Use uv instead of pip for python packages

dev/docker-compose.yaml

@@ -188,18 +188,22 @@ services:
 
   # Provision service - creates test data via Spark
   provision:
-    image: python:3.12-slim
+    image: ghcr.io/astral-sh/uv:python3.12-bookworm-slim
     networks:
       iceberg_test:
     depends_on:
       spark-iceberg:
         condition: service_healthy
-    entrypoint: ["/bin/sh", "-c", "pip install -q 'pyspark[connect]==4.0.1' && python3 /opt/spark/provision.py && touch /tmp/provision_complete && tail -f /dev/null"]
+    entrypoint: ["/bin/sh", "-c", "uv run /opt/spark/provision.py && touch /tmp/provision_complete && tail -f /dev/null"]
     volumes:
       - ./spark/provision.py:/opt/spark/provision.py:ro
+      - uv-cache:/root/.cache/uv
     healthcheck:
       test: ["CMD-SHELL", "[ -f /tmp/provision_complete ]"]
       interval: 2s
       timeout: 2s
       retries: 90
       start_period: 20s
+
+volumes:
+  uv-cache:

dev/spark/provision.py

@@ -15,6 +15,13 @@
 # specific language governing permissions and limitations
 # under the License.
 
+# /// script
+# requires-python = ">=3.12"
+# dependencies = [
+#     "pyspark[connect]==4.0.1",
+# ]
+# ///
+
 from pyspark.sql import SparkSession
 from pyspark.sql.functions import current_date, date_add, expr