Fix: allow manual restart while process is WAITING_RESTART

[jamesastound]

Title: Fix: allow manual restart while process is WAITING_RESTART (clear stale pid before start)

Summary

This PR fixes an issue where pm2 restart <id> would fail while a process was in WAITING_RESTART (e.g., due to --restart-delay or exp_backoff_restart_delay). The daemon sometimes retained a stale proc.process.pid in its clusters_db, causing startProcessId() to reject a start with "Process with pid already exists" or the CLI to display "Process not found".

What I changed

• lib/God/ActionMethods.js

  • God.startProcessId(id, cb) now checks whether proc.process.pid is actually running using God.checkProcess(pid). If it is not alive, it clears proc.process.pid = 0 and proceeds to start. If proc.process.pid is alive, it must still reject the start as before.

• test/programmatic/restart-delay.mocha.js

  • Add a programmatic test that reproduces WAITING_RESTART by launching a failing script (wrong.js) with restart_delay and waiting until the process enters waiting restart, then attempts pm2.restart(<pm_id>) while still waiting.

• test/programmatic/fixtures/restart-delay/wrong.js

  • Minimal reproduction file which throws immediately.

• Dockerfile.localpm2

  • A local Dockerfile for building an image with the modified pm2 (for maintainers/devs to reproduce the behavior and test).

• docker_test.sh

  • A script to run in the Docker image to validate the behavior non-interactively: start the failing script with restart_delay, wait for waiting restart, then restart the process by pm_id.

• test/unit.sh

  • Insert the restart-delay.mocha.js into the unit test runner.

Why this is needed

• A stale proc.process.pid that remains set in clusters_db while a process is in waiting restart can incorrectly block manual restarts, preventing users from intervening through pm2 restart <id>. Historically, PRs Fix the bug: process with "restart-delay" can't restart when manually stop Unitech/pm2#3572 and Fix the bug: process with "restart-delay" can't restart when manually stop Unitech/pm2#3584 partially addressed PID cleanup, but there were still edge cases where a value persisted and caused rejected starts. This fix ensures the start path is defensive against a stale pid.

How I tested

• Unit test: Added a new test under test/programmatic/restart-delay.mocha.js which polls until the process is waiting restart then restarts it by pm_id. The test verifies that the restart request succeeds.
• Docker integration (manual): The included Dockerfile.localpm2 builds a container with the modified local pm2 and runs docker_test.sh which performs the same reproduction and restart non-interactively.

Example behavior (Docker run)

• The docker run logs show:
  • PM2 daemon spawn
  • Start & crash of wrong.js into waiting restart
  • pm2 restart attempted while status waiting restart and the command returns success: [PM2] wrongtest ✓

Files changed - summary

• lib/God/ActionMethods.js (modified: add defensive pid check + clear stale pid if not alive)
• test/programmatic/restart-delay.mocha.js (new)
• test/programmatic/fixtures/restart-delay/wrong.js (new)
• test/unit.sh (updated to include new test)
• Dockerfile.localpm2 (new helper for reproducing changes in Docker)
• docker_test.sh (new helper to run non-interactive docker verification)

Backwards compatibility & risk

• This is a defensive runtime check—if proc.process.pid is not running, we clear it and proceed; this should not affect healthy process lifecycles.
• The main risk is if God.checkProcess incorrectly reports a process as dead in rare races; the code conservatively defaults to clearing the pid only when God.checkProcess returns false or throws.

CI / Reporting / Notes

• Please run the full unit suite and e2e tests to ensure no regressions:

  • bash test/unit.sh
  • bash test/e2e.sh

• You can reproduce using the included Docker container:

  # Build
  docker build -t pm2-local-mod:latest -f Dockerfile.localpm2 .
  
  # Run the integrated test
  docker run --rm -it pm2-local-mod:latest /bin/sh -lc "/usr/src/app/docker_test.sh"

• The change is small, and the tests added cover the scenario for restart with --restart-delay.

Related issues & PRs

• Fixes / relates to: Issue Stop a process that has set "restart-delay" then restart, throw "Process not found" Unitech/pm2#3571 (original report)
• References: PR Fix the bug: process with "restart-delay" can't restart when manually stop Unitech/pm2#3572, PR Fix the bug: process with "restart-delay" can't restart when manually stop Unitech/pm2#3584 (previous pid-clearing changes)