Environment

• OpenClaw version: 2026.2.22-2
• Node.js: v24.13.1
• OS: Ubuntu Linux 6.8.0-100-generic (x64)
• Host: VPS with ~8GB RAM

Issue

The openclaw-gateway service reached 6.0 GB peak memory (as reported by systemd memory peak) after only ~26 minutes of runtime. This is abnormally high for a gateway process.

Timeline

1. Gateway started at 17:24:42 UTC (PID 21315)
2. Normal operation: Telegram message routing, agent sessions (Claude Opus), tool calls (exec, memory_search)
3. SIGTERM received at 17:50:34 UTC (systemd stop)
4. systemd reported: Consumed 1min 15.038s CPU time, 6.0G memory peak, 0B memory swap peak

Additional observations

• After shutdown, orphaned child processes remained (bash, systemctl), preventing clean restart
• systemd attempted 6 restarts that all failed with gateway already running (pid XXXX); lock timeout after 5000ms and Port 18789 is already in use
• The stale process eventually exited, and the 7th restart succeeded
• A subsequent shorter session (~12 min) peaked at 1.0 GB

Possible causes

• Large context accumulation in agent sessions (Opus model with full system prompts + workspace files + tool results)
• Exec tool subprocess output buffering (large strings held in memory)
• Child process cleanup not fully handled on SIGTERM

Workaround

Added MemoryHigh=1536M and MemoryMax=2G to the systemd unit.

Logs

Feb 23 18:50:34 erebus node[21315]: 2026-02-23T17:50:34.728Z [gateway] signal SIGTERM received
Feb 23 18:50:34 erebus node[21315]: 2026-02-23T17:50:34.731Z [gateway] received SIGTERM; shutting down
Feb 23 18:50:35 erebus systemd[3281]: openclaw-gateway.service: Unit process 21922 (bash) remains running after unit stopped.
Feb 23 18:50:35 erebus systemd[3281]: openclaw-gateway.service: Unit process 21923 (systemctl) remains running after unit stopped.
Feb 23 18:50:35 erebus systemd[3281]: openclaw-gateway.service: Consumed 1min 15.038s CPU time, 6.0G memory peak, 0B memory swap peak.