Automatically install latest spark version (#2075)

* Automatically install latest pyspark version

* Better text

* Do not use shutil to keep behaviour

* Make setup_script cwd independent

* Use _get_program_version to calculate spark version

* Update setup_spark.py reqs

* Update setup_spark.py

* Add info about HADOOP_VERSION

* Add customization back

* Better text

* Specify build args when they are actually needed

* Better text

* Better code

* Better code

* Better text

* Get rid of warning

* Improve code

* Remove information about checksum

* Better text

.pre-commit-config.yaml

@@ -41,10 +41,12 @@ repos:
         args: [--config, ./mypy.ini]
         additional_dependencies:
           [
+            "beautifulsoup4",
             "numpy",
             "pytest",
             "requests",
             "urllib3",
+            "types-beautifulsoup4",
             "types-requests",
             "types-tabulate",
             "types-urllib3",

docs/using/specifics.md

@@ -42,18 +42,20 @@ ipython profile create
 You can build a `pyspark-notebook` image with a different `Spark` version by overriding the default value of the following arguments at build time.
 `all-spark-notebook` is inherited from `pyspark-notebook`, so you have to first build `pyspark-notebook` and then `all-spark-notebook` to get the same version in `all-spark-notebook`.
 
-- Spark distribution is defined by the combination of Spark, Hadoop, and Scala versions and verified by the package checksum,
+- Spark distribution is defined by the combination of Spark, Hadoop, and Scala versions,
   see [Download Apache Spark](https://spark.apache.org/downloads.html) and the [archive repo](https://archive.apache.org/dist/spark/) for more information.
 
-  - `spark_version`: The Spark version to install (`3.3.0`).
-  - `hadoop_version`: The Hadoop version (`3.2`).
-  - `scala_version`: The Scala version (`2.13`, optional).
-  - `spark_checksum`: The package checksum (`BFE4540...`).
-  - `openjdk_version`: The version of the OpenJDK (JRE headless) distribution (`17`).
+  - `openjdk_version`: The version of the OpenJDK (JRE headless) distribution (`17` by default).
     - This version needs to match the version supported by the Spark distribution used above.
     - See [Spark Overview](https://spark.apache.org/docs/latest/#downloading) and [Ubuntu packages](https://packages.ubuntu.com/search?keywords=openjdk).
-
-- Starting with _Spark >= 3.2_, the distribution file might contain the Scala version.
+  - `spark_version` (optional): The Spark version to install, for example `3.5.0`.
+    If not specified (this is the default), latest stable Spark will be installed.
+  - `hadoop_version`: The Hadoop version (`3` by default).
+    Note, that _Spark < 3.3_ require to specify `major.minor` Hadoop version (i.e. `3.2`).
+  - `scala_version` (optional): The Scala version, for example `2.13` (not specified by default).
+    Starting with _Spark >= 3.2_, the distribution file might contain the Scala version.
+  - `spark_download_url`: URL to use for Spark downloads.
+    You may need to use <https://archive.apache.org/dist/spark/> url if you want to download old Spark versions.
 
 For example, here is how to build a `pyspark-notebook` image with Spark `3.2.0`, Hadoop `3.2`, and OpenJDK `11`.
 
@@ -65,14 +67,14 @@ This recipe is not tested and might be broken.
 # From the root of the project
 # Build the image with different arguments
 docker build --rm --force-rm \
-    -t jupyter/pyspark-notebook:spark-3.2.0 ./images/pyspark-notebook \
+    -t my-pyspark-notebook ./images/pyspark-notebook \
+    --build-arg openjdk_version=11 \
     --build-arg spark_version=3.2.0 \
     --build-arg hadoop_version=3.2 \
-    --build-arg spark_checksum=707DDE035926A50B75E53FCA72CADA519F3239B14A96546911CB4916A58DCF69A1D2BFDD2C7DD5899324DBD82B6EEAB9797A7B4ABF86736FFCA4C26D0E0BF0EE \
-    --build-arg openjdk_version=11
+    --build-arg spark_download_url="https://archive.apache.org/dist/spark/"
 
 # Check the newly built image
-docker run -it --rm quay.io/jupyter/pyspark-notebook:spark-3.2.0 pyspark --version
+docker run -it --rm my-pyspark-notebook pyspark --version
 
 # Welcome to
 #       ____              __
@@ -81,7 +83,12 @@ docker run -it --rm quay.io/jupyter/pyspark-notebook:spark-3.2.0 pyspark --versi
 #    /___/ .__/\_,_/_/ /_/\_\   version 3.2.0
 #       /_/
 
-# Using Scala version 2.13.5, OpenJDK 64-Bit Server VM, 11.0.15
+# Using Scala version 2.12.15, OpenJDK 64-Bit Server VM, 11.0.21
+# Branch HEAD
+# Compiled by user ubuntu on 2021-10-06T12:46:30Z
+# Revision 5d45a415f3a29898d92380380cfd82bfc7f579ea
+# Url https://github.com/apache/spark
+# Type --help for more information.
 ```
 
 ### Usage Examples

images/pyspark-notebook/Dockerfile

@@ -16,49 +16,38 @@ USER root
 # Spark dependencies
 # Default values can be overridden at build time
 # (ARGS are in lowercase to distinguish them from ENV)
-ARG spark_version="3.5.0"
-ARG hadoop_version="3"
-ARG scala_version
-ARG spark_checksum="8883c67e0a138069e597f3e7d4edbbd5c3a565d50b28644aad02856a1ec1da7cb92b8f80454ca427118f69459ea326eaa073cf7b1a860c3b796f4b07c2101319"
 ARG openjdk_version="17"
 
-ENV APACHE_SPARK_VERSION="${spark_version}" \
-    HADOOP_VERSION="${hadoop_version}"
-
 RUN apt-get update --yes && \
     apt-get install --yes --no-install-recommends \
     "openjdk-${openjdk_version}-jre-headless" \
     ca-certificates-java && \
     apt-get clean && rm -rf /var/lib/apt/lists/*
 
-# Spark installation
-WORKDIR /tmp
-
-# You need to use https://archive.apache.org/dist/ website if you want to download old Spark versions
+# If spark_version is not set, latest stable Spark will be installed
+ARG spark_version
+ARG hadoop_version="3"
+# If scala_version is not set, Spark without Scala will be installed
+ARG scala_version
+# URL to use for Spark downloads
+# You need to use https://archive.apache.org/dist/spark/ website if you want to download old Spark versions
 # But it seems to be slower, that's why we use the recommended site for download
-RUN if [ -z "${scala_version}" ]; then \
-    curl --progress-bar --location --output "spark.tgz" \
-        "https://dlcdn.apache.org/spark/spark-${APACHE_SPARK_VERSION}/spark-${APACHE_SPARK_VERSION}-bin-hadoop${HADOOP_VERSION}.tgz"; \
-  else \
-    curl --progress-bar --location --output "spark.tgz" \
-        "https://dlcdn.apache.org/spark/spark-${APACHE_SPARK_VERSION}/spark-${APACHE_SPARK_VERSION}-bin-hadoop${HADOOP_VERSION}-scala${scala_version}.tgz"; \
-  fi && \
-  echo "${spark_checksum} *spark.tgz" | sha512sum -c - && \
-  tar xzf "spark.tgz" -C /usr/local --owner root --group root --no-same-owner && \
-  rm "spark.tgz"
+ARG spark_download_url="https://dlcdn.apache.org/spark/"
 
 # Configure Spark
-ENV SPARK_HOME=/usr/local/spark
-ENV SPARK_OPTS="--driver-java-options=-Xms1024M --driver-java-options=-Xmx4096M --driver-java-options=-Dlog4j.logLevel=info" \
-    PATH="${PATH}:${SPARK_HOME}/bin"
+ENV SPARK_VERSION="${spark_version}" \
+    HADOOP_VERSION="${hadoop_version}" \
+    SCALA_VERSION="${scala_version}" \
+    SPARK_DOWNLOAD_URL="${spark_download_url}"
 
-RUN if [ -z "${scala_version}" ]; then \
-    ln -s "spark-${APACHE_SPARK_VERSION}-bin-hadoop${HADOOP_VERSION}" "${SPARK_HOME}"; \
-  else \
-    ln -s "spark-${APACHE_SPARK_VERSION}-bin-hadoop${HADOOP_VERSION}-scala${scala_version}" "${SPARK_HOME}"; \
-  fi && \
-  # Add a link in the before_notebook hook in order to source automatically PYTHONPATH && \
-  ln -s "${SPARK_HOME}/sbin/spark-config.sh" /usr/local/bin/before-notebook.d/10spark-config.sh
+ENV SPARK_HOME=/usr/local/spark
+ENV PATH="${PATH}:${SPARK_HOME}/bin"
+ENV SPARK_OPTS="--driver-java-options=-Xms1024M --driver-java-options=-Xmx4096M --driver-java-options=-Dlog4j.logLevel=info"
+
+COPY setup_spark.py /opt/setup-scripts/
+
+# Setup Spark
+RUN /opt/setup-scripts/setup_spark.py
 
 # Configure IPython system-wide
 COPY ipython_kernel_config.py "/etc/ipython/"

images/pyspark-notebook/setup_spark.py

@@ -0,0 +1,107 @@
+#!/usr/bin/env python3
+# Copyright (c) Jupyter Development Team.
+# Distributed under the terms of the Modified BSD License.
+
+# Requirements:
+# - Run as the root user
+# - Required env variables: SPARK_HOME, HADOOP_VERSION, SPARK_DOWNLOAD_URL
+# - Optional env variables: SPARK_VERSION, SCALA_VERSION
+
+import os
+import subprocess
+from pathlib import Path
+
+import requests
+from bs4 import BeautifulSoup
+
+
+def get_all_refs(url: str) -> list[str]:
+    """
+    Get all the references for a given webpage
+    """
+    resp = requests.get(url)
+    soup = BeautifulSoup(resp.text, "html.parser")
+    return [a["href"] for a in soup.find_all("a", href=True)]
+
+
+def get_spark_version() -> str:
+    """
+    If ${SPARK_VERSION} env variable is non-empty, simply returns it
+    Otherwise, returns the last stable version of Spark using spark archive
+    """
+    if (version := os.environ["SPARK_VERSION"]) != "":
+        return version
+    all_refs = get_all_refs("https://archive.apache.org/dist/spark/")
+    stable_versions = [
+        ref.removeprefix("spark-").removesuffix("/")
+        for ref in all_refs
+        if ref.startswith("spark-") and "incubating" not in ref and "preview" not in ref
+    ]
+    # Compare versions semantically
+    return max(
+        stable_versions, key=lambda ver: [int(sub_ver) for sub_ver in ver.split(".")]
+    )
+
+
+def download_spark(
+    spark_version: str,
+    hadoop_version: str,
+    scala_version: str,
+    spark_download_url: Path,
+) -> str:
+    """
+    Downloads and unpacks spark
+    The resulting spark directory name is returned
+    """
+    spark_dir_name = f"spark-{spark_version}-bin-hadoop{hadoop_version}"
+    if scala_version:
+        spark_dir_name += f"-scala{scala_version}"
+    spark_url = spark_download_url / f"spark-{spark_version}" / f"{spark_dir_name}.tgz"
+
+    tmp_file = Path("/tmp/spark.tar.gz")
+    subprocess.check_call(
+        ["curl", "--progress-bar", "--location", "--output", tmp_file, spark_url]
+    )
+    subprocess.check_call(
+        [
+            "tar",
+            "xzf",
+            tmp_file,
+            "-C",
+            "/usr/local",
+            "--owner",
+            "root",
+            "--group",
+            "root",
+            "--no-same-owner",
+        ]
+    )
+    tmp_file.unlink()
+    return spark_dir_name
+
+
+def prepare_spark(spark_dir_name: str, spark_home: Path) -> None:
+    """
+    Creates a ${SPARK_HOME} symlink to a versioned spark directory
+    Creates a 10spark-config.sh symlink to source PYTHONPATH automatically
+    """
+    subprocess.check_call(["ln", "-s", f"/usr/local/{spark_dir_name}", spark_home])
+
+    # Add a link in the before_notebook hook in order to source PYTHONPATH automatically
+    CONFIG_SCRIPT = "/usr/local/bin/before-notebook.d/10spark-config.sh"
+    subprocess.check_call(
+        ["ln", "-s", spark_home / "sbin/spark-config.sh", CONFIG_SCRIPT]
+    )
+
+
+if __name__ == "__main__":
+    spark_version = get_spark_version()
+    spark_dir_name = download_spark(
+        spark_version=spark_version,
+        hadoop_version=os.environ["HADOOP_VERSION"],
+        scala_version=os.environ["SCALA_VERSION"],
+        spark_download_url=Path(os.environ["SPARK_DOWNLOAD_URL"]),
+    )
+    prepare_spark(
+        spark_dir_name=spark_dir_name, spark_home=Path(os.environ["SPARK_HOME"])
+    )

tagging/taggers.py

@@ -128,7 +128,12 @@ class JuliaVersionTagger(TaggerInterface):
 class SparkVersionTagger(TaggerInterface):
     @staticmethod
     def tag_value(container: Container) -> str:
-        return "spark-" + _get_env_variable(container, "APACHE_SPARK_VERSION")
+        SPARK_VERSION_LINE_PREFIX = r"   /___/ .__/\_,_/_/ /_/\_\   version"
+
+        spark_version = _get_program_version(container, "spark-submit")
+        version_line = spark_version.split("\n")[4]
+        assert version_line.startswith(SPARK_VERSION_LINE_PREFIX)
+        return "spark-" + version_line.split(" ")[-1]
 
 
 class HadoopVersionTagger(TaggerInterface):