feat: Adds Trino as a compute engine

Signed-off-by: Marcus Rosti <rostimarcus@gmail.com>
feast-dev · Marcus-Rosti · Sep 29, 2026 · Sep 30, 2026 · 2a0b1c838ca6de96f5146694f6d99df2a3a8dfe7
commit 2a0b1c838ca6de96f5146694f6d99df2a3a8dfe7
@@ -22,6 +22,7 @@ engines.
 |-------------------------|-------------------------------------------------------------------------------------------------|------------|------|
 | LocalComputeEngine      | Runs on Arrow + Pandas/Polars/Dask etc., designed for light weight transformation.              | ✅         |      |
 | SparkComputeEngine      | Runs on Apache Spark, designed for large-scale distributed feature generation.                  | ✅         |      |
+| TrinoComputeEngine      | Runs on Trino, designed for scalable feature generation using Trino SQL.                         | ✅         | [docs](../../reference/compute-engine/trino.md) |
 | SnowflakeComputeEngine  | Runs on Snowflake, designed for scalable feature generation using Snowflake SQL.                | ✅         |      |
 | LambdaComputeEngine     | Runs on AWS Lambda, designed for serverless feature generation.                                 | ✅         |      |
 | FlinkComputeEngine      | Runs on Apache Flink, designed for distributed feature generation through PyFlink Table API.    | ✅         |      |

@@ -0,0 +1,149 @@
+# Trino
+
+## Description
+
+Trino Compute Engine provides a distributed execution engine for batch materialization operations (`materialize` and `materialize-incremental`) and historical retrieval operations (`get_historical_features`).
+
+It is designed to handle large-scale data processing directly on Trino clusters without moving raw data to the client machine.
+
+### Design
+
+The Trino Compute engine is implemented as a subclass of `feast.infra.compute_engines.base.ComputeEngine`.
+
+The engine supports the following features:
+- **Pushdown SQL execution**: Compiles feature pipeline operations (filtering, deduplication, time-windowed aggregations, point-in-time joins, and transformations) into Trino SQL Common Table Expressions (CTEs) executed directly on the Trino cluster.
+- **Streaming online materialization**: Streams query results in chunks as PyArrow record batches directly to the online store via a thread pool, avoiding loading entire datasets into client memory.
+- **Atomic offline table writes**: Materializes offline tables using a temporary staging table and rename swap (`CREATE TABLE ...__staging AS ...; DROP TABLE ...; ALTER TABLE ...__staging RENAME TO ...`).
+- **Configuration inheritance**: Inherits connection details from the Trino offline store when both are configured.
+
+---
+
+## Example
+
+```yaml
+project: feast_trino_project
+registry: data/registry.db
+provider: local
+
+offline_store:
+  type: trino.offline
+  host: localhost
+  port: 8080
+  catalog: iceberg
+  dataset: feast_offline
+  user: feast_user
+
+batch_engine:
+  type: trino.engine
+  batch_size: 10000
+  write_concurrency: 4
+
+online_store:
+  type: redis
+  connection_string: localhost:6379
+```
+
+---
+
+## Example in Python
+
+```python
+from datetime import timedelta
+from feast import (
+    BatchFeatureView,
+    Entity,
+    Field,
+)
+from feast.aggregation import Aggregation
+from feast.infra.offline_stores.contrib.trino_offline_store.trino_source import TrinoSource
+from feast.transformation.mode import TransformationMode
+from feast.transformation.trino_transformation import TrinoTransformation
+from feast.types import Float32, Int32
+from feast.value_type import ValueType
+
+# 1. Define Entity
+driver = Entity(
+    name="driver_id",
+    value_type=ValueType.INT32,
+    join_keys=["driver_id"],
+    description="Driver identifier",
+)
+
+# 2. Define Trino Batch Source
+driver_stats_source = TrinoSource(
+    name="driver_stats_source",
+    table="iceberg.feast_offline.driver_hourly_stats",
+    timestamp_field="event_timestamp",
+    created_timestamp_column="created_timestamp",
+)
+
+# 3. Define Pure Trino SQL Transformation
+# The '{}' placeholder will be substituted with the upstream CTE/table name
+transform = TrinoTransformation(
+    mode=TransformationMode.TRINO_SQL,
+    udf="""
+    SELECT
+        driver_id,
+        event_timestamp,
+        conv_rate * 2.0 AS conv_rate,
+        acc_rate * 2.0 AS acc_rate
+    FROM {}
+    """,
+)
+
+# 4. Define Feature View with SQL Transformation & Aggregations
+driver_hourly_stats_fv = BatchFeatureView(
+    name="driver_hourly_stats",
+    entities=[driver],
+    ttl=timedelta(days=3),
+    feature_transformation=transform,
+    aggregations=[
+        Aggregation(column="conv_rate", function="sum"),
+        Aggregation(column="acc_rate", function="avg"),
+    ],
+    schema=[
+        Field(name="sum_conv_rate", dtype=Float32),
+        Field(name="avg_acc_rate", dtype=Float32),
+        Field(name="driver_id", dtype=Int32),
+    ],
+    online=True,
+    offline=True,
+    source=driver_stats_source,
+)
+```
+
+---
+
+## Retrieving and Materializing Features
+
+```python
+import pandas as pd
+from datetime import datetime, timezone
+from feast import FeatureStore
+
+fs = FeatureStore(repo_path=".")
+
+entity_df: pd.DataFrame
+
+# Lazy Historical Retrieval:
+# Compiles into a single Trino ANSI SQL query with CTEs.
+job = fs.get_historical_features(
+    entity_df=entity_df,
+    features=[
+        "driver_hourly_stats:sum_conv_rate",
+        "driver_hourly_stats:avg_acc_rate",
+    ],
+)
+
+# Inspect the compiled pure SQL query:
+print(job.to_sql())
+
+# Execute on cluster and return pandas dataframe:
+training_df = job.to_df()
+
+# Materialize to Online and Offline Stores:
+fs.materialize(
+    start_date=datetime(2025, 1, 1, tzinfo=timezone.utc),
+    end_date=datetime(2025, 1, 15, tzinfo=timezone.utc),
+)
+```
@@ -7,3 +7,4 @@ class DAGFormat(str, Enum):
     ARROW = "arrow"
     RAY = "ray"
     FLINK = "flink"
+    TRINO = "trino"
@@ -0,0 +1,6 @@
+from feast.infra.compute_engines.trino.compute import (
+    TrinoComputeEngine,
+    TrinoComputeEngineConfig,
+)
+
+__all__ = ["TrinoComputeEngine", "TrinoComputeEngineConfig"]