Skip to content

snapraid-sync — guard, precheck, then write parity

Nightly at 03:30 the timer starts the unit; the unit runs the guard, which runs the precheck, and only then snapraid sync. A failed precheck is a failed unit and pages ntfy.

Input → Movement → Output

Reads the three data disks and the content files. Recomputes parity for changed blocks (deleted blocks are flagged and recomputed too — a big exclusion rewrites hundreds of GB). Writes parity, then the content file to all four locations; autosave 100 checkpoints every 100 GB.

Why this shape

The one destructive failure mode of offline parity is syncing against a missing or mass-deleted disk: the loss becomes the new truth. The precheck refuses to sync when a mount is missing or read-only, SMART is unhealthy, the kernel logged I/O errors in 6 h, or snapraid diff shows > 500 removed / > 2,000 updated files.

Steps

  1. snapraid-sync.timer → snapraid-sync.service; ExecStart overridden to /usr/local/bin/snapraid-sync-guard.sh (snapraid-sync.service.d/override.conf).
  2. Guard runs /home/angel/snapraid-precheck.sh (snapraid-sync-guard.sh:4-10); on failure logger + exit 1 → OnFailure=snapraid-alert@%n (.d/alert.conf).
  3. Precheck: mounts rw (:7-12) → SMART health + pending/offline on sda–sdd, skipping a drive in standby (smartctl -n standby, added 09-09 so a no-op night does not leave parity spinning) → kernel I/O errors (:37-39) → snapraid status (:41) → snapraid diff thresholds DEL_LIMIT=500 UPD_LIMIT=2000 (:48-59, overridable via env for a known large change).
  4. exec /usr/bin/snapraid -l /var/log/snapraid/sync-%D.log sync (snapraid-sync-guard.sh:12).

Manual: sr (runs the unit, blocks, reports). For a known large change: sudo DEL_LIMIT=N /home/angel/snapraid-precheck.sh to see the numbers, then run the sync by hand with -l as item 06 did.

If you change this

  • Hits: the alert path (unit name is in OnFailure) · the rescue kit copies both scripts nightly (rescue-sync.sh:112) · scrub runs at Sun 04:00 and expects a synced array.
  • Does not hit: mergerfs, containers (sync reads through /mnt/diskN, is Nice=10 + idle I/O class).

Surfaces

Surface Role
ntfy plex-hgviaey failure only
/var/log/snapraid/ per-run logs
journalctl -u snapraid-sync precheck messages

See

  • Objects: snapraid-array
  • Source: systemctl cat snapraid-sync.service · /usr/local/bin/snapraid-sync-guard.sh · /home/angel/snapraid-precheck.sh