PR #44494: fix(daemon): trigger SIGUSR1 in-process restart when scheduling launchd handoff

• Repository: openclaw/openclaw
• Author: palomyates516-alt
• State: closed | merged: False
• Link: https://github.com/openclaw/openclaw/pull/44494

Description (problem / solution / changelog)

Summary

Fixes #44490 - Gateway restart fails with SIGTERM instead of SIGUSR1.

Root Cause

When openclaw gateway restart was called from within the LaunchAgent-managed gateway process, the code would schedule a detached handoff script but never send SIGUSR1 to trigger graceful in-process restart. The process would eventually receive SIGTERM from launchd's kickstart -k and terminate completely instead of restarting gracefully.

In restartLaunchAgent() (src/daemon/launchd.ts), when isCurrentProcessLaunchdServiceLabel() returns true:

• Before: Only scheduled handoff script, process never got SIGUSR1
• After: Send SIGUSR1 first to trigger in-process restart, then schedule handoff as fallback

Fix

Modified restartLaunchAgent() to send SIGUSR1 signal before scheduling the handoff:

1. First triggers in-process restart via SIGUSR1 (graceful shutdown + reinitialization)
2. Then schedules detached handoff script as fallback (ensures new instance starts if process exits)
3. Includes error handling: if SIGUSR1 fails, logs warning but continues with handoff

This ensures the gateway logs show [gateway] signal SIGUSR1 received instead of terminating with SIGTERM.

Test plan

• Type check passes (no new errors in src/daemon/launchd.ts)
• All 23 tests in src/daemon/launchd.test.ts pass
• All lifecycle and restart tests pass:
  • src/daemon/launchd-restart-handoff.test.ts (1 test)
  • src/cli/daemon-cli/lifecycle.test.ts (8 tests)
  • src/cli/gateway-cli/run-loop.test.ts (8 tests)

Manual verification

To test on macOS with LaunchAgent:

1. Start gateway: openclaw gateway run --bind loopback
2. From another terminal: openclaw gateway restart
3. Expected: Gateway logs show signal SIGUSR1 received and restarts gracefully
4. Before fix: Gateway would terminate with signal SIGTERM received and not restart

Changed files

• src/daemon/launchd.ts (modified, +18/-0)

---

PR #44495: fix: add missing 'xhigh' option to --thinking help text

• Repository: openclaw/openclaw
• Author: darrenzeng2025
• State: closed | merged: False
• Link: https://github.com/openclaw/openclaw/pull/44495

Description (problem / solution / changelog)

描述

修复 #44443: CLI --thinking help text missing 'xhigh' option

变更内容

• 更新 src/cli/cron-cli/register.cron-add.ts 中的help text，添加xhigh选项
• 更新 src/cli/cron-cli/register.cron-edit.ts 中的help text，添加xhigh选项

测试

• 本地测试通过
• pnpm build 通过
• 代码变更符合项目规范

AI辅助声明

• 此PR使用AI辅助生成
• 已进行充分测试
• 理解代码变更内容

Changed files

• .agent-team/agents/code-reviewer.md (added, +64/-0)
• .agent-team/agents/issue-finder.md (added, +47/-0)
• .agent-team/agents/issue-fixer.md (added, +49/-0)
• .agent-team/agents/pr-submitter.md (added, +83/-0)
• .agent-team/coordinator.ts (added, +670/-0)
• .agent-team/logs/coordinator.log (added, +13/-0)
• .agent-team/state/issue-44490-analysis.md (added, +1716/-0)
• .agent-team/state/progress.json (added, +13/-0)
• src/cli/cron-cli/register.cron-add.ts (modified, +1/-1)
• src/cli/cron-cli/register.cron-edit.ts (modified, +1/-1)

---

PR #44514: fix(daemon): trigger SIGUSR1 in-process restart when scheduling launchd handoff

• Repository: openclaw/openclaw
• Author: palomyates516-alt
• State: closed | merged: False
• Link: https://github.com/openclaw/openclaw/pull/44514

Description (problem / solution / changelog)

Summary

Fixes #44490 - Gateway restart fails with SIGTERM instead of SIGUSR1.

Root Cause

When openclaw gateway restart was called from within the LaunchAgent-managed gateway process, the code would schedule a detached handoff script but never send SIGUSR1 to trigger graceful in-process restart. The process would eventually receive SIGTERM from launchd's kickstart -k and terminate completely instead of restarting gracefully.

In restartLaunchAgent() (src/daemon/launchd.ts), when isCurrentProcessLaunchdServiceLabel() returns true:

• Before: Only scheduled handoff script, process never got SIGUSR1
• After: Send SIGUSR1 first to trigger in-process restart, then schedule handoff as fallback

Fix

Modified restartLaunchAgent() to send SIGUSR1 signal before scheduling the handoff:

1. First triggers in-process restart via SIGUSR1 (graceful shutdown + reinitialization)
2. Then schedules detached handoff script as fallback (ensures new instance starts if process exits)
3. Includes error handling: if SIGUSR1 fails, logs warning but continues with handoff

This ensures the gateway logs show [gateway] signal SIGUSR1 received instead of terminating with SIGTERM.

Test plan

• Type check passes (no new errors in src/daemon/launchd.ts)
• All 23 tests in src/daemon/launchd.test.ts pass
• All lifecycle and restart tests pass:
  • src/daemon/launchd-restart-handoff.test.ts (1 test)
  • src/cli/daemon-cli/lifecycle.test.ts (8 tests)
  • src/cli/gateway-cli/run-loop.test.ts (8 tests)

Manual verification

To test on macOS with LaunchAgent:

1. Start gateway: openclaw gateway run --bind loopback
2. From another terminal: openclaw gateway restart
3. Expected: Gateway logs show signal SIGUSR1 received and restarts gracefully
4. Before fix: Gateway would terminate with signal SIGTERM received and not restart

Changed files

• src/daemon/launchd.ts (modified, +18/-0)

---

PR #44518: fix(daemon): trigger SIGUSR1 in-process restart when scheduling launchd handoff

• Repository: openclaw/openclaw
• Author: palomyates516-alt
• State: open | merged: False
• Link: https://github.com/openclaw/openclaw/pull/44518

Description (problem / solution / changelog)

Fixes #44490 - Gateway restart fails to restart service (SIGTERM instead of SIGUSR1)

Root Cause

When restarting from within the gateway process (detected via isCurrentProcessLaunchdServiceLabel()), the code only scheduled a detached handoff script but never triggered the in-process graceful restart via SIGUSR1. This meant the gateway would receive SIGTERM from launchd instead of gracefully restarting.

Fix

Emit SIGUSR1 before scheduling the handoff script. This ensures:

1. The gateway gracefully restarts in-process (via SIGUSR1 handler)
2. The handoff script waits for the process to exit
3. A new instance starts after the current process terminates

Follows the same pattern used by emitGatewayRestart() in src/infra/restart.ts.

Test plan

• All launchd-related tests pass (launchd.test.ts 23 tests)
• Gateway restart from within process triggers SIGUSR1
• Handoff script still starts new instance after process exits

Changed files

• src/daemon/launchd.ts (modified, +18/-0)

---

PR #44531: fix: gateway restart sends SIGUSR1 via systemctl reload instead of SIGTERM (#44490)

• Repository: openclaw/openclaw
• Author: Mrxyx
• State: closed | merged: False
• Link: https://github.com/openclaw/openclaw/pull/44531

Description (problem / solution / changelog)

Problem

Running openclaw gateway restart when the gateway is managed by systemd stops the service but fails to restart it properly. The gateway log shows it received SIGTERM instead of SIGUSR1.

Root Cause

When the systemd service is loaded, runServiceRestart() in lifecycle-core.ts calls service.restart(), which maps to restartSystemdService() in systemd.ts. This function ran systemctl --user restart <unit>, which sends SIGTERM to stop the process, then starts a new one.

However, the gateway run-loop (run-loop.ts) treats SIGTERM as a stop signal — it shuts down gracefully and exits. The process never performs an in-process restart. Systemd eventually restarts it via Restart=always after RestartSec=5, but this is a cold restart, not the intended graceful restart.

The correct restart path is SIGUSR1, which triggers a graceful in-process restart: drain active tasks, close the server, then re-initialize. This is what restartGatewayWithoutServiceManager() already does for unmanaged processes.

Fix

1. systemd-unit.ts: Add ExecReload=/bin/kill -USR1 $MAINPID to the systemd unit template so systemctl reload sends SIGUSR1.
2. systemd.ts: Change restartSystemdService() to try systemctl reload first (graceful SIGUSR1). Falls back to systemctl restart for older unit files that lack ExecReload.

Why macOS is unaffected

On macOS, restartLaunchAgent() uses launchctl kickstart -k, which kills and restarts the process externally. With KeepAlive=true, launchd handles the full lifecycle — the gateway does not need to self-restart.

Files Changed

• src/daemon/systemd-unit.ts — Added ExecReload directive
• src/daemon/systemd.ts — Prefer reload over restart
• src/daemon/systemd.test.ts — Updated and added tests

Fixes #44490

Changed files

• src/daemon/systemd-unit.ts (modified, +3/-0)
• src/daemon/systemd.test.ts (modified, +41/-6)
• src/daemon/systemd.ts (modified, +13/-0)