fix(cleanup): exit 0 and always clear busy under the runner's bash -e

cleanup.sh

@@ -7,43 +7,47 @@
 # to cost 21-39 s per job here). JOB_STARTED (pre-job.sh) sweeps up after a
 # job whose runner was killed before this hook ran.
 #
-# Intentionally not using set -e: individual cleanup failures should not
-# prevent other cleanups from running.
+# The runner runs this hook under `bash -e`, whatever the shebang says, and a
+# non-zero exit fails the job ("Complete runner") even when every step passed.
+# A missing directory is normal here (a runner that never ran Gradle has no
+# ~/.gradle/caches), so every command ends in `|| true`, `busy` is cleared
+# first, and the hook always exits 0.
 #
 # KEEP: the slot caches (see preflight-entrypoint.sh) and the work dir, which
 #       lives in the slot (RUNNER_DATA_DIR/slots/<id>/work), not under /tmp.
 # DELETE: /tmp contents, Gradle daemon/journals/locks, runner _diag logs.
 
 # Empty dir $1 (/tmp), keeping the dir itself.
 clean_tmp() {
-  find "$1" -mindepth 1 -maxdepth 1 -exec rm -rf {} + 2>/dev/null
+  find "$1" -mindepth 1 -maxdepth 1 -exec rm -rf {} + 2>/dev/null || true
 }
 
 cleanup() {
   local before after
   echo "[cleanup] Running post-job cleanup..."
-  before=$(df -h / | awk 'NR==2 {print $4}')
+  # The job is over: the pruner may trim this slot again. First, so nothing
+  # below can leave the marker behind (it would make the next job look like a
+  # crash recovery).
+  if [ -n "${RUNNER_SLOT_DIR:-}" ]; then rm -f "$RUNNER_SLOT_DIR/busy" || true; fi
+  before=$(df -h / | awk 'NR==2 {print $4}') || true
 
   clean_tmp /tmp
 
   # Gradle daemon / journals / locks only. Keep modules-2, build-cache-*,
   # transforms-*. $HOME covers both the root (default) and RUN_AS_ROOT=false
   # (/home/runner) runner users.
-  rm -rf "$HOME"/.gradle/daemon/ "$HOME"/.gradle/caches/journal-* 2>/dev/null
+  rm -rf "$HOME"/.gradle/daemon/ "$HOME"/.gradle/caches/journal-* 2>/dev/null || true
   # Shallow on purpose: Gradle's lock files sit at most three levels under
   # caches/, and a find over the whole (multi-GB) tree is what made this slow.
-  find "$HOME"/.gradle/caches -maxdepth 3 -name '*.lock' -delete 2>/dev/null
+  find "$HOME"/.gradle/caches -maxdepth 3 -name '*.lock' -delete 2>/dev/null || true
 
   # Runner diagnostic logs
-  find /actions-runner/_diag -name '*.log' -delete 2>/dev/null
+  find /actions-runner/_diag -name '*.log' -delete 2>/dev/null || true
 
-  # The job is over: the pruner may trim this slot again.
-  if [ -n "${RUNNER_SLOT_DIR:-}" ]; then rm -f "$RUNNER_SLOT_DIR/busy"; fi
-
-  after=$(df -h / | awk 'NR==2 {print $4}')
+  after=$(df -h / | awk 'NR==2 {print $4}') || true
   echo "[cleanup] Post-job cleanup complete. Disk free: ${before} -> ${after}"
 }
 
 # Sourced (tests/): define the functions only.
 [ "${BASH_SOURCE[0]}" = "$0" ] || return 0
-cleanup
+cleanup || true

tests/cleanup.sh

@@ -0,0 +1,44 @@
+#!/bin/bash
+# The JOB_COMPLETED hook runs under `bash -e`, as the runner invokes it, and must
+# exit 0 and clear the slot's `busy` marker whatever the home looks like: no
+# Gradle cache at all (control runners, a runner that never ran Gradle), or a
+# Gradle cache with daemon state to drop. A non-zero exit fails the job in
+# "Complete runner"; a marker left behind makes every later job look like a
+# crash recovery. /tmp is not wiped here: clean_tmp is stubbed out.
+#   bash tests/cleanup.sh
+here="$(cd "$(dirname "${BASH_SOURCE[0]}")" && pwd)"
+# shellcheck source=tests/harness.sh
+source "$here/harness.sh"
+
+# Run cleanup.sh's main path under bash -e, with clean_tmp stubbed so the real
+# /tmp is left alone.
+hook() {
+  local home="$1" slot="$2"
+  HOME="$home" RUNNER_SLOT_DIR="$slot" bash -e -c '
+    source "$1"
+    clean_tmp() { :; }
+    cleanup
+  ' _ "$root/cleanup.sh" >/dev/null 2>&1
+}
+
+slot="$work/slot"
+mkdir -p "$slot"
+
+home="$work/empty"
+mkdir -p "$home"
+touch "$slot/busy"
+hook "$home" "$slot" || fail "exit $? with no gradle cache"
+gone "$slot/busy"
+
+home="$work/gradle"
+mkdir -p "$home/.gradle/daemon/8.10" "$home/.gradle/caches/journal-1" "$home/.gradle/caches/modules-2/files-2.1"
+touch "$home/.gradle/caches/modules-2/modules-2.lock" "$home/.gradle/caches/modules-2/files-2.1/keep.jar"
+touch "$slot/busy"
+hook "$home" "$slot" || fail "exit $? with a gradle cache"
+gone "$slot/busy"
+gone "$home/.gradle/daemon"
+gone "$home/.gradle/caches/journal-1"
+gone "$home/.gradle/caches/modules-2/modules-2.lock"
+kept "$home/.gradle/caches/modules-2/files-2.1/keep.jar"
+
+finish "cleanup exits 0 under bash -e and always clears busy"