diff --git a/tests/tpch/plot_results.ipynb b/tests/tpch/plot_results.ipynb
new file mode 100644
index 0000000000..aeb6d762aa
--- /dev/null
+++ b/tests/tpch/plot_results.ipynb
@@ -0,0 +1,345 @@
+{
+ "cells": [
+ {
+ "cell_type": "markdown",
+ "id": "0cdaec38-4a9e-4a25-b45e-1188903d219d",
+ "metadata": {},
+ "source": [
+ "# Plot TPC-H results"
+ ]
+ },
+ {
+ "cell_type": "code",
+ "execution_count": 1,
+ "id": "5e877284-731c-4105-bef4-4409b410dc24",
+ "metadata": {},
+ "outputs": [],
+ "source": [
+ "import pandas as pd\n",
+ "from plotting import from_db, latest, normalize, plot"
+ ]
+ },
+ {
+ "cell_type": "markdown",
+ "id": "048cfafb-573f-4ad5-b724-fdff3af83adf",
+ "metadata": {},
+ "source": [
+ "## Load data"
+ ]
+ },
+ {
+ "cell_type": "code",
+ "execution_count": 2,
+ "id": "01d857fc-7956-4592-ba34-6e325cbb217b",
+ "metadata": {},
+ "outputs": [],
+ "source": [
+ "df = from_db(\"/Users/hendrikmakait/Downloads/tpch_1000.db\")"
+ ]
+ },
+ {
+ "cell_type": "code",
+ "execution_count": 3,
+ "id": "5815ac3c-2ad6-4207-a977-31d61039d355",
+ "metadata": {},
+ "outputs": [],
+ "source": [
+ "spark_df = from_db(\"/Users/hendrikmakait/projects/coiled/benchmarks/benchmark.db\")\n",
+ "spark_df = spark_df[spark_df[\"library\"] == \"pyspark\"]"
+ ]
+ },
+ {
+ "cell_type": "code",
+ "execution_count": 4,
+ "id": "1130a571-db77-46e7-990f-b0913a22b5dc",
+ "metadata": {},
+ "outputs": [],
+ "source": [
+ "dask_df = from_db(\"/Users/hendrikmakait/Downloads/tpch_1000.db 3\")\n",
+ "dask_df = dask_df[dask_df[\"library\"] == \"dask\"]"
+ ]
+ },
+ {
+ "cell_type": "code",
+ "execution_count": 5,
+ "id": "c2ad0b96-86f8-40d3-8c83-ca064877c647",
+ "metadata": {},
+ "outputs": [],
+ "source": [
+ "combined_df = pd.concat([df, dask_df, spark_df])"
+ ]
+ },
+ {
+ "cell_type": "code",
+ "execution_count": 6,
+ "id": "36617fe5-c676-474a-b4d3-7eb139157d21",
+ "metadata": {},
+ "outputs": [],
+ "source": [
+ "df = latest(combined_df)"
+ ]
+ },
+ {
+ "cell_type": "markdown",
+ "id": "ba013c77-5173-42c5-881f-89a4455f01ac",
+ "metadata": {},
+ "source": [
+ "## Prepare data"
+ ]
+ },
+ {
+ "cell_type": "code",
+ "execution_count": 7,
+ "id": "677b16fc-aae5-4c21-abfd-aeb5505f403c",
+ "metadata": {},
+ "outputs": [],
+ "source": [
+ "df = normalize(df)"
+ ]
+ },
+ {
+ "cell_type": "markdown",
+ "id": "e6eb21e4-9754-48fd-a6b9-4429f4696928",
+ "metadata": {},
+ "source": [
+ "## Generate plots"
+ ]
+ },
+ {
+ "cell_type": "code",
+ "execution_count": 8,
+ "id": "1c83ccc4-ac11-4edf-bbea-62d613113e61",
+ "metadata": {},
+ "outputs": [
+ {
+ "data": {
+ "text/html": [
+ "\n",
+ "\n",
+ "\n",
+ ""
+ ],
+ "text/plain": [
+ "alt.VConcatChart(...)"
+ ]
+ },
+ "execution_count": 8,
+ "metadata": {},
+ "output_type": "execute_result"
+ }
+ ],
+ "source": [
+ "abs_chart = plot(df, [\"dask\", \"duckdb\", \"pyspark\"])\n",
+ "abs_chart.save(\"tpch-scale-1000-no-polars.svg\")\n",
+ "abs_chart"
+ ]
+ },
+ {
+ "cell_type": "code",
+ "execution_count": 9,
+ "id": "ab5492be-6f97-49cd-b248-c6f5cac71638",
+ "metadata": {},
+ "outputs": [
+ {
+ "data": {
+ "text/html": [
+ "\n",
+ "\n",
+ "\n",
+ ""
+ ],
+ "text/plain": [
+ "alt.VConcatChart(...)"
+ ]
+ },
+ "execution_count": 9,
+ "metadata": {},
+ "output_type": "execute_result"
+ }
+ ],
+ "source": [
+ "rel_chart = plot(df, [\"dask\", \"duckdb\", \"pyspark\"], column=\"relative_duration\")\n",
+ "rel_chart.save(\"tpch-scale-1000-no-polars-relative.svg\")\n",
+ "rel_chart"
+ ]
+ },
+ {
+ "cell_type": "markdown",
+ "id": "611c2644-c7b7-47ac-a2b2-1c47fa22360f",
+ "metadata": {},
+ "source": [
+ "## Compute summary statistics"
+ ]
+ },
+ {
+ "cell_type": "code",
+ "execution_count": 10,
+ "id": "34830787-2364-4541-8cf8-8adffbde9148",
+ "metadata": {},
+ "outputs": [
+ {
+ "data": {
+ "text/plain": [
+ "library\n",
+ "dask 1.000000\n",
+ "duckdb 1.091302\n",
+ "pyspark 1.255598\n",
+ "Name: relative_duration, dtype: float64"
+ ]
+ },
+ "execution_count": 10,
+ "metadata": {},
+ "output_type": "execute_result"
+ }
+ ],
+ "source": [
+ "df.groupby(\"library\").relative_duration.mean()"
+ ]
+ }
+ ],
+ "metadata": {
+ "kernelspec": {
+ "display_name": "Python 3 (ipykernel)",
+ "language": "python",
+ "name": "python3"
+ },
+ "language_info": {
+ "codemirror_mode": {
+ "name": "ipython",
+ "version": 3
+ },
+ "file_extension": ".py",
+ "mimetype": "text/x-python",
+ "name": "python",
+ "nbconvert_exporter": "python",
+ "pygments_lexer": "ipython3",
+ "version": "3.11.8"
+ }
+ },
+ "nbformat": 4,
+ "nbformat_minor": 5
+}
diff --git a/tests/tpch/plotting.py b/tests/tpch/plotting.py
new file mode 100644
index 0000000000..88503e4fde
--- /dev/null
+++ b/tests/tpch/plotting.py
@@ -0,0 +1,78 @@
+import altair as alt
+import pandas as pd
+
+LIBRARY_COLORS = {
+ "dask": "#5677a4",
+ "duckdb": "#e68b39",
+ "polars": "#d4605b",
+ "pyspark": "green",
+}
+
+
+def from_db(path):
+ df = pd.read_sql_table(table_name="test_run", con=f"sqlite:///{path}")
+
+ df = df[
+ (df.call_outcome == "passed")
+ & (df.path.str.contains("^tpch/test_(?:dask|duckdb|polars|pyspark)"))
+ & df.cluster_name
+ ]
+ df = df[["path", "name", "duration", "start", "cluster_name"]]
+ df["library"] = df.path.map(lambda path: path.split("_")[-1].split(".")[0])
+ df["query"] = df.name.map(lambda name: int(name.split("_")[-1]))
+ df["name"] = df.cluster_name.map(lambda name: name.split("-", 3)[-1])
+ df["scale"] = df.cluster_name.map(lambda name: int(name.split("-")[2]))
+ del df["path"]
+ del df["cluster_name"]
+ return df
+
+
+def latest(df, n=1):
+ df = df.sort_values(["query", "library"])
+
+ def recent(df):
+ return df.sort_values("start").tail(n)
+
+ df = df.groupby(["library", "query"]).apply(recent).reset_index(drop=True)
+ del df["start"]
+ return df
+
+
+def normalize(df):
+ dask_durations = df[df["library"] == "dask"].set_index("query")["duration"]
+ data = df.groupby("query").apply(
+ lambda group: group.assign(
+ relative_duration=group["duration"] / dask_durations[group.name]
+ )
+ )
+ return data.reset_index(drop=True)
+
+
+def subplot(df, column, libraries):
+ return (
+ alt.Chart(df)
+ .mark_bar()
+ .encode(
+ x="query:N",
+ y=f"{column}:Q",
+ xOffset="library:N",
+ color=alt.Color("library").scale(
+ domain=libraries,
+ range=[LIBRARY_COLORS[lib] for lib in libraries],
+ ),
+ tooltip=["library", column],
+ )
+ )
+
+
+def plot(df, libraries=None, column="duration"):
+ if libraries is None:
+ libraries = ["dask", "duckdb", "polars", "pyspark"]
+ plot = subplot(df[df["query"] < 12], column=column, libraries=libraries) & subplot(
+ df[df["query"] >= 12], column=column, libraries=libraries
+ )
+ return plot.properties(
+ title=f"TPC-H -- scale:{df.scale.iloc[0]} name:{df.name.iloc[0]}"
+ ).configure_title(
+ fontSize=20,
+ )