sparkmonitor 3.2.0

Jupyter Notebook & Lab extension to monitor Apache Spark jobs from a notebook

SparkMonitor

SparkMonitor is a Jupyter extension for monitoring Apache Spark jobs launched from notebooks. It displays live Spark metrics directly in the notebook interface, making it easier to understand, debug, and profile Spark workloads as they run.

It supports JupyterLab and classic Jupyter Notebook with PySpark 3.x and 4.x.

About

+

=

SparkMonitor adds an interactive monitoring panel below notebook cells that trigger Spark jobs, so you can inspect execution progress without leaving the notebook.

---

Requirements

• Python 3.x
• PySpark 3.x or 4.x
• JupyterLab 4.x or classic Jupyter Notebook 4.4.0 or later
• Spark Classic API mode
  • SparkMonitor works with the traditional Spark driver model used by PySpark.
  • It is not compatible with Spark Connect.

Features

• Live monitoring of Spark jobs launched from a notebook cell
• Job and stage table with progress bars and execution details
• Timeline view showing jobs, stages, and tasks over time
• Resource graphs for active tasks and executor core usage

Quick Start

Installation

Create and activate a virtual environment:

python -m venv venv
source venv/bin/activate

Install SparkMonitor together with PySpark and a notebook frontend.

For JupyterLab

pip install sparkmonitor pyspark jupyterlab

Enable the SparkMonitor IPython kernel extension:

ipython profile create
echo "c.InteractiveShellApp.extensions.append('sparkmonitor.kernelextension')" >> "$(ipython profile locate default)/ipython_kernel_config.py"

This only needs to be done once per IPython profile.

Using SparkMonitor in a Notebook

To use SparkMonitor, create your Spark session with the SparkMonitor listener enabled.

This requires two Spark configurations:

Configuration	Purpose
spark.extraListeners	Registers the SparkMonitor listener that collects Spark job metrics
spark.driver.extraClassPath	Points to the SparkMonitor listener JAR bundled with the sparkmonitor package

Example with a manually specified listener JAR path

If you already know the exact path to the matching SparkMonitor listener JAR in your current environment, you can set spark.driver.extraClassPath directly:

from pyspark.sql import SparkSession

spark = (
    SparkSession.builder.config(
        "spark.extraListeners",
        "sparkmonitor.listener.JupyterSparkMonitorListener",
    )
    .config(
        "spark.driver.extraClassPath",
        # Put the path to the matching SparkMonitor listener JAR here.
        "venv/lib/python3.13/site-packages/sparkmonitor/listener_spark4_2.13.jar",
    )
    .getOrCreate()
)

Example with automatic listener JAR detection

The most robust approach is to resolve the listener JAR path dynamically from the installed Python package instead of hardcoding the full environment path. The example below first checks SPARK_HOME and then falls back to the pyspark package layout used by pip install pyspark, where SPARK_HOME is often not set:

import os
from pathlib import Path

import pyspark
import sparkmonitor
from pyspark.sql import SparkSession


def iter_spark_jar_dirs() -> list[Path]:
    candidates = []

    spark_home = os.environ.get("SPARK_HOME")
    if spark_home:
        candidates.append(Path(spark_home) / "jars")

    candidates.append(Path(pyspark.__file__).resolve().parent / "jars")
    return [path for path in candidates if path.exists()]


def resolve_listener_jar(sparkmonitor_dir: Path) -> Path:
    for jars_dir in iter_spark_jar_dirs():
        for jar in jars_dir.glob("spark-core_*.jar"):
            # spark-core_2.13-3.5.8.jar => scala=2.13, spark_major=3
            scala_ver, spark_ver = jar.name.split("_")[1].split("-")[:2]
            spark_major = spark_ver.split(".")[0]
            if spark_major == "3" and scala_ver == "2.12":
                return sparkmonitor_dir / "listener_spark3_2.12.jar"
            if spark_major == "3" and scala_ver == "2.13":
                return sparkmonitor_dir / "listener_spark3_2.13.jar"
            if spark_major == "4" and scala_ver == "2.13":
                return sparkmonitor_dir / "listener_spark4_2.13.jar"

    raise RuntimeError(
        "Could not detect Spark/Scala version from SPARK_HOME or the pyspark installation"
    )


sparkmonitor_dir = Path(sparkmonitor.__file__).resolve().parent
listener_jar = resolve_listener_jar(sparkmonitor_dir)

spark = (
    SparkSession.builder.config(
        "spark.extraListeners",
        "sparkmonitor.listener.JupyterSparkMonitorListener",
    )
    .config("spark.driver.extraClassPath", str(listener_jar))
    .getOrCreate()
)

Important

The correct listener JAR depends on:

• the location of your Python environment
• your Spark major version and Scala version
• how Spark is installed (SPARK_HOME vs pip install pyspark)

You can inspect the installed package location with:

import sparkmonitor

print(sparkmonitor.__path__)

Then locate the corresponding listener JAR in that package directory:

• listener_spark3_2.12.jar for Spark 3 + Scala 2.12
• listener_spark3_2.13.jar for Spark 3 + Scala 2.13
• listener_spark4_2.13.jar for Spark 4 + Scala 2.13

If needed, you can also build the listener JAR yourself with sbt, as described in the development section below.

Development

To work on SparkMonitor locally:

# Install the package in editable mode
pip install -e .

# Build the frontend (see package.json for available scripts)
yarn run build:<action>

# Link the JupyterLab extension into your local Jupyter environment
jupyter labextension develop --overwrite .

# Watch frontend files for changes
yarn run watch

# Build the Spark listener JARs
cd scalalistener_spark3 # Spark 3 / Scala 2.12 and 2.13
sbt +package

cd ../scalalistener_spark4 # Spark 4 / Scala 2.13
sbt package

Troubleshooting

SparkMonitor panel does not appear

Check the following:

• sparkmonitor is installed in the same Python environment as your notebook kernel
• pyspark is installed
• jupyterlab or notebook is installed
• the IPython kernel extension is enabled
• your Spark session includes spark.extraListeners
• spark.driver.extraClassPath points to a valid listener JAR
• you are using Spark Classic, not Spark Connect

Wrong listener JAR selected

The listener JAR must match both your Spark major version and Scala version:

Spark 3 + Scala 2.12 -> listener_spark3_2.12.jar
Spark 3 + Scala 2.13 -> listener_spark3_2.13.jar
Spark 4 + Scala 2.13 -> listener_spark4_2.13.jar

Using the wrong listener JAR may prevent the listener from loading correctly.

Hardcoded virtual environment path does not work

Avoid hardcoding paths when possible. Environment-specific paths vary across systems, Python versions, and virtual environments. The dynamic path resolution example above is usually more portable.

SPARK_HOME is not set

This is expected in some setups, especially when Spark comes from pip install pyspark. In that case, use the pyspark package location to find the bundled Spark JARs instead of assuming SPARK_HOME/jars exists.

Project History

• The first version of SparkMonitor was written by krishnan-r as a Google Summer of Code project with the SWAN Notebook Service team at CERN.
• Further fixes and improvements were made by the CERN team and community contributors in swan-cern/jupyter-extensions/tree/master/SparkMonitor.
• Jafer Haider updated the extension for JupyterLab during an internship at Yelp.
• Work from the jupyterlab-sparkmonitor fork was later merged into this repository so that both JupyterLab and Jupyter Notebook are supported from a single package.
• Ongoing maintenance and development continue through the SWAN team at CERN and the community.

References

• PyPI package: sparkmonitor
• Releases: swan-cern/sparkmonitor releases