Incorporate referential integrity and data synchronization checks into Deequ's VerificationSuite

[rdsharma26]

The following two utilities should be part of Deequ's verification suite.

• https://github.com/awslabs/deequ/blob/master/src/main/scala/com/amazon/deequ/comparison/DataSynchronization.scala
• https://github.com/awslabs/deequ/blob/master/src/main/scala/com/amazon/deequ/comparison/ReferentialIntegrity.scala

[VenkataKarthikP]

@rdsharma26 could you add more details on what enhancement we are looking at here, I could take a stab at implementation. thanks

[mentekid]

Hi,

Here's some context for this issue. Deequ lets you run checks on your data by constructing a VerificationSuite object. You can build a VerificationSuite by calling onData(), which then exposes the addCheck() method. You can call that repeatedly to add multiple checks to your suite, for example:

val verificationResult = VerificationSuite()
  .onData(data)
  .addCheck(Check(CheckLevel.Error, "must have 5 rows").hasSize(_ == 5)
  .addCheck(Check(CheckLevel.Error, "must have no nulls").isComplete("id")
  .run()

All checks within the same Verification suite are processed before Spark is called, and Deequ comes up with a plan to calculate all the necessary statistics without making unnecessary passes over the data.

However, for the comparison operations @rdsharma26 is mentioning above, there is no Check object and therefore they cannot be added to a VerificationSuite. This means two things:

1. There is no way for Deequ to optimize the execution of these comparisons
2. The code to run any of these checks looks incongruent with any other Deequ syntax, as you need to directly invoke methods rather than constructing a cohesive suite of tests.

The ask here is to merge the two operations with the standard Deequ APIs, so a user can create a verification suite that contains a mix of cross-dataset and in-dataset tests. This will probably require a bit of refactoring in the VerificationRunBuilder, because unlike any Checks we have today, the cross-dataset checks require an additional reference dataset in addition to the primary dataset (passed using onData(), which returns a VerificationRunBuilder).

Let me know if that's not clear or you have any follow-up questions.

[VenkataKarthikP]

@mentekid thanks, I can take a stab at this, will circle back once PR is ready.

[chaurasiya]

Hello @VenkataKarthikP and @rdsharma26 is there any update on implementation of ReferentialIntegrity check as well?

[VenkataKarthikP]

@chaurasiya I have plans to do it, will open a PR.

[chaurasiya]

Hello @VenkataKarthikP and @rdsharma26, is it still in plan to include Referential Integrity check?