OverlayFS-based write reduction plugin for openmediavault 7

• Uses tmpfs + overlayfs for selected write-heavy system paths (e.g., /var/log, APT cache).
• Optional zram (compressed RAM) backing — a log2ram-style workspace that holds more data per MB of RAM. Like tmpfs it is volatile; use a shared folder for power-loss durability.
• Safe defaults: caches dropped at reboot; optional flush on shutdown & daily timer.
• Journald Storage=volatile option to keep logs in RAM.
• Salt-managed config at /etc/omv-writecache/config.yaml.
• Systemd oneshot units for mount/flush.
• Minimal web UI (Workbench YAML) under Services → WriteCache.

sudo /usr/sbin/omv-writecache mount
sudo /usr/sbin/omv-writecache flush
sudo /usr/sbin/omv-writecache unmount
sudo /usr/sbin/omv-writecache status

By default /var/log is cached with policy drop: cached writes are discarded on unmount/reboot and are never written back, so periodic flush/rotateflush runs never touch that overlay and journald is left alone.

If you instead set /var/log to flush, persist, or writeback (to keep journal logs across the cache cycle), each flush unmounts and remounts a fresh overlay under journald's active journal file. journald keeps that file open, so the unmount falls back to a lazy detach, and once the new overlay is in place journald notices its file identity changed, logs Journal file has been deleted, rotating, and rotates. rotateflush already runs journalctl --rotate && journalctl --sync beforehand to minimize this, but a small window (and the log message) can still occur.

To close that window, add an explicit restart to the services list and run with --restart:

systemd-journald = restart

This makes the plugin restart journald right after the new overlay is mounted instead of waiting on journald's own detection. systemd buffers the journal socket across the restart, so no messages are lost from the restart itself. Only add this if you've changed /var/log's policy away from the drop default — with drop, flushes never touch /var/log and the restart would just interrupt logging for no benefit.