unusual behaviour of SQL probes on a partitioned dataset

Tanguy

I've implemented a SQL probe on a partitioned dataset that leverages AWS Athena to query an AWS S3 dataset :

A check has been configured on the SQL result of this probe to ensure the value is equal to 1, essentially validating the absence of duplicate keys in my data.

I asked dataiku to automatically compute the SQL probe and the corresponding check for all the partitions when building the dataset. If a check were to fail, then an error is raised.

Here's the surprising aspect: Dataiku raises an error because it detects SQL probe values greater than 1. However, upon manually recomputing the SQL probe for a partition with a value greater than 1, the calculated value consistently appears to be 1.

It seems as though the SQL probe behaves incorrectly during runtime, causing the checks to erroneously raise errors in my scenario. This issue appeared when we upgraded from DSS v9 to v11.

Is there a potential issue, or am I overlooking something?

---

Operating system used: linux redhat

Turribeach

Is the check throwing a Warning or a Fail?

Tanguy

Indeed:

but the error appears to be a false positive when manually computing the SQL probe (and the corresponding check)!