Fix Sandbox not terminating processes properly when running in Lambda (probably?)

changelog.md

@@ -2,6 +2,14 @@
 
 ---
 
+## [3.7.2] 2021-06-22
+
+### Fixed
+
+- Attempted to fix Sandbox not terminating processes properly when running in Lambda due to Lambda not having *nix `ps`
+
+---
+
 ## [3.7.0 - 3.7.1] 2021-06-14
 
 ### Added

package.json

@@ -1,6 +1,6 @@
 {
   "name": "@architect/sandbox",
-  "version": "3.7.1",
+  "version": "3.7.2-RC.1",
   "description": "Architect dev server: run full Architect projects locally & offline",
   "main": "src/index.js",
   "scripts": {

src/invoke-lambda/spawn.js

@@ -1,4 +1,5 @@
 let { spawn } = require('child_process')
+let { readdirSync } = require('fs')
 let kill = require('tree-kill')
 let { template } = require('../lib')
 let { head } = template
@@ -7,6 +8,7 @@ const SIG = 'SIGINT'
 module.exports = function spawnChild (params, callback) {
   let { cwd, command, args, options, request, timeout, update } = params
   let functionPath = options.cwd.replace(cwd, '').substr(1)
+  let isInLambda = process.env.AWS_LAMBDA_FUNCTION_NAME
   let timedout = false
   let headers = {
     'content-type': 'text/html; charset=utf8;',
@@ -60,11 +62,33 @@ module.exports = function spawnChild (params, callback) {
         update.error(`${functionPath} (pid ${pid}) caught hanging execution with an error, attempting to exit 1`)
         code = 1
       }
-      kill(pid, SIG, () => {
+      if (!isInLambda) {
+        kill(pid, SIG, () => {
+          murderInProgress = false
+          update.debug.status(`${functionPath} (pid ${pid}) successfully terminated`)
+          if (!closed) done(code)
+        })
+      }
+      else {
+        // tree-kill relies on *nix `ps`, which Lambda doesn't have – but it does have /proc
+        // Node process.kill() + Lambda Linux /proc/<pid>/task/<tid> is mysterious, so this may not be the best or proper approach
+        try {
+          let tasks = readdirSync(`/proc/${pid}/task`)
+          tasks.forEach(tid => {
+            try { process.kill(tid) }
+            catch (err) {
+              // Task may have ended naturally or been killed by killing child.pid, I guess we don't really know
+              update.debug.status(`${functionPath} (pid ${pid}) did not kill task (tid ${tid})`)
+            }
+          })
+          update.debug.status(`${functionPath} (pid ${pid}) (possibly maybe) successfully terminated inside Lambda`)
+        }
+        catch (err) {
+          update.debug.status(`${functionPath} (pid ${pid}) failed to terminate inside Lambda: ${err.message}`)
+        }
         murderInProgress = false
-        update.debug.status(`${functionPath} (pid ${pid}) successfully terminated`)
         if (!closed) done(code)
-      })
+      }
     }
     else {
       update.debug.status(`${functionPath} (pid ${pid}) is not running (termination in progress: ${murderInProgress}; process closed: ${closed}`)

test/integration/events-test.js

@@ -243,8 +243,29 @@ test('invoke-lambda should respect timeout for sync functions and process should
   })
 })
 
+test('invoke-lambda should respect timeout for sync functions and process should be killed (inside actual AWS Lambda)', t => {
+  let isLinux = process.platform === 'linux'
+  t.plan(1)
+  if (isLinux) {
+    process.env.AWS_LAMBDA_FUNCTION_NAME = 'yep'
+    let fileThatShouldNotBeWritten = join(tmp, 'foo-sync')
+    arc.events.publish({
+      name: 'event-timeout-sync',
+      payload: { path: fileThatShouldNotBeWritten }
+    },
+    function done (err) {
+      if (err) t.fail(err)
+      else setTimeout(() => {
+        t.notOk(existsSync(fileThatShouldNotBeWritten), 'file not created by event as event timed out and process was terminated appropriately')
+      }, 1250) // 1s is the configured timeout of test/mock/normal/src/events/event-timeout-sync so we pad it a bit and check after delay
+    })
+  }
+  else t.pass('Skipped because !Linux')
+})
+
 test('Sync events.end', t => {
   t.plan(1)
+  delete process.env.AWS_LAMBDA_FUNCTION_NAME
   events.end(function (err, result) {
     if (err) t.fail(err)
     else t.equal(result, 'Event bus successfully shut down', 'Events ended')