Bug Description

Gateway heartbeat feature stops working after triggering 1-2 times. The timer does not reschedule after firing, causing heartbeat to permanently stop.

Timeline

• 2026-03-09 16:50: User asked about OpenClaw v2026.3.8 release notes
• 2026-03-09 17:02:12: Config hot reload (lossless-claw plugin update)
• 2026-03-09 17:02:14: Heartbeat first triggered (never worked correctly from the start)
• 2026-03-09 to 2026-03-13: Heartbeat intervals chaotic (minutes to hours, not periodic)
• 2026-03-14 10:14: Heartbeat triggered last time
• 2026-03-14 10:14 to 12:46: No heartbeat for 2.5 hours
• 2026-03-14 12:46: Gateway restarted
• 2026-03-14 12:46:00: Heartbeat triggered once after restart
• 2026-03-14 13:16: Expected heartbeat (MISSING)
• 2026-03-14 13:37: Still no heartbeat (51 minutes since last trigger)

Evidence

Log excerpts (2026-03-14):

2026-03-14T10:11:18.478+08:00 [heartbeat] started
2026-03-14T10:14:11.872+08:00 [heartbeat] started
# GAP: 2.5 hours with NO heartbeat
2026-03-14T12:46:00.236+08:00 [heartbeat] started (after Gateway restart)
# NO subsequent triggers (expected every 30m)

Configuration at time of failure:

{
  "agents": {
    "defaults": {
      "heartbeat": {
        "every": "30m",
        "activeHours": {
          "start": "07:00",
          "end": "24:00",
          "timezone": "Asia/Shanghai"
        },
        "target": "last",
        "directPolicy": "allow"
      }
    },
    "list": [
      {
        "id": "agent-qwen",
        "heartbeat": {}
      }
    ]
  }
}

Root Cause Analysis

1. Heartbeat never worked correctly

From first activation on 2026-03-09, intervals were chaotic (not periodic as configured with every: "1h")

2. Hot reload may corrupt timer

Config hot reloads on 2026-03-14 (10:05 and 10:13) were followed by heartbeat stopping completely

3. Timer not rescheduling

After each trigger, the timer should reschedule for next interval. Instead, it fires 1-2 times then stops permanently.

4. Gateway restart only temporarily fixes

Restart causes one trigger, then timer stops again.

🔍 CRITICAL: Related to Lossless-Claw Context Engine

Timeline correlation:

• 2026-03-08 20:15: Issue Plugin context engines can resolve before plugin registration (Lossless Claw / contextEngine slot) #40232 created — "Plugin context engines can resolve before plugin registration"
• 2026-03-09 17:02:12: lossless-claw plugin config hot reload triggered
• 2026-03-09 17:02:14: Heartbeat first started (immediately malfunctioned)
• 2026-03-10 to 2026-03-13: Multiple heartbeat + context engine issues reported simultaneously

Gateway log (2026-03-09 17:02):

17:02:12 [reload] config change detected (plugins.installs.lossless-claw.version...)
17:02:22 [lcm] Plugin loaded (enabled=true, db=/Users/neoshi/.openclaw/lcm.db)
17:02:23 [lcm] Plugin loaded (repeated)
17:02:30 [lcm] Plugin loaded (repeated)
17:02:32 [lcm] Plugin loaded (repeated)
17:02:33 [lcm] Plugin loaded (repeated)

Related Issues:

Lossless-Claw / Context Engine:

Heartbeat:

Hypothesis:

The lossless-claw context engine registration timing bug (#40232) may be causing heartbeat scheduler to fail:

1. Context engine resolves before plugin registration completes
2. Heartbeat scheduler depends on context engine state
3. State inconsistency causes timer to not reschedule after trigger
4. Repeated plugin reload (seen in logs) may corrupt scheduler state

Expected Behavior

Heartbeat should trigger periodically (every 30m or 1h as configured) during activeHours (07:00-24:00).

Actual Behavior

• Heartbeat triggers 1-2 times after Gateway start/restart
• Timer does not reschedule
• No further triggers until next restart
• Intervals are chaotic even when working

Impact

• Heartbeat-dependent features (daily calendar checks, weather reports, email monitoring) do not work
• Users must manually restart Gateway or use cron-based workaround

Environment

• OpenClaw version: v2026.3.8 (issue introduced in this version)
• OS: macOS (Darwin 25.3.0, arm64)
• Node: v25.8.1
• First occurrence: 2026-03-09 (coincides with v2026.3.8 update + lossless-claw hot reload)
• Plugin: @martian-engineering/lossless-claw (repeatedly loaded during hot reload)

Workaround

Use cron job instead of built-in heartbeat:

{
  "schedule": {
    "kind": "cron",
    "expr": "0,30 7-23 * * *",
    "tz": "Asia/Shanghai"
  },
  "payload": {
    "kind": "agentTurn",
    "message": "Execute heartbeat tasks"
  }
}

Request

1. Investigate Gateway heartbeat timer implementation — check setInterval/setTimeout logic
2. Check if hot reload properly resets/reschedules timer — config changes should not break timer
3. Investigate lossless-claw context engine registration timing (Plugin context engines can resolve before plugin registration (Lossless Claw / contextEngine slot) #40232) — may be related to heartbeat scheduler failure
4. Verify every duration parsing — ensure correct interval calculation
5. Consider adding heartbeat health monitoring — alert if no triggers in 2x expected interval
6. Review plugin reload behavior — repeated plugin load (seen in logs) may corrupt scheduler state