From 704d49389b00a4890bc404cd836a0fdd6f7dd440 Mon Sep 17 00:00:00 2001
From: =?UTF-8?q?Florian=20M=C3=BCller?= <florian@tomueller.de>
Date: Tue, 28 Jul 2026 21:52:42 +0200
Subject: [PATCH 1/3] fix(spark): propagate exit code of the entrypoint

---
 spark-k8s/stackable/run-spark.sh | 17 +++++++++++------
 1 file changed, 11 insertions(+), 6 deletions(-)

diff --git a/spark-k8s/stackable/run-spark.sh b/spark-k8s/stackable/run-spark.sh
index 8871d3f36..b066b3953 100755
--- a/spark-k8s/stackable/run-spark.sh
+++ b/spark-k8s/stackable/run-spark.sh
@@ -12,13 +12,18 @@ _handle_term() {
 }
 trap _handle_term TERM INT
 
-# `wait` returns immediately when the trap fires; loop until the child is actually gone.
-wait "$child_pid"
-while kill -0 "$child_pid" 2>/dev/null; do
-  wait "$child_pid" 2>/dev/null || true
+# Wait for the entrypoint and propagate its exit status.
+# A trapped signal makes `wait` return immediately with a status > 128 while the child keeps
+# running, so in that case wait again rather than taking the interrupted status.
+while true; do
+  wait "$child_pid"
+  result=$?
+  if [ "$result" -gt 128 ] && kill -0 "$child_pid" 2>/dev/null; then
+    continue
+  fi
+  break
 done
-result=$?
 
 eval "$_STACKABLE_POST_HOOK"
 
-exit $result
+exit "$result"

From a918d4e80121d8509d2d1ba3042d1d7882769f61 Mon Sep 17 00:00:00 2001
From: =?UTF-8?q?Florian=20M=C3=BCller?= <florian@tomueller.de>
Date: Wed, 29 Jul 2026 17:00:13 +0200
Subject: [PATCH 2/3] chore: add changelog entry

---
 CHANGELOG.md | 5 +++++
 1 file changed, 5 insertions(+)

diff --git a/CHANGELOG.md b/CHANGELOG.md
index 961f1b29c..b55d65177 100644
--- a/CHANGELOG.md
+++ b/CHANGELOG.md
@@ -4,11 +4,16 @@ All notable changes to this project will be documented in this file.
 
 ## [Unreleased]
 
+### Fixed
+
+- spark: Propagate the entrypoint's exit code so failed applications are no longer reported as successful ([#1595]).
+
 ### Removed
 
 - omid: remove 1.1.2 ([#1593]).
 
 [#1593]: https://github.com/stackabletech/docker-images/pull/1593
+[#1595]: https://github.com/stackabletech/docker-images/pull/1595
 
 ## [26.7.0] - 2026-07-21
 

From af40e4037b6f1e195046fac32436b5b5156a1d7e Mon Sep 17 00:00:00 2001
From: =?UTF-8?q?Florian=20M=C3=BCller?= <florian@tomueller.de>
Date: Thu, 30 Jul 2026 15:52:21 +0200
Subject: [PATCH 3/3] fix(spark): rewrite run-spark.sh to fit into existing
 script pattern

---
 spark-k8s/stackable/run-spark.sh | 59 ++++++++++++++++++++++----------
 1 file changed, 41 insertions(+), 18 deletions(-)

diff --git a/spark-k8s/stackable/run-spark.sh b/spark-k8s/stackable/run-spark.sh
index b066b3953..e63db49db 100755
--- a/spark-k8s/stackable/run-spark.sh
+++ b/spark-k8s/stackable/run-spark.sh
@@ -1,29 +1,52 @@
 #!/bin/bash
 
-eval "$_STACKABLE_PRE_HOOK"
-
-# Forward SIGTERM to the Spark entrypoint to support spark JVM's gracefully shutdown.
-/stackable/spark/kubernetes/dockerfiles/spark/entrypoint.sh "$@" &
-child_pid=$!
+prepare_signal_handlers() {
+  unset term_child_pid
+  unset term_kill_needed
+  trap handle_term_signal TERM INT
+}
 
+# Only ever invoked via the trap installed above.
 # shellcheck disable=SC2329
-_handle_term() {
-  kill -TERM "$child_pid" 2>/dev/null || true
+handle_term_signal() {
+  if [[ -v term_child_pid ]]; then
+    kill -TERM "${term_child_pid}" 2>/dev/null || true
+  else
+    term_kill_needed="yes"
+  fi
 }
-trap _handle_term TERM INT
 
-# Wait for the entrypoint and propagate its exit status.
+# Waits for the given child and returns *its* exit status.
 # A trapped signal makes `wait` return immediately with a status > 128 while the child keeps
-# running, so in that case wait again rather than taking the interrupted status.
-while true; do
-  wait "$child_pid"
-  result=$?
-  if [ "$result" -gt 128 ] && kill -0 "$child_pid" 2>/dev/null; then
-    continue
+# running its shutdown sequence, so in that case wait again instead of taking the interrupted
+# status.
+wait_for_termination() {
+  term_child_pid=$1
+  if [[ -v term_kill_needed ]]; then
+    kill -TERM "${term_child_pid}" 2>/dev/null || true
   fi
-  break
-done
+  while true; do
+    wait "${term_child_pid}"
+    term_child_status=$?
+    if [[ "${term_child_status}" -gt 128 ]] && kill -0 "${term_child_pid}" 2>/dev/null; then
+      continue
+    fi
+    return "${term_child_status}"
+  done
+}
+
+eval "$_STACKABLE_PRE_HOOK"
+
+# The signal handlers are installed before the child is started, so that a signal arriving in
+# between is not lost. SIGTERM is forwarded to the Spark entrypoint to let the Spark JVM shut
+# down gracefully.
+prepare_signal_handlers
+
+/stackable/spark/kubernetes/dockerfiles/spark/entrypoint.sh "$@" &
+
+result=0
+wait_for_termination $! || result=$?
 
 eval "$_STACKABLE_POST_HOOK"
 
-exit "$result"
+exit "${result}"