diff --git a/bin/dataproc.sh b/bin/dataproc.sh
new file mode 100755
index 0000000..dc7afaf
--- /dev/null
+++ b/bin/dataproc.sh
@@ -0,0 +1,96 @@
+#!/bin/bash
+
+# A testing script for verifying the spark-bigquery connector with the existing
+# mozaggregator code. This requires `gcloud` to be configured to point at a
+# sandbox project for reading data from `payload_bytes_decoded`.
+
+set -e
+
+REGION=us-west1
+MODULE="python_mozaggregator"
+NUM_WORKERS=${NUM_WORKERS:-1}
+
+
+function bootstrap() {
+    local bucket=$1
+
+    # create the package artifacts
+    rm -rf dist build
+    python setup.py bdist_egg
+    gsutil cp "dist/${MODULE}*.egg" "gs://${bucket}/bootstrap/${MODULE}.egg"
+
+    # create the initialization script and runner
+    mkdir -p bootstrap
+    cd bootstrap
+    echo "apt install --yes python-dev" > install-python-dev.sh
+    tee mozaggregator-runner.py >/dev/null << EOF
+# This runner has been auto-generated from mozilla/python_mozaggregator/bin/dataproc.sh.
+# Any changes made to the runner file will be over-written on subsequent runs.
+from mozaggregator import cli
+
+try:
+    cli.entry_point(auto_envvar_prefix="MOZETL")
+except SystemExit:
+    # avoid calling sys.exit() in databricks
+    # http://click.palletsprojects.com/en/7.x/api/?highlight=auto_envvar_prefix#click.BaseCommand.main
+    pass
+EOF
+    cd ..
+    gsutil cp bootstrap/* "gs://${bucket}/bootstrap/"
+}
+
+
+function delete_cluster() {
+    local cluster_id=$1
+    gcloud dataproc clusters delete ${cluster_id} --region=${REGION}
+}
+
+
+function create_cluster() {
+    local cluster_id=$1
+    local bucket=$2
+    requirements=$(tr "\n" " " < requirements/build.txt)
+
+    function cleanup {
+        delete_cluster ${cluster_id}
+    }
+    trap cleanup EXIT
+
+    gcloud dataproc clusters create ${cluster_id} \
+        --image-version 1.3 \
+        --num-preemptible-workers ${NUM_WORKERS} \
+        --properties ^#^spark:spark.jars=gs://spark-lib/bigquery/spark-bigquery-latest.jar#spark:spark.hadoop.fs.s3a.access.key=${AWS_ACCESS_KEY_ID}#spark:spark.hadoop.fs.s3a.secret.key=${AWS_SECRET_ACCESS_KEY} \
+        --metadata "PIP_PACKAGES=${requirements}" \
+        --initialization-actions \
+            gs://${bucket}/bootstrap/install-python-dev.sh,gs://dataproc-initialization-actions/python/pip-install.sh \
+        --region=${REGION}
+}
+
+
+function submit() {
+    cluster_id=$1
+    bucket=$2
+    # pass the rest of the parameters from the main function
+    shift 2
+    gcloud dataproc jobs submit pyspark \
+        gs://${bucket}/bootstrap/mozaggregator-runner.py \
+        --cluster ${cluster_id} \
+        --region ${REGION} \
+        --py-files=gs://${bucket}/bootstrap/${MODULE}.egg \
+        -- "$@"
+}
+
+
+function main() {
+    cd "$(dirname "$0")/.."
+    bucket=$(gcloud config get-value project)
+    cluster_id=test-mozaggregator
+    bootstrap $bucket
+    create_cluster $cluster_id $bucket
+    submit $cluster_id $bucket "$@"
+}
+
+
+if [[ "${BASH_SOURCE[0]}" == "${0}" ]]; then
+    main "$@"
+fi