Blame - content/docs/0.9.0/aql/manual.html - incubator-asterixdb-site

<h2><a name="a1._Introduction_Back_to_TOC"></a><a name="Introduction" id="Introduction">1. Introduction</a><font size="4"> <a href="#toc">[Back to TOC]</a></font></h2>

241

<p>This document is intended as a reference guide to the full syntax and semantics of the Asterix Query Language (AQL), the language for talking to AsterixDB. This guide covers both the data manipulation language (DML) aspects of AQL, including its support for queries and data modification, as well as its data definition language (DDL) aspects. New AsterixDB users are encouraged to read and work through the (friendlier) guide “AsterixDB 101: An ADM and AQL Primer” before attempting to make use of this document. In addition, readers are advised to read and understand the Asterix Data Model (ADM) reference guide since a basic understanding of ADM concepts is a prerequisite to understanding AQL. In what follows, we detail the features of the AQL language in a grammar-guided manner: We list and briefly explain each of the productions in the AQL grammar, offering examples for clarity in cases where doing so seems needed or helpful.</p></div>

242

243

<h2><a name="a2._Expressions_Back_to_TOC"></a><a name="Expressions" id="Expressions">2. Expressions</a> <font size="4"><a href="#toc">[Back to TOC]</a></font></h2>

<pre>Query ::= Expression

248

</pre></div></div>

249

<p>An AQL query can be any legal AQL expression.</p>

<pre>Expression ::= ( OperatorExpr | IfThenElse | FLWOR | QuantifiedExpression )

254

</pre></div></div>

255

<p>AQL is a fully composable expression language. Each AQL expression returns zero or more Asterix Data Model (ADM) instances. There are four major kinds of expressions in AQL. At the topmost level, an AQL expression can be an OperatorExpr (similar to a mathematical expression), an IfThenElse (to choose between two alternative values), a FLWOR expression (the heart of AQL, pronounced “flower expression”), or a QuantifiedExpression (which yields a boolean value). Each will be detailed as we explore the full AQL grammar.</p>

256

257

<h3><a name="Primary_Expressions"></a>Primary Expressions</h3>

<pre>PrimaryExpr ::= Literal

262

| VariableRef

263

| ParenthesizedExpression

264

| FunctionCallExpr

265

| DatasetAccessExpression

| ListConstructor

| ObjectConstructor

</pre></div></div>

<p>The most basic building block for any AQL expression is the PrimaryExpr. This can be a simple literal (constant) value, a reference to a query variable that is in scope, a parenthesized expression, a function call, an expression accessing the ADM contents of a dataset, a newly constructed list of ADM instances, or a newly constructed ADM object.</p>

270

271

<h4><a name="Literals"></a>Literals</h4>

<pre>Literal ::= StringLiteral

| IntegerLiteral

| FloatLiteral

| DoubleLiteral

| "null"

| "true"

| "false"

StringLiteral ::= ("\"" (<ESCAPE_QUOT> | ~["\""])* "\"")

283

| ("\'" (<ESCAPE_APOS> | ~["\'"])* "\'")

284

<ESCAPE_QUOT> ::= "\\\""

285

<ESCAPE_APOS> ::= "\\\'"

286

IntegerLiteral ::= <DIGITS>

287

<DIGITS> ::= ["0" - "9"]+

288

FloatLiteral ::= <DIGITS> ( "f" | "F" )

289

| <DIGITS> ( "." <DIGITS> ( "f" | "F" ) )?

290

| "." <DIGITS> ( "f" | "F" )

291

DoubleLiteral ::= <DIGITS>

292

| <DIGITS> ( "." <DIGITS> )?

293

| "." <DIGITS>

294

</pre></div></div>

295

<p>Literals (constants) in AQL can be strings, integers, floating point values, double values, boolean constants, or the constant value null. The null value in AQL has “unknown” or “missing” value semantics, similar to (though not identical to) nulls in the relational query language SQL.</p>

296

<p>The following are some simple examples of AQL literals. Since AQL is an expression language, each example is also a complete, legal AQL query (!).</p>

297

298

<h5><a name="Examples"></a>Examples</h5>

<pre>"a string"

42

</pre></div></div></div></div>

305

306

<h4><a name="Variable_References"></a>Variable References</h4>

<pre>VariableRef ::= <VARIABLE>

311

<VARIABLE> ::= "$" <LETTER> (<LETTER> | <DIGIT> | "_")*

312

<LETTER> ::= ["A" - "Z", "a" - "z"]

313

</pre></div></div>

314

<p>A variable in AQL can be bound to any legal ADM value. A variable reference refers to the value to which an in-scope variable is bound. (E.g., a variable binding may originate from one of the for or let clauses of a FLWOR expression or from an input parameter in the context of an AQL function body.)</p>

315

316

<h5><a name="Examples"></a>Examples</h5>

<pre>$tweet

$id

</pre></div></div></div></div>

323

324

<h4><a name="Parenthesized_Expressions"></a>Parenthesized Expressions</h4>

<pre>ParenthesizedExpression ::= "(" Expression ")"

329

</pre></div></div>

330

<p>As in most languages, an expression may be parenthesized.</p>

331

<p>Since AQL is an expression language, the following example expression is actually also a complete, legal AQL query whose result is the value 2. (As such, you can have Big Fun explaining to your boss how AsterixDB and AQL can turn your 1000-node shared-nothing Big Data cluster into a $5M calculator in its spare time.)</p>

332

333

<h5><a name="Example"></a>Example</h5>

<pre>( 1 + 1 )

</pre></div></div></div></div>

339

340

<h4><a name="Function_Calls"></a>Function Calls</h4>

<pre>FunctionCallExpr ::= FunctionOrTypeName "(" ( Expression ( "," Expression )* )? ")"

345

</pre></div></div>

346

<p>Functions are included in AQL, like most languages, as a way to package useful functionality or to componentize complicated or reusable AQL computations. A function call is a legal AQL query expression that represents the ADM value resulting from the evaluation of its body expression with the given parameter bindings; the parameter value bindings can themselves be any AQL expressions.</p>

347

<p>The following example is a (built-in) function call expression whose value is 8.</p>

348

349

<h5><a name="Example"></a>Example</h5>

<pre>string-length("a string")

354

</pre></div></div></div></div>

355

356

<h4><a name="Dataset_Access"></a>Dataset Access</h4>

<pre>DatasetAccessExpression ::= "dataset" ( ( Identifier ( "." Identifier )? )

361

| ( "(" Expression ")" ) )

362

Identifier ::= <IDENTIFIER> | StringLiteral

363

<IDENTIFIER> ::= <LETTER> (<LETTER> | <DIGIT> | <SPECIALCHARS>)*

364

<SPECIALCHARS> ::= ["$", "_", "-"]

365

</pre></div></div>

366

<p>Querying Big Data is the main point of AsterixDB and AQL. Data in AsterixDB reside in datasets (collections of ADM objects), each of which in turn resides in some namespace known as a dataverse (data universe). Data access in a query expression is accomplished via a DatasetAccessExpression. Dataset access expressions are most commonly used in FLWOR expressions, where variables are bound to their contents.</p>

367

<p>Note that the Identifier that identifies a dataset (or any other Identifier in AQL) can also be a StringLiteral. This is especially useful to avoid conficts with AQL keywords (e.g. “dataset”, “null”, or “type”).</p>

368

<p>The following are three examples of legal dataset access expressions. The first one accesses a dataset called Customers in the dataverse called SalesDV. The second one accesses the Customers dataverse in whatever the current dataverse is. The third one does the same thing as the second but uses a slightly older AQL syntax.</p>

369

370

<h5><a name="Examples"></a>Examples</h5>

<pre>dataset SalesDV.Customers

375

dataset Customers

376

dataset("Customers")

377

</pre></div></div></div></div>

378

379

<h4><a name="Constructors"></a>Constructors</h4>

<pre>ListConstructor ::= ( OrderedListConstructor | UnorderedListConstructor )

384

OrderedListConstructor ::= "[" ( Expression ( "," Expression )* )? "]"

385

UnorderedListConstructor ::= "{{" ( Expression ( "," Expression )* )? "}}"

386

ObjectConstructor ::= "{" ( FieldBinding ( "," FieldBinding )* )? "}"

387

FieldBinding ::= Expression ":" Expression

388

</pre></div></div>

389

<p>A major feature of AQL is its ability to construct new ADM data instances. This is accomplished using its constructors for each of the major ADM complex object structures, namely lists (ordered or unordered) and objects. Ordered lists are like JSON arrays, while unordered lists have bag (multiset) semantics. Objects are built from attributes that are field-name/field-value pairs, again like JSON. (See the AsterixDB Data Model document for more details on each.)</p>

390

<p>The following examples illustrate how to construct a new ordered list with 3 items, a new unordered list with 4 items, and a new object with 2 fields, respectively. List elements can be homogeneous (as in the first example), which is the common case, or they may be heterogeneous (as in the second example). The data values and field name values used to construct lists and objects in constructors are all simply AQL expressions. Thus the list elements, field names, and field values used in constructors can be simple literals (as in these three examples) or they can come from query variable references or even arbitrarily complex AQL expressions.</p>

391

392

<h5><a name="Examples"></a>Examples</h5>

<pre>[ "a", "b", "c" ]

{

"project name": "AsterixDB"

402

"project members": {{ "vinayakb", "dtabass", "chenli" }}

403

}

404

</pre></div></div></div>

405

406

407

<p>When constructing nested objects there needs to be a space between the closing braces to avoid confusion with the <tt>}}</tt> token that ends an unordered list constructor: <tt>{ "a" : { "b" : "c" }}</tt> will fail to parse while <tt>{ "a" : { "b" : "c" } }</tt> will work.</p></div></div></div>

408

409

<h3><a name="Path_Expressions"></a>Path Expressions</h3>

<pre>ValueExpr ::= PrimaryExpr ( Field | Index )*

414

Field ::= "." Identifier

415

Index ::= "[" ( Expression | "?" ) "]"

416

</pre></div></div>

417

<p>Components of complex types in ADM are accessed via path expressions. Path access can be applied to the result of an AQL expression that yields an instance of such a type, e.g., a object or list instance. For objects, path access is based on field names. For ordered lists, path access is based on (zero-based) array-style indexing. AQL also supports an “I’m feeling lucky” style index accessor, [?], for selecting an arbitrary element from an ordered list. Attempts to access non-existent fields or list elements produce a null (i.e., missing information) result as opposed to signaling a runtime error.</p>

418

<p>The following examples illustrate field access for a object, index-based element access for an ordered list, and also a composition thereof.</p>

419

420

421

<h5><a name="Examples"></a>Examples</h5>

<pre>({"list": [ "a", "b", "c"]}).list

(["a", "b", "c"])[2]

({ "list": [ "a", "b", "c"]}).list[2]

430

</pre></div></div></div></div></div>

431

432

<h3><a name="Logical_Expressions"></a>Logical Expressions</h3>

<pre>OperatorExpr ::= AndExpr ( "or" AndExpr )*

437

AndExpr ::= RelExpr ( "and" RelExpr )*

438

</pre></div></div>

439

<p>As in most languages, boolean expressions can be built up from smaller expressions by combining them with the logical connectives and/or. Legal boolean values in AQL are true, false, and null. (Nulls in AQL are treated much like SQL treats its unknown truth value in boolean expressions.)</p>

440

<p>The following is an example of a conjuctive range predicate in AQL. It will yield true if $a is bound to 4, null if $a is bound to null, and false otherwise.</p>

441

442

443

<h5><a name="Example"></a>Example</h5>

<pre>$a > 3 and $a < 5

448

</pre></div></div></div></div></div>

449

450

<h3><a name="Comparison_Expressions"></a>Comparison Expressions</h3>

<pre>RelExpr ::= AddExpr ( ( "<" | ">" | "<=" | ">=" | "=" | "!=" | "~=" ) AddExpr )?

455

</pre></div></div>

456

<p>AQL has the usual list of suspects, plus one, for comparing pairs of atomic values. The “plus one” is the last operator listed above, which is the “roughly equal” operator provided for similarity queries. (See the separate document on <a href="similarity.html">AsterixDB Similarity Queries</a> for more details on similarity matching.)</p>

457

<p>An example comparison expression (which yields the boolean value true) is shown below.</p>

458

459

460

<h5><a name="Example"></a>Example</h5>

<pre>5 > 3

</pre></div></div></div></div></div>

466

467

<h3><a name="Arithmetic_Expressions"></a>Arithmetic Expressions</h3>

<pre>AddExpr ::= MultExpr ( ( "+" | "-" ) MultExpr )*

472

MultExpr ::= UnaryExpr ( ( "*" | "/" | "%" | "^"| "idiv" ) UnaryExpr )*

473

UnaryExpr ::= ( ( "+" | "-" ) )? ValueExpr

474

</pre></div></div>

475

<p>AQL also supports the usual cast of characters for arithmetic expressions. The example below evaluates to 25.</p>

476

477

478

<h5><a name="Example"></a>Example</h5>

<pre>3 ^ 2 + 4 ^ 2

</pre></div></div></div></div></div>

484

485

<h3><a name="FLWOR_Expression"></a>FLWOR Expression</h3>

<pre>FLWOR ::= ( ForClause | LetClause ) ( Clause )* ("return"|"select") Expression

490

Clause ::= ForClause | LetClause | WhereClause | OrderbyClause

491

| GroupClause | LimitClause | DistinctClause

492

ForClause ::= ("for"|"from") Variable ( "at" Variable )? "in" ( Expression )

493

LetClause ::= ("let"|"with") Variable ":=" Expression

494

WhereClause ::= "where" Expression

495

OrderbyClause ::= "order" "by" Expression ( ( "asc" ) | ( "desc" ) )?

496

( "," Expression ( ( "asc" ) | ( "desc" ) )? )*

497

GroupClause ::= "group" "by" ( Variable ":=" )? Expression ( "," ( Variable ":=" )? Expression )*

498

("with"|"keeping") VariableRef ( "," VariableRef )*

499

LimitClause ::= "limit" Expression ( "offset" Expression )?

500

DistinctClause ::= "distinct" "by" Expression ( "," Expression )*

501

Variable ::= <VARIABLE>

502

</pre></div></div>

503

<p>The heart of AQL is the FLWOR (for-let-where-orderby-return) expression. The roots of this expression were borrowed from the expression of the same name in XQuery. A FLWOR expression starts with one or more clauses that establish variable bindings. A <tt>for</tt> clause binds a variable incrementally to each element of its associated expression; it includes an optional positional variable for counting/numbering the bindings. By default no ordering is implied or assumed by a <tt>for</tt> clause. A <tt>let</tt> clause binds a variable to the collection of elements computed by its associated expression.</p>

504

<p>Following the initial <tt>for</tt> or <tt>let</tt> clause(s), a FLWOR expression may contain an arbitrary sequence of other clauses. The <tt>where</tt> clause in a FLWOR expression filters the preceding bindings via a boolean expression, much like a <tt>where</tt> clause does in a SQL query. The <tt>order by</tt> clause in a FLWOR expression induces an ordering on the data. The <tt>group by</tt> clause, discussed further below, forms groups based on its group by expressions, optionally naming the expressions’ values (which together form the grouping key for the expression). The <tt>with</tt> subclause of a <tt>group by</tt> clause specifies the variable(s) whose values should be grouped based on the grouping key(s); following the grouping clause, only the grouping key(s) and the variables named in the with subclause remain in scope, and the named grouping variables now contain lists formed from their input values. The <tt>limit</tt> clause caps the number of values returned, optionally starting its result count from a specified offset. (Web applications can use this feature for doing pagination.) The <tt>distinct</tt> clause is similar to the <tt>group-by</tt> clause, but it forms no groups; it serves only to eliminate duplicate values. As indicated by the grammar, the clauses in an AQL query can appear in any order. To interpret a query, one can think of data as flowing down through the query from the first clause to the <tt>return</tt> clause.</p>

505

<p>The following example shows a FLWOR expression that selects and returns one user from the dataset FacebookUsers.</p>

506

507

508

<h5><a name="Example"></a>Example</h5>

<pre>for $user in dataset FacebookUsers

where $user.id = 8

return $user

</pre></div></div>

<p>The next example shows a FLWOR expression that joins two datasets, FacebookUsers and FacebookMessages, returning user/message pairs. The results contain one object per pair, with result objects containing the user’s name and an entire message.</p></div>

517

518

<h5><a name="Example"></a>Example</h5>

<pre>for $user in dataset FacebookUsers

523

for $message in dataset FacebookMessages

524

where $message.author-id = $user.id

return

{

"uname": $user.name,

"message": $message.message

529

};

530

</pre></div></div>

531

<p>In the next example, a <tt>let</tt> clause is used to bind a variable to all of a user’s FacebookMessages. The query returns one object per user, with result objects containing the user’s name and the set of all messages by that user.</p></div>

532

533

<h5><a name="Example"></a>Example</h5>

<pre>for $user in dataset FacebookUsers

538

let $messages :=

539

for $message in dataset FacebookMessages

540

where $message.author-id = $user.id

541

return $message.message

return

{

"uname": $user.name,

"messages": $messages

546

};

547

</pre></div></div>

548

<p>The following example returns all TwitterUsers ordered by their followers count (most followers first) and language. When ordering <tt>null</tt> is treated as being smaller than any other value if <tt>null</tt>s are encountered in the ordering key(s).</p></div>

549

550

<h5><a name="Example"></a>Example</h5>

<pre> for $user in dataset TwitterUsers

555

order by $user.followers_count desc, $user.lang asc

556

return $user

557

</pre></div></div>

558

<p>The next example illustrates the use of the <tt>group by</tt> clause in AQL. After the <tt>group by</tt> clause in the query, only variables that are either in the <tt>group by</tt> list or in the <tt>with</tt> list are in scope. The variables in the clause’s <tt>with</tt> list will each contain a collection of items following the <tt>group by</tt> clause; the collected items are the values that the source variable was bound to in the tuples that formed the group. For grouping <tt>null</tt> is handled as a single value.</p></div>

559

560

<h5><a name="Example"></a>Example</h5>

<pre> for $x in dataset FacebookMessages

565

let $messages := $x.message

566

group by $loc := $x.sender-location with $messages

return

{

"location" : $loc,

"message" : $messages

571

}

572

</pre></div></div>

573

<p>The use of the <tt>limit</tt> clause is illustrated in the next example.</p></div>

574

575

<h5><a name="Example"></a>Example</h5>

<pre> for $user in dataset TwitterUsers

580

order by $user.followers_count desc

limit 2

return $user

</pre></div></div>

<p>The final example shows how AQL’s <tt>distinct by</tt> clause works. Each variable in scope before the distinct clause is also in scope after the <tt>distinct by</tt> clause. This clause works similarly to <tt>group by</tt>, but for each variable that contains more than one value after the <tt>distinct by</tt> clause, one value is picked nondeterministically. (If the variable is in the <tt>distinct by</tt> list, then its value will be deterministic.) Nulls are treated as a single value when they occur in a grouping field.</p></div>

585

586

<h5><a name="Example"></a>Example</h5>

<pre> for $x in dataset FacebookMessages

591

distinct by $x.sender-location

592

return

593

{

594

"location" : $x.sender-location,

595

"message" : $x.message

596

}

597

</pre></div></div>

598

<p>In order to allow SQL fans to write queries in their favored ways, AQL provides synonyms: <i>from</i> for <i>for</i>, <i>select</i> for <i>return</i>, <i>with</i> for <i>let</i>, and <i>keeping</i> for <i>with</i> in the group by clause. The following query is such an example.</p></div>

599

600

<h5><a name="Example"></a>Example</h5>

<pre> from $x in dataset FacebookMessages

605

with $messages := $x.message

606

group by $loc := $x.sender-location keeping $messages

select

{

"location" : $loc,

"message" : $messages

611

}

612

</pre></div></div></div></div></div>

613

614

<h3><a name="Conditional_Expression"></a>Conditional Expression</h3>

<pre>IfThenElse ::= "if" "(" Expression ")" "then" Expression "else" Expression

619

</pre></div></div>

620

<p>A conditional expression is useful for choosing between two alternative values based on a boolean condition. If its first (<tt>if</tt>) expression is true, its second (<tt>then</tt>) expression’s value is returned, and otherwise its third (<tt>else</tt>) expression is returned.</p>

621

<p>The following example illustrates the form of a conditional expression.</p>

622

623

624

<h5><a name="Example"></a>Example</h5>

<pre>if (2 < 3) then "yes" else "no"

629

</pre></div></div></div></div></div>

630

631

<h3><a name="Quantified_Expressions"></a>Quantified Expressions</h3>

<pre>QuantifiedExpression ::= ( ( "some" ) | ( "every" ) ) Variable "in" Expression

636

( "," Variable "in" Expression )* "satisfies" Expression

637

</pre></div></div>

638

<p>Quantified expressions are used for expressing existential or universal predicates involving the elements of a collection.</p>

639

<p>The following pair of examples illustrate the use of a quantified expression to test that every (or some) element in the set [1, 2, 3] of integers is less than three. The first example yields <tt>false</tt> and second example yields <tt>true</tt>.</p>

640

<p>It is useful to note that if the set were instead the empty set, the first expression would yield <tt>true</tt> (“every” value in an empty set satisfies the condition) while the second expression would yield <tt>false</tt> (since there isn’t “some” value, as there are no values in the set, that satisfies the condition).</p>

641

642

643

<h5><a name="Examples"></a>Examples</h5>

<pre>every $x in [ 1, 2, 3 ] satisfies $x < 3

648

some $x in [ 1, 2, 3 ] satisfies $x < 3

649

</pre></div></div></div></div></div></div>

650

651

<h2><a name="a3._Statements_Back_to_TOC"></a><a name="Statements" id="Statements">3. Statements</a> <font size="4"><a href="#toc">[Back to TOC]</a></font></h2>

<pre>Statement ::= ( SingleStatement ( ";" )? )* <EOF>

656

SingleStatement ::= DataverseDeclaration

657

| FunctionDeclaration

| CreateStatement

| DropStatement

| LoadStatement

| SetStatement

| InsertStatement

| DeleteStatement

| UpsertStatement

| Query

</pre></div></div>

<p>In addition to expresssions for queries, AQL supports a variety of statements for data definition and manipulation purposes as well as controlling the context to be used in evaluating AQL expressions. AQL supports object-level ACID transactions that begin and terminate implicitly for each object inserted, deleted, upserted, or searched while a given AQL statement is being executed.</p>

668

<p>This section details the statements supported in the AQL language.</p>

669

670

<h3><a name="Declarations"></a>Declarations</h3>

<pre>DataverseDeclaration ::= "use" "dataverse" Identifier

675

</pre></div></div>

676

<p>The world of data in an AsterixDB cluster is organized into data namespaces called dataverses. To set the default dataverse for a series of statements, the use dataverse statement is provided.</p>

677

<p>As an example, the following statement sets the default dataverse to be TinySocial.</p>

678

679

680

<h5><a name="Example"></a>Example</h5>

<pre>use dataverse TinySocial;

685

</pre></div></div>

686

<p>The set statement in AQL is used to control aspects of the expression evalation context for queries.</p>

<pre>SetStatement ::= "set" Identifier StringLiteral

691

</pre></div></div>

692

<p>As an example, the following set statements request that Jaccard similarity with a similarity threshold 0.6 be used for set similarity matching when the ~= operator is used in a query expression.</p></div>

693

694

<h5><a name="Example"></a>Example</h5>

<pre>set simfunction "jaccard";

699

set simthreshold "0.6f";

700

</pre></div></div>

701

<p>When writing a complex AQL query, it can sometimes be helpful to define one or more auxilliary functions that each address a sub-piece of the overall query. The declare function statement supports the creation of such helper functions.</p>

<pre>FunctionDeclaration ::= "declare" "function" Identifier ParameterList "{" Expression "}"

706

ParameterList ::= "(" ( <VARIABLE> ( "," <VARIABLE> )* )? ")"

707

</pre></div></div>

708

<p>The following is a very simple example of a temporary AQL function definition.</p></div>

709

710

<h5><a name="Example"></a>Example</h5>

<pre>declare function add($a, $b) {

715

$a + $b

716

};

717

</pre></div></div></div></div></div>

718

719

<h3><a name="Lifecycle_Management_Statements"></a>Lifecycle Management Statements</h3>

<pre>CreateStatement ::= "create" ( DataverseSpecification

724

| TypeSpecification

725

| DatasetSpecification

726

| IndexSpecification

727

| FunctionSpecification )

728

729

QualifiedName ::= Identifier ( "." Identifier )?

730

DoubleQualifiedName ::= Identifier "." Identifier ( "." Identifier )?

731

</pre></div></div>

732

<p>The create statement in AQL is used for creating persistent artifacts in the context of dataverses. It can be used to create new dataverses, datatypes, datasets, indexes, and user-defined AQL functions.</p>

733

734

<h4><a name="Dataverses"></a>Dataverses</h4>

<pre>DataverseSpecification ::= "dataverse" Identifier IfNotExists ( "with format" StringLiteral )?

739

</pre></div></div>

740

<p>The create dataverse statement is used to create new dataverses. To ease the authoring of reusable AQL scripts, its optional IfNotExists clause allows creation to be requested either unconditionally or only if the the dataverse does not already exist. If this clause is absent, an error will be returned if the specified dataverse already exists. The <tt>with format</tt> clause is a placeholder for future functionality that can safely be ignored.</p>

741

<p>The following example creates a dataverse named TinySocial.</p>

742

743

<h5><a name="Example"></a>Example</h5>

<pre>create dataverse TinySocial;

748

</pre></div></div></div></div>

749

750

<h4><a name="Types"></a>Types</h4>

<pre>TypeSpecification ::= "type" FunctionOrTypeName IfNotExists "as" TypeExpr

755

FunctionOrTypeName ::= QualifiedName

756

IfNotExists ::= ( "if not exists" )?

757

TypeExpr ::= ObjectTypeDef | TypeReference | OrderedListTypeDef | UnorderedListTypeDef

758

ObjectTypeDef ::= ( "closed" | "open" )? "{" ( ObjectField ( "," ObjectField )* )? "}"

759

ObjectField ::= Identifier ":" ( TypeExpr ) ( "?" )?

760

NestedField ::= Identifier ( "." Identifier )*

761

IndexField ::= NestedField ( ":" TypeReference )?

762

TypeReference ::= Identifier

763

OrderedListTypeDef ::= "[" ( TypeExpr ) "]"

764

UnorderedListTypeDef ::= "{{" ( TypeExpr ) "}}"

765

</pre></div></div>

766

<p>The create type statement is used to create a new named ADM datatype. This type can then be used to create datasets or utilized when defining one or more other ADM datatypes. Much more information about the Asterix Data Model (ADM) is available in the <a href="datamodel.html">data model reference guide</a> to ADM. A new type can be a object type, a renaming of another type, an ordered list type, or an unordered list type. A object type can be defined as being either open or closed. Instances of a closed object type are not permitted to contain fields other than those specified in the create type statement. Instances of an open object type may carry additional fields, and open is the default for a new type (if neither option is specified).</p>

767

<p>The following example creates a new ADM object type called FacebookUser type. Since it is closed, its instances will contain only what is specified in the type definition. The first four fields are traditional typed name/value pairs. The friend-ids field is an unordered list of 32-bit integers. The employment field is an ordered list of instances of another named object type, EmploymentType.</p>

768

769

<h5><a name="Example"></a>Example</h5>

<pre>create type FacebookUserType as closed {

"id" : int32,

"alias" : string,

"name" : string,

"user-since" : datetime,

778

"friend-ids" : {{ int32 }},

779

"employment" : [ EmploymentType ]

780

}

781

</pre></div></div>

782

<p>The next example creates a new ADM object type called FbUserType. Note that the type of the id field is UUID. You need to use this field type if you want to have this field be an autogenerated-PK field. Refer to the Datasets section later for more details.</p></div>

783

784

<h5><a name="Example"></a>Example</h5>

<pre>create type FbUserType as closed {

"id" : uuid,

"alias" : string,

"name" : string

}

</pre></div></div></div></div>

794

795

<h4><a name="Datasets"></a>Datasets</h4>

<pre>DatasetSpecification ::= "internal"? "dataset" QualifiedName "(" QualifiedName ")" IfNotExists

800

PrimaryKey ( "on" Identifier )? ( "hints" Properties )?

801

( "using" "compaction" "policy" CompactionPolicy ( Configuration )? )?

802

( "with filter on" Identifier )?

803

| "external" "dataset" QualifiedName "(" QualifiedName ")" IfNotExists

804

"using" AdapterName Configuration ( "hints" Properties )?

805

( "using" "compaction" "policy" CompactionPolicy ( Configuration )? )?

806

AdapterName ::= Identifier

807

Configuration ::= "(" ( KeyValuePair ( "," KeyValuePair )* )? ")"

808

KeyValuePair ::= "(" StringLiteral "=" StringLiteral ")"

809

Properties ::= ( "(" Property ( "," Property )* ")" )?

810

Property ::= Identifier "=" ( StringLiteral | IntegerLiteral )

811

FunctionSignature ::= FunctionOrTypeName "@" IntegerLiteral

812

PrimaryKey ::= "primary" "key" NestedField ( "," NestedField )* ( "autogenerated ")?

813

CompactionPolicy ::= Identifier

814

PrimaryKey ::= "primary" "key" Identifier ( "," Identifier )* ( "autogenerated ")?

815

</pre></div></div>

816

<p>The create dataset statement is used to create a new dataset. Datasets are named, unordered collections of ADM object instances; they are where data lives persistently and are the targets for queries in AsterixDB. Datasets are typed, and AsterixDB will ensure that their contents conform to their type definitions. An Internal dataset (the default) is a dataset that is stored in and managed by AsterixDB. It must have a specified unique primary key that can be used to partition data across nodes of an AsterixDB cluster. The primary key is also used in secondary indexes to uniquely identify the indexed primary data objects. Random primary key (UUID) values can be auto-generated by declaring the field to be UUID and putting “autogenerated” after the “primary key” identifier. In this case, values for the auto-generated PK field should not be provided by the user since it will be auto-generated by AsterixDB. Optionally, a filter can be created on a field to further optimize range queries with predicates on the filter’s field. (Refer to <a href="filters.html">Filter-Based LSM Index Acceleration</a> for more information about filters.)</p>

817

<p>An External dataset is stored outside of AsterixDB (currently datasets in HDFS or on the local filesystem(s) of the cluster’s nodes are supported). External dataset support allows AQL queries to treat external data as though it were stored in AsterixDB, making it possible to query “legacy” file data (e.g., Hive data) without having to physically import it into AsterixDB. For an external dataset, an appropriate adapter must be selected to handle the nature of the desired external data. (See the <a href="externaldata.html">guide to external data</a> for more information on the available adapters.)</p>

818

<p>When creating a dataset, it is possible to choose a merge policy that controls which of the underlaying LSM storage components to be merged. Currently, AsterixDB provides four different merge policies that can be configured per dataset: no-merge, constant, prefix, and correlated-prefix. The no-merge policy simply never merges disk components. While the constant policy merges disk components when the number of components reaches some constant number k, which can be configured by the user. The prefix policy relies on component sizes and the number of components to decide which components to merge. Specifically, it works by first trying to identify the smallest ordered (oldest to newest) sequence of components such that the sequence does not contain a single component that exceeds some threshold size M and that either the sum of the component’s sizes exceeds M or the number of components in the sequence exceeds another threshold C. If such a sequence of components exists, then each of the components in the sequence are merged together to form a single component. Finally, the correlated-prefix is similar to the prefix policy but it delegates the decision of merging the disk components of all the indexes in a dataset to the primary index. When the policy decides that the primary index needs to be merged (using the same decision criteria as for the prefix policy), then it will issue successive merge requests on behalf of all other indexes associated with the same dataset. The default policy for AsterixDB is the prefix policy except when there is a filter on a dataset, where the preferred policy for filters is the correlated-prefix.</p>

819

<p>The following example creates an internal dataset for storing FacefookUserType objects. It specifies that their id field is their primary key.</p>

820

821

<h5><a name="Example"></a>Example</h5>

<pre>create internal dataset FacebookUsers(FacebookUserType) primary key id;

826

</pre></div></div>

827

<p>The following example creates an internal dataset for storing FbUserType objects. It specifies that their id field is their primary key. It also specifies that the id field is an auto-generated field, meaning that a randomly generated UUID value will be assigned to each object by the system. (A user should therefore not proivde a value for this field.) Note that the id field should be UUID.</p></div>

828

829

<h5><a name="Example"></a>Example</h5>

<pre>create internal dataset FbMsgs(FbUserType) primary key id autogenerated;

834

</pre></div></div>

835

<p>The next example creates an external dataset for storing LineitemType objects. The choice of the <tt>hdfs</tt> adapter means that its data will reside in HDFS. The create statement provides parameters used by the hdfs adapter: the URL and path needed to locate the data in HDFS and a description of the data format.</p></div>

836

837

<h5><a name="Example"></a>Example</h5>

<pre>create external dataset Lineitem('LineitemType) using hdfs (

842

("hdfs"="hdfs://HOST:PORT"),

843

("path"="HDFS_PATH"),

844

("input-format"="text-input-format"),

845

("format"="delimited-text"),

846

("delimiter"="|"));

847

</pre></div></div></div></div>

848

849

<h4><a name="Indices"></a>Indices</h4>

<pre>IndexSpecification ::= "index" Identifier IfNotExists "on" QualifiedName

854

"(" ( IndexField ) ( "," IndexField )* ")" ( "type" IndexType )? ( "enforced" )?

855

IndexType ::= "btree"

856

| "rtree"

857

| "keyword"

858

| "ngram" "(" IntegerLiteral ")"

859

| "fulltext"

860

</pre></div></div>

861

<p>The create index statement creates a secondary index on one or more fields of a specified dataset. Supported index types include <tt>btree</tt> for totally ordered datatypes, <tt>rtree</tt> for spatial data, and <tt>keyword</tt>, <tt>ngram</tt>, and <tt>fulltext</tt> for textual (string) data. An index can be created on a nested field (or fields) by providing a valid path expression as an index field identifier. An index field is not required to be part of the datatype associated with a dataset if that datatype is declared as open and the field’s type is provided along with its type and the <tt>enforced</tt> keyword is specified in the end of index definition. <tt>Enforcing</tt> an open field will introduce a check that will make sure that the actual type of an indexed field (if the field exists in the object) always matches this specified (open) field type.</p>

862

<p>The following example creates a btree index called fbAuthorIdx on the author-id field of the FacebookMessages dataset. This index can be useful for accelerating exact-match queries, range search queries, and joins involving the author-id field.</p>

863

864

<h5><a name="Example"></a>Example</h5>

<pre>create index fbAuthorIdx on FacebookMessages(author-id) type btree;

869

</pre></div></div>

870

<p>The following example creates an open btree index called fbSendTimeIdx on the open send-time field of the FacebookMessages dataset having datetime type. This index can be useful for accelerating exact-match queries, range search queries, and joins involving the send-time field.</p></div>

871

872

<h5><a name="Example"></a>Example</h5>

<pre>create index fbSendTimeIdx on FacebookMessages(send-time:datetime) type btree enforced;

877

</pre></div></div>

878

<p>The following example creates a btree index called twUserScrNameIdx on the screen-name field, which is a nested field of the user field in the TweetMessages dataset. This index can be useful for accelerating exact-match queries, range search queries, and joins involving the screen-name field.</p></div>

879

880

<h5><a name="Example"></a>Example</h5>

<pre>create index twUserScrNameIdx on TweetMessages(user.screen-name) type btree;

885

</pre></div></div>

886

<p>The following example creates an rtree index called fbSenderLocIdx on the sender-location field of the FacebookMessages dataset. This index can be useful for accelerating queries that use the <a href="functions.html#spatial-intersect"><tt>spatial-intersect</tt> function</a> in a predicate involving the sender-location field.</p></div>

887

888

<h5><a name="Example"></a>Example</h5>

<pre>create index fbSenderLocIndex on FacebookMessages(sender-location) type rtree;

893

</pre></div></div>

894

<p>The following example creates a 3-gram index called fbUserIdx on the name field of the FacebookUsers dataset. This index can be used to accelerate some similarity or substring maching queries on the name field. For details refer to the <a href="similarity.html#NGram_Index">document on similarity queries</a>.</p></div>

895

896

<h5><a name="Example"></a>Example</h5>

<pre>create index fbUserIdx on FacebookUsers(name) type ngram(3);

901

</pre></div></div>

902

<p>The following example creates a keyword index called fbMessageIdx on the message field of the FacebookMessages dataset. This keyword index can be used to optimize queries with token-based similarity predicates on the message field. For details refer to the <a href="similarity.html#Keyword_Index">document on similarity queries</a>.</p></div>

903

904

<h5><a name="Example"></a>Example</h5>

<pre>create index fbMessageIdx on FacebookMessages(message) type keyword;

909

</pre></div></div>

910

<p>The following example creates a full-text index called fbMessageIdx on the message field of the FacebookMessages dataset. This full-text index can be used to optimize queries with full-text search predicates on the message field. For details refer to the <a href="fulltext.html#toc">document on full-text queries</a>.</p></div>

911

912

<h5><a name="Example"></a>Example</h5>

<pre>create index fbMessageIdx on FacebookMessages(message) type fulltext;

917

</pre></div></div></div></div>

918

919

<h4><a name="Functions"></a>Functions</h4>

920

<p>The create function statement creates a named function that can then be used and reused in AQL queries. The body of a function can be any AQL expression involving the function’s parameters.</p>

<pre>FunctionSpecification ::= "function" FunctionOrTypeName IfNotExists ParameterList "{" Expression "}"

925

</pre></div></div>

926

<p>The following is a very simple example of a create function statement. It differs from the declare function example shown previously in that it results in a function that is persistently registered by name in the specified dataverse.</p>

927

928

<h5><a name="Example"></a>Example</h5>

<pre>create function add($a, $b) {

933

$a + $b

934

};

935

</pre></div></div></div></div>

936

937

<h4><a name="Removal"></a>Removal</h4>

<pre>DropStatement ::= "drop" ( "dataverse" Identifier IfExists

942

| "type" FunctionOrTypeName IfExists

943

| "dataset" QualifiedName IfExists

944

| "index" DoubleQualifiedName IfExists

945

| "function" FunctionSignature IfExists )

946

IfExists ::= ( "if" "exists" )?

947

</pre></div></div>

948

<p>The drop statement in AQL is the inverse of the create statement. It can be used to drop dataverses, datatypes, datasets, indexes, and functions.</p>

949

<p>The following examples illustrate uses of the drop statement.</p>

950

951

<h5><a name="Example"></a>Example</h5>

<pre>drop dataset FacebookUsers if exists;

956

957

drop index FacebookUsers.fbSenderLocIndex;

958

959

drop type FacebookUserType;

960

961

drop dataverse TinySocial;

962

963

drop function add;

964

</pre></div></div></div></div></div>

965

966

<h3><a name="ImportExport_Statements"></a>Import/Export Statements</h3>

<pre>LoadStatement ::= "load" "dataset" QualifiedName "using" AdapterName Configuration ( "pre-sorted" )?

971

</pre></div></div>

972

<p>The load statement is used to initially populate a dataset via bulk loading of data from an external file. An appropriate adapter must be selected to handle the nature of the desired external data. The load statement accepts the same adapters and the same parameters as external datasets. (See the <a href="externaldata.html">guide to external data</a> for more information on the available adapters.) If a dataset has an auto-generated primary key field, a file to be imported should not include that field in it.</p>

973

<p>The following example shows how to bulk load the FacebookUsers dataset from an external file containing data that has been prepared in ADM format.</p>

974

975

976

<h5><a name="Example"></a>Example</h5>

<pre>load dataset FacebookUsers using localfs

981

(("path"="localhost:///Users/zuck/AsterixDB/load/fbu.adm"),("format"="adm"));

982

</pre></div></div></div></div></div>

983

984

<h3><a name="Modification_Statements"></a>Modification Statements</h3>

985

986

<h4><a name="Insert"></a>Insert</h4>

<pre>InsertStatement ::= "insert" "into" "dataset" QualifiedName ( "as" Variable )? Query ( "returning" Query )?

991

</pre></div></div>

992

<p>The AQL insert statement is used to insert data into a dataset. The data to be inserted comes from an AQL query expression. The expression can be as simple as a constant expression, or in general it can be any legal AQL query. Inserts in AsterixDB are processed transactionally, with the scope of each insert transaction being the insertion of a single object plus its affiliated secondary index entries (if any). If the query part of an insert returns a single object, then the insert statement itself will be a single, atomic transaction. If the query part returns multiple objects, then each object inserted will be handled independently as a tranaction. If a dataset has an auto-generated primary key field, an insert statement should not include a value for that field in it. (The system will automatically extend the provided object with this additional field and a corresponding value.). The optional “as Variable” provides a variable binding for the inserted objects, which can be used in the “returning” clause. The optional “returning Query” allows users to run simple queries/functions on the objects returned by the insert. This query cannot refer to any datasets.</p>

993

<p>The following example illustrates a query-based insertion.</p>

994

995

<h5><a name="Example"></a>Example</h5>

<pre>insert into dataset UsersCopy as $inserted (for $user in dataset FacebookUsers return $user ) returning $inserted.screen-name

1000

</pre></div></div></div></div>

1001

1002

<h4><a name="Delete"></a>Delete</h4>

<pre>DeleteStatement ::= "delete" Variable "from" "dataset" QualifiedName ( "where" Expression )?

1007

</pre></div></div>

1008

<p>The AQL delete statement is used to delete data from a target dataset. The data to be deleted is identified by a boolean expression involving the variable bound to the target dataset in the delete statement. Deletes in AsterixDB are processed transactionally, with the scope of each delete transaction being the deletion of a single object plus its affiliated secondary index entries (if any). If the boolean expression for a delete identifies a single object, then the delete statement itself will be a single, atomic transaction. If the expression identifies multiple objects, then each object deleted will be handled independently as a transaction.</p>

1009

<p>The following example illustrates a single-object deletion.</p>

1010

1011

<h5><a name="Example"></a>Example</h5>

<pre>delete $user from dataset FacebookUsers where $user.id = 8;

1016

</pre></div></div></div></div>

1017

1018

<h4><a name="Upsert"></a>Upsert</h4>

<pre>UpsertStatement ::= "upsert" "into" "dataset" QualifiedName Query

1023

</pre></div></div>

1024

<p>The AQL upsert statement is used to couple delete (if found) with insert data into a dataset. The data to be upserted comes from an AQL query expression. The expression can be as simple as a constant expression, or in general it can be any legal AQL query. Upserts in AsterixDB are processed transactionally, with the scope of each upsert transaction being the upsertion (deletion if found + insertion) of a single object plus its affiliated secondary index entries (if any). If the query part of an upsert returns a single object, then the upsert statement itself will be a single, atomic transaction. If the query part returns multiple objects, then each object upserted will be handled independently as a tranaction.</p>

1025

<p>The following example illustrates a query-based upsertion.</p>

1026

1027

<h5><a name="Example"></a>Example</h5>

<pre>upsert into dataset Users (for $user in dataset FacebookUsers return $user)

1032

</pre></div></div>