JBoss.orgCommunity Documentation

Chapter 1. SQL Support

1.1. Identifiers
1.2. Expressions
1.2.1. Column Identifiers
1.2.2. Literals
1.2.3. Aggregate Functions
1.2.4. Case and searched case
1.2.5. Scalar subqueries
1.2.6. Parameter references
1.3. Criteria
1.4. SQL Commands
1.4.1. SELECT Command
1.4.2. INSERT Command
1.4.3. UPDATE Command
1.4.4. DELETE Command
1.4.5. EXECUTE Command
1.4.6. Procedural Relational Command
1.5. Temp Tables
1.6. SQL Clauses
1.6.1. SELECT Clause
1.6.2. FROM Clause
1.6.3. WHERE Clause
1.6.4. GROUP BY Clause
1.6.5. HAVING Clause
1.6.6. ORDER BY Clause
1.6.7. LIMIT Clause
1.6.8. INTO Clause
1.6.9. OPTION Clause
1.7. Set Operations
1.8. Subqueries
1.8.1. Inline views
1.8.2. Subqueries in the WHERE and HAVING clauses

Teiid supports SQL for issuing queries and for defining view transformations; see also Procedure Language for how SQL is used in virtual procedures and update procedures.

Teiid provides nearly all of the functionality of SQL-92 DML. SQL-99 and later features are constantly being added based upon community need. The following does not attempt to cover SQL exhaustively, but rather highlights SQL's usage within Teiid. See the grammar for the exact form of SQL accepted by Teiid.

SQL commands contain references to tables and columns. These references are in the form of identifiers, which uniquely identify the tables and columns in the context of the command. All queries are processed in the context of a virtual database, or VDB. Because information can be federated across multiple sources, tables and columns must be scoped in some manner to avoid conflicts. This scoping is provided by models, which contain the information for each data source or set of views.

Fully-qualified identifiers can always be used in SQL commands. Partially- or unqualified forms can also be used, as long as the resulting names are unambiguous in the context of the command. Different forms of qualification can be mixed in the same query.

Identifiers, literals, and functions can be combined into expressions. Expressions can be used almost anywhere in a query -- SELECT, FROM (if specifying join criteria, WHERE, GROUP BY, HAVING. However you currently cannot use expressions in an ORDER BY clause.

Literal values represent fixed values. These can any of the 'standard' data types.

Syntax Rules:

  • Integer values will be assigned an integral data type big enough to hold the value (integer, long, or biginteger).

  • Floating point values will always be parsed as a double.

  • The keyword 'null' is used to represent an absent or unknown value and is inherently untyped. In many cases, a null literal value will be assigned an implied type based on context. For example, in the function '5 + null', the null value will be assigned the type 'integer' to match the type of the value '5'. A null literal used in the SELECT clause of a query with no implied context will be assigned to type 'string'.

Some examples of simple literal values are:

  • ‘abc’

  • ‘isn’’t true’ - use an extra single tick to escape a tick in a string with single ticks

  • 5

  • -37.75e01 - scientific notation

  • 100.0 - parsed as double

  • true

  • false

  • '\u0027' - unicode character

There are 4 basic commands for manipulating data in SQL, corresponding to the CRUD create, read, update, and delete operations: INSERT, SELECT, UPDATE, and DELETE. In addition, procedures can be executed using the EXECUTE command or through a procedural relational command.

The SELECT command is used to retrieve records any number of relations.

All of these clauses other than OPTION are defined by the SQL specification. The specification also specifies the order that these clauses will be logically processed. Below is the processing order where each stage passes a set of rows to the following stage. Note that this processing model is logical and does not represent the way any actual database engine performs the processing, although it is a useful model for understanding questions about SQL.

  • FROM stage - gathers all rows from all tables involved in the query and logically joins them with a Cartesian product, producing a single large table with all columns from all tables. Joins and join criteria are then applied to filter rows that do not match the join structure.

  • WHERE stage - applies a criteria to every output row from the FROM stage, further reducing the number of rows.

  • GROUP BY stage - groups sets of rows with matching values in the group by columns.

  • HAVING stage - applies criteria to each group of rows. Criteria can only be applied to columns that will have constant values within a group (those in the grouping columns or aggregate functions applied across the group).

  • SELECT stage - specifies the column expressions that should be returned from the query. Expressions are evaluated, including aggregate functions based on the groups of rows, which will no longer exist after this point. The output columns are named using either column aliases or an implicit name determined by the engine. If SELECT DISTINCT is specified, duplicate removal will be performed on the rows being returned from the SELECT stage.

  • ORDER BY stage - sorts the rows returned from the SELECT stage as desired. Supports sorting on multiple columns in specified order, ascending or descending. The output columns will be identical to those columns returned from the SELECT stage and will have the same name.

  • LIMIT stage - returns only the specified rows (with skip and limit values).

This model can be used to understand many questions about SQL. For example, columns aliased in the SELECT clause can only be referenced by alias in the ORDER BY clause. Without knowledge of the processing model, this can be somewhat confusing. Seen in light of the model, it is clear that the ORDER BY stage is the only stage occurring after the SELECT stage, which is where the columns are named. Because the WHERE clause is processed before the SELECT, the columns have not yet been named and the aliases are not yet known.

Procedural relational commands use the syntax of a SELECT to emulate an EXEC. In a procedural relational command a procedure group names is used in a FROM clause in place of a table. That procedure will be executed in place of a normal table access if all of the necessary input values can be found in criteria against the procedure. Each combination of input values found in the criteria results in an execution of the procedure.

Multiple Execution

The usage of 'in' or join criteria can result in the procedure being executed multiple times.

Teiid supports creating temporary,or "temp", tables. Temp tables are dynamically created, but are treated as any other physical table.

Temp tables can be defined implicitly by referencing them in a SELECT INTO or in an INSERT statement or explicitly with a CREATE TABLE statement. Implicitly created temp tables must have a name that starts with '#'.

The following example is a series of statements that loads a temporary table with data from 2 sources, and with a manually inserted record, and then uses that temp table in a subsequent query.

CREATE LOCAL TEMPORARY TABLE TEMP (a integer, b integer, c integer); 
SELECT * INTO temp FROM Src1; SELECT * INTO temp FROM Src2; 
INSERT INTO temp VALUES (1,2,3); 
SELECT a,b,c FROM Src3, temp WHERE Src3.a = temp.b; 

See virtual procedures for more on temp table usage.

This section describes the clauses that are used in the various SQL commands described in the previous section. Nearly all these features follow standard SQL syntax and functionality, so any SQL reference can be used for more information.

When the into clause is specified with a SELECT, the results of the query are inserted into the specified table. This is often used to insert records into a temporary table. The INTO clause immediately precedes the FROM clause.


INTO table FROM ...

Syntax Rules:

  • The INTO clause is logically applied last in processing, after the ORDER BY and LIMIT clauses.

  • Teiid's support for SELECT INTO is similar to MS SQL Server. The target of the INTO clause is a table where the result of the rest select command will be inserted. SELECT INTO should not be used UNION query.

Teiid supports the UNION, UNION ALL, INTERSECT, EXCEPT set operation as a way of combining the results of commands.


command (UNION|INTERSECT|EXCEPT) [ALL] command [ORDER BY...]

A subquery is a SQL query embedded within another SQL query. The query containing the subquery is the outer query.

Supported subquery locations: