chore: add icon.ico so the exe gets the proper program icon

thor-watcher had no icon.ico, so build.bat's 'if exist icon.ico' check
always fell through to building without --icon → PyInstaller's default
Python/floppy icon. Reuse the shared Terra-Mechanics icon from
series3-watcher; build.bat now picks it up automatically (no script
change needed). Rides with the pending 0.4.1 build.

Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>

CHANGELOG.md

@@ -5,6 +5,38 @@ All notable changes to this project will be documented in this file.
 The format is based on [Keep a Changelog](https://keepachangelog.com/en/1.1.0/),
 and this project adheres to [Semantic Versioning](https://semver.org/spec/v2.0.0.html).
 
+## [0.4.1] - 2026-06-25
+
+### Fixed
+- **Auto-updater no longer dies when settings are saved.** Saving the Settings dialog restarts the watcher thread via `_restart_watcher`, which sets and then replaces the shared `stop_event`. Because the updater loop keyed off that *same* event, saving settings would terminate the auto-updater thread (a race — it could survive one save and die on the next). The watcher kept running and reporting heartbeats normally, but the agent silently stopped checking for / applying updates until the tray app was relaunched. The updater now keys off its own app-lifetime event (`_app_stop`) that watcher restarts never touch; it stops only on Exit or when applying an update.
+- Added `test_thor_tray.py` covering the tray thread lifecycle (updater survives a watcher restart; stops on app exit; `_exit`/`_do_update` signal the lifetime event).
+
+## [0.4.0] - 2026-06-25
+
+### Added — Mirror (dual-send)
+- **Optional best-effort mirror server.** Each heartbeat and each forwarded event can now be sent to a *second* ("mirror") server in addition to the primary, so a standby box (e.g. the office NAS) stays a continuous replica during a migration — making the eventual cutover a non-event. **Default off** — blank `mirror_api_url`/`mirror_sfm_url` means existing installs are unchanged after auto-update.
+- **Heartbeat mirror (best-effort).** After the primary heartbeat POST, the same payload is fired at `mirror_api_url` when set. Bounded by the existing API timeout and fully wrapped — it can never delay or fail the primary heartbeat; its result is ignored except for a debug log line.
+- **Event mirror (reliable, isolated).** Events are re-forwarded to `mirror_sfm_url` using a *separate* sha256 state file (`mirror_sfm_state_file`, blank → `<log_dir>/thor_forwarded_mirror.json`), so the mirror tracks its own delivery independently of the primary. A quick reachability probe (~3 s) skips the mirror pass when the mirror server is down, so a dead mirror never stalls the loop on per-event timeouts; skipped events stay pending in the mirror state and deliver when the mirror returns — **no data loss**.
+- **Isolation invariant (the one rule).** Nothing on the mirror path can delay or fail the primary: its own state file, its own try/except, the reachability guard + bounded timeouts, all exceptions swallowed-and-logged. The primary path stays exactly as reliable as before.
+- **Mirror tab in the Settings dialog** with `Mirror API URL` + `Mirror SFM URL` fields (blank = off). The mirror rides along with the primary — there is no separate enable flag (heartbeat-mirror active iff the primary heartbeat is on *and* `mirror_api_url` is set; event-mirror active iff primary forwarding is on *and* `mirror_sfm_url` is set).
+- New unit tests in `test_event_forwarder.py` covering mirror reachability, the down-mirror skip, separate-state idempotency, and the isolation invariant (a mirror failure leaves the primary's result + state untouched).
+
+### Configuration
+
+New `config.json` keys (all default-empty = off; existing 0.3.x deployments don't change behaviour on auto-update):
+
+| Key | Default | Notes |
+|---|---|---|
+| `mirror_api_url` | `""` | Second heartbeat destination — same URL form as `api_url` (blank = off) |
+| `mirror_sfm_url` | `""` | Second SFM base URL, e.g. `http://10.0.0.x:8200` (blank = off) |
+| `mirror_sfm_state_file` | `""` | Override the mirror's forwarded-sha256 state file. Blank → `<log_dir>\thor_forwarded_mirror.json` |
+
+### Changed
+- Bumped `VERSION` to `0.4.0`
+
+> **Operational note (migration seeding).** To make the mirror re-deliver only the gap since a one-time replica snapshot, copy the primary state file to the mirror state file and drop entries whose `forwarded_at` is after the snapshot. SFM-side dedup covers any overlap. See `docs/mirror-dual-send-design.md`.
+
+
 ## [0.3.1] - 2026-05-20
 
 ### Hot fix

README.md

@@ -1,8 +1,8 @@
 # Thor Watcher
 
-**Version:** 0.3.1
+**Version:** 0.4.1
 
-Micromate (Series 4) watcher agent for Terra-View fleet management. Runs as a Windows system tray application, scans THORDATA for Micromate unit activity, sends heartbeat data to Terra-View, and (optionally) forwards `.IDFH`/`.IDFW` event files to a seismo-relay SFM server.
+Micromate (Series 4) watcher agent for Terra-View fleet management. Runs as a Windows system tray application, scans THORDATA for Micromate unit activity, sends heartbeat data to Terra-View, and (optionally) forwards `.IDFH`/`.IDFW` event files to a seismo-relay SFM server. Heartbeats and events can optionally be **mirrored** to a second server (e.g. a standby/NAS) in addition to the primary.
 
 ---
 
@@ -29,7 +29,7 @@ build.bat
 ```
 
 Produces:
-- `dist\thor-watcher-0.3.1.exe` — upload to Gitea release
+- `dist\thor-watcher-0.4.1.exe` — upload to Gitea release
 - `dist\thor-watcher.exe` — use with Inno Setup
 
 Then run Inno Setup Compiler on `installer.iss` to produce `thor-watcher-setup.exe`.
@@ -71,6 +71,9 @@ Managed through the Settings dialog (right-click tray icon → Settings). A `con
 | `sfm_state_file` | string | `""` | Path to the sha256-keyed state file. Blank → `<log_dir>\thor_forwarded.json` |
 | `sfm_max_forwards_per_pass` | integer | `500` | Cap per pass to drip-feed large backfills |
 | `sfm_max_event_age_days` | integer | `365` | Skip event files older than this many days |
+| `mirror_api_url` | string | `""` | Second heartbeat destination (same URL form as `api_url`). Blank → heartbeat mirror off |
+| `mirror_sfm_url` | string | `""` | Second SFM base URL (e.g. `http://10.0.0.x:8200`). Blank → event mirror off |
+| `mirror_sfm_state_file` | string | `""` | Path to the mirror's sha256 state file. Blank → `<log_dir>\thor_forwarded_mirror.json` |
 
 ---
 
@@ -86,6 +89,21 @@ When `sfm_forward_enabled` is true and `sfm_url` is set, Thor Watcher walks the
 
 ---
 
+## Mirror (Dual-Send)
+
+Thor Watcher can post every heartbeat and event to a **second ("mirror") server** in addition to the primary — useful for keeping a standby box (e.g. the office NAS) as a continuous replica during a server migration, so the eventual cutover is a non-event. **Default off.**
+
+The mirror rides along with the primary — there is no separate enable flag:
+
+- **Heartbeat mirror** is active when the API heartbeat is enabled **and** `mirror_api_url` is set. After the primary heartbeat POST, the same payload is fired at `mirror_api_url`.
+- **Event mirror** is active when SFM forwarding is enabled **and** `mirror_sfm_url` is set. Events are re-forwarded to `mirror_sfm_url` using a **separate** sha256 state file (`mirror_sfm_state_file`, blank → `<log_dir>\thor_forwarded_mirror.json`).
+
+**The mirror can never delay or fail the primary.** The heartbeat mirror is fully wrapped and bounded by the API timeout. The event mirror runs a quick reachability probe (~3 s) first and skips the pass if the mirror server is down — skipped events stay pending in the mirror state and deliver when it returns, so nothing is lost. All mirror exceptions are swallowed-and-logged; the primary path stays exactly as reliable as before.
+
+- **Seeding for a migration.** To re-deliver only the gap since a one-time replica snapshot, copy the primary state file to the mirror state file and drop entries whose `forwarded_at` is after the snapshot. SFM-side dedup covers any overlap. See [`docs/mirror-dual-send-design.md`](docs/mirror-dual-send-design.md).
+
+---
+
 ## Tray Icon Colors
 
 | Color | Meaning |
@@ -121,7 +139,7 @@ Posted to `api_url` on each API interval:
 {
   "source_id": "THOR-PC",
   "source_type": "series4_watcher",
-  "version": "0.3.1",
+  "version": "0.4.1",
   "generated_at": "2026-03-20T14:30:00Z",
   "log_tail": ["...last 25 log lines..."],
   "units": [

installer.iss

@@ -3,7 +3,7 @@
 
 [Setup]
 AppName=Thor Watcher
-AppVersion=0.3.1
+AppVersion=0.4.1
 AppPublisher=Terra-Mechanics Inc.
 DefaultDirName={pf}\ThorWatcher
 DefaultGroupName=Thor Watcher

series4_ingest.py

@@ -1,5 +1,5 @@
 """
-Thor Watcher — Series 4 Ingest Agent  v0.3.1
+Thor Watcher — Series 4 Ingest Agent  v0.4.1
 
 Micromate (Series 4) ingest agent for Terra-View.
 
@@ -29,7 +29,7 @@ import event_forwarder
 
 # ── Version ───────────────────────────────────────────────────────────────────
 
-VERSION = "0.3.1"
+VERSION = "0.4.1"
 
 
 # ── Config ────────────────────────────────────────────────────────────────────

test_thor_tray.py

@@ -0,0 +1,81 @@
+"""
+Tests for thor_tray's thread lifecycle.
+
+Regression guard for the bug where saving settings — which restarts the
+watcher thread via ``_restart_watcher`` — also killed the auto-updater
+thread, because both shared a single ``stop_event``. The updater must key
+off a separate lifetime event (``_app_stop``) that a watcher restart never
+disturbs; it should only stop when the app actually exits.
+"""
+import sys
+import time
+import threading
+import unittest
+from unittest import mock
+
+# Thor's tray pulls in Windows-only GUI deps at import time; stub them so the
+# module imports on a headless dev/CI box.
+for _name in ("pystray", "PIL", "PIL.Image", "PIL.ImageDraw"):
+    sys.modules.setdefault(_name, mock.MagicMock())
+
+import thor_tray  # noqa: E402
+
+
+class TrayThreadLifecycle(unittest.TestCase):
+    def setUp(self):
+        # Keep _start_watcher cheap: a no-op watcher loop so restarting the
+        # watcher doesn't spin up the real THORDATA scanner / network heartbeat.
+        patcher = mock.patch.object(
+            thor_tray.watcher, "run_watcher", lambda state, stop_event: None
+        )
+        patcher.start()
+        self.addCleanup(patcher.stop)
+
+    def test_app_stop_is_separate_lifetime_event(self):
+        app = thor_tray.WatcherTray()
+        self.assertIsInstance(app._app_stop, threading.Event)
+        self.assertIsNot(app._app_stop, app.stop_event)
+        self.assertFalse(app._app_stop.is_set())
+
+    def test_exit_signals_app_stop(self):
+        app = thor_tray.WatcherTray()
+        icon = mock.MagicMock()
+        app._exit(icon, None)
+        self.assertTrue(app._app_stop.is_set())
+        icon.stop.assert_called_once()
+
+    def test_restart_watcher_leaves_watcher_stop_unset(self):
+        app = thor_tray.WatcherTray()
+        app._restart_watcher()
+        self.assertFalse(app.stop_event.is_set())
+
+    def test_updater_survives_watcher_restart_then_stops_on_exit(self):
+        app = thor_tray.WatcherTray()
+        app._icon = None
+        with mock.patch.object(thor_tray, "check_for_update", return_value=(None, None)), \
+                mock.patch.object(thor_tray, "_update_log"), \
+                mock.patch.object(app, "_tray_status", return_value="ok"):
+            t = threading.Thread(
+                target=app._icon_updater, daemon=True, name="test-updater"
+            )
+            t.start()
+            time.sleep(0.1)
+            self.assertTrue(t.is_alive(), "updater failed to start")
+
+            # A settings save restarts the watcher — it must NOT kill the updater.
+            app._restart_watcher()
+            time.sleep(0.2)
+            self.assertTrue(
+                t.is_alive(), "updater thread died on watcher restart (the bug)"
+            )
+
+            # Real app exit DOES stop the updater.
+            app._app_stop.set()
+            t.join(timeout=2)
+            self.assertFalse(
+                t.is_alive(), "updater thread did not stop on app exit"
+            )
+
+
+if __name__ == "__main__":
+    unittest.main()

thor_settings_dialog.py

@@ -1,5 +1,5 @@
 """
-Thor Watcher — Settings Dialog  v0.3.1
+Thor Watcher — Settings Dialog  v0.4.1
 
 Provides a Tkinter settings dialog that doubles as a first-run wizard.
 

thor_tray.py

@@ -1,5 +1,5 @@
 """
-Thor Watcher — System Tray Launcher  v0.3.1
+Thor Watcher — System Tray Launcher  v0.4.1
 Requires: pystray, Pillow, tkinter (stdlib)
 
 Run with:  pythonw thor_tray.py   (no console window)
@@ -334,7 +334,8 @@ def _show_cancel_message():
 class WatcherTray:
     def __init__(self):
         self.state = {}
-        self.stop_event = threading.Event()
+        self.stop_event = threading.Event()   # watcher lifecycle; replaced on restart
+        self._app_stop = threading.Event()    # app lifetime; only set on exit/update
         self._watcher_thread = None
         self._icon = None
         self._menu_lock = threading.Lock()
@@ -390,6 +391,7 @@ class WatcherTray:
                 subprocess.Popen(["explorer", HERE])
 
     def _exit(self, icon, item):
+        self._app_stop.set()
         self.stop_event.set()
         icon.stop()
 
@@ -478,7 +480,7 @@ class WatcherTray:
         # line in the log soon after startup; subsequent checks every ~5 min.
         update_check_counter = 27
 
-        while not self.stop_event.is_set():
+        while not self._app_stop.is_set():
             icon_status = self._tray_status()
 
             if self._icon is not None:
@@ -504,7 +506,7 @@ class WatcherTray:
                     self._do_update(url)
                     return
 
-            self.stop_event.wait(timeout=10)
+            self._app_stop.wait(timeout=10)
 
     def _do_update(self, download_url=None):
         """Notify tray then apply update. If url is None, fetch it first."""
@@ -519,6 +521,7 @@ class WatcherTray:
 
         success = apply_update(download_url)
         if success:
+            self._app_stop.set()
             self.stop_event.set()
             if self._icon is not None:
                 self._icon.stop()