Featured Post

Applying Email Validation to a JavaFX TextField Using Binding

This example uses the same controller as in a previous post but adds a use case to support email validation.  A Commons Validator object is ...

Monday, May 14, 2012

Efficient Lookups with Talend Open Studio's Hash Components

If you are using the same data source in several Talend Open Studio subjobs, consider loading the data data into an internal data structure for use throughout a job.

A HashMap is a Java data structure maintained in RAM.  Talend Open Studio lets you load manipulate data in HashMaps using the tHashInput and tHashOutput components.  Because the data exists in RAM, lookups are faster, especially if secondary storage is based on spinning hard disks.

Without a HashMap

This Talend Open Studio job loads data from two spreadsheets, EmployeeHires and EmployeeTerminations, into a target table, EmployeeActions.  The spreadsheet sources contain a data (hireDate and terminationDate) that is used as a key into a table called BusinessDates.  Although the date could simply be carried over into the target table (without the lookup), many data warehouses maintain date information in a separate table.  This is because calendar-related business information is merged with the timestamp.

This data includes flags for Holiday, Payday, and Weekday that supplement that timestamp (Month, Day, Year, Quarter) fields.  The spreadsheet has been loaded into the MS SQL Server table "BusinessDates".

Basic Date Fields PLUS Business-specific Information
A typical Talend Open Studio job will use BusinessDates as a lookup table.  The main flow comes from two sources: an Employee Hires spreadsheet, and an Employee Terminations spreadsheet.

BusinessDates is Repeated for Each Subjob
The Employee Hires spreadsheet is similar to the Terminations spreadsheets with hireDate replaced with terminationDate.

Employee Action Source (Hires)

Hash Alternative

The preceding job works.  5 Hire and 3 Termination records are written to the database.  However, the job has a drawback.  Even if caching is enabled, the lookup is read at least once for each subjob, leading to poorer performance.  An alternative is to use the Talend Hash components: tHashInput and tHashOutput.

Job Rewritten with Talend Hash Component
This version creates a tMap with the tHashOutput_1 component which is loaded by a database input, BusinessDates.  I flag three columns as keys: year, month, day.  This is for informational purposes; any fields can be used in the later tMap joins.

The tHashOutput component is configured as follows.

tHashOutput Configuration
The schema -- with the 3 informational keys -- used in the tHashOutput follows.


The tHashOutput schema can now be applied to joins (tMap) by adding tHashInput components as lookup flows.  This is the configuration of tHashInput_1 which is identical to tHashInput_2.  More inputs can be added for other data loading subjobs.

tHashInput Configuration Links to tHashOutput
 In the UI, you must define a schema for both the tHashInput and tHashOutput components.  I do this by setting the value to the Repository, then changing the value to Built-in and re-defining the keys from "id" to "year/month/day".

The join is performed as an Inner Join on three columns.  Because the dates are represented differently, I use a TalendDate routine.  Note the important +1 which dealing with the zero-based Java month and day values.

Multi-Part Join on a Hash Map Input
The job loads 8 records in the database.  The Hires are flagged with "HI" and the Terminations with "TE".

Data Loading Results
RAM-based data structures can provide a performance improvement since slower spinning disks aren't involved in data reads.  This is a good pattern when you're dealing with data from different sources (different dbs, spreadsheets, etc).  If your data processing is table-to-table in the same database, huge performance improvements can be made with the ELT* components that keep all processing within the database, eliminating the network latency of pulling the data into Talend Open Studio's JVM.



6 comments:

  1. This blog is a greta resource. Quick question, is it more efficient to use the hash tool or the tReplicate for a task like this?

    ReplyDelete
  2. You can implement this job with tReplicate by routing the main flow from the Business Dates tMSSqlInput component to both of the tMaps. Functionally, you'll get the same results, but won't need the tHashInput initialization subjob.

    I believe that the hash version is more efficient because it creates fewer Java objects during the main loops (from t*Input on). tHashInput doesn't add code to any of the main loops (there is no main .javajet in Component Designer terms). There is a "new" operator in tReplicate_main.javajet that I think will make a difference.

    I use tReplicate when I'm producing exactly the same output in a different format. For example, you can use tReplicate to produce two text files with different character sets, UTF-8 and ISO8859_1 (for legacy systems).

    This isn't a cue to rip out all tReplicates, especially absent any performance problems. But if you have a job with a lot of replicated connections, large memory use, or performance that's degrading with increased input, give the hash components a try.

    ReplyDelete
  3. Carl, what I need to do to clear the thashinput / thashoutput from the memory after using it, the reason I am asking is I am calling multiple jobs in a single job (main job) and I use thashinput/output in all those jobs, while executing the main job I am getting some wierd errors, (if I am execuing the jobs individually with out calling the main job it works fine). Please let me know what to do toclear it from the memory and will try it and see. thank you.

    ReplyDelete
    Replies
    1. What are the weird errors?

      Delete
    2. Take a look at the Talend JIRA too. If you're on 4.2.3, this may apply: TDI-17859.

      Delete
  4. Hi Carl, I saw your post 'Manipulating a tHashOutput in Talend Open Studio", thank you. Since I am using 4.2.3 some of the properties of thash's were not there, I guess you should be using a different version, however I am checking TDI-17859 and see if it is having any solution.

    ReplyDelete