Orphaned Chrome processes block session restart and cause authentication re-login loop

[atani]

Problem

When the daemon process dies unexpectedly (crash, forced kill, or certain session-stop timing issues), the Chrome browser process it spawned becomes orphaned but continues running. This creates a cascading failure:

1. Orphaned Chrome holds the userDataDir lock — launchPersistentContext prevents reusing the same userDataDir while another Chrome instance is using it
2. New session can't reuse previous cookies — starting a new session either errors out ("Browser is already in use for ...") or requires a fresh userDataDir
3. Authentication state is lost — for SAML/SSO-protected sites, the session cookie (_fi_sess on GitHub Enterprise, for example) is a session cookie (expires=0), meaning it's destroyed when the Chrome process terminates
4. Process accumulation — repeating this cycle accumulates orphaned Chrome processes (I had 47 at one point)

Root cause analysis

The daemon spawns Chrome with detached: true + child.unref() (program.js:201-214), which correctly decouples the daemon from its parent. However, when the daemon exits (via session-stop → gracefullyProcessExitDoNotHang), the Chrome process it launched is not guaranteed to terminate:

• Context.disposeAll() calls browserContext.close(), which should close Chrome
• But if this takes longer than the 15-second watchdog timeout (watchdog.js:32), process.exit(0) is called while Chrome is still running
• The orphaned Chrome process has no parent to send SIGTERM to it

Evidence

$ playwright-cli session-list
  (no sessions)

$ playwright-cli open https://example.com
Error: Browser is already in use for /Users/me/.playwright-cli/ghe-profile

$ ps aux | grep ghe-profile | head -1
user  93516  ...  /Applications/Google Chrome.app/.../Google Chrome --user-data-dir=/Users/me/.playwright-cli/ghe-profile --remote-debugging-port=55665 ...

$ ps -o ppid= -p 93516
  → Parent process gone (orphan!)

Proposed solution

1. Kill orphaned Chrome on session-stop (critical)

Before or after sending the "stop" message to the daemon, detect and kill Chrome processes associated with the session's userDataDir. This could use:

• The CDP port recorded during launch to send Browser.close() via CDP
• Or lsof / process scanning to find Chrome processes using the specific userDataDir

2. Reconnect to orphaned Chrome via CDP (nice-to-have)

If a Chrome process is found with a known --remote-debugging-port, attempt CDP reconnection instead of launching a new instance. This would preserve the existing authentication state.

3. Cookie persistence across session restarts (nice-to-have)

Add cookies export <file> / cookies import <file> commands (similar to browser-use CLI) to save/restore authentication state. This would help with SAML/SSO sites where session cookies are lost on browser restart.

Environment

• OS: macOS (Darwin 25.3.0)
• playwright-cli: 0.0.61 (@playwright/cli)
• Browser: Google Chrome (system Chrome via --profile)

Related

• unable to open two session at once #293 — "unable to open two session at once" (same underlying launchPersistentContext single-instance constraint)