From fbd371fc03c67304c80af69b6f669bc676a1810a Mon Sep 17 00:00:00 2001
From: Alexander Nicolay <alexnly@icloud.com>
Date: Mon, 6 Jul 2026 07:56:24 +0200
Subject: [PATCH] Retry pending journal sync when the machine comes back online

---
 src/matebot/cli.py       | 15 ++++++++++++--
 src/matebot/sync.py      | 29 ++++++++++++++++++++++-----
 tests/test_sync_retry.py | 43 ++++++++++++++++++++++++++++++++++++++++
 3 files changed, 80 insertions(+), 7 deletions(-)
 create mode 100644 tests/test_sync_retry.py

diff --git a/src/matebot/cli.py b/src/matebot/cli.py
index ef2d9e5..bb1ab4c 100644
--- a/src/matebot/cli.py
+++ b/src/matebot/cli.py
@@ -112,12 +112,12 @@ def events(self):
 
             messenger = create_messenger(config)
 
-        def schedule_sync():
+        def schedule_sync(*, quiet: bool = False):
             if config.sync_enabled and config.data_repo:
                 asyncio.create_task(
                     sync_soon(
                         client, config.data_repo, messenger.send,
-                        site_title=config.site_title,
+                        site_title=config.site_title, state=state, quiet=quiet,
                     )
                 )
 
@@ -171,9 +171,20 @@ async def pump_events():
 
             async def pump_shots():
                 import re as _re
+                import time as _time
+
+                last_frame_at = 0.0
 
                 async def tee(source):
+                    nonlocal last_frame_at
                     async for frame in source:
+                        now = _time.monotonic()
+                        gap = now - last_frame_at if last_frame_at else None
+                        last_frame_at = now
+                        if state.get("sync_pending") and (gap is None or gap > 60):
+                            log.info("machine is back; retrying pending journal sync")
+                            state.set("sync_pending", False)  # avoid re-trigger storms
+                            schedule_sync(quiet=True)
                         cache_frame(frame)
                         await router.on_frame(frame)
                         yield frame
diff --git a/src/matebot/sync.py b/src/matebot/sync.py
index 98b0e81..df0443a 100644
--- a/src/matebot/sync.py
+++ b/src/matebot/sync.py
@@ -9,7 +9,6 @@
 
 from __future__ import annotations
 
-import asyncio
 import fcntl
 import json
 import logging
@@ -17,6 +16,8 @@
 import subprocess
 from pathlib import Path
 
+import aiohttp
+
 from .machine import GaggiMateClient, MachineError
 from .sitegen import generate
 
@@ -130,15 +131,33 @@ async def sync(
 
 
 async def sync_soon(
-    client: GaggiMateClient, repo: str | Path, notify, *, site_title: str = "Shot Journal"
+    client: GaggiMateClient, repo: str | Path, notify, *,
+    site_title: str = "Shot Journal", state=None, quiet: bool = False,
 ) -> None:
-    """Post-shot sync wrapper: run, report conflicts to the user, never raise."""
+    """Post-shot sync wrapper: run, report problems, never raise.
+
+    A failed sync is remembered in *state* (``sync_pending``) so the caller
+    can retry when the machine comes back online.
+    """
     try:
         await sync(client, repo, site_title=site_title)
     except SyncConflict as exc:
         await notify(f"⚠️ Shot journal sync hit a git conflict — fix it manually:\n{exc}")
+    except (TimeoutError, aiohttp.ClientError, OSError) as exc:
+        log.warning("sync failed, machine unreachable: %r", exc)
+        if state is not None:
+            state.set("sync_pending", True)
+        if not quiet:
+            await notify(
+                "📡 The machine went offline before the journal could sync — "
+                "I'll catch up as soon as it's back on."
+            )
     except Exception as exc:  # noqa: BLE001
         log.exception("sync failed")
-        await notify(f"⚠️ Shot journal sync failed: {exc}")
+        if state is not None:
+            state.set("sync_pending", True)
+        if not quiet:
+            await notify(f"⚠️ Shot journal sync failed: {exc!r}")
     else:
-        await asyncio.sleep(0)
+        if state is not None:
+            state.set("sync_pending", False)
diff --git a/tests/test_sync_retry.py b/tests/test_sync_retry.py
new file mode 100644
index 0000000..58de77a
--- /dev/null
+++ b/tests/test_sync_retry.py
@@ -0,0 +1,43 @@
+import aiohttp
+import pytest
+
+from matebot import sync as sync_mod
+from matebot.state import State
+from matebot.sync import sync_soon
+
+
+class Notify:
+    def __init__(self):
+        self.sent = []
+
+    async def __call__(self, text):
+        self.sent.append(text)
+
+
+@pytest.mark.asyncio
+async def test_machine_offline_sets_pending_flag(tmp_path, monkeypatch):
+    async def boom(client, repo, site_title="x"):
+        raise aiohttp.ClientConnectionError("machine off")
+
+    monkeypatch.setattr(sync_mod, "sync", boom)
+    state = State(tmp_path / "state.json")
+    notify = Notify()
+    await sync_soon(None, tmp_path, notify, state=state)
+    assert state.get("sync_pending") is True
+    assert "offline" in notify.sent[0]
+
+    # quiet retry path: no message even on repeated failure
+    await sync_soon(None, tmp_path, notify, state=state, quiet=True)
+    assert len(notify.sent) == 1
+
+
+@pytest.mark.asyncio
+async def test_success_clears_pending_flag(tmp_path, monkeypatch):
+    async def ok(client, repo, site_title="x"):
+        return True
+
+    monkeypatch.setattr(sync_mod, "sync", ok)
+    state = State(tmp_path / "state.json")
+    state.set("sync_pending", True)
+    await sync_soon(None, tmp_path, Notify(), state=state)
+    assert state.get("sync_pending") is False