fix(tui): log parent gateway lifecycle exits (#31051)

* fix(tui): log parent gateway lifecycle exits

Add parent-side breadcrumbs for TUI gateway shutdown and transport exits so future backend EOF/SIGTERM reports identify the parent action that caused them.

* chore(tui): retrigger lifecycle logging checks

Retry transient GitHub checkout failures on the lifecycle logging PR.

ui-tui/src/__tests__/gatewayClient.test.ts

@@ -34,6 +34,7 @@ class FakeWebSocket {
       options !== null &&
       'once' in options &&
       Boolean((options as { once?: unknown }).once)
+
     const entries = this.listeners.get(type) ?? []
 
     entries.push({ callback, once })
@@ -84,6 +85,7 @@ class FakeWebSocket {
 
     for (const entry of entries) {
       entry.callback(event)
+
       if (entry.once) {
         this.removeEventListener(type, entry.callback)
       }
@@ -170,6 +172,7 @@ describe('GatewayClient websocket attach mode', () => {
       method: 'event',
       params: { type: 'tool.start', payload: { tool_id: 't1' } }
     })
+
     gatewaySocket.message(eventFrame)
 
     expect(seen).toContain('tool.start')
@@ -193,6 +196,8 @@ describe('GatewayClient websocket attach mode', () => {
     gatewaySocket.close(1011)
 
     expect(exits).toEqual([1011])
+    expect(gw.getLogTail(20)).toContain('[lifecycle] websocket close code=1011')
+    expect(gw.getLogTail(20)).toContain('[lifecycle] transport exit code=1011')
   })
 
   it('rejects pending RPCs with websocket wording when the attached socket closes', async () => {
@@ -226,9 +231,10 @@ describe('GatewayClient websocket attach mode', () => {
     const req = gw.request('session.create', {})
     await vi.waitFor(() => expect(gatewaySocket.sent.length).toBeGreaterThan(0))
 
-    gw.kill()
+    gw.kill('test.shutdown')
 
     await expect(req).rejects.toThrow(/gateway closed/)
+    expect(gw.getLogTail(20)).toContain('[lifecycle] GatewayClient.kill reason=test.shutdown')
   })
 
   it('reattaches when HERMES_TUI_GATEWAY_URL rotates between requests', async () => {
@@ -279,6 +285,7 @@ describe('GatewayClient websocket attach mode', () => {
     gw.drain()
 
     expect(stderrLines.length).toBeGreaterThan(0)
+
     for (const line of stderrLines) {
       expect(line).not.toContain('hunter2')
       expect(line).not.toContain('channel=secret')
@@ -370,6 +377,7 @@ describe('GatewayClient websocket attach mode', () => {
     gw.drain()
 
     expect(stderrLines.length).toBeGreaterThan(0)
+
     for (const line of stderrLines) {
       expect(line).not.toContain('alice')
       expect(line).not.toContain('hunter2')

ui-tui/src/app/useMainApp.ts

@@ -1,4 +1,4 @@
-import { useApp, useHasSelection, useSelection, useStdout, useTerminalTitle, type ScrollBoxHandle } from '@hermes/ink'
+import { type ScrollBoxHandle, useApp, useHasSelection, useSelection, useStdout, useTerminalTitle } from '@hermes/ink'
 import { useStore } from '@nanostores/react'
 import { useCallback, useEffect, useMemo, useRef, useState } from 'react'
 
@@ -365,7 +365,7 @@ export function useMainApp(gw: GatewayClient) {
   const gateway = useMemo(() => ({ gw, rpc }), [gw, rpc])
 
   const die = useCallback(() => {
-    gw.kill()
+    gw.kill('app.die')
     exit()
     // Ink's exit() calls unmount() which resets terminal modes but does NOT
     // call process.exit().  Without an explicit exit the Node process stays
@@ -377,7 +377,7 @@ export function useMainApp(gw: GatewayClient) {
   }, [exit, gw])
 
   const dieWithCode = useCallback((code: number) => {
-    gw.kill()
+    gw.kill(`app.dieWithCode:${code}`)
     exit()
     process.exit(code)
   }, [exit, gw])
@@ -736,10 +736,13 @@ export function useMainApp(gw: GatewayClient) {
   const anyPanelVisible = SECTION_NAMES.some(
     s => sectionMode(s, ui.detailsMode, ui.sections, ui.detailsModeCommandOverride) !== 'hidden'
   )
+
   const thinkingPanelVisible =
     sectionMode('thinking', ui.detailsMode, ui.sections, ui.detailsModeCommandOverride) !== 'hidden'
+
   const toolsPanelVisible =
     sectionMode('tools', ui.detailsMode, ui.sections, ui.detailsModeCommandOverride) !== 'hidden'
+
   const activityPanelVisible =
     sectionMode('activity', ui.detailsMode, ui.sections, ui.detailsModeCommandOverride) !== 'hidden'
 

ui-tui/src/entry.tsx

@@ -43,23 +43,24 @@ setupGracefulExit({
     () => {
       resetTerminalModes()
 
-      return gw.kill()
+      return gw.kill('graceful-exit-cleanup')
     }
   ],
   onError: (scope, err) => {
-    const message = err instanceof Error ? `${err.name}: ${err.message}` : String(err)
+    const message = err instanceof Error ? `${err.name}: ${err.message}\n${err.stack ?? ''}` : String(err)
 
-    process.stderr.write(`hermes-tui ${scope}: ${message.slice(0, 2000)}\n`)
+    process.stderr.write(`hermes-tui lifecycle ${scope}: ${message.slice(0, 2000)}\n`)
   },
   onSignal: signal => {
     resetTerminalModes()
-    process.stderr.write(`hermes-tui: received ${signal}\n`)
+    process.stderr.write(`hermes-tui lifecycle: received ${signal}\n`)
   }
 })
 
 const stopMemoryMonitor = startMemoryMonitor({
   onCritical: (snap, dump) => {
     resetTerminalModes()
+    process.stderr.write(`hermes-tui lifecycle: memory critical exit heap=${formatBytes(snap.heapUsed)} rss=${formatBytes(snap.rss)}\n`)
     process.stderr.write(dumpNotice(snap, dump))
     process.stderr.write('hermes-tui: exiting to avoid OOM; restart to recover\n')
     process.exit(137)

ui-tui/src/gatewayClient.ts

@@ -21,6 +21,14 @@ const WS_CLOSED = 3
 const truncateLine = (line: string) =>
   line.length > MAX_LOG_LINE_BYTES ? `${line.slice(0, MAX_LOG_LINE_BYTES)}… [truncated ${line.length} bytes]` : line
 
+const describeChild = (proc: ChildProcess | null) => {
+  if (!proc) {
+    return 'pid=none'
+  }
+
+  return `pid=${proc.pid ?? 'unknown'} killed=${proc.killed} exitCode=${proc.exitCode ?? 'null'} signal=${proc.signalCode ?? 'null'}`
+}
+
 const resolveGatewayAttachUrl = () => {
   const raw = process.env.HERMES_TUI_GATEWAY_URL?.trim()
 
@@ -85,7 +93,7 @@ const asWireText = (raw: unknown): string | null => {
 // otherwise-malformed URLs that the WHATWG `URL` parser can't accept.
 // Used by the `redactUrl` fallback so embedded credentials are
 // scrubbed from log lines even when the URL is unparseable.
-const _USERINFO_FALLBACK_RE = /^([a-z][a-z0-9+.\-]*:\/\/)[^/?#@]*@/i
+const _USERINFO_FALLBACK_RE = /^([a-z][a-z0-9+.-]*:\/\/)[^/?#@]*@/i
 
 // Connection URLs (gateway, sidecar) often carry bearer tokens in the query
 // string. We surface them in user-facing log lines and the
@@ -191,6 +199,7 @@ export class GatewayClient extends EventEmitter {
     const ws = this.ws
     this.ws = null
     this.wsConnectPromise = null
+
     try {
       ws?.close()
     } catch {
@@ -239,6 +248,7 @@ export class GatewayClient extends EventEmitter {
   private handleTransportExit(code: null | number, reason?: string) {
     this.clearReadyTimer()
     this.closeSidecarSocket()
+    this.pushLog(`[lifecycle] transport exit code=${code ?? 'null'} reason=${reason ?? 'none'}`)
     this.rejectPending(new Error(reason || `gateway exited${code === null ? '' : ` (${code})`}`))
 
     if (this.subscribed) {
@@ -257,6 +267,7 @@ export class GatewayClient extends EventEmitter {
 
     if (typeof WebSocket === 'undefined') {
       this.pushLog(`[sidecar] WebSocket unavailable; skipping mirror to ${redactUrl(this.sidecarUrl)}`)
+
       return
     }
 
@@ -324,6 +335,7 @@ export class GatewayClient extends EventEmitter {
     env.PYTHONPATH = pyPath ? `${root}${delimiter}${pyPath}` : root
     this.startReadyTimer(python, cwd)
     this.proc = spawn(python, ['-m', 'tui_gateway.entry'], { cwd, env, stdio: ['pipe', 'pipe', 'pipe'] })
+    this.pushLog(`[lifecycle] spawned gateway child ${describeChild(this.proc)} python=${python} cwd=${cwd}`)
 
     this.stdoutRl = createInterface({ input: this.proc.stdout! })
     this.stdoutRl.on('line', raw => {
@@ -353,11 +365,14 @@ export class GatewayClient extends EventEmitter {
     this.proc.on('error', err => {
       // Skip stale errors on an already-replaced child.
       if (this.proc !== ownedProc) {
+        this.pushLog(`[lifecycle] stale child error ignored ${describeChild(ownedProc)} message=${err.message}`)
+
         return
       }
 
       const line = `[spawn] ${err.message}`
 
+      this.pushLog(`[lifecycle] child error ${describeChild(ownedProc)} message=${err.message}`)
       this.pushLog(line)
       this.publish({ type: 'gateway.stderr', payload: { line } })
       // Detach the reference up front so the late `exit` event for
@@ -369,14 +384,19 @@ export class GatewayClient extends EventEmitter {
       this.proc = null
       this.handleTransportExit(1, `gateway error: ${err.message}`)
     })
-    this.proc.on('exit', code => {
+    this.proc.on('exit', (code, signal) => {
       // start() can replace `this.proc` while an old child is still
       // tearing down. Skip stale exits so we don't clear the new
       // startup timer or reject newly-issued pending requests.
       if (this.proc !== ownedProc) {
+        this.pushLog(
+          `[lifecycle] stale child exit ignored ${describeChild(ownedProc)} code=${code ?? 'null'} signal=${signal ?? 'null'}`
+        )
+
         return
       }
 
+      this.pushLog(`[lifecycle] child exit ${describeChild(ownedProc)} code=${code ?? 'null'} signal=${signal ?? 'null'}`)
       this.handleTransportExit(code)
     })
   }
@@ -400,6 +420,7 @@ export class GatewayClient extends EventEmitter {
       let settled = false
 
       this.ws = ws
+
       const connectPromise = new Promise<void>((resolve, reject) => {
         ws.addEventListener(
           'open',
@@ -454,9 +475,12 @@ export class GatewayClient extends EventEmitter {
         // new ready timer or reject the new pending requests on behalf
         // of a stale socket.
         if (this.ws !== ws) {
+          this.pushLog(`[lifecycle] stale websocket close ignored code=${ev.code}`)
+
           return
         }
 
+        this.pushLog(`[lifecycle] websocket close code=${ev.code}`)
         this.ws = null
         this.wsConnectPromise = null
         this.handleTransportExit(ev.code, `gateway websocket closed${ev.code ? ` (${ev.code})` : ''}`)
@@ -483,14 +507,17 @@ export class GatewayClient extends EventEmitter {
     this.resetStartupState()
 
     if (this.proc && !this.proc.killed && this.proc.exitCode === null) {
+      this.pushLog(`[lifecycle] replacing live gateway child ${describeChild(this.proc)}`)
       this.proc.kill()
     }
+
     this.proc = null
     this.closeGatewaySocket()
     this.closeSidecarSocket()
 
     if (attachUrl) {
       this.startAttachedGateway(attachUrl)
+
       return
     }
 
@@ -686,8 +713,11 @@ export class GatewayClient extends EventEmitter {
     })
   }
 
-  kill() {
+  kill(reason = 'requested') {
-    this.proc?.kill()
+    const proc = this.proc
+    const killed = proc?.kill()
+
+    this.pushLog(`[lifecycle] GatewayClient.kill reason=${reason} ${describeChild(proc)} killResult=${killed ?? 'none'}`)
     this.closeGatewaySocket()
     this.closeSidecarSocket()
     this.clearReadyTimer()