fix: Amend relation to python dictionaries approach as using polars (via arrow) is leading to dates to be transformed to datetimes

stevenhsd · stevenhsd · commit 5b1aff997c79 · 2026-01-12T22:37:04.000Z
diff --git a/docs/json_schemas/contract/components/field_error_type.schema.json b/docs/json_schemas/contract/components/field_error_type.schema.json
@@ -0,0 +1,21 @@
+{
+    "$schema": "https://json-schema.org/draft-07/schema",
+    "$id": "data-ingest:contract/components/field_error_type.schema.json",
+    "title": "field_error_detail",
+    "description": "The error type for a field when a validation error is raised during the data contract phase",
+    "type": "object",
+    "properties": {
+        "error_type": {
+            "description": "The type of error the details are for",
+            "type": "string",
+            "enum": [
+                "Blank",
+                "Bad value",
+                "Wrong format"
+            ],
+            "additionalProperties": {
+                "$ref": "field_error_detail.schema.json"
+            }
+        }
+    }
+}
diff --git a/pyproject.toml b/pyproject.toml
@@ -52,6 +52,7 @@ include-groups = [
 [tool.poetry.group.dev.dependencies]
 commitizen = "4.9.1"
 pre-commit = "4.3.0"
+ipykernel = "^7.1.0"
 
 [tool.poetry.group.test]
 optional = true
diff --git a/src/dve/core_engine/backends/implementations/duckdb/duckdb_helpers.py b/src/dve/core_engine/backends/implementations/duckdb/duckdb_helpers.py
@@ -273,3 +273,16 @@ def get_all_registered_udfs(connection: DuckDBPyConnection) -> set[str]:
     """
     connection.sql("CREATE TEMP TABLE IF NOT EXISTS dve_udfs (function_name VARCHAR)")
     return {rw[0] for rw in connection.sql("SELECT * FROM dve_udfs").fetchall()}
+
+
+def duckdb_rel_to_dictionaries(
+    entity: DuckDBPyRelation, batch_size=1000
+) -> Iterator[list[dict[str, Any]]]:
+    """Iterator converting DuckDBPyRelation to lists of dictionaries.
+    Avoids issues where dates are getting converted to datetimes using polars as intermediate."""
+    # TODO - look into float conversion - floats that can't be stored exactly in binary
+    # TODO - are given to nearest approximation. Tried Decimal, causes issues in arrays
+    # TODO - with templating (as in complex fields, repr used when str called in jinja templating).
+    cols: tuple[str] = tuple(entity.columns) # type: ignore
+    while rows := entity.fetchmany(batch_size):
+        yield [dict(zip(cols, rw)) for rw in rows]
diff --git a/src/dve/core_engine/backends/implementations/duckdb/rules.py b/src/dve/core_engine/backends/implementations/duckdb/rules.py
@@ -23,6 +23,7 @@
 from dve.core_engine.backends.implementations.duckdb.duckdb_helpers import (
     DDBStruct,
     duckdb_read_parquet,
+    duckdb_rel_to_dictionaries,
     duckdb_write_parquet,
     get_all_registered_udfs,
     get_duckdb_type_from_annotation,
@@ -511,23 +512,25 @@ def notify(self, entities: DuckDBEntities, *, config: Notification) -> Messages:
         if config.excluded_columns:
             matched = matched.select(StarExpression(exclude=config.excluded_columns))
 
-        for record in matched.df().to_dict(orient="records"):
-            # NOTE: only templates using values directly accessible in record - nothing nested
-            # more complex extraction done in reporting module
-            messages.append(
-                FeedbackMessage(
-                    entity=config.reporting.reporting_entity_override or config.entity_name,
-                    original_entity=config.entity_name,
-                    record=record,  # type: ignore
-                    error_location=config.reporting.legacy_location,
-                    error_message=template_object(config.reporting.message, record),  # type: ignore
-                    failure_type=config.reporting.legacy_error_type,
-                    error_type=config.reporting.legacy_error_type,
-                    error_code=config.reporting.code,
-                    reporting_field=config.reporting.legacy_reporting_field,
-                    reporting_field_name=config.reporting.reporting_field_override,
-                    is_informational=config.reporting.emit in ("warning", "info"),
-                    category=config.reporting.category,
+        for chunk in duckdb_rel_to_dictionaries(matched):
+            for record in chunk:
+                # NOTE: only templates using values directly accessible in record - nothing nested
+                # more complex extraction done in reporting module
+                messages.append(
+                    FeedbackMessage(
+                        entity=config.reporting.reporting_entity_override or config.entity_name,
+                        original_entity=config.entity_name,
+                        record=record,  # type: ignore
+                        error_location=config.reporting.legacy_location,
+                        error_message=template_object(config.reporting.message,
+                                                      record), # type: ignore
+                        failure_type=config.reporting.legacy_error_type,
+                        error_type=config.reporting.legacy_error_type,
+                        error_code=config.reporting.code,
+                        reporting_field=config.reporting.legacy_reporting_field,
+                        reporting_field_name=config.reporting.reporting_field_override,
+                        is_informational=config.reporting.emit in ("warning", "info"),
+                        category=config.reporting.category,
+                    )
                 )
-            )
         return messages
diff --git a/tests/features/movies.feature b/tests/features/movies.feature
@@ -31,7 +31,7 @@ Feature: Pipeline tests using the movies dataset
         Then The rules restrict "movies" to 4 qualifying records
         And there are errors with the following details and associated error_count from the business_rules phase
             | ErrorCode       | ErrorMessage                                           | error_count |
-            | LIMITED_RATINGS | Movie has too few ratings ([6.1])                      | 1           |
+            | LIMITED_RATINGS | Movie has too few ratings ([6.5])                      | 1           |
             | RUBBISH_SEQUEL  | The movie The Greatest Movie Ever has a rubbish sequel | 1           |
         And the latest audit record for the submission is marked with processing status error_report
         When I run the error report phase
@@ -67,7 +67,7 @@ Feature: Pipeline tests using the movies dataset
         Then The rules restrict "movies" to 4 qualifying records
         And there are errors with the following details and associated error_count from the business_rules phase
             | ErrorCode       | ErrorMessage                                           | error_count |
-            | LIMITED_RATINGS | Movie has too few ratings ([6.1])                      | 1           |
+            | LIMITED_RATINGS | Movie has too few ratings ([6.5])                      | 1           |
             | RUBBISH_SEQUEL  | The movie The Greatest Movie Ever has a rubbish sequel | 1           |
         And the latest audit record for the submission is marked with processing status error_report
         When I run the error report phase
diff --git a/tests/test_core_engine/test_backends/test_implementations/test_duckdb/test_duckdb_helpers.py b/tests/test_core_engine/test_backends/test_implementations/test_duckdb/test_duckdb_helpers.py
@@ -1,20 +1,26 @@
 """Test Duck DB helpers"""
+
+import datetime
 import tempfile
 from pathlib import Path
+from typing import Any
 
 import pytest
 import pyspark.sql.types as pst
 from duckdb import DuckDBPyRelation, DuckDBPyConnection
 from pyspark.sql import Row, SparkSession
 
-from dve.core_engine.backends.implementations.duckdb.duckdb_helpers import _ddb_read_parquet
+from dve.core_engine.backends.implementations.duckdb.duckdb_helpers import (
+    _ddb_read_parquet,
+    duckdb_rel_to_dictionaries)
 
 
 class TempConnection:
     """
     Full object would be a DataContract object but this simplified down to meet min requirements
     of the test.
     """
+
     def __init__(self, connection: DuckDBPyConnection) -> None:
         self._connection = connection
 
@@ -25,7 +31,7 @@ def __init__(self, connection: DuckDBPyConnection) -> None:
         ("movie_ratings"),
         ("movie_ratings/"),
         ("file://movie_ratings/"),
-    ]
+    ],
 )
 def test__ddb_read_parquet_with_hive_format(
     spark: SparkSession, temp_ddb_conn: DuckDBPyConnection, outpath: str
@@ -38,11 +44,13 @@ def test__ddb_read_parquet_with_hive_format(
                 Row(movie_name="Hot Fuzz", avg_user_rating=7.7, avg_critic_rating=6.5),
                 Row(movie_name="Nemo", avg_user_rating=8.8, avg_critic_rating=7.6),
             ],
-            pst.StructType([
-                pst.StructField("movie_name", pst.StringType()),
-                pst.StructField("avg_user_rating", pst.FloatType()),
-                pst.StructField("avg_critic_rating", pst.FloatType()),
-            ])
+            pst.StructType(
+                [
+                    pst.StructField("movie_name", pst.StringType()),
+                    pst.StructField("avg_user_rating", pst.FloatType()),
+                    pst.StructField("avg_critic_rating", pst.FloatType()),
+                ]
+            ),
         )
         out_path = str(Path(temp_dir_path, outpath))
         test_data_df.coalesce(1).write.parquet(out_path)
@@ -51,3 +59,39 @@ def test__ddb_read_parquet_with_hive_format(
 
         assert isinstance(ddby_relation, DuckDBPyRelation)
         assert ddby_relation.count("*").fetchone()[0] == 2  # type: ignore
+
+
+@pytest.mark.parametrize(
+    "data",
+    (
+    
+        [
+            {
+                "str_field": "hi",
+                "int_field": 5,
+                "array_float_field": [6.5, 7.25],
+                "date_field": datetime.date(2021, 5, 3),
+                "timestamp_field": datetime.datetime(2022, 6, 7, 1, 2, 3),
+            },
+            {
+                "str_field": "bye",
+                "int_field": 3,
+                "array_float_field": None,
+                "date_field": datetime.date(2021, 8, 11),
+                "timestamp_field": datetime.datetime(2022, 4, 3, 1, 2, 3),
+            },
+        ],
+    
+    ),
+)
+def test_duckdb_rel_to_dictionaries(temp_ddb_conn: DuckDBPyConnection,
+                                    data: list[dict[str, Any]]):
+    _, con = temp_ddb_conn
+    test_rel = con.query("select dta.* from (select unnest($data) as dta)",
+                                                  params={"data": data})
+    res: list = []
+    for chunk in duckdb_rel_to_dictionaries(test_rel):
+        res.extend(chunk)
+    
+    assert res == data
+        
diff --git a/tests/testdata/movies/movies.dischema.json b/tests/testdata/movies/movies.dischema.json
@@ -21,7 +21,7 @@
                     },
                     "duration_minutes": "int",
                     "ratings": {
-                        "type": "NonNegativeFloat",
+                        "type": "float",
                         "is_array": true
                     },
                     "cast": {
diff --git a/tests/testdata/movies/movies.json b/tests/testdata/movies/movies.json
@@ -36,7 +36,7 @@
     "year": 2020,
     "genre": ["Fantasy", "Family"],
     "duration_minutes": 110,
-    "ratings": [6.1],
+    "ratings": [6.5],
     "cast": [
       { "name": "R. Williams", "role": "Cat", "date_joined": "2016-05-06" },
       { "name": "T. Brown", "role": "Dog", "date_joined": "2016-05-07" }
