fix(install): survive a reboot

Two independent reasons the bridge stayed down after a reboot, both found by
actually rebooting a Venus device instead of reading the script:

- /service is a tmpfs, so the symlink is gone at boot and rc.local re-runs
  install.sh. The old check ('no symlink' = first install) then dropped the
  'down' marker again and supervise left the bridge off. A persistent
  .installed marker now separates a genuine first install from a boot re-entry;
  uninstall.sh removes it again.
- Venus only executes /data/rc.local when it carries the executable bit
  (/etc/init.d/custom-rc-late.sh). install.sh only chmod'ed the file when it
  created it, so an existing 0644 rc.local disabled the autostart silently.
  It is now chmod'ed on every install.

Verified by two reboots of the testberry: before the fix no service and no
/run/venus/custom-rc marker, after it the bridge was up 35 seconds after boot
with DeviceInstances 49/55/56 and its counters unchanged.

Author: okuegow

CHANGELOG.md

@@ -46,6 +46,14 @@ adheres to [Semantic Versioning](https://semver.org/spec/v2.0.0.html).
   write, so no migration step is needed.
 
 ### Fixed
+- **The bridge now survives a reboot.** `/service` is a tmpfs, so rc.local
+  re-runs `install.sh` on every boot - and that looked like a first install,
+  which dropped the `down` marker again and left the bridge off until someone
+  noticed the chargers missing in VRM. A persistent `.installed` marker now
+  tells the two cases apart.
+- `install.sh` makes `/data/rc.local` executable on every run. Venus only runs
+  it when it is (`/etc/init.d/custom-rc-late.sh`: `if [ -x /data/rc.local ]`),
+  and a non-executable file silently disabled the autostart.
 - `service/run` no longer restarts the bridge in a tight loop after a
   configuration error. daemontools respawns immediately, and that spin is
   enough to trip the Venus load watchdog (`/etc/watchdog.conf`: `max-load-5 =

install.sh

@@ -11,29 +11,47 @@ chmod 755 "$SCRIPT_DIR/service/run" "$SCRIPT_DIR/service/log/run"
 # Ensure log directory exists (multilog needs it writable)
 mkdir -p "/data/log/$SERVICE_NAME"
 
-# Symlink into daemontools. On FIRST install we drop a 'down' marker so
+# Symlink into daemontools. On the FIRST install we drop a 'down' marker so
 # supervise does not auto-start the bridge before the operator has run
-# seed_state.py (otherwise fresh DIs get allocated and the legacy title->DI
-# mapping is lost). On re-install (e.g. rc.local re-entry after reboot) the
-# symlink already exists -> no 'down' file is created -> service comes back up.
+# seed_state.py (otherwise fresh DeviceInstances get allocated and the legacy
+# title->DI mapping is lost).
+#
+# /service is a tmpfs, so after a reboot the symlink is gone and rc.local runs
+# this script again. That must NOT look like a first install - it used to drop
+# the 'down' marker again and the bridge stayed off until someone noticed the
+# chargers missing in VRM. The marker file below survives the reboot and tells
+# the two cases apart.
+INSTALL_MARKER="$SCRIPT_DIR/.installed"
 if [ ! -L "/service/$SERVICE_NAME" ]; then
-    touch "$SCRIPT_DIR/service/down"
-    ln -s "$SCRIPT_DIR/service" "/service/$SERVICE_NAME"
-    echo "Created /service/$SERVICE_NAME (in 'down' state - not running yet)"
-    FIRST_INSTALL=1
+    if [ -f "$INSTALL_MARKER" ]; then
+        rm -f "$SCRIPT_DIR/service/down"
+        ln -s "$SCRIPT_DIR/service" "/service/$SERVICE_NAME"
+        echo "Re-created /service/$SERVICE_NAME after reboot (service will start)"
+        FIRST_INSTALL=0
+    else
+        touch "$SCRIPT_DIR/service/down"
+        ln -s "$SCRIPT_DIR/service" "/service/$SERVICE_NAME"
+        echo "Created /service/$SERVICE_NAME (in 'down' state - not running yet)"
+        FIRST_INSTALL=1
+    fi
 else
     echo "Service symlink already exists, skipping"
     FIRST_INSTALL=0
 fi
+touch "$INSTALL_MARKER"
 
 # Persist across firmware updates via rc.local
 filename=/data/rc.local
 if [ ! -f "$filename" ]; then
     touch "$filename"
-    chmod 755 "$filename"
     echo "#!/bin/bash" >> "$filename"
     echo >> "$filename"
 fi
+# Venus only runs /data/rc.local when it is executable
+# (/etc/init.d/custom-rc-late.sh: `if [ -x /data/rc.local ]`). An existing but
+# non-executable file is the silent way to lose the bridge after a reboot, so
+# fix the mode on every install, not just when we create the file.
+chmod 755 "$filename"
 grep -qxF "$SCRIPT_DIR/install.sh" "$filename" || echo "$SCRIPT_DIR/install.sh" >> "$filename"
 
 # The legacy single-loadpoint bridges leave their own line in rc.local; their

uninstall.sh

@@ -71,6 +71,9 @@ fi
 
 # Reset any leftover 'down' marker so a re-install starts cleanly
 rm -f "$SCRIPT_DIR/service/down"
+# Forget that we were ever installed, so a later install.sh starts from the
+# guarded first-install path again instead of coming straight up.
+rm -f "$SCRIPT_DIR/.installed"
 
 echo
 if [ "$FAILED" = "0" ]; then