This extension is EXPERIMENTAL. The behaviour of this extension -- including the names of its functions and anything else documented about this extension -- may change without notice in a future release of PHP. Use this extension at your own risk.
In order to use the Relational Data Access Service for Service Data Objects, you will need to understand some of the concepts behind SDO: the data graph, the data object, the disconnected way of working, the change summary, XPath and property expressions, and so on. If you are not familiar with these ideas, you might want to look first at the section on SDO. In addition, the Relational DAS makes use of the PDO extension to isolate itself from specifics of different back-end relational databases. In order to use the Relational DAS you will need to be able to create and pass a PDO database connection; for this reason you might also want to take a look at the section on PDO.
The job of the Relational DAS is to move data between the application and a relational database. In order to do this it needs to be told the mapping between the database entities - tables, columns, primary keys and foreign keys - and the elements of the SDO model - types, properties, containment relationships and so on. You specify this information as metadata when you construct the Relational DAS.
The first step is to call the Relational DAS's constructor, passing the metadata that defines the mapping between database and SDO model. There are examples of this below.
The next step might be to call the executeQuery() or executePreparedQuery() methods on the Relational DAS, passing either a literal SQL statement for the DAS to prepare and execute, or a prepared statement with placeholders and a list of values to be inserted. You may also need to specify a small amount of metadata about the query itself, so that the Relational DAS knows exactly what columns will be returned from the database and in what order. You will also need to pass a PDO database connection.
The return value from executeQuery() or executePreparedQuery() is a normalised data graph containing all the data from the result set. For a query that returns data obtained from a number of tables, this graph will contain a number of data objects, linked by SDO containment relationships. There may also be SDO non-containment references within the data.
Once the query has been executed and the data graph constructed, there is no need for either that instance of the the Relational DAS or the database connection. There are no locks held on the database. Both the Relational DAS and the PDO database connection can be garbage collected.
Quite possibly the data in the data graph will go through a number of modifications. The data graph can be serialised into the PHP session and so may have a lifetime beyond just one client-server interaction. Data objects can be created and added to the graph, the data objects already in the graph can be deleted, and data objects in the graph can be modified.
Finally, the changes made to the data graph can be applied back to the database using the applyChanges() method of the Relational DAS. For this, another instance of the Relational DAS must be constructed, using the same metadata, and another connection to the database obtained. The connection and the data graph are passed to applyChanges(). At this point the Relational DAS examines the change summary and generates the necessary INSERT, UPDATE and DELETE SQL statements to apply the changes. Any UPDATE and DELETE statements are qualified with the original values of the data so that should the data have changed in the database in the meantime this will be detected. Assuming no such collisions have occurred the changes will be committed to the database. The application can then continue to work with the data graph, make more changes and apply them, or can discard it.
There are other ways of working with the data in the database: it is possible to just create data objects and write them to the database without a preliminary call to executeQuery(), for example. This scenario and others are explored in the Examples section below.
The installation instructions for all the SDO components are in the SDO install section of the SDO documentation.
In any case, the essential facts are that the Relational DAS is written in PHP and it should be placed somewhere on the PHP include_path .
Your application will of course need to include the Relational DAS with a statement like this:
The Relational DAS requires that the SDO extension be installed. The SDO extension requires a version of PHP 5.1, and the Relational DAS requires a recent version that contains an important fix for PDO. The most up-to-date information about required levels of PHP should be found in the changelog for the package on PECL. At the time of writing, though, the Relational DAS requires the most recent beta level of PHP 5.1, that is PHP 5.1.0b3.
The Relational DAS uses PDO to access the relational database, and so should run with a variety of different relational databases. At the time of writing it has been tested in the following configurations
MySQL 4.1.14, on Windows. The Relational DAS operates correctly with the php_pdo_mysql driver that comes with the pre-built binaries for PHP 5.1.0b3.
MySQL 4.1.13, on Linux. It is necessary to have the most up-to-date PDO driver for MySQL, which comes built in to PHP 5.1.0b3. It may be necessary to uninstall the usual driver that would have come from PECL using pear uninstall pdo_mysql . You will need to configure PHP with the --with-pdo-mysql option.
DB2 8.2 Personal Edition, on Windows. The Relational DAS operates correctly with the php_pdo_odbc driver that comes with the pre-built binaries for PHP 5.1.0b3.
DB2 8.2 Personal Developer's Edition, on Linux. The Developer's Edition is needed because it contains the include files needed when PHP is configured and built. You will need to configure PHP with the --with-pdo-odbc=ibm-db2 option.
The Relational DAS applies changes to the database within a user-delimited transaction: that is, it issues a call to PDO::beginTransaction() before beginning to apply changes, and PDO::commit() or PDO::rollback() on completion. Whichever database is chosen, the database and the PDO driver for the database must support these calls.
There are the following limitations in the current release of the Relational DAS:
No support for nulls. There is no support for SQL NULL type. It is not legal to assign PHP NULL to a data object property and the Relational DAS will not write that back as a NULL to the database. If nulls are found in the database on a query, the property will remain unset.
Only two types of SDO relationship. The metadata described below allows the Relational DAS to model just two types of SDO relationship: multi-valued containment relationships and single-valued non-containment references. In SDO, whether a property describes a single- or multi-valued relationship, and whether it is containment or non-containment, are independent. The full range of possibilities that SDO allows cannot all be defined. There may be relationships that it would be useful to model but which the current implementation cannot manage. One example is a single-valued containment relationship.
No support for the full range of SDO data types. The Relational DAS defines all primitive properties in the SDO model as being of type string. SDO defines a richer set of types containing various integer, float, boolean and data and time types. String is adequate for the purposes of the Relational DAS since the combination of PHP, PDO and the database will ensure that values passed as strings will be converted to the proper type before being put in the database. This does affect some scenarios in which the Relational DAS has to work with a data graph that has come from or will go to a different DAS.
Only one foreign key per table. The metadata only provides the means to specify one foreign key per table. This foreign key may be mapped to one of the two types of SDO relationship supported. Obviously there are some scenarios that cannot be described under this limitation - it is not possible to have two non-containment references from one table to another for example.
This section illustrates how the Relational DAS can be used to create, retrieve, update and delete data in a relational database. Many of the examples are illustrated with a three-table database that contains companies, departments within those companies, and employees that work in those departments. This example is used in a number of places within the SDO literature. See the examples section of the Service Data Objects specification or the Examples section of the documentation for the SDO extension.
The Relational DAS is constructed with metadata that defines the relational database and how it should be mapped to SDO. The long section that follows describes this metadata and how to construct the Relational DAS. The examples that follow it all assume that this metadata is in an included php file.
The examples below and others can all be found in the Scenarios directory in the Relational DAS package.
The Relational DAS throws exceptions in the event that it finds errors in the metadata or errors when executing SQL statements against the database. For brevity the examples below all omit the use of try/catch blocks around the calls to the Relational DAS.
These examples all differ from the expected use of SDO in two important respects.
First, they show all interactions with the database completed within one script. In this respect these scenarios are not realistic but are chosen to illustrate just the use of the Relational DAS. It is expected that interactions with the database will be separated in time and the data graph serialised and deserialised into the PHP session one or more times as the application interacts with an end user.
Second, all queries executed against the database use hard-coded queries with no variables substituted. In this case it is safe to use the simple executeQuery() call, and this is what the examples illustrate. In practice, though, it is unlikely that the SQL statement is known entirely ahead of time. In order to allow variables to be safely substituted into the SQL queries, without running the risk of injecting SQL with unknown effects, it is safer to use the executePreparedQuery() which takes a prepared SQL statement containing placeholders and a list of values to be substituted.
This first long section describes in detail how the metadata describing the database and the required SDO model is supplied to the Relational DAS.
When the constructor for the Relational DAS is invoked, it needs to be passed several pieces of information. The bulk of the information, passed as an associative array in the first argument to the constructor, tells the Relational DAS what it needs to know about the relational database. It describes the names of the tables, columns, primary keys and foreign keys. It should be fairly easy to understand what is required, and once written it can be placed in a php file and included when needed. The remainder of the information, passed in the second and third arguments to the constructor, tells the Relational DAS what it needs to know about the relationships between objects and the shape of the data graph; it ultimately determines how the data from the database is to be normalised into a graph.
The first argument to the constructor describes the target relational database.
Each table is described by an associative array with up to four keys.
|name||The name of the table.|
|columns||An array listing the names of the columns, in any order.|
|PK||The name of the column containing the primary key.|
|FK||An array with two entries, 'from' and 'to', which define a column containing a foreign key, and a table to which the foreign key points. If there are no foreign keys in the table then the 'FK' entry does not need to be specified. Only one foreign key can be specified. Only a foreign key pointing to the primary key of a table can be specified.|
This metadata corresponds to a relational database that might have been defined to MySQL as:
create table company ( id integer auto_increment, name char(20), employee_of_the_month integer, primary key(id) ); create table department ( id integer auto_increment, name char(20), location char(10), number integer(3), co_id integer, primary key(id) ); create table employee ( id integer auto_increment, name char(20), SN char(4), manager tinyint(1), dept_id integer, primary key(id) );
or to DB2 as:
create table company ( \ id integer not null generated by default as identity, \ name varchar(20), \ employee_of_the_month integer, \ primary key(id) ) create table department ( \ id integer not null generated by default as identity, \ name varchar(20), \ location varchar(10), \ number integer, \ co_id integer, \ primary key(id) ) create table employee ( \ id integer not null generated by default as identity, \ name varchar(20), \ SN char(4), \ manager smallint, \ dept_id integer, \ primary key(id) )
Note that although in this example there are no foreign keys specified to the database and so the database is not expected to enforce referential integrity, the intention behind the co_id column on the department table and the dept_id column on the employee table is they should contain the primary key of their containing company or department record, respectively. So these two columns are acting as foreign keys.
There is a third foreign key in this example, that from the employee_of_the_month column of the company record to a single row of the employee table. Note the difference in intent between this foreign key and the other two. The employee_of_the_month column represents a single-valued relationship: there can be only one employee of the month for a given company. The co_id and dept_id columns represent multi-valued relationships: a company can contain many departments and a department can contain many employees. This distinction will become evident when the remainder of the metadata picks out the company-department and department-employee relationships as containment relationships.
There are a few simple rules to be followed when constructing the database metadata:
All tables must have primary keys, and the primary keys must be specified in the metadata. Without primary keys it is not possible to keep track of object identities. As you can see from the SQL statements that create the tables, primary keys can be auto-generated, that is, generated and assigned by the database when a record is inserted. In this case the auto-generated primary key is obtained from the database and inserted into the data object immediately after the row is inserted into the database.
It is not necessary to specify in the metadata all the columns that exist in the database, only those that will be used. For example, if the company table had another column that the application did not want to access with SDO, this need not be specified in the metadata. On the other hand it would have done no harm to specify it: if specified in the metadata but never retrieved, or assigned to by the application, then the unused column will not affect anything.
In the database metadata note that the foreign key definitions identify not the destination column in the table which is pointed to, but the table name itself. Strictly, the relational model permits the destination of a foreign key to be a non-primary key. Only foreign keys that point to a primary key are useful for constructing the SDO model, so the metadata specifies the table name. It is understood that the foreign key points to the primary key of the given table.
Given these rules, and given the SQL statements that define the database, the database metadata should be easy to construct.
The Relational DAS uses the database metadata to form most of the SDO model. For each table in the database metadata, an SDO type is defined. Each column which can represent a primitive value (columns which are not defined as foreign keys) are added as properties to the SDO type.
All primitive properties are given a type of string in the SDO model, regardless of their SQL type. When writing values back to the database the Relational DAS will create SQL statements that treat the values as strings, and the database will convert them to the appropriate type.
Foreign keys are interpreted in one of two ways, depending on the metadata in the third argument to the constructor that defines the SDO containment relationships. A discussion of this is therefore deferred until the section on SDO containment relationships below.
The second argument to the constructor is the application root type. The true root of each data graph is an object of a special root type and all application data objects come somewhere below that. Of the various application types in the SDO model, one has to be the application type immediately below the root of the data graph. If there is only one table in the database metadata, the application root type can be inferred, and this argument can be omitted.
The third argument to the constructor defines how the types in the model are to be linked together to form a graph. It identifies the parent-child relationships between the types which collectively form a graph. The relationships need to be supported by foreign keys to be found in the data, in a way shortly to be described.
The metadata is an array containing one or more associative arrays, each of which identifies a parent and a child. The example below shows a parent-child relationship from company to department, and another from department to employee. Each of these will become an SDO property defining a multi-valued containment relationship in the SDO model.
Foreign keys in the database metadata are interpreted as properties with either multi-valued containment relationships or single-valued non-containment references, depending on whether they have a corresponding SDO containment relationship specified in the metadata. In the example here, the foreign keys from department to company (the co_id column in the department table) and from employee to department (the dept_id column in the employee table) are interpreted as supporting the SDO containment relationships. Each containment relationship mentioned in the SDO containment relationships metadata must have a corresponding foreign key present in the database and defined in the database metadata. The values of the foreign key columns for containment relationships do not appear in the data objects, instead each is represented by a containment relationship from the parent to the child. So the co_id column in the department row in the database, for example, does not appear as a property on the department type, but instead as a containment relationship called department on the company type. Note that the foreign key and the parent-child relationship appear to have opposite senses: the foreign key points from the department to the company, but the parent-child relationship points from company to department.
The third foreign key in this example, the employee_of_the_month , is handled differently. This is not mentioned in the SDO containment relationships metadata. As a consequence this is interpreted in the second way: it becomes a single-valued non-containment reference on the company object, to which can be assigned references to SDO data objects of the employee type. It does appear as a property on the company type. The way to assign a value to it in the SDO data graph is to have a graph that contains an employee object through the containment relationships, and to assign the object to it. This is illustrated in the later examples below.
The following set of examples all use the Relational DAS to work with a data graph containing just one application data object, a single company and the data just to be found the company table. These examples do not exercise the power of SDO or the Relational DAS and of course the same result could be achieved more economically with direct SQL statements but they are intended to illustrate how to work with the Relational DAS.
For this very simple scenario it would be possible to simplify the database metadata to include just the company table - if that were done the second and third arguments to the constructor and the column specifier used in the query example would become optional.
Example 1. Creating a data object
The simplest example is that of creating a single data object and writing it to the database. In this example a single company object is created, its name is set to 'Acme', and the Relational DAS is called to write the changes to the database. The company name is set here using the property name method. See the Examples section on the SDO extension for other ways of accessing the properties of an object.
Data objects can only be created when you have a data object to start with, however. It is for that reason that the first call to the Relational DAS here is to obtain a root object. This is in effect how to ask for an empty data graph - the special root object is the true root of the tree. The company data object is then created with a call to createDataObject() on the root object. This creates the company data object and inserts it in the graph by inserting into a multi-valued containment property on the root object called 'company'.
When the Relational DAS is called to apply the changes a simple insert statement 'INSERT INTO company (name} VALUES ("Acme");' will be constructed and executed. The auto-generated primary key will be set into the data object and the change summary will be reset, so that it would be possible to continue working with the same data object, modify it, and apply the newer changes a second time.
Example 2. Retrieving a data object
In this example a single data object is retrieved from the database - or possibly more than one if there is more than one company called 'Acme'. For each company returned, the name and id properties are echoed.
In this example the third argument to executeQuery(), the column specifier is needed as there are other tables in the metadata with column names of name and id. If there were no possible ambiguity it could be omitted.
Example 3. Updating a data object
This example combines the previous two, in the sense that in order to be updated the object must first be retrieved. The application code reverses the company name (so 'Acme' becomes 'emcA') and then the changes are written back to the database in the same way that they were when the object was created. Because the query searches for the name both ways round the program can be run repeatedly to find the company and reverse its name each time.
In this example the same instance of the Relational DAS is reused for the applyChanges(), as is the PDO database handle. This is quite alright; it also alright to allow the previous instances to be garbage collected and to obtain new instances. No state data regarding the graph is held the Relational DAS once it has returned a data graph to the application. All necessary data is either within the graph itself, or can be reconstructed from the metadata.
Example 4. Deleting a data object
Any companies called 'Acme' or its reverse 'emcA' are retrieved. They are then all deleted from the graph with unset.
In this example they are all deleted in one go by unsetting the containing property (the property defining the containment relationship). It is also possible to delete them individually.
The following set of examples all use two tables from the company database: the company and department tables. These examples exercise more of the function of the Relational DAS.
In this series of examples a company and department are created, retrieved, updated, and finally deleted. This illustrates the lifecycle for a data graph containing more than one object. Note that this example clears out the company and department tables at the start so that the exact results of the queries can be known.
You can find these examples combined into one script called 1cd-CRUD in the Scenarios directory in the Relational DAS package.
Example 5. One company, one department - Create
As in the earlier example of creating just one company data object, the first action after constructing the Relational DAS is to call createRootDataObject() to obtain the special root object of the otherwise empty data graph. The company object is then created as a child of the root object, and the department object as a child of the company object.
When it comes to applying the changes, the Relational DAS has to perform special processing to maintain the foreign keys that support the containment relationships, especially if auto-generated primary keys are involved. In this example, the relationship between the auto-generated primary key id in the company table and the co_id column in the department table must be maintained. When inserting a company and department for the first time the Relational DAS has to first insert the company row, then call PDO's getLastInsertId() method to obtain the auto-generated primary key, then add that as the value of the co_id column when inserting the department row.
Example 6. One company, one department - Retrieve and Update
In this case the SQL query passed to executeQuery() performs an inner join to join the data from the company and department tables. Primary keys for both the company and department tables must be included in the query. The result set is re-normalised to form a normalised data graph. Note that a column specifier is passed as the third argument to the executeQuery() call enabling the Relational DAS to know which column is which in the result set.
Note that the co_id column although used in the query is not needed in the result set. In order to understand what the Relational DAS is doing when it builds the data graph it may be helpful to visualise what the result set looks like. Although the data in the database is normalised, so that multiple department rows can point through their foreign key to one company row, the data in the result set is non-normalised: that is, if there is one company and multiple departments, the values for the company are repeated in each row. The Relational DAS has to reverse this process and turn the result set back into a normalised data graph, with just one company object.
In this example the Relational DAS will examine the result set and column specifier, find data for both the company and department tables, find primary keys for both, and interpret each row as containing data for a department and its parent company. If it has not seen data for that company before (it uses the primary key to check) it creates a company object and then a department object underneath it. If it has seen data for that company before and has already created the company object it just creates the department object underneath.
In this way the Relational DAS can retrieve and renormalise data for multiple companies and multiple departments underneath them.
Example 7. One company, two departments - Retrieve and Delete
In this example the company and department are retrieved and then deleted. It is not necessary to delete them individually (although that would be possible) - deleting the company object from the data graph also deletes any departments underneath it.
Note the way that the company object is actually deleted using the PHP unset call. The unset has to be performed on the containing property which in this case is the company property on the special root object. You must use:
The following examples use all three tables from the company database: the company, department, and employee tables. These introduce the final piece of function not exercised by the examples above: the non-containment reference employee_of_the_month.
Like the examples above for company and department, this set of examples is intended to illustrate the full lifecycle of such a data graph.
Example 8. One company, one department, one employee - Create
In this example a company is created containing one department and just one employee. Note that this example clears out all three tables at the start so that the exact results of the queries can be known.
Note how once the company, department and employee have been created, the employee_of_the_month property of the company can be made to point at the new employee. As this is a non-containment reference, this cannot be done until the employee object has been created within the graph. Non-containment references need to be managed carefully. For example if the employee were now deleted from under the department, it would not be correct to try to save the graph without first clearing or re-assigning the employee_of_the_month property. The closure rule for SDO data graphs requires that any object pointed at by a non-containment reference must also be reachable by containment relationships.
When it comes to inserting the graph into the database, the procedure is similar to the example of inserting the company and department, but employee_of_the_month introduces an extra complexity. The Relational DAS needs to insert the objects working down the tree formed by containment relationships, so company, then department, then employee. This is necessary so that it always has the auto-generated primary key of a parent on hand to include in a child row. But when the company row is inserted the employee who is employee of the month has not yet been inserted and the primary key is not known. The procedure is that after the employee record is inserted and its primary key known, a final step is performed in which the the company record is updated with the employee's primary key.
Example 9. One company, one department, one employee - Retrieve and update
The SQL statement passed to the Relational DAS is this time an inner join that retrieves data from all three tables. Otherwise this example introduces nothing that has not appeared in a previous example.
The graph is updated by the addition of a new department and employee and some alterations to the name properties of the existing objects in the graph. The combined changes are then written back. The Relational DAS will process and apply an arbitrary mixture of additions, modifications and deletions to and from the data graph.
Example 10. One company, two departments, two employees - Retrieve and delete
The company is retrieved as a complete data graph containing five data objects - the company, two departments and two employees. They are all deleted by deleting the company object. Deleting an object from the graph deletes all the object beneath it in the graph. Five SQL DELETE statements will be generated and executed. As always they will be qualified with a WHERE clause that contains all of the fields that were retrieved, so that any updates to the data in the database in the meantime by another process will be detected.
You may be interested in seeing the SQL statements that are generated in order to apply changes back to the database. At the top of the SDO/DAS/Relational.php you will find a number of constants which control whether the process of constructing and executing the SQL statements is to be traced. Try setting DEBUG_EXECUTE_PLAN to TRUE to see the generated SQL statements.
The Relational DAS provides two classes: the Relational DAS itself and the subclass of Exception that can be thrown. The Relational DAS has four publicly useful calls: the constructor, the createRootDataObject() call to obtain the root object of an empty data graph, the executeQuery() call to obtain a data graph containing data from a relational database, and the applyChanges() call to write changes made to a data graph back to the relational database.
The only object other than an SDO_DAS_Relational_Exception with which the application is expected to interact.
__construct - construct the Relational DAS with a model derived from the passed metadata
createRootDataObject - obtain an otherwise empty data graph containing just the special root object
executeQuery - execute an SQL query passed as a literal string and return the results as a normalised data graph
executePreparedQuery - execute an SQL query passed as a prepared statement, with a list of values to substitute for placeholders, and return the results as a normalised data graph
applyChanges - examine the change summary in the data graph and apply those changes back to the database, subject to an assumption of optimistic concurrency
Is a subclass of PHP's Exception. It adds no behaviour to Exception. Thrown, with useful descriptive text, to signal errors in the metadata or unexpected failures to perform SQL operations.