " love at&t its 3G is good:)" }
{ "uname": "BramHatch", "message": " can't stand t-mobile its voicemail-service is OMG:(" }
{ "uname": "BramHatch", "message": " dislike iphone the voice-command is bad:(" }
By default, AsterixDB evaluates equijoin queries using hash-based join methods that work well for doing ad hoc joins of very large data sets (http://en.wikipedia.org/wiki/Hash_join). On a cluster, hash partitioning is employed as AsterixDB's divide-and-conquer strategy for computing large parallel joins. AsterixDB includes other join methods, but in the absence of data statistics and selectivity estimates, it doesn't (yet) have the know-how to intelligently choose among its alternatives. We therefore asked ourselves the classic question---WWOD?---What Would Oracle Do?---and in the interim, AQL includes a clunky (but useful) hint-based mechanism for addressing the occasional need to suggest to AsterixDB which join method it should use for a particular AQL query.
The following query is similar to Query 2-A but includes a suggestion to AsterixDB that it should consider employing an index-based nested-loop join technique to process the query:
use dataverse TinySocial;
for $user in dataset FacebookUsers
for $message in dataset FacebookMessages
where $message.author-id /*+ indexnl */ = $user.id
return {
"uname": $user.name,
"message": $message.message
};
The expected result is (of course) the same as before, modulo the order of the instances. Result ordering is (intentionally) undefined in AQL in the absence of an order by clause. The query result for our sample data in this case is:
{ "uname": "EmoryUnk", "message": " love sprint its shortcut-menu is awesome:)" }
{ "uname": "MargaritaStoddard", "message": " dislike iphone its touch-screen is horrible" }
{ "uname": "IsbelDull", "message": " like samsung the plan is amazing" }
{ "uname": "MargaritaStoddard", "message": " can't stand at&t the network is horrible:(" }
{ "uname": "WillisWynne", "message": " love sprint the customization is mind-blowing" }
{ "uname": "IsbelDull", "message": " like t-mobile its platform is mind-blowing" }
{ "uname": "VonKemble", "message": " dislike sprint the speed is horrible" }
{ "uname": "MargaritaStoddard", "message": " like verizon the 3G is awesome:)" }
{ "uname": "EmoryUnk", "message": " love verizon its wireless is good" }
{ "uname": "MargaritaStoddard", "message": " can't stand motorola the touch-screen is terrible" }
{ "uname": "MargaritaStoddard", "message": " can't stand at&t its plan is terrible" }
{ "uname": "BramHatch", "message": " can't stand t-mobile its voicemail-service is OMG:(" }
{ "uname": "BramHatch", "message": " dislike iphone the voice-command is bad:(" }
{ "uname": "WoodrowNehling", "message": " love at&t its 3G is good:)" }
{ "uname": "SuzannaTillson", "message": " like iphone the voicemail-service is awesome" }
(It is worth knowing, with respect to influencing AsterixDB's query evaluation, that nested for clauses---a.k.a. joins--- are currently evaluated with the "outer" clause probing the data of the "inner" clause.)
In order to support joins between tables with missing/dangling join tuples, the designers of SQL ended up shoe-horning a subset of the relational algebra into SQL's from clause syntax---and providing a variety of join types there for users to choose from. Left outer joins are particularly important in SQL, e.g., to print a summary of customers and orders, grouped by customer, without omitting those customers who haven't placed any orders yet.
The AQL language supports nesting, both of queries and of query results, and the combination allows for an arguably cleaner/more natural approach to such queries. As an example, supposed we wanted, for each Facebook user, to produce a record that has his/her name plus a list of the messages written by that user. In SQL, this would involve a left outer join between users and messages, grouping by user, and having the user name repeated along side each message. In AQL, this sort of use case can be handled (more naturally) as follows:
use dataverse TinySocial;
for $user in dataset FacebookUsers
return {
"uname": $user.name,
"messages": for $message in dataset FacebookMessages
where $message.author-id = $user.id
return $message.message
};
This AQL query binds the variable $user to the data instances in FacebookUsers; for each user, it constructs a result record containing a "uname" field with the user's name and a "messages" field with a nested collection of all messages for that user. The nested collection for each user is specified by using a correlated subquery. (Note: While it looks like nested loops could be involved in computing the result, AsterixDB recogizes the equivalence of such a query to an outerjoin, and it will use an efficient hash-based strategy when actually computing the query's result.)
Here is this example query's expected output:
{ "uname": "MargaritaStoddard", "messages": [ " dislike iphone its touch-screen is horrible", " can't stand at&t the network is horrible:(", " like verizon the 3G is awesome:)", " can't stand motorola the touch-screen is terrible", " can't stand at&t its plan is terrible" ] }
{ "uname": "IsbelDull", "messages": [ " like samsung the plan is amazing", " like t-mobile its platform is mind-blowing" ] }
{ "uname": "EmoryUnk", "messages": [ " love sprint its shortcut-menu is awesome:)", " love verizon its wireless is good" ] }
{ "uname": "NicholasStroh", "messages": [ ] }
{ "uname": "VonKemble", "messages": [ " dislike sprint the speed is horrible" ] }
{ "uname": "WillisWynne", "messages": [ " love sprint the customization is mind-blowing" ] }
{ "uname": "SuzannaTillson", "messages": [ " like iphone the voicemail-service is awesome" ] }
{ "uname": "NilaMilliron", "messages": [ ] }
{ "uname": "WoodrowNehling", "messages": [ " love at&t its 3G is good:)" ] }
{ "uname": "BramHatch", "messages": [ " dislike iphone the voice-command is bad:(", " can't stand t-mobile its voicemail-service is OMG:(" ] }
Not all joins are expressible as equijoins and computable using equijoin-oriented algorithms. The join predicates for some use cases involve predicates with functions; AsterixDB supports the expression of such queries and will still evaluate them as best it can using nested loop based techniques (and broadcast joins in the parallel case).
As an example of such a use case, suppose that we wanted, for each tweet T, to find all of the other tweets that originated from within a circle of radius of 1 surrounding tweet T's location. In AQL, this can be specified in a manner similar to the previous query using one of the built-in functions on the spatial data type instead of id equality in the correlated query's where clause:
use dataverse TinySocial;
for $t in dataset TweetMessages
return {
"message": $t.message-text,
"nearby-messages": for $t2 in dataset TweetMessages
where spatial-distance($t.sender-location, $t2.sender-location) <= 1
return { "msgtxt":$t2.message-text}
};
Here is the expected result for this query:
{ "message": " love t-mobile its customization is good:)", "nearby-messages": [ { "msgtxt": " love t-mobile its customization is good:)" } ] }
{ "message": " hate verizon its voice-clarity is OMG:(", "nearby-messages": [ { "msgtxt": " like motorola the speed is good:)" }, { "msgtxt": " hate verizon its voice-clarity is OMG:(" } ] }
{ "message": " can't stand iphone its platform is terrible", "nearby-messages": [ { "msgtxt": " can't stand iphone its platform is terrible" } ] }
{ "message": " like samsung the voice-command is amazing:)", "nearby-messages": [ { "msgtxt": " like samsung the voice-command is amazing:)" } ] }
{ "message": " like verizon its shortcut-menu is awesome:)", "nearby-messages": [ { "msgtxt": " like verizon its shortcut-menu is awesome:)" } ] }
{ "message": " like motorola the speed is good:)", "nearby-messages": [ { "msgtxt": " hate verizon its voice-clarity is OMG:(" }, { "msgtxt": " like motorola the speed is good:)" } ] }
{ "message": " like sprint the voice-command is mind-blowing:)", "nearby-messages": [ { "msgtxt": " like sprint the voice-command is mind-blowing:)" } ] }
{ "message": " can't stand motorola its speed is terrible:(", "nearby-messages": [ { "msgtxt": " can't stand motorola its speed is terrible:(" } ] }
{ "message": " like iphone the voice-clarity is good:)", "nearby-messages": [ { "msgtxt": " like iphone the voice-clarity is good:)" } ] }
{ "message": " like samsung the platform is good", "nearby-messages": [ { "msgtxt": " like samsung the platform is good" } ] }
{ "message": " like t-mobile the shortcut-menu is awesome:)", "nearby-messages": [ { "msgtxt": " like t-mobile the shortcut-menu is awesome:)" } ] }
{ "message": " love verizon its voicemail-service is awesome", "nearby-messages": [ { "msgtxt": " love verizon its voicemail-service is awesome" } ] }
As another example of a non-equijoin use case, we could ask AsterixDB to find, for each Facebook user, all Twitter users with names "similar" to their name. AsterixDB supports a variety of "fuzzy match" functions for use with textual and set-based data. As one example, we could choose to use edit distance with a threshold of 3 as the definition of name similarity, in which case we could write the following query using AQL's operator-based syntax (~=) for testing whether or not two values are similar:
use dataverse TinySocial;
set simfunction "edit-distance";
set simthreshold "3";
for $fbu in dataset FacebookUsers
return {
"id": $fbu.id,
"name": $fbu.name,
"similar-users": for $t in dataset TweetMessages
let $tu := $t.user
where $tu.name ~= $fbu.name
return {
"twitter-screenname": $tu.screen-name,
"twitter-name": $tu.name
}
};
The expected result for this query against our sample data is:
{ "id": 1, "name": "MargaritaStoddard", "similar-users": [ ] }
{ "id": 2, "name": "IsbelDull", "similar-users": [ ] }
{ "id": 3, "name": "EmoryUnk", "similar-users": [ ] }
{ "id": 4, "name": "NicholasStroh", "similar-users": [ ] }
{ "id": 5, "name": "VonKemble", "similar-users": [ ] }
{ "id": 6, "name": "WillisWynne", "similar-users": [ ] }
{ "id": 7, "name": "SuzannaTillson", "similar-users": [ ] }
{ "id": 8, "name": "NilaMilliron", "similar-users": [ { "twitter-screenname": "NilaMilliron_tw", "twitter-name": "Nila Milliron" } ] }
{ "id": 9, "name": "WoodrowNehling", "similar-users": [ ] }
{ "id": 10, "name": "BramHatch", "similar-users": [ ] }
The expressive power of AQL includes support for queries involving "some" (existentially quantified) and "all" (universally quantified) query semantics. As an example of an existential AQL query, here we show a query to list the Facebook users who are currently employed. Such employees will have an employment history containing a record with a null end-date value, which leads us to the following AQL query:
use dataverse TinySocial;
for $fbu in dataset FacebookUsers
where (some $e in $fbu.employment satisfies is-null($e.end-date))
return $fbu;
The expected result in this case is:
{ "id": 1, "alias": "Margarita", "name": "MargaritaStoddard", "user-since": datetime("2012-08-20T10:10:00.000Z"), "friend-ids": {{ 2, 3, 6, 10 }}, "employment": [ { "organization-name": "Codetechno", "start-date": date("2006-08-06"), "end-date": null } ] }
{ "id": 2, "alias": "Isbel", "name": "IsbelDull", "user-since": datetime("2011-01-22T10:10:00.000Z"), "friend-ids": {{ 1, 4 }}, "employment": [ { "organization-name": "Hexviafind", "start-date": date("2010-04-27"), "end-date": null } ] }
{ "id": 4, "alias": "Nicholas", "name": "NicholasStroh", "user-since": datetime("2010-12-27T10:10:00.000Z"), "friend-ids": {{ 2 }}, "employment": [ { "organization-name": "Zamcorporation", "start-date": date("2010-06-08"), "end-date": null } ] }
{ "id": 5, "alias": "Von", "name": "VonKemble", "user-since": datetime("2010-01-05T10:10:00.000Z"), "friend-ids": {{ 3, 6, 10 }}, "employment": [ { "organization-name": "Kongreen", "start-date": date("2010-11-27"), "end-date": null } ] }
{ "id": 6, "alias": "Willis", "name": "WillisWynne", "user-since": datetime("2005-01-17T10:10:00.000Z"), "friend-ids": {{ 1, 3, 7 }}, "employment": [ { "organization-name": "jaydax", "start-date": date("2009-05-15"), "end-date": null } ] }
{ "id": 7, "alias": "Suzanna", "name": "SuzannaTillson", "user-since": datetime("2012-08-07T10:10:00.000Z"), "friend-ids": {{ 6 }}, "employment": [ { "organization-name": "Labzatron", "start-date": date("2011-04-19"), "end-date": null } ] }
{ "id": 8, "alias": "Nila", "name": "NilaMilliron", "user-since": datetime("2008-01-01T10:10:00.000Z"), "friend-ids": {{ 3 }}, "employment": [ { "organization-name": "Plexlane", "start-date": date("2010-02-28"), "end-date": null } ] }
As an example of a universal AQL query, here we show a query to list the Facebook users who are currently unemployed. Such employees will have an employment history containing no records with null end-date values, leading us to the following AQL query:
use dataverse TinySocial;
for $fbu in dataset FacebookUsers
where (every $e in $fbu.employment satisfies not(is-null($e.end-date)))
return $fbu;
Here is the expected result for our sample data:
{ "id": 3, "alias": "Emory", "name": "EmoryUnk", "user-since": datetime("2012-07-10T10:10:00.000Z"), "friend-ids": {{ 1, 5, 8, 9 }}, "employment": [ { "organization-name": "geomedia", "start-date": date("2010-06-17"), "end-date": date("2010-01-26") } ] }
{ "id": 9, "alias": "Woodrow", "name": "WoodrowNehling", "user-since": datetime("2005-09-20T10:10:00.000Z"), "friend-ids": {{ 3, 10 }}, "employment": [ { "organization-name": "Zuncan", "start-date": date("2003-04-22"), "end-date": date("2009-12-13") } ] }
{ "id": 10, "alias": "Bram", "name": "BramHatch", "user-since": datetime("2010-10-16T10:10:00.000Z"), "friend-ids": {{ 1, 5, 9 }}, "employment": [ { "organization-name": "physcane", "start-date": date("2007-06-05"), "end-date": date("2011-11-05") } ] }
Like SQL, the AQL language of AsterixDB provides support for computing aggregates over large amounts of data. As a very simple example, the following AQL query computes the total number of Facebook users:
use dataverse TinySocial;
count(for $fbu in dataset FacebookUsers return $fbu);
In AQL, aggregate functions can be applied to arbitrary subquery results; in this case, the count function is applied to the result of a query that enumerates the Facebook users. The expected result here is:
10
Also like SQL, AQL supports grouped aggregation. For every Twitter user, the following group-by/aggregate query counts the number of tweets sent by that user:
use dataverse TinySocial;
for $t in dataset TweetMessages
group by $uid := $t.user.screen-name with $t
return {
"user": $uid,
"count": count($t)
};
The for clause incrementally binds $t to tweets, and the group by clause groups the tweets by its issuer's Twitter screen-name. Unlike SQL, where data is tabular---flat---the data model underlying AQL allows for nesting. Thus, following the group by clause, the return clause in this query sees a sequence of $t groups, with each such group having an associated $uid variable value (i.e., the tweeting user's screen name). In the context of the return clause, due to "... with $t ...", $uid is bound to the tweeter's id and $t is bound to the set of tweets issued by that tweeter. The return clause constructs a result record containing the tweeter's user id and the count of the items in the associated tweet set. The query result will contain one such record per screen name. This query also illustrates another feature of AQL; notice that each user's screen name is accessed via a path syntax that traverses each tweet's nested record structure.
Here is the expected result for this query over the sample data:
{ "user": "ChangEwing_573", "count": 1 }
{ "user": "ColineGeyer@63", "count": 3 }
{ "user": "NathanGiesen@211", "count": 6 }
{ "user": "NilaMilliron_tw", "count": 1 }
{ "user": "OliJackson_512", "count": 1 }
As for joins, AsterixDB has multiple evaluation strategies available for processing grouped aggregate queries. For grouped aggregation, the system knows how to employ both sort-based and hash-based aggregation methods, with sort-based methods being used by default and a hint being available to suggest that a different approach be used in processing a particular AQL query.
The following query is similar to Query 9-A, but adds a hash-based aggregation hint:
use dataverse TinySocial;
for $t in dataset TweetMessages
/*+ hash*/
group by $uid := $t.user.screen-name with $t
return {
"user": $uid,
"count": count($t)
};
Here is the expected result:
{ "user": "OliJackson_512", "count": 1 }
{ "user": "ColineGeyer@63", "count": 3 }
{ "user": "NathanGiesen@211", "count": 6 }
{ "user": "NilaMilliron_tw", "count": 1 }
{ "user": "ChangEwing_573", "count": 1 }
In some use cases it is not necessary to compute the entire answer to a query. In some cases, just having the first N or top N results is sufficient. This is expressible in AQL using the limit clause combined with the order by clause.
The following AQL query returns the top 3 Twitter users based on who has issued the most tweets:
use dataverse TinySocial;
for $t in dataset TweetMessages
group by $uid := $t.user.screen-name with $t
let $c := count($t)
order by $c desc
limit 3
return {
"user": $uid,
"count": $c
};
The expected result for this query is:
{ "user": "NathanGiesen@211", "count": 6 }
{ "user": "ColineGeyer@63", "count": 3 }
{ "user": "NilaMilliron_tw", "count": 1 }
As a last example of AQL and its query power, the following query, for each tweet, finds all of the tweets that are similar based on the topics that they refer to:
use dataverse TinySocial;
set simfunction "jaccard";
set simthreshold "0.3";
for $t in dataset TweetMessages
return {
"tweet": $t,
"similar-tweets": for $t2 in dataset TweetMessages
where $t2.referred-topics ~= $t.referred-topics
and $t2.tweetid != $t.tweetid
return $t2.referred-topics
};
This query illustrates several things worth knowing in order to write fuzzy queries in AQL. First, as mentioned earlier, AQL offers an operator-based syntax for seeing whether two values are "similar" to one another or not. Second, recall that the referred-topics field of records of datatype TweetMessageType is a bag of strings. This query sets the context for its similarity join by requesting that Jaccard-based similarity semantics (http://en.wikipedia.org/wiki/Jaccard_index) be used for the query's similarity operator and that a similarity index of 0.3 be used as its similarity threshold.
The expected result for this fuzzy join query is:
{ "tweet": { "tweetid": "1", "user": { "screen-name": "NathanGiesen@211", "lang": "en", "friends_count": 39339, "statuses_count": 473, "name": "Nathan Giesen", "followers_count": 49416 }, "sender-location": point("47.44,80.65"), "send-time": datetime("2008-04-26T10:10:00.000Z"), "referred-topics": {{ "t-mobile", "customization" }}, "message-text": " love t-mobile its customization is good:)" }, "similar-tweets": [ {{ "t-mobile", "shortcut-menu" }} ] }
{ "tweet": { "tweetid": "10", "user": { "screen-name": "ColineGeyer@63", "lang": "en", "friends_count": 121, "statuses_count": 362, "name": "Coline Geyer", "followers_count": 17159 }, "sender-location": point("29.15,76.53"), "send-time": datetime("2008-01-26T10:10:00.000Z"), "referred-topics": {{ "verizon", "voice-clarity" }}, "message-text": " hate verizon its voice-clarity is OMG:(" }, "similar-tweets": [ {{ "iphone", "voice-clarity" }}, {{ "verizon", "voicemail-service" }}, {{ "verizon", "shortcut-menu" }} ] }
{ "tweet": { "tweetid": "11", "user": { "screen-name": "NilaMilliron_tw", "lang": "en", "friends_count": 445, "statuses_count": 164, "name": "Nila Milliron", "followers_count": 22649 }, "sender-location": point("37.59,68.42"), "send-time": datetime("2008-03-09T10:10:00.000Z"), "referred-topics": {{ "iphone", "platform" }}, "message-text": " can't stand iphone its platform is terrible" }, "similar-tweets": [ {{ "iphone", "voice-clarity" }}, {{ "samsung", "platform" }} ] }
{ "tweet": { "tweetid": "12", "user": { "screen-name": "OliJackson_512", "lang": "en", "friends_count": 445, "statuses_count": 164, "name": "Oli Jackson", "followers_count": 22649 }, "sender-location": point("24.82,94.63"), "send-time": datetime("2010-02-13T10:10:00.000Z"), "referred-topics": {{ "samsung", "voice-command" }}, "message-text": " like samsung the voice-command is amazing:)" }, "similar-tweets": [ {{ "samsung", "platform" }}, {{ "sprint", "voice-command" }} ] }
{ "tweet": { "tweetid": "2", "user": { "screen-name": "ColineGeyer@63", "lang": "en", "friends_count": 121, "statuses_count": 362, "name": "Coline Geyer", "followers_count": 17159 }, "sender-location": point("32.84,67.14"), "send-time": datetime("2010-05-13T10:10:00.000Z"), "referred-topics": {{ "verizon", "shortcut-menu" }}, "message-text": " like verizon its shortcut-menu is awesome:)" }, "similar-tweets": [ {{ "verizon", "voicemail-service" }}, {{ "verizon", "voice-clarity" }}, {{ "t-mobile", "shortcut-menu" }} ] }
{ "tweet": { "tweetid": "3", "user": { "screen-name": "NathanGiesen@211", "lang": "en", "friends_count": 39339, "statuses_count": 473, "name": "Nathan Giesen", "followers_count": 49416 }, "sender-location": point("29.72,75.8"), "send-time": datetime("2006-11-04T10:10:00.000Z"), "referred-topics": {{ "motorola", "speed" }}, "message-text": " like motorola the speed is good:)" }, "similar-tweets": [ {{ "motorola", "speed" }} ] }
{ "tweet": { "tweetid": "4", "user": { "screen-name": "NathanGiesen@211", "lang": "en", "friends_count": 39339, "statuses_count": 473, "name": "Nathan Giesen", "followers_count": 49416 }, "sender-location": point("39.28,70.48"), "send-time": datetime("2011-12-26T10:10:00.000Z"), "referred-topics": {{ "sprint", "voice-command" }}, "message-text": " like sprint the voice-command is mind-blowing:)" }, "similar-tweets": [ {{ "samsung", "voice-command" }} ] }
{ "tweet": { "tweetid": "5", "user": { "screen-name": "NathanGiesen@211", "lang": "en", "friends_count": 39339, "statuses_count": 473, "name": "Nathan Giesen", "followers_count": 49416 }, "sender-location": point("40.09,92.69"), "send-time": datetime("2006-08-04T10:10:00.000Z"), "referred-topics": {{ "motorola", "speed" }}, "message-text": " can't stand motorola its speed is terrible:(" }, "similar-tweets": [ {{ "motorola", "speed" }} ] }
{ "tweet": { "tweetid": "6", "user": { "screen-name": "ColineGeyer@63", "lang": "en", "friends_count": 121, "statuses_count": 362, "name": "Coline Geyer", "followers_count": 17159 }, "sender-location": point("47.51,83.99"), "send-time": datetime("2010-05-07T10:10:00.000Z"), "referred-topics": {{ "iphone", "voice-clarity" }}, "message-text": " like iphone the voice-clarity is good:)" }, "similar-tweets": [ {{ "verizon", "voice-clarity" }}, {{ "iphone", "platform" }} ] }
{ "tweet": { "tweetid": "7", "user": { "screen-name": "ChangEwing_573", "lang": "en", "friends_count": 182, "statuses_count": 394, "name": "Chang Ewing", "followers_count": 32136 }, "sender-location": point("36.21,72.6"), "send-time": datetime("2011-08-25T10:10:00.000Z"), "referred-topics": {{ "samsung", "platform" }}, "message-text": " like samsung the platform is good" }, "similar-tweets": [ {{ "iphone", "platform" }}, {{ "samsung", "voice-command" }} ] }
{ "tweet": { "tweetid": "8", "user": { "screen-name": "NathanGiesen@211", "lang": "en", "friends_count": 39339, "statuses_count": 473, "name": "Nathan Giesen", "followers_count": 49416 }, "sender-location": point("46.05,93.34"), "send-time": datetime("2005-10-14T10:10:00.000Z"), "referred-topics": {{ "t-mobile", "shortcut-menu" }}, "message-text": " like t-mobile the shortcut-menu is awesome:)" }, "similar-tweets": [ {{ "t-mobile", "customization" }}, {{ "verizon", "shortcut-menu" }} ] }
{ "tweet": { "tweetid": "9", "user": { "screen-name": "NathanGiesen@211", "lang": "en", "friends_count": 39339, "statuses_count": 473, "name": "Nathan Giesen", "followers_count": 49416 }, "sender-location": point("36.86,74.62"), "send-time": datetime("2012-07-21T10:10:00.000Z"), "referred-topics": {{ "verizon", "voicemail-service" }}, "message-text": " love verizon its voicemail-service is awesome" }, "similar-tweets": [ {{ "verizon", "voice-clarity" }}, {{ "verizon", "shortcut-menu" }} ] }
In addition to loading and querying data, AsterixDB supports incremental additions to datasets via the AQL insert statement.
The following example adds a new tweet by user "NathanGiesen@211" to the TweetMessages dataset. (An astute reader may notice that this tweet was issued a half an hour after his last tweet, so his counts have all gone up in the interim, although he appears not to have moved in the last half hour.)
use dataverse TinySocial;
insert into dataset TweetMessages
(
{"tweetid":"13",
"user":
{"screen-name":"NathanGiesen@211",
"lang":"en",
"friends_count":39345,
"statuses_count":479,
"name":"Nathan Giesen",
"followers_count":49420
},
"sender-location":point("47.44,80.65"),
"send-time":datetime("2008-04-26T10:10:35"),
"referred-topics":{{"tweeting"}},
"message-text":"tweety tweet, my fellow tweeters!"
}
);
In general, the data to be inserted may be specified using any valid AQL query expression. The insertion of a single object instance, as in this example, is just a special case where the query expression happens to be a record constructor involving only constants.
In addition to inserting new data, AsterixDB supports deletion from datasets via the AQL delete statement. The statement supports "searched delete" semantics, and its where clause can involve any valid XQuery expression.
The following example deletes the tweet that we just added from user "NathanGiesen@211". (Easy come, easy go. :-))
use dataverse TinySocial;
delete $tm from dataset TweetMessages where $tm.tweetid = "13";
It should be noted that one form of data change not yet supported by AsterixDB is in-place data modification (update). Currently, only insert and delete operations are supported; update is not. To achieve the effect of an update, two statements are currently needed---one to delete the old record from the dataset where it resides, and another to insert the new replacement record (with the same primary key but with different field values for some of the associated data content).
AsterixDB supports record-level ACID transactions that begin and terminate implicitly for each record inserted, deleted, or searched while a given AQL statement is being executed. This is quite similar to the level of transaction support found in today's NoSQL stores. AsterixDB does not support multi-statement transactions, and in fact an AQL statement that involves multiple records can itself involve multiple independent record-level transactions. An example consequence of this is that, when an AQL statement attempts to insert 1000 records, it is possible that the first 800 records could end up being committed while the remaining 200 records fail to be inserted. This situation could happen, for example, if a duplicate key exception occurs as the 801st insertion is attempted. If this happens, AsterixDB will report the error (e.g., a duplicate key exception) as the result of the offending AQL insert statement, and the application logic above will need to take the appropriate action(s) needed to assess the resulting state and to clean up and/or continue as appropriate.
That's it You are now armed and dangerous with respect to semistructured data management using AsterixDB.
AsterixDB is a powerful new BDMS---Big Data Management System---that we hope may usher in a new era of much more declarative Big Data management. AsterixDB is powerful, so use it wisely, and remember: "With great power comes great responsibility..." :-)
Please e-mail the AsterixDB user group (asterixdb-users (at) googlegroups.com) if you run into any problems or simply have further questions about the AsterixDB system, its features, or their proper use.