From 77ef4d1aa2d877d883b0a4c201af7dd1a3062855 Mon Sep 17 00:00:00 2001
From: gmega <giuliano.mega@gmail.com>
Date: Tue, 17 Jun 2025 20:57:03 -0300
Subject: [PATCH] feat: add halt all processes on single failure

---
 src/process_monitor.bash       | 78 ++++++++++++++++++++++++++++------
 test/test_process_monitor.bats | 56 +++++++++++++++++++-----
 2 files changed, 110 insertions(+), 24 deletions(-)

diff --git a/src/process_monitor.bash b/src/process_monitor.bash
index 989a524..3b3f92c 100644
--- a/src/process_monitor.bash
+++ b/src/process_monitor.bash
@@ -9,6 +9,7 @@ source "${LIB_SRC}/utils.bash"
 
 _procmon_output=$(clh_output_folder "procmon")
 _procmon_pid=""
+_procmon_stop_mode=""
 
 _procmon_init_output() {
   rm -rf "${_procmon_output}" || true
@@ -16,13 +17,14 @@ _procmon_init_output() {
 }
 
 clh_start_process_monitor() {
-  _procmon_init_output
-
   if [ -n "$_procmon_pid" ]; then
     echoerr "[procmon] process monitor already started"
     return 1
   fi
 
+  _procmon_init_output
+  _procmon_stop_mode="$1"
+
   local pid=$$
   _pgid=$(ps -o pgid= -p ${pid} | sed 's/ //g')
   export _pgid
@@ -31,16 +33,30 @@ clh_start_process_monitor() {
   echoerr "[procmon] start"
 
   (
-    shutdown=false
-    while ! $shutdown; do
+    _procmon_pid=${BASHPID}
+    while true; do
       clh_get_tracked_pids
       for pid in "${result[@]}"; do
-        if ! kill -0 "${pid}"; then
-          echoerr "[procmon] ${pid} is dead"
-          rm "${_procmon_output}/${pid}.pid"
+        if kill -0 "${pid}"; then
+          continue
         fi
-        sleep 1
+
+        exit_code=$(cat "${_procmon_output}/${pid}.pid")
+        if [ -z "$exit_code" ]; then
+          echoerr "[procmon] ${pid} died with unknown exit code. Aborting."
+          _clh_halt "halted_no_return"
+        fi
+
+        if [ "$exit_code" -eq 0 ]; then
+          echoerr "[procmon] ${pid} died with exit code $exit_code."
+          rm "${_procmon_output}/${pid}.pid"
+          continue
+        fi
+
+        echoerr "[procmon] ${pid} is dead with exit code $exit_code. Aborting."
+        _clh_halt "halted_process_failure"
       done
+      sleep 1
     done
   ) &
   _procmon_pid=$!
@@ -65,7 +81,30 @@ clh_get_tracked_pids() {
   done
 }
 
-clh_stop_process_monitor() {
+clh_monitor_state() {
+  # No process ID, process monitor never ran.
+  if [ -z "$_procmon_pid" ]; then
+    echo "halted"
+    return 0
+  fi
+
+  # Process ID is set and process is running.
+  if kill -0 "$_procmon_pid"; then
+    echo "running"
+    return 0
+  fi
+
+  if [ -f "${_procmon_output}/procmon_exit_code" ]; then
+    exit_code=$(cat "${_procmon_output}/procmon_exit_code")
+    echo "$exit_code"
+    return 0
+  fi
+
+  echo "error_no_exit_code"
+  return 1
+}
+
+_clh_halt() {
   if [ -z "$_procmon_pid" ]; then
     echoerr "[procmon] process monitor not started"
     return 1
@@ -76,13 +115,26 @@ clh_stop_process_monitor() {
     return 1
   fi
 
-  if [ "$1" = "monitor_only" ]; then
+  echo "$1" > "${_procmon_output}/procmon_exit_code"
+
+  if [ "$_procmon_stop_mode" = "kill_on_exit" ]; then
+    echoerr "[procmon] stop process group. This will halt the script."
+    kill -s TERM "-$_pgid"
+  else
     echoerr "[procmon] stop monitor only. Children will be left behind."
     kill -s TERM "$_procmon_pid"
     await "$_procmon_pid"
     return 0
-  else
-    echoerr "[procmon] stop process group. This will halt the script."
-    kill -s TERM "-$_pgid"
   fi
+}
+
+clh_stop_process_monitor() {
+  _clh_halt "halted"
+}
+
+clh_exit() {
+  exit_code=$1
+  echoerr "[procmon] $BASHPID exit with code $exit_code"
+  echo "$exit_code" > "${_procmon_output}/${BASHPID}.pid"
+  exit "$exit_code"
 }
\ No newline at end of file
diff --git a/test/test_process_monitor.bats b/test/test_process_monitor.bats
index e58d649..b9c2a9d 100644
--- a/test/test_process_monitor.bats
+++ b/test/test_process_monitor.bats
@@ -9,9 +9,15 @@ setup() {
 }
 
 @test "should not start process monitor twice" {
+  assert_equal $(clh_monitor_state) "halted"
+
   assert clh_start_process_monitor
+  assert_equal $(clh_monitor_state) "running"
+
   refute clh_start_process_monitor
-  assert clh_stop_process_monitor "monitor_only"
+
+  assert clh_stop_process_monitor
+  assert_equal $(clh_monitor_state) "halted"
 }
 
 @test "should not stop the process monitor if it wasn't started" {
@@ -19,24 +25,25 @@ setup() {
 }
 
 @test "should keep track of process IDs" {
-  echo "hi"
   assert clh_start_process_monitor
 
   clh_get_tracked_pids
   assert [ ${#result[@]} -eq 0 ]
 
   (
-    while true; do
+    while [ ! -f "${_procmon_output}/sync" ]; do
       sleep 0.1
     done
+    clh_exit 0
   ) &
   clh_track_last_background_job
   p1=$!
 
   (
-    while true; do
+    while [ ! -f "${_procmon_output}/sync" ]; do
       sleep 0.1
     done
+    clh_exit 0
   ) &
   clh_track_last_background_job
   p2=$!
@@ -44,22 +51,49 @@ setup() {
   clh_get_tracked_pids
   assert [ ${#result[@]} -eq 2 ]
 
-  kill -s TERM "$p1"
-  kill -s TERM "$p2"
+  touch "${_procmon_output}/sync"
 
-  echo "Kill issued" > killissued
-
-  # This will hang the bats runner for some reason.
   await "$p1"
   await "$p2"
 
   # This should be more than enough for the process monitor to
   # catch the exits. The alternative would be implementing temporal
   # predicates.
-  sleep 3
+  sleep 1
 
   clh_get_tracked_pids
   assert [ ${#result[@]} -eq 0 ]
 
-  clh_stop_process_monitor "monitor_only"
+  clh_stop_process_monitor
 }
+
+@test "should stop the monitor and all other processes if one process fails" {
+  assert clh_start_process_monitor
+
+  (
+    while [ ! -f "${_procmon_output}/sync" ]; do
+      sleep 0.1
+    done
+    clh_exit 1
+  ) &
+  clh_track_last_background_job
+  p1=$!
+
+  (
+    while [ ! -f "${_procmon_output}/sync" ]; do
+      sleep 1
+    done
+    clh_exit 0
+  ) &
+  clh_track_last_background_job
+  p2=$!
+
+  touch "${_procmon_output}/sync"
+
+  await "$p1"
+  await "$p2"
+
+  sleep 1
+
+  assert_equal $(clh_monitor_state) "halted_process_failure"
+}
\ No newline at end of file