feat(build): add Spark v3.4.1 (#40)

* update to 3.4.1

.github/workflows/ci.yml

@@ -111,6 +111,30 @@ jobs:
           scala:        "2.13"
           with_hive:    "true"
           with_pyspark: "true"
+        - spark:        "3.4.1"
+          java:         "8"
+          hadoop:       "3.3.4"
+          scala:        "2.12"
+          with_hive:    "true"
+          with_pyspark: "true"
+        - spark:        "3.4.1"
+          java:         "8"
+          hadoop:       "3.3.4"
+          scala:        "2.13"
+          with_hive:    "true"
+          with_pyspark: "true"
+        - spark:        "3.4.1"
+          java:         "11"
+          hadoop:       "3.3.4"
+          scala:        "2.12"
+          with_hive:    "true"
+          with_pyspark: "true"
+        - spark:        "3.4.1"
+          java:         "11"
+          hadoop:       "3.3.4"
+          scala:        "2.13"
+          with_hive:    "true"
+          with_pyspark: "true"
     runs-on: ubuntu-20.04
     env:
       IMAGE_NAME: "spark-k8s"

CHANGELOG.md

@@ -3,8 +3,9 @@
 ## v3
 
 - (Temporarily drop support for R due to keyserver issues)
-- Only supports for for 3.1.3, 3.2.2, 3.3.0 (dropped 2.4.8).
+- Only supports for for 3.1.3, 3.2.2, 3.3.0, 3.4.1 (dropped 2.4.8).
 - Supports both Java 8 and 11 for Spark 3 builds.
+- Add Ubuntu-based image since the migration to eclipse-temurin for jre image source.
 
 ## v2
 

README.md

@@ -12,16 +12,22 @@ Debian:
 - `3.3.0`
 - `3.2.2`
 - `3.1.3`
+- `3.4.1`
 
 ## Note
 
 (R builds are temporarily suspended due to keyserver issues at current time.)
 
-All the build images here are Debian based as the official Spark repository now
-uses `openjdk:<java>-jdk-slim-buster` as the base image for Kubernetes build.
-Because currently the official Dockerfiles do not pin the Debian distribution,
-they are incorrectly using the latest Debian `bullseye`, which does not have
-support for Python 2, and its Python 3.9 do not work well with PySpark.
+Build image for Spark 3.4.1 is Ubuntu based because openjdk is deprecated and
+going forward the official Spark repository uses `eclipse-temurin:<java>-jre`
+where slim variants of jre images are not available at the moment.
+
+All the build images with Spark before v3.4.0 are Debian based as the official 
+Spark repository now uses `openjdk:<java>-jre-slim-buster` as the base image 
+for Kubernetes build. Because currently the official Dockerfiles do not pin 
+the Debian distribution, they are incorrectly using the latest Debian `bullseye`,
+which does not have support for Python 2, and its Python 3.9 do not work well 
+with PySpark.
 
 Hence some Dockerfile overrides are in-place to make sure that Spark 2 builds
 can still work.

make-distribution.sh

@@ -67,6 +67,15 @@ else
     DOCKERFILE_PY="./resource-managers/kubernetes/docker/src/main/dockerfiles/spark/bindings/python/Dockerfile"
 fi
 
+if [[ ${SPARK_MAJOR_VERSION} -eq 3 && ${SPARK_MINOR_VERSION} -ge 4 ]]; then  # >=3.4
+    # From Spark v3.4.0 onwards, openjdk is not the prefered base image source as it i
+    # deprecated and taken over by eclipse-temurin. slim-buster variants are not available
+    # on eclipse-temurin at the moment.
+    IMAGE_VARIANT="jre"
+else
+    IMAGE_VARIANT="jre-slim-buster"
+fi
+
 # Temporarily remove R build due to keyserver issue
 # DOCKERFILE_R="./resource-managers/kubernetes/docker/src/main/dockerfiles/R/Dockerfile"
 
@@ -83,7 +92,7 @@ TAG_NAME="${SELF_VERSION}_${SPARK_LABEL}_hadoop-${HADOOP_VERSION}_scala-${SCALA_
 #     build
 
 ./bin/docker-image-tool.sh \
-    -b java_image_tag=${JAVA_VERSION}-jre-slim-buster \
+    -b java_image_tag=${JAVA_VERSION}-${IMAGE_VARIANT} \
     -r "${IMAGE_NAME}" \
     -t "${TAG_NAME}" \
     -f "${DOCKERFILE_BASE}" \

templates/vars.yml

@@ -15,3 +15,8 @@ versions:
   java:   ['8', '11']
   hadoop: ['3.3.2']
   scala:  ['2.12', '2.13']
+
+- spark:  ['3.4.1']
+  java:   ['8', '11']
+  hadoop: ['3.3.4']
+  scala:  ['2.12', '2.13']