Kettle JDBC driver
The thin Kettle JDBC driver allows a Java client to query a Kettle transformation remotely using JDBC and SQL.
Available from Pentaho Data Integration Enterprise Edition 5.0 or higher.
As with most JDBC drivers, there is a server and a client component to the JDBC driver.
The server is designed to run as a Servlet on the Carte server or the Pentaho Data Integration server.
Where are the data service tables coming from?
You can configure a transformation step to serve as a data service in the "Data Service" tab of the transformation settings dialog:
When such a transformation gets saved there is a new entry created in either the local metastore (under the .pentaho/metastore folder) or in the Enterprise repository on the DI server (/etc/metastore).
As such, Carte or the DI Server will automatically and immediately pick up newly configured data services from the moment your service transformation is saved.
The carte configuration file accepts a <repository> which will also be scanned in addition to the local metastore.
Reminder: the default user/password is stored in the pwd/kettle.pwd file and is cluster/cluster.
During execution of a query, 2 transformations will be executed on the server:
- A service transformation, of human design built in Spoon to provide the service data
- An automatically generated transformation to aggregate, sort and filter the data according to the SQL query
These 2 transformations will be visible on Carte or in Spoon in the slave server monitor and can be tracked, sniff tested, paused and stopped just like any other transformation. However, it will not be possible to restart them manually since both transformations are programatically linked.
For this example we open the "Getting Started Transformation" (see the sample/transformations folder of your PDI distribution) and configure a Data Service for the "Number Range" called "gst". (comparable to the screenshot above)
Then we can launch Carte or the Data Integration Server to execute a query against that new virtual database table:
This query is being parsed by the server and a transformation is being generated to convert the service transformation data into the requested format:
The data which is being injected is originating from the service transformation:
So for each executed query you will see 2 transformations listed on the server.
The JDBC Client
The JDBC driver uses the following class:
The URL is in the following format:
For Carte, this is an example:
For the Data Integration server:
this example is for a the carte configuration file shown above.
The following standard options are available:
- webappname : the name of the web app (typically pentaho-di on the DI server) Make sure to specify this in the "Options" sections of the Kettle database connection dialog if you want to connect with PDI to a PDI Server (Carte or the DI server);
- proxyhostname : the proxy server for the HTTP connection(s)
- proxyport : the port of the proxy server
- nonproxyhosts : the hosts (comma seperator) for which not to use a proxy
- debugtrans : the optional name of a file in which the generated transformation will be stored for debugging purposes (example: /tmp/debug.ktr)
- debuglog : set to "true" to have the logging text of the remote SQL transformation will be written to the general logging channel once execution is finished.
Parameters for the service transformation can be set with the following format: PARAMETER_name=value (so with the option name prepended with "PARAMETER_")
Support for the SQL is minimal at the moment.
The following things are supported, please consider everything else unsupported:
- * is expanded to include all rows
- COUNT(DISTINCT field)
- DISTINCT <fields>
- IIF( condition, true-value or field, false-value or field)
- CASE WHEN condition THEN true-value ELSE false-value END
- Aggregates: SUM, AVG, MIN, MAX
- Alias both with the "AS" keyword and with one or more spaces seperated, for example SUM(sales) AS "Total Sales" or SUM(sales) TotalSales
- Constant expressions are possible, see below in the literals section.
- Calculations on the other hand are not possible yet, perform them in the service transformation for now.
- Strictly one service name, aliasing is possible
- You can omit the service name to query from an empty row or you can query from dual, for example "SELECT 1" or "SELECT 1 FROM dual" are the same.
- You can specify a schema (default is Kettle) but it is currently ignored. It will be translated to a namespace in the near future.
- nested brackets
- AND, OR, NOT if preceded by brackets, for example: NOT ( A = 5 OR C = 3 )
- precedence taken into account
- Literals (String, Integer)
- PARAMETER('parameter-name')='value' (always evaluates to TRUE in the condition)
- <=, =<
- >=, =>
- LIKE (standard % and ? wildcards are converted to .* and . regular expressions)
- REGEX (matches regular expression)
- IS NULL
- IS NOT NULL
- IN ( value, value, value, ... )
- You can put a condition on the IIF expression or it's alias if one is used. (please use identical string literals for expressions)
- GROUP BY
- Group on fields, not IIF() function
- Conditions should be placed on the aggregate construct, not the alias
- Please use identical strings for the expressions, the algorithm is not yet that smart. In other words, if you use "COUNT( * )" in the SELECT clause you should use the same "COUNT( * ) " expression in the HAVING clause, not "COUNT(*)" or any variant of it.
- You can place having conditions on aggregations that do not appear in the SELECT clause.
- ORDER BY
- You can order on any column in the result or not in the result
- You can order on IFF or CASE-WHEN expressions.
- Strings have single quotes around them, escaping quotes is not yet supported.
Dates have square brackets around them and the following formats are supported: [yyyy/MM/dd HH:mm:ss.SSS], [yyyy/MM/dd HH:mm:ss] and [yyyy/MM/dd]
- Number and BigNumber should have no grouping symbol and the decimal is . (example 123.45)
- Integers contain only digits
- Boolean values can be TRUE or FALSE
Besides the obviously plentiful limitations in the support for the SQL standard, there are a few noteworthy things to note:
- Grouping is done using a "Memory Group By" step which keeps all the groups in memory. If you expect large amounts of groups to be used, watch your memory consumption on the server. We're using the "memory group by" step to avoid doing a costly sort on the data.
- It is not possible to specify the same field twice in the same SELECT clause. (for whatever reason you might want to do that)
- calculations and functions like string concatenation and so on is not (yet) supported.
Clients typically need to following libraries to work:
- commons HTTP client
- commons code
- commons lang
- commons logging
- commons VFS (1.0)
These libraries can be found in your data-integration/lib folder with the appropriate version number (e.g. kettle-core-5.0.0.jar).
Since SQuirrel already contains most needed jar files, configuring it simply done by adding kettle-core.jar as a new driver jar file along with Apache Commons VFS 1.0 and scannotation.jar
The following jar files need to be added:
- commons HTTP client
- commons code
- commons lang
- commons logging
- commons VFS (1.0)
Pentaho Report Designer
Simply replace the kettle-*.jar files in the lib/ folder with new files from Kettle v5.0-M1 or higher.
Pentaho Schema Workbench
Replace the current kettle-*.jar files with the ones from Kettle v5 or later.
Pentaho Interactive Reporting
Interactive reporting runs off Pentaho Metadata so this advice also works there.
You need a BI Server that uses the PDI 5.0 jar files or you can use an older version and update the kettle-core, kettle-db and kettle-engine jar files in the /tomcat/webapps/pentaho/WEB-INF/lib/ folder
Pentaho Analyses (Mondrian): Analyzer / Saiku / JPivot
See Pentaho Interactive reporting: simply update the kettle-*.jar files in your Pentaho BI Server (tested with 4.1.0 EE and 4.5.0 EE) to get it to work.
Example of patching :
Fun fact: Mondrian generates the following SQL for the report shown above:
You can query a remote service transformation with any Kettle v5 or higher client. You can query the service through the database explorer and the various database steps (for example the Table Input step).
TODO: ask project owners to change the current old driver class to the new thin one.
Jaspersoft iReport Designer
Partial success as I'm getting some XML parsing errors. However, adding the aforementioned jar files at least allow you to get back query fields:
To be investigated.
The following things are next on the agenda:
- Caching of services data in memory with validity time-out
- Caching of queries in memory with validity time-out
- writing to a service transformation (INSERT INTO)