Serialize E2E signal cleanup ownership

Prevent opposite signals and normal finalization from racing detached cleanup or releasing its run lock.

Author: Bill Leoutsakos

apps/sim/e2e/foundation/build-manifest.spec.ts

@@ -124,12 +124,13 @@ test('orchestrator lock rejects live ownership and recovers stale descriptors',
   try {
     const lock = acquireE2eRunLock(lockPath)
     expect(() => acquireE2eRunLock(lockPath)).toThrow(/Another E2E orchestrator/)
-    lock.transfer(process.pid)
+    expect(lock.transfer(process.pid)).toBe(true)
     expect(() => acquireE2eRunLock(lockPath)).toThrow(/Another E2E orchestrator/)
     lock.retain('manual cleanup required')
-    lock.transfer(process.pid)
+    expect(lock.transfer(process.pid)).toBe(true)
     expect(() => acquireE2eRunLock(lockPath)).toThrow(/manual cleanup required/)
     lock.release()
+    expect(lock.transfer(process.pid)).toBe(false)
 
     mkdirSync(lockPath)
     expect(() => acquireE2eRunLock(lockPath)).toThrow(/is acquiring/)

apps/sim/e2e/scripts/run.ts

@@ -111,6 +111,7 @@ async function main(): Promise<void> {
   let canaryCoverageComplete = true
   let diagnosticsRetained = true
   let failed = false
+  let signalCleanupStarted = false
 
   const cleanup = (): Promise<void> => {
     if (cleanupPromise) return cleanupPromise
@@ -163,6 +164,8 @@ async function main(): Promise<void> {
   }
 
   const handleSignal = async (signal: NodeJS.Signals): Promise<void> => {
+    if (signalCleanupStarted) return
+    signalCleanupStarted = true
     failed = true
     const exitCode = signal === 'SIGINT' ? 130 : 143
     process.exitCode = exitCode
@@ -211,7 +214,9 @@ async function main(): Promise<void> {
         if (!cleanupProcess.pid) {
           throw new Error('Detached E2E cleanup supervisor started without a process ID')
         }
-        runLock.transfer(cleanupProcess.pid)
+        if (!runLock.transfer(cleanupProcess.pid)) {
+          throw new Error('Detached E2E cleanup supervisor could not acquire run-lock ownership')
+        }
         lockTransferred = true
         cleanupProcess.unref()
         console.error(`Detached E2E cleanup supervisor started as PID ${cleanupProcess.pid}`)
@@ -227,7 +232,8 @@ async function main(): Promise<void> {
     }
     process.exit(exitCode)
   }
-  // `once` is intentional: a second signal uses the OS default force termination.
+  // Both signal types share a synchronous single-flight guard so opposite signals cannot launch
+  // competing supervisors while the first handler awaits child-process startup.
   const handleSigint = (): void => void handleSignal('SIGINT')
   const handleSigterm = (): void => void handleSignal('SIGTERM')
   process.once('SIGINT', handleSigint)
@@ -412,8 +418,10 @@ async function main(): Promise<void> {
     process.off('SIGINT', handleSigint)
     process.off('SIGTERM', handleSigterm)
     setManagedProcessGroupObserver(null)
-    if (cleanupSucceeded) runLock.release()
-    else runLock.retain('normal cleanup failed; inspect diagnostics and clean resources manually')
+    if (!signalCleanupStarted) {
+      if (cleanupSucceeded) runLock.release()
+      else runLock.retain('normal cleanup failed; inspect diagnostics and clean resources manually')
+    }
     console.info(
       diagnosticsRetained
         ? `E2E ${failed ? 'failed' : 'completed'}; diagnostics: ${runDirectory}`

apps/sim/e2e/support/run-lock.ts

@@ -32,7 +32,7 @@ export interface E2eRunLock {
   path: string
   token: string
   setProcessGroupIds(processGroupIds: number[]): void
-  transfer(pid: number): void
+  transfer(pid: number): boolean
   retain(reason: string): void
   release(): void
 }
@@ -74,14 +74,16 @@ export function acquireE2eRunLock(
             ),
           })
         },
-        transfer(pid: number): void {
+        transfer(pid: number): boolean {
           const current = readDescriptor(lockPath)
-          if (current?.token !== descriptor.token) return
+          if (current?.token !== descriptor.token) return false
           writeDescriptor(lockPath, {
             ...current,
             pid,
             processStartIdentity: readProcessStartIdentity(pid),
           })
+          const transferred = readDescriptor(lockPath)
+          return transferred?.token === descriptor.token && transferred.pid === pid
         },
         retain(reason: string): void {
           retainE2eRunLock(lockPath, descriptor.token, reason)