hermes - 💡(How to fix) Fix hermes update fails to restart gateway on Linux systemd (exit code 75 handoff broken)

Bug Description

After hermes update completes on Linux (systemd user unit), the gateway is not restarted. The update log shows "draining..." but never prints the expected "✓ Restarted hermes-gateway" line. The gateway process exits with code 75 (TEMPFAIL) during the restart handoff and systemd's Restart=always + RestartSec=300s is the only thing that brings it back — after a 5-minute gap.

This is the Linux/systemd equivalent of #28135 ("hermes update via gateway leaves launchd service unrestarted on macOS"), which was fixed by #28341 (exit code 75 for launchd). But the Linux path is still broken.

Environment

• OS: Linux (Ubuntu 24.04, x86_64)
• Hermes: v0.14.0 (2026.5.16), installed as systemd user unit
• Update: from 116 commits behind to current main

Steps to Reproduce

1. Have Hermes gateway running as a systemd user service
2. Run hermes update (either from CLI or via gateway Telegram command)
3. Observe the update log — it shows "draining..." but never "✓ Restarted"
4. hermes gateway status shows gateway is not running
5. After ~5 minutes, systemd auto-restarts the gateway

Actual Behavior

Update log ends with no restart confirmation. Gateway is down for ~5 minutes.

Expected Behavior

Gateway should restart within seconds of the update completing.

Timeline from Logs (2026-05-23, UTC+7)

23:45:27  hermes update started (116 commits pulled)
23:46:58  Gateway received stop signal
23:47:03  Gateway shut down cleanly (4.81s teardown)
23:47:03  Replacement gateway starts (--replace flag)
23:47:05  Replacement exits with code 75
          Update log ends — NO restart confirmation
23:52:05  systemd auto-restarts after RestartSec=300s
23:52:12  Telegram reconnects

Exit Diagnostic

From gateway-exit-diag.log:

{"tag": "asyncio.run.SystemExit", "pid": 87094, "code": 75,
 "traceback": "...gateway/run.py:18159 in start_gateway\nSystemExit: 75"}

Analysis

The exit code 75 is intentional (#28341) — it signals the service manager to relaunch. But the handoff between the old gateway (exiting 75) and the new gateway (started with --replace) fails: the replacement also exits 75 immediately.

This is separate from #29421 (D-Bus hang on hermes gateway restart) and #30297 (fix for that via --no-block). The update flow does not use systemd_restart() — it goes through the gateway's internal self-restart mechanism, which is a different code path.

Workaround

Wait for systemd's Restart=always to recover (~5 minutes), or manually run:

systemctl --user reset-failed hermes-gateway
systemctl --user start hermes-gateway

Related

• #28135 — macOS equivalent (fixed by #28341)
• #29421 — D-Bus hang on restart command (fix pending: #30297)
• #8260 — restart returns before completion