Prediction Pipeline Errors

[Tom Butkevich]

Hello BQML Team,

I am receiving errors on the BQScriptExecutor step of the prediction pipeline for three models today.  All of the training steps ran successfully and the prediction pipelines ran successfully yesterday--- but today all three prediction pipelines are receiving this error in the logs:

Job: Check New Partition, Worker Class: BQScriptExecutor
Execution failed: WorkerException: Query error: division by zero: 1 / 0 at [106:5]

I am confused since they ran successfully yesterday and models/tables were created in BQ at that time. Would this be a BQ configuration issue? I found no help online for this error.

I am using BQML GA4 v1.2.

Thanks, Tom

[Instant BQML and Vertex Users]

Hi Tom,

Thank you for reaching out. I understand that you're encountering errors with the BQScriptExecutor step in the prediction pipeline for your three models in BQML GA4 v1.2. Specifically, you're seeing the following error:

Job: Check New Partition, Worker Class: BQScriptExecutor
Execution failed: WorkerException: Query error: division by zero: 1 / 0 at [106:5]

I realize this is confusing, especially since everything ran successfully yesterday.

This error is actually expected behavior in version 1.2 for users with Daily Only GA4 BigQuery exports. In this version, we added two new jobs to the pipeline: Check New Partition and Set Latest Partition. These jobs check if there's a new daily GA4 BigQuery export partition available to analyze. If the analysis for that day has already been completed, the Check New Partition job will fail because there's no new data to process.

We acknowledge that this user experience can be confusing. To address this, we made some modifications in version 1.3 of the pipelines to avoid this situation. I recommend upgrading to version 1.3 to benefit from these improvements. You can find the upgrade guide here.

However, please keep in mind that in version 1.3, we've removed the event object from the Measurement Protocol payload entirely to eliminate any influence on GA4 reporting. This means your Measurement Protocol payloads will arrive silently in GA4, and you won't be able to see them in the real-time reports. Rest assured, this doesn't affect functionality—you'll still be applying user properties with propensity scores to your users.

I hope this clarifies things. Please let me know if you have any other questions or need further assistance.

[Tom Butkevich]

Thank you.  The upgrade to version 1.3 solved the issue.