Fix cleanup on failure/interruption

.github/workflows/ci.yaml

@@ -65,7 +65,7 @@ jobs:
           # override/add settings for specific distros
           - { DEB_DISTRO: bookworm, DEBS: true, VERBOSE: bloom }
           - { DEB_DISTRO: noble, VERBOSE: true }
-          - { DEB_DISTRO: jammy, ARCH: arm64, BUILD_TIMEOUT: 3, ROS_SOURCES: good, DEBS: true, EXPECT_EXIT_CODE: 143 }
+          - { DEB_DISTRO: jammy, ARCH: arm64, BUILD_TIMEOUT: 4, ROS_SOURCES: good, DEBS: true, EXPECT_EXIT_CODE: 130 }
           # add failing jobs
           - { DEB_DISTRO: focal, ARCH: x64, ROS_SOURCES: broken, EXPECT_EXIT_CODE: 2, CONTINUE_ON_ERROR: false, DEBS: true }
           - { DEB_DISTRO: focal, ARCH: x64, ROS_SOURCES: broken, EXPECT_EXIT_CODE: 1, CONTINUE_ON_ERROR: true, COLCON_PKG_SELECTION: '--packages-up-to rostime' }

main.js

@@ -7,6 +7,7 @@
 const fs = require('fs')
 const child_process = require('child_process')
 const core = require('@actions/core')
+const ps = require('ps-node')
 
 // find paths of generic.sh and action .sh script
 function script_paths() {
@@ -27,22 +28,28 @@ function script_paths() {
 var [generic, action] = script_paths()
 var child = child_process.spawn(generic, [action], { stdio: 'inherit' })
 
-// kill child on signals SIGINT and SIGTERM
+function forward(signal) {
+  child.kill(signal)
+
+  ps.lookup({ ppid: child.pid }, function (err, resultList) {
+    if (err) { throw new Error(err) }
+    resultList.forEach(function (p) {
+      try { process.kill(p.pid, signal) } catch (error) { }
+    })
+  })
+}
 function handle(signal) {
   console.log('Forwarding signal ' + signal + ' to child process')
-  // Escalate signal INT -> TERM -> KILL
-  // https://github.com/ringerc/github-actions-signal-handling-demo
-  if (signal == 'SIGINT') { signal = 'SIGTERM' }
-  else if (signal == 'SIGTERM') { signal = 'SIGKILL' }
-  child.kill(signal)
+  forward(signal)
 }
+// kill child (and sub processes) on signals SIGINT and SIGTERM
 process.on('SIGINT', handle)
 process.on('SIGTERM', handle)
 
 // exit if child exits
 child.on('exit', function (exit_code, signal) {
   const expect = core.getInput('EXPECT_EXIT_CODE') || 0 // expected exit code
-  exit_code = exit_code !== null ? exit_code : 143
+  exit_code = exit_code !== null ? exit_code : 130
   const suffix = exit_code == expect ? ' (as expected)' : ' != ' + expect
   const msg = 'Process finished with code ' + exit_code + suffix
   exit_code == expect ? core.debug(msg) : core.warning(msg)
@@ -52,9 +59,11 @@ child.on('exit', function (exit_code, signal) {
 // cancel build after given timout (github default: 6h - 20min slack)
 const timeout_minutes = core.getInput('BUILD_TIMEOUT') || (6 * 60 - 20)
 function cancel() {
+  console.log('')
   core.warning('Cancelling build due to timeout')
-  child.kill('SIGTERM')
-  // escalate to SIGKILL after 5s
-  setTimeout(function () { child.kill('SIGKILL') }, 5 * 1000)
+  forward('SIGINT')
+  // escalate to SIGTERM after 5s
+  // https://github.com/ringerc/github-actions-signal-handling-demo
+  setTimeout(function () { forward('SIGTERM') }, 5 * 1000)
 }
 setTimeout(cancel, timeout_minutes * 60 * 1000)