Skip to content

Delete Generated Data

Info

Delete generated data is a paid feature. Try the free trial here.

Creating a data generator for Postgres and delete the generated data after using it.

Requirements

  • 5 minutes
  • Git
  • Gradle
  • Docker

Get Started

First, we will clone the data-caterer-example repo which will already have the base project setup required.

git clone git@github.com:pflooky/data-caterer-example.git

Plan Setup

Create a new Java or Scala class.

  • Java: src/main/java/com/github/pflooky/plan/MyAdvancedDeleteJavaPlanRun.java
  • Scala: src/main/scala/com/github/pflooky/plan/MyAdvancedDeletePlanRun.scala

Make sure your class extends PlanRun.

import com.github.pflooky.datacaterer.java.api.PlanRun;
...

public class MyAdvancedDeleteJavaPlanRun extends PlanRun {
    {
        var autoRun = configuration()
                .postgres("my_postgres", "jdbc:postgresql://host.docker.internal:5432/customer")  (1)
                .enableGeneratePlanAndTasks(true)                                                 (2)
                .enableRecordTracking(true)                                                       (3)
                .enableDeleteGeneratedRecords(false)                                              (4)
                .enableUniqueCheck(true)
                .generatedPlanAndTaskFolderPath("/opt/app/data/generated")                        (5)
                .recordTrackingFolderPath("/opt/app/data/recordTracking")                         (6)
                .generatedReportsFolderPath("/opt/app/data/report");

        execute(autoRun);
   }
}
import com.github.pflooky.datacaterer.api.PlanRun
...

class MyAdvancedDeletePlanRun extends PlanRun {

  val autoRun = configuration
    .postgres("my_postgres", "jdbc:postgresql://host.docker.internal:5432/customer")  (1)
    .enableGeneratePlanAndTasks(true)                                                 (2)
    .enableRecordTracking(true)                                                       (3)
    .enableDeleteGeneratedRecords(false)                                              (4)
    .enableUniqueCheck(true)
    .generatedPlanAndTaskFolderPath("/opt/app/data/generated")                        (5)
    .recordTrackingFolderPath("/opt/app/data/recordTracking")                         (6)
    .generatedReportsFolderPath("/opt/app/data/report")

  execute(configuration = autoRun)
}

In the above code we note the following:

  1. We have defined a Postgres connection called my_postgres
  2. enableGeneratePlanAndTasks is enabled to auto generate data for all tables under customer database
  3. enableRecordTracking is enabled to ensure that all generated records are tracked. This will get used when we want to delete data afterwards
  4. enableDeleteGeneratedRecords is disabled for now. We want to see the generated data first and delete sometime after
  5. generatedPlanAndTaskFolderPath is the folder path where we saved the metadata we have gathered from my_postgres
  6. recordTrackingFolderPath is the folder path where record tracking is maintained. We need to persist this data to ensure it is still available when we want to delete data

Postgres Setup

If you don't have your own Postgres up and running, you can set up and run an instance configured in the docker folder via.

cd docker
docker-compose up -d postgres
docker exec docker-postgresserver-1 psql -Upostgres -d customer -c '\dt+ account.*'

This will create the tables found under docker/data/sql/postgres/customer.sql. You can change this file to contain your own tables. We can see there are 4 tables created for us, accounts, balances, transactions and mapping.

Run

Let's try run.

cd ..
./run.sh
#input class MyAdvancedDeleteJavaPlanRun or MyAdvancedDeletePlanRun
#after completing
docker exec docker-postgresserver-1 psql -Upostgres -d customer -c 'select * from account.accounts limit 1'

It should look something like this.

   id   | account_number  | account_status | created_by | created_by_fixed_length | customer_id_int | customer_id_smallint | customer_id_bigint |   customer_id_decimal    | customer_id_real | customer_id_double | open_date  |     open_timestamp      | last_opened_time |                                                           payload_bytes
--------+-----------------+----------------+------------+-------------------------+-----------------+----------------------+--------------------+--------------------------+------------------+--------------------+------------+-------------------------+------------------+------------------------------------------------------------------------------------------------------------------------------------
 100414 | 5uROOVOUyQUbubN | h3H            | SfA0eZJcTm | CuRw                    |              13 |                   42 |               6041 | 76987.745612542900000000 |         91866.78 |  66400.37433202339 | 2023-03-05 | 2023-08-14 11:33:11.343 | 23:58:01.736     | \x604d315d4547616e6a233050415373317274736f5e682d516132524f3d23233c37463463322f342d34376d597e665d6b3d395b4238284028622b7d6d2b4f5042
(1 row)

The data that gets inserted will follow the foreign keys that are defined within Postgres and also ensure the insertion order is correct.

Check the number of records via:

docker exec docker-postgresserver-1 psql -Upostgres -d customer -c 'select count(1) from account.accounts'
#open report under docker/sample/report/index.html

Delete

We are now at a stage where we want to delete the data that was generated. All we need to do is flip two flags.

.enableDeleteGeneratedRecords(true)
.enableGenerateData(false)  //we need to explicitly disable generating data

Enable delete generated records and disable generating data.

Before we run again, let us insert a record manually to see if that data will survive after running the job to delete the generated data.

docker exec docker-postgresserver-1 psql -Upostgres -d customer -c "insert into account.accounts (account_number) values ('my_account_number')"
docker exec docker-postgresserver-1 psql -Upostgres -d customer -c "select count(1) from account.accounts"

We now should have 1001 records in our account.accounts table. Let's delete the generated data now.

./run.sh
#input class MyAdvancedDeleteJavaPlanRun or MyAdvancedDeletePlanRun
#after completing
docker exec docker-postgresserver-1 psql -Upostgres -d customer -c 'select * from account.accounts limit 1'
docker exec docker-postgresserver-1 psql -Upostgres -d customer -c 'select count(1) from account.accounts'

You should see that only 1 record is left, the one that we manually inserted. Great, now we can generate data reliably and also be able to clean it up.

Additional Topics

One class for generating, another for deleting?

Yes, this is possible. There are two requirements: - the connection names used need to be the same across both classes - recordTrackingFolderPath needs to be set to the same value

Define record count

You can control the record count per sub data source via numRecordsPerStep.

var autoRun = configuration()
      ...
      .numRecordsPerStep(100)

execute(autoRun)
val autoRun = configuration
  ...
  .numRecordsPerStep(100)

execute(configuration = autoRun)