Hello Spark Team and users. I'm currently exploring custom spark image building. I've found two places containing dockerfiles from the "original" apache project. The first one is well known github mirror https://github.com/apache/spark/tree/v3.5.1/resource-managers/kubernetes/docker/src/main/dockerfiles/spark The second one is related to SPARK-40513: https://github.com/apache/spark-docker Also documentation at https://spark.apache.org/docs/latest/running-on-kubernetes.html#docker-images notes dockerfile in distribution tar (which originates from the first repo) as good starting point for customization. However, starting with version 3.3.3 and SPARK-42505 the second repo's entrypoint script and Dockerfile diverge significantly from the main repo.
E.g. "pass through" command is no longer executed under tini supervision. Could you please guide me with the following questions: 1) Is the difference in dockerfile and entrypoint intended? Or just not synced yet? 2) Which one is better to use as a starting point for a custom image? 3) Why is tini not used as a supervisor for "pass through" command mode in the second repo? Really can't find any info behind this decision. Sincerely, Andrei Lopukhov.
